ECSS HB 32 24A Buckling of Structures

ECSS-E-HB-32-24A
24 March 2010
Space engineering
Buckling of structures
ECSS Secretariat
ESA-ESTEC
Requirements & Standards Division
Noordwijk, The Netherlands
ECSSEHB3224A
24March2010
Foreword
ThisHandbookisonedocumentoftheseriesofECSSDocumentsintendedtobeusedassupporting
material for ECSS Standards in space projects and applications. ECSS is a cooperative effort of the
EuropeanSpaceAgency,nationalspaceagenciesandEuropeanindustryassociationsforthepurpose
ofdevelopingandmaintainingcommonstandards.
ThishandbookhasbeenpreparedbytheECSSEHB3224AWorkingGroup,reviewedbytheECSS
ExecutiveSecretariatandapprovedbytheECSSTechnicalAuthority.
Disclaimer
ECSSdoesnotprovideanywarrantywhatsoever,whetherexpressed,implied,orstatutory,including,
butnotlimitedto,anywarrantyofmerchantabilityorfitnessforaparticularpurposeoranywarranty
that the contents of the item are errorfree. In no respect shall ECSS incur any liability for any
damages,including,butnotlimitedto,direct,indirect,special,orconsequentialdamagesarisingout
of,resultingfrom,orinanywayconnectedtotheuseofthisdocument,whetherornotbasedupon
warranty,businessagreement,tort,orotherwise;whetherornotinjurywassustainedbypersonsor
propertyorotherwise;andwhetherornotlosswassustainedfrom,oraroseoutof,theresultsof,the
item,oranyservicesthatmaybeprovidedbyECSS.
Publishedby: ESARequirementsandStandardsDivision
ESTEC,P.O.Box299,
2200AGNoordwijk
TheNetherlands
Copyright: 2010bytheEuropeanSpaceAgencyforthemembersofECSS
2
ECSSEHB3224A
24March2010
Change log
ECSSEHB3224A Firstissue
24March2010
3
ECSSEHB3224A
24March2010
Table of contents
Change log .................................................................................................................3
Acknowledgements.................................................................................................16
Introduction..............................................................................................................17
1 Scope.....................................................................................................................19
2 References ............................................................................................................19
3 Terms, definitions and abbreviated terms..........................................................20

3.1 Terms from other documents ................................................................................... 20
3.2 Terms specific to the present document .................................................................. 21
3.3 Abbreviated terms .................................................................................................... 22
Part 1: General .........................................................................................................23
4 Overview................................................................................................................24
4.1 Instability of structures: concepts and definitions ..................................................... 24
4.2 Stability of columns, beams and plates .................................................................... 25
4.2.1 Buckling of beams ...................................................................................... 25
4.2.2 Lateral torsional buckling............................................................................ 25
4.2.3 Buckling of plates ....................................................................................... 26
4.3 History of shell buckling............................................................................................ 26
4.3.1 Overview..................................................................................................... 26
4.3.2 Boundary conditions and nonlinear pre-buckling behaviour....................... 28
4.3.3 Postbuckling of the perfect shell................................................................. 28
4.3.4 Buckling and postbuckling behaviour of imperfect shells ........................... 30
4.4 Design load for thin-walled isotropic shells............................................................... 32
4.4.1 Overview..................................................................................................... 32
4.4.2 Empirical knock-down factors..................................................................... 32
4.4.3 The Southwell method................................................................................ 33
4.4.4 Numerical analysis with realistic geometrical imperfections....................... 34
4.4.5 Numerical analysis with worst geometrical imperfection ............................ 34
4
ECSSEHB3224A
24March2010
4.4.6 Numerical analysis with stimulating imperfections ..................................... 35
4.4.7 Physics based design load ......................................................................... 35
4.4.8 Overview of different approaches for modelling imperfections................... 37
4.5 Influence of non-traditional imperfections................................................................. 38
4.6 External pressure and torsion on cylindrical shells................................................... 39
4.7 Dynamic buckling ..................................................................................................... 39
4.8 Stiffened shells ......................................................................................................... 40
4.8.1 Overview..................................................................................................... 40
4.8.2 Imperfection sensitivity ............................................................................... 41
4.8.3 Orthotropic shell approach ......................................................................... 42
4.8.4 Eccentricity of stringers .............................................................................. 43
4.8.5 Boundary conditions ................................................................................... 43
4.8.6 Load cases ................................................................................................. 43
4.8.7 Optimization................................................................................................ 44
4.9 Plastic buckling......................................................................................................... 44
4.10 Composite shells ...................................................................................................... 45
4.11 References ............................................................................................................... 47
4.12 Abbreviated Symbols ............................................................................................... 63
5 Types of instability and failure behaviour of typical structural elements .......64

5.1 Introduction............................................................................................................... 64
5.2 Elastic buckling of columns ...................................................................................... 64
5.2.1 Overview..................................................................................................... 64
5.2.2 Buckling of columns with compound cross-sections .................................. 66
5.3 Buckling of thin plates .............................................................................................. 69
5.3.1 Overview..................................................................................................... 69
5.3.2 Buckling behaviour of stiffened panels ....................................................... 72
5.3.3 Buckling behaviour of sandwich plates....................................................... 73
5.4 Instability of axially compressed cylindrical panels .................................................. 75
5.5 Structural behaviour of thin shells ............................................................................ 77
5.5.1 Overview..................................................................................................... 77
5.5.2 Buckling of cylindrical shells ....................................................................... 77
5.5.3 Buckling of Shells of Revolution ................................................................. 79
5.6 References ............................................................................................................... 85
6 Analysis Methods .................................................................................................87

6.1 Introduction............................................................................................................... 87
6.2 Static equilibrium and stability .................................................................................. 87
6.2.1 Overview..................................................................................................... 87
5
ECSSEHB3224A
24March2010
6.2.2 Turning points and failure points ................................................................ 88
6.2.3 Types of loadings ....................................................................................... 89
6.2.4 Distinction between equilibrium and stability equations ............................. 90
6.2.5 Static stability in the sense of Hadamard and Euler ................................... 90
6.2.6 The non linear eigenvalue problem for Euler stability................................. 91
6.3 Static equilibrium and stability of conservative systems........................................... 91
6.4 Solution of nonlinear equilibrium equations.............................................................. 92
6.5 Finding critical points of the equilibrium path............................................................ 96
6.6 Stability at the critical points of a discrete conservative system ............................... 97
6.7 Imperfection analysis.............................................................................................. 101
6.8 Dynamic stability analysis ...................................................................................... 104
6.9 References ............................................................................................................. 112
7 Material characteristics......................................................................................113
7.1 Overview ................................................................................................................ 113
7.2 Linear elasticity and elastic properties ................................................................... 113
7.2.1 Overview................................................................................................... 113
7.2.2 Orthotropic elasticity ................................................................................. 114
7.2.3 Transversely-isotropic elasticity of UD materials...................................... 115
7.2.4 Isotropic elasticity ..................................................................................... 116
7.2.5 Viscoelasticity ........................................................................................... 116
7.3 Strength properties and hygro-thermal properties.................................................. 116
7.3.1 Strength properties ................................................................................... 116
7.3.2 Hygro-thermal properties.......................................................................... 117
7.4 Elastic and inelastic material behaviour ................................................................. 118
7.4.1 Overview................................................................................................... 118
7.4.2 Stress-strain curve of isotropic materials.................................................. 119
7.4.3 Special aspects with composites.............................................................. 120
7.4.4 Mapping of a stress-strain curve .............................................................. 121
7.4.5 Cyclic hardening behaviour ...................................................................... 121
7.5 Plasticity and damage ............................................................................................ 122
7.6 Material testing methods, test data, and evaluation ............................................... 122
7.6.1 Overview................................................................................................... 122
7.6.2 Determination of a design allowable......................................................... 124
7.6.3 Sources for finding property data ............................................................. 124
7.7 Some practical aspects for stability analyses ......................................................... 124
7.7.1 Effect of yielding and damage on stiffness ............................................... 124
7.7.2 Effect of temperature and moisture on property values............................ 126
6
ECSSEHB3224A
24March2010
7.7.3 Visco-plasticity, strain rate and impact ..................................................... 126
7.7.4 Miscellaneous........................................................................................... 126
7.8 References ............................................................................................................. 128
8 Design to stability and its verification by analysis ..........................................130

8.1 Introduction............................................................................................................. 130
8.2 Design development process ................................................................................. 130
8.3 Analysis pre-work and load input data ................................................................... 131
8.3.1 Overview................................................................................................... 131
8.3.2 Load analysis, limit load, and design limit load......................................... 132
8.4 Safety concept - Factors of safety (FoS) ................................................................ 132
8.5 Dimensioning load cases selection ........................................................................ 134
8.5.1 Margin policy and design loads ................................................................ 134
8.5.2 Design loads combinations relevant for buckling ..................................... 135
8.5.3 Dimensioning load cases.......................................................................... 136
8.6 Remaining input data ............................................................................................. 136
8.6.1 Overview................................................................................................... 136
8.6.2 Influences from manufacturing and manufacturing signatures................. 136
8.6.3 Boundary conditions ................................................................................. 137
8.6.4 Application of properties ........................................................................... 137
8.6.5 Specific definitions and notions ................................................................ 137
8.7 Stability design allowable and knock-down-factor (KDF)............................................. 138
8.7.1 Stability design allowable (design allowable of the buckling
resistance) ................................................................................................ 138
8.7.2 Knock-Down-Factor.................................................................................. 138
8.8 Analyses, modelling, and design aspects .............................................................. 138
8.8.1 General..................................................................................................... 138
8.8.2 Optimization and robust design ................................................................ 139
8.8.3 Idealisation of geometry and modelling .................................................... 139
8.8.4 Analysis solution types ............................................................................. 140
8.8.5 Imperfection sensitivity analysis (nonlinear) ............................................. 140
8.8.6 Choice of calculation method ................................................................... 141
8.9 Procedure of design verification ............................................................................. 142
8.9.1 Overview................................................................................................... 142
8.9.2 Verification policy...................................................................................... 143
8.9.3 Assessment of the structure ..................................................................... 144
8.10 References ............................................................................................................. 144
8.11 Abbreviated Terms and Symbols ........................................................................... 145
7
ECSSEHB3224A
24March2010
9 Influence of manufacturing and assembly processes on the buckling
load.....................................................................................................................146
9.1 Overview ................................................................................................................ 146
9.2 Metallic Structures Integral Machined Curved Panels......................................... 146
9.3 Metallic Structures Deformations due to Welding................................................ 147
9.4 Composite Structures ............................................................................................. 147
9.5 Assembly Stresses and Deformations ................................................................... 147
9.6 The Imperfection Data Bank, a Means to Obtain a Realistic Buckling Load .......... 148
9.6.1 Overview................................................................................................... 148
9.6.2 Imperfection surveys on large or full scale shells ..................................... 149
9.6.3 Comparison of the measured initial imperfections.................................... 159
9.6.4 Characteristic imperfection distributions................................................... 163
9.6.5 Probabilistic stability analysis ................................................................... 163
9.6.6 Conclusions .............................................................................................. 166
9.7 References ............................................................................................................. 167
10 Modelling aspects of numerical analysis .......................................................168

10.1 Introduction............................................................................................................. 168
10.2 Semi-analytical models - shooting method and finite difference method ............... 168
10.2.1 Overview................................................................................................... 168
10.2.2 Description of the numerical methods ...................................................... 169
10.2.3 Capabilities and scope of the programs ................................................... 169
10.3 Finite Element Model.............................................................................................. 170
10.3.1 Finite Element Model Generation ............................................................. 170
10.3.2 Finite Element Model Element Mesh........................................................ 170
10.3.3 Selection of Finite Elements ..................................................................... 170
10.3.4 Finite Element Model Boundary Conditions ............................................. 171
10.3.5 Finite Element Model Load applications................................................... 171
10.3.6 Implementation of discontinuities in the finite element model .................. 172
10.3.7 Finite Element Model Check..................................................................... 172
10.4 Thermo-Mechanical Loading .................................................................................. 173
10.5 Recommended Numerical Procedures .................................................................. 173
10.5.1 Overview................................................................................................... 173
10.5.2 Load Application ....................................................................................... 173
10.5.3 Iteration Schemes..................................................................................... 174
10.5.4 Convergence Criteria................................................................................ 176
10.5.5 Estimation of Bifurcation Points................................................................ 176
10.5.6 Post-Buckling Analysis ............................................................................. 177
8
ECSSEHB3224A
24March2010
10.6 Allowable Buckling Load ........................................................................................ 177
10.7 Finite Element Programmes with Buckling Analysis Capabilities ........................... 178
10.7.1 Overview................................................................................................... 178
10.7.2 ABAQUS/Standard ................................................................................... 179
10.7.3 ANSYS ..................................................................................................... 179
10.7.4 MSC.Marc/MSC.Nastran .......................................................................... 179
10.7.5 ABAQUS/EXPLICIT.................................................................................. 180
10.8 Guidelines to select a solution scheme and solution procedure............................. 180
10.8.1 Guidelines to select a solution scheme .................................................... 180
10.8.2 Guidelines for the solution procedure....................................................... 182
10.9 References ............................................................................................................. 182
11 Strategy for hierarchical high fidelity analysis applied to stability

analysis ..............................................................................................................184
11.1 Introduction............................................................................................................. 184
11.2 Hierarchical high fidelity analysis ........................................................................... 185
11.3 Flow chart............................................................................................................... 186
11.4 Description and General Guidance ........................................................................ 187
11.4.1 Overview................................................................................................... 187
11.4.2 Definition of the Problem .......................................................................... 187
11.4.3 A Priori Determination of Possible Failure Modes .................................... 187
11.4.4 Definition of the Strategy and Mathematical Model .................................. 187
11.4.5 Analytical Solution, Semi-Analytical Solution, and Finite Element
Solution of Idealized Structure.................................................................. 187
11.4.6 Finite Element Analysis of the Real Structure .......................................... 189
11.4.7 Test........................................................................................................... 192
11.4.8 Assessment of the Structure .................................................................... 192
11.5 Hierarchical High Fidelity Stability Analysis of Anisotropic Cylinders ..................... 193
11.5.1 Overview................................................................................................... 193
11.5.2 Level-1 Perfect Shell Buckling Analysis ................................................... 194
11.5.5 Imperfection Sensitivity Study .................................................................. 201
11.5.6 Single Axisymmetric Imperfection ............................................................ 201
11.5.7 Single Asymmetric Imperfection ............................................................... 206
11.5.8 Measured Initial Imperfections.................................................................. 210
11.5.9 Discussion of the Results ......................................................................... 214
11.5.10 Conclusions .............................................................................................. 214
9
ECSSEHB3224A
24March2010
11.6 References ............................................................................................................. 215
12 Buckling experimental methods and design verification by tests ...............218

12.1 Generalities of buckling tests ................................................................................. 218
12.2 Design output ......................................................................................................... 220
12.3 Objectives of the test.............................................................................................. 220
12.4 Test plan................................................................................................................. 221
12.5 Test procedures ..................................................................................................... 222
12.6 Test facility ............................................................................................................. 224
12.6.1 Overview................................................................................................... 224
12.6.2 Mounting system ...................................................................................... 224
12.6.3 Loading procedure.................................................................................... 224
12.6.4 Data acquisition ........................................................................................ 225
12.7 Documentation of the test ...................................................................................... 226
12.8 References ............................................................................................................. 227
13 Instrumentation ................................................................................................228
13.1 Introduction............................................................................................................. 228
13.2 Measurements........................................................................................................ 228
13.2.1 Classification of measurement techniques ............................................... 228
13.2.2 Measurements before the tests ................................................................ 229
13.2.3 Measurements during the tests ................................................................ 229
13.2.4 Measurements after the tests ................................................................... 229
13.3 Measurements system ........................................................................................... 230
13.3.1 Introduction............................................................................................... 230
13.3.2 Measurement of strain.............................................................................. 230
13.3.3 Displacement sensors .............................................................................. 233
13.3.4 Optical methods........................................................................................ 235
13.3.5 Force transducers..................................................................................... 238
13.3.6 Pressure transducers ............................................................................... 238
13.3.7 Temperature measurements .................................................................... 238
13.3.8 Accelerometers and vibration measurements .......................................... 239
13.3.9 Acoustic and thermal emission sensors ................................................... 239
13.3.10 Non destructive testing (NDT) .................................................................. 239
13.4 References ............................................................................................................. 241
Part 2: Structural elements with examples..........................................................243
14 Implementation of Stability Methods ..............................................................244

14.1 Overview ................................................................................................................ 244
10
ECSSEHB3224A
24March2010
14.2 Static versus kinetic approach................................................................................ 244
14.3 Problems requiring nonlinear analysis ................................................................... 250
14.4 Approximate Solutions of Bifurcation Problems ..................................................... 251
14.4.1 Overview................................................................................................... 251
14.4.2 The Rayleigh-Ritz Method ........................................................................ 252
14.4.3 Galerkins Method..................................................................................... 252
14.4.4 Finite Element Formulation of Bifurcation Problems ................................ 254
14.5 Computational Tools for Bifurcation Problems ....................................................... 255
14.6 References ............................................................................................................. 256
15 Columns, Beams, Arches and Rings ..............................................................258

15.1 Introduction............................................................................................................. 258
15.2 Columns ................................................................................................................. 258
15.2.1 Overview................................................................................................... 258
15.2.2 Long Column (elastic buckling) ................................................................ 260
15.2.3 Inelastic flexural instability ........................................................................ 262
15.2.4 Compression crippling .............................................................................. 267
15.2.5 Combined flexural and local instability ..................................................... 271
15.2.6 Torsional instability ................................................................................... 272
15.3 Lateral buckling of beams ...................................................................................... 275
15.4 The beam-column .................................................................................................. 277
15.4.1 Overview................................................................................................... 277
15.4.2 The "complete" differential equation of the beam-column ........................ 277
15.4.3 Stability under axial load........................................................................... 282
15.5 In-plane buckling of rings and arches..................................................................... 284
15.5.1 Overview................................................................................................... 284
15.5.2 The thin circular ring ................................................................................. 285
15.5.3 High circular arch...................................................................................... 288
15.6 References ............................................................................................................. 289
16 Flat and curved panels.....................................................................................290

16.1 Overview ................................................................................................................ 290
16.2 Flat Plates .............................................................................................................. 290
16.2.1 Introduction............................................................................................... 290
16.2.2 Basic Equations for Isotropic Plates ......................................................... 290
16.2.3 Applications for the Stability Equations .................................................... 292
16.2.4 Stability Equations for Orthotropic Plates ................................................. 300
16.2.5 Stability Equations for Sandwich Plates ................................................... 303
16.2.6 Failure of Plates........................................................................................ 308
11
ECSSEHB3224A
24March2010
16.3 Curved Panels........................................................................................................ 310
16.3.1 Introduction............................................................................................... 310
16.3.2 Compression loading................................................................................ 311
16.3.3 Curved Panels under Shear ..................................................................... 314
16.3.4 Curved Panels under Combined Loading................................................. 315
16.4 References ............................................................................................................. 316
17 Closed shells ....................................................................................................318

17.1 Introduction............................................................................................................. 318
17.2 Circular cylindrical shells ........................................................................................ 318
17.2.1 Level-1 solutions for isotropic shells......................................................... 320
17.2.2 Level-1 solutions for orthotropic shells ..................................................... 331
17.3 References ............................................................................................................. 339
Part 3: Structures ..................................................................................................341
18 Buckling design, analysis and testing of large structures ...........................342

18.1 Design .................................................................................................................... 342
18.1.1 General..................................................................................................... 342
18.1.2 Design aspects ......................................................................................... 342
18.1.3 Sub-structuring of large structures ........................................................... 343
18.2 Analysis and testing ............................................................................................... 344
18.2.1 Introduction............................................................................................... 344
18.2.2 Loading processing .................................................................................. 344
18.2.3 Flow chart for analysis and testing (logic) ................................................ 345
18.3 Examples to illustrate the use of the Flow Chart .................................................... 347
18.3.1 Overview................................................................................................... 347
18.3.2 Vulcain 2 Nozzle....................................................................................... 347
18.3.3 ARIANE 5 Main Cylindrical Structure ..................................................... 350
18.4 Validation of the Mathematical Model .................................................................... 351
18.5 Overview of Buckling Analysis Performed on the ARIANE Launchers................... 353
18.6 Abbreviated Terms ................................................................................................. 355
19 LH2-Tank ARIANE 5..........................................................................................356

19.1 Overview ................................................................................................................ 356
19.2 Presentation of the structure .................................................................................. 356
19.3 Sizing rules............................................................................................................. 357
19.4 Upper skirt .............................................................................................................. 358
19.4.1 Model used for the analysis...................................................................... 358
12
ECSSEHB3224A
24March2010
19.4.3 Expected failure mode.............................................................................. 359
19.4.4 Analysis method ....................................................................................... 359
19.4.6 Results...................................................................................................... 361
19.5 Y-rings .................................................................................................................... 361
19.5.1 Model used for the analysis...................................................................... 361
19.5.3 Expected failure modes ............................................................................ 361
19.5.4 Analysis method for the buckling.............................................................. 362
19.5.6 Results...................................................................................................... 362
19.6 Cylinder .................................................................................................................. 363
19.6.1 Overview................................................................................................... 363
19.6.2 Buckling analysis of the pressurized cylinder using the NASA SP-
8007.......................................................................................................... 363
19.6.3 Axisymmetric FEM analysis...................................................................... 364
19.7 Elongated lower skirt (ELS) .................................................................................... 365
19.7.1 Overview................................................................................................... 365
19.7.2 Linear analysis applying a KDF approach ................................................ 366
19.7.3 Nonlinear collapse analysis with imperfection sensitivity study ................ 367
19.8 Test Campaign ....................................................................................................... 374
19.8.1 Qualification test plan ............................................................................... 374
19.8.2 Test results ............................................................................................... 375
19.9 Conclusion.............................................................................................................. 378
19.10 References ............................................................................................................. 378
20 Buckling of the Inner Dome of the ARIANE 5 Upper Stage ESC-A...............379

20.1 Overview ................................................................................................................ 379
20.1.1 General..................................................................................................... 379
20.1.2 Buckling Modes: ....................................................................................... 381
20.2 Buckling investigation levels................................................................................... 382
20.2.1 Level-1 investigations ............................................................................... 382
20.2.2 Level-2 investigations ............................................................................... 388
20.2.3 Level-3 investigation................................................................................. 391
20.3 Conclusion.............................................................................................................. 395
20.4 References ............................................................................................................. 395
21 Non-Linear Stability Analysis of ARIANE 5 Front Skirt (JAVE) ....................396
13
ECSSEHB3224A
24March2010
21.1 Overview ................................................................................................................ 396
21.2 Description of the Structure .................................................................................... 397
21.3 Mathematical Model ............................................................................................... 400
21.3.1 Launcher Coordinate System and Sign Convention................................. 400
21.3.2 Geometry.................................................................................................. 400
21.3.3 Material Definition..................................................................................... 401
21.3.4 Loading..................................................................................................... 402
21.3.5 Geometric Boundary Conditions............................................................... 404
21.3.6 Imperfections ............................................................................................ 404
21.4 Finite Element Analysis .......................................................................................... 405
21.4.1 Solution Method........................................................................................ 405
21.4.2 Amendments to Extraction of Eigenvalues............................................... 406
21.4.3 Discretisation ............................................................................................ 406
21.5 Results Nonlinear Analyses of Load Case A151................................................. 407
21.5.1 Introduction and Chronology .................................................................... 407
21.5.2 Jobname A151_NEWTON_Evol20a_Imp .............................................. 408
21.5.3 Generation of Final Imperfect Reference Configuration ........................... 415
21.6 Summary and Conclusion ...................................................................................... 424
21.6.1 Summary .................................................................................................. 424
21.6.2 Conclusion................................................................................................ 424
22 Buckling Analysis and Qualification Static Load Testing of VEGA

Interstage 1/2 Structure ....................................................................................426
22.1 Overview ................................................................................................................ 426
22.2 Interstage 1/2 Assembly......................................................................................... 427
22.3 Imperfection Sensitivity Analyses ........................................................................... 428
22.3.1 Classical Buckling Load............................................................................ 428
22.3.2 Asymptotic Post Buckling Theory ............................................................. 429
22.3.3 Analyses Applying SRA Programs ........................................................... 431
22.4 Finite Element Analysis .......................................................................................... 433
22.4.1 Structural Mathematical Model ................................................................. 433
22.4.2 Buckling analyses of flight configuration................................................... 434
22.5 Test Campaign ....................................................................................................... 437
22.5.1 Overview................................................................................................... 437
22.5.2 Test Jig ..................................................................................................... 437
22.5.3 Test Correction Factors ............................................................................ 437
22.5.4 FE Analyses of Test Set-up...................................................................... 438
14
ECSSEHB3224A
24March2010
22.5.5 Static Test results ..................................................................................... 439
22.5.6 Correlation between test and FE analyses............................................... 440
22.6 Conclusions............................................................................................................ 440
22.7 References ............................................................................................................. 441
22.8 Abbreviated Terms and Symbols ........................................................................... 442
23 Stability analysis of the 3rd-stage skirts of the ELDO-A...............................443

23.1 Overview ................................................................................................................ 443
23.2 Structural description and design aspects.............................................................. 443
23.3 Input data ............................................................................................................... 445
23.3.1 Material..................................................................................................... 445
23.3.2 Loadings and dimensioning load cases.................................................... 447
23.3.3 Analysis of material behaviour.................................................................. 449
23.4 Stability failure modes and load-carrying capacity of the shell ............................... 452
23.5 Local failure of stringer profiles under compression elastic local buckling and
elastic-plastic crippling ........................................................................................... 453
23.5.1 AISI steel .................................................................................................. 453
23.5.2 Titanium.................................................................................................... 454
23.6 Panel instability: Elasto-plastic buckling of the stiffened shell between the ring
frames .................................................................................................................... 455
23.6.1 Assumptions: ............................................................................................ 455
23.6.2 Application to AISI steel corrugation:........................................................ 455
23.6.3 Application to titanium corrugation: .......................................................... 457
23.7 General instability of the stiffened shell .................................................................. 457
23.7.1 Assumptions ............................................................................................. 457
23.7.2 Methods.................................................................................................... 457
23.7.3 Van der Neut method ............................................................................... 458
23.7.4 Numerical analysis of the steel corrugation shell ..................................... 459
23.8 Structural testing .................................................................................................... 460
23.8.1 Sample tests with the corrugated skin...................................................... 460
23.8.2 Static structural tests with the corrugated orthotropic shell ...................... 460
23.9 General conclusions............................................................................................... 462
23.10 References ............................................................................................................. 463
15
ECSSEHB3224A
24March2010
Acknowledgements
ThisHandbookhasbeenauthoredandagreeduponbythefollowingpersons:
J.Arbocz,TUDelft
C.Bisagni,PolitecnicodiMilano
A.Calvi,ESAESTEC(Convenor)
E.Carrera,PolitecnicodiTorino
R.Cuntze,formerlyMANTechnologie
R.Degenhardt,DLRBraunschweigandPFHGttingen
N.Gualtieri,ThalesAleniaSpace
H.Haller,Intales
N.Impollonia,UniversitdiCatania
M.Jacquesson,CNES
E.Jansen,TUDelft
H.R.MeyerPiening,ETHZuerich
H.Oery,RWTHAachen
A.Rittweger,AstriumEADS
R.Rolfes,LeibnizUniversitaetHannover
G.Schullerer,MTAerospace
G.Turzo,CNES
T.Weller,Technion,Haifa
J.Wijker,DutchSpace
Thevaluablecontributionsofthefollowingpersonsarealsoacknowledged:
C. Huehne, DLR Braunschweig; D. Petry, Astrium EADS; H. G. Reimerdes, RWTH Aachen; K.
Rohwer,DLRBraunschweig;R.Zimmermann,DLRBraunschweig.
The ECSSEHB3224 has been prepared by merging the volunteer contributions of the authors.
Comments concerning the technical content of this handbook will be welcomed by the European
CooperationforSpaceStandardization,Noordwijk,theNetherlands,www.ecss.nl.
16
ECSSEHB3224A
24March2010
Introduction
Thehandbookisorganisedinthreeparts:
Part1:General.
Itincludeschapters4to13,andprovidesgeneralinformationonthestructuralstability.
Chapter 4 addresses the history of stability problems starting from Euler (1757), who
investigated buckling of beams as well as of plates. More recently, during the past 50
years, the discrepancy between experiments and linear buckling analysis as well as the
large scatter of the tests stimulated a large amount of research work. The focus was
especially on the influence of boundary conditions, of the nonlinear prebuckling
behaviour, of the postbuckling behaviour of the perfect shell and of the buckling and
postbucklingbehaviouroftheimperfectshell.
In chapter 5 the occurrence of different types of instabilities are discussed on hand of
relatively simple examples. The following topics are addressed: elastic buckling of
columns; buckling of thin plates; instability of axially compressed cylindrical panels;
structuralbehaviourofthinshells.
Thefirstscopeofchapter6istogiveasurveyofallthetypesofanalysiswhichcouldbe
used for the study of stability. A simple descriptive presentation is provided to fix a
commonlanguageandaclearunderstanding.Forrigorousderivationandcomprehensive
details,properreferenceisindicated.Asecondscopeistocategorizethephenomenaofloss
ofstability,byindicatingforeachtypetheappropriateanalysistobeperformed.
Chapter7addressesthefollowingtopics:linearelasticityandelasticproperties;strength
properties and hygrothermal properties; elastic and inelastic material behaviour;
plasticityanddamage;materialtestingmethods,testdata,andevaluation.
Chapter8describesthedesigntostabilityanditsverificationbyanalysis.Thefollowing
topics are included: the design development process; analysis prework and load input
data; safety concept and factors of safety; dimensioning load cases selection; stability
design allowable and knockdownfactors; analyses, modelling, and design aspects;
procedureofdesignverification,includingthedeterminationofthemarginofsafety.
Chapter 9 describes the influence of manufacturing and assembly processes on the
bucklingload.
Inchapter10theuseofthefiniteelementmethodandthenumericalproceduresapplied
to perform a buckling or postbuckling analysis are described with a focus on space
applications.Aguidetoselectthemostadequatesolutionschemeisprovided.
This handbook promotes the Hierarchical High Fidelity Analysis approach, which is
reportedinchapter11.Inpractice,inordertoarriveatareliablepredictionofthecritical
bucklingloadandtomakeanestimateofitsimperfectionsensitivitywhichcanbeused
with confidence, the structural analyst should proceed step by step from simple
analyticalsolutions(Level1)tomorecomplexmodelsandsolutionprocedures(Level3).
17
ECSSEHB3224A
24March2010
This chapter provides a detailed description of the hierarchical high fidelity analysis
approachincludingaflowchartestablishingastrategytohandlebucklingphenomena.
Chapter12isdedicatedtothebucklingexperimentalmethodsanddesignverificationby
tests.Experimentscanexplorethephysicalbehaviournearbuckling,atbucklingandin
thepostbucklingrangeandtheyalsoyieldempiricaldatauponwhichdesignguidelines
canbebased.Therealizationofthetestscanbedoneatseverallevels:testsonelementary
sample,testsonstructuralelementsandtestonfullscalestructure.
Thetestinstrumentationisdescribedinchapter13.Itdealswithmeasurementtechniques
instructuraltestingandmeasurementsystems.
Part2:Structuralelementswithexamples
Itincludeschapters14to17,anditisdevotedtostructuralelementswithexamples.Itreports
the mathematical formulae for calculating the buckling loads for the most common structural
elements,loadsandboundaryconditions.
Chapter 14 is an introductory chapter covering the following topics: static versus
kinematic approach; problems requiring nonlinear analysis; approximate solutions of
bifurcationproblems;computationaltoolsforbifurcationproblems.
The remaining chapters 15, 16 and 17, present the solutions with examples for the
followingstructuralelements:columns,beams,arches,andrings;flatandcurvedpanels;
closedshells.
Part3:Structures
Itincludeschapters18to23,andaddressestherealworldstructures.
Chapter18isanintroductorychapterdescribingthedesign,analysisandtestingaspects
of large aerospace structures, including the approach for the mathematical model
correlation and validation. Relevant flowcharts and examples are provided, mainly
relatedtothestructureoftheEuropeanlauncherARIANE5.
Theremainingchapters19to23reportvariouslargeexamplesprovidedbyvariousEuropean
aerospacecompanies.
Thebucklinganalysisandtestactivitiesperformedontheliquidhydrogen(LH2)tankof
theESCAARIANE5stagearepresentedinchapter19.
Chapter 20 describes the Inner Dome of the ARIANE 5 Upper Stage ESCA. This
structureissubjectedtocompressiveloadsincaseofinternalpressure.Forthisreasonthe
InnerDomeshellhasbeendesignedcarefullyagainstbuckling.
The results of the nonlinear stability analysis performed on the ARIANE5 Front Skirt
(JAVE)anditsadjacentstructuresaresummarizedinchapter21.Thechapterpresentsthe
main characteristics of the mathematical model, including the definition of material
models,loadings,geometricandphysicalimperfectionsandanalysistechnique.
Chapter 22 presents the results of the buckling analyses and static strength tests of the
Interstage1/2fortheVegalaunchvehicle.TheInterstage1/2isconstructedasmonocoque
andisbucklingcritical,hencetheneedtoobtainaccuratepredictionsforbucklingstrength,
includingtheinfluenceofinitialimperfections.TheInterstage1/2wasqualifiedbyastatic
loadtest,andthenwasfurtherloadedtofinalfailureinarupturetest.
Finally,chapter23reportstheresultsofthestabilityanalysisofthe3rdstageskirtsofthe
EuropeanlauncherELDOA.
18
ECSSEHB3224A
24March2010
1
Scope
TheECSSEHB3224recommendsengineeringpracticesforEuropeanprogramsandprojects.Itmay
be cited in contracts and program documents as a reference for guidance to meet specific
program/projectneedsandconstraints.
The target users of this handbook are engineers involved in design, analysis and verification of
launchers and spacecraft in relation to structural stability issues. The current knowhow is
documentedinthishandbookinordertomakethisexpertiseavailabletoallEuropeandevelopersof
spacesystems.
Itisaguidelinesdocument;thereforeitincludesadvisoryinformationratherthanrequirements.
2
References
Duetothestructureofthedocument,eachchapterincludesatitsendthereferencescalledinit.
19
ECSSEHB3224A
24March2010
3
Terms, definitions and abbreviated terms
3.1 Terms from other documents

Forthepurposeofthisdocument,thetermsanddefinitionsfromECSSSST0001apply.
Forthepurposeofthisdocument,thefollowingtermsanddefinitionsfromECSSEST32apply:
buckling
designallowable
designfactor
designlimitload(DLL)
designload(DL)
designparameters
designultimateload(DUL)
designultimatestress
designyieldload(DYL)
designyieldstress
detrimentaldeformation
factorofsafety(FOS)
failure
limitload(LL)
relievingloads
residualstress
stiffness
structuraldesign
structure
ultimatestrength
yieldstrength
Forthepurposeofthisdocument,thefollowingtermsanddefinitionsfromECSSEST3210apply:
localdesignfactor(KLD)
marginpolicyfactor(KMP)
modelfactor(KM)
projectfactor(KP)
testfactors(KAandKQ)
ultimatedesignfactorofsafety(FOSU)
yielddesignfactorofsafety(FOSY)
20
ECSSEHB3224A
24March2010
3.2 Terms specific to the present document

3.2.1 buckling analysis
methodtostudytheequilibriumstabilityofastructure,basedoneigenvalueproblemanalysis
NOTE The most general approach for the analysis of equilibrium stability
requiresthestudyofnonlinearequilibriumandstabilityequations;the
bucklinganalysisapproachisusuallybasedonspecificassumptionsand
formulationwhichleadtosolveeigenvalueproblems.
3.2.2 buckling mode or shape (with reference to buckling)

changeddeformedconfigurationofastructure,duetooccurrenceofbuckling
NOTE1 Buckling mode and shape is defined in shape and amplitude, depending
fromloadlevel.
NOTE2 Buckling mode can be a new equilibrium or the failed structure
configuration
3.2.3 buckling modes or shapes (with reference to buckling analysis)

eigenvectorsassociatedtotheeigenvaluesofabucklinganalysis
NOTE Bucklingmodesandshapesaredefinedonlyinshape,notinamplitude.
3.2.4 buckling resistance

limitstatecausedbylossofstabilityundercompressiveandshearmembraneloads
3.2.5 design buckling load

bucklingloadmultipliedbyadesignfactor
3.2.6 global buckling

bucklingwhosebucklingshapeisaffectingquiteallthestructure
NOTE Usuallytheglobalbucklingcausesstructurecollapse.
3.2.7 local buckling

bucklingwhosebucklingshapeisaffectingonlyaminorpartofthestructure
NOTE1 Examples of local buckling are intracel buckling of sandwich, inter rivet
buckling,andbeamlegbuckling
NOTE2 Usually the local buckling causes load path changes and not structure
collapse.
3.2.8 margin of safety (MOS)

ratioofthebucklingresistanceandthedesignbucklingloadminus1
21
ECSSEHB3224A
24March2010
3.3 Abbreviated terms

Thefollowingabbreviatedtermsaredefinedandusedwithinthishandbook:
Abbreviation Meaning
AIT assembly,integrationandtests
CAD computeraideddesign
CAE computeraidedengineering
CAM computeraidedmanufacturing
COG centreofgravity
DL designload
DLL designlimitload
DOF degreeoffreedom
DRD documentrequirementdefinition
DUL designultimateload
DYL designyieldload
FCI fracturecriticalitem
FEA finiteelementanalysis
FE finiteelement
FM flightmodel
FMECA failuremode,effectsandcriticalityanalysis
FOS factor(s)ofsafety
FOSU factor(s)ofsafetyatultimate
FOSY factor(s)ofsafetyatyield
LL limitload
KDF knockdownfactor
MOS marginofsafety
NDT nondestructivetest
NDI nondestructiveinspection
PFCI potentialfracturecriticalitem
NOTE Some additional abbreviated terms which are used in a specific chapter
aredefinedattheendoftherelevantchapter.
22
ECSSEHB3224A
24March2010
Part 1: General
23
ECSSEHB3224A
24March2010
4
Overview
4.1 Instability of structures: concepts and definitions

The stability is a property of the equilibrium configurations of a given structure, subjected to static
and/ordynamicloads.
The equilibrium is said stable if small pertubations do not cause significant, unexpected and
unwished changes of equilibrium configuration, otherwise the equilibrium is said unstable (it can
changeabruptly).
Although the above sentences are qualitative, nevertheless they clearly describe the experimental
evidence and they can highlight many key aspects related to the concept and study of stability of
structures.
Threefactsdeservespecialattention.
Firstofall,thepossibilityoffindingadditionalequilibriumconfigurationsisstrictlyrelatedto
the nonlinear nature of the stability problem. In the case of linear systems, in fact, only one
equilibrium configuration exists, which can be changed only by linear relationships of
configurationparameters(e.g.loadsanddisplacements).
Secondly,tostudythestabilityoftheequilibrium,itisnecessarynotonlytofindanequilibrium
configuration,butalsotostudypossiblechangesoftheseconfigurations,andstabilityequations
needtobeformulatedforthispurpose.Thesestabilityequationsarealsoknownasincremental
equilibriumequations.
Finally,thestudyofchangedconfigurationsmayormaynotbeperformedinthetimedomain,
i.e.stabilitymaybeinvestigateddynamicallyorstatically.Therearecaseswhereonlydynamic
approach is appropriate (e.g. nonconservative systems); however the static approach is
adequate for many stability problems of aerospace structures. Nevertheless, the most general
methodreliesonLiapunovsdynamicstabilitydefinition.
Variousmethodsandtechniques,whichhavebeenproposedandarenecessarytoproperlystudythe
stability of the equilibrium state, are discussed in detail in subsequent chapters (see in particular
Chapter6:Analysismethods).Theseconceptscanbesubdividedbythethreecriteriafordefiningor
detectinginstabilityofstructures,whichdifferincomplexityandgenerality[1]:
a. Thecriterionofnontrivialequilibriumstate(seeChapter6.5)
b. Thedynamicalcriterion(seeChapter6.8)
c. Thetotalpotentialenergycriterion(seeChapter6.3)
Forconservativeelasticsystemsallthreecriteriagiveidenticalresults.
Usingthecriterionofnontrivialequilibriumstatemeanstoaskifthereisanadjacentconfiguration
whereequilibriumisachieved.Thisisthecase,ifthestabilitydeterminantofasystemisequaltozero
foranontrivialsolution.Thestabilitydeterminantcanbeobtainedfromtheequilibriumconditionsor
bythesecondderivativeoftheenergyfunction.Themethoddoesnotworkforproblemsofstability
wherethereisnoadjacentequilibriumstateandfordynamicalbucklingingeneral.
24
ECSSEHB3224A
24March2010
Thedynamicalcriterionisthemostgeneralapproach.Theideaistodetermineifasystemreturnsto
the initial configuration after applying a deflection or if the deflections increase or converge to a
differentconfiguration.Mathematically,thiscanbedescribedbythefrequencydeterminantandthe
socalledBurchartdeterminant,whichbothhavetobeequaltozero.
The most widely used criterion is the total potential energy criterion, which uses the fact that for
stabilitytheenergyfunctionmustbeaminimumandforinstabilityitmustbeamaximum.Thus,the
critical state is reached, when the second derivative of the energy function is equal to zero. The
disadvantageofthiscriterionisthatitisrestrictedtoconservativesystems.
Alltheseconceptsarenowwellestablishedintodaysmechanicsofstructures,butanenormouseffort
has been dedicated to understand and describe the phenomenon. The history of stability problems
dates back to Euler (1757), who investigated buckling of beams as well as plates. Prominent
researchers like Bernoulli, Lagrange, Navier and Kirchoff further developed the theory of plate
buckling. The historical overview in this chapter (and the bibliography) gives an idea on this
development,byfocussingeffortsanddifferentconcepts.
4.2 Stability of columns, beams and plates
4.2.1 Buckling of beams

Euler[2]wasthefirstonewhosolvedthebucklingproblemofacantileverbeamin1744leadingtohis
famousformulas.Thecorrespondingeigenvalueproblemcanbesolvedeasilyfordifferentboundary
conditions(seee.g.[3],[4]).WhileEulerregardedtheperfectcolumn,Rivello[5]consideredaninitial
geometricimperfectionandconcludedthatforcolumnswithmeasurableimperfectionsthebuckling
load of the perfect column is sufficiently accurate. Further, Rivello determined the yielding stress
limits considering linearly elastic and idealized plastic material behaviour. Simitses [6] showed that
Rivelloscurvesin[5]canalsobeobtainedifthegeometricimperfectionisreplacedbyaneccentricity
oftheappliedload.
Experimentalevidenceindicatesthatforaslendernessration L A I > 80 Eulersformulasarequite

accurate, but for 80 > L A I > 20 to get reasonably accurate predictions of the buckling load the
use of Shanleys tangent modulus theory [7] is recommended. For smaller slenderness ratios failure
occursmainlybyplasticcrushing(seechapter4.9).
4.2.2 Lateral torsional buckling

Forbeamsthatarebentwithrespecttotheirstrongaxis,furthercasesofinstabilitycanoccur,ifthe
torsional rigidity IT and the moment of inertia Izz are relatively small compared to the moment of
inertiaIyyofthestrongaxis.Then,lateralbucklingcanoccur,whichmeansthatthebeamtwistsand
deflects out of the loading plane. The special case, that the beam is strained only with bending
momentswasfirstsolvedbyPrandtl[8]andalmostatthesametimebyMichell[9]in1899.Further
solutions were given by Timoshenko [10], Nylander [11] and Kindem [12]. Chawalla [13] showed
somepossibilitiesforpracticalapplicationsofthesesolutions.
The more general problem of lateral torsional buckling under bending and compression was first
mentioned by Wagner [14] and solved by Kappus [15],[16] in 1937. Thus, design rules for beams
which tend to lateral buckling could be adopted in the German guideline DIN 4114 [17] in 1952. A
multitudeofsolutionsofthelateralbucklingloadfordifferentloadcasesandboundaryconditionsare
listedbyPetersen[18].
25
ECSSEHB3224A
24March2010
4.2.3 Buckling of plates
Ifoneregardsacolumnasanassemblageofthinplates,thequestionoflocalstabilityoftheassembly
arises.Thisleadstothestabilityofinplaneloadedthinwalledshellsandplates.
FirstexperimentaltestsonplateswereexecutedbyFairbairnandHodgkinsonforthebridgeprojects
of Stephenson in 1845, the Britannia and Conway Tubular Bridges [19],[20]. The first one who
determined the buckling load of a simply supported plate analytically was Bryan [21] in 1891. He
derived the critical stress, which depends on the buckling coefficient kc. Gerhard and Becker [22]
calculatedbucklingcoefficientsforseveralcombinationsofboundaryconditionsandloadcases.Coan
[23] studied rectangular plates under uniformly displaced loaded edges including a halfwave sine
initial imperfection and using a large deflection theory. He found that the axial stress of the
compressedplateisuniformlydistributedoverthecrosssection,butafterbucklinghasoccurred,the
stresses increase almost only close to the edges of the plate whereas in the middle of the plate the
stressesmaintainalmostaconstantmagnitudeequaltothebucklingstress,iftheplateiscompressed
further.Ifoneassumesthatthetotalloadiscarriedbythestripesattheedges,onecandeterminethe
effective width. This concept was already described by Schuman and Back [24], then defined and
theoretically derived by von Krmn et al [25] and later improved by Sechler, Cox, Maguerre and
others[26][30]withexperimentsandtheory.Theconceptoftheeffectivewidth,whichwasproposed
by aeronautic engineers, was adapted to civil and marine engineering. Several theoretical and
experimentalstudieswereaddedfromallbranchesoverthecentury(seee.g.[31][36]).
Because for plates it is possible to increase the sustained load after buckling, the post buckling
behaviour plays an important role, if one wants to exploit this load carrying reserve. Increasing the
load after exceeding the buckling load leads to a change of the buckled form. There, one can
distinguishbetweenminorandmajorchangesinbuckledform.Aminorchangeisagradualsmooth
changeliketheflatteningofthetransversehalfwave,whereasamajorchangeconsistsofanabrupt
snapfromonebuckledformtoanother,liketheincreaseinthenumberofwavesalongthedirectionof
compressioninarectangularplate.Themajorchangesinpostbucklingpatternwerealreadynotedby
Sechler[29],[30]andStein[37],[38]whostudiedthechangeinwavelengthwiththeaidofasimplified
mathematicalmodel.Thisphenomenon,whichisoftencalledsecondarybucklingintheliterature,
wassubjectoffurtherinvestigationsonplatebuckling(seee.g.[39]).
4.3 History of shell buckling
4.3.1 Overview
For solving the buckling problem of circular cylindrical shells first analytical approaches were
publishedin1908byLorenz[40],in1910byTimoshenko[41]andin1914bySouthwell[42].Improved
methods were developed in the 1930ies, e.g. the wellknown checkerboard approach of Flgge [43].
Batdorfin1947[44]presentedthiscurvatureparameter Z = L / Rt ( 2
) 2
1 v ,whichmadeitpossibleto
combinethecylinderdimensionsandthematerialpropertiesinsuchawaythattheresultsofaseries
ofbucklingloadcalculationscouldbepresentedbyasinglecurve.Allpapersmentionedarefocussing
onlinearstabilityanalysismethodsforthebifurcationloadofperfectcylinders.
First tests of thinwalled elastic cylindrical shells were performed in 1933 by Lundquist [45] and in
1934 by Donnell [46]. During the 1950ies and 1960ies a large number of tests followed. The axially
compressed cylindrical shell represents one of the best known examples of the very complicated
stabilitybehaviourwhichcanoccurwiththinwalledconstructions.Thenatureoftheproblemiswell
illustrated in Figure 41 [47], where some of the experimental results for isotropic shells have been
plottedasafunctionofthe R / t (radiusoverwallthickness)ratio.AscanbeseeninFigure41,the
26
ECSSEHB3224A
24March2010
testsrevealawidescatterintheexperimentalresults,withexperimentalbucklingloadsforverythin
shells( R / t > 1000 ,say)aslowas20%ofthetheoreticalvalues.
Intheabsenceofabettersolution,forpracticalapplicationsallcurrentshelldesignmanualsrelyon
the socalled Lower Bound Design Philosophy that has been in use for over 50 years. These
referencesrecommendtheuseofanempiricalknockdownfactorwhichischosensuchthatwhenit
ismultipliedbytheperfectshellbucklingload,alowerboundtotheappropriatepartoftheexisting
testdataisobtained.ThelowerboundcurveshowninFigure41wasproposedin1965forisotropic
shellsbyWeingartenetal[47]as
P R
= 1 0.901(1 e ) for < 1500 41
Pcl t
1 R Et 2
where = , Pcl = 2 RN cl , N cl = , c = 3 (1 v ) .
2
16 t cR
Introducingtheempiricalknockdownfactor as
= 1 0.901(1 e ) 42
oneobtainsthewellknowncriticalbucklingloadformulaforanaxiallycompressedisotropicshell
Pc = Pcl 43
Figure41:Testresultsofaxiallyloadedcylinderswithdifferentslendernessratio
(from[47])
Thediscrepancybetweenexperimentsandlinearbucklinganalysisaswellasthelargescatterofthe
testsstimulatedalargeamountofresearchworkduringthepast50years.Thefocuswasespeciallyon
the influence of boundary conditions, of the nonlinear prebuckling behaviour, of the postbuckling
behaviouroftheperfectshellandofthebucklingandpostbucklingbehaviouroftheimperfectshell.
27
ECSSEHB3224A
24March2010
4.3.2 Boundary conditions and nonlinear pre-buckling behaviour
Initially,itwastriedtoexplainthediscrepancybydifferentboundaryconditionsorbythenonlinear
prebuckling state which is not covered by linear analyses. Historically, Ohira [48] and Hoff [49] in
1961werethefirsttopublishpaperswhereforsimplysupportedboundaryconditions(w=Mx=0,
Nx=N0)andusingNxy=0insteadofv=0theyobtainedsolutionsofthelinearizedstabilityequations,
whichwereabout50percentlowerthantheclassicalsolutionspublishedbyLorenz,Timoshenkoand
Southwell.
ForfreeboundariesNachbarandHoff[50]in1962publishedapaperwheretheyobtainedacritical
buckling load of 37 percent of the classical solution. As stated all theses solutions were based on
assumingamembraneprebucklingstate.HoffandSoong[51]in1965publishedtheresultsofavery
extensive investigation based on membrane prebuckling for eight possible combinations of edge
supportconditions.InthesameyearAlmroth[52]publishedtheresults ofasimilarinvestigationfor
the same eight combinations of edge support but he used a rigorous solution of the prebuckling
problem.Hisresultswereinmostcases1015percentlower.
Thus,theconclusionfromtheseresultsisthatboundaryconditionscanhaveasignificantinfluenceon
thebucklingloadobtained,buttheycannotaccountforthelargediscrepanciesbetweenthetheoretical
predictionsandtheavailableexperimentalresultsbythemselves.
4.3.3 Postbuckling of the perfect shell

Thedevelopmentofnonlinearbucklingtheorieswasanimportantprogresstowardsinterpretationof
thetestresults.Thebasicequationswereformulatedin1934byDonnell[46]andwerenotrestrictedto
small deformations. The quadratic terms of the radial displacement w were taken into account. For
solving the nonlinear equations the energy method was used and approximations for w in terms of
trigonometric series were made. Based on these equations von Krmn and Tsien performed first
postbucklinganalyses ofaxiallyloaded cylindersalreadyin 1941[53]. The solution of thenonlinear
DonnelltypeshellequationsbasedonaRitztechniqueshowedforthefirsttimethatforcertainvalues
of the loading parameter (= P/Pcl) there exist three possible equilibrium configurations [53](see
points GAC in Figure 410a). The form of the assumed radial displacement w was based on
experimentalevidenceconsistingofthreeparts
w ( x, y ) = f 0 + f1w1 ( x, y ) + f 2 w2 ( x, y ) 44
where f0 is introduced in order to allow the shell to expand radially, w1 represents the diamond
shapedstablepostbucklingpatternandw2theasymmetricchessboardlikebifurcationbucklingmode
oftheclassicaltheory.
LeggettandJones[54],Michielsen[55],Kempner[56]andAlmroth[57]calculatedtheloadshortening
curve withan increasingnumber of coefficients. The potential energy was minimized notonly with
respecttothesecoefficientsbutalsowithrespecttothe l x and l y ,thelengthsofthebucklesinaxial
andcircumferentialdirection,respectively.Thebucklingloaddecreasedfrom34percentofthelinear
bucklingload(accordingtovonKrmnandTsien)toonly10,8percent(accordingtoAlmroth)(see
Figure 42). An overview of the results was given by Hoff [58]. The search for improved solutions
came to an end in 1965 by the work of Hoff, Madsen und Mayers [59]. They revealed that the
postbucklingloadtendstowardszerowhenthenumberofcoefficientsisincreasedmoreandmore.
ThielemannandEsslingerconfirmedbytheoreticalandexperimentalinvestigations,thatthebuckling
loaddecreaseswithincreasinglengthofthecylinder[60].Fortheinfinitelylongcylindersinvestigated
so far, the length l x of the buckles in axial direction could become even larger than the cylinder
length due to the missingboundary conditions.Realistic postbuckling loads couldonly beachieved
by taking into account the finite length of the cylinder. Moreover, so far only regular postbuckling
28
ECSSEHB3224A
24March2010
patterns were assumed, that spread over the whole cylinder surface. Test of cylinders with finite
lengthrevealedalocallyconcentratedpattern,usuallywithtworowsofbucklesalongthelength.(see
Figure 43). In 1955 Yoshimura [61] confirmed with simple approximate calculations that a finite
cylinder tends to buckle with a concentrated pattern. The deformation mode with concentrated
buckles has a lower energy content than the regular, socalled von Krmn postbuckling pattern.
Esslinger[62]developedatheoryfordescribingtheconcentratedpatternbytakingintoaccountthe
boundary conditions. The postbuckling curve with minimum shortening was assumed to be the
appropriateone.
Investigationsonaxiallyloadedcylindricalshellsshowedthatinstableequilibriummodesexistinthe
postbuckling regime showing small radial displacements and loads below the buckling load.
Thereforetheshellsweresensitivetoperturbations.Thelargescatterofresultscanalsobetracedback
to this phenomenon. The structural behaviour was explained by Esslinger [63] using the idea of
perturbations and restoring forces. According to that in a axially loaded beam only the bending in
axial direction creates back tracing forces, whereas in a plate an additional force is present due to
bendingintransversedirection.Incylindricalshellsthecircumferentialmembraneforcegeneratesan
additionalbacktracingforceandthusleadstoveryhighbucklingstresses.However,thesupportby
the membrane forces is strongly influenced by small manufacturing imperfections. The support is
very much weakened in the postbuckling regime and finally vanishes fully, thus leading to a load
carryingcapabilitybelowthetheoreticalbucklingload.
Figure42:Calculatedpostbuckling Figure43:Postbucklingpatternofaxially
curve(from[117]) loadedmylarcylinder(from[60])
In the theoretical work either load or displacement was applied to the cylinder, whereas the
experimentsweredisplacementdriveninordertonotdamagethestructure.Whenthebucklingstrain
was reached the load decreased momentarily at constant strain. The cylinder achieved a new stable
equilibrium mode at a load well below the buckling load. The highly dynamic transition from the
stableprebucklingtoanotherstablepostbucklingmodewasvisualizedbyAlmrothetal.in1964[64]
andEsslingerin1970[65]usingahighspeedcamera.Startingfromaninitialsinglebucklethebuckles
spread over the whole surface of the cylinder. Several unstable deformation patterns were passed
beforethestablepostbucklingmode,asdepictedinFigure43,wasreached.
29
ECSSEHB3224A
24March2010
Figure44:Postbucklingcurves([69])
4.3.4 Buckling and postbuckling behaviour of imperfect shells

Koiter was the first, who in 1945 recognized the close relation between low buckling loads and
geometricalimperfections[66].Geometricalimperfectionsweredefinedasdeviationsfromtheperfect
cylindrical geometry. The thesis, written in Dutch language, was not recognized by the scientific
communityuntilRikstranslatedittoEnglish[67].Independently,andbasedon[46]and[68],Donnell
and Wan in 1950 [69] investigated the load shortening behaviour of imperfect shells up to the deep
postbuckling regime. They detected that already at small imperfection amplitudes w0 the actual
buckling load can be well below the linear one. (see Figure 44). Donnell and Wan assumed an
imperfection pattern affine to von Krmns postbuckling pattern. The work was thoroughly
evaluated by Arbocz [70]. Within the deep postbuckling range the postbuckling paths of imperfect
and perfect shells approached each other. The subsequent table shows publications which use
differentboundaryconditionsandimperfectionpatterns.
In contrast to Donnell and Wan, Koiter did not focus on the deep postbuckling regime but on the
initial postbuckling behaviour. A comprehensive presentation of Koiters theory was given by
BudianskyandHutchinson[77]andbyBudiansky[78],[79].Verydetaileddescriptionsofthetheory
were provided by Seide [80] and Masur [81], short overviews are given by Almroth and Brush [82]
andbyTvergaard[83].Achronologicaloverviewofpapersusingtheasymptoticapproachisgivenin
Table42.
30
ECSSEHB3224A
24March2010
Table41:Chronologicaloverviewofpapersprovidingfullloadshorteningcurves
forimperfectcylindricalshells(from[71])
Authors Literature Year Imperfectionpattern Amplitude

Donnell,Wan [69] 1950 Affinetopostbucklingpattern variable
Pflger [72] 1962 Affinetocriticaleigenmode R/200
Arboczetal. [73],[74] 1969/74 Measuredimperfectionpattern measured
Affinetomodeofcharacteristic
Esslinger,Geier [75] 1972 variabel
postbucklingcurve
Yamaki [76] 197284 Differentmeridianshapes variabel
Table42:Chronologicaloverviewofpapersdescribingtheasymptotic
postbucklingbehaviour(from[78]
Authors Literature Year Imperfectionpattern Amplitude

Koiter [66],[84] 1945 Eigenmodeaffineimperfections varying
Koiter [85] 1963 Axisymmetricimperfections varying
Tennyson,
[86] 1969 Localaxisymmetricimperfection varying
Muggeridge
Yamaki [76] 1984 Affinetocharacteristicpostbucklingshape varying
Jrcke [87] 1985 Affinetoperiodiceigenmode varying
Figure45:Equilibriumpathsforsymmetricpostbucklingbehaviour
KOITERintroducedforhisasymptoticapproachaseriesexpansionnearthelowest(critical)eigenvalue
=cofthestructure.Theinitialpostbucklingpathisdescribedusingtheloadparameterandthe
normalizedamplitudeofthefirstbucklingmodeby
31
ECSSEHB3224A
24March2010
/ c = 1 + a + b 2 +K 45
whereaandbarecalledthefirstandsecondpostbucklingcoefficient,respectively.Forthesymmetric
casea=0follows,whereasthecoefficientb,thesocalledbfactor,determinesthestructuralbehaviour
in the initial postbuckling regime. In case of b>0 the load can be further increased after having
reached the branching load. Thus the structure is insensitive to imperfections. For the unstiffened
cylindricalshellsb<0applies,whichmeansthatthestructurecollapseswhenreachingtheloadcand
instable equilibrium modes exist at lower load. (see Figure 45). The asymptotic approach has the
attractivefeaturethatthepostbucklingcoefficientsaandbonlydependontheperfectshell.Without
knowing shape and amplitude of the imperfections a qualitative prediction can be made of the
imperfectionsensitivityofthestructureintheinitialpostbucklingregime.
4.4 Design load for thin-walled isotropic shells
4.4.1 Overview
In the design process theinfluence of,usually unknown, imperfections has tobe taken into account
properlysothatasafestructureisconstructed.TypicalguidelineslikeNASASP8007[88]orDASt
Richtlinie 017 [89] recommend very conservative knockdown factors, however, they do not require
knowledge about pattern or even amplitude of the imperfections. Measuring imperfections and
introducing them into the numerical analysis is very costly, but promises less conservative design
loads. A definition would be highly desirable of a physics based design load, which quantitatively
describesimperfectionsensitivitywithoutrequiringknowledgeoftherealimperfections.
4.4.2 Empirical knock-down factors

Empirical methods were used for designing axially loaded cylindrical shells, since adequate
theoreticalanalysismethodsweremissing,whichcouldcovertheinfluenceofrealisticimperfections.
The first attempt to describe the loading capacity by evaluating existing test results using statistical
methodswasmadein1957byHarrisetal.[90].Almrothetal.[91]determinedtheknockdownfactor
as a function of the radius to wallthickness ratio for different safety levels P (see Figure 46). A
disadvantage of these statistical methods is the data basis, i.e. also those test results influence the
design,whicharenotrelevantforthestructureunderconsiderationduetoitsspecificmanufacturing
procedureoritsverydifferentsize[91].
Very conservative guidelines were established based on such statistical investigations. In aerospace
engineering NASA SP8007 (see [92]) is still used for designing axially loaded cylinders, the knock
down factors are depicted in Figure 41. The guideline DAStRichtlinie 017 is applicable to steel
structuresinallengineeringapplications.Theknockdownfactorsinbothguidelinesarebasedontest
results[94],thestatisticalevaluationofwhichisnotspecifiedindetail[47].Grimmeltetal[93]drew
theconclusion,thattheguidelinesareatleastpartiallybasedonthesametestresults,andtheknock
downfactorsofNASASP8007are90%fractiles,theonesinDAStRichtlinie017are99,9fractiles.A
summaryof guidelinesusedbyNASAfordifferentshellgeometriesisgivenbyNemethandStarnes
[92].
32
ECSSEHB3224A
24March2010
Figure46:Empiricalknockdownfactorsforaxiallyloadedcylinders(from[91])
4.4.3 The Southwell method

In 1932 Southwell published a method which allows determining the buckling load of a perfect
columnfromexperimentaltests[95].Hefoundalinearrelationofthelateraldisplacementandthe
ratioofandtheaxialforceP
1
= ( + 0 ) 46
P PE
where0istheamplitudeoftheinitialgeometricimperfectionandPEthebucklingloadoftheperfect
column. The plot /P versus (see Figure 47) is called the Southwell plot and its inverse slope
correspondstoPE.Hence,withthemeasurementofandPintestsitispossibletodeterminePE,the
bucklingloadoftheperfectcolumn.
Forfurtherdetailstheinterestedreadershouldconsultchapter4of[96].
Figure47:VonKrmnsdataoncompressedcolumnsplottedinthelinearform
bySouthwell(from[95])
33
ECSSEHB3224A
24March2010
Roorda showed that the extension of Southwells method to plates does not work in general, but
depends on the post buckling behavior of the regarded structure [97]. He noted that considering
nonlineareffectsthelinesintheSouthwellplotarenotstraight.Donnell[98],[99]wasthefirstwho
appliedtheSouthwellmethodtocylindricalshells,followedbyotherinvestigators,whoappliedthe
methodtoseveraltypesofshells(seee.g.[100][101]).
TheextensionofSouthwellsmethodtotheinelasticregionseemstobeimpossibleapriori,butWang
showed in 1948 that the application to plastic buckling of columns is possible providing that Er is
constant[102]whereErisacombinationoftheelasticmodulusandthetangentmodulusintheplastic
region.ThiswasconfirmedbySinger[103].Withouttheoreticalevidence,theSouthwellmethodhas
alsobeenappliedtoplasticbucklingofshells(seee.g.[104][106]).
4.4.4 Numerical analysis with realistic geometrical imperfections

The main emphasis in the literature was on geometrical imperfections since they were made
responsibleforthelargediscrepancybetweenanalysisandtest.However,withoutaprioriknowledge
of the shape and amplitude of the imperfections they cannot be taken into account in a numerical
analysis.Winterstetter andSchmidt[107]definethreedifferentapproaches forchoosinggeometrical
imperfections:stimulating,worstandrealisticimperfections.
Realistic geometrical imperfections in a numerical analysis are such, that they describe the existing
prebuckling deformations as good as possible [107]. This approach assumes that specific
manufacturingprocessesstatisticallyresultinbasicallythesameimperfections[73],[108].Methodsfor
accuratelymeasuringimperfectionsofcylindricalshellsandforformulatingthemastwodimensional
FourierseriesweredevelopedbyArbocz,Babcock,Sechler[70].Forprovingpracticalapplicabilitya
cylindrical shell with ten feet diameter and stiffened in circumferential and axial direction was
measuredbyArboczandWilliams[109].Allimperfectionswerestoredindatabanks[110].Assuming
acertainsafetylevelandusingstatisticalmethodstheloadcapacityofimperfectshellswaspredicted.
WhileVeldsetal.[111]usedtheMonteCarlomethod,ElishakoffandArbocz[112]appliedtheFirst
OrderSecondMoment(FOSM)method(seealso[113][115]).Couplingofstochasticmethodswiththe
finiteelementmethodswasfirstdemonstratedbyChryssanthopoulos[116].
4.4.5 Numerical analysis with worst geometrical imperfection

Theobjectiveofthisapproachwastofindtheimperfectionpattern,whichforacertainimperfection
amplitudecausedthelowestbucklingload[107].Initially,eigenformaffinepatternswereassumedto
beworstimperfections,lateraxisymmetricimperfectionswerebelievedtoshowthisbehaviour[117].
InKoitersgeneraltheorytheimperfectionpatternwasassumedtobethecriticalbucklingmode.With
givenbfactorandimperfectionamplitude = / t theknockdownfactorcouldbecalculatedfrom
3 ( 3)
3/ 2 1/ 2
s
( b ) s b < 0
1/ 2
1 = 47
c 2 c
(seeFigure48,where s = Ps / Pc and = / t ).ThisequationwasfirstpresentedbyKoiter[66],for
itsderivationmostauthorsrefertoCohen[118].ThisapproachwasusedbyHutchinsonandAmazigo
[119]andbyHutchinsonandFrauenthal[120]intheirworkoneccentricallystiffenedshells.
TheinfluenceofaxisymmetricimperfectionswasinvestigatedbyKoiterusingaspecifictheory[85].It
was further developed by Tennyson and Muggeridge [86] for local axisymmetric imperfections. A
goodoverviewofthemanypublicationsregardingtheinfluenceofgeometricalimperfectionsonthe
buckling behaviour of axially loaded cylindrical shells is given by Hutchinson and Koiter [121],
Tvergaard [83], Budiansky and Hutchinson [122] and Pignataro [123]. Deml [124] developed a FE
34
ECSSEHB3224A
24March2010
code, which calculated by rigorous mathematical methods the worst imperfection pattern for each
specificshellgeometry.Foraxiallyloadedcylindricalshellstheresultwasasinglebuckle(see[124]
[126]).
Figure48:Imperfectionsensitivityasafunctionofbfactorandimperfection
amplitude[119]
4.4.6 Numerical analysis with stimulating imperfections

Whilethescopeofrealisticgeometricimperfectionswastointroduceexistingimperfectionsintothe
numericalmodel,stimulatingimperfectionsweremeanttoprovidesimplegeometricpatterns,which
stimulatetherealphysicalbehaviouroftheshellandthusproviderealisticresultsfortheknockdown
factor [107]. Often axisymmetric or eigenform affine imperfections were selected for this approach.
RecommendationsfortheamplitudeofimperfectionsaregivenintheguidelinesDAStRichtlinie017
[89] and ENV 19931.6 [127]. However, this often results in a very conservative design [128]. The
influence of local axisymmetric imperfections resulting from welding seams were investigated by
Pircher et al. [129][131], Jrcke et al. [132], Teng and Rotter [133], Hautala and Schmidt [134] and
Bornscheueretal.[135].
4.4.7 Physics based design load

As an alternative to geometric imperfections it was attempted to establish a physics based lower
boundfortheloadcapacity.StartingfromaclosedformsolutionfortheloadshorteningcurveTsien
[136]formulatedasimpleenergycriterion.Failureshouldoccurataloadwherethewholepotential
energybeforeandafterbucklingisthesame(seeFigure49ab).Thesimplecriterionassumedthat
energy from an outside perturbation initializes buckling, but cannot be transferred to the system.
Tsien [137] in 1947 released this assumption and developed a stronger energy criterion stating, that
under outside energetic perturbation failure can occur at the earliest, if the loadshorteningcurve
begins to show multiple equilibrium configurations (see Figure 49cd). Brendel compared these
energeticcriteriawithmanyexperimentalinvestigationsandconcluded,thatthisapproachhadonly
limitedsuccessinprovidingboundsforthetestresults[138].
35
ECSSEHB3224A
24March2010
Simpleenergycriterion Strongenergycriterion
load shortening load shortening
Source:[136] [136],[61] [139],[137] [137],[75]

(a) (b) (c) (d)
Figure49:Lowerboundforthebucklingloadderivedfromloadshorteningcurve
(from[138])
IngeneralKrmnandTsien[139]definedfailureasthetransitionfromthestableprebucklingtothe
stable postbuckling configuration (point GC,Figure 410). To overcome the energy barrier
representedbytheunstableportionofthepostbucklingpath,theyreliedontheexternaldisturbances
always present around the test setup. Since it is difficult to determine the magnitudes of the
disturbancesinadvance,ithasbeensuggestedtousetheminimumofthepostbucklingcurveasasafe
design value in practical engineering application. Motivated by own experimental results, Esslinger
[63]cametothesameconclusion,thattheloadcapacityisequaltotheminimumpostbucklingload
(load in point N, Figure 49c). Later Esslinger and Geier suggested the postbuckling load at the
minimumshorteningasconservativeloadcapacity(loadinpointN,Figure49d)and,togetherwith
Garkisch,themaximumloadatminimumpostcriticalshorteningasprogressiveloadcapacity(loadin
pointG)[75].ThedesignloadderivedfromtestsbyEsslingerisequaltothepredictionsofthestrong
energycriterionbyTsien.
ContinuingthefailuredefinitionofKrmnandTsien,intheperturbationenergyconceptofDuddeck
etal.[140]usinganonlineareigenvalueanalysistheminimumdeformationenergySt(perturbation
energy)andtherelateddeformationpatternvSt(perturbationvector)iscalculatedwhichisrequired
fortransitionfromadefinedstableprebucklingreferenceconfiguration(pointG,Figure410a)toan
adjacent equilibrium configuration. Thus, if there is sufficient energy St available the transition is
from the stable prebuckling to the stable postbuckling configuration (point C) passing the unstable
postbucklingconfiguration(pointB).TherelationbetweenperturbationenergyStandknockdown
factorresultsfromalargenumberofcalculations(seeFigure410b).Theperturbationenergylevel
determines theimperfectionsensitivityofacertainstructureandisderivedfromexperiments.Using
thisvaluetheknockdownfactorandthedesignloadcanbeevaluatedusingthediagraminFigure
410b.Theperturbationvectorcanbeinterpretedasworstimperfectionpattern.
36
ECSSEHB3224A
24March2010

Figure410:(a)Loadshorteningcurvewithrelatedperturbationenergy;(b)
relationbetweenperturbationenergyandknockdownfactor
BasiccontributionstotheperturbationenergyconceptwereprovidedbyHillmann[141],Wagenhuber
[142] and Dinkler [143]. Based on this, Tranel carried out many numerical investigations on elastic
cylindricalshells[144].Schfer[145]andSpohr[146]extendedthescopeoftheconcepttocylinders
with plasticity, Knoke to cylinder cone configurations [147]. According to Knebel [148] the question
remainsopen,whetheralwaystheclosestneighbouringconfigurationisfound.
Rottner et al. [149], [150] applied a dynamic start deflection to the reference configuration. A large
numberofsimulationswasruntodetectthecriticalenergy,whichisrequiredtostimulatetransition
to a neighbouring equilibrium configuration. Similar to the perturbation energy concept the load
capacitycanbedeterminedbyanenergythreshold.
Further concepts tried to explain the influence of imperfections with a reduction of the membrane
stiffness [151][153]. The concept of quasi strain free bending by Fritz und Wittek [154] regards the
energydistributionatthecriticalbucklingpointandfromthatdrawsconclusionstothepostbuckling
behaviour.AccordingtoFritzandWittekthewholemembraneeffectdisappearsatthebucklingpoint,
and the energy balance comprises only bending parts in the postbuckling regime. Then, the design
load is lower than the minimum postbuckling load and therefore by far too conservative (Brendel
[138]). The method of direction dependent reducedmembrane bucklingisa further development of
thequasistrainfreebending.Apartofthemembraneenergyremainsinthestructure,whichresults
in a somewhat higher minimum postbuckling load. Wittek [153] states that this method provides a
good approximation of the minimum postbuckling load for such shell structures, that do not have
multiple mode switches before reaching the postbuckling minimum and that have almost linear
prebuckling behaviour. However, these prerequisites are not fulfilled by the unstiffened cylindrical
shell.
ThebasicideaofthelowerboundtheorydevelopedbyBatista and Croll [155],[156]iscomparableto
thedirectiondependentreducedmembranebucklingmethod.Theapproachesonlydifferintheparts
ofmembraneenergytakenintoaccount.
4.4.8 Overview of different approaches for modelling imperfections

Theknockdownfactorsforshellsofisotropicmaterialrecommendedinexistingguidelinesleadtoa
very conservative design and do not take into account pattern or amplitude of the existing
imperfections.Geometricalimperfectionscanbecoveredinthenumericalanalysiswhenpatternand
amplitudearemeasuredandaprobabilisticevaluationisperformed.However,thisapproachiscostly
as it should be adopted to each manufacturing process anew. For the approach with worst
imperfectionthepatterniscomputedwithhighnumericaleffort.Amplitudeorperturbationenergyis
specified orderived from tests.Forusing imperfection perturbations, patternandamplitude should
37
ECSSEHB3224A
24March2010
be specified such that a realistic load capacity is predicted. As before, this specification is very
difficult, since often sufficient test data for validation are missing. Some approaches for predicting
physicsbaseddesignloadswerederivedfromobservingthestructuralbehaviour.Thesignificanceof
theresultscannotbefullyassessedsofar.Itcanbestatedthatforisotropicshellssofarnoalternative
tothetraditionalknockdownfactorshasbeenestablished.
4.5 Influence of non-traditional imperfections

In literature imperfections are mostly understood as geometrical imperfections. Beside deviations
from perfect geometry also deviations from ideal, homogeneous load introduction occur in reality.
Thesenontraditionalimperfectionscanalsosignificantlyreducethebucklingload.
Alreadyin1932Flgge[43]investigatedtheinfluenceofinhomogeneousloadsp=p0+p1cosacting
on both boundaries on the buckling of isotropic cylindrical shells (see Figure 411a). He concluded,
thatassumingconstantloadingp=p0+p1leadstoasafedesign.1959AbirundNardo[157]published
analytical investigations on cylindrical shells with axial thermal stress varying in circumferential
direction.ForthesameproblemBijlaardandGallagher[158]presentedoneyearlatertheirresultsof
numerical calculations. Analytical solutions for cylinders with general axial loading, which was
developedinaFourierseriesincircumferentialdirection,weredevelopedin1974byPeter[159].All
mentioned publications drew the conclusion that the maximum ordinate of the load distribution
determinesthebucklingloadratherthantheintegralloading.
Uniform loading, interrupted by unloaded areas (see Figure 411b) was numerically investigated by
Guggenberger[160].Theratioofthelengthoftheloadedboundarytothecircumferencewasvaried
anditsinfluenceonthebucklingloadwasdetermined.Limitstatesforareuniformloading(=1)
andasingleload(0).Evensmallunloadedareasresultedinsignificantreductionofthebuckling
loadcomparedtouniformloading.Boundaryimperfectionsduetounevensupportwereinvestigated
numerically and experimentally by Ummenhofer [161] and Knebel [148] (see Figure 411c). Also in
thiscasesignificantreductionswereobserved.
Thielemann, Esslinger and Geier already in the 1960ies ran tests on cylinders from mylar foil (see
Figure 411d). The boundary perturbation was realized by placing pieces of thin cigarette paper
betweencylinderandloadintroductionringatdifferentpositionsalongthecircumference.Surprised
bydrasticloadreduction,theauthorscarriedoutalargenumberoffurthertests,whichconfirmedthe
initialresult.
(a) (b) (c) (d)
Figure411:Differentmethodsofapplyingnonuniformloading
Inthe1980iesGeier,ZimmermannandKleintestedthestructuralbehaviourofanisotropiccomposite
shellsfirstunderuniformloading[162].Thentheyintroducedloadimperfectionsusingthinmetallic
plates(shims)[163].Forcompositeshellstheydetectedimperfectionsensitivitywithrespecttofibre
38
ECSSEHB3224A
24March2010
orientations and stacking sequence, similar to shells under uniform loading [164]. A comprehensive
evaluation and numerical analysis of these socalled shim tests was initiated by Hhne et al. [165].
PractialrelevanceofnonuniformloadingfortheARIANE5designwashighlightedbyAlbusetal.
[166].
4.6 External pressure and torsion on cylindrical shells

SubsequenttohisstudiesonplatebucklingFairbairnwasthefirstwhocarriedoutextensiveandwell
documentedexperimentsoncylindricalshellsunderexternalpressurein1858[167].Hisexperiments
andformulaswerethebasisofboilerdesignandfurtherresearchforthefollowingdecades.Inthelast
decadesofthe19thcenturyBachcarriedoutamultitudeofbucklingexperimentsonseveralstructures
includingsphericalandtorisphericalshells[168].Furthercomprehensivetestontubesunderexternal
pressure were executed by Carman [169] and Steward [170]. With the results of the empirical
investigations Steward developed design rules, while Carman showed that for thin shells analytical
determinationofthebucklingloadgivespassableresults.Thefirsttheoreticaltreatmentofshellunder
externalpressurewaspublishedbySouthwell[171],[172]andalmostatthesametimebyvonMises
[173].Theirpredictionsofbucklingloadsgavegoodresultsforasetofthinwalledsteelshellstested
byvonMises[174],butdidnot agreewiththevarious tests onthickershellsexecutedatthattime.In
1932 Saunders and Windenburg [175] published a report about the problems and necessities in
modelingandtestingshells.Laterstudiesonshellsunderexternalpressurearetheexperimentaland
theoretical works of Tokugawa [176] and Sturm [177], which included stiffened shells. For an
extensivesummaryofthehistoryofshellbucklingunderexternalpressurethereadershouldconsult
chapter9of[178].
The problem of cylindrical shells under torsion wasfirst solved by Donnell[179], who obtained the
critical stress using his famous Donnell equations. Batdorf et al. [180] determined a more accurate
solutionbyapplyingaGalerkinapproachtoDonnellsequations.FurthersolutionsusingtheDonnell
aswellastheFlggeequationsaregivenbyYamaki[181].Thecorrespondingpostbucklingproblem
wastreatedtheoreticallybyLoo[182],Nash[183]andHayashi[184]andexperimentalybyNash[185],
Weingarten[186]andYamaki[187].
4.7 Dynamic buckling

ThefirstnotionofdynamicbucklingcanbetracedtotheinvestigationofKoningandTaub[188]in
the early 30s. Further monographs, which treat the various aspects of dynamic buckling were
published by Bolotin [189], Lindberg and Florence [190], Simitses [191] and AriGur, Weller and
Singer[192][194].
There are two different phenomena, which are assigned to the term dynamic buckling in the
literatureingeneral[178].Thefirstoneistheresponseofastructuretooscillatingloads,whichisalso
associatedtothetermvibrationbucklingandparametricresonance.Itdealswiththeeffectthat
the transverse vibration of a beam becomes unacceptably large at a critical combination of load
amplitude, load frequency and structural damping. This phenomenon was studied by Bolotin [189],
Simitses[191]andaswellbyEvanIwanowski[195].Moreattentionwaspaidtothephenomenonof
the transient response to a pulse load i.e. dynamic buckling under impact loads. It was found that,
unlike under static loading dynamic buckling depends on both magnitude of loading and its
duration. Depending on duration, usually a structure will buckle under loads the magnitudes of
which are higher than its static buckling load. The shorter the duration the higher the magnitude.
However,withinsrangeofloadingdurationcorrespondingtothestructuresfirstlateralfrequencya
structurecanbuckleunderloadsthataresmallerthanitsstaticbucklingload.Thistopicwasalready
39
ECSSEHB3224A
24March2010
investigated by Koning and Taub [188], but only in the last three decades a basic understanding of
buckling under impulse loads has been developed, supported by the development of highspeed
electronicsandphotographicinstrumentation[191],[192]
IthasalreadybeenrecognizedbyKoningandTaub[188]thatthedefinitionofdynamicbucklingis
only possible under consideration of small lateral geometric imperfections. Based on simple
imperfection sensitive models, Budiansky and Hutchinson [196], [197] proposed some theoretical
criteriaandestimatesfordynamicbucklingloads.Furtherdefinitionsofdynamicbucklingaregiven
by Hoff [198], Lindberg and Florence [190] and Maymon and Libai [199] and in Chapter 6.8. An
energy based criterion is given by Humphreys and Bodner [200], which, amongst others, was
discussedandcomparedindetailbySimitses[191].
FormoreinformationondynamicstabilityanalysistheinterestedreadershouldconsultSection6.8.
(a)columnsupportingoscillatingload,
(b)alongcolumnsubjectedtoasuddenappliedaxialload;from[190]
Figure412:columnsubjectedtodynamicload
4.8 Stiffened shells
4.8.1 Overview
Since stiffeners or ribs very significantly increase the buckling loads of plates, stiffened plates have
been extensively studied, starting with Timoshenkos investigations in the first decades of the 20th
century (see [201], [202] and [203]). In the thirties, forties and fifties many studies were primarily
motivated by aircraft design (for example [204], [205]) as well as by civil engineering problems (for
example[206],[207])Marineengineeringstimulatedmuchoftheresearchonstiffenedpanelsinthe
sixties and seventies (for example [208][210]), and the 197071 bridge disasters in Europe and
Australiaaddedtheirinducementtostudiesthatcontinuedintotheninetiesandbeyond.
Stiffeningcylindricalshellsbystringersinaxialdirectionorbyframesincircumferentialdirectionnot
onlyincreasesthetheoreticalbucklingloadbutalsodecreasesimperfectionsensitivity[120],[211].The
stiffening influence mainly depends on geometrical parameters (cross section and distance of
stringers,skinthickness,seeWellerundSinger[212]),oneccentricity(insideoroutsidestiffeners)and
on the stiffening direction (axial or circumferential). Already in 1949, Shanley [213] gave a purely
empirical design method for stiffened cylindrical shells, based on very few test results. In the
40
ECSSEHB3224A
24March2010
following, Peterson and Dow [214], Burns and Almroth [215] and Smith and Spier [216] enhanced
Shanleysmethod,butstillthesemethodswerenotapplicabletogeneraluse.
4.8.2 Imperfection sensitivity

In1956Koiter[217]performedanasymptoticpostbucklinganalysisofnarrow,longcylindricalpanels.
Stephens[218]extendedKoitersanalysistoincludeinternalpressureandfinitetorsionalstiffnessof
the stringers in 1971. Koiter and Skaloud [219] found that the loadcarrying capability of structures
withsimultaneouslocalplatebucklingandEulertypecolumnbucklingmaybeespeciallysensitiveto
initialimperfections.HutchinsonandAmazigo[220]concludedthatthemajoreffectofaxialstiffening
(of cylindrical shells) is to reduce imperfection sensitivity. Using Koiters general theory in 1973
Tvergaard [221] , [222] obtained asymptotic estimates of the imperfectionsensitivity of plates
reinforced by axial stiffeners on one side. A panel designed so that local buckling coincides with
buckling as an Euler column displays a high sensitivity to initial imperfections due to mode
interaction.ButtheoptimumusuallycorrespondstoadesignwithEulerloadlowerthanthecritical
load associated with local buckling of the skin between the stringers [223]. In the following, plates
with several kind of stiffeners were investigated experimentally and analytically by Koiter and
Pignataro[224],vanderNeut[225],Thompson,TulkandWalker[226]andTvergaardandNeedleman
[227], [228]. Byskov and Hutchinson [229] found that while a single column or a single plate is not
sensitive to initial imperfections, the general buckling load of axially stiffened cylindrical shells is
always sensitive to initial imperfections, whereas the effect of interaction between local and general
instabilityistoincreasethesensitivity.ForstiffenedcylindersWellerandSinger[230]determinedthat
theimperfectionsensitivitydependsonthegeometryofstiffenersandontheratioAs/bt,respectively,
where As: crosssection area of the stringer, b: circumferential distance between stringers, t: wall
thicknessoftheshell.
ImperfectionsensitivitywaseitherdescribedbyKoitersbfactororbytheratioofbucklingtolowest
postbuckling load. Using the latter one for orthotropic shells under axial loading March [231] and
SchnellandBrhl[232]found,thatinsomecasestheimperfectionsensitivitywassimilartoisotropic
cylinders, in other cases it was far lower. This stimulated Almroth [57] to perform comprehensive
calculationsfor determining the minimum postbuckling load of axially loaded orthotropic cylinders
(without imperfections) with parameters 0 p 2 and 0.5 s 8.0; p is a plate parameter
depending on the bending stiffness whereas s is a shell parameter depending on the membrane
stiffness.Almrothshowedthattheratioofbucklingtopostbucklingloadonlyweaklydependsonthe
parameters p and s ,butisdominatedbythemainstiffnessratiodependingonmembraneand
bendingstiffness.Cylinderswith1/1weredenotedascircumferentiallystiffened,suchwith1/<1
as axially stiffened. Test results of Gerard, Schulz and Singer et al. [233][235] as well as analytical
results of Thielemann [236] (see Figure 413) showed, that axially stiffened cylinders are more
sensitivetoimperfectionsthancircumferentiallystiffenedones.Amongthese,weaklystiffenedshells
were more sensitive than strongly stiffened ones. However, for given stiffener cross section and
eccentricity the linear buckling load of the perfect shell is usually higher for axial than for
circumferential stiffening, since axial stiffeners carry a significant part of the load (see Figure 413).
Regarding the load carrying capability of imperfect shells (this is according to Thielemann the
minimumpostbucklingload)noaprioriselectionofaxialorcircumferentialstiffeningcouldbemade,
since this would require knowledgeabout imperfection sensitivity. Neglecting optimization aspects,
ry[211]recommendedtouseshortcircumferentiallystiffenedshells,sincethesestructureshavelow
imperfectionsensitivityandcanthereforebedesignedsafelywithoutproblems.
41
ECSSEHB3224A
24March2010
(a) (b)
(From[236])
Figure413:Postbucklingbehaviourofaxiallyandcircumferentiallystiffened
cylindersunderaxialloadingnxandinternalpressurep
4.8.3 Orthotropic shell approach

Eventhoughtheeffectofeccentricityofstiffenershasalreadybeenobservedinthethirtiesandforties
the orthotropic shell approach for cylindrical shells was generally used [237]. Advanced by work at
NASA in the fifties and sixties also axially stiffened panels were analyzed as equivalent orthotropic
platesuntilabout10yearsago[223].
In 1963 Baruch and Singer [238] developed equations using smeared out stiffeners and in the
following, it has been shown that the smeared stiffeners theory is satisfactory for closely stiffened
(cylindrical)shellsthatfailbygeneralinstability[239][248].
Analyticalmethodsforcalculatingthebucklingloadofperfectshellsbasedonalinearprebuckling
configuration were developed by Geier [249], Seggelke and Geier [250] as well as Hutchinson and
Amazigo [119]. Hutchinson and Frauenthal took into account also nonlinear prebuckling
deformations[120].Inalltheseapproachestheinfluenceofdiscretestiffenerswassmeared[249]and
orthotropicmaterialbehaviourwasassumed[250].Anexplanationofthecomplicatedloadcarrying
behaviourasgivenbyGeier[249]wasdisplayedbyKollrandDulcska[117].Shellswereclassified
by Thielemann [236]andSchnell andBrhl[232] by the shell parameter s , by the plate parameter
p andbythemainstiffnessratio.Forisotropic,unstiffenedshells s = p ==1applies.
Wittrick [251] published an analysis of prismatic structures composed of flat plates in 1968, which
predictsthethreetypesofinstability:generalinstability/overallbuckling,localbucklingandcrippling.
Viswanathan et al [252] extended Wittricks analysis to allow orthotropic wall properties and
intermittent elastic beamtype support. Williams [253] extended Wittricks analysis to include
vibration and incorporated substructuring techniques and in 1974 Wittrick and Williams [254]
implementedtheirmethodsintheprogramVIPASA.
In [178] Singer, Arbocz and Weller summarized the three approaches for the analysis of stiffened
plates:
a. Smearingthestiffenersandconsideranorthotropicplate
Thisapproachhassomeconstraintsandisonlyapplicabletoanalyzegeneralinstability.
b. Considering the stiffeners as linear discontinuities (Dirac function). This approach was
developedforcylindricalshellsandconsiderslocalbuckling.
42
ECSSEHB3224A
24March2010
c. Dividing the cross sections into its elements, which leads to a series of elements with rigidly
connectededges
ThisapproachisimplementedinVIPASAbyWittrickandWilliamsandismoreefficientthan
generalpurposefiniteelementcodes.
4.8.4 Eccentricity of stringers

TheimportantinfluenceoftheeccentricityofstiffenerswasdetectedbyvanderNeut[255]in1947.
Van der Neut [225] and Seggelke and Geier [250] showed, that outside stiffeners lead to a higher
bucklingloadthaninsidestiffeners,independentfromtheirorientation.
In the sixties the eccentricity effect of stringer stiffened cylinders has been analyzed by Singer et al
[247],[256]andThielemannandEsslinger[248].AlsoEsslingerdeterminedtheimportantinfluenceof
theeccentricityofstringersandthatontheoutsidethestringersincreasethebucklingloadmorethan
ontheinside[257].Further,Esslingerconcludedthatthelongerashellisthesmalleristheinfluenceof
thestringersonthebucklingload.Singeretal.[240],[247],[258]foundthatdependingonloadingand
shell geometry exceptions are possible where inside stiffeners stiffen the shell more than outside
stiffeners.
ImperfectionsensitivityofstiffenedshellswasinvestigatedbyHutchinsonandAmazigo[119]aswell
as by Budiansky and Hutchinson [259] using Koiters bfactor. They revealed that outside stiffened
shellshaveahigherimperfectionsensitivitythaninsidestiffenedones.
4.8.5 Boundary conditions

For stringerstiffened cylindrical shells the effect of boundary conditions differs significantly from
isotropic and ringstiffened shells [119], [212], [260][262]. Contrary to isotropic cylinders, where
geometricimperfectionshavethemaininfluence,forstiffenedshellstheeffectofboundaryconditions
is predominant [263]. In 1964 Esslinger and Thielemann [264] found that axial and circumferential
stringer stiffened cylindrical shells are insensitive to the shear softness of the boundary condition
which was confirmed by Singer et al [265] and Weller [266]. In the early eighties Singer and co
workers [267], [268], [269] developed the nondestructive vibration correlation technique (VCT) to
determineboundaryconditions,loadeccentricitiesandbucklingloadsonacylindricalshell.
4.8.6 Load cases

TheeffectofloadeccentricitiesoncylindricalshellswasinvestigatedbyStuhlmanetalin1966[270],
followedbyWelleretal[261]andSinger[268]whofoundthatthedifferencesinbucklingloadsdueto
theeffectofloadeccentricitycanbeupto50%forsomepracticalconfigurations.
Tennyson[271]testedringstiffenedandstringerstiffenedcylindricalshellsunderaxialcompression
and external pressure and found that the influence of external pressure as well as the influence of
geometric imperfection on the buckling load is much lower for ringstiffened cylinders than for
stringerstiffened cylinders. Hutchinson and Amazigo [220] emphasized in 1967 that under certain
circumstances,axialstiffeningmaybeamoreefficientmeansofstrengtheningagainstbucklingunder
hydrostaticpressurethanringstiffening.
For isotropic cylinders Libai und Durban [272] concluded that a cosine distributed load does not
decreasethebucklingloadofcylindricalshellssignificantlycomparedtoaconstantloaddistribution.
43
ECSSEHB3224A
24March2010
4.8.7 Optimization
Because of the additional parameters, optimization is an important topic for stiffened structures.
Further,itturnedoutthatmodeinteractionhasasignificantinfluenceontheoptimumdesign[222],
[223],[225],[228],[273][275].In1969VanderNeut[276]analyzedthetwoflangecolumnmodel.He
determined that if the local plate buckling and the general column load are close, the structure
collapses suddenly associated with high imperfectionsensitivity. This model was used for several
optimizationproceduresonstiffenedpanelsintheseventies[277].
4.9 Plastic buckling

Onlyforsufficientlyslenderstructuresisitvalidtoassumethatyieldstressesarenotreachedbefore
instability occurs. For thicker components it is necessary to consider the effect of plasticity. For this
caseEngesser[278]proposedin1889todeterminetheEulerbucklingloadswithtangentmodulusEt
instead of the Youngs modulus E. Considre [279] and von Krmn [280] established the reduced
modulus or doublemodulus Er, which is a function of the Youngs modulus E and Et, where
Et<Er<E. Hence, this approach provides a less conservative buckling load. Shanley [7] showed by
analyticalandexperimentalinvestigationsthatthetwobucklingloads,determinedwithErandwith
Et, can be considered as an upper and a lower bound for the critical inelastic buckling load. An
extensivesummaryofthisdevelopmentisgivenbySewell[281].Hill[282]developedageneralized
tangent modulus approach leading to a bifurcation criterion for threedimensional elasticplastic
solids.
A simple model of elasticstrain hardening plastic material behaviour is given by Ramberg and
Osgood [283]. It describes a stressstrain curve with three material parameters, which have to be
determined experimentally and hence, it allows calculating the tangent modulus Et, which was
alreadydoneandtabulatedfornumerousmaterials.
Concerningplasticbucklingofplatesandshellsthereisoneparadox,whichisstillnotresolved.Itis
calledflowtheoryversusdeformationtheoryparadoxorplasticbucklingparadoxandhasbeen
comprehensively discussed by Hutchinson [284] and Tvergaard and Needleman [285]. It turned out
thatcalculationswiththeflowtheoryoverestimatethebucklingload.Bycontrast,thecorresponding
deformation theory, which is not physically acceptable, gives much better agreements with test
results.OneattempttoresolvethiscontradictionistheunavoidableimperfectionsapproachbyOnat
and Drucker [286]. They showed that already extremely small initial imperfections decrease the
bucklingloadsignificantly,buttheirtheorydoesnotexplaintheexperimentallydeterminedscattering
of the buckling load. Hutchinson and Budiansky [287] confirmed Onat and Drucker, but only for
materials with relatively low strain hardening. For high strain hardening the imperfections must be
much larger in order to obtain the required decrease of the buckling load. The corner theories are
anotherdirectionofapproaches,whichwereinitiatedbyBatdorf[288].ChristoffersenandHutchinson
[289]developedtheJ2cornertheorywhichwasemployedbyNeedlemanandTvergaard[290],who
showedthatthepredictedbucklingloadofthistheoryalwaysfallsbetweentheJ2 deformationtheory
andtheJ2flowtheory.Butthistheorycannotbeusedasadesignmethod,becauseonecannotestimate
theappropriatecornerapriori.
Thoughfurtherinvestigationsleadtomoreaccuratepredictionsofthebucklingload(seee.g.[291]),
theplasticbucklingparadoxcannotyetbeconsideredassettled(seechapter16of[178]).
44
ECSSEHB3224A
24March2010
4.10 Composite shells

Insection4.8itwasdiscussed,thatbucklingofaxiallyloadedstiffenedshellsisdecisivelyinfluenced
by eccentricity, orientation and size of stiffeners [117], [250], [292][294]. When coming to fibre
composite shells, the question of optimal fibre orientations arises. Zimmermann showed with
analyticalcalculations,thatthebucklingloadcanbesignificantlyinfluencedbyfibreorientationsand
stacking sequence [295]. The results were confirmed by the DEVILS project [296]. Geier et al. [297]
revealedthatthebucklingloadofcompositecylinderscanbedoubledbyonlychangingthestacking
sequence. In order to find a physical explanation, the authors referred to stiffened shells. The
composite shell with higher buckling load was compared to an outside stiffened shell, the one with
lowerbucklingloadtoaninsidestiffenedshell.
FordesigningshellstheguidelineNASASP8007[88]isoftenusedinaerospaceengineering.Itcanbe
used for calculating the load carrying capability of isotropic and orthotropic shells. Knockdown
factorsfororthotropicshellshardlydifferfromthoseforisotropicshells.
The ESA handbook [298] recommends the method of Almroth et al. [64]. There, the linear buckling
loadofperfectshellsisdeterminedusingasemianalyticalapproach.Theeffectiveratioofradiusto
wallthickness(R/t)eiscalculatedfrom(R/t)andfromthestiffeners.Usingthis,theknockdownfactor
can be extracted from Figure 46 when assuming a certain probability of failure. Applying the 99
percentprobabilitycurvewasrecommendedin[91].Thismethodturnedouttobeonlyalittlemore
conservative than the direct determination of the knockdown factor for equivalent isotropic shells.
Therefore, Bushnell [91] designed anisotropic shells using the empirical knockdown factors for
isotropicshells(seeFigure46).
Alternatively, the ESA handbook describes the calculation of a knockdown factor according to
Tennyson et al. using the bfactor [299][301]. The amplitude of axisymmetric imperfection
perturbations was derived from measured imperfections. Tennyson et al. found good comparison
betweentheiranalyticalresultsandtestresultsforasmallnumberofshells[300][302].Parameteric
studieswereperformedfororthotropiclaminatesoftype( 90,,90)in[301]andfortype( 90, 2,
90) in [302]. Khot and Venkayya [303][306] also carried out parametric studies for anisotropic
laminates ((,,0) and (,,90)) from different materials. It was observed that the imperfection
sensitivity depends on the fibre angle. (see Figure 414a). Shells that were close to the optimum
showed the largest sensitivity. However, under very large imperfections the dependency of the
bucklingloadonthefibreangledisappeared(Figure414b).SimilartotheinvestigationsofTennyson
etal.itturnedoutthatthesequenceofbucklingloadswasthesameforperfectandimperfectshells.
Theoptimaldesignfortheperfectshellwasstillvalidfortheimperfectshell.Thus,imperfectionsdid
nothavetobeincludedintheoptimizationprocess.
45
ECSSEHB3224A
24March2010
-1.6
w0/t = 0,0
1200
-1.2 w0/t = 0,06
Beullast (lbs/in)
b-Faktor (-)
800 w0/t = 0,2

-0.8
400 w0/t = 1,0

-0.4
0.0 0
0 45 90 0 45 90
Faserwinkel (in grad) Faserwinkel (grad)
(a) (b)
Figure414:bfactorandinfluenceofgeometricimperfectionsasafunctionof
fibreorientation(stackingsequence(,,0),from[307])
In[298]theresultsusingtheapproachesofAlmrothetal.andTennysonetal.werecompared.Since
largedeviationswererevealedinsomecases,furtherinvestigationswererecommended.
Almroth [307] discussed the question, whether or not cylinders from composite material are more
sensitive to imperfections than those from isotropic material. By evaluating many publications it
turnedout,thatunstiffenedcompositeshellsarelesssensitivethanisotropicshells.Therefore,using
theknockdownfactorsforisotropicshellswouldleadtoasafedesign[91].
HilburgerandStarnes[308]raisedtheproblem,thattheexistingtestresultsforisotropicshellsdidnot
comprise any information about the structural behaviour and the characteristic imperfections of
composite shells. Therefore, Hilburger and Starnes as well as Almroth stimulated a detailed
investigationof characteristicimperfections[307],[308].ArboczandHol[309],[310]andEsongetal.
[311] showed that also for composite shells characteristic imperfection patterns are associated with
certain fabrication processes. Accordingly, imperfection data were stored in a data bank [312].
Almroth [307] recommended to investigate, whether imperfections that are specific for composite
materials, e.g. delaminations, reduce the buckling load. Large delaminations can be avoided by
detectionwithnondestructivetesting.Smalldelaminations,accordingto[91],reducethestiffnessof
theshell.Inthe1990iesaccuratemeasurementsrevealedthatastronginfluenceonthebucklingload
is given not only by the traditional geometric imperfections [308] but also by nontraditional load
imperfections caused by unevenness of edges or inhomogeneous stiffness distributions [308], [313],
[314].
As a prerequisite for safe design based on numerical analysis, a validation of numerical models is
necessary.HilburgerandStarnestookintoaccounttraditionalandnontraditionalimperfections[308],
while Elghazouli et al. [315][317], Spagnoli et al. [318] and Chryssanthopoulos et al. [319] mainly
investigated geometrical imperfections. Stochastic approaches were applied by Arbocz und Starnes
[320], Chryssanthopoulos and Poggi. [321] and Elishakoff et al. [322]. A simplified probabilistic
approach was suggested by Hilburger et al. [323]. However, the large number of test data required
andtheexactknowledgeofshapeandmagnitudeofimperfectionsneededmakethemethodcostly.
Zimmermann[324]mentioned,thatgeometricalimperfectionsmayarisenotonlyfromfabricationbut
46
ECSSEHB3224A
24March2010
alsofromthemountingprocesses.Theseandothertypesofimperfectionsshouldalsobecovered by
thisapproach.
It can be concluded that for designing anisotropic cylindrical shells very conservative knockdown
factorsareappliedwhicharebasedontestresultsforisotropicshellsandthereforearenotsuitedfor
optimization. Almroth suggested to account for orthotropic material behaviour by a modified ratio
(R/t)e,forwhichtheknockdownfactorisevaluated.TennysonrecommendedtouseKoitersbfactor
for estimating the imperfection sensitivity and to determine the knockdown factor for an assumed
imperfection amplitude. Both methods, however, are not sufficiently validated so far. Numerical
methods like FEM are presently validated by test results. Stochastic approaches are very costly and
not suited for design purposes. Since test data for composite shells are very rare as compared to
isotropicshells,nopracticallyapplicabledesignconceptisavailableasyet.
Inordertobeabletohandlethenonlinearitiesinthematerialbehaviouroffibrecomposites,Hhne
developed an improved approach for determining the material properties of thin walled composite
structures[325].
A comprehensive explanation of the structural behaviour of cylindrical composite shells under
compression is given by Hhne et al. [326]. By numerical and experimental investigations Hhne
showed that the collapse of cylindrical shells always starts with a single buckle, as it was already
detectedbyEsslingerforisotropicshells[65].Usingthisfact,Hhnedevelopedanewdeterministic
approachforalowerboundofthebucklingload.
Ifoneregardsstiffenedcompositestructures,furtherconsiderationsaboutcasesoffailurehavetobe
made.Thus,inthefirstyearsofthiscenturyseveralapproachesforthemodellingandsimulationof
delaminationweredevelopedbyJarlas[327],Nilsonetal[328],Wagneretal[329],[330]andThomson
et al [331], [332]. Sjgren et al [333] determined the stiffness reduction caused by impact, which is
controlledbyfibrebreakage,andYap,Scott,ThompsonandHachenberg[334]analyzedskinstiffener
debonding.ThedurabilityofstiffenedcompositestructureswasalreadyinvestigatedbyWelleretal
[335], [336] and Frostig et al [337] in the early nineties. In 1996 Hambley [338] tested damaged
cylindricalshellstoinvestigatetheinfluenceofdamageonthefirstbucklingload.
In 2001 the POSICOSS project [339] developed fast and reliable procedures and a final design
guidelineforstiffenedfibrecompositecurvedpanels.ThefollowingprojectCOCOMAT[340]focused
on a reliable simulation of postbuckling and collapse considering structural degradation in order to
determinelessconservativebucklingloadsatthesamesafety.
4.11 References
[1] ElNaschie,M.S.,Stress,StabilityandChaosinStructuralEngineering:AnEnergyApproach,
McGrawHill,1990
[2] Oldfather,W.A.,Ellis,C.A.undBrown,D.M.,LeonhardEulersElasticCurves,Isis,20(1),72
160,1933
[3] Brush,D.O.undAlmroth,B.O.,Bucklingofbars,plates,andshells,McGrawHillNewYork,
1975
[4] Gere,S.T.J.M.undTimoshenko,S.,TheoryofElasticStability,McGrawHill,NewYork,1961
[5] Rivello,R.M.,Theoryandanalysisofflightstructures,McGrawHillNewYork,1969
[6] Simitses,G.J.,Anintroductiontotheelasticstabilityofstructures,PrenticeHall,1976
[7] Shanley,F.R.,Inelasticcolumntheory,JournaloftheAeronauticalSciences,14(5),261268,1947
[8] Prandtl,L.,Kipperscheinungen,TUMnchen,1899
47
ECSSEHB3224A
24March2010
[9] Michell,A.G.M.,Elasticstabilityoflongbeamsundertransverseforces,Philosophical
Magazine,48,298309,1899
[10] Timoshenko,S.P.,EinigeStabilittsproblemederElastizittstheorie,Zeitschr.Math.&Physik,58,
337385,1910
[11] Nylander,H.,TorsionalandlateralbucklingofeccentricallycompressedIandTcolumns,
Kungl.Tekn.Hgskolans,1949
[12] Kindem,S.E.,DrehungundKnickunggeraderStbemitoffenemProfilimelastischenBereich,
TapirsForlag,1949
[13] Chwalla,E.,KippungvonTrgernmiteinfachsymmetrischen,dnnwandigenundoffenen
Querschnitten,SitzungsberichtderAkademiederWissenschaftenWien.IIa,Bd.153,1944,1944
[14] Wagner,H.,VerdrehungundKnickungvonoffenenProfilen,Festschrift25J.THDanzig,1929
[15] Kappus,R.,DrillknickenzentrischgedrckterStbemitoffenenProfilenimelastischenBereich,
LuftfahrtForschung,14,444457,1937
[16] Kappus,R.,ZentrischesundexzentrischesDrehknickenvonStbenmitoffenemProfil,
Stahlbau,22,612,1953
[17] DIN4114:Stahlbau,Stabilittsflle(Knickung,Kippung,Beulung),Berechnungsgrundlagen,
DeutschesInstitutfrNormung,1953
[18] Petersen,C.,StatikundStabilittderBaukonstruktionen,Vieweg&SohnVerlag,1982
[19] Timoshenko,S.P.,HistoryofStrengthofMaterials,McGrawHill,1953
[20] Walker,A.C.,ABriefReviewofPlateBucklingResearch,BehaviorofThinWalledStructures,
375398,1984
[21] Bryan,G.H.,OntheStabilityofaPlanePlateunderThrustsinitsownPlane,withApplications
totheBucklingoftheSidesofaShip,ProceedingsoftheLondonMathematicalSociety,1(1),
541890
[22] Gerard,G.undBecker,H.,HandbookofStructuralStability,NACATN3781,NewYorkUniv.,
NewYork,1957
[23] Coan,J.M.,Largedeflectiontheoryforplateswithsmallinitialcurvatureloadedinedge
compression,JournalofAppliedMechanics,18,143151,1951
[24] Schuman,L.undBack,G.,StrengthofRectangularPlatesUnderEdgeCompression,TR356,
NACA,1930
[25] vonKrmn,T.,Sechler,E.E.undDonnell,L.H.,TheStrengthofThinPlatesinCompression,
ASMEAppliedMechanicsTransactions,54(2),5357,1932
[26] Cox,H.L.,BucklingofThinPlatesinCompression,R.&M.1554,AeronauticalResearch
Committee,1933
[27] Dickinson,H.B.undFischel,J.R.,MeasurementofStiffenerStressesandEffectiveWidthsin
StiffenedPanels,JournaloftheAeronauticalScience,6(6),249254,1939
[28] Maguerre,K.,DiemittragendeBreitedergedrcktenPlatte,LuftfahrtForschung,14(3),121128,
1937
[29] Sechler,E.E.,TheUltimateStrengthofThinFlatSheetinCompression,GALCITPublication,27,
1933
[30] Sechler,E.E.,StressDistributioninStiffenedPanelsunderCompression,Journalof
AeronauticalSciences,4(8),320323,1937
[31] Becker,H.,Goldman,R.undPazerycki,J.,CompressiveStrengthofShipHullGirders.PartI.
UnstiffenedPlates,SSC217,ShipStructuresComittee,1970
[32] Conley,W.F.,Becker,L.A.undAllnutt,R.B.,BucklingandUltimateStrengthofPlating
LoadedinEdgeCompression,1682,DTMBReport,1963
48
ECSSEHB3224A
24March2010
[33] Faulkner,D.,AReviewofEffectiveSteelPlatingforUseintheAnalysisofStiffenedPlatingin
BendingandCompression,JournalofShipResearch,19(1),117,1975
[34] Fukumoto,Y.undItoh,Y.,BasicCompressiveStrengthofSteelPlatesFromTestData,344,
Proc.Jpn.Soc.Civ.Eng.(JSCE),129139,1984
[35] Kollbrunner,C.F.,DasAusbeulenderaufeinseitigen,gleichmassigverteiltenDruck
beanspruchtenPlattenimelastischenundplastischenBereich,(Versuchsbericht)Mitt.Inst.
Baustatik,ETHZrich,HNo17,1946,Lehmann,Zrich
[36] Winter,G.,StrengthofThinSteelCompressionFlanges,TransactionsoftheASCE,112(2305),
527576,1947
[37] Stein,M.,ThePhenomenonofChangeinBucklePatterninElasticStructures,NASATechnical
ReportR39,1959
[38] Stein,M.,LoadsandDeformationsofBuckledRectangularPlates,NASATechnicalReportR40,
1959
[39] Uemura,M.undByon,O.I.,Secondarybucklingofaflatplateunderuniaxialcompression.Part
I:Theoreticalanalysisofsimplysupportedflatplate,InternationalJournalofNonlinear
Mechanics,12(6),355370,1977
[40] Lorenz,R.,AchsensymmetrischeVerzerrungenindnnwandigenHohlzylindern,Zeitschrift
VereinDeutscherIngenieure,52(43),17061713,1908
[41] Timoshenko,S.,EinigeStabilittsproblemederElastizittstheorie,Z.Math.Physik,58(4),337
385,1910
[42] Southwell,R.V.,Onthegeneraltheoryofelasticstability,Phil.Trans.Roy.Soc.London,Series
A(213),187244,1914
[43] Flgge,W.,DieStabilittderKreiszylinderschale,Ingenieurarchiv,3,463506,1932
[44] Batdorf,S.B.,ASimplifiedMethodofElasticStabilityAnalysisforThinCylindricalShellsI
DonnellsEquation,NACATN1341,June1947.
[45] Lundquist,E.E.,Strengthtestsofthinwalledduraluminiumcylindersincompression,NACA
ReportNo.473,June1933
[46] Donnell,L.H.,Anewtheoryforthebucklingofthincylindersunderaxialcompressionand
bending,Trans.ASME,56,795806,1934
[47] Weingarten,V.I.,Morgan,E.J.undSeide,P.,Elasticstabilityofthinwalledcylindricaland
conicalshellsunderaxialcompression,AIAAJournal,3(3),500505,1965
[48] Ohira,H.,Localbucklingtheoryofaxiallycompressedcylinders,ProceedingsoftheEleventh
JapanNationalCongressofAppliedMechanics,34Tokyo,Japan,1961
[49] Hoff,N.J.,BucklingofThinShells,ProceedingsofanAerospaceSymposiumofDistinguished
LecturersinHonorofDr.TheodorevonKrmnonthe80thAnniversary,InstituteofAerospace
Science,NewYork,1961
[50] Nachbar,W.undHoff,N.Y.,TheBucklingofaFreeEdgeofanAxiallyCompressedCircular
CylindricalShell,QuarterlyofAppliedMechanics,20,267277,1962
[51] Hoff,N.J.undSoong,T.C.,Bucklingofcircularcylindricalshellsinaxialcompression,
SUDAER204,StanfordUniversity,Calif.,August1964
[52] Almroth,B.D.,Influenceofedgeconditionsonthestabilityofaxiallycompressedcylindrical
shells,NASACR161,February1965
[53] vonKrmn,T.undTsien,H.S.,Bucklingofthincylindricalshellsunderaxialcompression,
JournalofAeronauticalSciences,8,302335,1941
[54] Leggett,D.M.A.undJones,R.P.N.,Thebehaviourofacylindricalshellunderaxial
compressionwhenthebucklingloadhasbeenexceeded,ARCR&M2190,1942
49
ECSSEHB3224A
24March2010
[55] Michielsen,H.F.,Thebehaviourofthincylindricalshellsafterbucklingunderaxial
compression,JournalofAeronauticalSciences,15,738744,1948
[56] Kempner,J.,Postbucklingbehaviourofaxiallycompressedcircularcylindricalshells,Journalof
AeronauticalSciences,21,329342,1954
[57] Almroth,B.O.,Postbucklingbehaviourofaxiallycompressedcircularcylinders,AIAAJournal,
1,630633,1963
[58] Hoff,N.J.,ThePerplexingbehaviorofthincircularcylindricalshellsinaxialcompression,Israel
JournalofTechnology,4(1),128,1966
[59] Hoff,N.J.,Madsen,W.A.undMayers,J.,Postbucklingequilibriumofaxiallycompressed
circularcylindricalshells,AIAAJournal,4,126133,1966
[60] Thielemann,W.F.undEsslinger,M.,Onthepostbucklingbehaviorthinwalledaxially
compressedcircularcylindricalshellsoffinitelength,70thAnniversarySymposiumonthe
TheoryofShellstohonorLloydHamiltonDonnell,Houston,1966
[61] Yoshimura,Y.,Onthemechanismofbucklingofacircularcylindricalshellunderaxial
compression,NACATechnicalMemorandum1390,1955
[62] Esslinger,M.,NachbeulrechnungdnnwandigerKreiszylinderschalen,DLRFB6725,DLR
Braunschweig,1967
[63] Esslinger,M.,EineErklrungdesBeulmechanismusvondnnwandigenKreiszylinderschalen,
DerStahlbau,12,366371,1967
[64] Almroth,B.O.,Holmes,A.M.C.undBrush,D.O.,AnExperimentalStudyoftheBucklingof
CylindersunderAxialCompression,ExperimentalMechanics,4(9),263270,1964
[65] Esslinger,M.,HochgeschwindigkeitsaufnahmenvomBeulvorgangdnnwandiger,
axialbelasteterZylinder,Sonderdruckaus:DerStahlbau,3,1970
[66] Koiter,W.T.,Overdestabiliteitvanhetelastischevenwicht,Ph.D.thesisinDutch,THDelft,
Netherlands,H.T.Paris,Amsterdam1954
[67] Koiter,W.T.,OntheStabilityofElasticEquilibrium,NASATTF10833,1967
[68] vonKrmn,T.,Dunn,L.G.undTsien,H.S.,Theinfluenceofcurvatureonthebuckling
characteristicsofstructures,JournalofAeronauticalSciences,JournalofAeronauticalSciences,
7,276289,1940
[69] Donnell,L.H.undWan,C.C.,Effectsofimperfectionsofthincylindersandcolumnsunderaxial
compression,J.Appl.Mech,17,7383,1950
[70] J.Arbocz,Theeffectofinitialimperfectionsonshellstability.In:ThinShellStructures,Theory,
ExperimentandDesign.(Ed.Y.C.FungandE.E.Sechler)205245,PrenticeHall,Englewood
Cliffs,NJ,1974
[71] Dsing,H.,StabilittlngsversteiftersthlernerKreiszylinderschalenunterzentrischem
AxialdruckTheoretischeGrundlagenundbaupraktischerBeulsicherheitsnachweis,
FachbereichBauwesen,UniversittGHEssen,1994
[72] Pflger,A.,ZuraxialgedrcktenKreiszylinderschale,Braun.Wiss.Gesellschaft,14,91109,1962
[73] Arbocz,J.undBabcock,C.D.,TheEffectofGeneralImperfectionsontheBucklingofCylindrical
Shells,JournalofAppliedMechanics,36(1),2838,1969
[74] Arbocz,J.undSechler,E.E.,OntheBucklingofAxiallyCompressedImperfectCylindrical
Shells,JournalofAppliedMechanics,41(3),737743,1974
[75] Esslinger,M.undGeier,B.,GerechneteNachbeullastenalsuntereGrenzederexperimentellen
axialenBeullastenvonKreiszylindern,Sonderdruckaus:DerStahlbau,41(12),1972
[76] Yamaki,N.,Elasticstabilityofcircularcylindricalshells,NorthHolland,Amsterdam,New
York,Oxford,1984
50
ECSSEHB3224A
24March2010
[77] Budiansky,B.undHutchinson,J.W.,DynamicBucklingofImperfectionsensitiveStructures,
Proc.XIInternat.Congr.Appl.Mech.,636651,Munich,1966
[78] Budiansky,B.,Dynamicbucklingofelasticstructures:criteriaandestimates,DynamicStability
ofStructures,Proc.ofanInternat.Conf.,83106,PergamonPress,NorthwesternUniv.,
Evanston,Illinois,1967
[79] B.Budiansky,TheoryofBucklingandPostBucklingBehaviourofElasticStructures.In
AdvancesinAppliedMechanics,14.(Ed.C.S.Yih)165,1974
[80] P.Seide,AreexaminationofKoiterstheoryofinitialpostbucklingbehaviorandimperfection
sensitivityofstructures.InThinShellStructures,Theory,Experiment,Design.(Ed.Y.C.Fung
andE.E.Sechler)5980,PrenticeHall,1974
[81] Masur,E.F.,BucklingofShellsgeneralintroductionandreview,ASCENationalStructural
EngineeringMeeting,Preprint2000,SanFrancisco,1973
[82] Almroth,B.O.undBrush,D.O.,Bucklingofbars,platesandshells,McGrawHillBook
Company,1975
[83] Tvergaard,V.,BucklingBehaviourofPlateandShellStructures,NorthHollandPublishingCo.,
1976
[84] Koiter,W.T.,ElasticStabilityandPostbucklingBehavior,Proc.Symp.NonlinearProblems,257
275,UniversityofWisconsinPress,1963
[85] Koiter,W.T.,Theeffectofaxisymmetricimperfectionsonthebucklingofcylindricalshells
underaxialcompression,Proc.Roy.Netherl.Acad.Sci.Ser.B.No.6,265279,1963
[86] Tennyson,R.C.undMuggeridge,D.B.,Bucklingofaxisymmetricimperfectcircularcylindrical
shellsunderaxialcompression,AIAAJournal,7(11),21272131,1969
[87] Jrcke,R.K.,ZurStabilittundImperfektionsempfindlichkeitelastischerSchalentragwerke
FiniteElementeFormulierungderAnfangsNachbeultheorie,MitteilungNr.855,Institutfr
konstruktivenIngenieurbau,RuhrUniversittBochum,1985
[88] Anonym,BucklingofThinwalledCircularCylinders,NASASP8007,August1968
[89] Anonym,BeulsicherheitnachweisefrSchalen,DeutscherAusschufrStahlbau,DASt
Richtlinie017,1992
[90] Harris,L.A.,Suer,H.S.,Skene,W.T.undBenjamin,R.J.,TheStabilityofThinWalledUnstiffened
CircularCylindricalCylindersUnderAxialCompressionIncludingtheEffectsofInternal
Pressure,JournalofAeronauticalSciences,24(8),587596,1957
[91] Bushnell,D.,Computerizedbucklinganalysisofshells,MartinusNijhoffPublishers,1985
[92] Nemeth,M.P.undStarnes,H.S.Jr.,TheNASAMonographsonShellStabilityDesign
RecommendationsAReviewandSuggestedImprovements,NASA/TP1998206290,NASA,
Hampton,Virginia,1998
[93] M.Grimmelt,D.Nendl,E.Sperlich,ApplicationofProbabilisticReliabilityAnalysisMethodsto
ARIANE5UnequipedBoosterCase,StructuralSafetyandReliability,ProceedingsofICOSSAR
89,the5thInternationalConferenceonStructuralSafetyandReliability,SanFransisco,August
711,1989
[94] Bornscheuer,F.W.,PlastischesBeulenvonKreiszylinderschalenunterAxialbelastung,Der
Stahlbau,9,257262,1981
[95] Southwell,R.V.,OntheAnalysisofExperimentalObservationsinProblemsofElasticStability,
ProceedingsoftheRoyalSocietyofLondon.SeriesA,135(828),601616,1932
[96] Singer,J.,Arbocz,J.andWeller,T.,BucklingExperiments:ExperimentalMethodsinBuckling
forThinWalledStructures,Volume1,Wiley,2002.
[97] Roorda,J.,SomeThoughtsontheSouthwellPlot,JournaloftheEngineeringMechanics
Division,ASCE,93,3748,1967
51
ECSSEHB3224A
24March2010
[98] Donnell,L.H.,Anewtheoryforthebucklingofthincylindersunderaxialcompressionand
bending,TransactionsoftheASME,56(11),795806,1934
[99] Donnell,L.H.,OntheapplicationofSouthwellsmethodfortheanalysisofbucklingtests,
McGrawHill,NewYork,1938
[100] Galletly,G.D.undReynolds,T.E.,ASimpleExtensionofSouthwellsMethodforDetermining
theElasticGeneralInstabilityPressureofRingstiffenedCylindersSubjecttoExternal
HydrostaticPressure,ProceedingsoftheSocietyforExperimentalStressAnalysis,13(2),141
152,1956
[101] Horton,W.H.,Nassar,E.M.undSinghal,M.K.,DeterminationoftheCriticalLoadsofShellsby
NondestructiveMethods,ExperimentalMechanics,17(4),154160,1977
[102] Wang,C.T.,InelasticColumnTheoriesandanAnalysisofExperimentalObservations,Journal
oftheAeronauticalSciences,15(5),283292,1948
[103] Singer,J.,OntheApplicabilityoftheSouthwellPlottoPlasticBuckling,Experimental
Mechanics,29(2),205208,1989
[104] Galletly,G.D.,Blachut,J.undMoreton,D.N.,InternallyPressurizedMachinedDomedEndsA
ComparisonofthePlasticBucklingPredictionsofDeformationandFlowtheories,Proceedings
oftheInstitutionofMechanicalEngineers,204(c3),169186,1990
[105] Newman,J.B.,InelasticColumnBucklingofInternallyPressurizedTubes,Experimental
Mechanics,13(7),265272,1973
[106] Sobel,L.H.,TheSouthwellMethodforPredictingPlasticBucklingLoadsforElbows,Journalof
PressureVesselTechnology,TransactionsASME,105,28,1983
[107] Winterstetter,T.A.undSchmidt,H.,Stabilityofcircularcylindricalsteelshellsundercombined
loading,ThinWalledStructures,40(10),893910,2002
[108] Arbocz,J.,AbouttheStateoftheArtofShellDesign,MemorandumM629,DelftUniversityof
Technology,1990
[109] Arbocz,J.undWilliams,J.G.,ImperfectionSurveysona10ftDiameterShellStructure,AIAA
Journal,15(7),949956,1977
[110] Abramovich,H.,Yaffe,R.undSinger,J.,EvaluationofStiffenedShellCharacteristicsfrom
ImperfectionMeasurements,JournalofStrainAnalysis,22(1),1723,1987
[111] Velds,E.,Wijker,J.undKlein,M.,Probabilisticbucklinganalysisusingthefiniteelement
method,EuropeanConferenceonSpacecraftStructures,Materials&MechanicalTesting,
Toulouse,France,2002
[112] Elishakoff,I.undArbocz,J.,BucklingofShellswithGeneralRandomImperfections,
MemorandumM401,DelftUniversityofTechnology,1981
[113] Amazigo,J.C.,BucklingunderAxialCompressionofLongCylindricalShellswithRandom
AxisymmetricImperfections,QuarterlyofAppliedMathematicsVol.26,No.4,pp.537566,
1969
[114] Bolotin,V.V.,Statisticalmethodsinthenonlineartheoryofelasticshells,Englishtranslation,
NASATTF85,1962
[115] Fraser,W.B.undBudiansky,B.,TheBucklingofaColumnWithRandomInitialDeflections,
JournalofAppliedMechanics,36,233240,1969
[116] Chryssanthopoulos,M.K.,Baker,M.J.undDowling,P.J.,StatisticalAnalysisofImperfectionsin
StiffenedCylinders,JournalofStructuralEngineering,117(5),19791997,1991
[117] Kollar,D.undDulacska,E.,Schalenbeulung,TheoryundErgebnissederSabilittgekrmmter
FlchentragwerkeWernerVerlag,Dsseldorf,1975
[118] Cohen,G.A.,EffectofaNonlinearPrebucklingStateonthePostbucklingBehaviorand
ImperfectionSensitivityofElasticStructures,AIAAJournal,6(8),16161619,1968
52
ECSSEHB3224A
24March2010
[119] Hutchinson,J.W.undAmazigo,J.C.,Imperfectionsensitivityofeccentricallystiffened
cylindricalshells,AIAAJournal,5(3),392401,1967
[120] Hutchinson,J.W.undFrauenthal,J.C.,Elasticpostbucklingbehaviorofstiffenedandbarelled
cylindricalshells,JournalofAppliedMechanics,36,784790,1967
[121] Hutchinson,J.W.undKoiter,W.T.,Postbucklingtheory,AppliedMechanicsReview,13531366,
1970
[122] B.BudianskyundJ.W.Hutchinson,BucklingandChallenge.In:TrendsinSolidMechanics.
(Ed.J.F.BesselingandA.M.A.vanderHeijden)SijthoffNoordhoff,Amsterdam,1980
[123] Pignataro,M.,Rizzi,N.undLuongo,A.,Stability,BifurcationandPostcriticalBehaviorofElastic
Structures,Elsevier,1991
[124] Deml,M.,EinFiniteElementKonzeptzurTraglastanalyseimperfektionssensitiver
Schalenstrukturen,TechnischeUniversittMnchen,1997
[125] Deml,M.undWunderlich,W.,Directevaluationoftheworstimperfectionshapeinshell
buckling,ComputerMethodsinAppliedMechanicsandEngineering,149(14),201222,1997
[126] M.Deml,DirectEstimationofShellBucklingLoads.InModellingandSimulationBased
Engineering,VolumeI.(Ed.S.N.AtluriandP.E.ODonoghue)TechSciencePress,1998
[127] Anonym,ENV19961.6,1996
[128] Winterstetter,T.A.,StabilittvonKreiszylinderschalenausStahlunterkombinierter
Beanspruchung,FachbereichBauwesen,UniversittGHEssen,2000
[129] Pircher,M.,Berry,P.A.,Ding,X.undBridge,R.Q.,Theshapeofcircumferentialweldinduced
imperfectionsinthinwalledsteelsilosandtanks,ThinWalledStructures,39(12),9991014,2001
[130] Pircher,M.undBridge,R.,Theinfluenceofcircumferentialweldinducedimperfectionsonthe
bucklingofsilosandtanks,JournalofConstructionalSteelResearch,57(5),569580,2001
[131] Pircher,M.,OShea,M.D.undBridge,R.Q.,Theinfluenceofthefabricationprocessonthe
bucklingofthinwalledsteelboxsections,ThinWalledStructures,40(2),109123,2002
[132] Jrcke,R.K.,Krtzig,W.B.undWittek,U.,Kreiszylinderschalenmitwulstartigen
Imperfektionen,DerStahlbau,8,241244,1983
[133] Rotter,J.M.undTeng,J.G.,ElasticStabilityofCylindricalShellswithWeldDepressions,Journal
ofStructuralEngineering,15,12441263,1989
[134] Hautala,K.undSchmidt,H.,BucklingTestsonAxiallyCompressedCylindricalShellsMadeof
VariousAustenticStainlessSteelsatAmbientandElevatedTemperatures,76,Universitt
GesamthochschuleEssen,Essen,1998
[135] Bornscheuer,F.W.,Hfner,L.undRamm,E.,ZurStabilitteinesKreiszylindersmiteiner
RundschweinahtunterAxialbelastung,DerStahlbau,10,313318,1983
[136] Tsien,H.S.,Atheoryforthebucklingofthinshells,JournalofAeronauticalSciences,9,373
384,1942
[137] Tsien,H.S.,LowerBucklingLoadintheNonlinearBucklingTheoryforThinShells,Quarterly
ofAppliedMathematics,5,236237,1947
[138] Brendel,B.,GeometrischnichtlineareElastoStabilitt,Nr.791,InstitutfrBaustatik,
UniversittStuttgart,1979
[139] vonKrmn,T.undTsien,H.S.,TheBucklingofSphericalShellsbyExternalPressure,Journal
ofAeronauticalSciences,7,4350,1939
[140] H.Duddeck,B.Krplin,D.Dinkler,J.HillmannundW.Wagenhuber,Berechnungdes
nichtlinearenTragverhaltensdnnerSchalenimVorundNachbeulbereich.In:Nichtlineare
BerechnungenimKonstruktivenIngenieurbau.(Ed.E.Stein)SpringerVerlag,
53
ECSSEHB3224A
24March2010
[141] Hillmann,J.,GrenzlastenundTragverhaltenaxialgestauchterKreiszylinderschalenimVor
undNachbeulbereich,TUBraunschweig,1985
[142] Wagenhuber,W.,ImperfektionssensitivittundrechnerischerNachweisderBeulsicherheit
dnnerSchalen,BerichtNr.8959,InstitutfrStatik,TUBraunschweig,1989
[143] Dinkler,D.,StabilittdnnerFlchentragwerkebeizeitabhngigenEinwirkungen,BerichtNr.
8852,InstitutfrStatik,TUBraunschweig,1988
[144] Tranel,G.,StabilittsnachweisbeliebigerSchalenmitdemKonzeptderStrenergie,Institutfr
Statik,TechnischeUniversittBraunschweig,1994
[145] Schfer,U.,GrenzlastenvonKreiszylinderschalenbeielastischplastischemMaterialverhalten,
BerichtNr.1497,InstitutfrStatikundDynamikderLuftundRaumfahrtkonstruktionen,
UniversittStuttgart,1997
[146] Spohr,I.,StrenergieKonzeptfrdenelastoplastischenBeulsicherheitsnachweisbeliebig
belasteterZylinderschalen,InstitutfrStatik,TechnischeUniversittBraunschweig,1998
[147] Knoke,O.,BeulwiderstndezusammengesetzterZylinderKegelSchalen,InstitutfrStatik,
TechnischeUniversittBraunschweig,2003
[148] Knebel,K.,StabilittvonStahlzylindernmitunilateralenRandbedingungenbeistatischenund
dynamischenBeanspruchungen,InstitutfrMechanik,UniversittFridericianaKarlsruhe(TH),
1997
[149] Rottner,T.,SensitivittsanalysednnwandigerStrukturen,InstitutfrMechanik,Universitt
FridericianaKarlstuhe(TH),2000
[150] Rottner,T.,Schweizerhof,K.undVielsack,P.,TransientAnalysistoComputePostBuckling
SolutionsforAxiallyLoadedCylindricalShellsandSensitivityInvestigations,Fourth
InternationalColloquiumonComputationofS
[151] Wittek,U.,LinearesundnichtlinearesStabilittsverhaltenderFlchentragwerkeausderSicht
dehnungsloserBeulzustnde,S77/1,ForschungsundSeminarberichtausdemBereichder
MechanikderUniversittHannover,1977
[152] Croll,J.G.A.undEllinas,C.P.,ReducedStiffnessAxialLoadBucklingofCylinders,
InternationalJournalofSolidsandStructures,19(5),461477,1983
[153] U.Wittek,AbschtzungskonzeptezurSchalenbeulung.In:NichtlineareBerechnungenim
konstruktivenIngenieurbau.(Ed.E.Stein)SpringerVerlag,1989
[154] Fritz,H.undWittek,U.,ZurStabilittderFlchentragwerke,Institutfrkonstruktiven
Ingenieurbau,RuhrUniversittBochum,1974
[155] R.C.BatistaundJ.G.A.Croll,Adesignapproachforaxiallycompressedunstiffenedcylinders.
InStabilityproblemsinengineeringstructuresandcomponents.(Ed.T.H.RichardsandP.
Stanley)377399,1979
[156] Croll,J.G.A.undBatista,R.C.,Explicitlowerboundsforthebucklingofaxiallyloaded
cylinders,InternationalJournalofMechanicalScience,23(6),331343,1981
[157] Abir,D.undNardo,S.V.,ThermalBucklingofCircularCylindricalShellsundercircumferential
TemperatureGradients,JournalofAerospaceSciences,26,803808,1959
[158] Bijlaard,P.P.undGallagher,R.H.,ElasticInstabilityofaCircularCylindricalShellUnder
ArbitraryCircumfirentialVariationofAxialStress,JournalofAerospaceSciences,27,854858,
1960
[159] Peter,J.,ZurStabilittvonKreiszylinderschalenunterungleichmssigverteiltenaxialen
Randbelastungen,Nr.19,Prof.Dr.Ing.habil.A.Pflger,Hannover,1974
[160] Guggenberger,W.,NichtlinearesBeulverhaltenvonKreiszylinderschalenunterlokaler
Axialbelastung,Heft6,InstitutfrStahlbau,HolzbauundFlchentragwerke,Fakulttfr
Bauingenieurwesen,TechnischeUniversittGraz,
54
ECSSEHB3224A
24March2010
[161] Ummenhofer,T.,StabilittsverhaltenimperfekterzylindrischerStahlsiloschalen
experimentelleundnumerischeUntersuchungen,5.Folge,Heft2,Karlsruhe,1997
[162] B.Geier,H.KleinundR.Zimmermann,Bucklingtestswithaxiallycompressedunstiffened
cylindricalshellsmadefromCFRP.InBucklingofshellstructuresonland,intheseaandinthe
air.(Ed.J.F.Julien)498507,1991
[163] Geier,B.,Klein,H.undZimmermann,R.,ExperimentsonbucklingofCFRPcylindricalshells
undernonuniformaxialload,ProceedingsofInternationalConferenceonComposites
Engineering,10431044,Hui,D.,1994
[164] Zimmermann,R.,Bucklingresearchforimperfectiontolerantfibercompositestructures,Proc.
ConferenceonSpacecraftStructures,Materials&MechanicalTesting,411416,GrandHotel
HuisterDuin,Noordwijk,TheNetherlands,
[165] Hhne,C.,Zimmermann,R.,Rolfes,R.undGeier,B.,Sensitivitiestogeometricalandloading
imperfectionsonbucklingofcompositecylindricalshells,ProceedingsEuropeanConferenceon
SpacecraftStructures,MaterialsandMechanicalTesting,Toulouse,1113December2002.
[166] Albus,J.,GomezGarcia,J.undOery,H.,ControlofAssemblyInducedStressesand
DeformationsduetotheWavinessoftheInterfaceFlangesoftheESCAUpperStage,19,52nd
InternationalAstronauticalCongress,Toulouse,
[167] Fairbairn,W.,OntheResistanceofTubestoCollapse,PhilosophicalTransactionsoftheRoyal
Society,148,389413,1858
[168] Bach,C.,UntersuchungenberdieFormnderungenunddieAnstrengungengewlbterBden,
ZeitschriftdesVereinsdeutscherIngenieure,43(51),15851593,1899
[169] Carman,A.P.,ResistanceofTubestoCollapse,PhysicalReview,21(6),381387,1905
[170] Steward,R.T.,CollapsingPressureofBessemerSteelLapWeldedTubes,ThreetoTenInchesin
Diameter,TransactionsoftheASME,27,730822,1906
[171] Southwell,R.V.,OntheCollapseofTubesbyExternalPressure,PhilosophicalMagazine,25(6),
687698,1913
[172] Southwell,R.V.,OntheGeneralTheoryofElasticStability,PhilosophicalTransactionsofthe
RoyalSocietyofLondon,213(SeriesA),187244,1914
[173] vonMises,R.,DerkritischeAuendruckzylindrischerRohre,ZeitschriftdesVereinsdeutscher
Ingenieure,58(19),750755,1914
[174] vonMises,R.,DerKritischeAussendruckfrallseitsbelastetezylindrischeRohre,Festschrift
zum70.GeburstagvonProf.Dr.A.Stodola,418430,1929
[175] Saunders,H.E.undWindenburg,D.F.,TheUseofModelsinDeterminingtheStrengthofThin
WalledStructures,TransactionsoftheASME,54,263275,1932
[176] Tokugawa,T.,ModelExperimentsontheElasticStabilityofClosedandCrossStiffened
CircularCylindersunderUniformExternalPressure,Proceedings,WorldEngineering
Congress,29,249249,1931
[177] Sturm,R.G.,AStudyofthecollapsingpressureofthinwalledcylinders,UniversityofIllinois,
EngineeringExperimentStation,1941
[178] Singer,J.,Arbocz,J.undWeller,T.,BucklingExperiments:ExperimentalMethodsinBucklingof
ThinwalledStructures,Volume2,Wiley,2002
[179] Donnell,L.H.,Stabilityofthinwalledtubesundertorsion,NACAReportNo.479,1933
[180] Batdorf,S.B.,Stein,M.undSchildcrout,M.,CriticalStressofThinWalledCylindersinTorsion,
NACATN1344,1947
[181] Yamaki,N.,BucklingofCircularCylindricalShellsUnderTorsion,TohokuUniversity,1966
55
ECSSEHB3224A
24March2010
[182] Loo,T.T.,Effectsoflargedeflectionsandimperfectionsontheelasticbucklingofcylinders
undertorsionandaxialcompression,Proceedingsof2ndU.S.NationalCongressonApplied
Mechanics,345357,1954
[183] Nash,W.A.,BucklingofInitiallyImperfectCylindricalShellsSubjectedtoTorsion,Journalof
AppliedMechanics,24,125130,1957
[184] Hayashi,T.undHirano,Y.,Postbucklingbehavioroforthotropiccylindersundertorsion,
TransactionoftheJapanSocietyforAeronauticalandSpaceSciences,7,4754,1964
[185] Nash,W.A.,AnExperimentalAanlysisoftheBucklingofthinInitiallyImperfectCylindrical
ShellsSubjecttoTorsion,ProceedingsoftheSocietyforExperimentalStressAnalysis,16,5568,
1959
[186] Weingarten,V.I.,Theeffectofinternalpressureandaxialtensiononthebucklingofcylindrical
shellsundertorsion,Proceeingsofthe4thU.S.NationalCongressofAppliedMechanics,827
827,1962
[187] N.Yamaki,Experimentsonthepostbucklingbehaviorofcircularcylindricalshellsunder
torsion.In.312330,Springer,1976
[188] Koning,C.undTaub,J.,StossartigeKnickbeanspruchungschlankerStbeimelastischen
BereichbeibeiderseitsgelenkigerLagerung,Luftfahrtforschung,10,551933
[189] Bolotin,V.V.,TheDynamicStabilityofElasticSystems,SanFrancisco:HoldenDay,1964
[190] Lindberg,H.E.undFlorence,A.L.,DynamicPulseBuckling:TheoryandExperiment,M.
Nijhoff,1987
[191] Simitses,G.J.,DynamicStabilityofSuddenlyLoadedStructures,Springer,1990
[192] AriGur,J.,Weller,T.undSinger,J.,ExperimentalStudiesofColumnsUnderAxialImpact,TAE
ReportNo.346TechnionIsraelInstituteofTechnology,AerospaceEngineering,1978
[193] AriGur,J.undSinger,J.,TheoreticalStudiesofColumnsUnderAxialImpactandExperimental
Verification,TAEReport377TechnionIsraelInstituteofTechnology,Departmentof
AeronauticalEngineering,1979
[194] AriGur,J.,Weller,T.undSinger,J.,Experimentalandtheoreticalstudiesofcolumnsunder
axialimpact,InternationalJournalofSolidsandStructures,18(7),619641,1982
[195] EvanIwanowski,R.M.,Ontheparametricresponseofstructures,APPLIEDMECHANICS
REVIEWS,18,699702,1965
[196] Budiansky,B.undHutchinson,J.W.,Dynamicbucklingofimperfectsensitivestructures,
AppliedMechanics,636657,1966
[197] Hutchinson,J.W.undBudiansky,B.,Dynamicbucklingestimates,AIAAJournal,4(3),525530,
1966
[198] Hoff,N.J.,DynamicStabilityofStructures,AGARDConferenceProceedings,2,15021536,1972
[199] Maymon,G.undLibai,A.,DynamicsandFailureofCylindricalShellsSunjectedtoAxial
Impact,AIAAJournal,15(11),16241630,1977
[200] Humphreys,J.S.undBodner,S.R.,Dynamicbucklingofshallowshellsunderimpulsive
loading,ASCEProceedingsJournaloftheEngineeringMechanicsDivision,88(EM2,Part
1),1736,1962
[201] Timoshenko,S.P.,berdieStabilittversteifterPlatten,DerEisenbau,12,147163,1921
[202] Timoshenko,S.P.undGere,J.M.,TheoryofElasticStability,McGrawHill,NewYork,1961
[203] Bleich,F.,BucklingStrengthofMetalStructures,McGrawHill,NewYork,1952
[204] Cox,H.L.,Thurston,F.R.undColeman,E.P.,CompressionTestonSevenPanelsofMonocoque
Construction,AeronauticcalResearchCouncil,2042,1945
56
ECSSEHB3224A
24March2010
[205] Seide,P.,TheEffectofLongitudinalStiffenersLocatedonOneSideofaPlateonthe
CompressiveBucklingStressofthePlateStiffenerCombination,NACATN2873,1953
[206] Barbr,R.,BeulspannungenvonRechteckplattenmitLngssteifenbeigleichmiger
Druckbeanspruchung,DerBauingenieur,17,268273,1936
[207] Klppel,K.undMller,H.,BeulwerteausgesteifterRechteckplatten,W.Ernstu.Sohn,Berlin,
1968
[208] H.Becker,A.Colao,R.GoldmanandPozerycki.CompressiveStrengthofShipHullGirders.H.
Becker,A.Colao,R.GoldmanandPozerycki1971,
[209] Faulkner,D.,CompressionTestsonWeldedEccentricallyStiffenedPlatePanels,SteelPlated
Structures,581617,1977
[210] Hetrakul,N.undYu,W.W.,ColdFormedSteelIBeamsSubjectedtoCombinedBendingand
WebCrippling,ThinWalledStructures,413426,1979
[211] H.ry,KonstruierenmitFaserverbundwerkstoffeninderRaumfahrt.InKonstruierenmit
VerbundundHybridwerkstoffen,VDIBerichte563.241270,VDIGesellschaft,Dsseldorf,
1985
[212] Weller,T.undSinger,J.,Experimentalstudiesonthebucklingunderaxialcompressionof
integrallystringerstiffenedcircularcylindricalshells,JournalofAppliedMechanics,44,721
730,1977
[213] Shanley,F.R.,Simplifiedanalysisofgeneralinstabilityofstiffenedshellsinpurebending,
JournalofAeronauticalScience,16,590592,1949
[214] Peterson,J.P.undDow,M.B.,Compressiontestsoncircularcylindersstiffenedlongitudinally
bycloselyspacedZsectionstringers,Memorandum21259L,NASA,1959
[215] Burns,A.B.undAlmroth,B.O.,Structuraloptimizationofaxiallycompressed,ringstringer
stiffenedcylinders,JournalofSpacecraft&Rockets,3(1),1925,1966
[216] Smith,G.W.undSpier,E.E.,Studyofstabilityofunpressurizedshellstructuresunderstatic
loading,GDCDDG67005,GeneralDynamics,SanDiego,1967
[217] Koiter,W.T.,Bucklingandpostbucklingbehaviorofacylindricalpanelunderaxial
compression,S.476,NationalAeronauticalResearchInstitute,Amsterdam,1956
[218] Stephens,W.B.,Imperfectionsensitivityofaxiallycompressedstringerreinforcedcylindrical
panelsunderinternalpressure,AIAAJournal,9,17131719,1971
[219] Koiter,W.T.undSkaloud,M.,Interventions,1962
[220] Hutchinson,J.W.undAmazigo,J.C.,Imperfectionsensitivityofeccentricallystiffened
cylindricalshells,AIAAJournal,5,392401,1967
[221] Tvergaard,V.,ImperfectionSensitivityofaWideIntegrallyStiffenedPanelunderCompression,
InternationalJournalofSolidsandStructures,9,177192,1973
[222] Tvergaard,V.,InfluenceofPostbucklingBehaviorinOptimumDesignofStiffenedPanels,
InternationalJournalofSolidsandStructures,9,15191534,1973
[223] Bushnell,D.,Computerizedbucklinganalysisofshells,MartinusNijhoffPublishers,1985
[224] Koiter,W.T.undPignataro,M.,AnAlternativeApproachtotheInteractionBetweenLocaland
OverallBucklinginStiffenedPanels,BucklingofStructures,133148,1974
[225] VanderNeut,A.,ModeInteractionwithStiffenedPanels,BucklingofStructures,117132,1976
[226] Thompson,J.M.T.,Tulk,J.D.undWalker,A.C.,Anexperimentalstudyofimperfection
sensitivityintheinteractivebucklingofstiffenedplates,BucklingofStructures,149159,1976
[227] Tvergaard,V.undNeedlemann,A.,ModeInteractioninanEccentricallyStiffenedElastic
PlasticPanelUnderCompression,BucklingofStructures,160171,1976
57
ECSSEHB3224A
24March2010
[228] Tvergaard,V.undNeedlemann,A.,Bucklingofeccentricallystiffenedelasticplasticpanelson
twosinglesupportsormultiplysupported,InternationalJournalofSolidsandStructures,11,
647663,1975
[229] Byskov,E.undHutchinson,J.W.,Modeinteractioninaxiallystiffenedcylindricalshells,AIAA
Journal,15(7),941948,1977
[230] Weller,T.undSinger,J.,Experimentalstudiesonthebucklingunderaxialcompressionof
integrallystringerstiffenedcircularcylindricalshells,JournalofAppliedMechanics,44,721
730,1977
[231] March,H.W.,BucklingofLongThinPlywoodCylindersinAxialCompression,ReportNo.
1322A,ForestProd.Lab.,1943
[232] Schnell,W.undBrhl,C.,DielngsgedrckteorthotropeKreiszylinderschalebeiInnendruck,
Journal,Heft7),201207,1959
[233] Gerard,G.,CompressiveStabilityofOrthotropicCylinders,JournalofAerospaceSciences,
11711189,1962
[234] Schulz,V.,ZurBeulstabilittanisotroperZylinderschalenausglasfserverstrktemKunststoff,
Bauingenieur,47(5),157163,1972
[235] Singer,J.,Arbocz,J.undBabcock,C.D.,BucklingofImperfectStiffenedCylindricalShellsunder
AxialCompression,AIAAJournal,9,6875,1971
[236] Thielemann,W.F.,NewDevelopmentsintheNonlinearTheoriesoftheBucklingofThin
CylindricalShells,AeronauticsandAstronautics,76120,1960
[237] Baker,E.H.,Kovalevski,L.andRish,F.L.,StructuralAnalysisofShells,ISBN0898741181,
McGrawHill,NewYork,1986
[238] Baruch,M.undSinger,J.,EffectofEccentricityofStiffenersontheGeneralInstabilityof
StiffenedCylindricalShellsunderHydrostaticPressure,JournalofMechanicalEngineering
Science,5(1),2327,1963
[239] Baruch,M.undSinger,J.,GeneralInstabilityofStiffenedCircularConicalShellsunder
HydrostaticPressure,AeronauticalQuarterly,16(2),187204,1965
[240] Baruch,M.,Singer,J.undWeller,T.,EffectofEccentricityofStiffenersontheGeneralInstability
ofCylindricalShellsunderTorsion,IsraelJournalofTechnology,4(1),144154,1966
[241] Block,D.L.,BucklingofEccentricallyStiffenedOrthotropicCylindersunderPureBending,TN
D3351,NASA,1966
[242] Burns,A.B.,StructuralOptimizationonAxiallyCompresserCylinders,ConsidderingRing
StringerEccentricityEffect,JournalofSpacecraft&Rockets,3(8),12631268,1966
[243] Bushnell,D.,ComputerizedAnalysisofShells,ComputersandStructures,18(3),471536,1984
[244] Card,M.F.undJones,R.M.,ExperimentalandTheoreticalResultsforBucklingofEccentrically
StiffenedCylinders,TND3639,NASA,1966
[245] Simitses,G.J.,BucklingofEccentricallyStiffenedCylindersunderTorsion,AIAAJournal,6(10),
18561860,1968
[246] Simitses,G.J.undCole,R.T.,GeneralInstabilityofEccentricallyStiffenedThinSphericalShells
underPressure,JournalofAppliedMechanics,37(4),11651168,1970
[247] Singer,J.,Baruch,M.undHarari,O.,InversionoftheEccentricityEffectinStiffenedCylindrical
ShellsBucklingunderExternalPressure,JournalofMechanicalEngineeringScience,8(4),363
373,1966
[248] Thielemann,W.undEsslinger,M.,berdenEinflussderExzentrizittvonLngssteifenaufdie
axialeBeullastdnnwandigerKreiszylinderschalen,DerStahlbau,34,3321965
[249] Geier,B.,DasBeulverhaltenversteifterZylinderschalen,Teil1:Differentialgleichungen,
ZeitschriftfrFlugwissenschaften,14(7),306323,1966
58
ECSSEHB3224A
24March2010
[250] Seggelke,P.undGeier,B.,DasBeulverhaltenversteifterZylinderschalen,Teil2:Beullasten,
ZeitschriftfrFlugwissenschaften,15(12),477490,1967
[251] Wittrick,W.H.,Aunifiedapproachtotheinitialbucklingofstiffenedpanelsincompression,
TheAeronauticalQuarterly,265283,1968
[252] Viswanathan,A.V.,Soong,T.C.undMiller,R.E.,Jr.,Bucklinganalysisforaxiallycompressed
flatplates,structuralsectionsandstiffenedplatesreinforcedwithlaminatedcomposites,CR
1887,NASA,1971
[253] Williams,F.W.,Computationofnaturalfrequenciesandinitialbucklingstressesofprismatic
plateassemblies,JournalofSoundandVibration,21(1),87106,1972
[254] Wittrick,W.H.undWilliams,F.W.,Bucklingandvibrationofanisotropicorisotropicplate
assembliesundercombinedloadings,InternationalJournalofMechanicalScience,16,209239,
1974
[255] VanderNeut,A.,TheGeneralInstabilityofStiffenedShellsunderAxialCompression,XIII,
Nat.Aero.Res.Inst.,Amsterdam,1947
[256] Singer,J.,Baruch,M.undHarari,O.,OntheStabilityofEccentricallyStiffenedCylindricalShells
underAxialCompression,InternationalJournalofSolidsandStructures,3,445470,1967
[257] Esslinger,M.,VersteifteundabgesetzteSchalen,Stabilittsuntersuchungenandnnwandigen
KreiszylinderschalenunteraxialsymmetrischerBelastung,ChemieIng.Techn.,45(11),770776,
1973
[258] J.SingerandM.Baruch.RecentStudiesonOptimisationforElasticStabilityofCylindricaland
ConicalShells,1966,
[259] Budiansky,B.undHutchinson,J.W.,ASurveyofSomeBucklingProblems,AIAAJournal,4,
15051510,1966
[260] Singer,J.undRosen,A.,Influenceofboundaryconditionsonthebucklingofstiffened
cylindricalshells,BucklingofStructures,ProceedingofIUTAMSymphosium,227250,
SpringerVerlag,HarvardUniversity,CambridgeMA,1974
[261] T.Weller,J.SingerundS.C.Batterman,Influenceofeccentricityofloadingonbucklingof
stringerstiffenedcylindricalshells.InThinShellStructures,Theory,ExperimentandDesign.
(Ed.Y.C.FungandE.E.Sechler)305324,1974
[262] Weller,T.,Combinedstiffeningandinplaneboundaryconditionseffectsonthebucklingof
circularcylindricalstiffenedshells,ComputersandStructures,9,116,1978
[263] Weller,T.,Combinedstiffeningandinplaneboundaryconditionseffectsonthebucklingof
circularcylindricalstiffenedshells,ComputersandStructures,9,116,1978
[264] Esslinger,M.undThielemann,W.,EinfluderRandbedingungenaufdieBeullastvon
Kreiszylinderschalen,DerStahlbau,33(12),1964
[265] Singer,J.undRosen,A.,TheInfluenceofBoundaryConditionsontheBucklingofStiffened
CylindricalShells,AFOSRTR722394,Technion,Haifa(Israel),1974
[266] Weller,T.,FurtherStudiesontheEffectofInPlaneBoundaryConditionsontheBucklingof
StiffenedCylindricalShells,AFOSRTR722394,Technion,Haifa(Israel),1974
[267] Singer,J.undAbramovich,H.,Vibrationtechniquesfordefinitionofpracticalboundary
conditionsinstiffenedshells,AIAAJournal,17(7),762769,1979
[268] Singer,J.,Vibrationandbucklingofimperfectstiffenedshellsrecentdevelopments,1983
[269] Singer,J.undAbramovich,H.,Vibrationtechniquesfordefinitionofpracticalboundary
conditionsinstiffenedshells,AIAAJournal,17(7),762769,1979
[270] Stuhlman,C.E.,DeLuzio,A.undAlmroth,B.,Influenceofstiffenereccentricityandendmoment
onstabilityofcylindersincompression,AIAAJournal,4,872877,1966
59
ECSSEHB3224A
24March2010
[271] Tennyson,R.C.,Theeffectofshapeimperfectionsandstiffeningonthebucklingofcircular
cylinders,BucklingofStructures,251273,1976
[272] A.LibaiandD.Durban.BucklingofCylindricalShellsSubjectedtoNonuniformAxialLoads.
2008,
[273] Sridharan,S.undPeng,M.H.,PerformanceofAxiallyCompressedStiffenedPanels,
InternationalJournalofSolidsandStructures,25(8),879899,1989
[274] A.VanderNeut.ModeInteractionwithStiffenedPanels.1976,
[275] VanderNeut,A.,OverallBucklingofZStiffenedPanelsinCompression,Collapse:The
BucklingofStructuresinTheoryandPractice,259268,CambridgeUniversityPress,1983
[276] VanderNeut,A.,Theinteractionoflocalbucklingandcolumnfailureofthinwalled
compressionmembers,TweelfthInternationalCongressofAppliedMechanics,389399,
Springer,Berlin,1969
[277] Thompson,J.M.T.undLewis,G.M.,Ontheoptimumdesignofthinwalledcompression
members,JournaloftheMechanicsandPhysicsofSolids,20,101109,1972
[278] Engesser,F.,berKnickfragen,SchweizerischeBauzeitung,26(4),2430,July1895
[279] Considere,A.,Resistancedespiecescomprimees,CongresInternationalDesProcedesde
Construction,1891
[280] vonKrmn,T.,DieKncikfestigkeitvonStben,PhysikalischeZeitschrift,9(4),136140,1908
[281] M.J.Sewell,ASurveyofPlasticBuckling.In.85197,H.Leipholzed.,1971
[282] Hill,R.,Ageneraltheoryofuniquenessandstabilityinelasticplasticsolids,
J.Mech.Phys.Solids,6(3),236249,1958
[283] Ramberg,W.undOsgood,W.R.,Descriptionofstressstraincurvesbythreeparameters,TN
902,NACA,1943
[284] Hutchinson,J.W.,PlasticBuckling,AdvancesinAppliedMechanics,(14),67144,1974
[285] Tvergaard,V.undNeedleman,A.,Onthefoundationsofplasticbuckling,Developmentsin
ThinWalledStructures,1,205233,1982
[286] Onat,E.T.undDrucker,D.C.,Inelasticinstabilityandincrementaltheoriesofplasticity,Journal
oftheAeronauticalSciences,20(3),181186,1953
[287] Hutchinson,J.W.undBudiansky,B.,AnalyticalandNumericalStudyoftheEffectsofInitial
ImperfectionsontheInelasticBucklingofaCruciformColumn,BucklingofStructures,Harvard
University,1974
[288] Batdorf,S.B.,Theoriesofplasticbuckling,JournaloftheAeronauticalSciences,20(3),405408,
1949
[289] Christoffersen,J.undHutchinson,J.W.,ClassofPhenomenologicalCornerTheoriesof
Plasticity,JournaloftheMechanicsandPhysicsofSolids,27(56),465487,1979
[290] Needleman,A.undTvergaard,V.,Aspectsofplasticpostbucklingbehavior,Mechanicsof
solids:TheRodneyHill60thanniversaryvolume.Oxford,PergamonPress,453498,1982
[291] Rubin,M.B.undBodner,S.R.,Anincrementalelasticviscoplastictheoryindicatingareduced
modulusfornonproportionalbuckling,InternationalJournalofSolidsandStructures,32(20),
29672987,1995
[292] vanderNeut,A.,TheGeneralInstabilityofStiffenedCylindricalShellsunderAxial
Compression,ReportS.314,NationaalLuchtvartlaboratorium,TheNetherlands,1946
[293] Esslinger,M.,BeulenundNachbeulenexzentrischversteifterdnnwandigerKreiszylinder
unteraxialsymmetrischerBelastung,FB7048,DLRBraunschweig,1970
60
ECSSEHB3224A
24March2010
[294] Arbocz,J.,PostBucklingBehaviourofStructuresNumericalTechniquesformore
ComplicatedStructures,in:LectureNotesinPhysics288,BucklingandPostBuckling,ISBN3
540183124,SpringerVerlagBerlinHeidelbergNewYork,pp.83142,1987
[295] Zimmermann,R.,OptimierungaxialgedrckterCFKZylinderschalen,FortschrittsberichteVDI,
Nr.207,1992
[296] MeyerPiening,H.R.,Farshad,M.,Geier,B.undZimmermann,R.,BucklingloadsofCFRP
compositecylindersundercombinedaxialandtorsionloadingexperimentsandcomputations,
CompositeStructures,53(4),427435,2001
[297] Geier,B.,MeyerPiening,H.R.undZimmermann,R.,Ontheinfluenceoflaminatestackingon
bucklingofcompositecylindricalshellssubjectedtoaxialcompression,CompositeStructures,
55(4),467474,2002
[298] Anonym,ESAHandbook,PSS031101,ESA,1992
[299] R.C.Tennyson,ThinWalledStructuresinAerospaceDesign.InDevelopmentsinThinWalled
Structures2.(Ed.J.RhodesandA.C.Walker)173210,ElsevierAppliedSciencePublishers,
Barking,1980
[300] R.C.TennysonundJ.S.Hansen,Optimumdesignforbucklingoflaminatedcylinders.In
Collapse:thebucklingstructuresintheoryandpractice.(Ed.J.M.T.ThompsonandG.W.
Hunt)409429,CambridgeUniversityPress,1983
[301] Tennyson,R.C.,Chan,K.H.undMuggeridge,D.B.,TheEffectofAxisymmetricShape
ImperfectionsontheBucklingofLaminatedAnisotropicCircularCylinders,Trans.Canadian
AeronauticsandSpaceInstitute,4(2),131139,1971
[302] Tennyson,R.C.,Muggeridge,D.B.,Chan,K.H.undKhot,N.S.,BucklingofFiberReinforced
CircularCylindersUnderAxialCompression,AFFDLTR72102,AirForceFlightDynamics
Laboratory,WrightPattersonAirForceBase,
[303] Khot,N.S.,Bucklingandpostbucklingbehaviourofcompositecylindricalshellsunderaxial
compression,AIAAJournal,8(2),229235,1970
[304] Khot,N.S.,Ontheinfluenceofinitialgeometricimperfectionsonthebucklingand
postbucklingbehaviouroffiberreinforcedcylindricalshellsunderuniformaxialcompression,
TechnicalReportAFFDLTR68136,AirForce
[305] Khot,N.S.,Ontheeffectsoffiberorientationandnonhomogenityonbucklingand
postbucklingequilibriumbehaviouroffiberreinforcedcylindricalshellsunderuniformaxial
compression,TechnicalReportAFFDLTR6819,
[306] Khot,N.S.undVenkayya,V.B.,Effectoffiberorientationoninitialpostbucklingbehaviour
andimperfectionsensitivityofcompositecylindricalshells,TechnicalReportAFFDLTR70
125,AirForceFlightDynamicsLabora
[307] Almroth,B.O.,DesignofCompositeMaterialStructuresAnEvaluationoftheStateoftheArt,
AFWALTR813102,LockheedMissiles&SpaceCompany,PaloAlto,California,1981
[308] Hilburger,M.H.undStarnes,H.S.Jr.,EffectsofImperfectionsontheBucklingResponseof
CompressionloadedCompositeShells,InternationalJournalofNonlinearMechanics,37,623
643,2002
[309] Arbocz,J.,Theimperfectionsdatabank,ameantoobtainrealisticbucklingloads,Bucklingof
Shells:AStateoftheArtColloquium,535567,SpringerVerlag,Berlin,1982
[310] Arbocz,J.undHol,J.M.A.M.,CollapseofAxiallyCompressedCylindricalShellswithRandom
Imperfections,ThinWalledStructures,23(14),131158,1995
[311] Esong,I.E.,Elghazouli,A.Y.undChryssanthopoulos,M.K.,Measurementtechniquesfor
bucklingsensitivecompositeshells,Strain,1998
61
ECSSEHB3224A
24March2010
[312] Arbocz,J.undAbramovich,H.,TheinitialimperfectiondatabakattheDelftUniversityof
TechnologyPartI,ReportLR290,DelftUniversityofTechnology,TheNetherlands,1979
[313] Arbocz,J.,Theeffectofimperfectboundaryconditionsonthecollapsebehaviorofanisotropic
shells,InternationalJournalofSolidsandStructures,37,68916915,2000
[314] Arbocz,J.,Starnes,H.S.Jr.undNemeth,M.P.,AHierarchialApproachtoBucklingLoad
Calculations,AIAAPaperNo.991232,Proceedingsofthe40thAIAA/ASME/ASCE/Structures,
StructuralDynamicsandMaterialsConference,St.Louis,1999
[315] Elghazouli,A.Y.,Esong,I.E.undChryssanthopoulos,M.K.,BucklingTestsonGFRPCylinders
underConcentricandEccentricCompression,TestSeriesI:Rovimat1200,CESLICReport
OR10,EngineeringStructuresLaboratories,ImperialCollege,
[316] Elghazouli,A.Y.,Spagnoli,A.undChryssanthopoulos,M.K.,BucklingTestsonGFRP
Cylinders,TestSeriesII:DF1400,CESLICReportOR11,EngineeringStructuresLaboratories,
CivilEngineeringDepartment,ImperialCollege,
[317] Elghazouli,A.Y.,Chryssanthopoulos,M.K.undSpagnoli,A.,Experimentalresponseofglass
reinforcedplasticcylindersunderaxialcompression,MarineStructures,11(9),347371,1999
[318] Spagnoli,A.,Elghazouli,A.Y.undChryssanthopoulos,M.K.,Numericalsimulationofglass
reinforcedplasticcylindersunderaxialcompression,MarineStructures,14(3),353374,2001
[319] Chryssanthopoulos,M.K.,Elghazouli,A.Y.undEsong,I.E.,ValidationofFEmodelsfor
bucklinganalysisofwovenGFRPshells,CompositeStructures,49(4),355367,2000
[320] Arbocz,J.undStarnesJr,J.H.,Futuredirectionsandchallengesinshellstabilityanalysis,Thin
WalledStructures,40(9),729754,2002
[321] Chryssanthopoulos,M.K.undPoggi,C.,StochasticImperfectionModellinginShellBuckling
Studies,ThinWalledStructures,23(14),179200,1995
[322] Elishakoff,I.,vanManen,S.,Vermeulen,P.G.undArbocz,J.,Firstordersecondmoment
analysisofthebucklingofshellswithrandomimperfections,AIAAJournal,25(8),11131117,
1987
[323] Hilburger,M.H.,Nemeth,M.P.undStarnesJr,J.H.,ShellBuckligDesignCriteriaBasedon
ManufacturingImperfectionSignatures,AIAAJournal,44(3),654663,2006
[324] Zimmermann,R.,MehrzieloptimierungbeulgefhrdeterCFKZylinderschalenmit
Imperfektionen,ZeitschriftfrangewandteMathematikundMechanik,75,605606,1995
[325] Hhne,C.,EinverbesserterAnsatzzurBestimmungderMaterialparameterdnnwandiger,
stabilittsgefhrdeterStrukturenausFaserverbundwerkstoff,IB1312004/44,Deutsches
ZentrumfrLuftundRaumfahrt(DLR),Braunschweig.
[326] Hhne,C.,Rolfes,R.,Breitbach,E.undTessmer,J.,RobustDesignofCompositeCylindrical
ShellsunderAxialCompressionSimulationandValidation,ThinWalledStructures,46,947
962,2008
[327] Jarlas,R.,Predictinginitiationofdelaminationgrowthwith3Dfiniteelementapproach,
Unknown,2000
[328] Nilsson,K.F.,Asp,L.E.,Alpman,J.E.undNystedt,L.,Delaminationbucklingandgrowthfor
delaminationsatdifferentdepthinaslendercompositepanel,InternationalJournalofSolids
andStructures,38,30393071,2001
[329] Wagner,W.,Gruttmann,F.undSprenger,W.,AFiniteElementFormulationfortheSimulation
ofPropagatingDelaminationsinLayeredCompositeStructures,Int.J.Num.Meth.Engng,1337
1359,2001
[330] Wagner,W.undGruttmann,F.,DelaminationGrowthAnalysisofCompositePanels,
submittetdtoThinWalledStructures,Unknown,2003
62
ECSSEHB3224A
24March2010
[331] Rajbhandari,S.P.,Scott,M.L.,Thomson,R.S.undHachenberg,D.,AnApproachtoModelling
andPredictingImpactDamageinCompositeStructures,2002
[332] Thomson,R.S.undScott,M.L.,ModellingDelaminationsinPostbucklingCompositeShear
Panels,ComputationMechanics,26(1),7589,2000
[333] Sjrgen,A.,Krasnikos,A.undVarna,J.,Experimentaldeterminationofelasticpropertiesof
impactdamageincarbonfibre/epoxylaminates,Composites:PartA,32,12371242,2001
[334] Yap,J.W.H.,Scott,M.L.,Thomson,R.S.undHachenberg,D.,TheAnalysisofSkintoStiffener
DebondinginCompositeAerospaceStructures,CompositeStructures,57(14),425435,2002
[335] Weller,T.,Messer,G.undLibai,A.,RepeatedBucklingofGraphiteEpoxyShearPanelswith
BondedMetalStiffeners,TAEReport546,Technion,Haifa(Israel),1984
[336] Weller,T.undSinger,J.,DurabilityofStiffenedCompositePanelsunderRepeatedBuckling,
InternationalJournalofSolidsandStructures,26(9/10),10371069,1990
[337] Frostig,Y.,Siton,G.undSegal,A.,PostbucklingBehaviorofLaminatedCompositeStiffeners
andStiffenedPanelsunderCyclicLoading,AIAAJournalofAircraft,28(7),471480,1991
[338] Hambly,E.T.,BucklingExperimentsonDamagedCylindricalShells,InternationalJournalof
SolidsandStructures,33,35391996
[339] Zimmermann,R.undRolfes,R.,POSICOSSimprovedpostbucklingsimulationfordesignof
fibrecompositestiffenedfuselagestructures,CompositeStructures,73(2),171174,2006
[340] Degenhardt,R.,Rolfes,R.,Zimmermann,R.undRohwer,K.,COCOMATimprovedmaterial
exploitationofcompositeairframestructuresbyaccuratesimulationofpostbucklingand
collapse,CompositeStructures,73(2),175178,2006
4.12 Abbreviated Symbols

ThefollowingabbreviatedsymbolsaredefinedandusedwithinthisChapter:
Symbol Meaning
A12 + A33 2
s Shellparameter=
A11 A22
D12 + 2 D33
p Plateparameter=
D11 D22
D11 A11
mainstiffnessratio=
D22 A22
WhereAij,Dijarestiffnesscoefficientsfororthtropicwallconstruction(seeSection17.2.2)
63
ECSSEHB3224A
24March2010
5
Types of instability and failure behaviour of
typical structural elements
5.1 Introduction
Allstructuraldesignersknowthatthedimensioningoftheirstructureincaseofstaticloadingis
basedonseveralbasiccriteria,namely:
a. strengthcriteria,whichareusedtoverifythatstressescausedbythespecifieddesignloadcases
donotexceedtheircorrespondingstrengthdesignallowables;
b. the stiffness criterion, which specifies design limits for deflections or strains or natural
frequencies for the different design load cases in order not to hinder proper operation or to
avoid undesirable and potentially dangerous behaviour such as flutter or mechanical
vibrations.
What often is overlooked is that by carrying out the usual stress and deformation analysis with the
manyeasilyavailablefiniteelementcodesoneobtainsnoinformationastothestabilitybehaviourof
thestructure.Itisbynowwellknownthatthinwalledslenderstructures,orstructureswhichcontain
slendermemberssubjecttocompressiveorshearstresses,caninitiallyfailinoneofthemanypossible
instability modes, which in turn may significantly affect the strength or stiffness behaviour of the
whole structure. This is especially true for the current trends in design where with the use of
structural optimization techniques one is often producing highly stressed structures of very slender
proportions.
Withthesuddenandoftenunexpectedoccurrenceofpartialortotalstructuralfailureduetodifferent
forms of (at least initially) elastic instabilities, in the past, one relied on socalled buckling tests to
providethedataforthedevelopmentofsafeandreliabledesignrecommendations.
Currently, often numerical simulations are used in place of buckling tests. However, meaningful
numericalsimulationsonbucklingofstructurescanonlybecarriedoutbyananalystwithathorough
understandingofthebasicphenomenaofstructuralinstabilityandawarenessofthedifferenttypesof
buckling behaviour that can occur during the loading process of a structure. In the following, the
occurrenceofdifferenttypesofinstabilitieswillbediscussedonhandofrelativelysimpleexamples.
5.2 Elastic buckling of columns
5.2.1 Overview
The problem of a slender, perfectly straight, centrally compressed column, built in vertically at the
baseandfreeattheupperend(seeFigure51a)hasbeenfirstsolvedbyLeonardEulerin1744[1].He
foundasthesmallestcriticalload
64
ECSSEHB3224A
24March2010
2 EI
Pcr = 51
4 L2
where E is Youngs modulus, I is the minimal moment of inertia of the crosssection and the
correspondingbucklingmodeisshowninFigure51b.Eulerhasassumedinhisworkthatthecross
section of the column does not distort during buckling and failure and that the wavelength of the
bucklingmodeisoftheorderofthecolumnlength.
Theperfectcolumnassumptionisunrealistic.Usinganinitialimperfectionoftheform
x
w o (x) = W01 sin 52
L
andalargedeflectiontheoryRivello[2]hasobtainedtheresultsshowninFigure52.Fromthisfigure
one can draw the conclusion that the straight position is the only equilibrium configuration for a
columnwithvanishinglysmallimperfectionsuntil P = Pc .Closetoandat P = Pc thedeflectionsofa
columnwithvanishinglysmallimperfectionsgrowveryfastandareapproximatelygivenbyequation
(14.56) of [2] until on the compression loaded side the stresses in the extreme fibre exceed the
proportionallimit.AscanbeseenfromFigure52alsocolumnswithmeasurableimperfectionsdonot
bend appreciably until P is very nearly equal to Pcr . Due to these rapidly increasing bending
deformations the stresses on the compression loaded side soon exceed the yield strength and in
practicalapplicationscollapseofslendercolumnsoccursatPslightlybelowbutcloseto Pcr .
ThedottedcurvesinFigure52,representingtheyieldstrengthlimits,werealsoobtainedbyRivello
[2] for a column with homogeneous crosssectionusing an idealized linearly elastic perfectly plastic
material behaviour. The dotted curves were computed for an idealized H section made of 7075T6
aluminium alloy, whereby it was assumed that the web has negligible resistance in bending and
extensionbutisrigidinshear.Ithasbeenshownin[3]thatifoneconsiderseccentricallyappliedaxial
loading in place of geometric initial imperfections, one obtains curves similar to the ones shown in
Figure 52. The instability theory of Euler accurately describes the buckling behaviour of slender
columns with solid or thickwalled cross sections. To obtain a direct measure of slenderness it is
customarytorewriteEulersformulaas
P E
cr = cr = 2 53
A (L / )2
where = radiusofgyrationofthecrosssection (= I / A ) .Experimentalevidenceindicatesthatfor

valuesoftheslendernessration L / > 80 Eulersformulapredictsthebucklingloadofcolumnsquite
accurately. For values of the slenderness ratio 20 < L / < 80 one can get reasonably accurate
predictionsbyusingShanleystangentmodulus theory[4],whichessentiallyconsistsofreplacingin
equation (53) the modulus of Elasticity E by the tangent modulus E t . Finally, for values of the
slenderness ratio L / < 20 failure occurs mainly by plastic crushing of the crosssection and cr is
equaltothecompressivestrengthofthematerial.Formetalsoneusuallyuses cr = cy ,where cy
isthecompressiveyieldstrengthofthematerial.
For thinwalled columnsEulers assumptions thatthe crosssection does notdistort during buckling
and that the wavelength of the buckle is of the order of the column length should be reexamined.
Suchcolumnscanbethoughtofasanassemblageofthinplates.
65
ECSSEHB3224A
24March2010
Figure51:Eulersproblem
Figure52:Nonlinearbehaviourofperfectandimperfectcolumns
5.2.2 Buckling of columns with compound cross-sections

The different failure modes that can occur with a thinwalled column of varying length can best be
illustratedbyconsideringthelippedchannelsectionshowninFigure53.
66
ECSSEHB3224A
24March2010
Figure53:Alippedchannelsection
Assumingthatthelateraldeflectionofthecrosssectionfromthelineofactionofthecompressiveload
variessinusoidallyalongthelengthofthecolumn,thensufficientlylongcolumnswillbuckleinglobal
or overall modes, where the half wavelength l of the sinusoidal buckle is equal to the length of a
simplysupportedcolumn.
Considering Figure 54 one sees that, depending on the value of the slenderness parameter l / b
global buckling can take the form of a flexural mode (Euler mode, l / b 50) , in which the cross
section translates but does not rotate. Also possible is the form of a torsional mode (l / b = 10) , in
which the crosssection rotates but does not translate. There is also a third global mode, called the
torsionalflexuralmode,inwhichthesectionbothrotatesandtranslates.
In most applications of thinwalled open sections, there exists at least one axis of symmetry, as
illustrated in Figure 53. For torsionalflexural instability of such cases, for example in [5], the
followingcharacteristicequationhasbeenderived
{ }
(P Pz ) I0 / A) (P Py ) (P P ) P2 y 20 = 0 54
where
2EIy 2EIz 2E
Py = ; Pz = ; P = (A / I0 ) + GJ 55
L2 L2 L2

and
I0 = Iy + Iz + A(y 20 + z20 )
=WagnerTorsionBendingConstant
J =TorsionalConstant
y 0 , z0 =coordinatesoftheShearCentreS.C.
If Pz is the smallest of the three roots of equation (54), the column will buckle in pure bending.
Otherwise,thebucklingwillbecombinedbendingandtwisting.
Ifthecrosssectionhastwoaxesofsymmetry, y 0 = z0 = 0 andequation(54)simplifiestotheform
(P Pz ) (P Py ) (P P ) = 0 56
In this case the three roots are Py , Pz and P and the column will buckle in pure bending or pure
twisting,dependingonwhichofthethreerootsisthesmallest.
AlsoillustratedinFigure54arecasesoflocalinstabilityinwhichthecrosssectiondistortswithout
translationorrotation.Thus,whentheslendernessratio l / b = 0.75 onegetsalocalbucklingmode,
67
ECSSEHB3224A
24March2010
whereallthecornersbetweentheplateelementsremainstraightwhilethecentresoftheplatesdeflect
outofplaneasshown.Interestinglyenough,whentheslendernessratio l / b = 4.0 onegetsanother
formoflocalbuckling,whereonlytwoofthecornersbetweentheplateelementsremainstraight.This
type of buckling is called flange buckling or distortional buckling. For a detailed discussion of the
theoryofbucklingofthinwalledcolumnstheinterestedreadershould consultV.Z. Vlasovsseminal
contributionfirstpublishedinRussianin1959[7].
Figure54:Bucklingbehaviourofthinwalledcolumns(from[6])
Testonshort,thinwalledcolumnsshowthatoften,afterlocalbucklinghasoccurred,thecolumnsstill
have the ability to carry a greater load before general failure takes place. Further, it appears that in
cases where local buckling occurs at relatively low stress levels, the stresses at general failure (or
crippling)willbenoticeablyhigher.Ontheotherhand,iflocalbucklingtakesplaceatrelativelyhigh
meanstresslevels(say,at 0.7 cy )thenthecriticalbucklingandthecripplingstressesarepractically
thesame.Figure55displaysthestressdistributiononathinwalledcrosssectionafterlocalbuckling
has occurred but prior to crippling or failure. Bending deflections become large after the flanges
buckle, and crippling occurs when the stresses at the supported (essentially straight) edges of the
flangesreachthecompressiveyieldstress cy .
The nonlinear behaviour associated with large displacements and plasticity has prevented the
developmentofasatisfactoryanalyticalsolutionforthecripplingstress crip ofarbitrarythinwalled
crosssections.HencesemiempiricalformulasareusedwhicharediscussedinChapter6ofRef.[8].
68
ECSSEHB3224A
24March2010
Figure55:Stressdistributionafterlocalbuckling
5.3 Buckling of thin plates
5.3.1 Overview
Thebucklingload ofasimplysupportedrectangularthinflatplateofwidthbandlengtha,subjected
to a uniform compressive force per unit length N = h on the edges x = 0 and x = a while the
boundaries y = 0 and y = b areunrestrainedagainstinplanemotion(seeFigure56)has firstbeen
derivedbyG.H.Bryanin1891[9].Usingthedeflectionmodeshape
mx ny
w = Wmn sin sin where m,n = 1,2,... 57
a b
Figure56:Platesubjectedtoinplanecompressiveloading
hefoundasthesmallestcriticalstress
2
2E h
c = k 58
2
12(1 ) b
wheretheplatebucklingfactor k istheminimumvalueof
69
ECSSEHB3224A
24March2010
2
mb a
k mn = + n2 59
a mb
obtainedforagivenplateaspectrationa/bbyproperselectionoftheintegerwavenumbersmandn.
From equation (59) it is obvious that the minimum value of k mn occurs when n = 1 . To minimize
equation(59)withrespecttom,oneplots k asafunctionofa/bfordifferentvaluesofm,asshownin
Figure 57. The minimum value of k , which is then used in equation (58) is given by the lower
envelopeofthecurves,indicatedinFigure57bythesolidline.
Figure57:Compressivebucklingcoefficientsforsimplysupportedplates
The buckling loads of uniform rectangular plates under constant normal edge forces have been
determined for various boundary conditions either by solving the appropriatedifferential equations
or by using the RayleighRitz method. In these simple cases the inplane section forces equal the
appliededgesforcesandthebucklingstresscanbecalculatedfromequation(58).However,thevalue
ofthe bucklingcoefficient k dependsuponthe typeofloadingandtheedgerestraints.Resultsfrom
Figure 14 of [10] are shown in Figure 58 and give the values of k as a function of a/b for uniaxial
compressionwithvariouscombinationsofsimplysupported,clampedandfreeedges.
Noticethat k isessentiallyindependentoftherestraintattheloadededgeswhen a / b > 3 .However,
inthesecases k dependsstronglyupontherestraintoftheunloadededges.The buckling factor k is
nearlyconstantforlongplates (a / b > 3) .
Itisnaturallyunrealistictoassumethattheplateisperfectlyflat.Usingalargedeflectiontheoryand
aninitialimperfectionoftheform
x y
w 0 (x, y) = W11 sin sin 510
a b
Coan [11] obtained solutions for the buckling and the postbuckling behaviour of rectangular plates
withuniformlydisplacedloadededgesandeitherundistortedorstressfreeunloadededges.Ascan
be seen from Figure 59a, in the postbuckling region the axial compressive stress x is no longer
uniformlydistributedovertheloadededgesasitisbeforebucklingoccurs.Insteadithasamaximum
valueatthesimplysupportedunloadededgesthatareheldstraight.Ofconsiderable importanceare
theinplanestresses y thatariseinthepostbucklingregion.Noticethatinthecentralregionofthe
plate the y stresses are tensile in character and thus they stiffen the plate considerably against
furtherlateral deflection.Thesemembranestressestogetherwiththefactthattheunloadededgesare
70
ECSSEHB3224A
24March2010
restrainedagainstoutofplanedeflectionexplainwhytheplate,unlikethecolumn(wherethereare
no such middle surface forces), can carry axial loads that are much higher than the buckling load.
Noticefurtherthattherearenoresultantforcesintheydirectionthustheunloadededgesarefreeto
moveuniformly.Ontheotherhand,ascanbeseenfromFigure59b,iftheunloadededgesofaplate
are traction free then they deform, that is a contraction occurs at the central region. The absence of
membraneforcesintheydirectionaccountsforthefactthatsuchaplatecarriessmallerpostbuckling
loadsthanthoseofaplatewithstraightunloadededges.
Figure58:Compressivebucklingfactorforrectangularplateswithvarious
boundaryconditions(from[10])
Figure59:Postbucklingstressdistributionsforplateswithuniformlydisplaced
loadededges(from[11])
Thebending(outofplane)deformationatthecentreofperfectandimperfectsquareplatessubjected
touniformendshorteningareshowninFigure510.Comparingthesecurveswiththecorresponding
columncurvesofFigure52oneseesthat,unlikeforcolumns,forplatessizeablepostbucklingstresses
are possible. Notice that following buckling, the stiffness of the plate decreases; however, failure
occurs only when x at the unloaded edges reaches the yield strength of the material used. The
71
ECSSEHB3224A
24March2010
bucklingandpostbucklingbehaviourofplatessubjectedtoshearloadingisdiscussedinRef.[10]in
greatdetail.
Figure510:Nonlinearbehaviourofperfectandimperfectplates(from[12])
5.3.2 Buckling behaviour of stiffened panels

Thediscussionthatfollowsisapplicabletobothplaneandcurvedstiffenedpanels(seealsoRef.[13]).
Stiffened panels exhibit various buckling and failure modes. There are modes, usually with short
waves,whichinvolvetheskin,andleavethestiffenersessentiallyundeformed;bladestiffenershavea
tendency towardslateral buckling,a longwave mode; wallsinother kinds of stiffeners may buckle
into short ripples; or parts of stringers may buckle together with the skin in a common shortwave
length.Ingeneral,apanelbuckledintooneofthesemodescanbeloadedbeyondthecorresponding
bucklingload.Therewillbeadegradationoftheeffectivestiffnessofthebuckledpartsduetotheout
ofplanedeflections.Asaconsequencethestressesequilibratingtheadditionalloadwillbedistributed
in a different way, as compared to the fundamental state. The limit of the load carrying capacity is
likely to be attained with buckling into a mode involving both skin and stiffeners, a type which
frequentlyiscalledoverallorglobalbuckling,seeFigure511.
72
ECSSEHB3224A
24March2010
Figure511:Bucklingbehaviourofstiffenedpanels
Twomorefailuremodesmaylimitthestrengthofstiffenedpanelspriortooverallbuckling:
a. thestressesinthebuckledpartmayexceedthestrengthofthematerial,
b. in nonintegrally stiffened plates the stiffeners might separate from the skin due to high peel
andshearstressesinthestringer/skinjoint.
Finally, a remark is considered appropriate concerning stiffened panels which were designed
according to the simultaneous mode philosophy. In this design philosophy it is supposed that an
optimumisachievedifseveralbucklingmodesoccuratthesame(primary)bifurcationload.Ithasto
beexpectedthatthesepanelsfailinacatastrophicwaybeforetheenvisagedbucklingloadisreached,
since mode coupling increases the sensitiveness to initial imperfections [14], [15]. In terms of
probabilistic, the joint failure probability of the combined acting modes is higher than that for each
singlemode.
5.3.3 Buckling behaviour of sandwich plates

Structural sandwich constructions are formed by bonding two thin face sheets to a comparatively
thickcoreasdepictedinFigure512.Thesefacingsprovidepracticallyalltheoverallbendingandin
planeextensionalrigiditytothesandwich.Thecoreservestopositionthefacesatlocationsremoved
from the neutral axis, provides virtually all the transverse shear rigidity of the sandwich, and
stabilizesthefacingsagainstlocalbuckling.
73
ECSSEHB3224A
24March2010
Figure512:Sandwichplateconstruction
Sandwichplatesmayfailineither:
a. anOverallModeofBucklingsimilartothebucklingofunstiffenedplatesshowninFigure513
or
Figure513:Overallbucklingmodeofasandwichplate
b. avarietyoflocalmodesasillustratedinFigure514andasdescribedbelow.
1. Intracellular Buckling (Face Dimpling) This is a localized mode of instability which
occurs only when the core is NOT continuous. As depicted in Figure 514(A), in the
regionsdirectlyabovecorecells(suchasthoseofahoneycombcore),thefacingsbuckle
inplatelikefashionwiththecellwallsactingasedgesupports.Theprogressivegrowth
of these buckles can eventually precipitate the buckling mode identified below as face
wrinkling.
2. FaceWrinklingThisisalocalizedmodeofinstabilitywhichmanifestsitselfintheform
ofshortwavelengthbucklesinthefacings,whicharemostlyconfinedtoindividualcells
ofcellulartypecores,andinvolvesthetransverse(normaltothefacings)strainingofthe
corematerial.AsshowninFigure514(B),oneshouldconsiderthepossibleoccurrenceof
wrinkleswhichmaybeeithersymmetricalorantisymmetricalwithrespecttothemiddle
surfaceoftheoriginalundeformedsandwich.
As shown in Figure 515, final failure from wrinkling will usually result either from
crushingthecore,ortensileruptureofthecoretofacingbond,ortensileruptureofthe
coreproper.
3. Shear Crimping Shear crimping is often referred to as a local mode of failure but is
actually a special form of general instability for which the buckle wavelength is very
short due to the low transverse shear modulus for the core. This phenomenon occurs
quite suddenly and usually causes the core to fail in shear, it may also cause a shear
failureinthecoretofacingbond.
74
ECSSEHB3224A
24March2010
Figure514:Localizedinstabilitymodes
Figure515:UltimatefailuresprecipitatedbyFaceWrinkling
Unified treatments from which both the overall and the local modes of buckling can be calculated
havebeendeveloped(e.g.Ref.[16].Acomprehensivereviewandcompilationofanalysisdata,which
serveasadetailedstateoftheartassessment,arecontainedinRef.[17].
5.4 Instability of axially compressed cylindrical panels

Koiterin1956[13]performedanasymptoticpostbucklinganalysisofnarrow,longcylindricalpanels
under axial compression where the longitudinal edges are simply supported. He found that if the
panelissufficientlynarrow(thatis 1 )thebifurcationstressisgivenby
2Eh2
cr = (1 + 4 ) 511
3(1 2 )b
where
75
ECSSEHB3224A
24March2010
1 4 b
= 12(1 2 ) 512
2 Rh
isameasureforthetotalcurvatureofthepanel.Thegeometricparametersinequations(511)and(5
12) are shown in Figure 516, with simple support conditions presumed to exist along adjacent
stringers.Figure517displaysloadendshorteningcurvesforpanelswithvariousvaluesofthetotal
curvature .Noticethatinthisfigure representsthepostbifurcationslope.Theslopefor = 0
correspondstothatofaflatplate,andas increasesthepostbifurcationslopedecreases.
Figure516:Curvedpanelnomenclature.Panelisconsideredlongandsimply
supportedalongstraightedgeswherestringersexist
Figure517:Tangentstopostbucklingcurveforseveralvaluesoftotalcurvature
(fromKoiter[13])
Criticalloadsonpanelswith > 0.64 maybesensitivetoinitialimperfections,withthissensitivity

increasingtowardthevalue foracompletecylindricalshellas increases,asshowninFigure518a
and Figure 518b. These plots correspond to panels with initial imperfections in the form of the
classicalbucklingmodewithimperfectionamplitudeequalto10%oftheshellwallthickness.
76
ECSSEHB3224A
24March2010
Figure518:Loadvs.endshorteningcurvesfor=0.7and=1.0(fromBrushand
Almroth[5])
Actually, a noticeable sensitivity to imperfections can only be determined from a full postbuckling
analysis.Especiallywithstiffenedcurvedpanelsthepostbucklingpathoftenrisesagainafteronlya
small load drop at the bifurcation point. In this case the structure cannot be considered as
imperfection sensitive despite of the negative slope at the bifurcation point (see section 5.5.2, last
paragraph).
5.5 Structural behaviour of thin shells
5.5.1 Overview
Thinwalled structures, like panels and shells, unstiffened or stiffened, are important elements for
lightweightairframesandspacecraftstructures.Inthefollowingthetermshellwillbeusedforall
typesofsuchtwodimensionalstructures.Frequentlytheshellsfailbybucklingunderloadscausing
compressiveorshearstresses.
Depending on size, support conditions and material properties shells may reach their maximum
strength at initial buckling, or will fail in the postbuckling regime after having undergone
considerable outofplane deflections. The analysis of both initial buckling, and postbuckling
behaviour,shouldbebasedonnonlinearkinematicsofdeformation,i.e.onageometricallynonlinear
theory of shells. Uncertainties in defining constitutive laws, e.g. for laminated composite or closely
stiffened shells, and difficulties in establishing nonlinear shell theories, as well as in solving their
equations,necessitatecarefulcheckingoftheoreticalresultsagainstcarefullyevaluated testresultsas
experimentalevidence.Confidenceindesignverificationcanonlybeachievedinthisway.
5.5.2 Buckling of cylindrical shells

SometypicalfeaturesofthebucklingbehaviourofshellsareillustratedinFigure519.Thediagram
showstheappliedloadasafunctionofthedisplacementinthedirectionoftheload,atthepointof
load application. For a longitudinally compressed cylindrical shell it can be thought of as a plot of
loadvs.enddisplacement.Thelinesinthediagramrepresentequilibriumstates.
77
ECSSEHB3224A
24March2010
Attheverybeginningofloadingtheresponseislinearlyelastic,butathigherloadsadeviationfrom
linearity may be observed. At a certain level another solution branches off while (in theory) the
primary solution, termed the fundamental state, can be continued to higher loads where more
branchingsolutionsmaybedetected.Thepointsatwhichothersolutionsbranchoffthefundamental
statearecalledbifurcationpointsandtheloadatthelowestbifurcationpointiscommonlycalledthe
bucklingload.
Figure519:Bucklingbehaviourofaxiallycompressedcylindricalshells
Thefundamentalstatesbeyondthelowestbifurcationpointareunstable.Hence,theactualcurveof
load vs. end displacement will follow the bifurcation solution path where increasing outofplane
displacementswilldevelopintheshell.Eventuallyasecondarybifurcationpointmaybemetmeaning
that the deformation mode will change again more or less rapidly. Often the postbuckling state
bifurcatingatthebucklingloadisalsounstable.Inthiscasetheshellwillallofasuddenjumpintoa
stableequilibriumstateinadeeplybuckledconfiguration,or,asaconsequenceoflargedeformations,
itwillbreakdowncompletely.
Inrealityanyshellwillbeslightlyimperfectingeometryorloading.Theimperfectionswillcausethe
behaviourtodeviatefromthetheoreticalpathoftheperfectshell,andtofollowaneighbouringcurve
thatpassesthebifurcationstatesgradually(dashedline).
It is worth mentioning that upon following the fundamental path a maximum of the load can be
attained without meeting bifurcation points. That maximum is called a limit point. Encountering
imperfectionsintheformofthefirstbifurcationmodemayenforcelimitpointbehaviour.
The behaviour discussedabove is typical of the multitude of solutions of nonlinear boundary value
problems.Asarule,thefundamentalsolutionsexhibitcertainregularities:
a. For shells of revolution subjected to axisymmetric loads the displacements and stress
distributionswillbeaxisymmetric,too.
b. Doublysymmetricshellswilldeforminadoublysymmetricway.
c. Flatpanelsunderinplaneloadswillnotexperienceoutofplanedeflections.
78
ECSSEHB3224A
24March2010
The bifurcation solutions break the regularity of the fundamental solution and bring about
considerableoutofplanedisplacementstotheeffectthatthechangeofdimensionduetocompressive
loadsisachievedpartlybygeometricshortening.
The bifurcation points of the fundamental solution are found upon solving a homogeneous linear
boundaryvalueproblemforthebifurcationmode.Itscoefficientsdependonthefundamentalsolution
and,hence,ontheappliedload.Solutionsarepossibleonlyforcertaindiscretevaluesoftheload.In
otherwords:Thebifurcationpointsaredeterminedbyeigenvalueproblems.
Ingeneralthecoefficientsoftheequationsgoverningtheeigenvalueproblemarevariablealongthe
shell surfaceand, moreover, dependin a nonlinearway on themagnitude of the applied load. This
typeofproblemisoftentermedanonlineareigenvalueproblem.Inordertoescapethedifficulties
involvedinthesolutionofnonlineareigenvalueproblemscertainsimplificationsarefrequentlymade:
Often it is justified to assume that stresses and displacements in the fundamental state grow in
proportiontotheload.Thisassumptionhastheeffectofrenderingtheeigenvalueproblemlinear.The
rapidvariationofthecoefficientsovertheshellsurfacearestilltobetakenintoaccount.
Mostoftheclassicalsolutionstoshellbucklingproblemswereestablishedbyneglectingedgeeffects
in the fundamental state. Due to this assumption the coefficients of the governing equations are
constantalongtheshellorvaryonlysmoothly,andclosedformsolutionscanbeobtainedforregular
shellconfigurations.
Thebifurcationsolutionpathsrepresentthepostbucklingbehaviouroftheshells.Anindicationof
theinitialpostbucklingbehaviour,viz.theslopeofthebifurcationpathatthebifurcationpoint,can
beobtainedatarelativelylowexpense[18],[19].Itisvaluableforestimatingtheeffectivestiffnessof
buckled plane or slightly curved panels, but for highly curved panels and shells its usefulness is
doubtful, since their stiffness may change quickly in the postbuckling range. The advanced
postbucklingbehaviourcanbedeterminedonlybysolvingnonlinearboundaryvalueproblemswhich
requireconsiderablecomputationaleffort.
5.5.3 Buckling of Shells of Revolution

5.5.3.1 Overview
Besidescircularcylindricalshellsmanystructuralapplicationsofthinwalledshellsconsistofgeneral
shellsofrevolution,themiddlesurfaceofwhichisobtainedbyrotatingaplanecurveaboutanaxisin
the planeofthecurve(seeFigure520andFigure521).Thelinesofprincipalcurvatureonashellof
revolutionarecalledtheparallelsandthemeridians.Theparallelsareformedbytheintersectionof
planes normal to the axis of revolution with the shell surface, whereas the meridians are the
intersectionswiththeshellsurfaceofplanesthatcontaintheaxisofrevolution.
Pointsonthemiddlesurfacearereferredtocoordinates and ,where denotestheanglebetween
the axisofrevolutionandanormaltothesurface,whereas isthecircumferentialcoordinate.The
principalradiiofcurvatureofthesurfaceinthe and directionare R and R ,respectively.In
certainapplicationsliketheshallowsphericalcapitisconvenienttointroduceanadditionalvariabler
definedas
r = R sin 513
Stabilityequationsforsymmetricallyloadedshellsofrevolutioncanbefound,forinstance,in[5].
79
ECSSEHB3224A
24March2010
Figure520:Shellofrevolutionnotationandsignconvention
Figure521:Meridianofashellofrevolution
5.5.3.2 Buckling Pressure of Spherical Shells

BucklingpressuresofsphericalshellswereinitiallycalculatedbyZoellyin1915[20]andSchwerin in
1922[21],who assumedthatthedisplacementsweresymmetricalwithrespecttoadiameter,andby
vanderNeutin1932[22],whoconsideredunsymmetricaldisplacements.Theirsolutionswerebased
on the assumption of infinitesimal displacements from the linear prebuckling solution, and
w / R = p (R / h ) /[(1 2 )E] ,andwereobtainedintermsofLegendrefunctions.Inallcasesthecritical
valuesofthemembranestressandpressurewere
1 Eh
cl = 514
3(1 2 ) R
h
pcl = 2cl 515
R
whereRistheradiusofthesphere,hitsthickness,ptheexternalpressure, themembranestress,w
the inward radial displacement, and E and the modulus and Poissons ratioof the shell material,
respectively.
80
ECSSEHB3224A
24March2010
The critical buckling stress is thus identical to the critical buckling stress for an axially loaded
cylindrical shell. As was the case for the cylindrical shell, it was soon found that the experimental
values were considerably lower than the theoretical values. Early tests also revealed that buckling
initiatesatsomespotatwhichasmalldimpleformswhichsubtendsasmallsolidangle.Thisallows
tomodeltheproblemofbucklingofacompletesphericalshellwiththeuseofashallowsphericalcap
clampedatitsedge.
5.5.3.3 Shallow Shells Subjected to Uniform External Pressure

ThegeometryofaclampedshallowsphericalshellisshowninFigure522.Astheexternalpressureis
increasing,dependingonthevalueofthesocalledshallownessparameter
H R
= 2[3(1 2 )]1 / 4 ( )1 / 2 = [12(1 2 )]1 / 4 ( )1 / 2 2 sin 516
h h 2
where H is the rise of the cap above the plane in which the edge lies, one gets different types of
behaviour.ThisisillustratedinFigure523wheretheloaddeflectioncurvesareshownforspherical
caps clamped at the boundary, using linear and nonlinear theories for calculating the axisymmetric
prebuckling deformations. The open circles on the linear loaddeflection lines indicate bifurcation
bucklingattheclassicalpressureforthecompletesphericalshell(seeequation(515)withthesame
radiustothicknessratioasthesphericalcap.
Figure522:Geometryofsphericalcapunderexternalpressure
NOTE1 With = 0 (the flat circular plate) there is no similarity at all with the
behaviourofthecompletesphericalshell.Theloaddeflectioncurveexhibitsa
stiffening characteristic which results from the buildup of inplane tension as
theplatedeforms(seeFigure523(a)).
NOTE2 With lessthanabout3.5theloaddeflectioncurvehasnohorizontaltangent
and no bifurcation point so that there is no loss of stability on the primary
equilibriumpath(seeFigure523(b)).
NOTE3 For lessthanabout6thereisaxisymmetricsnapthrough,butnobifurcation
buckling(seeFigure523(c))
NOTE4 For > 6 bifurcation buckling into a nonsymmetric mode occurs at a lower
loadthaneitheraxisymmetricsnapthroughofthecaporclassicalbucklingofa
completesphericalshell(seeFigure523(d,e,f)).
81
ECSSEHB3224A
24March2010
NOTE5 As increases above 7 the prebuckling behaviour becomes more and more
linear. Figure 523(f) corresponds to a configuration in which the cap can no
longerbecalledshallow.Thenonuniformityofprebucklingbehaviouroccurs
in a relatively narrow band or boundary layer near the edge. Any further
increase in results in no further alteration in the curves or locations of the
bifurcation points presented in Figure 523(f). No matter how high is, the
behaviouroftheincompletesphericalshellclampedatitsboundarywillnever
be the same as that of a complete spherical shell because the presence of the
boundary gives rise to edge buckling at the pressure from 80% to 90% of the
classicalvalue p cl .
For actual spherical shells and shallow caps random imperfections play a major role in the loss of
stability under uniform external pressure. Figure 524 demonstrates that the effect of initial
imperfectionsisjustassevereasinthecaseofcylindricalshellssubjectedtoaxialcompression.
82
ECSSEHB3224A
24March2010
H
Shallownessparameter = 2[3(1 2 )]1 / 4 ( )1 / 2)
h
(fromBushnell[28])
Figure523:Deflectioncurvesandbifurcationbucklingofcapsunderexternal
pressurewithvariousvaluesoftheshallownessparameter.
83
ECSSEHB3224A
24March2010
(fromKaplan[29])
Figure524:Earlyexperimentalresultsforclampedsphericalcapsunderexternal
pressure
5.5.3.4 Domed (torispherical) End-closures under Internal Pressure

Thisproblemisofspecialinteresttodesignersofpressurevessels,manyofwhichhavetorispherical
domes as endclosures. The possibility of nonaxisymmetric buckling due to internal pressure was
first predicted by Galletly in 1959 [30]. Buckling is caused by the occurrence of compressive
circumferentialstresses,whichareinducedbytheinternalpressureoverpartsoftheendclosure.
Using the membrane equations of an axisymmetric shell of revolution with no torque acting on the
torispherical endclosure shown in Figure 525 (a torispherical shell is a toroidal shell jointed to a
spherical cap) Galletly obtained the following expressions for the meridian and the hoop stress
resultants
1
N = p R 517
2
1 R
N = p R (2 ) 518
2 R
As can be seen in Figure 525 both radii of curvature are positive, therefore for internal pressure
( p > 0) the meridional stress resultant N will always be positive (tensile). Notice that the hoop
stressresultant N canbepositiveornegative,dependingontheratio R / R .Fromequation(517)
itisevidentthatif
R
>2
R
then N will benegative (compressive). The existenceof compressive hoop stresses due tointernal
pressureindicatesthatbucklingwithanasymmetricbucklingmodemayoccur.
To calculate the critical buckling pressure one usually solves the linearized stability equations,
whereby here, as can be seen in Figure 525 the prebuckling stress resultants N and N are not
84
ECSSEHB3224A
24March2010
constantbutvarywith .Tosolvetheresultingvariablecoefficientlinearizedstabilityequationsfor
therathercomplexmeridionalgeometryofthetorisphericalshellsonereliesonnumericalmethods.
Figure525:Pressurevesselwithtorisphericalendclosurenotationandsign
convention
5.6 References
[1] Euler,L.,Decurviselastics,LeonhardEulersElasticCurves,translatedandannotatedby
W.A.Oldfather,C.A.EllisandD.M.Brown,reprintedfromIsis,20,(58),1933,TheSt.Catherine
Press,Bruges,Belgium.
[2] Rivello,R.M.,TheoryandAnalysisofFlightStructures,McGrawHill,NewYork,1969.
[3] Simitses,G.J.,AnIntroductiontotheElasticStabilityofStructures,PrenticeHall,Englewood
Cliffs,NewJersey,1976.
[4] Shanley,F.R.,InelasticColumntheory,JournaloftheAeronauticalSciences,14,May1947,pp.
261268.
[5] Brush,D.O.andAlmroth,B.O.,BucklingofBars,PlatesandShells,McGrawHill,NewYork,
1975.
[6] Notenboom,R.P.,FiniteStripElementsinThinPlateBucklingAnalysis,ReportLR642,Delft
UniversityofTechnology,FacultyofAerospaceEngineering,Delft,theNetherlands,September
1990.
[7] Vlasov,V.Z.,ThinwalledElasticBeams,2ndEdition,(translatedfromRussian).NSF,
WashingtonD.C.andU.S.Dept.ofCommercebytheIsraelProgrammeforScientific
Translation,Jerusalem,1961.
[8] Singer,J.,Arbocz,J.andWeller,T.,BucklingExperiments,ExperimentalMethodsinBuckling
ofThinWalledStructures,Vol.1,JohnWiley&Sons,NewYork,ISBN0471956619,1998.
[9] Bryan,G.H.,OntheStabilityofaPlanePlateunderThrustsinitsOwnPlanewithApplications
totheBucklingoftheSidesofaShip,Proc.LondonMath.Soc.,Vol.22,1891,pp.5467.
[10] Gerard,G.andBecker,H.,HandbookofStructuralStability,Part1:BucklingofFlatPlates,
NACATN3781,1957.
[11] Coan,J.M.,LargeDeflectionTheoryforPlateswithSmallInitialCurvaturesLoadedinEdge
Compression,ASMEJournalofAppliedMechanics,18,(2),June1951,pp.143151.
85
ECSSEHB3224A
24March2010
[12] Hu,P.C.,Lundquist,E.E.andBatdorf,S.B.,EffectofSmallDeviationsfromFlatnesson
EffectiveWidthandBucklingofPlatesinCompression,NACATN1124,1946.
[13] Koiter,W.T.,BucklingandPostBucklingBehaviourofaCylindricalPanelunderAxial
Compression,ReportS.476,NationalAeronauticalResearchInstitute(NLR),Amsterdam,The
Netherlands,14May1956.
[14] Neut,A.vander,Theinteractionoflocalandoverallbucklingofthinwalledstructures,9th
Congr.Internat.Counc.Aeron.Sci.,Haifa,1974.ICASPaperNo.7407.
[15] Koiter,W.T.,Generaltheoryofmodeinteractioninstiffenedplateandshellstructures,
TechnischeHogeschoolDelft,AfdelingderWerktuigbouwkunde,ReportWTHD91,1976.
[16] Benson,A.S.andMayers,J.,GeneralInstabilityandFaceWrinklingofSandwichPlates
UnifiedTheoryandApplications,AIAAJournal,Vol.5,No.4,April1967,pp.729739.
[17] Sullins,R.T.,Smith,G.W.andSpier,E.E.,ManualforStructuralStabilityAnalysisofSandwich
PlatesandShells,NASACR1457,1969.
[18] Koiter,W.T.,Onthestabilityofelasticequilibrium,(inDutch),ThesisDelftUniversityof
Technology,1945,Engl.Translations:NASATT10,833(1967)andAFFDLTR7025(1970).
[19] Budiansky,B.,Theoryofbucklingandpostbucklingbehaviorofelasticstructures,Advances
inAppliedMechanics(editedbyChiaShunYih),Vol.14,NewYork,SanFrancisco,London:
AcademicPress1974,ISBN0120020149.
[20] Zoelly,R.,UebereinKnickungsproblemanderKugelschale,ThesisETHZrich.Switzerland,
1915.
[21] Schwerin,E.,ZurStabilittderDnnwandigenHohlkugeluntergleichmssigem
Aussendruck,Z.Angew.Math.Mech.,Vol.2,1922,pp.8191.
[22] vanderNeut,A.,DeElastischeStabiliteitvandenDunwandigenBol(inDutch),Ph.D.Thesis,
DelftUniversityofTechnology,1932.
[23] Kaplan,A.,Fung,Y.C.,Anonlineartheoryofbendingandbucklingofthinelasticshallow
sphericalshells,NACATN3212,1954.
[24] Klppel,K.,Jungbluth,O.,BeitragzumDurchschlagproblemdnnwandigerKugelschalen,
DerStahlbau,Vol.VI,1953,pp.121130.
[25] Homewood,R.H.,Brine,A.C.,Johnson,A.E.,Jr.,ExperimentalInvestigationoftheBuckling
InstabilityofMonocoqueShells,ExperimentalMechanics,Vol.1,1961,pp.8896.
[26] Bellifante,R.J.,Bucklingofsphericalcapsunderuniformexternalpressure,DouglasAircraft
MissileandSpaceSystemsEngineeringReportSM38938,1962.
[27] Tsien,H.S.,ATheoryfortheBucklingofThinShells,J.Aeronaut.Sci.,Vol.9,1942,pp.373
384.
[28] Bushnell,D.,Computerizedbucklinganalysisifshells,MartinusNijhoffPublishers,
Dordrecht/BOSTON/LANCASTER,1985,ISBN9024733099.6.
[29] Kaplan,A.,BucklingofSphericalShells,fromThinShellStructures,Theory,Experimentand
Design,Y.C.FungandE.E.Sechler(eds.)PrenticeHall,Inc.,EnglewoodCliffs,N.J.,1974,pp.
248288.
[30] Galletly,G.D.,TorisphericalshellsACautiontoDesigners,ASMEJournalofEngineering
Industry,BI,(1),February1959,pp.5166.
86
ECSSEHB3224A
24March2010
6
Analysis Methods
6.1 Introduction
To study the stability of the equilibrium state requires the solution of two different problems. One
proceeds
a. todetermineatwhichloadleveltheequilibriumlosesitsstability,and
b. to determine if the loss of stability causes also the failure of the structure with unacceptable
stressordeformations,orifthestructureiscapableofsomeadditionalloadcarryingcapability
inanewstableconfiguration(postbuckling).
Althoughtheproblemsareclear,severaldifficultiesarise,andthefirstoneisrelatedtoafundamental
question:whattypeofanalysisisadequatetoapproachtheseproblems?
Theanswerisnotquiteobvious.Ononeside,theanalysisofthefailureofthestructureandthelossof
load carrying capability is expected to be nonlinear, involving both geometrical and material
nonlinearity. But, on the other side, the loss of stability is sometime coincident with failure, and for
somecasesofsuchtype,wellknownlinearanalyseshavebeendemonstratedtobesufficienttofind
loss of stability. In addition, in some other cases, a static analysis is not adequate to evaluate the
stabilityoftheequilibriumandthestudyofselfexcitedvibrationsisnecessary.
Thefirstscopeofthischapteristogiveasurveyofallthetypesofanalysiswhichcouldbeusedfor
thestudyofstability.Asimpledescriptivepresentationwillbeprovided,tofixacommonlanguage
andaclearunderstanding.Forrigorousderivationandcomprehensivedetails,properreferenceswill
beindicated.
A second scope is to categorize the phenomena of loss of stability, by indicating for each type the
appropriateanalysistobeperformed.
6.2 Static equilibrium and stability
6.2.1 Overview
In this chapter we suppose that the structural model (sometimes named system) is a discrete model
(e.g. aFEMmodel)andwe follow extensively theexposition of C.A. Felippa, which provides much
moredetailsinhislectures[1].TheNcomponentsofthestatevectoru(e.g.thenodaldisplacements)
describetheconfigurationofthesystemundertheactionofphysicalquantities(e.g.externalforces).
Thesequantitiesaredescribedbyseveralcontrolparameters,whicharethecomponentsofthecontrol
vector.Thecontrolvectorreducestoascalarifthesystemiscontrolledbyoneparameteronly.
Eachcoupleofuandcorrespondingtoanequilibriumstateisanequilibriumpointinthehyperspace
defined by u and . When varies, the equilibrium point moves on trajectories called equilibrium
87
ECSSEHB3224A
24March2010
paths.Projectionsofthesepathscanbeobtainedbychoosinganyreferencevaluesuandamongthe
componentsofuand.Thegraphobtained,calledalsoresponsediagram,isasthatofFigure61andis
usefultodepicteffectivelythepropertiesofstructuralequilibrium.
Figure61:Typicalequilibriumpath
Specializedequilibriumpointsofinterestforstabilityarethecriticalpoints,whicharedistinguishedin:
a. Limit points, at which the tangent to the equilibrium path is horizontal, i.e. parallel to the
deflectionaxis,and
b. Bifurcationpoints,atwhichtwoormoreequilibriumpathscross.
At critical points the relation between the load and the associated deflection is not unique, and
physicallythestructurebecomesuncontrollableormarginallycontrollable.
6.2.2 Turning points and failure points

NOTE Someotherspecialequilibriumpoints,notrelatedspecificallytostability
study,are:
theturningpoints,atwhichthetangenttotheequilibriumpathisvertical,i.e.parallel
totheloadaxis.Forloadcontrolledsystemsthesearenotcriticalpointsandhaveless
physical significance, but are of interest for some structures. They have some
computational significance, however, because they can affect the performance of
certainsolutionmethods.
thefailurepoints,atwhichapathsuddenlystopsorbreaksbecauseofe.g.material
failure. The phenomenon of failure can be local or globalin nature. In the first case
(e.g,failure ofanoncriticalstructurecomponent)thestructurecanregainfunctional
equilibrium after dynamically jumping to another equilibrium path. In the latter
case the failure is catastrophic or destructive and the structure does not regain
functionalequilibrium.
Because u is not always a single valued function of (either in general see Figure 61 and in
particularattheneighbourhoodofcriticalpoints),themostappropriatedescriptionoftheloadpath
aretheparametricequations:
88
ECSSEHB3224A
24March2010
u = u(t ) , = ( t ) 61
where t is a scalar parameter. If t is interpreted as a physical time, (t) describes the history of the
loadingandu(t)thecorrespondinghistoryofthestructuralconfiguration.
Theneedofintroducinganditshistoryisstrictlyrelatedtononlinearnatureoftheproblem.While
for a linear problem one or more (separately computed) equilibrium points are sufficient for a full
description(becauselinearsuperpositionmaybeappliedtodescribethecompleteloadscenario),fora
nonlinearsystemthesequenceofloadapplicationisimportant.
Under the action of the external load vector p(u,) (in general dependent directly on u and ), the
structure deforms to develop the equilibrating internal force vector f(u). The static equilibrium
equationscanthenbewritteninthefollowingforcebalanceequation:
f (u ) = p (u, ) 62
whichcanbewrittenintheformoftheforceresidualequation:
r (u, ) = 0 63
byintroducingtheresidualforcevectorr=pf.
6.2.3 Types of loadings

Some special cases of external loads are important. Firstly, very common are onecontrol
parameter external loads p(u,). If the vector p(u,) does not depend on u, the system is called
separable.Ifinadditionitislinearintheloadingissaidtobeproportional,and theincremental
loadvectorp/=qisconstant.
As we will see in the following paragraphs, the incremental form of the previous equation is
convenient, both for nonlinear nature and for the stability study. By indicating the derivatives with
respecttothescalarparametertwithadot,thefollowingincrementalequilibriumequationapplies
Ku& Q& = 0 64
where:
r
K= 65
u
isthetangentialstiffness,and
r
Q= 66

istheincrementalloadmatrix.
Forasystemwithasinglecontrolparameter,ift=,theincrementalequilibriumequationtakesthe
morefamiliarform:
K du = q d 67
where
r
q= 68

istheincrementalloadvector.
Bydefiningtheincrementalvelocityvectorv=du/d,theequation67canbewritten:
89
ECSSEHB3224A
24March2010
K (u ) v = q 69
According to Euler the equilibrium is stable if for a given q, v is a unique solution; then the
equilibriumisnotstablewhenthefollowingsystemhasnonzerosolutionsz:
K (u ) z = 0 610
whichisonlypossibleif:
det(K(u)) = 0 611
Inotherwords,ifwesolvethegeneraleigenvalueproblemforK(u)
K (u)z i = i z 612
weareinterestedinzeroeigenvalues.
6.2.4 Distinction between equilibrium and stability equations

As outlined already in this paragraph, it is important to emphasize the distinction between
equilibriumandstabilityequations.
Equilibriumequationscanbedefinedasresidualequationsofgeneralizedforces(asinthissection),
by means of the total potential energy for conservative systems (as in the next section) or by a
comparisonbetweenexternalandinternalwork.
Stabilityequationscanbederivedasincrementalequationsfromtheresidualequations.
In continuum mechanics the equilibriumresidual equations are complemented by compatibility
equationsandconstitutivelawsandresultinasetofgoverningequations,partlyindifferentialform.
6.2.5 Static stability in the sense of Hadamard and Euler

An incremental form of equilibrium, as equation 69, can also be used to define static stability
followingHadamard (18651963):equilibriumisstableifvvariescontinuouslywithq,i.e.if acertain
norm of v is lower than an arbitrary value , whenever the norm of q is chosen lower than . The
Hadamard definition is more coherent with the dynamic definition (e.g. Liapunov), but the Euler
approach has the advantage of defining immediately a condition, like Eq. 610 and Eq. 611, to be
verifiedforanonstableequilibrium.Howeveroneshouldnoticethatthisconditionisaffectedbythe
followingrestrictivehypotheses:
theuniquenessofsolutionisinvestigatedonlyforunchangedcontrolparameters,while
ingeneralthecheckshouldbeperformedalsoforconstantstateparameter.
the existence of multiple adjacent equilibrium points does not give sufficient
informationonthestability;additionalconsiderationsareneededtoestablishstability,
e.g.workbalanceonincrementedequilibriumstate(asthetotalpotentialenergystudy
allowsforconservativesystems);
theincrementalequationlikeequation69,althoughnonlinearlydependentfromstate
andcontrolparameters,isalinearization,usingthetangentasanapproximationofthe
equilibriumpath.
In conclusion, to follow the Euler approach (which studied all the possible shapes of an initially
straightelastica)andfollowingLagrangecontribution(whichintroducedthenotionofbifurcation),a
completenonlinearanalysisshouldbenecessarytofindallthepossibleadjacentequilibriumpoints,
andadditionalconsiderationsarerequiredtoassessstability.
90
ECSSEHB3224A
24March2010
6.2.6 The non linear eigenvalue problem for Euler stability
Theequation610requiresthesolutionofanonlineareigenvalueproblem.
6.3 Static equilibrium and stability of conservative

systems
Theconservativesystemsareveryimportantcases,frequentlyencounteredinengineeringpractice.
Asystemisconservativeifitsatisfiesthefollowingtwoconditions:thestructuredissipatesnoenergy
(internally conservative) and the work done by external load is path independent (externally
conservative).Withthesehypotheses,onecandefineaninternalenergyU(commonlycalledthestrain
energy),aloadpotentialP(thenegativeofwhichistheexternalworkW)andatotalpotentialenergy=
UP.Theinternalforcevectorf,theexternalloadvectorp,theforceresidualrandthetangentstiffness
matrixKhavethefollowingexpressions:
P U 2
p= , f = , r = , K = 613
u u u u 2
Theequilibriumequationr=0isthenequivalenttothestationaryconditionofU:

U = u = 0 , u 0 or =0 614
u u
The static stability approach searches for equilibrium points for which non zero incremental
displacements z are possible for unchanged load. If we restrict to the most common case of one
controlparametersystem,byusingequation67,thisconditioncanbewritten:
K (u ) z = 0 , z0 615
whichispossibleonlyif
det(K (u)) = 0 616
and,accountingforequation613,itisequivalentto
2 2 2
2U = u = 0 , 2 u 0 or = 0 617
u 2 u 2
Thepreviousequationcanbeusedtodefinethecriticalpointsofconservativesystems.Theyarefound
on an equilibrium path for a critical value cr of the load parameter. Because the system is
conservative, K is symmetric and equal to its transpose; then by transposing equation 615, one
obtains
z T K (u) = 0 618
Premultiplyingbothsidesofequation67byzTandusingequation618thefollowingexpressionfor
thework,performedbytheincrementalloadvectorq,isderived
z T qd = 0 619
Twocasesarepossible:if
zT q 0 620
thend=0,i.e.thetangenttoequilibriumpathisparalleltou,andthecriticalpointisalimitpoint;and
if
91
ECSSEHB3224A
24March2010
zT q = 0 621
(i.e.whenzisorthogonaltoq),thecriticalpointisabifurcationpoint.
Both limit and bifurcation points can be single or multiple valued. The bifurcation points can be
asymmetric, stablesymmetric and unstable symmetric. For a qualitative description of the different
behaviourofthetotalpotentialenergyseesection6.6.
6.4 Solution of nonlinear equilibrium equations

Tointroducethebasicideasofthesolutionmethods,letusconsideraonecontrolparametersystem
andmakereferencetoFigure62andFigure63.

P3

P2

P1
u0, 0 u1 u2 u3 u
Figure62:Purelypredictormethod
P1 C1
u0, 0 u1 u
Figure63:Predictorcorrectormethod
92
ECSSEHB3224A
24March2010
Let us suppose that a starting equilibrium condition (u0,0) and the tangential stiffness K0=K(u0) are
known. Foran increment of the control parameter, we can predict anapproximation of the new
equilibriumpoint,bymovingalongthetangenttotheequilibriumpathatthepointP1.Thealgorithm
usedtofindP1iscalledpredictorandcanbeappliedforthesubsequentincrementstoapproximate
theequilibriumpath.
Theapproximationofsuchpurelypredictormethodsisaffectedbyadrifterrorhighlydependingon
the increments i. To improve the approximation, after each prediction, a corrector algorithm is
applied,tofindpointsCi,closertotheequilibriumpathalongaselectedprojectionline(paralleltou
in the example of the Figure 63). Note that both Pi and Ci are only approximations of equilibrium
points:thecorrectorisaniterativealgorithmusedtoreducethe residualofequilibriumequation63
toarequestedlevel.
Moregenerally,thesolutionmethodsusebothapredictortogetherwithacorrectoralgorithm,finding
thesequenceoftheapproximatesolutionpoints
ui,i 622
If the solution point S= Pn at step n is known, to find the point Pn+1, the predictor applies the
incrementalequation
K du = q d 623
bymovingontheequilibriumpathtangenttothepointPn+1,theintersectionofthetangentwiththe
surfacedefinedbytheconstraintequation
c(u n , n ) = 0 624
Then an iterative corrector moves along the constraint surface towards the equilibrium path, by
buildingthesequence:
u kn , kn 625
to find the next approximate equilibrium point C n +1 (u n +1 , n +1 ) , satisfying with a requested

accuracytheresidualequation:
rn (u n +1 , n +1 ) = 0 626
Note that both equation 623 and equation 626 are systems of N equations, with N+1 unknown
quantities(Ncomponents ofuand):theconstraintequationistheadditionalconditionnecessaryto
solvetheproblem.
Todefinethealgorithmthefollowingstepsareused:
a. anintegrationschemeofequation623
b. aconstraintequation624forthepredictor
c. aniterativecorrectorforbetterapproximationofresidualequation624
d. arequestedaccuracywithadefinednormforresidualr.
TheEulerforwardschemeisthemostsimpleandcommonlyusedintegrationschemeforthetangent
stiffnessequation623,whichbecomes:
K n (u n +1 u n ) = q n 627
ThreetypicalstrategiesofconstraintequationsareschematicallyrepresentedinFigure64,Figure65
andFigure66.
93
ECSSEHB3224A
24March2010
n=ln Corrector
Predictor
Casea
Figure64:Typicalcorrectorstrategies(1of3)
Corrector
Predictor
un=ln
Caseb
94
ECSSEHB3224A
24March2010
sn=ln
Corrector
Predictor
Casec
Thecasea)referstotheloadcontrolmethodwhichspecifiesavaluelnfortheloadincrementn:
n = l n 628
Thecaseb)referstothestatecontrolmethodwhichspecifiesavalueforaselectednormofu:
u n = ln 629
Thecasec)referstothearclengthcontrolmethodwhichspecifiesalengthsnonthetangentpath
s n = l n 630
IterativecorrectoralgorithmscanbeobtainedbyaTaylorexpansionaboutthelastequilibriumpoint
(iterationkofpredictionn)ofthesystemofequations:
r (u, ) = 0
631
c(u, ) = 0
asfollows
rnk + K kn (u kn +1 u kn ) q(kn +1 kn ) = 0
632
c nk + a T (u kn +1 u kn ) + g (kn +1 kn ) = 0
withthedefinitions:
r r c c
K= q= aT = g= 633
u u
The matrix of the system of equations 632 is called augmented stiffness matrix, because it is of order
N+1ifthetangentstiffnessmatrixKisoforderN.
Several tests can be used to check the accuracy of the solution and to stop the iterations along the
constraint surface. For example, the iterations can be stopped by checking the size of the residual
forces
95
ECSSEHB3224A
24March2010
rk r 634
orbycheckingthesizeofthestatevectorincrementofthelasttwoiterations:
u k +1 u k u 635
6.5 Finding critical points of the equilibrium path

The study of the stability of the equilibrium of a structure subject to static or quasistatic loads
requiresthesolutionoftwoproblems:
a. tofindcriticalpointsoftheequilibriumpath,and
b. toevaluatethestabilityofthesepoints.
Thischapterisdedicatedtothefirstproblem;thesection6.6discussesthesecondproblem.
The first problem can be approached by the direct methods, which detect the critical points without
tracingtheequilibriumpaths,orbytheindirectmethods,whichdetectthecriticalpointswhiletracing
theequilibriumpath.
A direct method searches for critical points among equilibrium points. It does not trace the
equilibrium path, but it uses the residual force equilibrium equations with a set of constraints, the
criticalpointtestfunction(CPTF),whichcharacterizesthecriticalpoints.
The most common CPTF is the equation of the eigenvalues of the stiffness matrix K. Then a direct
methodusuallyusesthefollowingsetofequations:
r (u, ) = 0 636
K (u, ) z = 0 637
(z ) = 0 638
wherethelastequationisanaddedconstrainttoruleoutthetrivialsolution z = 0 (i.e. z T z = 1 can

beused).
In other words, because the critical points are characterized by the tangent stiffness matrix K being
singular,aCPTFessentiallymonitorsKtoobtainanestimateofitsproximitytoacriticalpoint.The
smallesteigenvalueofKisanaccurateestimatorofKssingularity.Achangeofsignofthesmallest
eigenvalue detects K becoming singular. The monitoring of the eigenvalues of K, or the problem of
finding null eigenvalues of K, is nonlinear because: K is evaluated at equilibrium points on a
nonlinearequilibriumpath,anditisanonlinearfunctionofu,whichinturnisanonlinearfunctionof
.Asaconsequence,becauseacompleteresponseanalysiscannotbeavoided,onlyindirectmethods
areadequatetosolvetheproblemoffindingthecriticalpoints.However,becauseananalysisofsuch
natureiscomputationallyveryexpensive,someapproximationscanbeaccepted.Themostimportant
istheLinearBucklingEigenvalue(LBE)approximation,whichcanbeformulatedasfollows.
Withreferencetoacertainreferenceequilibriumpoint u 0 letusconsideraTaylorapproximationofK
withrespectto.theincrementalloadparameter:
K ( ) = K 0 + K 1 + 2K 2 + 3K 3 + ... 639
Byjusttakingthelinearterms,alineareigenvalueproblemisobtained:
(K 0 + K 1 ) z = 0 640
Herein K 0 is indicated also as K M and is called the material stiffness, being the stiffness at the
96
ECSSEHB3224A
24March2010
reference equilibrium point; and K 1 is indicated also as K G and is called the geometric or
differential stiffness, linearly dependent on the load parameter and K1 is also evaluated at the
referenceequilibriumpoint.
If,inaddition,theunloadedstructureconditionistakenasreferenceequilibriumcondition,thelinear
prebucklingproblem(LPB)isobtained,requiringthefollowingsteps:
Assemblethelinearstiffness K 0 andsolvethelinearstaticproblem K 0 u = q 0 for=1,and

obtaintheinternalforcesandstresses.
Computethereferencegeometricmatrix K 1 andthegeometricstiffnessmatrix K G = K 1
Solvethelinearbucklingeigenvalueproblem (K 0 + K 1 ) z = 0
The LPB method is a true direct method, because it does not require the solution of any nonlinear
equilibriumequation.Anintermediateapproximationispossibleifasreferenceequilibriumpoint,a
pointistakenonanequilibriumpathotherthantheunloadedstructureconditionandaLBEproblem
is formulated. In this case prebuckling nonlinear effects are accounted for and a LBE is applied,
subsequent to some steps of nonlinear equilibrium analysis. Figure 67 summarizes the different
situationsforLPBandLBEaccountingfornonlinearprebucklingeffects.
linear
nonlinear
Figure67:Eigenvalueanalysiswithlinearandnonlinearprebuckling
6.6 Stability at the critical points of a discrete

conservative system
As explained in section 6.3 the static stability of a conservative mechanical system can be tested
completely using a static criterion, by investigating the second variation of the potential energy. A
stationary value of the total potential energy with respect to the state variables is necessary and
97
ECSSEHB3224A
24March2010
sufficient for the equilibrium of the system and a relative minimum of the total potential energy is
necessaryandsufficientforthestabilityofanequilibriumstate.
Fordiscreteconservativesystems,inaddition,aclassicaltreatiseofThompsonandHunt[2]provides
generalconclusiononthestabilityatfourtypesofcriticalpoints,basedonthestudyofthepotential
energy. These results are summarized in this section by using response plots, with the following
conventions:
a. thecontrolparameteristheverticalaxis
b. arepresentativedisplacementuordeformationmodeamplitudeisshownalongthehorizontal
axisasstateparameter
c. heavylinesdenoteequilibriumpaths
d. continuouslinesdenotestablepaths
e. dashedlinesdenoteunstablepaths
f. shadedlinesdenotetotalpotentialenergy(whichdependsonu,forafixedloadparameter)
NOTE These figures illustrate the physics well, but if we are dealing with a
system with many degrees of freedomcare shouldbe taken in drawing
conclusionsfromtheseschematicfigures.Onanactualplotofagainst
oneofthestateparameters,thelimitpointisnormallyseenasasmooth
maximum,butitshouldberealizedthatforacertainchoiceofthestate
parameter ui the point might appear as a sharp cusp. The smooth
maximumofapathinthreedimensionalspacecanforexamplebeseen
asacuspiftheeyeisdirectedalongthehorizontaltangenttothepath.
Figure 68 represents the case of a limit point. At a fixed value = 1<cr the total potential energy
(u, 1 ) hasaminimumwithrespecttothestateparameteruonthestablerisingregionofthepath
andamaximumontheunstablefallingregion.Astheprescribedvalueof isincreased,themaximum
and the minimum approach each other and coalesce when = cr. At this critical point the total
potentialenergyhasahorizontalpointofinflexion.Atahighervalueof=2 > cr ,therearenolocal
equilibriumstatesandthetotalpotentialenergy hasnostationarypoint.Theequilibriumstateatthe
limitpointisunstable(becausethetotalpotentialenergyhasnotaminimum),andtheabsenceoflocal
equilibrium states at values of >cr implies that a physical system under slowly increasing will
eventuallysnapthrough,dynamically.
Figure 69 shows the case of an asymmetric bifurcation point. The initially stable fundamental
equilibrium path that emanates from the origin loses its stability on intersecting a distinct and
continuoussecondary(postbuckling)equilibriumpath.TheintersectionpointBisacriticalpointof
bifurcationtype.Anasymmetricbifurcationpointischaracterizedbythefactthatbothpathshavea
nonzeroslopewithrespecttoatB.Withvaryingthepathsexhibitaphenomenoncalledexchangeof
stability.For1<crthetotalpotentialenergy hasaminimumwithrespecttouonthestableregionof
thefundamentalpathandamaximumwithrespecttouontheunstableregionofthepostbuckling
path.As isincreased,themaximumandminimumfinallycoalescesothatat=crthetotalpotential
energy hasahorizontalpointofinflexionatthecriticalequilibriumstate.At valuesoverthecritical
one the maximum and minimum exchange places. Since an unstable branch emanates from B, the
critical equilibrium state is unstable. In the presence of small disturbances a physical system under
slowly increasing would snap dynamically from this critical equilibrium state despite of the
existenceofstableequilibriumstatesathighervaluesof.
Symmetric bifurcation points are characterized by the fact that the intersecting path has zero slope
withrespecttothecontrolparameteratB.Thesepointsmaybedividedintostableorunstable,
98
ECSSEHB3224A
24March2010
depending on whether the intersecting postbuckling path is rising or falling and the total
potentialenergyhasaminimumoramaximum.
Figure 610 depicts the case of an stablesymmetric bifurcation point. Here a fundamental equilibrium
pathrisingmonotonicallyfromthereferencestateisseentointersectastablerisingsecondary(post
buckling) path that passes smoothly through the critical equilibrium state with zero slope. The
continuationofthefundamentalpathbeyondBisunstable.Thetotalpotentialenergyfor=1 < cr has
a single stationary value with respect to u, namely the minimum on the stable region of the
fundamentalpath,andasthe value of is increased this minimum is transformed into two minima
andonemaximum.Thecriticalequilibriumstateisneutrallystableandthesecondarypathisstable,
soaphysicalsystemunderslowlyincreasing wouldexhibitnodynamicsnapbutwouldfollowthe
stable rising postbuckling path, the direction taken depending on the small disturbances or
imperfectionswhichareinevitablypresent.
Incaseofanunstablesymmetricbifurcationpoint(seeFigure611),thefundamentalpathintersectsan
unstablefallingpathwhichasinthepreviouscasehasazeroslopeatthecriticalequilibriumstate.At
a prescribed value of = 1 < cr the total potential energy has now three stationary values with
respecttou,namelytwomaximaontheunstablepostbucklingorsecondarypath,andaminimumon
the stable region of the fundamental path. These three stationary points transform into a single
maximumwithincreasing.Thecriticalequilibriumstateisseentobeunstable,soaphysicalsystem
would snap dynamically from the critical equilibrium state, the direction taken depending on the
postulatedsmalldisturbancesorimperfections.
(From[1])
Figure68:Potentialenergyandstabilityatalimitpoint
99
ECSSEHB3224A
24March2010
(From[1])
Figure69:Potentialenergyandstabilityatanasymmetricbifurcationpoint
(From[1])
Figure610:Potentialenergyandstabilityatastablesymmetricbifurcationpoint
100
ECSSEHB3224A
24March2010
(From[1])
Figure611:Potentialenergyandstabilityatanunstablesymmetricbifurcation
point
6.7 Imperfection analysis

Usuallystructuralanalysismodelsdonotaccountformultipleimperfectionsoftherealstructures,
as for example deviations of medium surfaces from ideal geometry, variation of thickness, load
eccentricity, material nonhomogeneous properties and flexibility of boundary conditions. These
omissions may be accepted, and included among analysis uncertainties covered by safety factors,
whenever they do not affect significantly structural behaviour and, as a consequence, do not cause
excessive discrepancies between analysis results of perfect models and test measurements of
imperfect real structures (usually differences between analytical predictions and test results are
acceptediflowerthan10%fordisplacementsand20%forstress).
Thisisnotalwaystrueforstructurestabilityanalysis.Forbucklingofshellsbigdiscrepanciesbetween
analysisandtestwereemphasizedsince1932byFlgge[3]andsubsequentinvestigatorsrecognized
the imperfections as responsible for these differences. In particular shell stability has been
discovered to be very sensitive to deviations of the medium surface from ideal geometry (in the
following referred to simply as shape or geometric imperfections), as indicated for example in the
Figure612,reproducedfromBabcock[4].Thisfigurereferstoacylinderunderaxialloadandshows
the relative importance of different types of imperfections, and that a small geometric imperfection
can substantially reduce the stability values, because the curve of shape imperfection starts with a
verticaltangent.
More generally, high sensitivity to the presence of small geometric imperfections is a
phenomenon associated with certain types of critical points, as can be summarized with the help of
thefollowingfigureswhichrepresenttheeffectofimperfectionsonlimitandbifurcationpoints.
101
ECSSEHB3224A
24March2010
Figure612:Influenceofimperfectionsonthestabilityofaxiallyloadedcylinders
(from[4])
IntheFigure613andFigure614,ontheleftaprojectionoftheloadpathisdisplayed,byusing,as
usually,whichrepresentsthecontrolparameter(e.g.theloadmultiplier)andurepresentingastate
parameter(e.g.displacementinloaddirection).Boldlinesrepresenttheresponseoftheperfectsystem
whereas light lines represent the responses ofimperfect systems for fixed values of adimensionless
imperfection parameter (e.g. amplitude of shape deviation as percentage of minimum thickness).
Furthermorecontinuouslinesidentifystableequilibriumpathportionswhereasdashedlinesidentify
unstableportions.Inthesamefigures,ontheright, cristhevalueofatcriticalpoints,plotted
againsttheimperfectionparameter .
Figure613describestheeffectsofimperfectionsatalimitpoint:thechangesinducedon equilibrium
pathdonotaffectthenatureofcriticalpoint,itisalwaysalimitpointbothforperfectandimperfect
structure.Thecriticalvalue at limitpointL movesalongacurvesassketchedintherightplot,which
maintains a finite and nonzero slope even for 0. Changes of L are usually moderate (1020%)
andthestructurecouldbetypifiedasmildlyimperfectionsensitive.
For asymmetric bifurcation points the role of imperfection is much more significant, as shown in
Figure 614. Now, for > 0 , the imperfection modifies the nature of the critical point from a
bifurcationpointoccurringatCtoalimitpointatS;Thecurveontherighthasnowaninfiniteslope
as 0 and the drop from C to S can be very high (e.g. more than 70%) also for a small
imperfection, typifying the structure as highly sensitivity to initial positive imperfections. For < 0 ,
the structure apparently exhibits no instability in the vicinity of the bifurcation point and follows a
stable rising path; however rapid growth of the deflections arises as the critical load level of the
102
ECSSEHB3224A
24March2010
perfect system is reached and the equilibrium is unreliable in the presence of small dynamic
disturbances.
Figure613:Effectofinitialimperfectionsatalimitpoint
Figure614:Effectofinitialimperfectionsatanasymmetricbifurcationpoint
TypicalplotsforanunstablesymmetricbifurcationpointareshowninFigure615.Bothpositiveand
negativeimperfectionscausemodificationfrombifurcationtolimitpoints.Duetoasharpcuspofthe
curvecrvs ,significantreductionofstabilitycriticalvaluescanarisealsoforsmallimperfections,
andalsothiscaseischaracterizedbyahighsensitivitytoinitialimperfections.
TypicalpicturesforastablesymmetricbifurcationpointareshowninFigure616.Bothpositiveand
negativeimperfectionshavesimilareffects,eachyieldingacontinuouslystableandrisingequilibrium
pathsasshown.Therefore,imperfectstructuresofthistypemovealongstableequilibriumpaths,but
displayamorerapidgrowthofthedeflectionsasthecriticalloadoftheperfectsystemisapproached.
103
ECSSEHB3224A
24March2010
Inconclusion,duetohighimperfectionsensitivityofasymmetricandunstablesymmetricbifurcation
points,structuralstabilityanalysisshouldaccountforimperfections.
Figure615:Effectofinitialimperfectionsatanunstablesymmetricbifurcation
point.
Figure616:Effectofinitialimperfectionsatastablesymmetricbifurcationpoint
6.8 Dynamic stability analysis

Dynamicstabilityanalysisisrelevant,in themostgeneralsense,tothestabilityofagenericmotionand,
inamorerestrictedsense,to thestabilityoffreevibrationsaroundastaticequilibriumstate(i.e.atrest
state).Inthefirstsenseitcanbe applied forexampletotheresponseofastructuretoloadsvarying
with the time in such a way that inertia forces cannot be ignored (i.e. impulse loads). In the second
sense it is required when static approaches described in previous paragraphs are not adequate to
checkstabilityofstaticequilibrium,butitisnecessarytostudyaperturbationmotionstartingfroma
104
ECSSEHB3224A
24March2010
static equilibrium state (i.e. stability of equilibrium of nonconservative systems). Lyapunovs
definition of dynamic stability can be applied to both cases. It states essentially the continuity of
motion,consideredintermsofdisplacementsandvelocities,withrespecttotheinitialconditions.
Letusthenconsiderthephysicaltimetasstateparameteranddescribethemotionbythevector
u(t )
y (t ) = 641
u& (t )
& (t ) = du / dt ,withtheinitialconditions
includingboththestatevectoru(t)anditsderivative u
u(0)
y0 = 642
u& (0)
Letusconsideranothermotionx(t) withinitialconditions x 0 ,anddefineanorm(x,y)tomeasure
thedistanceofthemotions andanorm0(x0,y0)tomeasurethedistanceofinitialconditions.Thenwe
cangivethefollowingdefinitions.
AmotionisLyapunovstableif:
> 0, > 0 : 0 (x 0 , y 0 ) < ( x, y ) < 643
ThemotionisLyapunovasymptoticallystableifitisstableandx(t) y(t)ast.
By words: a motion is Lyapunov stable if any other motion starting sufficiently close, still remains
closetoit(intermsofdisplacementandvelocity);amotionisLyapunovasymptoticallystableifany
other motion starting sufficiently close, will asymptotically approach it. The Figure 617 gives a
graphicalrepresentationofLyapunovsstabilitydefinitionbothinthehistoryplane(yvst)andinthe
phase plane ( u
& vs u). In particular note the graphical representation of free vibration near a
configurationofstaticequilibrium.
Lyapunovs definition, although simple and elegant, is not completely satisfactory from a practical
point of view, because it does require a minimum lower bound to ,andfor this reasonastructure
could be practically unstable even if theoretically stable. But the most challenging aspect is the
difficultyofcheckingthestabilityconditionasrequestedbythedefinition.
105
ECSSEHB3224A
24March2010
y y2 y
y2
y1 2
y
y1
2 y0 2
t t
0
Motion stability Equilibrium dynamic stability
u& u& u&

Static
instability
u u u
Dynamic
instability
Asymptotically stable Stable Unstable
Figure617:GraphicalrepresentationofLyapunovsstabilitydefinitions
ThedirectLyapunovmethodisancillarytothesedifficulties.Thismethodisbasedonfindingfunctionals
V(y)whicharesimilarforpropertiesandusetothetotalpotentialenergy.AccordingtoaLyapunov
theoremthesufficientconditionforstabilityistheexistenceofafunctionalV(y)withtheproperties:
V ( y ) > 0, V ( 0) = 0 644
V dy
V& (y ) = 0 645
y dt
ThestabilitycriterionofLyapunovfunctionalsisrepresentedgraphicallybyFigure618
106
ECSSEHB3224A
24March2010
y
V1< V2< V3< V4
grad(V) V& < 0

grad(V)
V y
V1
V2
V
V4 3
Figure618:GraphicalrepresentationofLyapunovsfunctionalcriterion
Unfortunately no general procedure to create these functionals is known, except for linear systems,
andforthisreasonthemethodisgenerallydifficulttobeapplied,notonlyinnumericalsimulation,
butalsoincontinuummechanicstheory.
Alternative possibility is given by the socalled indirect Lyapunovs methods, which are based on
linearizationofmotionequationsandstudyofchangesofmotionwithinputdata.
To describe such a methods, we can rewrite the force balance equation 62 in a new form more
appropriatetodynamics,byaddingtotheinternalforcestheinertiaforces(DAlembertsprinciple),
by indicating the direct dependence of internal forces from u and u& , and by indicating the direct
dependencyofexternalforcesfromthetime.Theresultingnonlineardynamicequationsare:
&& + f (u, u& ) = q(t )
Mu 646
Letusconsideranothermotion w (t ) = u (t ) + d (t ) obtainedbyaddingaperturbationmotiond(t)to
the reference motion u(t), and linearize the internal forces as in statics, but accounting for a direct
dependencyfromvelocity:
f & f
& ) = f (u, u& ) +
f (w, w d+ d 647
u& u
f
By introducing, as usual, the tangent stiffness matrix K = and in addition the damping matrix
u
f
C= ,thelinearizeddynamicequationscanthenbewritten:
u&
Mw&& + f (u, u& ) + Cd& + Kd = q(t ) 648
and can be split into equations 646 valid for reference solution u(t) and the linearized differential
equationsfortheperturbationmotion d (t )
&& + Cd& + Kd = 0
Md 649
towhichthefollowingnonhomogenousinitialconditionsapply:
107
ECSSEHB3224A
24March2010
d(0) = d 0 d& (0) = d& 0 650
todefinecompletelythelinearproblemoffreevibrations.
Wecansearchforsolutionsofthetype
d(t ) = z e pt 651
&& = p d and d& = pd inEq.649yields

Replacing d
2
( p 2 M + pC + K )z = 0 652
whichistheeigenproblemthatgovernslineardynamicstability.
Theeigenvaluesareingeneralcomplexinconjugatepairs
p = j , j = 1 653
Todistincteigenvaluescorrespondeigenmodesofthetype
e t cos t , e t sin t 654
Foraneigenvalueofmultiplicityr,thereareeigenmodesofthetype
t e t cos t t r 1e t sin t 655
From these results the stability condition is made simple and strictly related to the real part of the
eigenvalues.Ifoneofthefollowingconditionsarise,foratleastoneeigenmode:
Re( p ) > 0 (fordistincteigenvalues), 656
or
Re( p ) 0 (formultipleeigenvalues), 657
theninstabilityoccurs,becausethemotionexponentiallygrowswithtime.Moreoveriftheeigenvalue
isreal(=0),theinstabilityisofdivergencetype,otherwiseitisofoscillatorytypecalledflutter.
Graphicalrepresentationsoftheeigenvaluesp() as isvariedaretherootlocusplotsonthecomplexp2
orpplanes(seeFigure619andFigure620),ortherootamplitudeplotswhichshowthemagnitudeofp
thatis|p()| vs (see Figure621andFigure622).
108
ECSSEHB3224A
24March2010
Figure619:Rootlocusplotofdivergenceinstability(from[1])
Figure620:Rootlocusplotofflutterinstability(from[1])
109
ECSSEHB3224A
24March2010
Figure621:Rootamplitudeplotofdivergenceinstability(from[1])
Figure622:Rootamplitudeplotofflutterinstability(from[1])
In the case of static (not time dependent) loads, the commercial codes give the possibility to easily
performfullnonlinearstaticanalysis.Ifthereisalsothepossibilityofswitching,atacertainlevelof
110
ECSSEHB3224A
24March2010
load, to a freevibration analysis, indirect Lyapunov methods and variation of frequencies can be
appliedtocheckthedynamicstability.Althoughreducedtosolveeigenvalueproblems,however,the
computationeffortissignificantandstrictlyspeakingnotpracticallypossible,becausealleigenvalues
arerequiredtoverifythestabilityconditions.Moreover,astudyofdependencyofeigenvalueswith
respecttoloadparameterhastobeinvestigated.TheninpracticebothdirectandindirectLyapunov
methodsaredifficulttobeapplied,andeveniftheycontributedefinitivelytoclarifythetermsofthe
problem,sometimeotherapproachesareusedinpractice.
For the case of transient loads, the commercial codes do not always allow easily such sophisticated
analyses as required by direct or indirect Lyapunov methods. Nevertheless the possibility of
performing nonlinear transient analysis allow a procedure firstly suggested by Budiansky and Roth
[5],suggestingastabilitycriterionqualitative,butfairlywelldefined,andeasytobeapplied.
They considered a clamped spherical cup, loaded by a pressurevarying stepwise in the time, asin
Figure623(a).Thetransientnonlinearresponseisthencomputedanddisplacementoutputhistories
areobtainedfordifferentlevelsofpressureamplitude,asinFigure623(b).Theproposedcriterionis
toconsiderlossofstabilitywhen,forasmallincrementofloadamplitude,thereisalargeincrementin
output displacements. In Figure 623 (c), the maximum output displacement is plotted vs pressure
amplitude,showingajumpforapressureamplitudewhichisdefinedascritical.
Figure623:(a)Rectangularpressurehistory;(b)responsehistoriesforvarious
pressure;(c)variationofmaximumresponsewithpressure
Imperfection sensitivity also arises for dynamic stability and the most easy way of accounting for
imperfectionsseemstobethecontinuationmethods,requiringsolutionofthefulltransientnonlinear
equation (e.g. the BudianskyRoth approach). Nevertheless an initial postbuckling Koitertype
approachhasbeenrecentlyproposed[6].Thegreatestamountofinvestigationsreferstotransientload
conditionswhichcanrevealahighersensitivitywithrespecttothecorrespondingstaticload.Figure
624(takenfrom[7])representsthecaseofastructureofthetypethatbifurcatessymmetricallyand
hasastaticbucklingloadPslessthanPcifitisperfect.Dynamicanalysisshowsthatthesamestructure
withthesameimperfectionwouldhaveundersteploadadynamicbucklingloadPd.Thenasshown
111
ECSSEHB3224A
24March2010
in the figure dynamic buckling is more sensitive to imperfection, but the dynamic buckling never
occursataloadlevellessthan70%ofthestaticone.
Figure624:Dynamicbucklingduetosteploading
Forfurtherinformationondynamicbuckling,particularlyunderimpacttypeloading,seedetailed
informationanddiscussionsinChapter18of[8].
6.9 References
[1] FelippaA.C.,Lecturenotesonnonlinearfiniteelementmethod,ReportNo.CUCSSC9616
collectinglecturesatUniversityofColorado,Boulder,2001.
[2] ThompsonJ.M.T.andHuntG.W.,Ageneraltheoryofelasticstability,JohnWiley&Sons,
1973.
[3] Flgge,W.,DiestabilittderKreisszylinderschale,Ing.Arch,Vol.3,(1932),pp.463506.
[4] BabcockC.D.,Shellstability,JournalofAppliedMechanics,Vol.50,pp.935940,Dec.1983.
[5] BudianskyB.andRothS.,Axisymmetricdynamicbucklingofclampedshallowspherical
shells,NASAcollectedpapersonstabilityofshellsstructures,TN1510,(1962),pp.597606.
[6] JansenE.L.,Aperturbationmethodfornonlinearvibrationsofstructures,XXIICTAM,1521
August2004,Warsaw,Poland.
[7] BudianskyB.andHutchinsonJ.H.,Asurveyofsomebucklingproblems,NASACR66071,
(1966).
[8] SingerJ.,ArboczJ.andWellerT.,BucklingExperiments:ExperimentalMethodsinBucklingof
ThinWalledStructures,Volume2,JohnWiley&Sons,2002
112
ECSSEHB3224A
24March2010
7
Material characteristics
7.1 Overview
Structuralstabilitydependsontheappliedloadandgeometricparameterslikeslenderness,boundary
conditions and imperfections. Furthermore, it is influenced by material parameters. As long as the
load intensity is small enough to keep the material within the linear elastic range the material
parameters are merely factors for the stability limit. These factors, however, can be affected by
environmentalconditionsliketemperatureandmoisturechangesaswellasbytheloadingvelocity.
Beyondtheelasticlimitsthematerialcanshowanonlinearloaddeflectionbehaviourwhichshouldbe
considered in the stability analysis. Also the material strength is of importance. Surpassing the
strengthlimitlocallyresultsinachangeofthestructureleadingtodifferentstabilitybehaviour.
7.2 Linear elasticity and elastic properties
7.2.1 Overview
Generally, material properties depend on the effects of the environmental conditions, and on
procurement,manufacturingprocessandverification.Forcompositesinaddition,thetypeofrequired
properties depends on the level on which the analysis is executed. Is there for instance an analysis
model applied on the laminate wall level or on lamina level (macromechanical quantities for a
laminabasedFEA).Laminamodellingcanusethepropertiesofthelaminaconstituentsmatrixand
fibre(micromechanics)asmaterialinput.
Basically,itisassumedthatthestrainatamaterialpointinanelasticbodyiscompletelydetermined
bythecurrentstressandthetemperaturethere.Incasethatthethermalstateisignoredthestrainata
material point depends only on the stress at that material point, thus =(). The relation 71 is
therebyestablishedbyalinearoperator,thesocalledlinearelasticitytensor

ij = C ijkl kl 71
Itisthesimplestmodelforaconstitutiveequationwhichhowever,isessentialforstabilityanalyses.
ThelinearelasticitytensorrepresentsthegeneralisedHookeslawby9linearequationswith9x9=81
elasticity coefficients. With application of ij=ji and ij=ji and exploitation of occurring material
symmetry conditions, the number of independent elasticity coefficients (proportional to elasticity
propertiesandengineeringconstants)reduces,seeforinstanceRef[1].
113
ECSSEHB3224A
24March2010
Table71:Elasticityproperties(here,engineeringconstants)ofstructuralmaterials

ElasticityProperties

direction
1 2 3 12 23 13 12 23 13
orplane
general.
9 E1 E2 E3 G12 G 23 G13 12 23 13 comments
orthotr.
G = E /( 2 + 2 )
UD,
|| = || E / E||
5 noncrimp E|| E E G|| G G|| || ||
quasiisotropic
fabrics
23plane
EW E3 GWF GW 3 G|W 3 WF W3 W3
6 fabrics EF Warp=Fill
fabrics EW
9 EF E3 GWF GW 3 GF 3 WF F3 W3 WarpFill
general
GM=EM/(2+2M)
EM EM E3 GM GM 3 GM 3 M M3 M3 1isperpendicularto
5 mat
quasiisotropicmat
plane
2 isotropic E E E G G G G=E/(2+2)
E:=Youngsmodulus,:=Poissonsratio,G:=Shearmodulus.1or||:=paralleltothe
fibre,2or :=transversaltothefibre;3:=thicknessdirection;W:=Weft,F:=Fill,M:=
Mat. v :=(here)largerPoissonsratio
In Table 71 these characteristic elasticity properties are summarized. The respective number of
elasticitypropertiesofamaterialisrepresentedtherebythesocalledengineeringconstantsnormally
utilized as analysis input. The first column gives the number of independent properties. Shadowed
entriesindicatehowmanymutuallyindependentpropertiesamaterialpossesses.Forinstanceinthe
caseofanisotropicmaterialmodelthereexistonlytwoproperties,namelyEand.
7.2.2 Orthotropic elasticity

Withintheframeworkofbucklingitisnotrecommendedduetomissingtestexperiencetoestablish
an anisotropic material model of higher level than the general orthotropic one. Besides this, it would
require a huge amount of effort to estimate all necessary material properties. And their accuracy is
essentialtopassthematerialstabilitycriteriavisualizedbelow.
Eq. 72 defines the elasticity coefficients in the stressstrain relationship of the general orthotropic
material,whichisthemostgenerallinearmaterialmodel,wherebyC12=C21,C13=C31andC23=C32.Itis
takenfor modelling of e.g. formed ductile metal sheets and woven fabrics. Thereare9independent
coefficients, which represent the entries in the stiffness matrix [C] and the compliance matrix [S],
explainedbelow.
Asconventionsareapplied:1:=longitudinaldirectionofametalsheetorthelaminafibredirection,2:=
transversefibredirectionacrossthewidthoftheplane,3:=throughthicknessdirection.
114
ECSSEHB3224A
24March2010
1 c11 c12 c13 0 0 0 1

2 c 21 c 22 c 23 0 0 0 2

3 c31 c32 c33 0 0 0 3
= or { } = [C ] { } 72
23 0 0 0 c 44 0 0 23

13 0 0 0 0 c55 0 13

12 0 0 0 0 0 c66 12
Fororthotropicelasticitythesocalledmaterialstabilityrequiresthefollowingrestrictions:
E1 , E2 , E3 , G23 , G13 , G12 > 0 , 12 < E1 / E2 , 13 < E1 / E3 , 23 < E2 / E3 ,and
1 12 21 23 32 31 13 2 21 32 13 > 0 .
Under plane stress conditions, such as in a shell element, these restrictions simplify to:
E1 , E2 , G23 , G13 , G12 > 0 , 12 < E1 / E2 .
Theserestrictionshavetobecheckedbeforeperforminganynumericalanalysis.
Using engineering notation, engineering constants, and engineering stresses the strainstress
relationshipinlinearelasticityreadsasEq73.
1 21 31
E 0 0 0
E2 E3
1
12 1 32
1 0 0 0 1

E 1 E2 E3
2
2 13 23 1
0 0 0 or{}=[S]{}
3 E 1 E2 E3 3 73
=
23 1
0 0 23
13 G 23 13

12 1
( symm.) 0 12
G13

1
G12
Thissocalledcomplianceformulationisadvantageousfortestdataevaluation.
ThePoissonratiosarelinkedtogetherbytheMaxwellBettirelationship.

21 E1 = 12 E 2 and 31 E1 = 13 E 3 . 74

7.2.3 Transversely-isotropic elasticity of UD materials

Transverseisotropyisaspecialclassoforthotropicity,whichisappliedforthemodellingofe.g.UD
materials. Thereby, the direction 1 is aligned with the reinforcement (parallel ||) and direction 2
transverse ( ) to the reinforcement. Due to the fact that the 23 plane is a plane of isotropy, the
followingisvalid: c33 = c22 , c13 = c12 , c55 = c66 , c44 = ( c22 c23 ) / 2 .
Thematerialstabilityrequirestherestrictions:
115
ECSSEHB3224A
24March2010
E1 , E2 , G23 , G12 > 0 , 12 > 1 , 23 < E2 / E3 ,

1 12 21 23 32 31 13 2 21 32 13 > 0 .
TheMaxwellBettirelationsintroducedbeforearestillvalid.
Switchingtotheselfexplaining,symbolicallydenoted(donebytheoldcompositeresearchershalfa
centuryago)propertiesthefollowingfourelasticityconstantshavetobeprovidedwhenperforminga
2Danalysis E , E , G , .Inthecaseof3DanalysesthetransversePoissonsratio ofthequasi
isotropicplaneisadditionallyneeded.
7.2.4 Isotropic elasticity

Itisthesimplestmodelforaconstitutiveequationanditispreferablyappliedformetallicmaterials.
Thematerialsymmetryconditionsleadto c11 = c22 , c23 = c12 , c66 = ( c11 c12 ) / 2 andthenumberof
independentpropertiesisreducedto2.
The material stability criterion requires that E > 0 , G > 0 , and 1 < < 0.5 . Values of
approaching0.5indicatenearlycompressiblebehaviour.
7.2.5 Viscoelasticity
Thismodelisappliedtomaterialswhichexhibitbothelasticandviscousbehaviour.Thelatteroneis
caused by the dissipation of energy due to several internal loss mechanisms (e.g. generation of
substructures in metals, bond breakage, dislocations). Viscoelastic constitutive models typically are
used when modelling polymeric materials. For details see Schapery [2] and e.g. VDI2014 guideline
[3].
7.3 Strength properties and hygro-thermal properties
7.3.1 Strength properties

7.3.1.1 Metals
Strength properties for onset of yielding R0.2 and onset of fracture Rm heavily depend on
manufacture, environment, grain direction, tempering. The susceptibility to environmental effects
suchastemperature,humidity,radiation,andcyclicloadinghastobetakenintoaccountwhentesting
and designing. By the establishment of the data one has to consider the most recent edition of the
normativedocumentsindicatedinStandardECSSE3208[4].Foraspecificprojecttherequireddata
maybetailoredtomatchtheparticularprofileandcircumstancesoftheprojectinagreementwiththe
customer.
Essential for structural stability in case of the usually ductile lightweight isotropic materials is
yielding and in this context due to bending both the yield strengths, for tension R p 0.2 and for
compression Rc 0.2 areneeded.
7.3.1.2 Composites
Strengthpropertiesheavilydependoncuringcycle,environment,andfibredirection.Fortheactually
usedbrittlecompositesnoyieldstrengthequivalentstrengthpropertyexistsfortheformulationofa
116
ECSSEHB3224A
24March2010
singlecommonyieldfailurecondition(seesectionplasticity).InterFibreFailure(IFF),tosomeextent,
replaces yielding. Even in a 2D strength analysis there are five strength properties, two for fibre
t c t c
(fracture)failure(FF), R , R ,andthreeforinterfibrefailure(IFF), R , R , R .
7.3.2 Hygro-thermal properties

7.3.2.1 General
Hygrothermal properties are essential for the engineer when designing to stability. These are in
generaltheCTEandtheCMEproperties.Table72showswhichpropertiesarerequiredforwhattype
ofmaterial.Thermalandmoisturephenomenacanbeidealisedbysimilarapproacheswhichcanbe
visualized by the units: CTE by a change in temperature[ mm /( mm K ) ] andCME by a changein
moisture[ mm /( mm %) ].
7.3.2.2 Metals
Aluminium, steel and titanium change their behaviours with varying temperature especially when
goingintothecryogenicdomain[5].
Table72:Hygrothermalpropertiesforstructuralmaterials.
T:=Thermal,M:=Moisture
Hygrothermalproperties
direction,or
1 2 3 1 2 3
plane
general
9 T1 T 2 T 3 M1 M2 M3
orthotropic
UD,
5 noncrimp T || T T M || M M
fabrics
6 fabrics TW TW T 3 MW MW M3
fabrics
9 W F F MW MF M3
general
5 mat TM TM TM 3 MM MM MM 3
isotropic
2 T T T M M M
forcomparison
Coefficients of Thermal Coefficients of Mois-

Expansion (CTE) ture Expansion (CME)
7.3.2.3 Composites
Thedifferenceinthermalormoistureexpansion(forcompositesessential)betweentheconstituentsof
acompositeoracoatedheterogeneousmetallicmaterialcaninducelargestressesorstrainsandcan
117
ECSSEHB3224A
24March2010
lead to failure. Therefore, temperature and relative humidity of the manufacture and storage
environments should be controlled and monitored in order to provide the designer with reliable
material data. This is the more mandatory when conditioning test specimens. Further, high stresses
canbecausedbyoccurringhightemperaturegradients.
The constituents of the composite possess different CTEs. Their effects on filamentmatrix level are
assumed to be included in the CTE property values of the lamina. The CTEs of a UD lamina are
denoted T , T ,see[5].
Composites can absorb significant amounts of moisture in the polymeric matrix and in case of
aramide fibres (in addition these are brushing under compression) in the fibre, too. Also in the
filamentmatrixinterfacemoisturecanbestoredandreleased.ThemoisturecontentMisdefinedas
theratioofthemoisturetakenupbythecompositetothedrymassofthecomposite.TheCMEsofthe
UDlaminamaterial M , M aredefinedinasimilarwaytotheCTEs.
WithintheoperationalregimethecoefficientsCTEsandCMEsareusuallyaveraged.CTEaswellas
CME may essentially vary with the temperature and moisture content, see [6], HSB sheets 3710501
and 3710302. For instance, GFRP and CFRP can even change the behaviour to the opposite when
goingfromroomtemperaturedowntocryogenictemperature.
7.4 Elastic and inelastic material behaviour
7.4.1 Overview
Linear analyses are convenient approximations which are often adequate fordesign purposes. They
havetheadvantagethattheprincipleofsuperpositionofloadcasescanbeapplied.Thisstatementis
validaslongastheboundaryconditionsremainunchangedandthedeformationsaresmall.Thereare
threesourcesofnonlinearity:material,boundaryandgeometricnonlinearity.
This section is focused on the available models used for the characterisation of materially nonlinear
phenomena with restriction to stability analysis. In advance, it is however stated that beside metals
manymaterialsincludingcompositescan,withinlimits,becharacterisedbylinearelasticity.Aslong
asthestressstateremainsinthesocalledelasticrangethematerialresponseisnearlylinearlyelastic.
For explanations to this subject it is referred to the literature. An extensive representation of these
topicscanbefoundinReferences[7],[8]and[9].
Inthecaseofaninelasticbodythestrainresponseinthematerialpointcannolongerbedetermined
bythestressstateandtemperaturealone.Moreover,theresponsealsodependsoninternalvariables
, such that = (,T,). Ifthe internal variables remain constant at constantstressand temperature,
thenthelocalstate(,T,)iscalledalocalequilibriumstate.Ifthisrequirementisviolatedtheprocess
isirreversible,whichisprovenbyeffectsofcreepandrelaxation.
Itisfurtherobservedthatwithincreasingplasticyieldingthevolumechangebecomesmoreandmore
negligible(0.5).Thereforetheplasticdeformationisprimarilycharacterisedbyadistortion,which
canbedescribedbytheseconddeviatorinvariant J 2 ofthestressstate.
Theconditionthatplasticstrains willoccurisdefinedbytheexistenceofayieldfunctionF(),
pl
wheretheadmissiblestateisconstrainedtolieontheboundaryofF,thusF()=1.Mandatoryforthe
accurate definition of the yield surface is whether the material is dense (nonporous) or porous by
voidsinvirginmaterialorvoidsnucleatedandgrownunderatriaxialtensilestressstate,e.g.nearthe
cracktipsofametal.However,isotropicmaterialmodelswhichalsoincludethebehaviourwithinthe
latermentionedstrainsofteningdomainarenormallynotusedforbucklinganalysis,becausethelocal
118
ECSSEHB3224A
24March2010
predictions would violate applicable design requirement limits. Ductile behaving materials are
normallynotusedbeyondthecompressiveyieldstrength.
7.4.2 Stress-strain curve of isotropic materials

The transition from elastic to plastic behaviour occurs at a certain stress level, known as the elastic
limit point, where the proportionality between stress and strain starts to be violated. For isotropic
materialswithductilebehaviourthisstresslevelisengineeringlikedefinedfordesignpurposeswith
theoccurrenceof0.2%permanentplasticstrain(0.2%offset),whichisobtainedafterunloading.The
associatedstressesaretheyieldstrength R0.2 .Thefullstressstraincurveisdeterminedfromtestdata
(seeFigure71a).Duetothescatter,themeasuredindividualcurvescanbedifferent.Whentestinga
largenumberofspecimens,enoughdatacanbeobtainedtodeterminethescatterbandofthestress
strain curve of the particular material and the average stressstrain curve needed for an analysis
providing the engineer with the average structural behaviour which represents the highest
expectance.
Maximum point of the curve is the (ultimate) tensile strength Rm Rmt determined from the
maximum load as Rm = Fm / A0 . The plastic strain at Rm is called the plastic strain at the end of
uniform elongation ( Aus in %). Beyond this strain limit, full plasticity occurs, necking begins in the
tensilerodspecimen,andstressesarenolongeruniaxial,[10].
Uptothecompressiveyieldstrength Rc 0.2 thecompressivestressstraincurvehasashapesimilarto

thatofthetensilestressstraincurve.Thevaluesofthemechanicalpropertiesareusuallydifferentfor
tensionandcompression.Athighcompressivestresses,alargepseudoincreaseofthecrosssectional
area of the test specimen can occur (barrelling), with a theoretical increase of the compressive
engineeringstress.Note:Acompressivestrengthdoesnotexistformaterialswithductilebehaviour,
c
andtheultimatecompressivestrength R m existsinbrittlecases,only.
Duetobending,tensileandcompressivestressstraincurveshavetobeusedforstabilityanalyses.In
thetensiledomainsocalledengineeringandtruestressstraincurves(forcollapseloadanalyses)are
employed,coveringtherange R m .Atruestressistheaveragestressintheinstantaneousoriginal
t
minimumcrosssectionalarea.Itcanbecomputedtogetherwiththetruestrainfromtheengineering
stressandstrainasfollows(see[6],HBSsheets5211002)

true
= F / A = ( 1 + ) , true
= ln( 1 + ) . 75

Thetrueultimatestrengthisgivenby(seeFigure71b).
Rmtrue = Rm ( 1 + Rm ) 76
Allstrainsandstresses,above,areaccordingtotheirdefinitionuniaxialquantities.
119
ECSSEHB3224A
24March2010
Figure71:a)Testdataofaductilematerialb)Engineeringvstruestressstrain
curve
In nonlinear FEA often true stresses are the output. This requires as input a true stresstrue strain
curveandaswellasforaconsistentjudgementatruestrengthvalue.
7.4.3 Special aspects with composites

Thefullfibredominatedstressstraincurve
t
( ) andthemainlyfibredominatedcurve ( )
t c c
canbeconsideredlinearintheusualstabilityanalysisapplicationregimes.Twoofthethreematrix
dominatedcurvesinFigure72arenonlinear,
c
( ) and ( ) ,only ( ) islinear.ForUD
c

t

t

composites, the combined quasiyielding effect of the three inter fibre failure (IFF) modes (multiple
modes)canreplacetheMisesyieldingmode(singlemode)fortheductileisotropicmaterials.
Eachcurveinthefigurecanbenormalizedbythecorrespondingstrength.Thenacurve,obtainedfor
onemode,representsasocalledmodestresseffortcurveoftheUDmaterial, Eff = / R .Astress
effortEff=100%meansexhaustionofthematerial.TheIFFrelatedstressstraincurvesofaUDlamina
in Figure 72 exhibit a strainhardening (isolated lamina) and a strainsoftening part (if lamina is
embedded).
Figure72:ExamplefornonlinearstressstraincurvesofaCFRPUDlamina
embeddedinthelaminatestack(insitubehaviourreflectingcurve,strain
softeningpart,assumed)
120
ECSSEHB3224A
24March2010
7.4.4 Mapping of a stress-strain curve
Widely applied as mapping function (because a parameter source is available and because it often
describestheusuallightweightmaterialswell),istheRambergOsgood(R/O)relationshipwhichisa
monotonously increasing function. Using this relationship, the elastic strain and additional plastic
strainaredeterminedseparately.Theelasticstrainfollowsfrom = / E ,theplasticstrainportionis
assumedtofollowthepowerrelationship(see[6],HBSsheets5211002).
ln ( Aus / 0 ,2% )
n
77
= + 0.002 with n =
E R0.2 ln (Rm / R0.2 )
An example for the application of this equation is shown in Figure 71 b. Using Eq. 77, the stress
straincurveis determined by the mechanical properties E and R0.2 and by the material dependent
R/Oexponent n .TypicalR/OcurvesarebasedonanaverageEandanaverage R0.2 (mindthebar).
Valuesof n arederivedfromtestdata.Fortensionandcompressionthesameequationisused,with
different values of E , R0.2 and n for tension and compression. Applying engineering stresses and
strains,theR/Orelationshipoften yieldsanadequateapproximationofthetestdataforstressesupto
the yield strength. If true stresses and strains are used, the region where a good approximation is
obtainedcanbeextendeduptothestrainwherelocalneckingstartsat Rm .
The value of n depends on material, condition, product form and grain direction. Usually, only
typicalvaluesofEandncanbeobtainedfromliteratureandmaterialdatasheets.Thisresultsinaso
calledtypical R/Ocurvegoingthroughthatpoint,whichreflectsthetypical(average)yieldstrength.
Ifnotestdataisavailabletodeterminethevalueof n ,valuesfromliteratureareused.Typicalvalues
of n arepresentedintheMMPDS2(formerMILHdbk5)[11].Itshouldbenotedthatavalueof n is
always related to specific values of E and R p 0.2 . For a typical stressstrain curve the value of n is
differentfromthevalueforaminimumcurve.Also,dataforthecompressioncurvearedifferentand
mighthavetobeassessed.
7.4.5 Cyclic hardening behaviour

If the stress is slightly above the elastic limit, irreversible strains will occur after the process of
unloading.Thematerialisthencharacterisedbypermanentstrain.Incaseofreloadingthelevelofthe
yield strength will increase against the virgin value. This behaviour is defined as cyclic hardening
behaviour of a material. Cyclic hardening effects are to be considered if they are essential before
applyingtheDesignBucklingLoad.Therearetwotypes:isotropicandkinematichardening.
Isotropichardeningofamaterialcanbeapplied,if
thecentreoftheelasticrangeremainsattheorigin,independentoftheloaddirection
itislinearintheamountofplasticflowandindependentofitssign.
The yield function is therefore characterized by F = F (,), where > 0 represents the cyclic
hardening (not to be mixed up with strainhardening) parameter. The simplest evolutionary
equationforthehardeningparameteris:

& = & pl
= & dt .
pl
78

The isotropic hardening model is a coarse simplification of the real behaviour of materials.
Often,isotropichardeningisnotsufficientforproblemswherecyclicloadingsarespecified.
121
ECSSEHB3224A
24March2010
At kinematic hardening it can be observed that the centre of the elastic range undergoes a
motionintothedirectionoftheplasticflow(Bauschingereffect).Consequently,incaseofcyclic
loadings above the proportionality limit the yield strengths become different between tension
andcompression.ForthismaterialmodeltheyieldfunctionisthereforetobeextendedtoF=F
(,,),whererepresentsthecentreoftheyieldsurface.
7.5 Plasticity and damage

An increasingly strained ductile behaving material usually undergoes the initiation of damage
followed by the onset of progressive damage. The initiation is caused by little crystal texture
dislocations or flaws at the microscale which increase in size during the yielding process. So, both
phenomenaplasticityanddamageareencounteredbeyondthelinearregimeofastressstraincurve
andcanbeseentohavethesameeffect:Ductilebehavingmetalspossessnonlinearstressstraincurves
duetoyieldingandbrittlecompositesduetomicrodamage.Thislowersthestiffness.
Onset of yielding of nonporous metallic isotropic materials can be described by Mises (accurate:
HuberHenckyMises) shear yielding. Thereby it is assumed that for yielding the specific strain
energycausedbythedeviatorofthestressstateisresponsible.Subsequentyieldingleadstoagrowth
oftheMisescylinder.Fullyplasticsolutionsareonlyappliedinspecialcaseswheretheknowledgeof
the plastic collapse load is essential for proving and/or saving a stabilityendangered structural
component.
In case of anisotropic yielding (metal sheet where the grain directions can have different properties
duetothemanufacturingprocesssuchasrolling)thenHillswellknownyieldsurfaceformulationis
applied. It is a simple extension of the Mises formulation containing six additional material
parameters which are determined by tests in different directions. For more details the reader is
referredtotherespectiveliteratureandtothetheorymanualoftheappliedFEcode.
Actual UD materialcomposed brittle behaving composites show yielding phenomena only
marginally, by the constituent matrix on the microscale, and therefore these can be omitted. In the
observedmicrostructuraldamageprocessesfilament,matrixandinterfaceareinvolved.Thetaskisto
formulate a damage initiation condition analogously to the Mises shear yield condition which
considers the materials stiffness loss due to progressive degradation represented by the growing
Misescylinderwhentheequivalentstressexceeds R0.2 .
Addressingbrittlecomposites,thesingleyieldfailuremodeofthemetalispracticallyreplacedbythe
multiple interfibrefailure modes which interact micromechanically and form after interaction a
commonsinglequasiyieldconditionagain,[12].Thisconditionisadamageevolutionconditionand
shouldbeconsideredwhentacklingthenonlinearitywithitsstiffnessloweringeffect.
In the WorldWideFailureExercise (WWFEI [13]) two failure condition sets (criteria) matched the
test results best: Pucks ActionPlane failure conditions, and Cuntzes Failure Mode Concept. For
compression, the currently running WWFEII [12] will provide designers with validated, reliable
failureconditions.TsaiWuiserrorproneinthecompressiondomain.
7.6 Material testing methods, test data, and evaluation
7.6.1 Overview
Before executing tests, it is recommended to assure that the material is still commercially available.
Significantarethestaticuniaxialmechanicalpropertytests:tensile,compression,shear,bending,and
122
ECSSEHB3224A
24March2010
impact.Alsoimportantistheknowledgeofthephysicalproperties.Offurtherinterestforstabilityis
creep testing in order to determine the viscoelastic properties. For isotropic materials testing is
standardized,see[14],[15].Forcomposites,theexecutionoftestswithtestspecimenssubjectedtoa
multiaxialstressstateismuchmorecomplicatedandthefailurepredictionevenfor2Dstressstatesis
not sufficiently validated. For these materials formspecific tests unique to filament winding, to
braiding,ortothicksectioncompositeshadtobedeveloped,seee.g.[16].
If data from different test series are available one can use the statistical acceptance rules to decide
whethertheymightbecombinedinonelotornot(seee.g.[16],sheets04200).So,asthereareseldom
enough reliable properties available when dimensioning, the socalled population sampling is a
helpfultooltoincreasereliabilitybyutilizingallinformationavailable([6],[16]).
Dependent on the needed thickness of isotropic material, flat specimens (for thin sheet material) or
roundbarsaretested:E.g.see[15],[16],[17]forstandardtestmethods.
Figure73:Coordinateaxes
In case of buckling application clad sheet and plates and unclad plates are to be distinguished. As
directionsL(longitudinal),LT(longitudinaltransversal),andsometimesST(thicknessdirection)are
considered in order to apply accurate dedicated properties. This is similar to the composites. Heat
treatment,applicability,andoperationaltemperaturelimitsarealsotoberegarded.
UD laminas have widespread aerospace applications. Its test specimens are socalled isolated
specimenswhichactaccordingtotheweakestlinkprinciple.Intension/compression/torsiontestrigs
thepropertiesaremeasured.Thereferences[16],[18]and[19]arevaluablesourcesforfindingthetest
specimens and information about 2D stress states testing. Special effort is put on multiaxial
compressiontestsintheWWFEII[12].
Wovenfabricsareimportantforstructureswithdoublecurvatures(antennadishesetc.)andirregular
shapes where they demonstrate the draping advantages. There are biaxial and triaxial singleply
woven fabric composites applied in structural components. Here the reader is referred to the
StructuralMaterialsHandbook[18]whichisgoingtobeissuedbyECSS.Inplanecompressiveaswell
as tensile mechanical properties of woven fabrics are inferior to those of UDcomposite laminates.
Theirvaluesdependonweavetype(plain,twill,satin)andtheratioofthereinforcementinwarpand
filldirection.Layupandtypeofweaveinfluencethefibrevolumecontentofthefinalcompositeand
therebythestiffness.
Theoftenappliedsandwichesconsistofthefacematerialwhichmaybeastackofdifferentlyoriented
UDlaminaeorofwovenfabrics.Themostcommontypesofcorematerialsare:Nomexhoneycomb,
vinylsheetfoam,polyurethanefoam,corrugatedshells,andendgrainBalsa.Thesandwichesshould
betested[15]todeterminecompressiveandbendingstiffnessaswellasthestrengths.Forthecoresit
shouldbenotedthatthestiffnessinlongitudinaldirectioncanbeprettydifferentfromthoseinlateral
orinthicknessdirection.
In many companies individual company test standards are used, sometimes created due to missing
officialstandards.Thesestandardsshouldbeapprovedbytherespectiveaerospaceauthority.
For further information on testing of aerospace structures the interested reader should consult
chapters12and13ofthishandbook.
123
ECSSEHB3224A
24March2010
7.6.2 Determination of a design allowable
Minimum values for certain mechanical properties are assigned as part of the procurement
specificationrequirements.
In the case of strength properties, distinct statisticallybased minimum values are assigned for final
designverification.Theyaretermeddesignallowablesorbetterstrengthdesignallowablesbecausethey
areonlystrengthrelated.Suchadesignallowableisavalue,abovewhichatleast99%(Avalue)or
90% (B value) of the population of values is expected to fall, with a 95% confidence level. The
necessarydatabasisforthestrengthdesignallowablesisdescribedintheactualissueofMMPDS02
[11], formerly MILHDBK5 [20]. For a preliminary statistically based strength value the minimum
numberoftestdatais20foreachbatchoutof3batches.
In the design allowablesdefining standard MILHDBK5 stability as another resistance quantity
besidesstrengthwasnotaddressed.
From predimensioning until final verification different property data, sometimes available in
handbooks,areused.Theyaredefinedasfollows:
Svalue:Procurement,valueonthesafeside.
A, Bvalue: Strength design allowables= statistically defined according to MILHDBK5.
Numberofdifferentbatchesisrequired.
T99/T90value: Material strength values. The determination follows the same statistical
procedure as with the strength design allowables. However, the data volume and batch
requirementsarelessstringent.A>S,onlyallowedifpremiumselectionofmaterialisapplied.
NormallyT99 A=S.
7.6.3 Sources for finding property data

Theliteraturesurveyattheendofthischaptercontainssomesourceswherematerialpropertiescan
be found. These are: Ref. [16] for polymer matrix composites, Ref. [11] for metal, inReferences [18],
[21] and [22] for sandwich, and in Ref. [5] for cryogenic materials. Company sources are welcomed
suchasRohacell[23]forsandwichmaterial.
7.7 Some practical aspects for stability analyses
7.7.1 Effect of yielding and damage on stiffness

Buckling is faced by the strength failure mode onset of yielding, due to the fact that the utilized
metallic light weight materials rarely show brittle behaviour. Increasing yielding locally and later
globallywhenloadinguptoDesignBucklingLoadlevel,isaccompaniedbyastiffnessreduction.In
this context, for the stability analyst of interest is how the materials stiffness in a local material
instability point of the buckling structure really decreases with increasing stress effort Eff of the
material.
Figure 74 demonstrates that design allowable R and average strength R lead to different stress
effortsindesignverification(considersthedesignFoS)andinmodellingofmaterialdamaging(50%
value=highestexpectance).
Whereas in the case of the metallic light weight materials an overload should only lead to a local
yielding(less0.2%,inthecriticalmaterialpointsofthestructure)thesituationforthebrittle(actually
applied) composites, most often CFRP (carbon fibre reinforced plastic), leads to a partial micro
124
ECSSEHB3224A
24March2010
mechanicaldamageinorbetweenthelaminae,whichmaybetermedpseudoorquasiyielding.Both
situationsaffectaredistributionofthelocalstresses,howevermuchearlierforthelaminatethanfor
theisotropicmetalduetothehigherstressconcentrationcausedbytheanisotropy.
Theinterfibrefailure(IFF)modesshowbrittlebehaviourhowevertheredundantlaminatewallmay
behavequasiductile.Initiationofabucklingfailuremodeisprimarilylinkedto2Dstatesofstress.
Eff = eq / R .,D:=damage,MoS:=MarginofSafety
Figure74:Visualizationofthedevelopmentofstresseffort,strengthvalue,
equivalentstrength,anddamage.
Veryessentialforstabilityisadelaminationinthelaminatewallbecausethecompressivestiffnessis
reducedafteronsetofdelamination.Delaminationcanoccurintensileshearcasesandcompression
shearcases,andfurtherifbendsinthestructurearestretchedorcompressedwhichgeneratesstresses
acrossthewallthickness.Consideringdelaminationa3Dstressstateistobeused.Therearespecific
failure conditions to predict onset of delamination such as from Hashin [24]. Also the lamina 3D
failure conditions of Puck [25] and Cuntze [12], used in the WorldWide Failure Exercise, can be
applied for the prediction of the laminate failure mode delamination. A practical condition to
predictdelaminationgrowthanditseffectonfurtherstiffnesslossisnotyetavailable.
Thetransferofisolatedlaminadatatothelaminateanalysisisusualpractice.Theprerequisiteisthat
theeffectsofthepropertiesinthelaminateareequaltothoseinthelamina.Ingeneral,thisisnottrue
becausethelaminatepossessesredundancyoverthelaminastack.Therefore,somehealingeffectsare
actinginthelaminate,whichusuallyleadtoalowercoefficientofvariation(CoV)thanarefoundfor
thelaminaewhereasthemeanvaluecanbelower.Embeddedlaminasbehavemorebenignthanthe
weakestlinkdrivenisolatedones.
Foroftenusedcladaluminiumalloyplatesusually2valuesofthemodulusofelasticityarepresented,
theprimaryandthesecondarymodulus.Theinitialslope,orprimarymodulus,denotesaresponseof
both the lowstrength cladding and higher strength core elastic behaviours. For commonly used
aluminium alloy sheets, differences between both moduli are 5 10 %. For the analysis, usually the
(lower) secondary modulus is used. This means that for low stresses a stiffness is used which is
slightlytoosmall.
125
ECSSEHB3224A
24March2010
7.7.2 Effect of temperature and moisture on property values
Metallic CTE (coefficients of thermal expansion) are temperaturesensitive properties. It is observed
that the CTE of the aluminium alloy AA 2219 show the largest change with temperature variation
down to cryogenic temperature. The CTE of steel and titanium shows the same characteristic,
howeverthedropisdistinctivelysmaller.
The influence of temperature on the yield strength is different and again depends on the specific
material. Therefore, it is to be checked whether the value will increase (most often in cryogenic
regime) or decrease in the specified temperature operational regime. Especially in the cryogenic
regimethereisnocleartendency.Youngsmodulusofthemetalsremainsconstantwiththetendency
of a very little increase at cryogenic temperatures. The decrease of ductility caused by lower
temperature leads to a reduction of the elongation for nearly every metal except for the aluminium
alloyAA2219whoseductilityincreaseswithdecreaseoftemperature.
CompositeCTEsandCMEsaresensitivetotemperatureandmoisture.Thetendencycanevenchange
betweenCFRPandGFRPwhichisessentialfordesignandshouldbechecked.Essentialforstabilityis
the fact that there should not be a moisture pickup possible which would violate the load transfer
capability of the adhesive. There is a temperature dependency of the stressstrain relationship, the
yieldandthetensilestrength,thefailurestrainsandfortheYoungsmodulusaverylittleincreaseat
cryogenictemperatures.
The properties of the builtin composite materials are susceptible to changes induced by the
absorption,e.g.moistureuptakebeforethelaunchofthespacecraft,andthendesorptionofpolymer
basedmaterialsoccurringasthespacevacuuminduceshygrothermalstresses.
7.7.3 Visco-plasticity, strain rate and impact

Atimedependentcreepbehaviourshouldbeavoidedbyanaccuratestabilitydesign.Requirementin
stability design is to avoid viscoplastic behaviour. Therefore, the designer is forced to check the
selectedmaterialanditsbehaviour,aswellasthepredictedstresslevel.
Fastloadingshaveaneffectonthematerialsresponse.Oftenthepropertiesincrease.
Impactofacertainlevelcausesdamageandaffectsreductionsofthedifferentmechanicalproperties
and of the performance. In addition, for laminates the impact damage is often barely visible and its
judgementisdifficult.
7.7.4 Miscellaneous
Metalliclightweightmaterialsusuallyappliedinaerospacearebehavingductile.Theyexhibit
notmuchsensitivitytostressconcentration.Residualstressescausedbymanufacturingareto
beconsideredinstabilityanalysiswhenjudgedtobeeffective.
Ductilemetalsyield.Thereforewithrespecttothedesignrequirementsnostrainsbeyond Rc 0.2
arepermittedinusualstabilitydesign.Forstabilityanalyses(compression,shear)ofmetalsitis
usualtoapplyintheplasticregion

E
= 0.5 sec (0.5 el ) 79
E

Building block of a laminate is the lamina which is relatively brittle in case of the usually
applied fibre reinforced plastics (FRP). Nevertheless, the laminate can behave quasiductile
according to the redundancy built in by the stack and its stacking sequence. UDcomposed
126
ECSSEHB3224A
24March2010
laminatesexhibithighsensitivitytostressconcentration.Theeffectofmicrocrackgrowthalters
elasticitypropertiesaswellastheCTEandCME,thevirginvalueschange.
Itisgenerallyassumedthattheeffectofthemicromechanicalresidualstressfromcuringonthe
strength in the lamina material test specimen is the same as in the (embedded) lamina of the
real structural wall. Therefore, this residual stress is regarded to be incorporated in the
macromechanicalmaterialtestdataapplied.Howeverindimensioning,theresidualstressesof
thelaminatearedeterminedandtakenintoaccountbysuperimposingthemontotheexternal
loadstresses.
Thebehaviourofanembeddedlaminaisdifferent(duetoredundancywithinthestack)tothat
ofanisolatedUDtestspecimen(nonembedded=notredundant)laminawhereoneobtainsthe
propertydatafrom.
Futureloadcarryingstructuralcomponentswillbebuiltfromwovenfabrics,too.Suchlaminae,
if consisting of traditional FRP, behave brittle especially due to the fibre crossing effect. They
experience still at low stresses IFF (inter fibre failure) and thereby degradation. This is to be
considered in stability analysis with respect to stiffness reduction. Fabrics exhibit high
sensitivitytostressconcentration.
In nonlinear FEA often true stresses are the output. This requires as input a true stresstrue
straincurveandaswellasforaconsistentjudgementatruestrengthvalue.
Adesigncurveforyieldingofahomogeneousisotropicmaterialor,ashere,adesigncurvefor
IFF of a UD material is most often assumed to be a shrunken test datamapping curve; the
average (mean, if enough test results) strength values are replaced by the strength design
allowablesR(nobarover)(seeFigure75).
Fromengineeringpointofview,forlinearelasticanalysestheinitialmodulusforcompression
andtensioncanbetakenthesame.
Figure75:Designcurve,mappingcurve,andloadstressesofaUDmaterial.
Mapping: F ( i , Ri ) = F ( 2 , 21 , Rt , Rc , R ) = 1 ,
Designcurve: F ( i , Ri ) = F ( 2 , 21 , Rt , Rc , R ) = 1 .
127
ECSSEHB3224A
24March2010
7.8 References
[1] Jones,R.M.,MechanicsofCompositeMaterials,Taylor&Francis,Inc.1999,Philadelphia,PA,
ISBN156032712X.
[2] Schapery,R.A.,StressAnalysisofViscoelasticCompositeMaterials,JournalofComposite
MaterialsJuly1967,pp.228267.
[3] Anonymous,VDI2014guideline,sheet3DevelopmentofFibreReinforcedPlastic
Components,Analysis.BeuthVerlag,Berlin;(inGermanandEnglish),2006.
[4] Anonymous,ECSSE3208,Materials,2008.
[5] Anonymous,CryogenicMaterialsDataHandbook,AFMLTDR64280VolumeI,1970.
[6] Anonymous,HSBHandbookFundamentalsandMethodsfortheDesignandEvaluationof
AeronauticalStructures.TechnicalInformationLibrary,Hannover,2008.
[7] Crisfied,M.A.:NonlinearFiniteElementAnalysisofSolidsandStructures;Vol1,JohnWiley&
SonsLtd,1994.
[8] SimoJ.C.andHughesT.J.R.:ComputationalInelasticity.Volume7,InterdisciplinaryApplied
Mathematics,Springer,1998.
[9] ChenW.F.andZhangH.:StructuralPlasticity.Springer,1991.
[10] Moosbrucker,C.:Representationofstressstrainbehaviour.ASMInternational2002,2ndedition.
[11] Anonymous,MMPDS2,MetallicMaterialPropertiesDevelopmentandStandardization
(MMPDS).Battelle.FormerlyMILHDBK5MetallicMaterialsandElementsforAerospace
VehicleStructures.
[12] Cuntze,R.G:ThePredictiveCapabilityofFailureModeConceptbasedStrengthConditionsfor
LaminatesComposedofUDLaminaeunderTriaxialStressStates,(WWFEII).PartA.
CompositesScienceandTechnology,2008(tobepublished).
[13] Cuntze,R.G:ThePredictiveCapabilityofFailureModeConceptbasedStrengthCriteriafor
MultidirectionalLaminates,(WWFEI).PartB.CompositesScienceandTechnology64,487516,
2004.
[14] Anonymous,ECSSQST7045,Standardmethodsformechanicaltestingofmetallicmaterials,
2008.
[15] Anonymous,EN20021and2,Aerospaceseries:Metallicmaterials,testmethods,tensile
testing,2006.
[16] Anonymous,MILHDBK171E,PolymerMatrixComposites:Vol.IGuidelinesfor
CharacterizationofstructuralMaterials;Vol.IIMaterialProperties;Vol.IIIUtilizationof
Data,Dept.ofDefence(DOD),TechomingPublishingUSA,1997.
[17] Anonymous,ASTME8/E8M08,StandardTestMethodsforTensionTestingofMetallic
Materials
[18] Anonymous,ECSSEHB3220,StructuralMaterialsHandbook(tobepublished)
[19] Tarnopolskii,Y.M.andKincis,T.:StaticTestMethodsforComposites.VanNostrand,NewYork.
[20] Anonymous,MetallicMaterialsandElementsforFlightVehicleStructures,MilitaryHandbook
MILHDBK5A,Feb.8,1966.
[21] Anonymous,EngineeringSciencesDataUnit(ESDU)AerospaceCollection:MetallicMaterials
DataHandbookseriesandCompositeseries
[22] Anonymous,MILHDBK23,SandwichComposites,1968.SupersedingMILHDBK23,PartI,
ANC23,PartIIandMILHDBK23,PartIII.
[23] Anonymous,RohacellTechnicalDataSheets,EvonikDegussa
128
ECSSEHB3224A
24March2010
[24] Hashin,Z.,1980,FailureCriteriaforUnidirectionalFibreComposites,J.Appl.Mech.47,329
334.
[25] Puck,A.andSchurmann,H.,2002,FailureAnalysisofFRPLaminatesbyMeansofPhysically
BasedPhenomenologycalModels,Comps.Sci.andTechnology,62,16331662.
129
ECSSEHB3224A
24March2010
8
Design to stability and its verification by
analysis
8.1 Introduction
Design verification is the process whereby a structural design is comprehensively examined and
qualificationtestedtoensurethatitwillperformintherequiredway,beforeandduringoperational
use.
The verification process is a fundamental step in the development of a successful space system. A
combination of different verification methodologies such as test, analysis, inspection is used
throughout the various phases in product development. The development logic of the structural
productanditsdevelopmentphasesareshowninFigure81.
8.2 Design development process

Following Reference [1] designing isan iterative process whereby various concepts are evolvedand
evaluated against a set of specified design requirements and constraints from manufacturing. The
requirements essentially consist of the Functional Requirements which describe What must be done?
and of the Operational Requirements How absolutely must it be done? During Conceptual Design,
supportedbytradestudies,thefeasibilityandanestimationofcostandriskisestablished.
Structural integrity should be provided during design development, manufacture, and service. In
stability,asimpleloadrelatedMoSdeterminationisnotalwayspossible.Moreovertheconsequences
ofabuckledconfigurationhavetobeinvestigatedwithrespecttochangedequilibriumconditionsand
design requirements. The following questions can appear: Is the branch to a new equilibrium path
characterised by a benign behaviour or not? Does the structure in the deformed state violate any
strengthoroperationalrequirement?
Inthestabilitydesignprocessonefacesnonlinearbucklingphenomenaevenwhendisregardingthe
prebuckling state. Therefore this topic has to be handled very carefully. Experience, wellfounded
theories and last but not least tests performed in the past should be if available involved in the
decisionsduringthedimensioningprocess.
130
ECSSEHB3224A
24March2010
Figure81:Generaldescriptionofadesignfindingprocess
8.3 Analysis pre-work and load input data
8.3.1 Overview
Analysis aims to predict and therefore models to a certain level the response of a structure. The
accuracy is strongly dependent on the input values. At least adequately defined properties and an
estimateofthescatterexpectedforeachdesignparametershouldbeavailable.
131
ECSSEHB3224A
24March2010
Robustdesignisrequiredwithidentificationofthemoststabilitysensitivedesignparameters.Robust
means: being not sensitive to modifications of relevant design parameters. Robustness furthermore
reducesthedevelopmentrisk.
Completeanalyses,usingastructuralmodel(usuallytermeddynamicmathematicalmodel)ofthefull
structuralsystem(e.g.launcherwithspacecraft)areperformedtoderiverequirementinformationfor
Limit Loads (LL) and Limit Pressures (LP).These system analyses result in the design requirements
the designer has to tackle in the derivation of the Dimensioning Load Cases (DimLC) which are the
basisforthedesignverificationinthefollowingstressandstrainanalysis.Thissectionwilldealwith
staticloading,only.Itaimsatpresentingtherationalebehindit.
8.3.2 Load analysis, limit load, and design limit load

Structuresexperienceavarietyofloadingconditions(socalledactions),dependingontheparticular
role and function, including dead weight conditions on Earth, manufacturing, testing, assembling,
transportation,operation.Toresistallimposedloadingconditionswithoutunacceptabledistortionor
structural failure the strength and stiffness of materials, and the geometry of the structural part,
should be adequate for the application. This condition should be retained under the influence of all
environmentalconditions(suchastemperature,humidity,vacuum,radiation,atomicoxygen,debris,
lightning,impact,etc.)whilstbeingoptimizedtobemassandcostefficient.
Themaintaskinanyloadanalysisistoestablishallloadeventsthestructureislikelytoexperience(=
load history). This includes the estimation of all induced thermal, mechanical (static, cyclic, and
dynamic) and acoustical environment of the structure as well as of the corresponding lifetime
requirements (duration, number of cycles), as specified by the customer, or by an authority or a
commonstandard.Thelifeofaspacestructuree.g.includesallgroundphaseoperations.
In the establishment of a LL there is some uncertainty. In case of a launcher, this uncertainty is
typicallycounteractedbythe(uncertainty)factorKMPwhichconsidersboththemodeluncertaintyKM
andtheprogrammaticuncertaintyintheprojecttargetsKP,seeReference[2].TheKPfactorreflects
the socalled project margin policy which is normally considered in the value of K P but sometimes
also in an additionally prescribed margin of safety MoS. The Kfactors are of the magnitude of the
FactorsofSafetyFoS,andcandecreasewiththematurityoftheproject.So,thefollowingdefinitions
shouldbekeptinmind(seesection8.4):
Limit Loads (LL): Loads which consider all expected load events and induced thermal,
mechanicalandacousticalenvironmentofthestructure,andwhicharestatisticallydetermined.
Model Factor (KM): Factor, by which the Limit Loads are multiplied to take into account
inaccuraciesanduncertaintiesofthesystemmodel
ProjectFactor(KP):Factorwhichtakesintoaccountthematurityofthedesignanditspossible
evolution and programmatic margins which cover project uncertainties or some growth potential
whenrequired
Design Limit Load (DLL): =LL.KM.KP (example launcher). Limit load multiplied by design
factors.
8.4 Safety concept - Factors of safety (FoS)

In the design process the uncertainty of individual parameters is usually considered by using fixed
deterministic FoS, which act as load increasing, multiplying factors and should be called more
correctlystructuraldesignFoS.Accordingtotheuncertaintiesmentionedabove,suitableFoSjare
132
ECSSEHB3224A
24March2010
taken to ensure reliability of the structural part. These FoS are chosen on the basis of long term
experience with structural testing. Depending on the risk consequences respecting severity and
frequencyofoccurrenceofafailuredifferentclassesofFoSareapplied,e.g.formannedspacecrafts
higherFoSareusedthanforunmannedspacecrafts.
A FoS is a factor by which DLLs are multiplied in order to account: for many uncertainties and
inaccuraciesinherentinstructuralanalysis,numericalmodelling,andcomputation;foruncertainties
in manufacturing process, material properties and failure conditions; for purposes of analytical
assessment, test qualification and for the limited accuracy of the design with respect to strength,
stiffness,andstability.Itshouldbekeptinmind:
a. Inthedesignprocesstheuncertainty(notthetestscatter,howeverthetestscatteruncertainty!)
ofthedesignparametersisconsideredbyusingafixeddeterministicFoS,whichactsasaload
increasingfactorontheDLL.
b. IngeneraltheFoSisnotaccountingforloaduncertaintiesunlessinthedetaileddesignahigher
reliabilityisprescribedthanhasbeenappliedwhenderivingtheLLs.
c. The FoS cannot cover basic error sources. It is a minimum value specified on the assumption
thatmanufacturingprocess,analysistools,modelling,etc.arequalified.
d. The buckling FoS does not cover the effect of dimensional tolerances, load misalignment,
incorrectlyspecifiedboundaryconditionsandimperfections.
e. Aerospacesometimesstillworkswithoneoverall(global,common)FoS.Theuncertaintieson
the action side and on the resistance side are then accounted for by just one FoS. However,
ESTEC practically runs the simplest version of the socalled Partial Safety Factor Concept (2
parameters)discriminatinguncertaintiesofloads(seeKM,KP)fromthoseindesign(FoS).
ThepreviouslymentionedDLLsaremultipliedbyvariousFoSinordertoobtaintheDesignUltimate
Load (DUL), Design Buckling Load (DBL), or the Design Yield Load (DYL). All these loads are
distinct,butdifferentdesignloads:DLLforfatigueanalysis(flightloadlevel),DYLforonsetofyielding
which practically means a deformation limit for the structure, and DUL, DBL for ultimate fracture,
stabilityfailure.Inthiscontext,dimensioningaspectsrequirethatthestructurehastowithstand:
DLL withoutfatiguefailure
DYL withoutdetrimentaldeformation
DUL withoutfracture,collapse
DBL withlocalbuckling,butwithoutglobalbucklingor(plastic)collapse,
andwithregardingspecifiedlimitsofusagefromtheoperationalrequirements.
Table 81 depicts an example for a FoS table in the launcher domain. Incorporated are besides the
usual design FoS for load and pressure, the buckling FoS, the test factors, and further the socalled
special(design)factorsforjoints,welds,fittingswhichareappliedinordertosimplifythedesignin
case of structural discontinuities. Such factors account for specific uncertainties linked to analysis
complexity. In the actual case the Technical Specification will refer to References [2] and [3] and
usuallywillprovideprojectassignedfactors,too.Arelativelylowvalueofjult=1.25istakenifastatic
qualification test is foreseen. In the case of untested hardware or design by analysis only a higher
valueisrequired,e.g.avalueof1.5.Incaseofbuckling,usuallyhighervaluesarespecifiedthangiven
for ultimate. This value is judged for each structure. Its choice depends on the loadings and their
combinationsandsequences.
133
ECSSEHB3224A
24March2010
Table81:SchemeforbucklingFoS,specialfactors,andtestfactorsforpressure
(valuesarefixedineachproject)
FoSY FoSU
for for Qualif. proof
Structuretype/ FoSY FoSU verification verification Special Proof jA
sizingcase jp0.2 jult byanalysis byanalysis factors jQ proof
only only
Metallicstructures 1.1 1.25 1.25 1.5 2 1.5
FRPstructures
1.25 1.5 2 1.?
(uniformmaterial)
FRPstructures
1.25 1.5 1.2
(discontinuities)
Sandwichstruct.
Facewrinkling 1.25 1.5

1.25 1.5
Intracellbuckl.
Honeycombshear 1.25 1.5
Glass/ceramic
2.5 5.0
structures
Buckling 1.5 2.0
NOTE1:FoSvaluesjarebasedonlongengineeringexperience.ThereliabilityassessmentoftheFoS,appliede.g.in
aerospaceoutlinedthattheyarelocatedinthefailureprobabilityrangeof107through109.Differentindustry,
however,hasdifferentriskacceptanceattitudes.
NOTE2:FoSisnotapplicabletoloadsofthermaloriginforwhichj=1isused,duringlaunchphase.Duringonorbit
phases,thermalloadsarecombinedwithmechanicalandpressureloadswhentheyareadditivebutarenot
combinedwhentheyarerelieving.Thisisconsideredinstabilityanalysis,wherestabilizingloadingsareapplied
asminimumvalues.
8.5 Dimensioning load cases selection
8.5.1 Margin policy and design loads

Especially in stability design the user responsibility in each project forces to consider the margin
policy when selecting the load cases. In this context, deciding for a stability design driving
Dimensioning Load Case (DimLC) requires the consideration of the associated failure modes.
Thereforetheapplicationofafailuremodeeffectanalysis(FMEA)isrecommended.Duetothisfact
this task can only reasonably be established by taking into account the KMP factors (see Figure 82,
fromRef.[2]).Thefullylinkednonlinearproceduremakessensitivitystudieswithestimationofthe
most influential design parameters (sensitive drivers) necessary and thereby helps to judge the risk
because a preliminary joint failure view is mandatory and possible. Thereby, the designer can
effectively support the DLL generation process. In the sensitivity studies the design parameters are
assumedtobeindependentfromeachother.
134
ECSSEHB3224A
24March2010
Ifnotestresultsareavailable,thestructuralmodelcannotbevalidated(thedesigncannotbeverified).
Thussomeuncertaintiesareinherentinthemodel.Forthatreasonthestillmentionedmarginpolicyis
introducedtocoverthesemodeluncertainties(includingkinematicalandstaticboundaryconditions).
The margin depends strongly on the type of structure. If the structure exhibits a lot of sensitive
parameters,themarginisincreasedonbasisofthesensitivityanalysisaboveviaKPor,inalater
design phase, as an additional MoS; in some cases these factors are specified for a single structural
partonly.
Itshouldbekeptinmind:MarginpolicyshouldnotbemixedupwiththeKnockDownFactorpolicy.
Itisbasedonexperience,sensitivitystudies,andtrendanalyses.
Figure82:Derivationofdesignloads,DimLCandaccesstodesign(Example:
launcher)
Selecting load cases requires the knowledge of the loading sequence. For instance thermal loading
withorwithoutyielding,staticloadingwithorwithoutyielding,cyclicloading.
With the knowledge of the structural behaviour (linear prebuckling behaviour, nonlinear pre
bucklingbehaviour,switchingdeformationpattern,imperfectionsensitivity)andwiththeknowledge
of its driving stability failure indicating points (bifurcation point, limit point) the load selection
process has to be determined. The more the structure of concern behaves nonlinearly the more care
shouldbetaken.
8.5.2 Design loads combinations relevant for buckling

From all potentially designrelevant load combinations those have to be extracted which might be
finally really design decisive. The selection has to consider the stability failure mode and the
sensitivity(influence)ofadrivingparametertoit.
Socalledloadcombinationfactorsplayanimportantroleintheassessmentofthestructuralintegrity.
As it cannot be a priori decided which combination is mainly responsible for an instability
phenomenon,differentstudieswithdifferentloadcombinationfactorshavetobeperformed.Forthat
reasonthe i sensitivecombinationshavetobesoughtafteraccordingto

Lvector = DLLi FOSi i 81
i

withtheloadcombinationfactors i .InformationonthesefactorscanbefoundinRef.[4].
135
ECSSEHB3224A
24March2010
8.5.3 Dimensioning load cases
In practice, often quick decisions have to be made during the design development phases and for
theseonlyasubsetoftheformerdesigndrivingloadcasescanberespected.Thesecasesarereferred
toastheDimensioningLoadCases(DimLC).
ADimLCspecifiesaloadcasewhichisdecisiveforthedesignandisbasedonasingledesignloadoradesign
loadcombination.
In the case of stability analyses the decision for a distinct DimLC is typically based on knowledge
aboutthelinearornonlinearprebucklingbehaviour,apossibleswitchingdeformationpattern,and
imperfection sensitivity of the influential design parameters. Further, a preknowledge should be
availableaboutthebucklingfailuremodesandthelocationofbifurcationpointandlimitpoint.
NOTE DimLCsarerequestedinordertosupportquickengineeringdecisionsin
case of input changes and to avoid an analysis overkill as well as an
overkillofanalysisdataevaluation.ThinkingintensivelyaboutDimLCs
improves the understanding of the structural behaviour, strengthens
engineeringjudgmentandthereforepaysoff.
8.6 Remaining input data
8.6.1 Overview
Onebasicinputisthegeometryoftheloadcarryingstructureanditslocalthicknesses.Thisincludes
thetolerancebandfordeterminingtheminimumthicknesses.
8.6.2 Influences from manufacturing and manufacturing signatures

Thissubchapterisstronglylinkedtochapter9butitisnoteworthymentioningthispointheretooin
order to underline its importance. Manufacture naturally produces physical and geometrical
imperfections which dominate the stability design as very important design parameters of the
dimensioning process. Such manufacturing signatures from the chosen manufacturing process,
including ovalizations of a crosssection, delaminations and flaws should be controlled with great
care.
ThesensitivityofadesignparameterwasabasicissueintheDimLCselection(seesection0.1).Which
designparameter,suchasanelasticitymodulusoranimperfection,mightbesensitivetothebuckling
failuremodeinminddependsonthemanufacturingprocessandonthetypeofthestructuralelement
(smoothshell,stiffenedshelletc.)andtheappliedmaterial.
For light weight structures the geometrical imperfections are the most important parameters when
investigating instability phenomena. Depending on the type of the structure the geometrical
imperfectionsdifferentlyinfluencetheloadcarryingcapacity.Tobeconsideredareforinstance:
Wallthicknessandovalizations,
wavinessandeccentricloadintroduction,
misalignments.
Thephysicalimperfectionsarerelatedtothematerialpropertiesandtocertainstatevariables.These
arecausedbythemanufacturingprocesseslike:
Forgingorwelding,
136
ECSSEHB3224A
24March2010
machiningorforming,
manufacturingofthelayupandmolding.
Alltheseprocessesinfluencethevalueandthedistributionof:
Thematerialproperties(e.g.strength,elasticitymoduli...)
Theinducednonzerostatevariables(e.g.residualstresses).

Boundaryconditionswhichcomprisesupportconditionsor/andloadingconditionsarechoseninsuch
awayastoensurethattheyachievearealisticoratleastaconservativemodeloftheconstruction.
The boundary condition is a very sensitive parameter determining the analytical or numerical
instabilitypoint.
8.6.4 Application of properties

Areliabledesigndemandsreliableinputvaluesfortheproperties,andforthegeometricaldata.Thus
the questionis: which value should one use: a minimum or a maximum value or an average value.
Theseaveragevaluesareapproximationsofthemeanvalueandarealsotermedtypicalvalues.Very
seldomonegetssuchalargenumberoftestdatatoobtainarealstatisticalmean=50%expectance.
Regarding the properties used during analysis, it is important to know the statistical basis of them.
Propertieswhichareimportantparticularlyfornonlinearmaterialbehaviourareexplainedinchapter
7.
8.6.5 Specific definitions and notions

Before establishing further contents some definitions and notions should be recalled due to the fact
thattheymightbedifferentlyused:
imperfection: deviations from geometry (perfect shape), thickness, perfect material, load
introductionanddistribution,mechanicalperformanceofinsertsetc.(Approacheswithrespect
togeometricimperfectionsfornumericalsimulationofshellstructures:Stimulatinggeometric
imperfections like welded seams are local perturbations which stimulate the characteristic
physical shell buckling behaviour. Worst geometric imperfections have a mathematically
determined worst possible imperfection pattern like the single buckle, [5]. Realistic geometric
imperfectionsaredeterminedbymeasurementafterfabricationandinstallation[6]
plastic collapse: structural failure linked to a limit point where the structure reaches a global
maximumbyhighlyplasticstraining.
stabilitydesignvalue:valueofthebucklingresistancetakingintoaccounttheuncertaintiesof
themodelbasedpredictions(isanapproximationofastabilitydesignallowablewhichisused
in design. It is never a statistical value when analytically or numerically derived. Associated
KDFsmaybeobtainedfromstatisticalevaluationofasufficientnumberoftestresults).
137
ECSSEHB3224A
24March2010
8.7 Stability design allowable and knock-down-factor (KDF)
8.7.1 Stability design allowable (design allowable of the buckling resistance)

A sufficient number of relevant tests are the only basis to determine a (statistically MILHDBK5
based) stability design allowable. The computation of a buckling load, transferred into an equivalent
bucklingresistance,canneverdelivermorethananapproximationofastabilitydesignallowable.So,
asastabilitydesignallowableobtainedfromtestseriesisusuallynotavailableoneonlycanachievea
valuewhichcanbecalledstabilitydesignvalue.
8.7.2 Knock-Down-Factor
Aknockdownfactorisareductionfactorusedtoderiveadesignbucklingresistancefromapredicted
theoretical buckling load value. It will cover the effects which are not considered in the applied
structuralmodel.TheKDFhasnothingtodowiththemarginpolicy.
The application of the KDF primarily takes place in the classical eigenvalue extraction (linear or
nonlinear),wherethecriticalbucklingloadvalueisdecreasedbytheKDF.
TheKDFmethodusuallyisbasedontheanalysisoftheideal,perfectstructure.Itisthemostsimple
andhistoricallythefirstapproachtocorrectstabilityanalysisresultsachievedforgeneralinstabilityor
globalbuckling.AKDFaccountsforthedifferencebetweenpredictedvalueandtestresult.
Sufficient test data are rarely available. In such a case KDF data are referenced to a known
imperfectionlevel,acertaingeometryandaspecifictypeofload.
KDFs consider the influence of geometrical imperfections, of residual stresses and simplifications in
theanalysismethod(boundaryconditions,prebucklingdeformations)andshouldatleastcorrectthe
differencebetweencalculationmodelandtestwiththeaverageorwith50%expectance.
The KDF is differently defined. In case of NASA SP8007 the slenderness ratio dependent KDF is
based on a lowerbound design curve of the cylindrical shell test data. Here one might say that the
KDFisappliedtoacharacteristicbucklingresistance.ProbabilisticaspectsofKDFsarepresentedin[6]
and[7].
8.8 Analyses, modelling, and design aspects
8.8.1 General
Instabilitydesignitisdiscriminatedbetweenthedifferentlyjeopardizedstructuralelements.
Figure 83 displays the loaddeflection behaviour of a shell, a slightly curved plate and a column.
Thesestructuralelementshaveadifferentpostbucklingbehaviour:nonstableandstable.Thismakes
itpracticallyimpossibletogenerallydefineastatisticallybasedstabilitydesignallowableanalogousto
thestrengthdesignallowables.
138
ECSSEHB3224A
24March2010
Figure83:Equilibriumpathofstructuralelementswithdifferentpostbuckling
behaviour
Theresistanceofabucklingendangeredcomponentisdeterminedbyacarefulanalysisconsidering
thescatterofalldrivingdesignparameters.
Itisdistinguished,whetherthestructuralcomponentisgloballyaffectedbythefailuremechanismor
locally like the flange of a girder beam. Local buckling is normally permitted, if the design
requirementsarefulfilled.Thatincludesthefunctionalityandtheserviceabilityofthestructure.Inthe
local analysis, the residual stresses are considered. In this context, there are some limiting bounds
whichmayterminatethecombinedfullynonlineargeometricmaterialanalysis:
Onsetofdetrimentaldeformation:Onsetofglobalyielding,ofessentialmicrodamage
Onsetoffracture:materialstrengthfailurewhichisclearlyidentifiedforallmaterialsincaseof
buckling;oracollapse(mechanism,lossofstaticequilibrium)
Onsetofessentialdegradation:interfibrefailureofcomposites<finalfailure.
Althoughitispossibletoanalysethepostbucklingbehaviour,theopenquestionisalways:Atwhich
pointshouldtheengineerfixthecomputedbucklingloadifnosevereglobalbucklingwasdetected
before?
8.8.2 Optimization and robust design

The design has to strike a balance between lifecycle cost, schedule, and acceptable, verifiable
performance,see[8].Thepurposeofanalysisistounderstandthestructuralproblemandgaininsight
not primarily generate numbers. In this context a compromise is aimed at. Thereby it helps to
developdesignsthatarerobust,ortoleranttouncertaintiesinenvironmentsandintheassumptions.
OptimizationhastoaimatRobustDesign.Thisresultsinastructuralbehaviourwhichisinsensitiveagainst
thevariationofthemostimportantparameters.
Sensitivity investigations, multidisciplinary optimization and judgement of robustness are
prerequisitesforadesignwiththeobjectiveofHighFidelitystabilityanalysis.
8.8.3 Idealisation of geometry and modelling

A difficult problem in structural analysis [8] is that of idealizing a structure, or representing it in a
formweknowhowtoanalyzeoptimally.Onehastomakeassumptions,whetherastabilitymodele.g.
139
ECSSEHB3224A
24March2010
isashellbucklingproblemforwhichonecanfindappropriatesolutionsinhandbooksorforwhich
onehastoperformaFEA.Theassumptionsdefinethequalityoftheoutput.
Basicobjectiveofthedesigneristheprovisionofamodelwhichallowsgivinganswerstotheposed
projecttask.Thismeansaqualifiedresponsetothetaskforobtainingthenecessaryinformationand
required accuracy on stresses, strains, buckling load predictions, mode shapes or displacements.
Simplifyingisrequiredasfarthattheprogrammedeadlineswithrespecttotimeandmoneyaremet.
Someaspectsarerespectedthereby:
Physicshavetobemodelledaccurately.Simpledesignsarelesscostlyandmorereliable.Badmodellingcannotbe
covered by the design FoS which have to guarantee quality of the design and of the structure test in order to
achieveacertainlevelofstructuralreliabilityforthe(flight)hardware!
8.8.4 Analysis solution types

Good designs are obtained at the quickest by proceeding step by step (see chapter 11) from simple
analytical solutions, called here Level1 solutions, to more and more complex models and solution
procedures up to a Level4. Only such a hierarchical approach leads to the needed high fidelity
desiredinstabilityanalysis.Thereby,thenextstepistakenaftersatisfactionoftheprecedingresult.
Theenvisagedanalysislevelsare:
Level1:Analyticalsolutionofidealstructure(perfectgeometry)
Level2:Semianalyticalsolutionofidealstructure
Level3a:FEsolutionofidealstructure
Level3b:FEAofrealstructure(imperfectgeometry,KDF1.00).
The value of the KDF approaches 1.00 the more accurate the applied stability analysis is when
followingthestrategyforahierarchicalhighfidelityanalysis.
The simplest method to determine computed bifurcation points and limit points is the eigenvalue
extractionmethod(classicalnumericalbucklinganalysis).Thismethodfacesseveralessentialtopics:
Carefulmodellingofthestructureincludingboundaryconditions
Considerationofliterature,experiencewithrespecttomeshing(elements,fineness)
Meshconvergencestudywithdifferentrefinements
Parameterstudiesofboundaryconditions
Plausibility check of the stability failure modes in conjunction with the stress interpretation
(membraneforces,shellbendingmoments,)
Similaritycheckswithliterature,experience,handbooks,
Checkofnegativeeigenvalues
ConsiderationofimperfectionsbyapplyingKDFsonthecomputedcriticalbucklingload.
8.8.5 Imperfection sensitivity analysis (nonlinear)

This method is linked to geometrical nonlinear and if applicable to material nonlinear analysis. As
shown by Koiter, for light weight structures the instability failure is sensitive to imperfections as
shown inFig. 48. Toinvestigate thisimportant effect, the two mainimperfectionsgeometricaland
physicalareinvestigatedinanonlinearanalysis.
140
ECSSEHB3224A
24March2010
NoticethatLevel1andLevel2solutionsareanalyticalsolutions,whilethefollowingessentialtopics
areconcernedwithFEsolutionaspects.
Geometricalimperfections:
In case hardware measurements are available the results including a statistical
backgroundshouldbeimplementedintothemodel.
If no hardware measurements are available, eigenmodes can be used instead in an
imperfection sensitivity study. This method is linked to [9] and is called pitchfork
buckling.
Physicalimperfections:
Residual(inselfequilibrium)stressesfrommanufacturingprocess(seesection8.6.2)
Effectofmanufacturingontheorthotropicmaterialdataandbehaviour.
Everyseriousnonlinearanalysislookingforlimitpointsand/orbifurcationpointsisaccompaniedbyan
appropriate eigenvalue extraction at certain loading states. Besides the essential topics from above
someadditionalonesshouldbeaccountedfor:
Carefulmodellingofthestructureincludingboundaryconditions
Considerationofliterature,experiencewithrespecttomeshing(elements,fineness)
Meshconvergencestudywithdifferentrefinements
Parameterstudiesofboundaryconditions
Decisionofmethod(Newton,arclength)
Plausibility check of stability failure modes in conjunction with stressing and strength failure
modesinterpretation(membraneforces,shellbendingmoments,)
Similaritycheckswithliterature,experience,handbooks,
Findingzeroeigenvalues(bifurcationpoints)
Findingstableequilibriumpath
Findingoftherelevantparametermakingthestiffnessmatrixsingular(structure,material,)
8.8.6 Choice of calculation method

The choice of the method is always an important point when beginning to design a structure.
Dependingontheloading,thetypeofstructure,thematerialused,theengineerhastodecidewhich
analysismethodtouse.
Forthatreasontheengineeralwayskeepsinmindtoweightheadvantagesanddrawbacksgivenby
tests,analyticalsolutions,andnumericalmethods.
Thecalculationmethodsaredescribedinchapters10and11.
141
ECSSEHB3224A
24March2010
8.9 Procedure of design verification
8.9.1 Overview
Asthedimensioningprocessofastructureisthemostimportantphasefortheengineersomehintsfor
decisionsaretriedtobegiveninthefollowingflowchartdetail(seeFigure84).Itmaybeappliedtoa
structuralelementaswellastoalargestructure.Itisguidedbythedevelopmentlogicpresentedin
Figure81whichmightbealsotermedgeneraldescriptionofadesignfindingprocess.
Figure84:Guidelineforstabilitydesignverification
142
ECSSEHB3224A
24March2010
8.9.2 Verification policy
Verification demonstrates through a dedicated process that the product meets the specific
requirements and is capable of sustaining its operational role during the project life cycle, reference
[10].
A satisfactory completion of the verification process is the basis for a contractual acceptance of the
productbythecustomer,thecertification.
Theobjectivesoftheverificationprocessareto:
demonstratethequalificationofdesignandperformance,bymeetingallspecifiedrequirements
(includingtimelines)atallthespecifiedlevels;
ensure that the product is in agreement with the qualified design, is free from workmanship
flawsandacceptableforuse;
verifythattheaerospacesystemincludingtools,proceduresandresourceswillbeabletofulfil
missionrequirements;
confirmproductintegrityandperformanceafterparticularstepsoftheprojectlifecyclesuchas
foranorbiter,theprelaunch,inorbit,andpostlandingloadcases.
Verification is accomplished by one or more of the following verification methods: Test, Analysis
including Similarity, Reviewofdesign, Inspection and in addition, Former Experience. A project
related optimal model philosophy such as the Prototype Philosophy with models such as QM, EM,
STM or the Protoflight Philosophy (reference [10]) is applied to achieve a high confidence in the
productverification.
Inthecaseofstabilitycriticaldesign,verificationisperformedbythemethods:
analysiswithcrosscheckingand/or
qualificationtestwithe.g.aQM
sensitivitystudies
flight.
For big structures and if no test model is foreseen the case design verification by analysisonly is
encountered. Then, the real design verification takes place later with a successful flight. For that
reason,alltheinvestigationsdoneas:
LoadcaseselectionandAnalyses
SensitivitystudiesandTrendanalyses
aretakenintoaccount.
Additionalcrosschecksareperformedby:
asimilartest
experience
crosscheckingofastructuralmodel.
143
ECSSEHB3224A
24March2010
8.9.3 Assessment of the structure
Instabilityanalysisalotofdifferentaspectsareconsideredfortheassessmentofthestructuralload
carryingcapacityandareidentifiedandinterpreted,like:
Eigenvalues
Limitpointsandbifurcationpoints
Loadpaths
Convergencebehaviour
Sensitivitystudies
Conditionofthestiffnessmatrixand
Materialbehaviour.
Finally a scalar value can be determined from the analysis results when usually applying the
deterministicsafetyformat.ThisvalueiscalledMarginofSafety(MoS).ItiscomputedforDBLfrom
theequations:
a. Linearcase(loadispermittedtobereplacedbystress)
stability design allowable (stress level) 82

MoS = 1
stress at DBL
b. Nonlinearcase
stability design allowable

MoS = 1 83
DBL
Becauseofthefactthatastabilitydesignallowableisrarelyavailableitistobereplacedonthesafe
sidebyastabilitydesignvalue.
NOTE ThequalityoftheMoSdependsstronglyonthepointslistedabove.Itis
much more important than the figure of the MoS itself. The chance of
failure does not dramatically increase when the figure turns slightly
negative,seereference[11].
8.10 References
[1] Anonymous,ESAPSS03212AideMemoireonStructuralMaterialsandSpaceEngineering,
Issue1,1995.
[2] Anonymous,ECSSEST3210Structuralfactorsofsafetyforspacecrafthardware,2009
[3] Anonymous,ECSSEST32Structuralgeneralrequirements,2008.
[4] Anonymous,EurocodeEN1990,GrundlagenderTragwerkspannung,2002.
[5] Hhne,C.,Rolfes,R.andTessmer,J.:Anewapproachforrobustdesignofcompositecylindrical
shellsunderaxialcompression,ProceedingsEuropeanConferenceonSpacecraftStructures,
MaterialsandMechanicalTesting,913May2005,Noordwijk,ESASP581(2005).
[6] Degenhardt,R.,Bethge,A.andKrger,L.:ProbabilisticAspectsofBucklingKnockDown
FactorsTestandAnalysis.ESADLR19709/06/NL/IAFinalReport.2007
[7] Arbocz,J.andHilburger,M.W.,TowardaProbabilisticDesignCriterionforBucklingCritical
CompositeShells,AIAAJournal,Vol.43,No.8,August2008,pp.18231827.
144
ECSSEHB3224A
24March2010
[8] Sarafin,T.P.andLarson,W.L.:SpaceCraftStructuresandMechanismsfromconceptto
launch.SpaceTechnologyseries.KluwerAcademic,published1997,2ndprint.
[9] Arbocz,J.:PostBucklingBehaviourofStructures:NumericalTechniquesformoreComplicated
Structures.InLectureNotesinPhysics,Ed.H.Arakietal.,SpringerVerlag,Berlin,pp.84142,
1987.
[10] Anonymous,ECSSEST1002,Verification.
[11] Cuntze,R.G.:Isacostlyredesignreallyjustifiedifslightlynegativemarginsareencountered.
JournalKonstruktion,Konstruktion2005March,pp.7782andApril,pp.9398.
8.11 Abbreviated Terms and Symbols

ThefollowingabbreviatedtermsandsymbolsaredefinedandusedwithinthisChapter:
DBL DesignBucklingLoad
DimLC DimensioningLoadCases
DLL DesignLimitLoad(=LLKMKP)
DUL DesignUltimateLoad
DYL DesignYieldLoad
EM EngineeringModel
FE FiniteElement
FEA FiniteElementAnalysis
FMEA FailureModeEffectAnalysis
FoS FactorOfSafety
KDF Knockdownfactor
KM ProgrammaticUncertaintyFactor
KMP UncertaintyFactor
KP ProjectUncertaintyFactor
LL LimitLoad
LP LimitPressure
MoS MarginofSafety
QM QualificationModel
STM StructuralThermalModel
i LoadCombinationFactor
145
ECSSEHB3224A
24March2010
9
Influence of manufacturing and assembly
processes on the buckling load
9.1 Overview
Thebucklingofshellstructuresisinfluencedbygeometricalimperfectionsandresidualstresses,both
dependingontheselectedmanufacturingandassemblyprocesses.
Generally orthotropic stiffened axisymmetric structures are used in space, e.g. cylindrical shells,
conicalshells,sphericalshellsforpropellanttankdomes(orellipticalorotherdomeshapes).
The orthotropic stiffening is possible in differential design or integral design. Differential design
means that stringer and ring stiffeners are riveted or/and bonded or welded (in case of metallic
material)totheskin.Anintegraldesignmeans
a. integralmachinedpartsoutofthickplates,e.g.wafflegridorisogridtypeofshellstructures
b. Sandwichstructures
c. Integralstiffeninge.g.corrugatedshell
d. Compositestructures
Anyhow,theshellsegmentsindifferentialorintegraldesignareassembledandconnectedbybolted,
riveted,bondedorwelded(formetallicmaterial)connections.
9.2 Metallic Structures Integral Machined Curved

Panels
The typical process is to take a plane thick plate and to form it into a basic shape (curved panel or
doublecurveddome).Forcurvedpanelsitisusuallydonebythreerollers,foradoublecurveddome
itcanbedonebyspinforming(pressingarolleragainstarotatingplateandformthematerialovera
tool) or forging. After the forming process a heat treatment can be applied to reduce the residual
stresses.Finallythemachiningprocessisapplied.
Thefollowingitemsshouldbehandledwithextracare:
a. Evenincaseofheattreatmentsignificantresidualstressescanappear.Theycanbringthepart
outofgeometricaltoleranceafterreleasingthemachinedpart(springbackeffect).Evenifthe
geometricaltoleranceisacceptableitisrecommendedtomeasuretheresidualstressesduring
the qualification of the manufacturing and to consider the residual stresses in the design
verificationanalyses.AnexampleisthewafflegridstiffenedinnerdomeoftheLH2propellant
tankoftheARIANE5upperstageESCA.ThematerialisAl2219.Afterthespinformingand
heattreatmentprocessthemeasuredresidualstressesreached50MPa.
146
ECSSEHB3224A
24March2010
b. The local skin thickness and the thickness of the stiffeners due to the machining should be
considered. There is a minimum thickness for machining (thin thickness is risky for the
machiningintermsoffailure;repairsolutionsandproceduresshouldbeincludedinthedesign
verificationprocess).Itisknownthatthethicknesscanvarystronglywithinthetoleranceband.
9.3 Metallic Structures Deformations due to Welding

Shellstructuresusuallyareassembledbyweldingofpanels,e.g.
a. thecircumferenceofcylindricalshellscanbebuiltbyweldingtogethercurvedpanelsresulting
inlongitudinalweldlines
b. sphericaldomestructurescanconsistofweldedsegments(likepartsofanorange)resultingin
meridionalweldlines
c. axisymmetric shell segments (cylindrical parts, conical parts, spherical parts) can be welded
togetherresultingincircumferentialweldlines
Duetotheheatimpactduringweldingthereisashrinkingintheheatedzoneaftercoolingleadingto
local deformation. Additional geometrical mismatch before welding should be taken into account
whichgiveslocaldisturbingstressesunderloading.Botheffectshaveofcourseimpactonthestrength
oftheweldinglinebutcaninfluencebucklingloadstoo(geometricalimperfection).
9.4 Composite Structures

Compositestructureshaveskins(facesheetsofasandwichpanel,skinofastringerorringstiffened
shell, skin of a corrugated shell) which are built up by fibre reinforced plastic layers with different
fibreorientationofthelayers.
After the hardening process in an Autoclave oven under heat and pressure there can be significant
deformationdependingonthelaminatecharacteristiclayerorientation.Theycanbereducedif:
a. the layers layup with the different fibre orientation is symmetrical to the mid plane and
balanced to one axis in the plane (symmetric laminates) and if the fibre volume of a layer is
homogeneous,aswellasthethicknessandthefibreorientationisrealizedinthemanufacturing
asdesigned.
Ifthe designofthelayersisnotsymmetricaspringbackofthepanelswillbepresentafterhardening
whichleadstofinalgeometricdeformation(imperfections)oradditionalconstraintsaftermounting.
9.5 Assembly Stresses and Deformations

If closed shell structures are assembled by bolted interfaces (e.g. stages of launchers) the small
geometrictolerancesoftheinterfacescanleadtohighmountingstresseswhichshouldbetakeninto
consideration for the buckling loads too. A certain level of mounting stresses should be considered
generally in the design verification process. Additionally an interface control process for mounting
stressescanbenecessary.
For the ARIANE 5 the specified interface tolerance for the 5.4 m diameter interface is +/ 1 mm in
longitudinaldirectionandageneralmountingstresslevelof40N/mmfortheaxialforcefluxhasbeen
considered in general design verification. Within this tolerance bound of +/ 1 mm a lot of axial
deformation shapes can be present (depending on the manufacturing process) which lead to non
acceptablestresses(morethan40N/mmaxialforceflux).
147
ECSSEHB3224A
24March2010
Theactualshapecanbemeasuredandtheinducedstressesduetomounting(allgapswillbeclosed
bytorquingthebolts)canbecalculatedinadvance(byforceddisplacementanalysis).Byclosingthe
axial gaps, warping stresses (axial force fluxes) will be produced. If a harmonic analyses of the
circumferentialdistributionofthegapprofilewillbeperformedthewarpingstressesforeachwave
numbercanbecalculatedseparatelyincaseofaxisymmetricshellstructures.Thewarpingstressesof
highwavenumberswilldecayrapidly,butforlowwavenumbers,especiallyforwavenumber2,the
impact can be present in the entire shell. The warping stress has to be considered for strength
verificationandforbucklinganalysis.
Theprocessformountingmorethan2structuresiscomplicatedbecausemountingstresses(warping
stresses)atthefirstmountedinterfaceprocedureproduceovalization(deformationintheplaneofthe
crosssection)atthefreeinterfaces.Iftheseinterfaceswillbemounted,againwarpingstresseswillbe
induced.
Nevertheless,theseeffectscanbeconsideredintheanalysis.Incaseofverificationproblems(negative
margins)thereisstillthepossibilityofshimming(fillingthegaps)oftheinterfaces,akindofrepair
solution.
9.6 The Imperfection Data Bank, a Means to Obtain a

Realistic Buckling Load
9.6.1 Overview
In the following an attempt will be made to show that the results of detailed initial imperfection
measurementsonfullscalestructures canbeusedeffectivelytodevelopimproveddesigncriteriafor
weight critical, low margin of safety, axially compressed cylindrical shells. As is known such
structures are usually buckling critical. A typical design procedure used for the layout of such
structurescanbesummarizedasfollows:
a. Layoutthepreliminarydimensions
b. Selectawallconstructionandastiffeningconcept
c. Use one of the many shell of revolution codes to calculate the buckling load of the perfect
structure taking into account the appropriate boundary conditions and the prebuckling
deformations
d. Selectaknockdownfactortoaccountfortheimperfectionspresentinthefinishedproduct
e. Applytheappropriatesafetyfactor.
Currentlyavailablecomputercodescanusuallymodeltheperfectstructureingreatdetailincluding
discretestiffeners,realisticendattachments,variouswallconstructionsetc.Thusthebucklingloadof
theperfectstructurecanbecalculatedveryaccurately.Thegreatunknownintheaboveprocedureis
themagnitudeoftheknockdownfactorbywhichthebucklingloadpredictionfortheperfectshell
(with all its 9 significant figures) should be multiplied in order to arrive at a reliable buckling load
predictionfortherealstructure.
AllcurrentshelldesignmanualsarebasedonthesocalledLowerBoundDesignPhilosophy.That
is, they recommend the use of an empirical knockdown factor, which is so chosen that when it is
multipliedbytheclassicalbucklingloadalowerboundtoallexistingexperimentaldataisobtained.
For applications where the total weight of the structure is of no concern the Lower Bound Design
Approach is simple to apply, and in most cases it will provide a safe buckling load prediction.
148
ECSSEHB3224A
24March2010
However,inthosecaseswherethedesignisweightcriticaloneisusuallyforcedtoacceptasmaller
marginofsafetyandhenceamorerefinedmethodofdesignandanalysisiscalledfor.
Extensiveresearchoverthepastfewdecadeshasshownthatforaxiallycompressedthincylindrical
shellsthelargestportionoftheknockdownfactorisduetoinitialimperfectionsinshellgeometry,
withthicknessvariations,plasticbehaviour,initialstressesandpoorloaddistributioncontributingthe
remainingpart.(Itisassumedherethattheeffectofrealisticboundaryconditionshasbeenproperly
accountedfor.)
Ithasalsobeenestablishedthatiftheinitialgeometricimperfectionsofashellareknownthenwith
the current generation of nonlinear structural analysis codes it is possible to predict the actual
buckling load within about 10%, a margin that is within the accuracy expected for imperfection
sensitivebucklingloadcalculations.
9.6.2 Imperfection surveys on large or full scale shells

9.6.2.1 Overview
Ingeneral,ithasnowbeenacceptedthatinordertoapplythetheoryofimperfectionsensitivitywith
confidence one should know the type of imperfections that occur in practice. Thus the initial
imperfectionsurveyscarriedoutonlaboratoryscaleshellsbytheCaltechgroupandothershasbeen
followed by imperfection measurements on large scale and full scale structures in different parts of
the world. When carrying out such measurements one should always remember that Koiters work
hasshownconclusivelythattheknockdownfactordependsnotonlyonthemagnitudebutalsoon
the shape of the initial imperfections. Thus it is not sufficient to spot check the shell surface for the
maximum imperfection amplitude by carrying out selected circumferential and/or axial scans. One
should always provide for sufficient crossreference data, so that later the individual scans can be
pieced together to a complete surface map of the measured structure via numerical techniques ona
digitalcomputer.
9.6.2.2 Imperfection surveys at NASA-langley

In1977ArboczandWilliams[1]publishedtheresultsofdetailedimperfectionsurveyscarriedouton
a large (radius 1534.0 mm, length 2387.6 mm, wall thickness 2.54 mm) integrally ring and stringer
stiffened shell. Figure 91 shows the test specimen with the equipment used to carry out the
imperfectionscansvisibleonthesideoftheshell.Thetechniqueemployedmeasuredthedeviationof
the cylinderouter surface relative to an imaginarycylindrical referencesurface. Physically, this was
accomplishedwitha3.0mlongaluminiumguiderailsupportedontheoutsidediameterofthetwo
steelendrings,whichweremachinedsoaccuratelythattheymatchedtowithin0.15mm.Atrolleycar
carryingadirectcurrentdifferentialtransformerinstrumentwasslowlydrivenalongtheguiderailby
anelectricmotor.Thepositionofthecarwaselectronicallymeasuredusingapotentiometer.Accuracy
of the displacement measurements was to within 0.05 mm. The guide rail was moved stepwise
aroundthecylindercircumferencein5degreeincrementsthusyieldingatotalof72discretescansfor
thecompletecylinder.Discretedigitaldatawasrecordedevery6.0mmalongthecylinderlength,thus
yielding343datepointsforatypical2057.4mmscan.Thisdigitizeddatawererecordedonamagnetic
tapeandusedlaterfordatareduction.
149
ECSSEHB3224A
24March2010
Figure91:InitialimperfectionsurveyinstrumentationatNASALangley[1]
In all cases where one attempts to measure the exact shape of a shell, before one can determine the
initialimperfectiononehastodefinetheperfectcylinder.Intheabovecase,theradialdeviationsfrom
the imaginary cylindrical reference surface, defined by the two very accurately machined steel end
ringsandtherigidaluminiumrailconnectingthemweremeasured.Nextthedatareductionprogram
described in Reference [2] was used to find the bestfit cylinder to the measured data of the initial
imperfection scans. Using the method of leastsquares, this program computed the eccentricities Y1
and Z1 , the rigid body rotations 1 and 2 and the mean radius R (see Figure 92). Finally, the
measureddisplacementswererecomputedwithrespecttothenewlyfoundperfectcylinder.These
valueswerethenusedtopreparethe3dimensionalplotoftheinitialimperfectionsshowninFigure
93.Thethreeweldedseamsareclearlyvisibleinthelongitudinaldirection.
Figure92:Definitionoftheperfectcylinder
150
ECSSEHB3224A
24March2010
Figure93:MeasuredinitialshapeofLangleyshell[1]
9.6.2.3 Imperfection surveys at Georgia Institute of Technology

In the early 70s Horton and his cowriters carried out a series of shell buckling tests at Georgia
Institute of Technology. Both small and large scale specimens were tested. The large shells (radius:
945.8mm,length:2743.2mm,wallthickness:0.613mm)weremadeof7075T6aluminiumalloy,each
consisting of 6 identical panels. On the inside they were reinforced by 312 closely spaced Zshaped
stringers (seeFigure94).Oneedgeofeachpanelwasjoggled,andtwostringerswererivetedalong
eachjointline.Theremainingstringerswereattachedtothesheetwithadhesive.Theshellswereheld
circular by two heavy end rings, which were rolled out of [shaped extruded sections, and by 7
equally spaced Zshaped rings on the outside. On two of the large shells complete imperfection
surveysconsistingof32equallyspacedcircumferentialscanswerecarriedout.Forthispurposestiff
endplateswithcentralbearingswereattachedtothespecimen,whichthenwasmountedinaheavy
framework,withitsaxishorizontal,insuchafashionthatitcouldberotatedaboutitsaxis.A3meter
long precision straight edge was positioned parallel to the shell and served as a reference beam. A
LVDT(linearvoltagedifferentialtransducer)wasthenattachedtothestraightedgeatdifferentaxial
positions. The shell was rotated about its axis and the variation of the profile recorded at equal
intervals alongthelengthoftheshell.AblockdiagramoftheoverallsystemisgiveninFigure95.A
3dimensionalviewofthemeasuredinitialimperfectionsreferencedtothepreviouslydescribedbest
fitcylinderisshowninFigure96.
151
ECSSEHB3224A
24March2010
Figure94:ConstructionofHortonsshellHO1[3]
Figure95:BlockdiagramofHortonsimperfectionsurveyinstrumentation
152
ECSSEHB3224A
24March2010
Figure96:MeasuredinitialshapeofHortonsshellHO1[3]
9.6.2.4 Imperfection surveys at sifferent aerospace companies in the USA

In carrying out imperfection surveys on a manufacturing floor one has to be able to adjust to the
realitiesofworkingonatightschedule.Also,inmostcases,onehastousetherotationhardwarethat
ispresentatthemanufacturingsiteinordertomaketherequiredaxialandcircumferentialscansof
the test specimen. This data is then later pieced together by an appropriate data reduction program
using a computer. In order to ascertain that the data recorded is of good quality a partial data
reductionshouldbedoneonsite.
The Caltech group has used the two piece portable rail system shown in Figure 97 to carry out
imperfectionsurveyson flighthardwareatdifferentaerospacecompanies.Onthesamefigureoneof
the test shells is shown mounted in a rotation device. The rail consists of a standard Ibeam that is
connectedwithtaperedpinsatthecenter.Itissupportedbyacenterstandwhichismovableinthe
vertical (z) direction and by two end stands which have xyz positioning capability. The
displacementtransducerismountedonacartwhichslidesalongtherail.Thecartisspringloadedso
that its precision bearing rollers are pressed against the reference axis (the front edge of the top
flange).Therailcanbealignedopticallyusingatheodoliteandanopticaltargetmountedonthecart.
Deviations from straightness are determined using this instrument. The initial error can be
determinedwiththepresentmethodtoabout0.01mmoverthe7.6mlengthofthebeam.Theexact
shapeofthebeamisrecordedandlaterremovedfromthemeasureddataduringthedatareduction
process. The shape of the shell generator is thus measured with respect to this rail by means of an
LVDT of sufficient range to determine not only the expected initial imperfections but also any
misalignmentofthereferencerailwithrespecttothebestfitperfectshellaxissystem.
153
ECSSEHB3224A
24March2010
Figure97:Measurementofafullscaleaerospaceshell[4]
ThepositionofthecartwiththeLVDTalongtherailisdeterminedbyablackandwhitetapeattached
to the rail. The tape is read by a light emitting photodiode pair, which generates a square wave to
triggerthedataacquisitionsystem.ThecartandtherailarealsoshowninFigure98.Thedata(LVDT
andpositionsignal)arecarriedtothedataacquisitionsystembyaflexiblecablewhichfollowsthecart
onaslidewire.Thecartismovedbyasteelcabledrivenbyavariablespeedmotorandlimitedby
switchesmountedontherail.
Figure98:Instrumentcardandpositiontapeonreferencebeam
154
ECSSEHB3224A
24March2010
Figure99showsa3dimensionalplotofthemeasuredinitialimperfectionsreferencedtothebestfit
cylinder from a large (radius 1212.1 mm, lengths 6454.1 mm, wall thickness 1.549 mm) integrally
stiffenedshellconstructedoutofthreepieces.Thelinesaredrawninthecircumferentialdirectionbut
the data were collected from axial scans. The three welded seams are clearly visible in the
imperfectionplot.Thedisturbanceatthecenterresultsfromaringframeatthislocationwhichtends
to minimize the weld imperfection. The waviness is caused by the pockets of the stiffened shell. A
betterviewofthisisshowninFigure910,wheretheregularpatternofthepocketsisclearlyseen.The
largejumpsattheendoftheshellarearesultoftheenddomesoftheshellbeingforcedfitinsidethe
cylindricalsection.Inall3dimensionalplotspositiveimperfectionsareoutward.Anexceptionofthis
rule is Figure 99 and Figure 910 where positive imperfections are pointed inward. This change in
orientationbecamenecessarybecauseoftheformofthemeasuredinitialimperfectionsnearthetwo
edges.AcloserlookattheaxialcrossplotshowninFigure910revealsthattheinitialimperfections
consistof3maincomponents,namelyalargestepfunctionlikecomponentduetotheuniformradial
displacement produced by the forcedfit domes, a large halfwave sine component in the axial
directionplusasmallshortwaveimperfectioncomponentwhichaccountsforthewaviness.Sincefor
stability calculations the uniform radial expansion is of no consequence the edge zones have been
eliminated from the data. Thus the Fourier decomposition of the measured data was done with the
initialimperfectionsshowninFigure911.Herepositiveimperfectionsarepointedoutwardasusual.
The measured initial imperfections referenced to the best fit cylinder from another large (radius
1527.4 mm, length 6047.7 mm, mean wall thickness: 2.629 mm) stringer stiffened shell made out of
four pieces are shown in Figure 912. The four longitudinal welded seams are once again clearly
visible.Also,since this shell has onlylongitudinal stiffeners the short wave waviness caused by the
intersectingtransversestiffenersobservedonthepreviousshellisabsent.Ontheotherhandthisshell
hasmorewavinessinthecircumferentialdirection.
Figure99:MeasuredinitialshapeofaerospaceshellX1
155
ECSSEHB3224A
24March2010
Figure910:TypicalmeridonalshapeofaerospaceshellX1
Figure911:MeasuredinitialshapeofaerospaceshellX1withedgezones
removed
156
ECSSEHB3224A
24March2010
Figure912:MeasuredinitialshapeofaerospaceshellX2
9.6.2.5 Imperfection surveys on aerospace shells in Europe

TheAerospaceStructuresGroupattheDelftUniversityofTechnologyusestheonepieceportablerail
systemshowninFigure913tocarryoutimperfectionsurveysontheARIANEinterstageI/IIandII/III
shells. Here the shell is positioned upright on a 2piece turntable, with the reference beam placed
paralleltoitonanadjustabletripod.
157
ECSSEHB3224A
24March2010
Figure913:InitialimperfectionsurveysystemfromtheTUDelft
There are 3 LVDT pickups installed on the reference beam. The fixed ones on the top and on the
bottomarebearingagainstthemachinedendrings.Thethirdoneisinstalledonacarriagewhichis
movedalongthebeambyanelectricdrivetorecordtheshapeofthecorrespondingshellgenerator.
Next the shell is rotated to a new position followed by another axial scan. The process is continued
untilthewholesurfacehasbeensurveyedandrecorded.Theexactshapeofthereferencebeamhas
been measured optically and it is removed from the measured data during the data reduction step.
The axial position of the carriage is recorded by an electrooptical device which scans a strip with
equallyspacedcutouts.Theresultingsquareshapedsignalisusedtodigitizethedatainintervalsof
10mm.Specialcareistakentorecordpossiblerandomrigidbodydisplacementsoftheshellassembly
with respect to the fixed position of the reference beam during rotating the turntable to a new
circumferentialposition.Thisisaccomplishedbymonitoringtheplanardisplacementsofacalibrated
circularringplacedinthecenteroftheturntable.Thesedisplacementsarethenalsoremovedfromthe
measureddataduringdatareduction(seeReference[5]foradetaileddescription).
Figure 913 shows an ARIANE interstage II/III shell (radius 1300.0 mm, length 2730.0 mm, wall
thickness1.2mm)whichisbuiltupoutof8identicalpanels.Thejointsbetweenadjacentpanelsare
joggled and on the outside there are 120 equally spaced hatshaped stringers riveted to the shell
surface.Thewholestructureisheldcircularbytwoprecisionmachinedendringsontheoutsideand5
equally spaced Ushaped rings on the inside. A 3dimensional view of the measured initial
imperfections referred to the socalled best fit cylinder is shown in Figure 914. It should be
mentioned here that in all 3dimensional views the measured initial radial imperfections have been
normalizedbythecorrespondingwallthickness.
158
ECSSEHB3224A
24March2010
Figure914:MeasuredinitialshapeoftheshellAR231[5]
9.6.3 Comparison of the measured initial imperfections

9.6.3.1 Overview
AtthemomenttherearetwoinitialImperfectionDataBanksinexistence,oneattheDelftUniversity
of Technology and one at the Technion in Haifa. The purpose of establishing these data banks is
twofold.First,alltheimperfectiondataobtainedatdifferentlaboratoriesbydifferentinvestigatorsis
presentedinidenticalformsothatacomparisonandcriticalevaluationismadepossible.Second,with
thecontinuedadvancesbeingmadeinthedevelopmentofpowerfulnonlinearshellanalysiscodesthe
need for realistic imperfection distributions to be used for collapse load predictions is expected to
increase.TheImperfectionDataBanksareintendedtoprovidethisinformation.
In all cases the measured initial imperfections are represented by the following two double Fourier
series:
N N
k x ly ly
W(x, y) = t cos L Akl cos
R
+ Bkl sin
R
91
k =0 l =0
referredtoasthehalfwavecosinerepresentation,and
N N
k x ly ly
W(x, y) = t sin C cos
L kl R
+ Dkl sin
R
92
k =0 l =0
calledthehalfwavesinerepresentation.HereR,Landtareshellradius,lengthandwallthickness;x
and yareaxialand circumferential coordinates; kand l are integers denoting the number of axial
halfwavesandthenumberoffullwavesinthecircumferentialdirection,respectively.Inallcasesthe
measuredimperfectionsarereferredtothesocalledbestfitcylinders.
Theadoptionofastandardrepresentationforthemeasuredinitialimperfectionsisnecessaryinorder
to be able to compare the different imperfection distributions, in an attempt to find characteristic
imperfection distributions that are associated with the different fabrication processes. It should be
159
ECSSEHB3224A
24March2010
stressed here that in all cases the Fourier coefficients are normalized by the corresponding wall
thicknessesoftheshellbodies.
9.6.3.2 Shell fabricated by Welded Seams

Manyaerospacetanksareconstructedoutofafixednumberofcurvedpartsjoinedinitiallybywelded
seamstoformacylinder.Latertheendsareclosedbyappropriateendpieces.Thecurvedpartsare
made out of initially thick walled flat plates, which are rolled to desired shape after the chosen
stiffeningpatternhasbeenintegrallymachinedintothem.Figure915showsthevariationofthehalf
wavesineFouriercoefficientsasafunctionofthecircumferentialwavenumber l (forselectedaxial
halfwavenumbersk)ofthelargeintegrallyringandstringerstiffenedshellmeasuredatLangley(see
Figure93forthecorrespondinginitialimperfectionmap).TheamplitudesoftheFourierharmonics
with a single half wave in the axial direction have a distinct maximum at 9 circumferential waves.
Only those harmonics that are integer multiples of the 3 longitudinal welds have significant
amplitudes. All Fourier coefficients with more than a single half wave in the axial direction are
comparatively much smaller. Also, because the heavy end rings have been machined to very close
tolerances the shell exhibits practically no ovalization. (That is the amplitude of the Fourier
coefficientswith l = 2 isverysmall.)
Figure915:CircumferentialvariationofthehalfwavesineFourierrepresentation
(shellLA1[1]
Figure 916 shows the variation of the halfwave sine Fourier representation for a large integrally
stiffenedaerospaceshell(seeFigure911forthecorrespondinginitialimperfectionmap),whichisalso
assembled out of 3 curved parts. In this case the amplitudes of the Fourier harmonics with a single
halfwaveintheaxialdirectionhaveadistinctmaximumat l = 3 ,whichcorrespondstothenumber
of longitudinal welds. Only harmonics that are integer multiples of 3 have significant amplitudes.
Also,theamplitudesoftheFouriercoefficientsdecaywith increasingwavenumbersbothintheaxial
and in the circumferential direction. Further since the circular end domes are forcedfit into the
cylindricalpart,theshellpossessesanegligiblysmall l = 2 component(ovalization).
160
ECSSEHB3224A
24March2010
(aerospaceshellX1)
FinallyFigure917showsthevariationofthehalfwavesineFourierrepresentationfortheotherlarge
integrally stiffened aerospace shell (see Figure 912 for the corresponding initial imperfection map),
which is assembled out of 4 curved parts. In this case the end domes were butwelded to the
cylindrical part. This resulted in a relatively large ovalization. As a matter of fact the Fourier
harmonics with a single half wave in the axial direction have a distinct maximum at l = 2 (outof
roundnessorovalization).Besides l = 2 and l = 6 onlythoseharmonicsthatareintegermultiplesof
4havesignificantamplitudes.AlsointhiscasetheamplitudesoftheFouriercoefficientsdecaywith
increasingwavenumbersbothintheaxialandinthecircumferentialdirection.
(aerospaceshellX2)
9.6.3.3 Shells fabricated by Riveted Seams

Thefixednumberofcurvedpanelscanalsobeassembledintoacircularcylindricalshellbymeansof
joggledandrivetedjoints.OneexampleofsuchashellistheonetestedbyHorton.Figure918shows
161
ECSSEHB3224A
24March2010
the variation of the halfwave sine Fourier representation for this shell (see Figure 96 for the
correspondinginitialimperfectionmap).TheamplitudesoftheFourierharmonicswithasinglehalf
wave in the axial direction have 2 distinct maxima, one at l = 2 (outofroundness) and another at
l = 6 (which corresponds to the number of panels the shell is assembled from). The Fourier
coefficientswithmorethanasinglehalfwaveintheaxialdirectionarecomparativelymuchsmaller.
(HortonsshellHO1[3])
Figure 919 shown the variation of the halfwave sine Fourier representation for the ARIANE
interstage II/III shell measured by the Aerospace Structures Group of the Delft University of
Technology (see Figure 914 for the corresponding initial imperfection map). The amplitudes of the
Fourier harmonics with a single half wave in the axial direction have a distinct maximum at l = 8
(whichcorrespondstothenumberofpanelstheshellisassembledfrom).Thereisalsoasizable l = 2
(outofroundness)component.AllotherFouriercoefficientsarecomparativelymuchsmaller.
(aerospaceshellAR231[5])
162
ECSSEHB3224A
24March2010
9.6.4 Characteristic imperfection distributions
Imbert [6] has established that the amplitudes of the Fourier coefficients representing the measured
imperfectionsoftheCaltechelectroplatedisotropicshellsdecaywithincreasingwavenumbersboth
intheaxialkandcircumferentialdirection l .Heshowedthatsuchadistributioncanberepresented
satisfactorybytheanalyticalimperfectionmodelproposedbyDonnellandWan[7]
X
kl = Ck2l + Dk2l = 93
k ls
r
where Ckl and Dkl are the halfwave sine Fourier coefficients, and the coefficients X, r and s are
determinedbyleastsquarefittingthemeasureddata.Thustheabovementionedformulasummarizes
thecharacteristicimperfectiondistributionforthespecificfabricationtechniquedescribed.
Itappearsfromtheresultspresentedinsections9.6.2and9.6.3thatforfullscaleaerospaceshellsbuilt
upoutofafixednumberofcurvedpanelstheinitialimperfectiondistributionswillbedominatedby
two components only, if the joints are riveted. Using the halfwave sine representation both
componentswillhaveasinglehalfwaveintheaxialdirectionandrespectively2,and l fullwavesin
the circumferential direction, where l is the number of full length panels out of which the shell is
assembled.Byusingaccuratelymachinedrigidendringsthe l = 2 outofroundnesscomponentcan
besignificantlyreducedinsize.ThusthevariationofthemeasuredharmonicsshowninFigure918
and Figure 919 can be considered as the characteristic initial imperfection distribution for this
particulartypeoffabricationprocess.
Ontheotherhand,ifthelongitudinalseams,neededtoassemblethecurvedparts,areweldedthen
theresultingimperfectiondistributionsaremorecomplicated.Thedominantimperfectionharmonics
continue to have a single half wave in the axial direction and the number of full waves in the
circumferentialdirectionisobviouslyinfluencedbythenumberoffulllengthpanelsoutofwhichthe
shell isassembled. However, the welding procedure used has also strong influence on the resulting
numberoffullwaves.ThusboththeLangleyshellLA1andthefirstaerospaceshellX1(radiusR=
1212.1 mm) consist of 3 full length curved panels, both have negligibly small outofroundness (the
l = 2 Fourier coefficients are very small), however for the Langley shell LA1 the largest Fourier
coefficient has 9 full waves in the circumferential direction (3 times the number of welded seams),
whereasforthefirstaerospaceshellX1thelargestFouriercoefficienthas3fullwaves(equaltothe
numberofweldedseams).Apparentlytheweldingprocedureusedbythetwoaerospacecompanies
were similarbecausealsothesecondaerospaceshellX2radius(R=1527.4mm)hasalargeFourier
coefficient with the same number of full waves in the circumferential direction as the number of
curvedpanelsoutofwhichitisassembled,namely4.However,theinitialimperfectiondistributionof
thisshellhasalsocomparativelylargeharmonicswith l = 2 (outofroundness)and l = 6 fullwaves
inthecircumferentialdirection,besidestheharmonicsthatareintegermultiplesof 4,the numberof
weldedseams.
Thus for shells assembled out of a fixed number of curved panels by welded seams an important
questionremainstobeanswered;namely,whenisthenumberoffullwavesofthedominantFourier
coefficientinthecircumferentialdirectionequaltothenumberofweldedseamsandwhenwillitbe
equaltothenumberofweldedseamstimesaninteger.
9.6.5 Probabilistic stability analysis

As pointed out in the beginning of section 9.6 the great unknown in any design procedure is the
selectionoftheknockdownfactorbywhichthebucklingloadpredictionoftheperfectshellshould
bemultipliedbyinordertoarriveatareliablebucklingloadpredictionfortherealstructure.Ithas
alsobeensuggestedthattheexistenceofInitialImperfectionDataBanksmakesitpossibletoassociate
statisticalmeasureswiththedifferentmethodsoffabrication,whichinturncanbeusedtoderivethe
163
ECSSEHB3224A
24March2010
correspondingReliabilityFunctions R() .Here R() isdefinedastheprobabilitythatthe(random)
bucklingloadofthestructure willexceed(orbeatleastequalto)thespecifiedloadlevel .Thus
R() = Pr ob( ) 94
where is the normalized load parameter (=P/Pcl, where Pcl = 2 Rt cl and

Et
cl = ; c = 3(1 2 ) )andthenormalizedrandombucklingload.
cR
AscanbeseenfromFigure920,theknowledgeofthereliabilityfunctionmakesitpossibletofindthe
allowableloaddefinedasthenormalizedloadlevel forwhichthedesiredhighreliability(say0.95)
is achieved, for the whole ensemble of shells produced by a given manufacturing process. The
correspondingknockdownfactoristhusa.
Figure920:Determinationoftheknockdownfactorabymeansofareliability
functionR()
Several investigators have studied the static and dynamic buckling of imperfection sensitive
structures with small random initial imperfections. For an authoritative review the reader should
consult Amazigos paper from 1976 [8]. However, it was not until 1979 that a method has been
proposedbyElishakoff[9],whichmadeitpossibletointroducetheresultsoftheinitialimperfection
surveysroutinelyintotheprobabilisticstabilityanalysis.
Basically Elishakoff suggested to utilize the Monte Carlo Method for the solution of the stability
problem of axially compressed cylindrical shells with random initial imperfections. When applying
this method the initial imperfections are expanded in terms of the usual double Fourier series and
then the Fourier coefficients are treated as random variables. Next the mean functions and the
variancecovariance matrices of the experimental sample are calculated. Then by using a special
numerical procedure the Fourier coefficients of the desired large sample of random initial
imperfectionshapesaresimulated.Thisisfollowedbyadeterministicbucklingloadanalysisforeach
of the simulated shells. Finally from the resulting histogram of non dimensional buckling loads the
reliability function representing the probability (i.e. fraction of an ensemble) of the buckling load
exceedingthespecifiedloadiscalculated.
TherelativeeasewithwhichonecanderivethereliabilityfunctionviatheMonteCarloMethod,once
a sufficiently large sample of initial imperfection measurements on nominally identical shells is
available,willbeillustratedbyconsideringaxisymmetricimperfections.HavingNrealizationsofthe
measuredimperfections
164
ECSSEHB3224A
24March2010
w(x)m x
W(x)m = = Ai(m) cos i (m = 1, 2,...,N) 95
t i
L
One calculates first, by taking ensemble averages, the estimated mean of the Fourier coefficients
Ai(m)
1 N (m)
Ai(e) = A 96
N m =1 i
andthentheestimatedvariancecovariancematrix
N
1
(e) = A(m) A(e) A(m) A (e) 97
jk N 1 m =1 j j k k
Since (e)
jk
isanonnegativesymmetricmatrix,thereforeitcanbedecomposedasaproductoflower
anduppertriangularmatricesbyCholeskysdecomposition[10].
(e) = C C T
jk
98
Nextthevector {A} ofthesimulatedinitialimperfectionsisobtainedasfollows
{A} = C {r} + {A(e) } 99
where
{A(e) } =estimatedmeanVector
{r} =randomvector
Thersarenormallydistributedrandomnumberswithzeromeanandunitvariancegeneratedbythe
computer.
Taking, for example, 1000 different rs, one gets 1000 different As, that is different simulated shells
with the As as the Fourier coefficients of the initial imperfections. For each of the created initial
imperfections one then carries out a deterministic buckling load calculation generating the buckling
loadhistogramofthegroupofshellsunderconsiderationshowninFigure921.Havingdefinedthe
reliabilityfunction R() astheprobabilitythatthebucklingload willexceedtheprescribedvalue
one then proceeds to calculate R() from the histogram of the buckling loads by the frequency
interpretation(i.e.fractionofanensemble)yieldingthedotsshowninFigure922.Theaccuracyofthe
Monte Carlo Method can be seen in the close coincidence with the solid curve which represents a
closedformsolutionintermsoferrorfunctionsforthesamecasepublishedbyRoordaandHansen
[11].
165
ECSSEHB3224A
24March2010
Figure921:Histogramofthenondimensionalbucklingloads[12]
Figure922:Comparisonofanalyticalreliabilityfunction[11]withresultsof
MonteCarlosimulation[12]
The feasibility of using the Monte Carlo Method to derive reliability functions using the data of
experimentalimperfectionsurveysassembledintheImperfectionDataBankattheDelftUniversityof
TechnologyhasbeendemonstratedbyElishakoffandArboczforbothaxisymmetric[12]andgeneral
asymmetricimperfections[13].
9.6.6 Conclusions
Because it is simple to apply and since for many cases it provides a safe and reliable buckling load
prediction the Lower Bound Design Philosophy will continue to be widely used also in the future,
especially for standard applications where the total weight of the structures is of no dominant
concern.
However,sincestructuraloptimizationinvaryinglyseemstoleadtothinwalledconfigurationswhose
buckling loads are affected by initial imperfections, therefore one can expect that the interest in
developingimproveddesignprocedureswillcontinuetooccupythescientificcommunity.Thesenew
approaches leading to possible improvements in our buckling load prediction capability for
imperfection sensitive structures will require in all cases extensive and the more detailed the better
knowledgeoftheactualimperfectionsthatarepresentintherealstructures.
It is encouraging to see that the need for detailed imperfection surveys on full scale and laboratory
scalestructuresandtheusefulnessofestablishingInitialImperfectionDataBanksisbeingrecognized
bymoreandmoreinvestigators.Forinstance,thisattitudehasbeenadoptedbytherecentPOSICOSS
[14]andCOCOMAT[15]projects,respectively.
166
ECSSEHB3224A
24March2010
TheideaofgeneratingreliabilityfunctionsviatheMonteCarloMethodwhichdisplaythedegrading
effect on the buckling load of the expected initial imperfection distribution characteristic of a given
fabrication process, seems to offer the means of combing the Lower Bound Design Philosophy with
thenotionofGoodnessClasses,thusshellsmanufacturedbyaprocess,whichproducesinherentlya
lessdamaginginitialimperfectiondistributionwillnotbepenalizedbecauseofthelowexperimental
results obtained with shells produced by another process, which generates a more damaging
characteristicimperfectiondistribution.
Lookingintothefuture,itistobeexpectedthattheexistenceofextensivedataoncharacteristicinitial
imperfection distributions classified according to fabrication processes, the availability of improved
versions of the present generation nonlinear structural analysis codes, the probabilistic stability
approachviathereliabilityfunctionsandthegreatlyincreasedcomputationspeedofferedbytheso
calledsupercomputerswillfinallyresultinaseriesofimproveddesignrecommendationswhichwill
incorporatethelatesttheoreticalfindingsandmakethemroutinelyaccessibletothedesigners.
9.7 References
[1] Arbocz,J.;Williams,J.G.;Imperfectionsurveysona10ftdiametershellstructure.AIAA
Journal,vol.15,no.7,pp.949956,1977.
[2] Arbocz,J.:Theeffectofgeneralimperfectionsonthebucklingofcylindricalshells.Ph.D.thesis,
CaliforniaInstituteofTechnology,Pasadena,1968.
[3] Horton,W.H.:Ontheelasticstabilityofshells.NASACR145088,1977.
[4] Babcock,C.D.Jr.;Arbocz,J.:Imperfectionmeasurementsoflargescaleshells.GALCITreport
SM787,CaliforniaInstituteofTechnology,Pasadena,1978.
[5] Sebek,R.W.L.;ImperfectionsurveysanddatareductionofARIANEinterstagesI/IIandII/III.Ir.
Thesis,TWDelft,DepartmentofAerospaceEngineering,1981.
[6] Imbert,J.:Theeffectofimperfectionsonthebucklingofcylindricalshells.Aeronautical
engineerthesis,CaliforniaInstituteofTechnology,1971.
[7] Donnell,L.M.andWan,C.C.,Effectofimperfectionsonbucklingofthincylindersandcolumns
underaxialcompression,J.Appl.Mech.,vol.17,pp.795806,1950.
[8] Amazigo,J.C.:Bucklingunderaxialcompressionoflongcylindricalshellswithrandom
axisymmetricimperfections.Quart.Appl.Math.,vol.26,no.4,pp.537566,1969.
[9] Elishakoff,I.,Bucklingofstochasticallyimperfectfinitecolumnsonanonlinearelastic
foundationareliabilitystudy,JournalofAppliedMechanics,vol.46,No2,pp.411416,1979.
[10] Schwarz,H.R.,NumericalAnalysisAComprehensiveIntroduction,Wiley&Sons,ISBN0
471920657,NewYork,1989.
[11] Roorda,J.;Hansen,J.S.:Randombucklingbehaviorinaxiallyloadedcylindricalshellswith
axisymmetricimperfections.JournalofSpacecraft,vol.9,no.2,pp.8891,1972.
[12] Elishakoff,I.;Arbocz,J.:Reliabilityofaxiallycompressedcylindricalshellswithrandom
axisymmetricimperfections.ReportLR306,DelftUniversityofTechnology,1980.
[13] Elishakoff,I.,vanManen,S.,Vermeulen,P.G.andArbocz,J.,FirstOrderSecondMoment
AnalysisoftheBucklingofShellswithRandomImperfections,AIAAJournal,vol.25,no.8,pp.
11131117,1987.
[14] Zimmermann,R.undRolfes,R.,POSICOSSimprovedpostbucklingsimulationfordesignof
fibrecompositestiffenedfuselagestructures,CompositeStructures,73(2),171174,2006
[15] Degenhardt,R.,Rolfes,R.,Zimmermann,R.undRohwer,K.,COCOMATimprovedmaterial
exploitationofcompositeairframestructuresbyaccuratesimultationofpostbucklingand
collapse,CompositeStructures,73(2),175178,2006
167
ECSSEHB3224A
24March2010
10
Modelling aspects of numerical analysis
10.1 Introduction
The Finite Element Method is the main modelling tool when analyzing a practical structure. In the
present chapter the use of the Finite Element method and the numerical procedures applied to
performabucklingorpostbucklinganalysisaredescribedinamoreorlessqualitativemannerwitha
focus on space applications (spacecraft, launch vehicles, etc.). The theoretical background of
performing buckling and postbuckling analyses is discussed in more detail in Chapter 6 of the
handbook.
First of all it is very important to consider the objectives of the planned Finite Element buckling
analysis in terms of what kind of information is needed to gather (geometry, material properties,
boundary conditions, loads,) and what is expected as output (deliverables like mathematical
models, reports, ), mostly laid down in some kind of task or work package. All activities to be
performed within the frame of a work package are to be arranged to achieve a balance between
planning,costsandtechnicaloutput.
In the present handbook, the use of a hierarchical analysis is considered to be an indispensable
engineeringapproachwhenacomplicatednonlinearbucklinganalysisisperformed.Thishierarchical
approach is described in Chapter 11. Central in the hierarchical approach is the identification of an
idealized structureforwhichanalyticalandsemianalyticalsolution methodsareavailable.Typically,
for spacecraft and launch vehicles, shells of revolution constitute the idealized structure. The
characteristicsofthecorrespondingsemianalyticaltoolsaredescribedinSection10.2.Thesubsequent
sectionsfocusonaspectsrelatedtotheFiniteElementmodelling.
10.2 Semi-analytical models - shooting method and finite

difference method
10.2.1 Overview
Startingfromthegoverningpartialdifferentialequationsforaspecificshelltheory(seeChapter17),
the semianalytical models for shells of revolution correspond to ordinary differential equations for
themeridiandirectionoftheshellofrevolution.Thenumericalapproachesthatareusedtosolvethe
resultingordinarydifferentialequations(withthecorrespondingboundaryconditionsatthetwoshell
edges)canbedividedintotwomaingroups:
a. TheShootingMethod(NumericalIntegration)
b. TheFiniteDifferenceMethod
168
ECSSEHB3224A
24March2010
10.2.2 Description of the numerical methods
The Finite Difference method is based on the reduction of the continuum to a system with a finite
numberofdegreesoffreedom.Thereductionisachievedbysubstitutionofalgebraicexpressionsfor
the unknown functions and their derivatives. These expressions contain function values at discrete
points.ForthederivationofFiniteDifferenceexpressions,aTaylorseriesexpansioncanbeused.The
approachresultsinasetofalgebraicequationsthatcanbesolvedusingthetechniquesavailablefrom
linear algebra. There is a close correspondence between Finite Difference procedures and Finite
Element procedures. Finite Difference equations can be interpreted as stiffness relations. A typical
representativeoftheFiniteDifferenceprogramsforshellanalysisisBOSOR4[1].
The Shooting Method is based on the numerical integration of the governing ordinary differential
equationsbymeansofinitialvaluesolvers.IntheShootingMethod[2],aninhomogeneousboundary
valueproblemisconvertedintoasequenceofinitialvalueproblemswhicharesolvedbynumerical
integration using standard routines. Guesses for the unknown boundary values are iteratively
adjusted until all prescribed boundary conditions are satisfied. In this way, the boundary value
problem has been reduced to the solution of a system of (nonlinear) equations for the unknown
boundaryvalues.Thusingeneral,theshootingprocedureconsistsoftwosteps,whichcanberepeated
inaniterativeprocedureuntilconvergencehasbeenachieved.Forlinearproblems,thesolutionisin
principle obtained in one step. To avoid the numerical problems caused by a rapid growth of the
initial value solutions, one can break the shell into an assemblage of shell segments. In this
modification,MultipleShooting,thegrowthofthesolutionsisthencontrolledbydividingtherange
ofintegrationintoanumberofsmallerintervals.Atypicalrepresentativeofthenumericalintegration
programsisSRA[3].
In the Shooting Method the governing equations are typically converted to sets of firstorder
differential equations and solved through the use of standard integration routines, based on for
instance the RungeKutta method or AdamsMoulton method. In Finite Difference analysis it is
convenient to reduce the equations to firstorder or secondorder equations before constructing the
FiniteDifferenceexpressions.
The programs based on the Shooting Method are closely corresponding to traditional analytical
approaches. Eigenvalue calculation can be done by using determinant plotting [4] or using mode
iteration[3].Dependingontheformulationoftheshootingmethod,acloserconnectionwithmatrix
formulation is possible. The programs based on Finite Differences immediately lead to matrix
formulation,andthestandardtoolsthatareavailableinmatrixformulation(astheyarealsoexploited
inaFiniteElementformulation)canbeapplied.
10.2.3 Capabilities and scope of the programs

ByapplicationoftheShootingMethodortheFiniteDifferenceMethod,veryaccuratesolutionscanbe
obtainedforthegivensetsoftwopointboundaryvalueproblems.Themostadvancedversionsofthe
programs based on numerical integration and finite differences should have the following analysis
capabilities,
a. linearstaticanalysis
b. eigenvalue/bucklinganalysis
c. (fully)nonlinearstaticanalysis
d. asymptotic/bfactoranalysis
e. nonlinearfundamental/prebucklingstate
f. inclusionofplasticity
169
ECSSEHB3224A
24March2010
g. asymmetricloading,livepressureloading
h. vibrationanalysis
andshouldincludethefollowingaspects,
orthotropicmaterial/anisotropicmaterial
generalmeridionalshape
discontinuitiesalongthemeridionaldirection,shellbranching
discreterings
smearedstiffeners
variousboundaryconditionsattheedges(includingageneralelasticconstraint)
10.3 Finite Element Model
10.3.1 Finite Element Model Generation

Thesize(details)ofthefiniteelementormathematicalmodelofashellstructuretoperformastability
investigationinordertoverifyanalyticallythestrengthrequirementsposedtothatshellstructureisof
coursestronglydependentonthedevelopmentphase(A,BorC)andavailablebudgetsallocatedto
the (space) project. Even the computing resources within the company or institute should be
considered. The technicalcontents should bein balance with the financialand planning constraints.
The project phases should follow the hierarchical approach which is discussed in more detail in
chapter 11. Do not start in the early phase of the project with too detailed finite element or
mathematicalmodels.Inthecourseoftheprojectoneshouldincreasethelevelofdetailstocoveras
well as possible the design aspects, geometry, material properties, load application, boundary
conditions,initialimperfections,postbuckling,thermaleffects,etc.
Preprocessingprogrammes(e.g.MSC.PATRAN,FEMAP,ABAQUS/CAE,)areappliedtogenerate
thefiniteelementmodelinaneconomicalway.
One should always remember that the mathematical idealization (finite element model) is always a
compromiseofthereality!
10.3.2 Finite Element Model Element Mesh

The final finite element mesh to be used during the buckling analyses is based on previous
convergence studies to assure that the expected buckling modes (wavelengths), expected stress
concentrations,expectednonlinearmaterialeffects(e.g.plasticityinlocalareas)canbeanalysedinan
adequate manner. The chosen detail of the finite element mesh, consisting of 0D, 1D, 2D and 3D
finite elements should be based upon studies using simple axialsymmetry modelling techniques
applying the finite difference method (in house programmes, BOSOR4, ) and in sequence an axi
symmetricfiniteelementmodelasanupbeattothe(final)finiteelementmodelwhichrepresentsthe
realstructure(withholes).
10.3.3 Selection of Finite Elements

Theselectionofthetypeoffiniteelementdependsuponexperienceoftheanalystandtheapplication
of a finite element software package and of course of the structural elements (bar, plate, shell,
170
ECSSEHB3224A
24March2010
machinedpart,)usedinthedesignofthestructuretobeanalysed.Itisgoodtoreadliteratureabout
thedevelopmentoffiniteelements(e.g.MacNeal[5])andbenchmarkstudiesfrequentlypublishedby
NAFEMS.
Shellstructuresshouldbeidealisedintofiniteelementsbyapplyingshellorplateelements,however,
the trend becomes to apply solid elements, because of easy transfer from CAD parts to the pre
processor of the finite element programme. It is an easy task to fill transferred solids with an
enormous number of solid elements such that computer resources are not sufficient and long wall
clock times can be expected. This is not a very good approach the analysis tradeoffs. Besides the
aspect ratio of solid elements in shell type structures can cause numerical problems during the
analysis.
Ringframestostiffentheshellstructure,canbemodelledbybeamswithorwithoutoffsetstolimit
the number of elements, however, more and more shell and plate elements are used and off set
characteristicsareautomaticallyincorporatedintothefiniteelementmodel.
Stringerstoreinforceshellstructurescanbemodelledliketheringframes.
Ringsareingeneralmodelledusingplateorshellelements,however,moreandmoresolidelements
areused.
Thecouplingbetweensolidandshellandbeamfiniteelementsshouldbedoneverycarefullybecause
nodesinsolidelementsrepresent3degreesoffreedomwhilethenodesinbeamandshellelements
represent 56 degrees of freedom. If the coupling of degrees of freedom is not treated properly
mechanisms in the finite element may jeopardise the numerical procedure of the stiffness matrix
inversion.
Itisdifficulttoprovidedetailedguidelinesforthefiniteelementanalysttosetupafiniteelementor
mathematicalmodeltoperformbucklinganalysis.Besidesthisbucklinghandbookcompanyandon
jobexperiencescanbeused.
10.3.4 Finite Element Model Boundary Conditions

The structural component for which a buckling load prediction has to be made to establish an
allowableloadtobecomparedwiththeactualloadsis,ingeneral,aloadcarryingpart(probablythe
primarystructure)ofthatstructure(e.g.centralcylinderofspacecraft,fueltankofstageofalaunch
vehicle). This structural part is extracted from the structural subsystem and converted into a finite
elementmodelandtheadjacentstructureissimulatedeitherbykinematicsconstraintsortheadjacent
structure is represented by simplified finite element models. The application of adjacent structures
andboundaryconditionsaremostlyprescribed(specified).
10.3.5 Finite Element Model Load applications

The number of design loads and combination of design loads can be minimized by deriving worst
caseloads,envelopingmostoftheloadcombinations.Thedesignloadcasesaremostlydeliveredto
theanalyst.
Thisloadspecificationis,ingeneral,translatedinrunningloadsforthenormalload,bendingmoment
andshearforce.Inadditiontothatuniformlydistributedmechanicalloadsandthermalloadscanbe
applied.Discreteloadisdistributedoverarepresentativearea.
Fluidinertialoadsdependsuponthegravityfactorandistreatedinastaticanalysisaspressureloads.
The loads are treated as dead (conservative) loads or follower forces (pressure stay normal to the
surface).Ingeneral,deadloadswilldo!
171
ECSSEHB3224A
24March2010
10.3.6 Implementation of discontinuities in the finite element model
Thethinwalledstructureisneveraperfectmonocoquestructurewheretheneutralplaneforbending
isnotchanging.Inreallifestructuresthepositionoftheneutralplaneisvaryingdueto,forexample:
a. Ring/shellconnections
b. Stringershellconnections
c. Ringframeshellconnections
d. Reinforcementsaroundmanholes
e. Rivetsandbolts
Offsettingastructural partfromanotherstructuralpartismostlyincorporatedinthefiniteelement
model applying rigid body elements or with very stiff elastic elements. These artificial off sets may
jeopardize the nonlinear analysis. In case when temperature gradients are to be applied the
introduction of rigid elements will probably cause erroneous results if no thermal expansion is
allowed in the rigid elements. A combination of mechanical and thermal loads should be handled
verycarefully.
Itsoundsalittlebitfoolishbutrepresentingyourstructureusingonlysolidelementswillencompass
problemswithdiscontinuitiesintheneutralplane.
Neglecting bolt or rivet connections may result in too stiff finite element models. This can be
investigatedanalyzingdetailedsubmodelsrepresentingtheconnectedparts.
10.3.7 Finite Element Model Check

It is common practice that a finite element or mathematical model representing the structure to be
analysed,isverifiedapplyingmoreorlessstandardchecks.
At first many checks can already be done in the preprocessor used to generate the mathematical
model;e.gverificationoffreesurfacesandboundaries,duplicateelements,normals,connectivity,the
shapeofthe2Dand3Delements,etc.
Runningthefiniteelementpackageadditionalcheckscanbedonetoinvestigateforhiddencontraints.
A freefree modal analysis yields normally 6 rigid body modes. If there are less, not expected
displacementcontraintsareapplied,andif therearemore,unwantedmechanismsareintroducedin
themathematicalmodel.Thehiddenconstraintscanalsobedetectedinstaticwaycalculatingthe6x6
rigidbodystrainenergymatrix.
[U ] = 1 [ R ]T [K ][ R ] = [0] ,
2
where [U]isthematrixofstrainenergies, [ R ] isthematrixof6rigidbodymodeswithrespecttoa
selectedposition,and [K ] isthestiffnessmatrix.
Anotherverypowerfulmathematicalmodelcheckisthestressfreethermoelasticdeformationcheck.
This check verifies model adequacy to perform thermal stress analysis and can be used to find
artificial stiffness introduced in the model. Typically an isothermal expansion analysis is performed
with statically determined boundary conditions. All the thermal coefficients of expansion as well as
Youngs and Poissons moduli are changed to a single value of a dummy isotropic material (e.g.
aluminium alloy). A uniform temperature increase is applied to the model. If the model is clean,
thereshouldbenorotations,reactionloads,elementforces,orstresses.
172
ECSSEHB3224A
24March2010
It is the responsibility of the analysts to mimize as much as possible potential errors in the
mathematicalmodel.
Thedensityofthemeshofthemathematicalmodelshouldbecapabletodescribeadequatelybuckling
modesdependentuponwavesincircumferentialandlongitudinaldirections.Thismeshadequacyto
describe sufficiently accurately buckling modes can be tested for example with axisymmetric
mathematicalmodels(e.g.BOSOR4).Thatmeanstheconvergencestudiesshouldbedoneusingless
complexsolverslikeBOSOR4andothers.
10.4 Thermo-Mechanical Loading

For thermomechanical buckling analysis the distribution of the temperature field all over and
through the structure is necessary. For space applications the temperature distribution is calculated
withspecialpurposethermalanalyzersbasedonthelumpedparametermethod(LPM),e.g.ESATAN,
SINDA,etc..Duetothehighlynonlinearcharacterofthethermalanalysis(temperaturetothepower
4),ingeneral,themathematicalmodelsforthethermalanalysesarecourserthanthestructuralfinite
element representation of the structure. Hence, the temperature distribution obtained by the LPM
programmeshouldbedepictedon thestructuralfiniteelementmodel.TheSINASprogramme(ESA
development), based on the prescribed average temperatures, is quite convenient to perform such a
temperature interpolation. On the other hand the temperature distribution can be calculated by the
finiteelementmethodusingathermalfiniteelementmodel.Theadvantageofusingthefiniteelement
methodforboththethermalandstructuralanalysisisthatnotemperatureinterpolationisneeded.
Thetemperaturefieldcanbeconsideredtobeequivalenttomechanicalloadsandthereforethesame
procedurestocalculatethestabilitycharacteristicsofstructurescanbeapplied.
10.5 Recommended Numerical Procedures
10.5.1 Overview
Itisverydifficulttosaywhichnumericalprocedureshouldbeusedtosolvetheequilibriumequation
ofanonlinearsystem.Theapplicationofcertainnumericalproceduresisstronglydependentonthe
typeofproblemthathastobesolved,forinstance
a. Linearprebucklinganalysis
b. Nonlinearprebucklinganalysis
c. Bifurcationanalysis
d. Postbucklinganalysisaftertheintroductionofartificialorrealmeasuredinitialimperfections.
10.5.2 Load Application

Forlinearstructuralfiniteelementanalysistheloadapplicationisveryclear.Themaximumloador
combinationofloadswillbeappliedandstructuralresponsescalculated.
173
ECSSEHB3224A
24March2010

Linear analysis
n
Nonlinear
2
1
Figure101:Loaddisplacementdiagram
However,thepathofequilibrium(relationloadversusresponse)foranonlinearrespondingstructure
isnotastraightlineasillustratedinFigure101.Theremightbeanoccasionwheretheappliedload
willbebeyondthepathofequilibrium.Hencethesolutionwilldiverge.Topreventthis,thearc
lengthorcontinuationmethods(Riks,Crisfield,Ramm,etc.)willbeused.
The same phenomena can occur also when applying an enforced displacement u . Again this is
illustratedinFigure101.
10.5.3 Iteration Schemes

Because of the nonlinear nature of the equations of equilibrium to achieve the prebuckling state,
beforethebifurcationanalysiscanstart,usuallyanincrementalsolutionprocedureisadoptedtotrace
the entire response of the structure. Using the incremental method the external load is applied in a
sequenceofloadincrementswhicharesmallenoughforthestructuretobeassumedtoberesponding
linearlywitheachincrement[6].Duetothenonlinearnatureoftheequilibriumequationtoobtainthe
equilibriumstatewithinanincrementalloadstepincrementaldisplacementiterationstepshavetobe
done. Amongst many other iteration schemes the most well known iteration schemes used are the
following(seeillustrationinFigure102):
a. StandardNewtonRaphsoniterationscheme(NR)
ItisusuallyveryexpensivetoapplytheNRiterationschemebecauseaftereveryiterationstep
thetangentstiffnessmatrixisassembledandfactorized.Thetypicalconvergencepropertiesare
showninFigure102.
b. ModifiedNewtonRaphsoniterationscheme(mNR)
ThemNRiterationschemediffersfromtheNRiterationschemeinthatthe stiffnessmatrixis
only updated and factorized occasionally. The typical convergence properties are shown in
Figure102.
c. QuasistaticNewtoniterationscheme(QN)
The basic idea of the QN iteration scheme is a good approximation for the tangent stiffness
matrix modifying the previous factorized tangent stiffness matrix. Refactorization of the
updated stiffness matrix can be avoided. The typical convergence properties are shown in
Figure102.
174
ECSSEHB3224A
24March2010
d. NewtonRaphsoniterationschemewithartificialdamping(NRwithdamping)
SometimesNRcalculationsmaynotconvergeinthepostbucklingarea.Theuseofadditional
artificialdampinghelpstoovercomethisproblem.However,onehastobecarefulwiththesize
of damping. A short parametric study for the determination of the appropriate value is
recommended.
ItisagoodthingtostudytheTheoreticalandUsersManualofthefiniteelementprogrammeused
inorder tofindoutwhatthedefaultsettingsare.Itisadvisedtorunsimpledemonstrationproblems
provided either by the finite element programme software supplier or benchmarks provided by
NAFEMSorboth.
Newton-Raphson Modified Newton-

method Raphson method
u n u n
u n 1 u n0 u1n u n 1 u n0 u1n
u n
Broyden-
Fletcher
Goldfarb-Shanno
(BFGS) Quasi-
Newton method
u n 1 u n0 u1n
Figure102:Iterationmethods[6]
175
ECSSEHB3224A
24March2010
10.5.4 Convergence Criteria
Theexperienceisthatthefiniteelementsoftwaresupplierprovidesmanycriteriawhichindicatethe
solution step has been converged. The default settings with the finite element programme used is
mostlyinvestigatedingreatdepthandthereforeitisadvisedtousethesedefaultsettings.
Itisalwaysgoodtoreadsomebackgroundinformationabouttheseconvergencesettingswithinyour
standardfiniteelementprogrammeandtoplayaroundwithit.
10.5.5 Estimation of Bifurcation Points

The estimation of the bifurcation points to investigate the load carrying capability of the buckling
sensitivestructureshouldbedoneinaverysystematicmanner.Thefiniteelementmodelidealization
ofthestructurehasalreadybeencheckedforsuitabilityforbucklinganalysis(meshdensity,boundary
conditions,loadapplication,materialproperties,etc.).Theanalysisofthelowestbifurcationpointis
illustratedinFigure103.
Bifurcation point
Linear analysis
Point of
linearization (pre-
buckling load)
Non-linear analysis Load increment
K = K t ( + ) K t ( )

K t ( )
Figure103:Analysisapproachofbifurcationpoint(typicallyshellstructures)
Theengineersstartingpointwillbetheclassicallinearbifurcationpointanalysistoestimatetheorder
ofmagnitudeofthelowestbucklingload.Themagnitudeoftheprebucklingloadhasnoinfluenceon
the linear buckling analysis result. This linear buckling load is used as a point of orientation in the
sequential nonlinear prebucking analysis followed by the bifurcation (eigenvalue) analysis. It is a
good practice to perform a nonlinear prebuckling analysis applying a load or a set of loads
simultaneously or sequentially up to 80% of the linear bifurcation buckling load. During the
bifurcationanalysis step the maximum appliedload is incremented by . The tangent stiffness
matrix increment K is used as geometrical stiffness matrix in the eigenvalue
problem [{K ( ) + K}]{} = {0} . To achieve the lowest buckling load the load is incremented
by min ,where min isthelowesteigenvalue.Itisagoodpracticetoincreasethemaximumload
as close as possible to the bifurcation point. Restart procedures can be applied to reduce
computationtime(wallclocktime)tocontinuetheanalysisafterthelastloadstep.
176
ECSSEHB3224A
24March2010
10.5.6 Post-Buckling Analysis
Koiter [7] showed in his PhD work (1945) the significant influence of initial imperfections on the
buckling load. The theoreticalaspectsof his workwill not be discussedin this BucklingHandbook.
The interested reader should consult Chapter 3 of Ref. [8] or Chapter 2 of Ref. [9] for a short
introduction to Koiters imperfection sensitivity theory. The sensitivity of the buckling load with
respect to initial imperfection can be studied by the socalled Koiter postbuckling factors a and b .
Thesensitivityanalysisshouldbedonebeforethedetailedfiniteelementanalysesevenstart.
Bifurcation point
Linear analysis
Limit point
Non-linear analysis
Equilibrium path
Post-buckling analysis
Figure104:Postbucklinganalysis
To force the response of the buckling sensitive structure in the direction of the bifurcated path of
equilibrium (see Figure 104) initialimperfection(manufacturingsignature) isadded to thenominal
geometryofthestructure.Thiscanbedoneintwoways:
a. Theshapeinitialimperfectionisbuildupfromacombinationofthelowestandlowerbuckling
modes.Theamplitudeoftheimperfectionisincaseofthinwalledshellstructuresproportional
tothewallthicknessoftheshell;e.g.10%100%ormore.Forastart50%isagoodnumber.
b. Ifpossiblethemeasuredinitialimperfectionsisapplied(seealsoChapter9).
In general initial imperfections should be added to reflect the likely manufacturing signature of the
real structure. In doing so, bifurcation points are transformed to limit points, but these can be very
sharpbecauseofthesmallimperfectionsizesensitivity.
10.6 Allowable Buckling Load

Toverifyastabilitycapabilityofthe(thinwalled)structureundercompressionandthermalloadsthe
allowablebucklingload(stressdistribution)iscomparedwiththeoccurringstressdistributioncaused
bytheapplieddesign(limit)loads.Ingeneralapositivemarginofsafety(MoS)isachieved.TheMoS
valueisdefinedbythefollowingequation
177
ECSSEHB3224A
24March2010
Allowable _ load
MoS = 1 0, 101
FoS * Design _ load
where the factor of safety is indicated by FoS. For the applications of the factors of safety (FoS) the
interestedreadershouldconsultthestandardECSSEST3210.
The crucial parameter in the verification process of the loaded structure is the determination of the
allowable(compression)load.
Theallowablebucklingloaddependsonthetypeofstabilityanalysis(seeFigure105):
a. Linearbifurcation
b. Nonlinearbifurcation
c. Postbucklinganalysis
TheuncertaintiesinthebucklinganalysisapproachareexpressedinanassociatedFoScorresponding
to the type of analysis. It is clear one strives to calculate the allowable load by applying a post
bucklinganalysisdependingonthekindandmagnitudeoftheinitialimperfections(manufacturing
signature),probablyincombinationwiththeKoiterpostbucklingcoefficientsaandb.
Bifurcation point

Limit point
Allowable load
Figure105:Allowableloaddefinition
10.7 Finite Element Programmes with Buckling Analysis

Capabilities
10.7.1 Overview
InthissectionsomeofCommercialOntheShelf(COTS)availablefiniteelementsoftwarepackagesare
mentionedandtheircapabilitieswithrespecttononlinearbucklinganalysisarebrieflyrecalled.The
following (non exhaustive) list of finite element codes applied in the aerospace industry, as well as
researcharementioned:
178
ECSSEHB3224A
24March2010
a. ABAQUS
b. ANSYS
c. MSC.Marc
d. MSC.Nastran
e. ABAQUS/EXPLICIT
10.7.2 ABAQUS/Standard
ABAQUS/Standard contains the capability for estimating elastic buckling by eigenvalue extraction.
This estimation is typically useful for stiff structures, where the prebuckling response is almost
linear. The buckling load estimate is obtained as a multiplier of the pattern of perturbation loads,
whichareaddedtoasetofbasestateloads.Thebasestateofthestructuremayhaveresultedfrom
any type of response history, including nonlinear effects. It represents the initial state to which the
perturbation loads are added. The response to the perturbation loads must be elastic up to the
estimatedbucklingloadvaluesfortheestimatestobereasonable.FormoreinformationseeABAQUS
TheoryManual,section2.3.
Acombinationofscaledlinearbucklemodescanbeusedasgeometricalimperfectionforanonlinear
postbucklinganalysis.FortheloadincrementsABAQUSofferstheRiksRammarclengthmethodto
covertheinstablesolutionduringthebucklingphase.
Calculations in the postbuckling area may not converge if the NewtonRaphson method or Riks
methodisapplied.ABAQUSofferstheNewtonRaphsonmethodwithartificialdamping(STABILIZE
command)whichhelpstoovercomethisproblem(cf.Section10.5.3).
10.7.3 ANSYS
The Finite Element Code ANSYS can be used for linearized and nonlinear buckling analyses. Block
Lanczos and Subspace eigenvalue algorithms are available for the determination of buckling loads
and related modes. For nonlinear postbuckling as well as snapthrough analyses forcecontrolled,
displacement controlled and arclength (Crisfield) methods are available. The NewtonRaphson
methodisusedassolverwhenforceordisplacementcontrolisapplied.
GeometricimperfectionscaneasilybeintroducedintothemodelusingtheresultsofANSYSanalyses
(e.g. buckling modes, eigenmodes). In this context the upgeomfunction allows the specification of
theinitialdeformationamplitudes.Theoptionofanaccompanyingeigenvalue(buckling)analysisis
available in ANSYS. Additionally, ANSYS allows the consideration of nonlinear material behaviour
(e.g.plasticity,viscoelasticity)andcontact(e.g.3Dshell)fornonlinearbucklinganalyses.Composite
materialsincludingrelatedfailureanalysis(e.g.TsaiWu,Puck)canalsobetreatedwiththesoftware.
10.7.4 MSC.Marc/MSC.Nastran
MSC.Softwareoffersthreeimplicitsolutionsforsimulationofgeometricnonlinearbehaviourofpost
buckling:
a. MSC.NastranSOL106
b. MSC.NastranSOL600
c. MSC.Marc
179
ECSSEHB3224A
24March2010
For all three solutions the post buckling behaviour can be simulated including nonlinear material
behaviour such as plasticity. For all three solutions composite material formulation can be used.
DelaminationanddamageisavailableinMarcandSOL600(version2005r3).
In SOL600 and Marc, a scaled linear buckle mode can be used as geometrical imperfection for a
nonlinearpostbucklinganalysis.
For the load increments all three solutions offer an arclength method to cover the instable solution
duringthebucklingphase.Followingmethodsareavailableinallsolutions:
1. Crisfield
2. RiksRamm
3. ModifiedRiksRamm
CrisfieldwithautomaticswitchtothemethodRiksRammfornonrealroots(SOL600andMarconly).
ContactcanbemodelledinSOL106withNodetoNodeContact(withGapelements)whileinSol600
orMarcafreesurfacecontactcanbeusedforpostbucklingsimulations.
InadditiononecanuseinSOL600andMarcdampingschemestosimulatebucklinginsteadofusing
arclengthmethods.
Differentinertialdampingschemescanbeusedinaquasistaticanalysisforsimulatingpostbuckling:
inertialdampingbasedonthepredictedstrainenergy.
userscalableinertialdampingmaybeintroduced(optional,onlyactiveduringinstabilities).
Thesedampingmethodscanalsobeusedinacoupledthermomechanicalanalysis(MSC.Marconly)
orthermalbuckling(SOL600andMarconly).
ForthermalbucklinginSOL106thearclengthmethodscanbeused.
10.7.5 ABAQUS/EXPLICIT
ABAQUS/EXPLICIT solves the equations of equilibrium governing the dynamic response by an
explicit integration operator, using the explicit central difference formula. The buckling phenomena
can be investigated, simulating the dynamic of a slow buckling test with an assigned displacement,
which is what often happens in the experimental tests. In this way, the dynamic analysis is able to
followthecurveofthereactionforceversustheimposeddisplacementinthepostbucklingfield.The
dynamicanalysisdoesnotrequireanyparticularadditionalinput.Inparticular,theintegrationtime
step is automatically computed by the code, and the value of the buckling loads does not result
sensitivetotheartificialstiffeningparametersandtothestructuraldamping.
10.8 Guidelines to select a solution scheme and solution

procedure
10.8.1 Guidelines to select a solution scheme

In this section some guidelines to select solution schemes are provided. These guidelines are partly
takenfrom[10].ForadditionalinformationtheinterestedreadershouldconsultReferences[11]and
[12].
180
ECSSEHB3224A
24March2010
Table101:Guidetoselectsolutionscheme
SolutionScheme Usewhen Remarks

Linearizedbuckling a. Analyzing thin walled structures Tendstooverpredictbuckling
analysis undercompressionloading load
b. The structure is not imperfection

sensitive
c. Initial imperfections are
negligible
d. Only the shape of the buckled
configuration is required for the
buckling load and the actual
displacements
e. Nopostbucklingisrequired
f. Thematerialresponseiselastic
Linearizedbuckling g. Analyzing thin walled structures Preload should be
analysiswithpreloading undercompressionloading applied using nonlinear
analysis
h. Initial imperfections are not
negligible The closer the preload is
totheactualcriticalload,
i. Only the shape of the buckled
the more accurate the
configuration is required for the
bucklingload
buckling load and the actual
displacements
j. No postbucklng information is
required
Arclengthmethod k. Limit points exist along the Solution will tend to
equilibriumpath oscillate around a
bifurcationpoint
l. Plasticitymaybepresent
Bifurcation points may
m. Post buckling information
be removed by the
required
introduction of
imperfections, usually
introduced as scaled
mode shapes obtained
from previous
eigenvalue analysis,
eitherlinearornonlinear
May still encounter
convergence difficulties
in the presence of sharp
limitpoints
181
ECSSEHB3224A
24March2010
10.8.2 Guidelines for the solution procedure
The following guidelines to perform a buckling analysis were developed during a design review of
theVEGAinterstage 1 / 2 .Thestepbystepanalysisapproachisdefinedasfollows:
a. Apply or incorporate specified forces (Normal force, bending moment, shear force) and
specified boundary conditions or adjacent structures in the mathematical or finite element
model.
b. Performalinear(linearized)bucklinganalysistoestablishtheupperboundofthebucklingload
c. Perform a nonlinear buckling (bifurcation) analysis and determine the lowest buckling mode
closetothebifurcationpoint.Thismodemayindicatethemostcritical(worst)directionifholes
weakenthestructure.
d. Apply besides the normal load the shear load and the bending moment in the most
sensible/critical direction. Rerun the nonlinear buckling analysis and determine the buckling
modesassociatedwiththelowestbucklingeigenvalues.
e. Repeat the previous step d. for two other lateral directions, however, closely to the direction
usedinstepd.
f. Select from the the analyses in the three lateral load directions the direction with the lowest
calculatedbifurcationload.
g. Calculate and visualize the buckling modes for the selected lateral load direction. Sufficient
m bucklingmodesclosetothebifurcationpointshouldbecalculated.
h. Plot the resulting buckling eigenvalues of step g. in the socalled curve. Select from this
curveintherange 1.2 lowest andavailablebucklingmodesthesignificantbucklingmodes
in that range ( n < m modes selected). Each of the n buckling modes will be applied as an
initial imperfection on the perfect geometry of the shell. The maximum imperfection per
buckling mode is scaled to 50% of the wall thickness. The scaling factor depends on the
manufacturingprocessandprocedures.
i. Runthe n nonlinearpostbucklinganalyseswiththeselectedimperfectionsaswellasforthe
linearcombinationoftheimperfectionsagainscaledto50%(orotherwise)ofthewallthickness.
Consequently n + 1 nonlinearpostbucklinganalysehavetobecarriedout.
NOTE For each nonlinear buckling or post buckling analysis, it has to be
checked that the maximum stress just before the bifurcation point and
limitpointdoesnotexceedtheyieldlimit(onlymetallicstructures)ofthe
material.Iftheyieldlimitisexceeded,thematerialnonlinearityhastobe
includedintothebucklinganalysis.
10.9 References
[1] Bushnell,D.(1972)Stress,Stability,andVibrationofComplexBranchedShellsofRevolution:
AnalysisandUsersManualforBOSOR4,NASACR2116
[2] Keller,H.,NumericalMethodsforTwoPointBoundaryValueProblems,BlaisdellPublishing
Co.,Waltham,Mass.,1968
[3] Cohen,G.,ComputerAnalysisofAsymmetricFreeVibrationsofRingStiffenedOrthotropic
ShellsofRevolution,AIAAJournal,3(12):23052312,1965
[4] Booton,M.,BucklingofImperfectAnisotropicCylindersunderCombinedLoading,UTIAS
Report203,UniversityofToronto,1976
182
ECSSEHB3224A
24March2010
[5] MacNeal,R.H.(1994)FiniteElements:TheirDesignandPerformance,MarcelDekker,Inc.,
ISBN0824791622
[6] Hinton,E.(1992)NAFEMS,IntroductiontoNonlinearFiniteElementAnalysis,NAFEMS,ISBN
187437600
[7] Koiter,W.T.,OntheStabilityofElasticEquilibrium,NASATTF10833,1967
[8] Singer,J.,Arbocz,J.andWeller,T.,BucklingExperimentsExperimentalMethodsinBuckling
ofThinWalledStructures,Volume1,JohnWiley&Sons,ISBN0471956619,NewYork,1998
[9] Arbocz,J.,PotierFerry,M.,Singer,J.AndTvergaard,V.,BucklingandPostBuckling,Four
LecturesinExperimental,NumericalandTheoreticalSolidMechanics,SpringerVerlag,ISBN3
540183124,BerlinHeidelbergNewYork,1987
[10] Falzon,B.G.,Hitchings,D.(2006),AnIntroductiontoModellingBucklingandCollapse,
NAFEMSpublication,ISBN1874376182
[11] Riks,E.(1970)OntheNumericalSolutionofSnappingProblemsintheTheoryofElastic
Stability.DepartmentofAeronauticsandAstronauticsStanfordUniversity,AFOSR702258TR
[12] Thorton,E.A.(1996)ThermalStructuresforAerospaceApplication,AIAAEducationSeries,
ISBN1563471906
183
ECSSEHB3224A
24March2010
11
Strategy for hierarchical high fidelity
analysis applied to stability analysis
11.1 Introduction
Itisgenerallyagreedthat,inordertomakethedevelopmentofAdvancedSpaceTransportationSystems
a success and to achieve the very ambitious performance goals (like every generation of vehicles 10x
saferand10xcheaperthanthepreviousone),itisnecessarytomakefullandefficientuseofthetechnical
expertiseaccumulatedinthepast50yearsorso,andtocombineitwiththetremendouscomputational
powernowavailable.Itisobviousthatwiththestrictweightconstraintsusedinspaceapplicationsthese
performance goalscanonlybeachievedwithanapproachoftencalledhighfidelityanalysis,where
theuncertaintiesinvolvedinadesignaresimulatedbyrefinedandaccuratenumericalmodels.Inthe
end the use of high fidelity numerical simulation will also lead to overall cost reduction, since the
analysis and design phase will be completed faster and only the reliability of the final configuration
needstobeverifiedbystructuraltesting.
Thelightweightshellstructuresusedinaerospaceapplicationsareoftenbucklingcritical.Thebuckling
load calculations are usually carried out by one of the many currently available finite element based
computercodes(e.g.[1],[2]).Inordertoreducecomputerexecutiontime,bucklinganalysesareoften
doneusingonlythesmalldisplacementstiffnessmatrixK0.Thisapproachisused,despitethefactthat
theinitialstabilityproblemsoformulatedcanonlygivephysicallymeaningfulanswersiftheelastic
solutionsbasedonK0(atleastapproximately)areidenticallyequaltozero[3].
When the qualitative nature of the expected behavior is completely unknown, the stability of the
structureshouldbeinvestigatedusingthefulltangentstiffnessmatrixKTinordertoguaranteeaccurate
andreliablebucklingloadandbucklingmodepredictions.Inordertodiscovertheloadlevelatwhich
KT ceases to be positive definite (that is, the load level when buckling occurs), a stepbystep analysis
procedureisneeded.
Inaddition,itisimperative(thoughoftencompletelyneglected),thatatthebeginningofanystability
investigationtheaccuracyofthediscretemodelusedshouldbecheckedagainstavailableanalyticalor
semianalytical results. This step is part of a mandatory study needed in order to establish the
dependence ofthe bucklingloadpredictionsonthe meshdistributionused.Furthermore,ashasbeen
pointed out in the past by Byskov [4], if one carries out imperfection sensitivity investigations, which
involveanextensionofthesolutionintothepostbucklingresponseregion,furthermeshrefinementis
neededsincethewavelengthofthedominantlargedeformationpatternoftendecreasessignificantly.
Finally,wheneveroneisengagedinshellstabilityanalysisitisespeciallyimportantthatoneisawareof
thepossibledetrimentaleffectsofawholeseriesoffactors,thathavebeeninvestigatedextensivelyinthe
late1960sandtheearly1970s.Thusforanaccurateandreliablepredictionofthecriticalbucklingloadof
arealstructure,oneshouldaccountnotonlyfortheinfluenceofinitialimperfections(e.g.,[5],[6])andof
the boundary conditions (e.g. [7]), but one should also consider the effects of stiffener and load
eccentricity(e.g.,[8])andtheprebucklingdeformationscausedbytheedgerestraints(e.g.[9],[10]).
184
ECSSEHB3224A
24March2010
11.2 Hierarchical high fidelity analysis

WhenoneusestheHighFidelityAnalysisapproachtocarryoutbucklingloadcalculationstheperson
performing the analysis should demonstrate beyond reasonable doubt that the results obtained are
accurateandreliable.IthasbeenshownbyArboczetal.[11],[12]thatinordertoarriveatareliable
predictionofthecriticalbucklingloadandtomakeanestimateofitsimperfectionsensitivitywhich
canbeusedwithconfidence,oneshouldproceedstepbystepfromsimpleanalyticalsolutions(Level
1)tomorecomplexmodelsandsolutionprocedures(Level3).
Central in the hierarchical approach is the possibility of identification of an idealized structure that
correspondstotherealstructureunderconsideration.Anidealizedstructureisareferencemodelthatis
a simplification of the real structure under consideration but represents important characteristics of
the geometry and behavior of the real structure and for which analytical and/or semianalytical
methodshavebeendeveloped.
Formanypracticalstructuralcomponentsthisreferencemodelisavailableinanaturalway.Forspace
applications (spacecraft and launch vehicles), different types of shells of revolution constitute the
idealizedstructures.Shellsofrevolutionarethemostcomplicatedidealizedstructuresthatexistand
they form the idealization of many important structural components in various branches of
engineering. The analysis of shells of revolution is therefore playing a key role in the hierarchical
approach,whichconsistsofmethodsatthreelevelsofcomplexity.
Level1solutionsuseasingletermdoubleFourierseriesapproximationtoreducethesolutionofthe
stability problem, which is formulated in terms of partial differential equations, to algebraic
eigenvalue problems. The effect of edge restraint is neglected (one uses a membrane prebuckling
solution) and the assumed field functions satisfy approximately the classical SS3 boundary
conditions. This approach yields, for instance, for isotropic shells the socalled classical solution
(Lorenz[13],Timoshenko[14],Southwell[15]).
m
m Nc
c = = 1.0 111
N cl
where
Eh2
Nc = , c = 3 (1 - 2 ) 112
l cR
This buckling load is a repeated eigenvalue with a high degree of multiplicity. The locus of the
repeatedeigenvaluesisthesocalledKoitercircle[5].
Level2 solutions eliminate the circumferential or ydependence of the assumed solution by a
truncated Fourier decomposition in the circumferential direction. The resulting system of nonlinear
ordinarydifferentialequationsissolvednumerically,wherebyboththespecifiedboundaryconditions
and the effect of edge restraint are rigorously satisfied. Thus, with this approach, the only
approximationusedfordeterminingtheshellresponseistheonethatrepresentsthevariationofthe
solution in the circumferential direction by a single harmonic with n full waves, whereby a search
over the range of integer values of n, an nsearch is used to establish which wave number is the
criticalone.Theinfluenceofthedifferentboundaryconditionscanthusbeinvestigated.
TheLevel3solutionsarebasedeitheronatwodimensionalfinitedifference(Almrothetal.[16])or
finite element ([2] and [17]) formulation. In both cases, if one uses the appropriate meshes, one can
obtain rigorous solutions where all nonlinear effects are properly accounted for. Thus initially a
convergence study should be carried out in order to establish the mesh size needed to model
accuratelytheresponseandbucklingbehavioroftheshellinquestion.
185
ECSSEHB3224A
24March2010
11.3 Flow chart

Inordertoestablishastrategytohandlebucklingphenomena,thestepsinFigure111areproposed.
The logic will be given as a flow chart, to emphasize the importance of sequence. The hierarchical
approachisreflectedintheitems3,4,and5oftheflowchart.Theflowchartisfurtherexplainedin
thesubsequentsections.
Figure111:Flowchartforhighfidelityanalysis
186
ECSSEHB3224A
24March2010
11.4 Description and General Guidance
11.4.1 Overview
In this section, the flow chart showing the strategy for nonlinear buckling analysis is described in
moredetail.
11.4.2 Definition of the Problem

a. Structuralcharacteristics
b. BoundaryConditions(kinematical,statical)
c. LoadTypes(Pointloads,distributedloads)
d. Estimationofstructuralloadcarryingbehaviour(membrane,bending,)
e. Materialbehaviour(isotropic,anisotropic,nonlinearelastic,inelastic,)
11.4.3 A Priori Determination of Possible Failure Modes

a. Structuralbifurcation(classicalbuckling)
b. Nonlineareffects
c. Materialbifurcation
11.4.4 Definition of the Strategy and Mathematical Model

a. Definitionofidealizedstructure:AnalyticalModel,SemianalyticalModel,andFiniteElement
Modelofidealizedstructure
b. DefinitionofFiniteElementModel
11.4.5 Analytical Solution, Semi-Analytical Solution, and Finite Element

Solution of Idealized Structure
Clearassumptionstoconsiderpotentialrestrictionsduringinterpretationandassessment
Clearselectionofthechosenparameters
Whichphenomenashouldandcanbeconsidered
Checkofformulasused
11.4.5.1 LEVEL 1 Analytical solution of idealized structure

1. Fourierseriesdiscretizationanalyticalsolutions
2. SeealsoChapters15,16and17forspecificstructuralelements
AttheLevel1,methodsbasedonseriesexpansions(trialfunctionmethods)areused.Thetwo
most commonly used methods are the Galerkin method, which starts from the governing
differentialequations,andtheRayleighRitzmethod,whichstartsfromanenergyexpression.
Sets of algebraic equations are obtained. Often simple support boundary conditions are
assumed,incombinationwithamembraneprebucklingstate.
(a) Theory
187
ECSSEHB3224A
24March2010
(b) Results
(1) Bucklingloadcharts(eigenvaluemaps)
(2) Prebuckling,bucklingandpostbucklingmodes
(c) ImperfectionSensitivityStudies
(1) Singleaxisymmetricimperfection
(2) Singleasymmetricimperfection
(3) Multimodeimperfection
(d) EffectofParameterUncertainties
(e) Limitations
11.4.5.2 LEVEL 2 Semi-analytical solution of idealized structure

1. Fourier series in one direction (circumferential direction), numerical discretization in
seconddirection(meridionaldirection)Semianalytical(analyticalnumerical)solutions
foraxisymmetricstructures
2. NonlinearanalysisusingshellofrevolutioncodeslikeBOSOR,SRAorsimilarcodes
3. SeealsoChapters15,16and17forspecificstructuralelements
AttheLevel2,aonedimensionaldiscretizationisobtainedafteraFourierdecompositioninthe
circumferential direction of the shell has been carried out. Very accurate solutions can be
obtained, including the effects of boundary conditions and a nonlinear prebuckling state, by
solving the resulting sets of ordinary differential equations for the meridional direction
numericallybymeansoftheShootingMethodortheFiniteDifferenceMethod(seeRef.[18]).
ThesesolutionsformreferencesolutionsfortheFiniteElementmodeloftheidealizedstructure.
WhentheFiniteElementresultsoftheidealizedstructureareinsatisfactoryagreementwiththe
resultsoftheLevel2analysis,theFiniteElementmodeloftherealstructurecanbedevelopedon
thebasisoftheFiniteElementmodeloftheidealizedstructure.Thiscanbeseenasanessential
firststepinthedevelopmentoftheFiniteElementmodeloftherealstructure.
(a) Theory
(b) Results
(1) Bucklingloadcharts
(2) Prebuckling,bucklingandpostbucklingmodes
(c) InfluenceofBoundaryConditions
(1) Classicalboundaryconditions
(2) Elasticboundaryconditions
(d) RefinedImperfectionSensitivityStudies
(1) Singleaxisymmetricimperfection
(2) Singleasymmetricimperfection
(e) CorrelationwithLevel1results
(f) Limitations
11.4.5.3 LEVEL 3 Finite Element solution of idealized structure

1. Twodimensionalnumericaldiscretization
2. Refinednonlinearanalysisoftheidealized,axisymmetricstructureusingcodesforshells
withgeneralshapelikeSTAGS,ABAQUSetc.
188
ECSSEHB3224A
24March2010
3. SeealsoChapter8
4. SeeforfurtherdetailsSubsection10.3
(a) Theory
(b) FiniteElementModelDevelopmentincludingElementSelection
(c) MeshConvergenceStudies
(d) CharacterizeErrorsduetoModellingAssumptions
(e) Results
(f) CorrelationwithLevel2results
NOTE ONLY when the correlation of Level 3 Idealized structure results
with the LEVEL2 results is satisfactory should one proceed with the
executionoftheFiniteElementanalysisoftherealstructure(Level3
Realstructure).
11.4.6 Finite Element Analysis of the Real Structure

a. Twodimensional(orthreedimensional)numericaldiscretization
b. Refinednonlinearanalysisoftherealstructure.Thefiniteelementmodeloftherealstructureis
basedontheFiniteElementmodelusedfortheidealizedstructuremodifiedtoinclude
1. localdiscontinuitieslikeholes,etc.
2. stiffeners
3. measured(orassumed)surfaceimperfections
4. measured(orassumed)boundaryimperfections.
c. ReferalsotoChapter8
1. Theory
2. FiniteElementModelDevelopmentincludingElementSelection
The selection of the right elements is an essential task. In order to investigate the
phenomenaofconcern,theuseroffiniteelementcodesshouldinformhimselfaboutthe
abilityoftheelementused.Iftheelementcodedoesnotprovideelementbenchmarks,it
is recommended to perform a benchmark test on a given small example. Such
benchmarksaregiveninseverallibraries.Forexample:
NAFEMS, independent international authority on the use of computer modeling and
simulationmethodsandtools.
Somepropertiesofelementscanbecheckedapriori:
(a) Numberofnodes,DOFs
(b) Displacementelement
(c) Mixedformulation
(d) Integrationpointsinthearea
(e) Integrationpointsinthesection
(f) Reducedintegrated
(g) Hourglasscontrol
(h) Fullyintegrated
189
ECSSEHB3224A
24March2010
(i) Shearlockingregularizations
3. MeshConvergenceStudies
4. CharacterizeErrorsduetoModellingAssumptions
(a) Clear assumptions to consider potential restrictions during interpretation and
assessment
(b) Checkofelementsused
(c) Whichphenomenashouldandcanbeconsidered?
(d) Checkofthemodel(plausibilitychecks,convergencechecks)
5. Results
TheessentialstepsforaFiniteElementanalysisaresummarizedintheTable111.
190
ECSSEHB3224A
24March2010
Table111:Essentialstepsforafiniteelementanalysis
Whataretheelementsabletomap?
Interpretationoffirstbifurcationpoints(local,global)
Checkoftheequilibriumpath
NecessityofNonlinearinvestigations(geometricaland/ormaterial)
Typeoffailure:
Localbuckling
Globalbuckling
Plasticcollapse
Whatisthedriver?
Thestructure(thestiffnessmatrix)
Thematerial(thematerialtensordominatesthestiffnessmatrix)
Astabilitypoint(bifurcationorlimitpoint)isgiven[19],[20]:
K ( u )ij j = 0i
113
det K ( u )ij = 0
K stiffnessmatrix
u vectorofstatevariables
Eigenvector
Inordertodifferentiatethelimitpointfromthebifurcationpoint[14]:
Limitpoint
T
i Pi 0 114
Bifurcationpoint
T
i Pi = 0 115
P vectorassociatedwiththepatternofappliedloads
Incaseofconsiderationofmaterialplasticity,Druckersstabilitypostulatehastobetakenintoaccount:
& jl Tijkl &ik 0 116
T Materialtensor
& Straingradienttensor
Typeofequilibrium:
Static
Dynamic
Incaseofclassicalsnapthroughphenomena,atleastasimpledynamiccheckrunshouldbedone.
191
ECSSEHB3224A
24March2010
11.4.7 Test
a. Testprocedure
b. ReferalsotoChapter9
1. Initialimperfectionmeasurements
(a) Shellwallgeometricimperfections
(b) Shellwallthicknessimperfections
(c) Shellendshapeorloadingsurfaceimperfection
2. Pretestloadingplatenadjustments
3. Requiredtestdataforanalysiscorrelation
4. Specimenparameteruncertaintyorvariabilitycharacterization
(a) Materialproperties
(b) Fibervolumefractioninfibermatrixcomposites
(c) Imperfectionmeasurementuncertainty
11.4.8 Assessment of the Structure

a. Computationalassessmentoftherealstructure
b. Comparisonwithtest
c. RiskAssessment
1. Possibilisticanalyses
2. Probabilisticanalyses
TheessentialstepsinthecomputationalassessmentofthestructurearesummarizedinTable112.
192
ECSSEHB3224A
24March2010
Table112:Essentialstepsforacomputationalassessmentofthestructure
Structuralfailure
Whichgeometricalandphysicalimperfectionshavebeenapplied?
Imperfectionsensitivityinvestigationbyacoupleofdifferentshapes
Istheamplitudeofimperfectionanimportantindicator?
Isthereanydangerthatthereferenceconfigurationswitchesintoanabsolutelydifferentdeformation
pattern(farawayfromimperfectionorlineartheory),whichcannotbediscoveredinthepresent
model.
Is it due to the bending effects that the initial buckling problem switches to a highly nonlinear
deformationproblem?
Ifmaterialnonlinearitydominatesthefailure
Isthechosenconstitutivelawsufficienttodescribethephenomena?
Shouldtheconstitutivelawbeenhanced?
Strainlocalisation
Meshdependency
Solverproblems
Errorestimationandassessment
Whichregularisationmethodscanbeappliedtobypasssomedrawbacksinmodels,assumptionsand
methods?
Isacrosscheckingprocedurenecessary?
Ifyes,how
Withotherelements
Withotheranalyticalorempiricalresults
Incaseofdynamicanalyses
Implicit
Explicitalgorithms
Isthereanydangerforsolutiondrifting?
11.5 Hierarchical High Fidelity Stability Analysis of

Anisotropic Cylinders
11.5.1 Overview
A recent test series of composite shells carried out at NASA Langley Research Center is used to
illustrate how such a hierarchical approach to buckling load calculations can be carried out. The
platform for the multilevel computations, needed for an accurate prediction of the critical buckling
loadsandareliableestimationoftheirimperfectionsensitivity,isprovidedbyDISDECO[21].With
this open ended, hierarchical, interactive computer code the user can access from his workstation a
successionofprogramsofincreasingcomplexity.
193
ECSSEHB3224A
24March2010
InthefollowingitwillbeshownthatwiththehelpofDISDECO,theDelftInteractiveShellDEsign
COde, the shell designer can study the buckling behavior of a specified shell, calculate its critical
bucklingloadquiteaccuratelyandmakeareliablepredictionoftheexpecteddegreeofimperfection
sensitivity ofthe critical bucklingload. The proposed procedureconsists ofahierarchical approach,
where the analyst proceeds stepbystep from the simpler (Level1) methods used by the early
investigatorstothemoresophisticatedanalyticalandnumerical(Level2andLevel3)methodsused
presently.
11.5.2 Level-1 Perfect Shell Buckling Analysis

The geometric and material properties of the 8ply,composite shell with symmetrical layup of Ref.
[22]arelistedinTable113.
Table113:GeometricpropertiesofNASAlayeredcompositeshellC3 [ 45 / 0 / 90]s
t tota l ( = h ) =0.04 in(=1.01600 mm )
L =14.0 in(=355.600 mm )
R =8.0 in(=203.200 mm )
E11 =18.5x 106 psi(=12.7552x 10 4 N/ mm2 )
E22 =1.64x 106 psi(=1.13074x 10 4 N/ mm2 )
G12 =0.87x 106 psi(=5.9984x 103 N/ mm2 )
12 =0.3
Note:Symmetricallayupwith8plysofequalthicknesses(=0.005in)
Assumingaperfectshell W = 0 andthefollowingmembraneprebucklingstate
W
(0) *
= hWv = hA12
c
117
2
F
(0) =
Eh 1 2
y
cR 2
where
=

cl
=
Nx
N cl
; cl =
Eh
cR
; N cl = cl h and (
c = 3 1
2
)
then the nonlinear equations governing the prebuckling state are identically satisfied and the
linearizedstabilityequationsreducetoasetofequationswithconstantcoefficients.Ithasbeenshown
inRef.[23]thatbyassuminganasymmetricbifurcationmodeoftheform
(1)
W = h sin m
x
L
cos
n
R
( y K x) 118
where
m=k=numberofaxialhalfwaves
n= l =numberofcircumferentialfullwaves
K =Khotsskewednessparameter
onecanreducethesolutionofthelinearizedstabilityequationstoanalgebraiceigenvalueproblem.
Notice that the eigenvalue mn depends besides on the wave numbers m and n also on Khots
194
ECSSEHB3224A
24March2010
skewednessparameter K ,arealnumber.Thecriticalloadparameter c isthelowestofallpossible
eigenvalues.Thusfinding c involvesnotonlyasearchovertheintegervaluedwavenumbersmand
nbutthesearchhastoberepeatedoverarangeofpossiblepositiveandnegativerealnumbersfor K .
Using the Level1 computational module AXBIF [23] a search over integer valued axial halfwave
numbers m and over a range of possible positiveand negative real numbers K yielded the lowest
eigenvalueslistedinTable114forthespecifiedcircumferentialwavenumbersn.
Table114:BucklingloadsoftheNASAlayeredcompositeshellC3[22]
BucklingloadmapfortheperfectshellusingAXBIF[15]
( N cl = 2239.342 lb / in )
n=4 m
c =0.371647(m=7, K = 3.322)
n=5 m
c =0.370329 (m=7, K = 2.662)
n=6 m
c =0.370073 (m=7, K = 2.179)
n=7 m
c =0.366056 (m=1, K = 0.011)
n=8 m
c =0.370173 (m=6, K = 1.681)
n=9 m
c =0.372036 (m=6, K = 1.489)
n=10 m
c =0.372927 (m=5, K = 1.348)
n=11 m
c =0.375303 (m=4, K = 1.218)
n=12 m
c =0.369137 (m=1, K = 0.581)
n=13 m
c =0.370143 (m=1, K = 0.656)
n=14 m
c =0.376835 (m=1, K = 0.736)
Noticethatbesidestheabsoluteminimumof cm =0.366056atn=7thereisalocalminimumof cm =
0.369137at n = 12 .
To facilitate the interpretation of the numerical results obtained, DISDECO provides the user with
variousgraphicalinterfaces.
Thustheresultsofthesearchforthecritical(lowest)bucklingload c canbedisplayedinacontour
mapasshowninFigure112.Usingmembraneprebucklingthecriticaleigenvalueis(seealsoTable
114)
cm = 0.366056
withm=1halfwavesintheaxialdirectionandn=7fullwavesinthecircumferentialdirection.Inorder
to provide a quick overview of the distribution of eigenvalues, the values displayed in the contour
plotarerenormalized.ThusinFigure112thefollowingrenormalizedeigenvaluesareplotted
mn
cm = 119
0.366056
Noticethatthecriticalbucklingloadcanbecalculatedusingasimplemultiplication
N cm = cm N cl = 0.366056 ( 2239.342) = 819.725 lb/in
195
ECSSEHB3224A
24March2010
Figure112:DistributionofLevel1membraneprebucklingbucklingloads
NASAquasiisotropiccompositeshellC3[22]

To investigate the effects of edge constraint and of different boundary conditions on the critical
buckling load of the perfect shell ( W = 0) one has to switch to the Level2 computational module
ANILISA[24].Inthismoduletheaxisymmetricprebucklingstateisrepresentedby
W
( 0) = hWv + hw0 ( x )
2 1110
F
(0) =
1 2 2
Eh
y + R f0 ( x )

cR 2
IthasbeenshowninRef.[24]thatwiththeseassumptionstheprebucklingproblemisreducedtothe
solution of a single fourth order ordinary differential equation with constant coefficients, which
always admits exponential solutions. Closed form solutions for simply supported and clamped
boundaryconditionshavebeenpublishedintheliterature[25].
Foranisotropicshellsthelinearizedstabilityequationsadmitseparablesolutionsoftheform
(1) = h w
W 1 ( x ) cos n + w2 ( x ) sin n
2 1111
F
(1) =
ERh
f1 ( x ) cos n + f 2 ( x ) sin n
c
y
where = .
R
UsingageneralizationofStodolasmethod[26]firstpublishedbyCohen[27]theresultingnonlinear
eigenvalue problem is reduced to a sequence of linearized eigenvalue problems. The resulting
196
ECSSEHB3224A
24March2010
ordinary differential equations are solved numerically by a technique known as parallel shooting
over Nintervals [28]. Notice that by this approach the effect of edge restraint and the specific
boundary conditions are satisfied rigorously. To find the critical load parameter c an nsearch is
carriedout,wherebyoneshouldbecarefultofindnotalocalminimumbuttheabsoluteminimum.As
canbeseenfromtheresultspresentedinTable115thensearchusingmembraneprebucklinganda
rigoroussatisfactionofSS3 ( N x = v = w = M x = 0) boundaryconditionsforthestabilityproblemnow
yieldsalocalminimumof c = 0.364776 atn=7andanabsoluteminimumof c = 0.364417 atn=11.
ThemostaccurateLevel2solutionsareobtainedwhenoneemploysarigorousnonlinearprebuckling
analysis. As can be seen from the results listed in Table 115, for this particular shell the critical
bucklingloadswithnonlinearprebucklingarealwayslowerthanthecorrespondingresultsobtained
usingamembraneprebucklinganalysis.Specifically,thelocalminimumof cnl = 0.337119 atn=7is
about8%lower,whereastheabsoluteminimumof cnl = 0.328629 atn=11isabout11%lower.Notice

that the critical load Nc can be calculated easily by multiplying the lowest eigenvalue c by the
normalizingfactor Ncl = 2239.342 lb/inyielding
Nc = c Ncl = 735.913 lb/in(n=11)
In Figure 113 the critical buckling modes using membrane and rigorous nonlinear prebuckling are
depicted. Notice that the solutions with nonlinear prebuckling differ significantly from the ones
obtained using membrane prebuckling, especially at n = 11 where one observes a typical edge
bucklingtypebehavior.
197
ECSSEHB3224A
24March2010

a.Membraneprebuckling,n=7 b.Nonlinearprebuckling,n=11
cm = 0.364776 cnl = 0.328629
Figure113:BucklingmodesoftheaxiallycompressedNASAcompositeshellC3
SS3B.C. N x = N 0 , v = W = M x = 0; N cl = 2239.342 lb / in .
Table115:BucklingloadsoftheNASAlayeredcompositeshellC3
( N cl = 2239.342 lb / in )BucklingloadmapfortheperfectshellusingANILISA[24]
(B.C. N x = v = W = M x = 0 )
Prebuckling Membrane Nonlinear

n=6 m
c =0.371960 nl
c =0.347394
n=7 m
c =0.364776 nl
c =0.337119
n=8 m
c =0.378759 nl
c =0.339721
n=9 m
c =0.371493 nl
c =0.330981
n=10 m
c =0.367525 nl
c =0.329194
n=11 m
c =0.364417 nl
c =0.328629
n=12 m
c =0.364609 nl
c =0.330315
n=13 m
c =0.368600 nl
c =0.334096
n=14 m
c =0.376402 nl
c =0.339446
198
ECSSEHB3224A
24March2010
Toverifytheearlierpredictionsthefinitedifferenceversion[29]ofthewellknownshellanalysiscode
STAGS [30] will be used. Due to the slightly skewed buckling pattern predicted by the Level1 and
Level2computationsoneisforcedtomodelthewholeshell.
Initially a convergence study should be carried out in order to establish the mesh size needed for
accuratemodelingofthebucklingbehavioroftheshellinquestion. Forthispurposetheasymmetric
bifurcationfromanonlinearprebucklingpathoptionwasused,wherebytheearlierresultsobtained
withtheLevel2moduleANILISAlistedinTable115serveasareference.
Intheconvergencestudy,atfirst, forafixednumberofmeshpointsintheaxialdirection(NR=161)
the number of mesh points in the circumferential direction (NC) was increased until the bifurcation
load approached a horizontal tangent. As can be seen from Figure 114 the results converge to a
limiting value from below at about NC = 241. Next, for a fixed number of mesh points in the
circumferential direction(NC =161) the number ofrows (NR) was varied. This time convergence is
fromaboveandascanbeseenfromFigure114thehorizontaltangentisreachedataboutNR=241.
Figure114:STAGSAconvergencestudyNASAquasiisotropiccompositeshellC3
SS3B.C. N x = N 0 , v = W = M x = 0
Using a mesh of 161 rows and 161 columns (a model with 79707 D.O.F.s and a maximum semi
bandwidthof637)andSS3boundaryconditions ( N x = N o ,v = W = M x = 0) thefollowing3lowest
eigenvalueswereobtained:
(1) (1) (1)

c = 0.325524 ( n = 11) N c = c N cl = 728.960 lb / in
c
( 2) ( 2) ( 2)
= 0.326816 ( n = 12 ) N c = c N cl = 731.854 lb / in 1112
( 3)
c
( 3) ( 3)
= 0.326842 ( n = 10 ) N c = c N cl = 731.910 lb / in
Details of the critical buckling mode are displayed in Figure 115. In the circumferential scan the
circles in Figure 115(a) denote the position of the mesh points. The accepted practice is that one
199
ECSSEHB3224A
24March2010
shouldhaveatleastfivemeshpointsperhalfwaveforproperconvergence.Noticethatthesequence
of the 3 lowest buckling loads and the corresponding buckling modes agree closely with the
predictionsobtainedwiththeLevel2moduleANILISA(seealsoTable115).
a. Circumferentialtraceofthecritical(lowest)bucklingmode.
b. Axialtraceofthecritical(lowest)bucklingmode.
Figure115:BucklingmodeoftheaxiallycompressedNASAcompositeshellC3
SS3 B.C. N x = N 0 , v = W = M x = 0
200
ECSSEHB3224A
24March2010
11.5.5 Imperfection Sensitivity Study
Thatinitialimperfectionsmaydecreasetheloadcarryingcapacityofthinwalledshellstructuresisby
nowwidelyknownandaccepted.However,inordertocalculatetheeffectofinitialimperfectionsone
shouldknowtheirshapeandamplitude,aninformationthatisrarelyavailable.
Intheabsenceofinitialimperfectionmeasurements,asafirststeponeshouldestablishwhetheragiven
shellloading combination is imperfection sensitive, and if the answer is positive to estimate how
damagingcertaincharacteristicimperfectionshapesare.
11.5.6 Single Axisymmetric Imperfection

11.5.6.1 Overview
BasedonKoiterspioneeringworkontheeffectofinitialimperfections[5],[31]thesimplestimperfection
modelconsistsofasingleaxisymmetricimperfection
x
W = h1 cos i 1113
L
whereiisanintegerdenotingthenumberofhalfwavesintheaxialdirectionand 1 istheamplitudeof
theaxisymmetricimperfectionnormalizedbytheshellwallthicknessh.
If one assumes that both the axial load and the boundary conditions are independent of the
circumferentialcoordinate,thentheprebucklingsolutionwillalsobeaxisymmetric,afactthatsimplifies
thesolutionconsiderably.
11.5.6.2 Level-1 Analysis of Axisymmetric Imperfection

Neglecting the effect of the prebuckling boundary conditions the nonlinear equations governing the
prebucklingstateadmitthefollowingaxisymmetricsolutions
W
( 0) = hWv + hw0 ( x )
2 1114
F
(0) =
Eh 1
y + f0 ( x )
2
cR 2
where
x
w0 ( x ) = h1 cos i
c L
i
f0 ( x ) =
(1 + B21* i2 ) Eh3 cos i x
c 2 * 1
2 i A22 c L 1115
i
( )
2
* 2
1 2 * 1 + B
21 i
c = i D11 +
i 2 2 *
2 i A22

Noticethatthelinearizedstabilityequationsbecomenowasetofequationswithvariablecoefficients.
* * *
Thereducedwavenumber i andthenormalizedstiffnesscoefficients A 22 , B 21 and D11 arealllisted
inRef.[23].
201
ECSSEHB3224A
24March2010
IthasbeenshowninRef.[23]thatbyassuminganasymmetricbifurcationmodeoftheform
(1)
W = h sin m
x
L
cos
n
R
( y K x) 1116
where
m=k=numberofaxialhalfwaves
n= l =numberofcircumferentialfullwaves
K =Khotsskewednessparameter
aGalerkintypeapproximatesolutionyieldsfortheeigenvalues(read,bucklingloads) oftheproblem
acharacteristicequationintheformofacubicpolynomial
3
( 2
) {
mn + 2c C11 i =2m + 2mn + c + C 2 C1 1 i =2 m c
i i i
( ) } 1117
{ ( ) }
2 2
mn + C 21 i = 2 m + C3 C 4 im 1 c = 0
i
where
i =2m = 1 if i = 2m ;

=0 otherwise
im = 1 if i = m

=0 otherwise
andtheconstants C1 , C 2 ,... arelistedinAppendixCofRef.[23].
Hereitshouldberememberedthatonewillonlygetanynoticeabledegradinginfluenceoftheassumed
axisymmetricimperfectionif 1 isnegativeandifthecouplingconditioni=2missatisfied.Thephysical
explanationforthiscanbefoundinKoiters1963paper[31].Furthermore,inordertoobtainthesmallest
realrootofEq.1117,foragivenaxisymmetricimperfection 1 annsearchiscarriedout.Itshouldalso
be noticed that the terms involving the Kronecker delta i=2 m are all linear in 1 , and thus they
dominatethebucklingbehavioroftheshellwithaxisymmetricimperfection.
AssumingthatthemostlikelyaxisymmetricimperfectionofthesteelmandrelusedtolayuptheNASA
compositeshellC3isgivenby
x
W = h1 cos 2 1118
L
the Level1 DISDECO computational module AXBIF generated the solid curve shown in Figure 116.
Noticethatthecurveisrenormalizedby cm = 0.366056 ,thecriticalLevel1bucklingloadoftheperfect
shellcomputedusingAXBIF[23]withmembraneprebucklingfor K = 0.011 andn=7.Noticealsothat
an initial imperfection amplitude equal to the wall thickness of the shell (1 = 1.0) generates a
knockdownfactorof c = c / cm = 0.486 ,resultinginthefollowingratherlowbucklingload
c = 0.486 cm = 0.486(0.366056) = 0.177903

N c = c N cl = 0.177903(2239.342) = 398.386 lb/in
with K =0.011andn=7fullwavesinthecircumferentialdirection.
202
ECSSEHB3224A
24March2010
Figure116:Imperfectionsensitivityforaxisymmetricimperfectionunderaxial
compressionSS3 B.C. N x = No , v = W = M x = 0; Ncl = 2239.342 lb / in .

Sincetheexternalloading,theboundaryconditionsandtheassumedinitialimperfection(seeEq.1118)
areaxisymmetric,thereforetheprebucklingsolutionwillalsobeaxisymmetric.IthasbeenshowninRef.
[32]thatbyassuming
W
( 0) = hWv + hw0 ( x )
{ }
2 1119
F
(0) =
Eh

1
y + R f0 ( x )
2 2
cR 2
the solution of the nonlinear partial differential equations governing the prebuckling state can be
reducedtothesolutionofasinglefourthorderordinarydifferentialequationwithconstantcoefficients,
whichcanbesolvedroutinely.
Foranisotropicshellstheresultinglinearizedstabilityequationsadmitseparablesolutionsoftheform
(1) = h w
W 1 ( x ) cos n + w2 ( x ) sin n
2 1120
F
(1) =
ERh
f1 ( x ) cos n + f2 ( x ) sin n
c
y
where = .
R
Solution proceeds as outlined in Ref. [32]. Using an updated version of the Level2 computational
module ANILISA [33] and SS3 (N x = v = W = M x = 0) boundary conditions one obtains the results
presentedinTable116.Noticethatarigorousnonlinearprebucklinganalysiswasusedandannsearch
wascarriedoutforeachspecifiedaxisymmetricimperfectionamplitude 1 .
203
ECSSEHB3224A
24March2010
Table116: BucklingloadsoftheNASAlayeredcompositeshell
C3 (Ncl = 2239.342 lb / in )AxisymmetricimperfectionusingANILISA[33]
(B.C. : N x = v = W = M x = 0)
1 c c 1 c c
~0. 0.328630(n=11) 0.898 0.1 0.322995(n=7) 0.882
0.001 0.328637(n=11) 0.898 0.2 0.299860(n=7) 0.819
0.002 0.328645(n=11) 0.898 0.3 0.274795(n=7) 0.751
0.005 0.328669(n=11) 0.898 0.4 0.251290(n=7) 0.686
0.01 0.328707(n=11) 0.898 0.5 0.229985(n=7) 0.628
0.02 0.328781(n=11) 0.898 0.6 0.210880(n=7) 0.576
0.04 0.328915(n=11) 0.899 0.7 0.193829(n=7) 0.530
0.06 0.329031(n=11) 0.899 0.8 0.178656(n=7) 0.488
0.07 0.328367(n=7) 0.897 0.9 0.165188(n=7) 0.451
0.08 0.326688(n=7) 0.890 1.0 0.153264(n=7) 0.419
0.09 0.324896(n=7) 0.888
Where
c
c = 1121
0.366056
The values of Table 116 are plotted as the dashed curve in Figure 116. A comparison of the results
obtainedviatheLevel1moduleAXBIF(solidcurve)andtheLevel2moduleANILISA(dashedcurve)
shows that also in the case of axisymmetric imperfections a rigorous prebuckling analysis should be
used. Especially for very small initial imperfection amplitudes ( 1 < 0.1) the Level1 predictions are
inaccurate and overestimate the critical buckling load. Notice further that both curves have been
normalized by cm = 0.366056 , the critical Level1 buckling load of the perfect shell computed using
membrane prebucklingbyAXBIF[23]for K = 0.011 andn=7.Thiswaytheeffectofusingarigorous
prebucklinganalysisbecomeseasilydiscernible.
204
ECSSEHB3224A
24March2010

1 = 0.06; cnl = 0.329031 (n = 11) 1 = 0.07; cnl = 0.328367 (n = 7)
W x
Figure117:Criticalbucklingmodesforaxisymmetricimperfections = 1 cos 2
h L
Itisinterestingtoseethatforsmallenoughinitialimperfectionamplitudes ( 1 < 0.07 ,say)thecritical

bucklingloadoftheshellisinsensitive totheinitialimperfectionshape specifiedby Eq.1118.Notice
that the critical buckling modes have n = 11 full waves in the circumferential direction, and as can be
seeninFigure 117,asomewhatskewedbucklingmodeshapewhichis dominatedbyedgebuckling.
However,largerinitialimperfections ( 1 0.07 ,say)forcetheshelltorespondinanothermodeshape
with n = 7 full waves in the circumferential direction and with practically straight nodal lines.
Interestingly enough now the critical buckling load of the shell is sensitive to the axisymmetric initial
imperfection shape specified by Eq. 1118 and an initial imperfection amplitude equal to the wall
thickness of the shell (1 = 1.0) generates a knockdown factor of c = c / cm = 0.419 . It predicts
thusthefollowingratherlowbucklingload
c = 0.419 cm = 0.419(0.366056) = 0.153377
N c = c N cl = 0.153377( 2239.342) = 343.465 lb/in
with a very slight skewedness of the buckling pattern and n = 7 full waves in the circumferential
direction.

Recalling that since both the axial load and the boundary conditions are independent of the
circumferentialcoordinate,thereforetheprebucklingsolutionwillalsobeaxisymmetric,andonecanuse
onceagaintheasymmetricbifurcationfromanonlinearprebucklingpathoption.Bymodelingthefull
shellthecodecanchooseitselfthecriticalnumberoffullwavesinthecircumferentialdirection.Non
searchiscarriedout.Usingauniformlyspacedmeshof161rowsand161columnsandtheuserwritten
subroutineoptionWIMPtointroducethefollowingaxisymmetricimperfection
205
ECSSEHB3224A
24March2010
x
W = 1.0 h cos 2 1122
L
(rememberSTAGSdefines W positiveoutward)thefollowingcriticalbifurcationloadwasfound
N c = 332.746 lb/in
The critical buckling mode has n = 7 full waves in the circumferential direction and no visible
skewedness. The nondimensional bifurcation load of the shell with axisymmetric imperfection is for
1 = 1.0
Nc 332.746
c = = = 0.148591
N cl 2239.342
Therenormalizedbifurcationloadis
Nc
cnl = where N cl = cnl N cl
N cl
thus
nl c 0.148591
c = = = 0.456 1123
nl
c 0.325524
NoticethattheLevel3renormalizationisdoneusing cnl = 0.325524 ,thecriticalLevel3bucklingload

oftheperfectshellcomputedusingSTAGSAwithnonlinearprebucklinganda161x161mesh.
Notice also that the Level2 ANILISA prediction (cnl = 0.419, n = 7) agrees closely with the Level3
STAGSAprediction (cnl = 0.456, n = 7) .ThedifferenceispartlyduetothefactthatANILISAusesthe
Donnell type nonlinear shell equations, whereas STAGSA employs the higher order MarloweFlgge
equations.
11.5.7 Single Asymmetric Imperfection

11.5.7.1 Overview
The effect of a single asymmetric initial imperfection can be investigated either by solving the full
nonlinear response problem or by employing the well known LyapunovSchmidtKoiter [5] reduction
technique.Wheninvestigatingthedegradingeffectofasinglemodeasymmetricimperfection
x y
W = t 2 sin m cos n 1124
L R
wheremandnareintegersdenotingthenumberofaxialhalfwavesandthenumberofcircumferential
fullwaves,respectively,instabilityoccursatthelimitpointoftheprebucklingstateinthegeneralized
loaddeformation space. Assuming that the eigenvalue problem for the critical (lowest) buckling load
c willyieldauniqueasymmetricbucklingmode W (1) ,thenforanimperfectshell ( 2 0) theshape
ofthegeneralizedloaddeflectioncurveinthevicinityofthebifurcationpoint = c isgivenbythe
followingasymptoticexpansion
( c ) = c a 2 + cb 3 c 2 ( c ) 2 + O (2 ) 1125
Expressionsforthepostbucklingcoefficientsaandbandtheimperfectionformsfactorsand
arederivedinReferences[34],[35].Ifthelimitpointisclosetothebifurcationpoint,thenthemaximum
206
ECSSEHB3224A
24March2010
load s thatthestructurecancarrypriortobucklingcanbeevaluatedfromEq.1125bymaximizing
withrespectto .Forcaseswherethefirstpostbucklingcoefficientaiszero,thisanalysisyieldsthe
modifiedKoiterformula[35]
3
(1- s )3/2 = -3 2 b[1- (1- s )] 2 1126
2
where s = s / c .
Noticethat,ifthesecondpostbucklingcoefficientbispositive,Eq.1126hasnorealsolutions.Thus
thebucklingloadofthespecifiedshellloadingcombinationisnotsensitivetosmallasymmetricinitial
imperfections of the shape given by Eq. 1124. If, however, the second postbuckling coefficient b is
negative, the equilibrium load decreases following buckling and the buckling load of the real
structure s issensitivetotheasymmetricinitialimperfectionspecifiedbyEq.1124.
11.5.7.2 Level-1 Analysis of Asymmetric Imperfection

Forthecompositeshellunderinvestigation,ascanbeseenfromthepartialresultslistedinTable114,
therearemanyeigenvaluesonlyslightlyhigherthanthecriticaloneof c = 0.366056 form=1,n=7
and K = 0.011 . Hence, strictly speaking, the proposed form of the perturbation expansion given by
Eqs. 1125 is not applicable, since the nonlinear interaction between the many nearly simultaneous
eigenmodes is not accounted for. Thus the following results, where one considers the eigenfunctions
correspondingtocertaincriticaleigenvalueschosenoneatthetime,canatbestgiveanindicationasto
theseverityoftheexpectedimperfectionsensitivity.
AssuminginitiallyanasymmetricimperfectionaffinetothecriticalbucklingmodeoftheperfectNASA
compositeshellC3ascomputedbytheLevel1computationalmoduleAXBIF(seealsoTable114)
x 7
W = h 2 sin cos ( y + 0.011x ) 1127
L R
andusingtheLevel1computationalmoduleBFACT[23]tocarryouttheinitialpostbucklinganalysis
yieldsthefollowingresults
c = cm = 0.366056 (m = 1, n = 7, K = 0.011)
b=0.048874 = = 1.0
SubstitutingthesevaluesintoEq.1126,onecanplotthedegradingeffectofanasymmetricimperfection
oftheshapegivenbyEq.1127asafunctionofitsamplitude 2 .AscanbeseenfromFigure118an
initialimperfectionamplitudeequaltothewallthicknessoftheshell ( 2 = 1.0) generatesaknockdown
factorof s = s / cm = 0.541 ,resultinginthefollowingratherlowbucklingload
s = 0.541 cm = 0.541(0.366056) = 0.198036
Ns = s N cl = 0.198036 (2239.342) = 443.471 lb/in
Notice that the imperfection form factors " "and"" are identically equal to 1.0 because BFACT uses
membrane prebuckling to calculate the necessary first and second order fields and the assumed
asymmetricimperfectionshapeofEq.1127isaffine tothebucklingmode.PleasenoticethatinFigure
118thecollapseloadisrenormalizedby cm = 0.366056 ,thecriticalLevel1bucklingloadoftheperfect
shellcomputedusingmembraneprebucklingbyAXBIF[13]for K = 0.011 and n = 7 .
207
ECSSEHB3224A
24March2010
11.5.7.3 Level-2 Analysis of Asymmetric Imperfection
To investigate the effects of edgeconstraint and/or different boundary conditions on the imperfection
sensitivityofthecriticalbucklingloadoftheNASAcompositeshellC3onehastoswitchtotheLevel2
moduleANILISA[24]andrunitspostbucklinganalysisoption.Inthismodule,asdescribedearlier,the
axisymmetricprebucklingstateisrepresentedbyEqs.1110,thebucklingmodesbyEqs.1111andthe
postbucklingstateby
W
( 2) = h w ( x ) + w ( x ) cos n + w ( x ) sin n

1128
2
F
( 2) =
ERh
f ( x ) + f ( x ) cos 2 n + f ( x ) sin 2 n
c
where = y / R .DetailsofthecomputationalproceduresusedarereportedinRefs.[24],[32].
Next, let us assume that the specified asymmetric imperfection is affine to the critical buckling mode
obtainedbytherigorousLevel2perfectshellanalysisdiscussedearlier
W = h 2 w1 ( x ) cos11 + w2 ( x ) sin 11 1129
where = y / R andthecomponentfunctions w1 (x) and w 2 (x) areshowninFigure113b.Running

ANILISAwithrigorousprebucklingandSS3boundaryconditions (N x = No , v = W = M x = 0) yields
thefollowingresults
c = cnl = 0.328629 (n = 11)
b = 0.37569 ; = 0.34019 ; = 0.20039
UsingEq.1126toplotthedegradingeffectoftheasymmetricimperfectionspecifiedbyEq.1127asa
functionofitsamplitude 2 oneobtainstheresultsdisplayedinFigure119asasolidline.Thefactthat
for an imperfection shape affine to (similar to) the buckling mode with an amplitude of 2 = 1.0 one
obtainsaloadcarryingcapacityofonly snl = 0.123 appearsunrealistic.
HereoneshouldrememberthatKoitersImperfectionSensitivityTheoryisasymptoticallyexact,thatis,
ityieldsaccuratepredictionsforsufficientlysmallimperfections,wherebywhatissufficientlysmallcan
varyfromcasetocase.Also,Eq.1126wasobtainedbyusingtheperturbationexpansiongivenbyEq.
1125,wheretermsoforder () areneglected.
208
ECSSEHB3224A
24March2010
Figure118:Level1imperfectionsensitivitycalculationusingBFACT[23]
x 7
Asymptotictheory: W = h 2 sin cos (y + 0.011x).
L R
Figure119:Level2imperfectionsensitivitycalculationusingANILISA[24]
Asymptotictheory: W = h2 w1 (x) cos11 + w 2 (x) sin11 .
y y

R R
AsitcanbeseenfromthedottedcurveplottedinFigure119,byusingmoreadvancedcomputational
modulessuchasCOLLAPSE[36],whereafullnonlinearsolutionisusedandtermsuptoandincluding
2
order ( ) arekept,oneobtainsmorereasonablepredictions.
209
ECSSEHB3224A
24March2010
Noticethatuptoabout 2 = 0.05 theasymptoticpredictionsfromANILISAandthenonlinearresultsof

COLLAPSEagreeveryclosely.Thusonecansaythatinthiscasetherangeofvalidityoftheasymptotic
solutionis 0 2 0.05 .
11.5.8 Measured Initial Imperfections

11.5.8.1 Overview
Inorder toapply thetheoryofimperfectionsensitivitywith confidence, oneshouldknowthetypeof
imperfectionsthatoccurinpractice.In1969ArboczandBabcock[37]publishedtheresultsofbuckling
experimentswhere,forthefirsttime,theactualinitialimperfectionsandtheprebucklinggrowthofthe
midsurface of electroplated isotropic shells were carefully measured and recorded by means of an
automatedscanningmechanism.Thistechniquehasbeenmeanwhilegenerallyacceptedandiswidely
used.
11.5.8.2 Midsurface Initial Imperfections

AscanbeseenfromFigure1110,themeasuredinitialmidsurfaceimperfectionsoftheNASAlayered
compositeshellC3[22]showarathergeneraldistributiondominatedbyan n = 2 mode.Onecanuse
thefollowinghalfwavecosinedoubleFourierseries
W 18 x 41 y 46 ' y 75 x y
= Wi 0 cos i + W0l cos l + W0l sin l + Wk l cos k cos l
h i =1 L l=1 R l=1 R k ,l L R
1130
83 ' x y
+ Wk l cos k sin l
k ,l L R
to represent the measured initial imperfections accurately, where Fourier coefficients with absolute
valueslessthan0.001areneglected.
Aconvenientmeasureofthesizeoftheinitialimperfectionsistheirrootmeansquare(RMS)value.By
definition
1 2 R L 2
W ( x, y ) dxdy
2
rms = 1131
2 RL 0 0
Thus
2
rms

h
1
2
2 1 2
( '2
)2 2
= i Wi 0 + 4 k,l Wk l + Wk l = axi + asy 1132
FortheNASAlayeredcompositeshellC3,theRMSvaluesare
axi = 0.01018 ; asy = 0.65822
UsingthesemeasuredRMSvalueswiththepreviouslydiscussedinitialimperfectionmodelsyieldsthe
followingbucklingloadpredictions:
210
ECSSEHB3224A
24March2010
Figure1110:MeasuredinitialimperfectionsofNASAlayeredcompositeshellC3
[22].
Figure1111:Measuredloadingsurfaceimperfections(fromHilburgerandStarnes
[38]).
x
Forthelongwaveaxisymmetricimperfection W = h 1 cos 2
L
if 1 = axi = 0.01018 thenfromMANILISA[33] cnl = 0.328709 (n = 11)
N cnl = cnl N cl = 736.092 lb/in
211
ECSSEHB3224A
24March2010
x
Fortheshortwaveaxisymmetricimperfection W = h 1 cos14
L
if 1 = axi = 0.01018 thenfromMANILISA[33] cnl = 0.315141(n = 7)
N cnl = cnl N cl = 705.708 lb/in
x y
Turningnowtothelongwaveasymmetricimperfection W = h 2 sin cos n
L R
if 2 = asy = 0.65822 thenfromCOLLAPSE[39] snl = 0.141503(n = 10)
Nsnl = snl N cl = 316.874 lb/in.
Comparing these buckling load predictions with the experimental buckling load of
Nexp = 678.040 lb/in,itappearsthatboththelongwave( i = 2, 8.6%higher)andtheshortwave( i = 14,
4.1%higher)axisymmetricimperfections predictan upperbound,whereasthe longwaveasymmetric
imperfections(53.3%lower)predictsaveryconservativelowerbound.
Foramoreaccurateestimateofthebucklingloadoneshouldusethemeasuredinitialimperfectionsin
codes like STAGS [29], [30] to carry out 2dimensional nonlinear collapse analysis. Employing a user
written subroutine WIMP to input the double Fourier series of Eq. 1130, the 161x161 STAGS model
yielded with SS3 boundary condition (N x = N o , v = W = M x = 0) a collapse load of
Ns = 682.152 lb/in, whereas the same model with C4 boundary condition (u = u o , v = W = w, x = 0)
yieldedacollapseloadof Ns = 722.092 lb/in.Noticethatalsothesevaluesaresignificantlyhigherthan
theexperimentalbucklingloadof Nexp = 678.040 lb/in.
Summarizing one can state, that in this case the results of the imperfection sensitivity investigation
indicate,thatforlayeredcompositeshellsusingthetraditionalmidsurfaceinitialimperfectionsmaynot
be sufficient for a close correlation between the analytical predictions and the experimental buckling
load.
11.5.8.3 Boundary and other non-traditional imperfections

IthasbeenshowninReferences[39]and[40]thatforisotropicshellsitispossibletoachieveexcellent
correlation between the analytical predictions and the experimental results if the boundary
imperfections are included in the (numerical) analysis. A similar investigation for composite shells by
HilburgerandStarnes[38]hasshownthatthenonlinearanalysisprocedureusedinthisstudy,which
included traditional and nontraditional initial imperfections can be used successfully to determine
accuratehighfidelitydesignknockdownfactorsthatcanbeusedforpredictingcompositeshellbuckling
andcollapseloadsinthedesignprocess.Thenontraditionalimperfectionsconsistedofboundary(shell
end) geometric imperfections (see Figure 1111), shellwall thickness variations, material property
variations,localshellwallplygapsassociatedwiththefabricationprocess(seeFigure1112),variations
inloadsappliedtotheendoftheshell,andelasticboundarysupportconditions.
Further,inarecentpaperHilburger,NemethandStarnes[22]haveshownconvincinglythatinvesting
time and effort in carefully measuring the initial imperfections of a set of shells fabricated by a given
manufacturing process can be used to establish the manufacturing imperfection signature needed for
derivingasciencebasedknockdownfactorrepresentativeofthesaidprocess.
212
ECSSEHB3224A
24March2010
Figure1112:Typicallaminatemodelgeometryincludingalaminaplygap(from
HilburgerandStarnes[32]).
In Eq. 1130 the measured initial imperfection data is represented by the following halfwave cosine
Fourierseries
N1 N2 N3
i =1 k =0 l =1
(
w ( , ) = Ai cos i + Ak l cos k cos l + Bk l cos k sin l ) 1133
where
W(, ) x y
w(, ) = , = , = , 0 x L , 0 2
h L R
ForthesakeofsimplicitywerewriteEq.1133inanalternativeway,replacingthedoublesummationin
Eq.1133byasinglesummation
N1 N
w ( , ) = Ai cos i + ( Ar cos k r cos l r + Br cos k r sin l r ) 1134
i =1 r =1
wherethequantitiesindexedwithrarechosensoastoensuretheequivalenceofthetwoseriesgivenby
equations1133and1134and N = N 2 xN3 .
Themeanvalueof w(,) isgivenby
N1 N
w ( , ) = w ( , ) = Ai cos i +
i =1 r =1
{ Ar }
cos k r cos l r + Br cos k r sin l r 1135
where denotes the mean. The mean values of the Fourier coefficients can be evaluated by the
followingensembleaveragesforasampleofexperimentallymeasuredinitialimperfections
1 M (m) 1 M (m) 1 M (m)

Ai = Ai ; Ar = Ar ; Br = Br 1136
M m=1 M m =1 M m =1
Thevarianceof w(, ) isdefinedby
2
var w ( , ) = w w ( , ) =
2
{w ( , ) w ( , ) } 1137
213
ECSSEHB3224A
24March2010
Thecorrespondingstandarddeviationis,bydefinition
w ( , ) = w w ( , ) = var w ( , ) 1138
Using six test specimens C1 C6, described in detail in Ref. [22], the mean value w (, ) and the
standarddeviation w (, ) werecalculatednumerically.Theeffectofdifferentimperfectionshapeson
theresponseofthethree8plyshellsC1C3wasinvestigatedusingauserwrittensubroutineWIMPto
introducetheimperfectionsintotheSTAGS[30]computercode.
Specifically the mean imperfection shape w with C4 (u = u o , v = W = W, x = 0) boundary
conditions predicted a collapse load for the shell C3 which was 1.3% higher that the experimental
bucklingload.However,themeanimperfectionplusonestandarddeviationshape w + w predicteda
collapse load which was 6% lower than the experimental buckling load. This value represents a
satisfactory(notveryconservative)lowerboundforshellC3.
11.5.9 Discussion of the Results

The study of the buckling behavior of the NASA quasiisotropic anisotropic shell C3 was undertaken
withtheobjectiveofestablishingarationalanalysisprocedureforincorporationofinitialimperfections
intothedesignofbucklingcriticalstructures.Forashellofrevolutionunderaxiallysymmetricloading
therearethefollowingfouritemsnecessaryfortheapplicationoftheproposedprocedure:
a. Abucklinganalysiscapabilityforshellsofrevolutionsubjectedtoanaxiallysymmetricloading
condition.(AcomputercodesuchasSRA[27]orBOSOR[41]willsatisfythisrequirement.)
b. An imperfection sensitivity analysis capability for shells of revolution subjected to an axially
symmetricloadingcondition.(AcomputercodesuchasSRA[27]orSTASOR[42]willsatisfy
thisrequirement.)
c. A geometrically nonlinear analysis capability for a general shell, which can represent the
asymmetry introduced by the general asymmetric initial imperfections (The STAGS [29], [30]
computercodewillsatisfythisrequirement.)
d. Adesigninitialimperfectionormanufacturingsignature.Atthepresenttimethisisthemost
elusiveitem.
Thereareafewinitialimperfectiondatabanks[43],[44]buttheycontainmostlyinitialimperfectiondata
on laboratory scale shells. The imperfection model introduced by Imbert [45], following an idea by
DonnellandWan[46]isoneapproachtoderiveanalyticalinitialimperfectionmodelsfrommeasured
initialimperfections,butithasyettobeextendedtolargercylindricalshellstructures.
11.5.10 Conclusions
Byrelyingonaseriesoftheoreticalresultsofvariousdegreeofsophisticationpublishedintheliterature,
thehierarchicalshellbucklinganalysisapproachusedinChapter11.5hasresultedinaseriesofbuckling
load predictions of increasing accuracy. It was shown that in order to be able to arrive at a reliable
predictionofthecriticalbucklingloadofaparticularshellusedasanexampleandtomakeanestimate
ofitsimperfectionsensitivitywhichcanbeusedwithconfidence,oneshouldproceedinastepbystep
manner from simple to more complex models and solution procedures in order to understand the
responsecharacteristicsoftheshellofinterest.
Inparticular,inordertopredictthecriticalbucklingloadaccuratelyandtomakeareliableestimateofits
imperfectionsensitivity,thenonlineareffectscausedbytheedgerestraintconditionsshouldbeincluded
intheanalysis.
214
ECSSEHB3224A
24March2010
Themostapproximateoftheanalysesdescribedherein,theLevel1solutionswhichneglecttheeffects
causedbytheedgerestraints,canstillbeusedtogreatadvantagetoestablishtheapproximatebehavior
ofashellsubjectedtothespecifiedexternalloadingcondition.
If there is no information available about the expected manufacturing signature, the initial
imperfectionsthatacertainmanufacturingprocesswillproduce,onecanonlyusethetraditionallower
bound design philosophy. However, it has been shown conclusively that the tools for a reliable and
accurate simulation of the buckling behavior of axially compressed thin walled shells are available, if
thereisatleastsomeinformationavailableabouttheexpectedmanufacturingsignatureoftheshell.It
isalsoclear,themoreinformationonehasthemoreaccurateandreliablethetheoreticalbucklingload
predictions become. Thus, for a more specific prediction of the final collapse load, the final goal of a
High Fidelity Buckling Load Analysis, one should conduct a refined Level3 analysis including
measured values of the traditional shellwall imperfections and variations in the shellend or loading
surfacegeometryandtheothernontraditionalimperfectionslistedearlier.
11.6 References
[1] Anonymous,NASTRAN,TheMacNealSchwendlerCorporation,LosAngeles,California.
[2] Anonymous,ABAQUS,Hibbitt,Karlsson&Sorensen,Inc.,Providence,RhodeIsland.
[3] Zienkiewicz,O.C.,TheFiniteElementMethod,(Thethird,expandedandrevisededitionof
TheFiniteElementMethodinEngineeringScience),McGrawHillBookCompany(UK)
Limited,London,1977,p.504.
[4] Byskov,E.,SmoothPostbucklingStressesbyaModifiedFiniteElementMethod,DCAMM
ReportNo.380,TechnicalUniversityofDenmark,Lyngby,1988.
[5] Koiter,W.T.,OntheStabilityofElasticEquilibrium,Ph.D.Thesis(inDutch),THDelft,The
Netherlands,H.J.Paris,Amsterdam,1945,EnglishtranslationNASATTF10,1967,pp.1833.
[6] Budiansky,B.andHutchinson,J.W.,DynamicBucklingofImperfectionSensitiveStructures,
Proceedings11thIUTAMCongress,Munich,1964,JuliusSpringerVerlag,Berlin,1966,pp.636
651.
[7] Hoff,N.J.,BucklingofThinShells,ProceedingsofanAerospaceSymposiumofDistinguished
LecturersinHonorofTheodorevonKrmnonhis80thAnniversary,InstituteofAerospace
Sciences,NewYork,1961,pp.142.
[8] Stuhlman,C.E.,DeLuzio,A.andAlmroth,B.,InfluenceofStiffenerEccentricityandEnd
MomentonStabilityofCylindersinCompression,AIAAJournal,Vol.4,No.5,May1966,pp.
872877.
[9] Fischer,G.,berdenEinflussdergelenkingenLagerungaufdieStabilittdnnwandiger
KreiszylinderschalenunterAxiallastundInnendruck,Z.f.Flugwissenschaften,Vol.11,1963,
pp.111119.
[10] Stein,M.,TheInfluenceofPrebucklingDeformationsandStressesontheBucklingofPerfect
Cylinders,NASATR190,February1964.
[11] Arbocz,J.,Starnes,J.H.andNemeth,M.P.,AHierarchicalApproachtoBucklingLoad
Calculations,Proceedingsofthe40thAIAA/ASME/ASCE/AHS/ASCStructures,Structural
DynamicsandMaterialsConference,April1215,1999,St.Louis,MO,pp.289299.
[12] Arbocz,J.,andStarnes,J.H.,BucklingLoadCalculationsoftheIsotropicShellA8usinga
HighFidelityHierarchicalApproach,Proceedingsofthe43rdAIAA/ASME/ASCE/AHS/ASC
Structures,StructuralDynamicsandMaterialsConference,Denver,2002,AIAApaper2002
1513.
[13] Lorenz,R.,DienichtachsensymmetrischeknickungdnnwandigerHohlzylinder,Physikalische
Zeitschrift,Vol.12,No.7,p.241,April1911.
215
ECSSEHB3224A
24March2010
[14] Timoshenko,S.,EinigeStabilittsproblemederElastizittstheorie,ZeitschriftfrMathematik
undPhysik,Vol.58,No.4,p.337,June1910.
[15] Southwell,R.V.,OntheGeneralTheoryofElasticStability,PhilosophicalTransactionsofThe
RoyalSociety,London,SeriesA,Vol.213,No.A501,p.187,August1913.
[16] Almroth,B.O.,Brogan,F.A.,Miller,E.,Zele,F.andH.T.Peterson:Collapseanalysisforshellsof
generalshape,UsersManualfortheSTAGSAcomputercode,AirForceFlightDynamicsLab.,
WrightPattersonAFB,AFFDLTR718,1973.
[17] Almroth,B.O.,Brogan,F.A.andG.M.Stanley,StructuralAnalysisofGeneralShells,Vol.II,
UsersInstrucionsforSTAGSC1,LockheedReportLMSCD633873,revised1983.
[18] Keller,H.,NumericalMethodsofTwoPointBoundaryValueProblems,BlaisdellPublishing
Co.,Waltham,Mass.,1968.
[19] Wriggers,P.andSimo,J.C.,AgeneralProcedureforthedirectComputationofTurningand
BifurcationPoint,Int.J.Num.Meth.Eng.,Vol.30,pp.155176,1990.
[20] Spence,AandJepson,A.D.,Foldsinthesolutionoftwoparametersystemsandtheir
calculation,Part1,SIAMJ.Numer.Anal.,Vol.22,pp.347368,1985.
[21] Arbocz,J.andHol,J.M.A.M.,ShellStabilityAnalysisinaComputerAidedEngineering(CAE)
Environment,Proceedings34thAIAA/ASME/ASCE/AHS/ASCStructures,StructuralDynamics
andMaterialsConference,April1922,1993,LaJolla,California,pp.300314.
[22] Hilburger,M.W.,Nemeth,M.P.andStarnes,J.H.,ShellBucklingDesignCriteriaBasedon
ManufacturingImperfectionSignatures,AIAAJournal,Vol.44,No.3,March2006,pp.654663.
[23] Arbocz,J.,TheEffectofInitialImperfectionsonShellStabilityAnUpdatedReview,Report
LR695,DelftUniversityofTechnology,FacultyofAerospaceEngineering,TheNetherlands,
September1992.
[24] Arbocz,J.andHol,J.M.A.M.,KoitersStabilityTheoryinaComputerAidedEngineering
(CAE)Environment,Int.J.SolidsandStructures,Vol.26,No.9/10,pp.945973,1990.
[25] Booton,M.,BucklingofImperfectAnisotropicCylindersunderCombinedLoading,UTIAS
ReportNo.203,UniversityofToronto,1976.
[26] VonKrmn,T.andBiot,M.A.,MathematicalMethodsinEngineering,McGrawHillBook
Co.,NewYork,London,1940,p.317.
[27] Cohen,G.A.,ComputerAnalysisofAsymmetricBucklingofRingstiffenedOrthotropicShells
ofRevolution,AIAAJournal,Vol.6,No.1,January1968,pp.141149.
[28] Keller,H.,NumericalMethodsofTwoPointBoundaryValueProblems,BlaisdellPublishing
Co.,Waltham,Mass.,1968.
[29] Almroth,B.O.,Brogan,F.A.,Miller,E.,Zele,F.andPeterson,H.T.,CollapseAnalysisforShells
ofGeneralShape;II.UsersManualfortheSTAGSAComputerCode,TechnicalReport
AFFDLTR718,AirForceFlightDynamicsLaboratory,WrightPattersonAirForceBase,Ohio,
March1973.
[30] Brogan,F.A.,Rankin,C.C.andCabiness,H.D.,STagsAccessRoutinesSTARReference
Manual,ReportLMSCP032595,Version2.0,LockheedPaloAltoResearchLaboratory,Palo
Alto,California,June1994.
[31] Koiter,W.T.,TheEffectofAxisymmetricImperfectionsontheBucklingofCylindricalShells
underAxialCompression,Koninkl.Ned.Akad.Wetenschap.Proc.B66,1963,pp.265279.
[32] Arbocz,J.andHol,J.M.A.M.,ANILISAComputationalModuleforKoitersImperfection
SensitivityTheory,ReportLR582,DelftUniversityofTechnology,FacultyofAerospace
Engineering,TheNetherlands,January1989.
216
ECSSEHB3224A
24March2010
[33] Romkes,A.,StabilityandImperfectionSensitivityofAnisotropicCylindricalShellsunder
GeneralBoundaryConditions,MemorandumM809,DelftUniversityofTechnology,Faculty
ofAerospaceEngineering,TheNetherlands,July1997.
[34] Hutchinson,J.W.andAmazigo,J.C.,ImperfectionSensitivityofEccentricallyStiffened
CylindricalShells,AIAAJournal,Vol.5,No.3,March1967,pp.392401.
[35] Cohen,G.A.,EffectofaNonlinearPrebucklingStateonthePostbucklingBehaviorand
ImperfectionSensitivityofElasticStructures,AIAAJournal,Vol.6,No.8,August1968,pp.
16161619.
[36] Arbocz,J.,deVries,J.andHol,J.M.A.M.,OntheBucklingofImperfectAnisotropicShellswith
ElasticEdgeSupportsunderCombinedLoadingPartI:TheoryandNumericalAnalysis,
MemorandumM849,DelftUniversityofTechnology,FacultyofAerospaceEngineering,The
Netherlands,May1997.
[37] Arbocz,J.andBabcock,C.D.,Jr.,TheEffectofGeneralImperfectionsontheBucklingof
CylindricalShells,J.Appl.Mech.,Vol.36,1969,pp.2838.
[38] Hilburger,M.W.andStarnes,J.H.Jr.,EffectsofImperfectionsontheBucklingResponseof
CompressionLoadedCompositeShells,Proceedings41stAIAA/ASME/ASCE/AHS/ASC
Structures,StructuralDynamicsandMaterialsConference,36April2000,Atlanta,Georgia,
(paperAIAA20001382).
[39] Arbocz,J.andHol,J.M.A.M.,OnaVerifiedHighFidelityAnalysisforAxiallyCompressed
OrthotropicShells,Proceedings46thAIAA/ASME/ASCE/AHS/ASCStructures,Structural
DynamicsandMaterialsConference,1821April,2005,Austin,TX,(paperAIAA20052302).
[40] Arbocz,J.andStarnes,J.H.,Jr.,OnaVerifiedHighFidelityAnalysisforAxiallyCompressed
CylindricalShells,Proceedings45thAIAA/ASME/ASCE/AHS/ASCStructures,Structural
DynamicsandMaterialsConference,1922April2004,PalmSprings,CA(paperAIAA2004
1712).
[41] Bushnell,D.,Stress,StabilityandVibrationofComplexBranchedShellsofRevolution:
AnalysisandUsersManualforBOSOR4,NASACR2116,1972.
[42] Weustink,A.P.D.,StabilityAnalysisofAnisotropicShellsofRevolutionunderAxisymmetric
LoadwithGeneralBoundaryConditions,MemorandumM881,DelftUniversityof
Technology,FacultyofAerospaceEngineering,TheNetherlands,November2000.
[43] Arbocz,J.,ImperfectionDataBank,aMeanstoObtainRealisticBucklingLoads,in:Buckling
ofShells,ProceedingsofaStateoftheArtColloquium,E.Ramm(ed.),Stuttgart,May67,1982,
SpringerVerlag,BerlinHeidelbergNewYork,1982,pp.535567.
[44] Yaffe,R.,Singer,J.andAbramovich,H.,FurtherInitialImperfectionMeasurementsof
IntegrallyStringerStiffenedCylindricalShellsSeries2,TAEReport404,FacultyofAerospace
Engineering,Technion,IsraelInstituteofTechnology,Haifa,Israel,1981.
[45] Imbert,J.,TheEffectofImperfectionsontheBucklingofCylindricalShells,Aeronautical
EngineerThesis,CaliforniaInstituteofTechnology,1971.
[46] Donnell,L.H.andWan,C.C.,EffectofImperfectionsonBucklingofThinCylindersand
ColumnsunderAxialCompression,J.App.Mech.,Vol.17,1950,pp.7383.
217
ECSSEHB3224A
24March2010
12
Buckling experimental methods and
design verification by tests
12.1 Generalities of buckling tests

With the rapid development in computers in the last decades the question of why continue to do
experiments?hasoftenbeenaskedinmanyfieldsofappliedmechanics,[1],[2].
Asthecomputationaltoolsimprovedandexpanded,theideathatcomputersimulationscanreplace
theexperimentshasbeenvoicedmoreandmoreoften.
Here it is worth to reflect in more detail on the purpose of experiments in the computer era. To
reinforce the assumption that the computer does not replace the experiments, it is possible to
enumerateeightprimarymotives.Theseeightmotivesareapplicableforalltypesoftests(elementary
tests,structuraltestsandscale1tests).
a. To better understand the buckling and postbuckling behaviour and the primary factors
affectingit.
Inadditiontothebucklingloads,carefulexperimentsinwhichtheparametersarevariedoneat
atimeyieldthebehaviourofthestructurejustbefore,atandafterbuckling,andaccentuatethe
main parameters affecting this behaviour. Such a philosophy of research type experimental
programshasbeenimplementedinsometestprograms.Basedontheseobservedparameters
numericalschemescanbedeveloped,verified,andcanalsobeemployedforexperimentson
thecomputertoextendtherangeoftheparameterstested.
Itshouldberememberedthatcomputermethods,likeforexamplefiniteelementanalysis,can
convergetononrealisticbehaviour,unlessthephysicalphenomenaarewellunderstood,orat
leastwelldescribedbyappropriateexperimentation,topermitreliablemodelling.
b. Tofindnewphenomena.
Thisreasonisadirectextensionofthefirstone.Inbucklingandpostbucklingexperiments,the
newphenomenaarelikelytobeunexpectedbehaviourpatternsormodeinteractions.
c. Toobtainbetterinputsforcomputations.
The mathematical models employed in modern large multipurpose computer programs can
simulate real structures fairly closely for buckling, but the simulation depends very much on
theinputofcorrectboundaryconditions,inparticularjointsorbonds,onmaterialproperties,
imperfections, residual stresses and load applications. This has been emphasized by recent
experience and definitely applies also to postbuckling. Better inputs can be provided by
subsidiary tests like stubcolumn test for properties of columns or stiffeners, or multiaxial
materialtestsformorecomplicatedstructuresorloadingconditions.
Often improved inputs can be obtained from appropriate nondestructive tests: for example,
boundary conditions by vibration correlation techniques, imperfection shapesand amplitudes
by imperfection scans, load transfers and eccentricities by strain measurements and vibration
218
ECSSEHB3224A
24March2010
correlationtechniques,residualstressesbyXraytechniquesetc.Fullyautomatedrecordingin
experiments has just begun and much closer interaction between test and computation is
developing.
d. Toobtaincorrelationfactorsbetweenanalysisandtestandformaterialeffects.
Even when large powerful programs are employed, test results can still differ considerably
from predictions. These differences are partly due to inaccuracies of inputs and partly to
variations in buckling and postbuckling behaviour of the mathematical model and the
structures tested. They can all be lumped for design purposes in a correlation factor. The
advantageofsuchacorrelationfactoristheoverallcorrelationitprovidesforthedesigner,but
itsweaknessisthatitiscompletelyreliableonlyforthestructurestested.Howeveritshouldbe
noted that the correlation factor is accepted to be valid for the full batch of the product
(intendingthestructuresthathavethesamedesignandmanufacturingprocesses).
Onecanstatisticallyevaluatealargenumberofteststoobtainoveralllowerboundcorrelation
factors,amethodemployedextensivelyforshells,wheretheyarecalledknockdownfactors,
but this results in very conservative design. Other statistical evaluations are extensively
employedforcolumnsandplates,especiallyforcivilandmarineengineeringdesigncode,and
these often tend to be conservative. Hence correlation factors should be more specialized.
Since many experiments are on laboratory scaled structures, extensive studies comparing the
results of laboratory scale and large scale tests should be available in order to reassure the
experimenterandtoguidethedesigner,inparticularfordynamicloading.
Correlation type experiments will therefore continue to be a major task of research and
industriallaboratoriesforquitesometimetocome,astheyprovidethedesignerwithessential
correction factors which include the effects of new materials and manufacturing techniques
and,tosomeextent,bridgethegapbetweenthebucklingandpostbucklingbehaviourofthe
computationalmodelandtherealisticstructures.
e. Tobuildconfidenceinmultipurposecomputerprograms.
Extensiveexperimentalverificationisanessentialelementtobringforthconfidenceinalarge
computer program. This is therefore a primary motive for buckling and postbuckling
experiments,whichbecomesmoreimportant,astheprogramsbecomemoresophisticatedand
ambitious. Though some developers of programs have promoted and applied extensive
experimentalconfirmation,morecorrelationsoftheresultsobtainedfromcomputerprograms
withtestresultsarerequired.
f. Totestnovelideasofconstructionorverycomplicatedelementsofastructure.
Exploratorytestsofnewconceptshavebeenusedextensivelybyaeronautical,civil,mechanical
andoceanengineers,andwillcontinuetobeanimportanttool.Furthermore,ifthestructureis
elaborate and has many openings with complicated stiffening and load diffusion elements,
model testing may sometimes be less expensive and faster than computation with a large
multipurposeprogram,eveninthedetaileddesignstate.
g. Forbucklingincomplexproblems.
Experimentscanthereforebepreferableinareaswherecomputationiscumbersome,expensive,
anddifficulttointerpretreliably,thoughexperimentstoopresentmanydifficulties.Examples
are buckling under dynamic loading, buckling in fluidstructures interaction problems,
bucklinginthermomechanicalloading
Theoryandnumericalcomputationsshouldfollowtheseexperimentsclosely,toreinforceand
broadenthepartialunderstandingofphenomenathattheexperimentswillprovide.
219
ECSSEHB3224A
24March2010
h. Forverificationbytestsoffullscalestructures.
This is the typical industrial task, which will continue till model experiments are sufficiently
advanced and integrated with computation to eliminate the necessity for them. Here
computerizationofdataacquisitionandreductionhasmadegreatstrides,andhassignificantly
advancedtheaccuracyofmeasurementandinterpretation.
12.2 Design output

Testsshouldbepreparedandperformedwiththe objectiveofgivingdataforthedesignphase.Thus,
therealizationofthetestscanbedoneatseverallevels:testsonelementarysample,testsonstructural
elementsandtestonfullscalestructure.
The tests on elementary samples allow the determination of the buckling critical load and the first
understanding of the material behaviour under buckling. Samples are loaded up to final failure.
Resultsareusedtodeterminetheallowablesusedinthedimensioningofthefullstructure.
Thetestsonstructuralelementsallowtounderstandandtovalidatetheglobalbehaviour.Ingeneral,
theeffortsappliedtothesepanelsareelementaryloads,combinedornot,butnotrepresentativeofthe
loading combination of the full structure. These tests on structural elements can also be used for
analyses of sensitivity to various parameters (geometrical, material or boundary conditions for
instance).
The test on a complete structure is performed in the frame of the qualification phase, where the
objectiveisthevalidationofthedimensioningchoices,especiallythroughthemodelcorrelation.This
typeoftestismorecomplexandthereforemoredelicatetorealize:itneedstotakeintoaccountthe
wholeloadsappliedtotherealstructure.Theseloadsaredeterminedwithinthephaseofselectionof
dimensioningcases(seeChapter8.4).Throughthecombinationofelementaryloadcases(mechanical,
pressure,thermal)respectingthetimeconsistency,thedimensioningloadcaseforbucklingcanalso
be extracted. This selection if performed through analysis with representative adjacent structures in
ordertoensurerealisticboundaryconditions.Therefore,forthebucklingtest,itisnecessarytotake
into account the load history (i.e. loading sequence representative from the inservice life) and
representativeboundaryconditions.
Thus, for each of the level of tests, specific methods should be followed with dedicated models (if
needed),dedicatedinstrumentationanddevotedtestfacilities.
12.3 Objectives of the test

Generallyspeaking,experimentsaimnotonlytoverifytheories.Theyexplorethephysicalbehaviour
near buckling, at buckling and in the postbuckling range and they also yield empirical data upon
whichdesignguidelinescanbebased.
Thedesignofaproperexperimentisdeterminedbyitsmainpurpose.
If it aims at verifications of a theory, the experiment should be carried out under as perfect
conditionsaspossible,withspecimensmadeasaccuratelyandmeasuredascarefullyaspossible,and
fromamaterialwhosecompositioncanbeconvenientlycontrolledandmeasured,andwithboundary
conditionsthatcanbedeterminedasaccuratelyaspossibleandsimulatedadequatelyinthetheory.If
thephysicalbehaviouristobeexplored,andprimarilyinthepostbucklingrange,specimensmadeof
materials that behave elastically much beyond the buckling loads, may be preferable though their
behaviour significantly differs from that of structural materials used in practice. All these
understandingphasescanusesometestsatelementaryleveloratstructuralelementslevel.
220
ECSSEHB3224A
24March2010
Ontheotherhand,ifthedataobtainedintheexperimentistobeemployedfordesignguidelines,the
specimensshouldsimulatetherealstructure,aswellasboundaryconditionsandenvironment.Inthat
case,thetestshouldbeperformedatfullscalelevel.
Besides,itcanbethatsomeoftheseimpressivelargescaletestsarecontractual.Whentheytakeplace
thedesignandtoolingissoadvancedthatonlydesignmodificationsaretolerable.Amayorpurpose
isalsotheverificationofthestiffnessmodeland,hence,ademonstrationthatclearancerequirements
aremetandresidualdistortionsremainwithinacceptablelimits(yieldconditions).
12.4 Test plan

Table121isanattempttodescribeatypicaltestplan,indicatingwhatshouldbetakenintoaccount
for performing a buckling test. It is organized into elements depending on phases (preparatory
phase and implementation and testing phase, cf. rows of the table) as well as on the aspects
structure,measurementsandtestfacility(cf.columnsofthetable).Attheendofthetablethe
activitiesconcerningtheseaspectsaremoreandmoremerged.
221
ECSSEHB3224A
24March2010
Table121:Typicaltestplan
Objectivesofthetest
Preparatoryphase
Objectivesandconstraintsof Responsestobemeasured Controlofthetest
structuredesign Methodsofmeasurement Loadingprocedure
Loadingandboundary (sensors) Load/displacement
conditions Compatibilityofmethods introductionandlinkageofthe
Scaling Instrumentationscheme structure
Dataacquisition
Implementationandtestingphase
Design,manufactureand Designandmanufactureofthe Designandmanufactureofthe
inspectionofthestructure sensorassembly testsetup
Applicationofthe Calibrationofsensors Operatingprogram
load/displacementintroduction Softandhardwarefordata andadaptationofloading
Takingactualdata acquisition controlparameters
Applicationofstraingauges, Checkofoperatingprogram
etc. andloadingprocess
Sensorfunctiontest Mountingofthestructure
Installationofmeasurement
devices
Checkofmeasurements
Structure Measurements Testfacility
Documentationofthetestsetup
Finalsystemcheck
Testing
Dataprocessingandpreliminarytestevaluation
Documentationofthetestingparametersandofthetestresults
12.5 Test procedures

Inthetestprocedure,someimportantpointsshouldberemembered.
Theseareobviouslyonlygeneralguidelines.
a. Preparationofthestructures
Thestructuresshouldbepreparedaccordingtothetestthatisperformed.
Forexample,itcouldbenecessarythatbothendsofthespecimenaremilled,orthatbaseplanesare
weldedtothespecimenends,matchingthegeometriccentreofthespecimentothecentreofthebase
plate. The welding procedure should be such that compressive residual stresses at the flange tips
causedbytheweldingareminimized.
222
ECSSEHB3224A
24March2010
b. Initialdimensions
Detailedinitialmeasurementofthestructureandofitsgeometricimperfectionsisimportant.So
itisconvenienttoaccuratelymeasurethem.
c. Aligningthestructure
Aligning the structure within the testing machine is an important step in testing procedure,
priortoloading.
Insomecases,thestructureshouldbealignedcentrallyloaded.Inthisapproachthespecimenis
aligned under load such that the axial stresses are essentially uniform. The objective in this
alignment method is to maximize the load by minimizing the bending stresses caused by
geometricalimperfectionsofthespecimen.
Inothercases,thestructureiscarefullyalignedgeometrically,butnospecialeffortismadeto
secure a uniform stress distribution over the critical cross section. Geometric alignment is
performed with respect to a specific reference point on the cross section. The method of
geometricalignmentisrecommendedforcolumnsasitis,generally,simplerandquicker.Asa
matter of fact, in recent years the first method has practically disappeared, and geometric
alignment with exact measurement of initial outofstraightness, coupled with analytical
strengthpredictions,isgenerallyused.
d. Instrumentation
It is usually desirable to measure the more important deflections and twists to compare the
behaviourofthestructure,withtheoreticalpredictionsofbehaviour.
Theinstrumentationhaschangedmarkedlyinrecentyearsduetoprogressmadeinmeasuring
techniquesanddataacquisitionsystem,anditisnowpossibletoobtainautomaticrecordings
andplottingofthemeasurements.Suchrecordingsaremoreconvenientandmoreprecisethan
manualreadings.
The most important records needed are the applied loads and the corresponding lateral
displacements,twist,andshortening.
Strainsaremeasuredusingstraingauges.Withmodernmultichanneldataloggersmorestrain
gaugescanbereadilyusedtoobtainadditionalcheckdataandthusimprovethereliabilityof
thetest.
Theoverallshorteningandthelateraldeflectionscanberecordedbytheuseofpotentiometers
orLVDTs.
Adedicatedchaptercomesbackonthedifferentmeansofinstrumentation(seeChapter13).
e. Testing
After the specimen isaligned in the testing machine, the test is usuallystarted with an initial
load of 1/20 to 1/15 of the expected maximum load. At this load all measuring devices are
adjustedforinitialreadings.
Further load is applied slowly and the corresponding deflections are recorded instantly. The
curveareplotteduntilthemaximumloadisreached.
Oneshouldnotethatingeneralthedynamiccurveislargerthanthestaticone.Thismeansthat
aspecimencansustainaconsiderablyhigherbucklingloadiftheloadisappliedrapidly.
f. Presentationoftestdata
Theresultsofthetestaremostclearlypresentedindiagrammaticform.Forexample,themid
heightloaddeflectioncurvecanbereported.
The behaviour of the specimen under load well into the postbuckling region is determined
withtheassistanceofmeasurementsoflateraldeflectionsandstrainsatselectedpoints.These
measurementscanbecomparedtotheoreticalpredictions.
223
ECSSEHB3224A
24March2010
g. Evaluationofthetestresults
Thetestresultscanbeevaluatedbycomparingtheexperimentalloaddeflectionbehaviour,or
axialstrainbendingstrainbehaviour,andthetheoreticalprediction.
A preliminary theoretical prediction can be made on simplified assumptions of material
properties,residualstresses,etc.Thepredictioncanbeimprovediftheactualresidualstresses
andthevariationsinmaterialpropertiesareusedintheanalysis.
12.6 Test facility
12.6.1 Overview
Thetestfacilityshouldbeabletoallowtotestthestructureaccordingtotheobjectivesofthetests;e.g.
to understand basics of the buckling phenomenon, or to produce an experimental reference for the
purposeofsoftwarevalidation,ortoverifytherequiredbehaviourofareallargestructure.
12.6.2 Mounting system

Themountingsystemshouldfollowtheobjectivesofthetests.
Inthecaseofverificationtests,themountingsystemshouldreproduceasmuchaspossiblethereal
conditionsofthestructure,whereasforsoftwarevalidationteststhepredominantaspectisthatthe
boundaryconditionsarewellknown.
Inanycase,theboundaryconditionsofthespecimenshouldbewelldefined.
Thetolerancesofthemovingpartsshouldbelimitedtoafewmicrons.
Thetestrigshouldaccommodatethestructureinamannerthatdoesnotgiverisetoresidualstresses
andedgedistortions.
Much care should be taken in order to minimize the energy stored in load application devices, i.e.
reactionpoints(tobeasstiffasfeasible),whiffletrees,jacksandaccumulators.Itiscertainlywiseto
makeattachmentsasrigidaspossible.
Thesystemshouldprovideauniformclampingforcedistribution,andnosignificantstressesshould
be generated by the clamping method or device. For example, in pressure tests with air the
pressurizedvolumeshouldbeminimizedforsafetyreasonsandtoallowquickunloading.
12.6.3 Loading procedure

Theloadintroductionshouldbechosentomeetthetestobjectives.
Loadapplicationcanbedoneviastructuralhardpointsorsimulatedinterfaces.
Inthegeneralcaseofcompressedstructuresitisdesirabletohaveconstantloaddistribution,andthe
buckling test is better performed in a displacementcontrolled mode. Consequently, the load level,
whichshouldbeappliedsmoothlytothestructure,dependsonlyontheimposeddisplacementand
onthestructuralresponse.
If the buckling test requires that buckling occurs, load or displacement increments should be small
enough to allow an accurate estimation of the buckling load. Indeed, many tests are performed to
demonstratethatbucklingatverificationloadlevelisnotdetrimentalorimpairingthefunctionofthe
structure.
224
ECSSEHB3224A
24March2010
The recommendations for good practice in the case of static loading tests on structures (which
includebuckling)arelistednext:
Theloadsshouldbeappliedbyhydraulicorscrewjacks.Asystemofcontrolshouldbeprovided.
Theloadmeasuringsystemshouldpresentanoverallaccuracyandrepeatability.
Loadingjacksandloadcellsshouldbeprovidedwithsphericalbearingsorsuitablealternatives.
The reaction bearings should be designed so that the position and directions of the load reactions
relativetothespecimenremainwithindefinedlimitsduringloading.
Where additional loading or reaction points are needed to provide moment and shear in required
portions,localstiffeningisneededtoformasatisfactoryloadpathintothespecimen.
The stability of the test specimen (in the test rig) when under load (in the event of catastrophic
collapse)shouldbecarefullyconsidered.
Theloadingjacksshouldbeoperatedunderdeflectioncontrolduringtheinelasticstagesofthetest,
butloadcontrolduringthepredominantlyelasticstagesisoptional.
The load should be applied in increments interrupted with pauses for the making of measurements
andobservations.
Therateofapplicationofloadshouldbesuchthatinthemosthighlystressedpartofthespecimen,
therateofchangeofstrainshouldnotexceed300microstrainperminute.
Aftereachincrementofloading,sufficienttimeshouldbeallowedforthestructuretoreachastable
conditionbeforemeasurementsaremade.
If after any increment of loading the need for a close examination of the structure delays the
applicationofthenextincrement,thenanadditionalsetofloadandgaugemeasurementsshouldbe
made.
The entire loading history of the structure should be recorded and this should include the loads
sustainedateachincrementofloading.Thefirstapplicationsetofloadisparticularlyimportant.
When loading to collapse has been commenced and substantial plastic flow has occurred then the
programoftheincrementalloadingshouldnotbeinterrupteduntilthetestiscomplete.
Time related recordings of the model behaviour should be made at selected stations as the collapse
conditionisapproached.Thecollapseloadshouldbedefinedasthemaximumloadsustained.
12.6.4 Data acquisition

12.6.4.1 Overview
Bothhardwareandsoftwarearenecessaryfordataacquisition.
12.6.4.2 Hardware for data acquisition

Thecircuitryandrecordinginstrumentsareessentialelementsofmanymeasurementsystems.
The multichannel versions of the circuitry and recording instruments are usually called data
acquisitionsystem.
Dataacquisitionsystemcanbeclassifiedintothreegroups:
a. Basicinstrumentationsystem.
The basicinstrumentation system, forone ora fewsensors,includes the essential elementsof
one(ormore)channelsystemwithoutsophisticatedscanningdevices.
225
ECSSEHB3224A
24March2010
b. Dataloggers.
Dataloggersserveprimarilyforacquiringandrecordingthedatameasuredbytensofsensors,
though most manufacturers offer addon scanners for expanding the capacity to hundreds of
channels.
c. Computerbaseddataacquisitionsystems.
Themoderncomputerbaseddataacquisitionsystemsfeaturehighsamplingratesandalarge
channelcapacity,theycanswamptheuserwithinformation.
12.6.4.3 Software for data acquisition

Thedataacquisitionsoftwarecanbedividedintothefollowingcategories:
a. Softwareforofflinedataanalysis
b. Softwareforonlinedataanalysis
Offlineanalysesareexecutedafterthetestisfinished.Theyarebasedonadhocdevelopedtoolsor,
morefrequently,onstandalonedataanalysissoftwarepackages.Theonlyrequirementthatshouldbe
verifiedisthattestdataareallowableinaformatcompatiblewiththechosendataanalysispackages.
The online data analysis is much more critical and generally based on specifically developed tools
abletocommunicatewithallthehardwaredevicesusedduringtest.Inparticular,thesoftwareshould
guarantee,duringtestingactivity,thefullcontrolofallthefollowingparameters:
1. Applieddisplacement(force);
2. Displacement(force)rate;
3. Resultingdisplacements(forces)inalltheselectedmeasurementpoints;
4. Deformations reached in all the selected measurement points equipped with strain
gauges;
5. Pressureofhydraulicram;
6. Deformedshape(ifmeasured);
7. Loaddisplacementcurve
12.7 Documentation of the test

Eachtestshouldbedocumentedandshouldcontainatleastthefollowinginformation:
a. Structureidentification.
b. Description of the structure: dimensions, materials, manufacturing method, preparation
method,storelife,initialstateofthestructureasreportedintheNDTpretestreport.
c. Description of the testing equipment and conditions: mounting system, loading procedure,
measurement systems, temperature conditions, measurement plan explaining and describing
thechoiceofthemeasurementsystemandthelocationofthesensor.
d. Alldataacquired,before,duringandafterthetests,intheformoftablesand/orgraphs.
e. Descriptionofthebucklingmode,failuremode.
f. Place,operatoranddateoftest.
g. Anydeviationand/oranomaliesofthetest.
h. Comments,photographs,etc.ifany.
226
ECSSEHB3224A
24March2010
12.8 References
[1] SingerJ.,ArboczJ.andWellerT.BucklingExperimentsExperimentalMethodsinBucklingofThin
WalledStructures.Vol.1andVol.2.NewYork,JohnWileyandSons,2002.
[2] KobayashiA.S.HandbookonExperimentalMechanics.SEM,Bethel,CT,USA,1993.
227
ECSSEHB3224A
24March2010
13
Instrumentation
13.1 Introduction
Theinstrumentsdescribedhereunderarenotespeciallydevotedtobucklingandareusuallyusedfor
general mechanical tests (see Ref. [1] and [2]). Selected advanced measurement systems which are
appropriateforbucklingexperiments,arealsodescribedinRef.[3].
The instrumentation is used before, during and after tests (see Chapter 12 Buckling experimental
methodsanddesignverificationbytests).
13.2 Measurements
13.2.1 Classification of measurement techniques

Themeasurementtechniquesinstructuraltestingcanbeclassifiedintothreegroups:
a. Point techniques are experimental approaches in which measurements of the response of the
structurearemade,withvariousmeasurementdevicesorsensors,atoneormorelocationson
thestructure.Thebehaviourofthewholestructure,orapartofit,isthenobtained,insucha
point technique, by assembly of the different point data, their reduction, synthesis and
comparisonwithanalyticalornumericalpredictions.
b. Wholeorfullfieldtechniquesareexperimentaltechniquesthatdirectlypresenttheresponseofthe
whole structure, or a part of it, or in other words perform themselves the required synthesis.
Thesetechniques(forexampleMoir,photoelasticcoating,holography,modalanalysis)usually
employ some optical system to observe a large portion of the structure and are therefore
generallyrelatedtothebehaviourofsurfacesofthestructurethatcanbescannedbycamerasor
otheropticaldevices.
c. Nondestructive test techniques (NDT) can present both point or whole field response of the
structure. They differ, however, from the previous techniques by their emphasis on the non
destructivenatureofthetest,i.e.byensuringthatthestructurewillremainentirelyintactand
in the case of instability will not buckle or collapse. Usually these NDT methods involve the
application of some type of energy to the structure (mechanical vibrations, X rays, ultrasonic
signals, etc.) or rely on some specific response of the structure (such as acoustic or thermal
emissions)asthediagnosticindicator.
228
ECSSEHB3224A
24March2010
13.2.2 Measurements before the tests
13.2.2.1 Inspections
Thefollowinginspectionofthestructurearesuggested:
a. Geometricimperfectionsontheinternalandexternalsurfaces(bydimensionalcontrol)
b. Thicknessvariation(bydimensionalcontrol)
c. Defectsanddelaminations(byNDT)
d. Weight
Ifinstallationofthestructuralelementintherigcausesthemeasurableimperfectionstochange,then
theresultingstressfieldandgeometricimperfectionsshouldalsobeevaluated.
13.2.2.2 Pre-test activities

Beforethetests,itismandatoryto:
a. storethearticleinconstantenvironmentalconditionspriortotesting
b. castthearticlestressfreeintotherig
c. havethehydraulicsconstantlyrunning
d. shadesunradiationorotherheatsource
e. setdeformationreferencetozero
13.2.3 Measurements during the tests

It is important that during the test, the load and the displacement are measured directly on the
structuralelement,andthattheaccuracyandtheuniformityoftheloadingprocessarecontinuously
verified.
Thefollowingmeasurementsaresuggested:
a. Loads
b. Selecteddisplacements
c. Postbucklingpattern
d. Selectedstraingaugesmeasurements
Strainrecordingshouldbeappliedinanareawhichis(locally)buckled.Thisbucklingareaisassessed
throughpredictiontask.Thestraingaugesshouldbeappliedinabacktobackconfiguration.
Inaddition,frequencyresonancemonitoringcanalsobeconsidered,fordynamictestespecially.
Besides, in cyclic tests it can be observed that stringer, flanges or panels distort repeatedly. This
observation can lead to fracture mechanics considerations. Again, deflections are to be monitored
whilethereareminorchangestomonitorlocalstraincycles.Otherwise,repairschemescanbetested
andcrackgrowthassumptionsevaluated.
13.2.4 Measurements after the tests

Thefollowinginspectionofthestructurearesuggestedafterthetests:
a. Failure
229
ECSSEHB3224A
24March2010
b. Delaminations
Therehastobevisualinspection,aswellasinspectionthroughNDT(nondestructivetesttechniques).
13.3 Measurements system
13.3.1 Introduction
Usually,thefollowingmeasurementsaretaken:
a. Strains
b. Displacements
c. Force
d. Pressure
e. Temperature
13.3.2 Measurement of strain

13.3.2.1 Overview
Strainisprobablytheprimaryquantitymeasuredinexperimentalmechanicssinceitisthegeometric
quantity that defines the deformation of the loaded structural element. With the aid of the relevant
stressstrainrelation,thestrainalsoindicatesthestressactingontheelementandthereforerepresents
anindicatorofpossiblefailureatapoint.Inthecaseofbucklingthestrainindicatesitsinitiationas
wellasthegrowthofthebucklingdeformation.
Strainisusuallydeterminedbymeasurementofonedisplacementcomponentafterloading,sayu,
overasmallportionofthestructuralelementalongtheshortlinesegmentl0.Thisdisplacementuis
convertedtostrainthroughdivisionbytheoriginallengthofthesegmentl0(i.e.x=u/l0).
Strainmeasuredinthismannerisnotexact,sincethemeasurementisoversomefinitelengthl0and
not at a point, as required by the definition for strain. The error involved depends upon the strain
gradientandthelengthofthelineelementl0,whichrepresentsthegaugelength.
Sincethelengthofthelinesegment(thegaugelength)maycausesignificanterrors,muchefforthas
beenexpendedinreducingl0.Butasthesizeisreduced,mechanicaldifficultiesappearandthestrain
tobemeasuredbecomesaverysmallquantity,accompaniedbystringentaccuracyrequirements.The
positionshouldbeaccuratelydeterminedthroughpredictionanalysis.
Thetaskofthestraingaugeinmeasuringstrainisessentiallytodeterminethedisplacementbetween
two points some distance l0 apart. Strain gauges can be classified into four groups according to the
principleemployedintheirconstruction:
a. Electrical
b. Semiconductor
c. Optical
d. Fiberoptics
230
ECSSEHB3224A
24March2010
13.3.2.2 Electrical strain gauges
Since the forties, electrical bonded resistance strain gauges have practically conquered the entire
worldofstrainmeasurement.Electricalstraingaugesareessentiallyinstrumentsindicatinganystrain
inthebodytowhichtheyareattachedbyaproportionalchangeinsomeelectricalcharacteristicofthe
gauge. The electrical variables commonly employed are resistance, inductance and capacitance, but
thepiezoelectriceffecthasalsobeenwidelyusedinrecentyears.
However, the dominant type of electrical strain gauge nowadays is the electricalresistance strain
gauge(Figure131).
Figure131:Typicalmetalfoilelectricalresistancestraingauge
Atypicalmetalfoilbondedresistancestraingaugeconsistsofthestrainsensingelement,thegrid,a
thinfilmthatservesasaninsulatorandacarrierforthestrainsensinggrid,thematrixandtabs(or
terminals)forleadwireconnections.
Standard metal foil strain gauges have a normal temperature range of 75C to 175C, with many
typeshavingextendedtemperatureranges,suchas270Cto250C.Moststraingaugestodayareself
temperature compensated (STC) for use on structural materials with specific thermal expansion
coefficients.Thestrainrangeformoststraingaugesis1.5to3percent,withseveraltypesextending
to5percentforgaugesofover3mmlength.
For measurements of postyield strains, or in the deep postbuckling range, special highelongation
strain gauges are available, with up to 20 percent strains for over 3 mm length. Most bonded
resistancestraingaugeshavearesistanceof120ohms,thoughhigherresistancegaugesaresometimes
employed,primarilyintransducerapplicationsandoncompositestructure.
The sensitivity, or output produced by a resistance strain gauge, is called the gauge factor and is
definedby
GF=(R/R0)/(l/l0) 131
whereR0istheinitialunrestrainedresistanceofthegauge.Nominalgaugefactorsforcommonlyused
alloys are: for constantan alloy (in STC form) GF = 2.05; for annealed constantan (used in high
elongationgauges)GF=2.00;fornickelchromiumalloyGF=2.1;andforisoelasticalloyGF=3.2.
Because of the many possible configurations of metal foil strain gauges (see Figure 132), one can
selectanearoptimumconfigurationofasingleelementgauge,twoelementorthreeelementrosette
orothersuitablelayoutforeachspecificapplication.
231
ECSSEHB3224A
24March2010
a)singleelementgauge;b)singleelementgauge;c)singleelementgauge;d)twoelementrosettes;
e)twoelementrosette;f)twoelementstackedrosette;g)threeelementrosette;
h)threeelementrosette;i)threeelementstackedrosette;j)torquegauge
k)diaphragmgauge;l)stressgauge;m)singleelementgaugesforuseinconcrete.
Figure132:Metalfoilelectricalresistancestraingauges
Straingaugesareusuallybondedtothestructurewithspecialadhesives,butweldablestraingauges
arealsoavailableinseveralconfigurations.
There are also some special applications in which the strain gauges are unbonded, the freefilament
wire gauges, used primarily in hightemperature environments. They consist of a single filament of
wire,whichisflamesprayedorbrushpaintedontoaceramicadhesive.
13.3.2.3 Semiconductor strain gauges

Semiconductorstraingaugesrepresentadifferentcategoryofgauges,whosemaincharacteristicisa
muchhighersensitivity.Theyweredevelopedinthefiftiesandsincetheearlysixties,semiconductor
or piezoelectric strain gauges of many types have been commercially available from several
manufactures.
The backed gauges usually have phenolic or epoxy type backing and nickelplated copper ribbon
leads or printedcircuit type soldering tabs. The unbacked, or unbonded gauges, often have gold
wires,whichareseveralcentimeterslongforeaseofhandling.
In addition to their very high sensitivity, the semiconductor materials have another advantage over
metallic alloys in strain gauges. Since their resistivity is about 1000 times that of the metallic alloys
employed in strain gauges, the semiconductor strain gauges do not need grids but are very short
elements.Thesemiconductorgaugesarethereforesuitableformeasurementofextremelysmallstrains
and hence are extensively used in miniaturized or high signal output transducers. Typical
232
ECSSEHB3224A
24March2010
semiconductor strain gauges range from 1 to 13 mm in length, and their filament crosssectional
dimensionsareabout0.13x0.013mm.
13.3.2.4 Optical strain gauges

Since the beams of light are weightless, free of friction and easier to manipulate than systems of
mechanicallevers, considerable effortswere expended in the first half of the 20th century to employ
lightbeamamplificationinvariousinstrumentsandinparticularthoseformeasurementofstrain.
A wellknown example of these is the Tuckerman optical strain gauge, in which the very sensitive
mirrorcombinationofthegaugeresultsinhighaccuracy(withinafewmicrostrains).Itisoneofthe
fewmechanicalopticalgaugesalsosuitableformeasurementofdynamicstrains.
Sincetheearlyseventies,newopticalstraingaugeshavebeendevelopedusingcoherentlight(from
gas and ruby lasers). The diffraction strain gauge is a typical example of these developments. It is
comparativelyeasytoinstall,providedthestructurecanbeobservedduringthetest.Ifthebladesare
madeofthesamematerialasthetestspecimen,thegaugeisautomaticallytemperaturecompensated,
whichisanadvantageforstrainmeasurementatelevatedtemperatures.
Theinterferometricstraingaugeisanotherexampleofopticalstraingaugesexploitingthecapabilities
ofcoherentlightsources.Itmeasuresthestrainwithoutanyreinforcingeffortsorbondingproblems,
while,asinthecaseofthediffractionstraingauge,temperaturecompensationisautomatic.
13.3.2.5 Fiber-optics strain sensors

Opticalfibersensorsarearelativelynewtechnologythatwasmadepossiblebythedevelopmentof
lowloss, singlemode optical fibers in the early seventies. Since then a wide variety of optical fiber
sensorsfortemperature, magneticfields,displacement,accelerationandotherapplications,including
strain,havebeendeveloped.
The primary advantages offered by optical fiber sensors are high sensitivity, immunity to
electromagneticinterference,relativeinsensitivitytoharshenvironments,largebandwidthandhigh
datarates,geometricversatility,compatibilitywithopticalfibertelemetry,andnoJouleheating.These
advantageshavecontributedtotherapiddevelopmentofopticalfibersensorsandhavesignificantly
enhancedthedesignofopticalfibersensorsofalltypes.
13.3.2.6 Strain gauge circuits and instrumentation

Thestraingaugesystemincludes,inadditiontosensors,theappropriatecircuitry(mostlyemploying
thebasicWheatstonebridgecircuit)andrecordinginstruments.Themaincircuitcomponentsarethe
strain gauges, lead wires, connecting circuit joints (including both solder and mechanical binding
posts)andincertaincircumstancesswitchesandsliprings.Theycanalsobetheareaswhereproblems
ariseinoperation.
13.3.3 Displacement sensors

13.3.3.1 Overview
Though strain is actually determined from the measurement of a local inplane displacement, the
average displacements over larger distances are often not the quantity of primary interest. But in
buckling and postbuckling tests displacements, and in particular outofplane displacements,
characterize the wave form and amplitude of the buckling deformation, and therefore the
measurementofdisplacementisofprimeimportanceinthiscase.
233
ECSSEHB3224A
24March2010
The initial geometric imperfections, i.e. the initial outofplane displacements, strongly influence the
buckling strength of structural elements, in particular of shells. Hence, their measurement has been
adoptedasanessentialtaskinanybucklingexperiment,especiallyinthecaseofshells.
Similarly,themeasurementoftheprebucklingandpostbucklinggrowthoftheseinitialoutofplane
displacements,aswellasthedeterminationofthebucklingandpostbucklingshapesfromthesedata,
areconsideredimperativeelementsofallbucklingtests.Also,integratedinplanedisplacementslike
endshorteningarerequisitemeasurementsbecausetheycharacterizetheprebuckling,bucklingand
postbuckling behaviour as well as the actual boundary conditions of structural elements. Since
vibrationsareoftenemployedtosimulategeneralinstabilityofstructuresordeterminetheireffective
boundary conditions as in vibration correlation techniques, vibration modes are also mandatory
measurementsinmanytests.
Hence the measurement of displacements is one of the fundamental assignments in buckling
experimentsandmanytypesofprobesandrecordinginstrumentshavethereforebeendevelopedfor
thispurpose.
13.3.3.2 Potentiometers
The simplest type of displacement transducers is variableresistance, or resistive potentiometers
(Figure133).Thesimplestformofpotentiometeristheslidewireresistor.
Linearpotentiometersareavailableinseverallengthsuptoabout1mandwitharesolutiondownto
0.001 mm. Similar potentiometers for angular displacement measurements, sometimes called rotary
potentiometers,arealsocommerciallyavailableforawiderange.
Figure133:Linearandangularpotentiometers
13.3.3.3 LVDT
Generally, the most commonly used displacement transducers are Linear Variable Displacement
Transducers (LVDT) (Figure 134). They were the first electrically based displacement measurement
systemsandhavebeenextensivelyemployedsincethelatethirties.
Sensitivitiesvaryfrom0.16to2.6Volt/mmofdisplacement,thehighersensitivitiesbeingassociated
withshortstrokeLVDTsof 2mmoperatingrangeandtheloweroneswithlongstrokeLVDTsof
150500mmoperatingrange.
The LVDT removes no energy from the structural element being measured if there is no friction
betweenthecoreandshellofthetransducer,whichcanbeensuredbycarefulcorealignment.
234
ECSSEHB3224A
24March2010
Figure134:LinearVariableDisplacementTransducer(LVDT)
13.3.3.4 Other displacement sensors

CloselyrelatedtoLVDTsisafamilyofdisplacementormotionpickupscalledvariableinductance,
variablereluctanceorvariablepermeance,transducer.
The physical shape and size of a Linear Variable Reluctance Transducer (LVRT) are very similar to
thoseofanLVDT.
There are other types of displacement pickups, such as capacitance transducers, eddy current
nonconductingtransducers,piezoelectrictransducersandinductiveproximitysensors.Mostofthese
transducersarenoncontactpickups,whichmakesthemindispensableforverythinwalls.However,
some of them can cause problems that require special consideration (such as the temperature
sensitivityofthecapacitancesensorortheseeingofthestiffenersbyeddycurrentpickups).
13.3.4 Optical methods

13.3.4.1 Overview
Optical methods possess the advantage to observe relatively large portions of the structure and
directly present their total response. They are therefore primarily fullfield techniques, but they can
alsobeemployedaspointtechniques.
Theopticalmethodsusedinstructuraltestingcanbeclassifiedintofivegroups:
a. Basicmethods,includinghighspeedphotography
b. Photoelasticityandphotoelasticcoatings
c. Moirinterferometry,includingintrinsicMoir,shadowMoir,projectionMoirandreflection
Moir
d. Holographicinterferometryandspecklemethods
e. Photogrammetry
13.3.4.2 Basic optical methods

Basicopticalmethodsincludealltheusesoftheweightlesslightbeamsininstruments.
However, for buckling experiments, highspeed photography is evidently the most important basic
optical method. Two types of photographic techniques can be applied: still photography, such as
singleflashstopactionormultiflashtechniques,andmotionpicturephotography.
235
ECSSEHB3224A
24March2010
Highspeed motion picture photography is employed in the study of buckling initiation and post
bucklingbehaviourofstructuralelements.Relativelysmall,portablecameraswithpicturefrequencies
upto20000frames/secondsareavailableandwidelyusedtoday.Larger,morecomplexandexpensive
cameraswithpicturefrequenciesuptoseveralmillionframes/secarealsoattainableforspecialhigh
speedrequirements.
Brittlecoatingcanbeconsideredasanopticalmethodsinceitisafullfieldtechniquethatdependson
visualorphotographicobservation.Thismethodconsistsofapplyingabrittlecoatingonthesurface
oftheparttobetestedandanalyzingthecracksinthecoatingthatappearduetoloading.Itcanbe
used to locate areas of large strain, determine directions of principal stresses, measure stress
concentrations and indicate localized plastic yielding. In buckling tests its application is limited
primarily to relatively thickwalled structures and to plastic buckling, since in thinwalled elements
thecoatingwouldstiffenthestructureandthusindicateafictitioushigherbucklingstrength.
13.3.4.3 Photoelasticity and photoelastic coatings

Photoelasticity is an experimental technique for strain and stress analysis that utilizes the effect of
strainorstressuponlight.Photoelasticityinfersstressesfromtheiropticaleffect,calledbirefringence
or double refraction. This optical effect was observed and studied in the middle of the nineteenth
centurybutbecameanexperimentalmethodforstrainandstressonlyinthefirsthalfofthetwentieth
century,withthedevelopmentoftheappropriateresinsandplastics.
Inrecentyearsthetraditionalapplicationsofphotoelasticityhavemostlybeentakenoverbymodern
numerical techniques, but new developments in lasers, fiber optics and digital methods for pattern
recognition, as well as computerized data acquisition and analysis, have enhanced and invigorated
theuseofthistechnique.
13.3.4.4 Moir methods

TodaythewordMoirisusedtodenotetheinterferencefringescreatedbythesuperpositionoftwo
or more geometric patterns. The Moir effect occurs whenever two similar, but not quite identical,
arraysofequallyspacedlinesordotsareplacedsothatonearraycanbeviewedthroughtheother.In
experimental mechanics the Moir technique is defined as the utilization of the Moir patterns to
measuredisplacementsofsurfacesandtheirstrains(Figure135).
Figure135:Moireffectandpatterns:foralignedgridsofdifferentpitchandfor
identicalgridsmutuallyinclined
TheMoirtechniquescanbeclassifiedintofourbasicmethods:
a. IntrinsicMoirmethod,orInPlaneMoirmethod,providingdisplacementsofthepointsofa
surfacewithrespecttotheirinitialposition.TheMoirpatternsinthismethodaretheresultof
geometric interference only. They essentially magnify the distortions of the motifs (like
equispaced parallel lines) to provide a visual picture of the variable throughout the viewed
236
ECSSEHB3224A
24March2010
area. The primary purpose of this method is the determination of inplane strains and
displacement.
b. ShadowMoir.Thismethodprovidesdisplacementsofthepointsoftheobservedsurfacewith
respect to a reference surface. The purpose of the method is the measurement of outofplane
displacements.
ToobtainashadowMoirpattern,thesurfacetobemeasuredispreparedbysprayingitwitha
matt white paint. A linear reference grating, madeup of black bars and clear spaces on aflat
glassplate,isplaceddirectlyinfrontofthesurface.Acollimatedlightbeamisdirectedatan
oblique incidence through the master grating onto the test surface. The shadow of the master
grating on the test surface serves as the spacing grating. When the test surface is viewed (or
photographed) at a right angle to the surface, Moir fringes form as a result of interference
between the lines of the master and its shadow. The outofplane displacements can then
determinedfromtheMoirfringes.
Thecollimatedlightcanbereplacedbyapointlightsource,butthenthecamera(orviewpoint)
shouldbeequidistantfromthegrating.
ForbucklingandpostbucklingexperimentstheshadowMoirmethodiswidelyemployedto
outlineandmeasurethebucklepattern.
c. ProjectionMoir.ThismethodisessentiallyanextensionoftheshadowMoirtechniqueandis
sometimes considered to belong to the same group. In this method a specimen with a matt
white surface is illuminated by an array of uniformly spaced walls of light. The illuminated
specimenisphotographedtwice,firstintheunloadedconditionandthenonanotherfilmafter
the loads have been applied. After development, the two films, when superimposed, yield a
Moir pattern. There are also other variations of projection Moir, in which different gratings
areprojectedontothetestsurface.
Projection Moir, like shadow Moir, measures outofplane displacements. Projection Moir
hastheadvantageovertheshadowmethodthatnoelementoftheapparatushastobecloseto
thetestsurface,butattheexpenseofmorecomplicatedopticalsystems.
d. Reflection Moir, provides the slopes of the surface with respect to the reference state. The
methodisusedinthestudyofplatesinbendingandsimilarproblems.Theplatetobestudied
ispolishedononesidesoastoactasamirror.Aflatmastergrating,withaholeinitscenter,is
placed at a distance from the plate. A camera is aimed at the plate through this hole and a
doubleexposureismade,firstundeformedandthendistortedunderload,thesuperpositionof
thetwogratingreflectedimagesproducingMoirfringes.
In buckling studies outofplane displacements are of principal interest, and therefore only
those Moir techniques that measure them (like shadow Moir) are usually considered.
However,highsensitivityMoir,andinparticularMoirinterferometry,whichfocusesonin
plane deformations, has experienced outstanding progress in recent years. Hence, though the
new methods relate mainly to crack propagation, dynamic loading, thermal stresses and
advanced composites,experimentalists in the fieldof structuralstabilityshould watch outfor
further developments in highsensitivity Moir that have potential application to their future
tests.
These methods can be used as a qualitative measurement system or as a quantitative measurement
system.Inthefirstcase,theyneedacomplementarymethodtomonitortheoutofplanedisplacement
forbuckling.
237
ECSSEHB3224A
24March2010
13.3.4.5 Holographic interferometry and speckle methods
Holographicinterferometricmethodsrequireastableopticaltable.Howevertheintroductionoffiber
optics in both holographic and speckle optical systems in recent years has minimized the vibration
isolationrequirementstoonlythelaser,thelaunchingopticsandthephotographicplate.
Holographicinterferometrycanbeappliedtosurfacedeformationmeasurements.
There are many useful applications of holographic nondestructive evaluation, such as detection of
flawsincomposites.Timeaverageholographicinterferometryisusedformodalanalysisofvibrating
bodies.
Thefuturetrendseemstobetowardcombinationswithdigitalimageprocessing.
13.3.4.6 Photogrammetry
Thispowerfulmethodallowsthreedimensionalmeasurementofdeformationfields.Pointsorclouds
ofpointsareappliedtothesurfacethedeformationsofwhicharetobemeasured.Digitalcamerasat
different positions take photos of the undeformedand the deformed surfaces. The positions of each
pointbeforeandafterdeformationareidentified,andthedeformationsarecalculatedbytriangulation
procedures.Useofhighspeeddigitalcamerasenableshighspeedmeasurementofdeformationfields.
13.3.5 Force transducers

A variety of transducers for measuring forces and torque are commercially available. Load cells of
differentconfigurationsarewidelyusedinallstructuraltests.
In most force transducers strain gauges are used as the sensor, but sometimes LVDTs, linear
potentiometersorquartzsensorsareemployedinstead.Overallaccuracyof0.1percentareroutinely
specified for commercial load cells, but higher accuracies, of 0.02 0.05 percent, are obtained with
precisionloadcells.
13.3.6 Pressure transducers

Pressuretransducersareessentiallysimilartoforceandtorquetransducers,beingdevicesthatconvert
pressure in an electrical signal through measurement of either displacement, strain or piezoelectric
response.
Pressure can be measured in three basic ways: absolute, differential or gauge. Absolute pressure is
measured relative to very good vacuum. Differential pressure is the pressure difference across a
boundary, measured relative to the pressure on one side of that boundary. If this latter pressure is
atmospheric,thedifferentialpressureisidenticaltothegaugepressure(thatismeasuredrelativeto
thelocalatmosphere).
13.3.7 Temperature measurements

Severaltypesofsensorsareavailablefortemperaturemeasurements,suchasresistancetemperature
detectors (RTD), used when high sensitivity is required, expansion thermometers, bimetallic
thermometers and pressure thermometers (employed in control circuits and for longterm stability),
thermistors,pyrometers(usedincaseofextremelyhightemperature)andthermocouples.
Thermocouplesarethemostfrequentlyemployedsensors,sincetheyarelowcosttransducersthatare
easy to fabricate and install, but because of their low signal output the recording instruments used
needtothaveahighimpedancetominimizethecurrentinthecircuit.
238
ECSSEHB3224A
24March2010
13.3.8 Accelerometers and vibration measurements
The most important sensor for vibration, shock and absolutemotion measurement is the
accelerometer. Measurements of acceleration are required to characterize the dynamic response of
structuralelementsandassemblies.Accelerationmeasurementsprovidemostofthedatainshockand
vibrationteststhatareperformedtoverifytheintegrityofstructuralcomponents.Accelerometersare
alsothemostcommonlyusedtransducerstosensevibratorymotion.
Fivetypesofaccelerometersarecommonlyemployed:
a. Piezoelectric(crystalnaturalormanufactured)
b. Piezoresistive(strainmeasurementswithsemiconductors)
c. Servo
d. Variablecapacitance
e. Massandforcebalancesystems
13.3.9 Acoustic and thermal emission sensors

Acoustic emission is the general term applied to sounds that are generated in a material which is
under stress. In particular, it is an important tool for composite structures to detect fiber fracture,
fibermatrix debonding, matrix cracking and formation of delaminations. It should be recalled that
delaminationsincompositestructureareusuallydetectedbyultrasonictechniques.
The development in digital image processing in the eighties have promoted new methods of
experimental stress analysis by thermal emission, based on measuring very small changes in
temperaturewithascanninginfrareddetector.Onesuchcommerciallyavailablesystemistherecent
SPDT system. This system is a unique infrared differential thermography system for thermoelastic
stressanalysis (TSA)and thermal nondestructive evaluation. Bycoupling special highspeed image
processingelectronicswithahighperformanceinfraredarraydetector,thesystemiscapableofrapid
imaging of stresses. The rapidity of the acquisition of these images makes the TSA economically
attractive.
13.3.10 Non destructive testing (NDT)

13.3.10.1 Overview
Nondestructivetestinghasinrecentyearsbecomeawidelyusedtoolnotonlyfordetectingflawsand
defects, but also for characterizing and evaluating them as to type and size. This change from
detection to characterization came about because of the interest in estimating the remaining life of
components.
13.3.10.2 Penetrant inspection

Itisfast,simple,andinexpensive.Theprimaryadvantageisthatitcanbeusedonmanyshapesand
usuallydoesnotrequirealargeamountoftrainingtoimplement.Themethodisapplicableonlytothe
detection of surfaceconnected defects and, in particular, is most effective for detecting closed flaws
suchascracks,anditisusedonlyformetallicstructures.
Liquid penetrant inspection depends mainly on the ability of a liquid to be absorbed preferentially
intosmallcavitiesthatareopentothesurface.Typically,thecavitiesofinterestareexceedinglysmall,
atleastinonedimension,andarenoteasilyvisible.Theprimaryfactorsthatdeterminethepropensity
for a liquid to enter a cavity depend on the relative ratios of surface tension and the ability of the
239
ECSSEHB3224A
24March2010
liquid to wet the surface. Good penetrant fluids have small surface contact angles and high surface
tension.
Theabilityofthepenetranttowetthesurfaceisbyfarthemostimportantpropertythatdetermines
thesuccessofaparticularpenetrantsystem.Similarly,otherfactors,suchascleanliness(thepresence
of a material with poor wetting characteristics), can significantly reduce the effectiveness of a
penetrant to indicate surface flaws. For this reason, the success of penetrant inspection often clearly
dependsoncleaningandremovingforeignmatterfromthesurfacetobeinspected.
13.3.10.3 Magnetic inspection

In magnetic inspection, flaws are identified by detecting disturbed magnetic fields. The method is
applicableonlytometallicstructures,andhasevolvedintoseveralapproaches.
Traditionally, magnetic particle has been the most widely used method for magnetic inspection. In
magneticparticleinspection,thedisturbedmagneticfieldisdetectedusingsmallmagneticparticles,
whichoutlineaflaw.Themagneticfieldfluxisinducedeitherbycausingthedirectpassageofcurrent
orbyusinginducedcurrent.Theflawdetectioncapabilityisdependentontherelativeorientationof
the flaw and the direction of current flow. This limitation occurs because the field induced by the
currentmustbeperpendiculartotheflawplaneiffluxleakageistooccur.
An extension of magnetic particle inspection is called flux leakage or magnetic perturbation. The
disturbed fields produced by flaws are detected using small magnetometers such as a coil or Hall
device.Thisadaptationismorequantitativethantheclassicapproachbecauseitisimplementedusing
instrumentationandhasthecapabilityofbeingautomated.
13.3.10.4 Radiographic inspection

The primary advantage is that it can detect the existence of flaws in the volume as well as some
surface flaws. It is also very useful in composite structures, as it can detect delamination as well as
inplanefailures.
Altough several types of Xray source are used, the most common one is the Xray tube. This tube
generatesXraysbybombardingatarget,usuallytungsten,withenergizedelectrons.
The method can be expensive to implement. The costs are normally associated with the necessary
capital equipment and purchasing, processing, and storing film. Larger users of the Xray process
reducecosts.
An advantage of the radiographic method is that it produces a permanent record for reference at a
latertime.
13.3.10.5 Ultrasonic inspection

Ultrasonicinspectionisoneofthemorewidelyusednondestructivetestingmethods.Itcanbeused
totestbothmetallicstructuresandcompositestructures.
Theprimaryadvantagesofultrasonictestingaretheabilitytodeterminethedepth,ordistance,toa
flaw. Another important advantage is the ability to conduct a nonhazardous inspection. The
disadvantagesoftheultrasonicmethodarerelatedmostlytolimitedsensitivityproducedbymaterial
andphysicalpropertiessuchasgrainsizeandsurfaceroughness.
The basic equipment typically used for ultrasonic inspection consists of a scanning device and a
transducerforsendingthesound,whichincludespulsersandreceivers.
Practicalultrasonictestingmethodsareconcernedwiththewayultrasonicbeamspropagatesthroughtest
materialsandthemethodsbywhichtheultrasonicinformationisdisplayed.Thetwoprimarymethodsfor
controlling ultrasonic beams propagating through media are pulse echo and through transmission. The
pulseechomethodisasinglesidedapproachwherebythetransmittedandreceivedsignalsoccuronthe
240
ECSSEHB3224A
24March2010
same side and are produced and received by the same transducers. In throughtransmission mode, the
objecttobeinspectedislocatedbetweenthetransmitterandthereceiver.
ThecommondisplaysforultrasonicinformationaretheA,B,andCscans.TheAscanissimplyan
amplitudeversustime representation of the received information. The display can be scaled in
distancesothatflawdepthcanbereaddirectly.TheBscanisacollectionofAscansthatshowsall
flawsinacrosssection.TheCscanisaplanviewofreturnedsignalsfromagiventhicknessrange.
13.4 References
[1] SingerJ.,ArboczJ.andWellerT.BucklingExperimentsExperimentalMethodsinBucklingofThin
WalledStructures.Vol.1andVol.2.NewYork,JohnWileyandSons,2002.
[2] KobayashiA.S.HandbookonExperimentalMechanics.SEM,Bethel,CT,USA,1993.
[3] DegenhardtR.,KlingA.,KleinH.,HillgerW.,GoettingCh.,ZimmermanR.,RohwerK.,Gleiter
A.,ExperimentsonBucklingandPostbucklingofThinWalledCFRPStructuresusing
AdvancedMeasurementSystems,InternationalJournalofStructuralStabilityandDynamics,Vol.
7,no.2(2007),pp.337358
241
ECSSEHB3224A
24March2010
(Thispageisintentionallyleftblank)
242
ECSSEHB3224A
24March2010
Part 2: Structural elements with examples
243
ECSSEHB3224A
24March2010
14
Implementation of Stability Methods
14.1 Overview
Wheninvestigatingthestabilitybehaviourofastructureunderagivenloadoneisreallyconcerned
whetherthecorrespondingequilibriumconfigurationisstableorunstable.Thus,atfirst,theanalyst
usuallysolvesfortheequilibriumconfigurationandtheninvestigateswhethertheequilibriumstate
foundisstableorunstable.
Referring to the loaddisplacement curves shown in Figure 519 each point on a path represents an
equilibrium position of the structure. From the form of the curves it is obvious that the governing
equilibrium equations are nonlinear. At parts of the loaddisplacement curves the equilibrium is
stable, at other parts it is unstable. The critical load is defined as the smallest load at which the
equilibriumofthestructurefailstobestableastheloadisslowlyincreasedfromzero.
Thecriticalloadmayoccuratthelimitpointofthefundamentalequilibriumpath,thatis,atthepoint
where the load is a relative maximum. Another possibility for reaching the critical load occurs
when the primary (or fundamental) path emanating from the origin is intersected by a secondary
equilibriumpath.Atthepointofintersection,thesocalledbifurcationpointtheequilibriumequations
havemultiplesolutions,onecorrespondingtoeachbranch.
Thusthestructuralanalyst,inprinciple,alwaysdealswithtwosetsofequations,onewhichgoverns
equilibriumandthesecondwhichyieldsinformationaboutthestabilitybehaviour.Theequilibrium
equationsareoftennonlinear,whereasinmostcasesthestabilityequationsusedarelinearized.
Thefactthatforinstabilityinvestigationsonereliesonnonlinearequilibriumequationsisduetothe
conceptofinstabilityused,namelythatatthecriticalloadmorethanoneequilibriumpositionexists.
Sinceforlineartheoryofelasticitythereisauniquenessproof(thatis,forgivenloadthereisoneand
onlyonesolution),obviouslyonecannotbasethederivationofstabilityanalysisonit.
In engineering applications, the nonlinear equilibrium equations are usually derived either by
establishingtheequilibriumofforcesandmomentsonaslightlydeformedelement,orbyusingthe
stationarypotentialenergycriterion(see,forexample,[1]and[2]).Ontheotherhand,thelinearized
stability equations are obtained either by the method of adjacent equilibrium, or by the minimum
potentialenergycriterion.Inthefollowingthedifferentapproacheswillbeillustratedbyexamples.
14.2 Static versus kinetic approach

The energy criteria of equilibrium and stability, which state that a conservative system is in
equilibrium if its total potential is stationary, and the equilibrium is stable if its total potential is a
minimum,isappliedinthefollowingtofindingthecriticalloadoftheprismaticcolumnsubjectedto
compressive end loads P shown in Figure 141. Its total potential energy may be written (see, for
instance,[2])
= Um + Ub + p 14-1
where
244
ECSSEHB3224A
24March2010
L
EA 2
2 x
Um = extensional energy = dx 14-2
0
L
EI 2
2 x
Ub = bending energy = dx 14-3
0
p = potential of the applied load
L
14-4
= P[u(L) u(o)] = P u,x dx
0
and
x = extensional strain of the centroidal axis
1 2 14-5
= u,x + w,
2 x
x = curvature of the centroidal axis = w,xx 14-6
Thefactthatifthecolumnisinequilibriumitstotalpotentialenergyassumesastationaryvalueyields
thefollowingvariationalproblem
L
= (Um + Ub + p ) = F(x,u,u,x , w, w,x , w,xx )dx = 0 14-7
0
The condition that = 0 implies that the integrand (the functional F) in Eq. 147 satisfies the
correspondingEulerequationsofthecalculusofvariation,whichinthiscaseare
F d F
= 0 14-8
u dx u,x
F d F d2 F
+ = 0 14-9
w dx w,x dx 2 w,xx
Calculatingtherequiredpartialderivatives,substitutionandregroupingyields
d d
(EA x ) + P = 0 14-10
dx dx
d2 d
(EIw,xx ) (Nw,x ) = 0 14-11
dx 2 dx
Thefirstoftheseequationscanbeintegratedyielding N = EA x = constant = P .Thesecondequation

becomesthenfor EI = constant
EIw,xxxx +Pw,xx = 0 14-12
This is the equilibrium equation of an axially compressed perfectly straight column. For any w that
satisfies this equation and the specified boundary conditions at x = 0 and x = L , the total potential
energy is stationary. Whether is also a relative minimum (that is > 0) will next be
investigated.
The character of the total potential energy for a given equilibrium configuration may be
determinedbyexaminationofthechangeintotal potentialenergy correspondingtoanarbitrary
infinitesimalvirtualdisplacementofthestructurefromthegiven equilibriumposition.Intermsofa
Taylorseriesexpansionthechangeinthetotalpotentialenergyis
245
ECSSEHB3224A
24March2010
1 2 1
= + + 3 + ... 14-13
2! 3!
wherethetermsontherightarelinear,quadratic,etc.,respectively,inthesmallvirtualdisplacements.
Recallingthatthefirstordertermvanishesidenticallyforequilibriumconfigurations,hencethesign
of is governed by the sign of the second variation. For sufficiently small values of the applied
load,it canbeshownthatthesecondvariationispositivedefinite.Thecriticalloadisdefinedasthe
smallestloadforwhichthesecondvariationnolongerispositivedefinite.
Toobtaintheexpressionfor 2 oneassumesinequations141to14-6that

u = uo + u;
w = wo + w 14-14
w
where uo , w o denote the configuration whose stability is under investigation and u, are small
increments. Substituting into equation 141 and regrouping yields the following expression for the
secondvariation
L
1 2 1 2 3
= EA u, 2x +2w o,x u,
x +uo,x w, x + w2o,x w,
x w, 2x + EIw,
2xx dx 14-15
2 2 2
0
Recallingthatfortheundeflectedformofthecolumn
P
uo = x; wo = 0 14-16
EA
henceequation14-15becomes
L
2 = {EAu, x } dx
2
x +EIw, 2
xx Pw, 2
14-17
0
ThisquadraticformisseentobepositivedefiniteforsufficientlysmallvaluesoftheappliedloadP.
ThecriticalvalueofPisthesmallestloadforwhichthedefiniteintegralceasestobepositivedefinite.
The criterionfor thelimitof positivedefiniteness for a continuous system isattributed to Treffz[3].
Consideringequation14-17forasmallvalueof P, 2 > 0 forallnonzerovariations u,
w .Forlarge
values of P, 2 < 0 for some variations u,
w . As P is increased from zero, a value is reached (say,
P = Pc ) atwhich 2 isforthefirsttimezeroforatleastonevariation u,
w .Itisstillpositiveforall
w
other variations u, . Thus for P = Pc , 2 assumes a stationary value for the particular set of
w
variations u, .Then
(2 ) = 0 14-18
Hence, on the basis of Trefftz criterion, the stability equations for the critical load are given by the
Eulerequationsforthefunctionalinthesecondvariationexpression.Forafunctionaloftheformof
theintegrandinequation1417theEulerequationsaregivenbyequations148and149,where
2x +EIw,
F = EAu, 2xx Pw,
2x 14-19
Calculatingtherequiredpartialderivatives,substitutionandregroupingyields
246
ECSSEHB3224A
24March2010
u,xx = 0 14-20
xxxx +Pw,
EIw, xx = 0 14-21
These are the uncoupled stability equations of the axially compressed column. The variational
approachyieldsalsothenaturalboundaryconditionsthatareimpliedby (2 ) = 0 tohold.Thusat
x = 0,L
either = EAu,
N x=0 or u = 0 14-22
either = EIw,
M xx = 0 or w ,x = 0 14-23
either ,xxx + Pw
EIw ,x = 0 or w = 0 14-24
ThegeneralsolutionofEq.14-21is
w = C1 sinkx + C2 cos kx + C3 x + C 4 14-25
where k 2 = P / EI . For the clampedfree column of Figure 141 this solution should satisfy the
followingboundaryconditions
at x = 0 : = w,
w x =0 14-26
at x = L : w,xx = 0; xxx +k 2 w,
w, x = 0 14-27
Theseconditionsleadtofourhomogeneousalgebraicequationsinthefourconstants C1 ,..., C 4 .Fora

nontrivialsolutiontoexist(thatis,allfourconstantsarenotidenticallyequaltozero)thedeterminant
of the coefficients of the Ci ' s vanishes. Expansion of the determinant yields the following
characteristicequation
4 EI
cos kL = 0 Pn = (2n 1)2 n = 1, 2,... 14-28
4 L2
Thesmallestbucklingloadoccursfor n = 1 .Thus
2 EI
Pc = 14-29
4 L2
thevaluefoundbyEulerin1744.Thebucklingmodeis
x
w = C2 cos 1 14-30
2L
Thefactthatforthecolumnboththeequilibriumequation1412andthestabilityequation1421areof
identicalformisanexception.Theseequationsareusuallydifferent.Noticealsothattheformofthe
stabilityequationsdependsontheprebuckling(undeflected)solutionused.
Turning now to the problem depicted in Figure 141, where the compressive load P at the free end
doesnotremainfixedinitsdirectionbutfollowsthedeformationofthebodyinsomemanner,then
theworkdonebytheendloadPinreachingthefinal positionispathdependentandone isdealing
with a nonconservative force. That insuch casesastability based on the energy criteria mayfail to
yieldthecorrectanswercaneasilybedemonstratedforthepresentcase.Theboundaryconditionsat
thefreeendnoware
at x=L: ,xx = w
w ,xxx = 0 14-31
247
ECSSEHB3224A
24March2010
Figure141:Columnloadedbyafollowerforce
Applyingtheboundaryconditionsspecifiedbyequations1426and14-31tothegeneralsolutionfor
w givenbyequation1425yieldsthefollowingsystemoflinearequations
sin kL cos kL C1
=0 14-32
cos kL sin kL C2
Butherethedeterminantofthecoefficientmatrixisnotequaltozerosince
sin2 kL + cos 2 kL = 1 14-33
Hence the only admissible solution is the trivial one, namely C1 = C2 = C3 = C 4 = 0 . This would
implythatinthiscasethecolumndoesnotbuckle.Thisisobviouslyincorrect.
For nonconservative problems one always uses the kinetic approach, where one starts with the
equationsgoverningsmallfreevibrationsoftheelasticstructureatsomeleveloftheexternalloading
(treated as a fixed quantity) and then tries to find at what level of the external loading the free
vibrationsceasetobeboundedinthesmall.
If denotesthemassperunitvolumethentheequationofmotionofthecolumndepictedinFigure
141underaconstantaxialloadPis
w,xxxx +k 2 w,xx = w,tt 14-34
where
P A
k2 = ; = 14-35
EI EI
Usingseparationofvariables
w(x, t) = W(x)eit 14-36
Itisobtained
W,xxxx +k 2 W,xx 2 W = 0 14-37
anordinarydifferentialequationwithconstantcoefficients,whosesolutioncanbewrittenas
W(x) = C1 sin x + C2 cos x + C3 sinh x + C 4 cosh x 14-38
where
248
ECSSEHB3224A
24March2010
k 4 2
= 1+ 1+ 14-39
2 k4
k 4 2
= 1 + 1 + 14-40
2 k4
Applyingtheboundaryconditionsspecifiedbyequations1426and14-31oneobtainsthefollowing
characteristicequation
( 4 + 4 ) 222 cos L + cosh L + (2 2 ) sin L sinh L = 0 14-41
Butnowfromequations1439and1440
4 + 4 = k 4 + 22 14-42
22 = 2 14-43
2 2 = k 2 14-44
Substitutingintoequation1441andregroupingyields
2 + sin L sinh L + 22 (1 + cos L cos L) = 0 14-45
Where
P EI L4
= ; PE = 2 ; 2 = 2 14-46
PE L2 4
2
14-47
L = 1+ 1+ 4
2
2
14-48
L = 1 + 1 + 4
2
The transcendental equation 1445 can be solved repeatedly for the frequencies 1 and 2 by
assigningdifferentpositivevaluesto ,startingfromzero.FromtheresultsplottedinFigure142one
sees that the unloaded natural frequencies (and the corresponding eigenfunctions) of the column
changewithincreasingloading .Itisalsoclearthataslongas < 2.0316 (theloadatwhichthetwo
frequencies 1 and 2 coalesce) the motion is oscillatory and hence stable. For > 2.0316 the
frequencies become a complex conjugate pair. From equation 1436 it is evident that the negative
imaginarypartresultsinunboundedoscillationandishenceunstable.
ThusthecriticalvalueforthefollowerforcePshowninFigure141,alsocalledBecksproblemwho
wasfirsttoobtainthecorrectsolution(see[4]),is = P / PE = 2.0316 ,or
EI
Pc = 2.03162 14-49
L2
A comparison with Eulers solution for a fixed load given by equation 141 indicates that the same
columncancarryaboutaneighttimeslargerfollowerload.Formoreinformationaboutthestability
ofnonconservativestructuralconfigurationstheinterestedreadershouldconsult[5]and[6].
249
ECSSEHB3224A
24March2010
Figure142:Characteristiccurveforthecantilevercolumnloadedbyafollower
force.
14.3 Problems requiring nonlinear analysis

Althoughbucklingisanonlinearphenomenonformanyapplicationstheuseofthelinearizedstability
equations, which are amenable to analytical treatment, yield results that are suitable for design
purposes. As has been pointed out in [2], there are three situations, however, in which a nonlinear
analysisisused.
a. It often is assumed that the prebuckling deformation is rotation free and the primary
equilibriumpathsaregovernedbymembranestressstates.If,however,onewantstosatisfythe
boundary conditions from the outset, then the prebuckling deformation of cylindrical and
generalshellscontainsrotationfromthebeginningoftheloadingprocess(seeFigure172a).In
these cases the linearized stability equations have variable coefficients which are solved for
fromthenonlinearequilibriumequationsgoverningtheprebucklingstate.
b. Determinationofthebucklingloadconsistsofsolvingthelinearizedstabilityequationsforthe
critical load, at which the primary equilibrium path in the loaddisplacement plane is
intersectedbyasecondaryequilibriumpath.Ithasbeenshowninadvancedtextsonstability
(see, for instance, [7] and [8]) that equilibrium on the primary path becomes unstable at such
pointandthatstructuralbehaviourbeyondthebifurcationpointisgovernedbyconditionson
thesecondarypath.Therearecaseswherethebehaviorofthestructurecanonlybeexplainedif
the shape of the secondary path is known. Such a knowledge is needed to explain whya flat
plate develops considerable postbuckling strength, for example, but a cylindrical shell under
axialcompressionbucklesabruptlyandevenexplosively.InKoiterslinearizedtheoryforinitial
postbuckling behaviour [8] it is shown that the shape of the secondary equilibrium path near
bifurcation(seeFigure143)playsacentralroleindeterminingtheinfluenceofinitialgeometric
imperfections.Iftheinitialportionofthesecondarypathhasapositiveslope(likeforplates),
then the structure can develop considerable postbuckling strength and loss of stability of the
primary path does not result in structural collapse. However, when the initial portion of the
secondarypathhasanegativeslope(likeforcylindricalshells)theninmostcasesbucklingwill
occur violently and the magnitude of the critical load is subject to the degrading influence of
initial geometric imperfections. Unfortunately, the information given by Koiters theory is
limitedtotheimmediateneighbourhoodofthebifurcationpoint.Thusanonlinearsolutionis
carriedoutiftheshapeofthesecondaryequilibriumpathinthemoreadvancedpostbuckling
regionisneeded.
c. Finally, in the most general case, when both geometric and/or material nonlinearities are
includedintheanalysis,lossofstabilityoccursatalimitpointratherthanatabifurcationpoint.
250
ECSSEHB3224A
24March2010
In such cases the critical load is determined through solution of the nonlinear equations of
equilibrium.
Figure143:Bifurcationpointandlimitpointvianonlinearanalysis
14.4 Approximate Solutions of Bifurcation Problems
14.4.1 Overview
Inmostofthecasesconsideredinthischapterthefindingofthecriticalbucklingloadwillbereduced
tothesolutionofalinearizedeigenvalueproblem.Ingeneralthedifferentialequationsinvolvedcan
bewrittenas
L(w) M(w) = 0 14-50
whereLandMarelinear,homogeneousdifferentialoperatorsoforder2pand2qrespectively,with
p > q . Any solution w satisfies equation 1450 at every point of the region R. Associated with the
differentialequationtherearepboundaryconditionsthatthefunctionwalsosatisfiesateverypoint
oftheboundaryCoftheregionR.Theboundaryconditionsareofthetype
Bi (w) = 0 i = 1, 2, ..., p 14-51
where the Bi are linear, homogeneous differential operators involving derivatives normal to the
boundaryandalongtheboundarythroughorder 2p 1 .
The eigenvalue problem consists of finding the values of the parameter , for which there are
nonvanishingfunctionswwhichsatisfythedifferentialequation1450andtheboundaryconditions
specified by equation 1451. Such parameters are called eigenvalues (say, buckling loads) and the
correspondingfunctionsarecalledeigenfunctions(say,bucklingmodes).
Unfortunately, in general, the solution of the eigenvalue problem for continuous systems is not a
straightforward matter. Exact solutions have been found only for uniform systems with relatively
simpleboundaryconditions.Inothercasesonereliesonapproximatesolutions.
Theeigenvalueproblemdefinedbyequations1450and1451issaidtobeselfadjointif,foranytwo
arbitraryadmissibleorcomparisonfunctions w1 and w 2 ,thestatements
251
ECSSEHB3224A
24March2010
w1L(w2 )dR = w2L(w1 )dR 14-52

R R
w1M(w2 )dR = w2M(w1 )dR 14-53

R R
holdtrue.Whetheraspecifiedsystemisselfadjointornotcanbeestablishedbymeansofintegration
byparts.
Further,ifforanysuchcomparisonfunctionw
wL(w)dR 0 14-54
R
theoperatorLissaidtobepositive.TheoperatorLissaidtobepositivedefiniteiftheintegraliszero
onlyifwisidenticallyzero.ThereisasimilardefinitionwithrespecttotheoperatorM.IfbothLand
M are positive definite, the eigenvalue problem is said to be positive definite, in which case all
eigenvalues 1 arepositive.Forfurtherdetailsaboutthenatureofthedifferenttypesofeigenvalue
problemstheinterestedreadermayconsult[9]and[10].
14.4.2 The Rayleigh-Ritz Method

Oneoftheapproachesthatcanbeusedtoobtainanapproximationforthecriticalbucklingloadofa
structure without having to derive and solve the linearized stability equations is the RayleighRitz
method.ItsuseisbasedontheTrefftzcriterion,whichdefinesthecriticalloadasthesmallestloadfor
whichthesecondvariationofthepotentialenergyassumesastationaryvalue.
Toapplythismethodoneassumesasolutionintheformofalinearcombinationoftrialfunctions wi ,
whichsatisfyatleastallthegeometricboundaryconditionsoftheproblem.Hence
n
wn = aiwi 14-55
i =1
wherethe wi areknown,linearlyindependentfunctionsofthespatialcoordinatesovertheregionR
andthe ai areunknowncoefficientstobedetermined.Thisassumedform wn issubstitutedintothe
second variation of the potential energy 2 of the problem. After carrying out the integrals
involved,thecoefficients ai aredeterminedsoastorendertheexpressionforthesecondvariationof
thepotentialenergy 2 stationary.Thenecessaryconditionforthistooccuristhat

(2 ) = (2 )a1 + (2 )a2 + ... + (2 )an = 0 14-56
a1 a2 an
Sincethevariations a1 , a2 , , an arearbitrarynonzeroquantitiesthisconditionissatisfiedifand
onlyif

(2 ) = 0 i = 1, 2,...,n 14-57
ai
a set of homogeneous, linear algebraic equations. The simultaneous solution of these equations
constitutesamatrixeigenvalueproblemwhichcanbesolvedeasilybystandardmethods.
14.4.3 Galerkins Method

ToobtainanapproximatesolutionofaneigenvalueproblemonecanalsoemployGalerkinsmethod,
sonamedaftertheRussiannavalengineerwhofirstproposeditin1915(see[11]).Inthismethodone
252
ECSSEHB3224A
24March2010
attemptstofindanapproximatesolutionofthegoverningdifferentialequationdirectly.Thisisdone
byassumingasolutionintheformofaseriesofcomparisonfunctions
n
wn = aiwi 14-58
i =1
where the wi are known, linearly independent functions which satisfy all the boundary conditions
andare2ptimesdifferentiable,whereasthe ai areunknowncoefficientstobedetermined.Ingeneral
theseriessolutionwillnotsatisfythedifferentialequationdefiningtheeigenvalueproblemunless,by
somecoincidence,theassumedseriessolutioniscomposedoftheeigenfunctionsoftheproblem.Thus
uponsubstitutionoftheassumedsolutioninthelinearizedeigenvalueproblem
L(w) M(w) = 0 14-59
anerror n willbeobtainedsothat
n = L(wn ) (n)M(wn ) 14-60
where (n) is the corresponding estimate of the eigenvalue . At this point one requires that the
weighted error n integrated over the region R be zero. As weighting functions one uses the n
comparisonfunctions wi .Theseconditionscanbewrittenasfollows

n a j (wn )dR = 0 j = 1, 2, ..., n 14-61
R
Considernow
n n

a j (wn ) L (wn ) dR = ai w jL(wi )dR = K iai j = 1, 2, ..., n 14-62
R i =1 R i =1
wherethecoefficients Kij aregivenby
K ij = K ji = w j L(wi )dR 14-63

R
andaresymmetriciftheoperatorLisselfadjoint.Similarlyonecanwrite
n n

a j (wn ) M(wn ) dR = ai w jM(wi )dR = Mijai j = 1, 2, ..., n 14-64
R i =1 R i =1
wherethecoefficients Mij aregivenby
Mij = M ji = w jM(wi )dR 1 4-65

R
andaresymmetriciftheoperatorMisselfadjoint.
With equations 14-61 through 14-65 one can reduce the solution of the original continuous
eigenvalueproblemspecifiedbyequation14-59tothefollowingsystemofnsimultaneousequations
N
(K ij (n)Mij ) ai = 0 j = 1, 2, ..., n 14-66
i =1
which are known as Galerkins equations. They represent a matrix eigenvalue problem for an n
degreeoffreedomsystemwhichcanbesolvedeasilybystandardtechniques.
253
ECSSEHB3224A
24March2010
14.4.4 Finite Element Formulation of Bifurcation Problems
A discussion of the computational tools available for bifurcation problems would be incomplete
withoutmentioningtheverypopularfiniteelementdisplacementmethod.Inthefollowingtheenergy
criteriaofequilibriumandstabilitywillbediscussedintheformproposedbyZienkiewicz[12].
If a conservative system is described by n generalized coordinates qi , i = 1, 2, ..., n , then to total
potentialenergy ofthesystemcanbewritten
= (q1 , q2 , ..., qn ) 14-67
Equilibriumissatisfiedif
= 0 14-68
whichimpliesthefollowingsetofnnonlinearalgebraicequations;

=0 i = 1, 2, ..., n 14-69
qi
Theequilibriumconfigurationisstableifthetotalpotentialenergyisarelativeminimum,i.e.
2
2 = q q > 0 i, j, = 1, 2, ..., n 14-70
qiq j i j
where repeated indices indicate summation. Notice that in this case the associated positive definite
matrix Vij = 2 / qiq j has all positive eigenvalues (r) . If the matrix Vij evaluated at an
equilibrium point has any negative eigenvalues then the total potential energy function attains
local maxima in the directions of the corresponding eigenvectors and the system is in a state of
unstableequilibrium.
The transition from stable to unstable equilibrium occurs when at least one eigenvalue, say (1) ,
becomes zero. The matrix Vij is then singular and the corresponding point on equilibrium path is
calledasingular(orcritical)point.
Singularpointsindicateeitherthatthereisabifurcationoftheequilibriumpathintoother,stableor
unstable branches or that a limit point has been reached. It is therefore important to detect and
calculatesingularpointsinadditiontostablepointsonanequilibriumpath.
Turning now to the finite element formulation, let the displacements at any point within an elastic
bodybedefinedasacolumnvector {u} ,then
{u} = [H] {q} 14-71
where the components of [H], the shape functions, are so chosen as to give the appropriate nodal
displacementswhenthecoordinatesofthecorrespondingnodesareinsertedand{q}containsallthe
nodaldisplacements.Noticethatthisandthefollowingexpressionsaretobeinterpretedasapplying
tothewholestructureunderconsideration.
With the displacements at all points within the body known one can proceed to calculate the
generalizedstrains(extensionalstrainsandcurvatures),whichcanbewritteninmatrixnotationas
{} = [B] {q} = ([Bo ] + [BL ] ) {q} 14-72
where [B o ] isthematrixobtainedfromthelinearinfinitesimalstrainanalysisand [BL ] containsthe

contributions of the nonlinear strain components. Notice that [B o ] is independent of {q} whereas
[BL ] isusuallyalinearfunctionof{q}(see[12],p.414).
254
ECSSEHB3224A
24March2010
Next,assuminggenerallinearelasticbehavior,therelationshipbetweenstressesandstrainswillbeof
theform
[] = [C] {} 14-73
where [C] istheelasticitymatrixcontainingtheappropriatematerialproperties.
NextfollowingthedetailsasoutlinedbyZienkiewicz[12]thesecondvariationofthetotalpotential
energycanbewritteninthefollowingquadraticform
2 = {q} T [K T ] {q} > 0 14-74
where [K T ] = [K ] + [K o ] + [K L ] isthetangentstiffnessmatrixand
[K] = symmetric matrix dependent on the stress level 14-75
Thismatrixisknownastheinitialstressorgeometricmatrix.
[K o ] = [Bo ]T [C][Bo ] dV 14-76

v
[K L ] = [Bo ]T [C][BL ] + [BL ]T [C][Bo ] + [BL ]T [C][BL ]dV 14-77

v
Noticethat [K o ] representstheusualsmalldisplacementsstiffnessmatrix,whereasthematrix [K L ]
is due to the large displacements and is variously known as the initial displacement or large
displacementmatrix.
Thus,whenthefiniteelementdiscretizationisemployed,Eq.1474representsthestabilitycriterionof
anequilibriumconfiguration.Fromthetheoryofquadraticformsoneknowsthatastableequilibrium
configurationisensuredifthetangentstiffnessmatrix
[K T ] = [K ] + [K o ] + [K L ] 14-78
hasnonegativeeigenvalues.Acriticalpointisreachedwhen [K T ] hasatleastone zeroeigenvalue.

Thus the stability of an equilibrium configuration can be determined by solving the eigenvalue
problem
[K T ] {X (r) } = (r) { X (r) } 14-79
at the current equilibrium state, where (r) is the eigenvalue and {X (r) } is the corresponding
eigenvector.
Notice that the computation of the critical point is done in two steps. First, the equilibrium
configurationassociatedwithagivenloadlevelPiscomputed.Next,thestabilityofthisconfiguration
is examinedby calculating the eigenvalues of [K T (P)] , the tangent stiffness matrix evaluated at the
loadP.
Thismethodofdeterminingthestabilityofaconservativesystemisveryaccurate,howeveritcanbe
computationallyexpensivebecauseitinvolvesthesolutionofaquadraticeigenvalueproblemforthe
criticalload.Cheapermethodsofestimatingthecriticalloadareavailable.Thesemethodsareusually
referred to as linearized buckling analyses, where the critical load is calculated based on a linear
extrapolationofthebehaviorofthestructureatasmallloadlevel.
14.5 Computational Tools for Bifurcation Problems

The majority of stability problems that arise at present in practical structural applications cannot be
solved analytically. It might be possible, that after a number of simplifying assumptions have been
255
ECSSEHB3224A
24March2010
introducedoneisabletoobtainanapproximatesolutionviatheRayleighRitzorGalerkinsmethods
discussed earlier. However, in nontrivial applications these methods may require considerable
analyticalandcomputationaleffort.
Thusthepointissoonreachedwhereonelookstowardsthesupposedlyeasierapproachofferedby
todays general purpose computer codes. A word of caution is appropriate here. One should not
expect that complicated structural stability problems involving thinwalled plate and shell
components, where nonlinear effects play an important role, can be solved routinely without much
effortandthoughtbyanyofthemanycodesthatarecurrentlyavailable.Athoroughunderstanding
of the shell and stability theory involved supplemented by a good working knowledge of the
computational algorithms used are the prerequisites for the analyst to be able to arrive at the
appropriate solutions. Otherwise the chances are high that incorrect or unreliable solutions will be
obtained.
Therearemanycommerciallyavailablefiniteelementcodeswithbucklinganalysiscapabilitiessuch
asNASTRAN[13],ADINA[14],MARC[15],ANSYS[16],andABAQUS[17],justtonameafew.A
comprehensivereviewoftheseandothercodesisobviouslybeyondthescopeofthisbook.Interested
readersshouldconsult[18]and[19]forfurtherinformation.
14.6 References
[1] Brush,D.O.andAlmroth,B.O.,BucklingofBars,PlatesandShells,McGrawHill,NewYork,
1975.
[2] Yamaki,N.,ElasticStabilityofCircularCylindricalShells,ElsevierSciencePublishersB.V.,
Amsterdam,1984.
[3] Trefftz,E.,ZurTheoriederStabilittdeselastischenGleichgewichts,ZAMM,Vol.13,1933,
pp.160165.
[4] Beck,M.,DieKnicklastdeseinseitigeingespannten,tangentialgedrcktenStabes,ZAMP,Vol.
3,1952,pp.225228.
[5] Ziegler,H.,PrinciplesofStructuralStability,BlaisdellPublishingCo.,Waltham,
Massachusetts,1968.
[6] Bolotin,V.V.,NonconservativeProblemsoftheTheoryofElasticStability,APergamonPress
Book,TheMcMillanCo.,NewYork,1963.
[7] Thompson,J.M.T.andHunt,G.W.,AGeneralTheoryofElasticStability,JohnWiley&Sons,
London,1973.
[8] Koiter,W.T.,OntheStabilityofElasticEquilibrium,(inDutch),Ph.D.thesis,DelftUniversity
ofTechnology,1945,Engl.Translation:NASATTF10,1967,883p.andAFFDLTR7025,1970.
[9] Langhaar,H.L.,EnergyMethodsinAppliedMechanics,JohnWiley&Sons,NewYork,1962.
[10] Meirovitch,L.,AnalyticalMethodsinVibrations,TheMacMillanCo.,London,1967.
[11] Galerkin,B.G.,BeamsandPlates,(inRussian),VestnikInzhenerov,1,(19),1915,pp.897908.
[12] Zienkiewicz,O.C.,TheFiniteElementMethodinEngineeringScience,2ndEdition,McGraw
Hill,London,1971.
[13] NASTRAN,TheMacNealSchwendlerCorporation,815ColoradoBlvd,LosAngeles,California,
90041,USA.
[14] ADINA,ADINAR&D,Inc,71EltonAve.,Watertown,Massachusetts,02172,USA.
[15] MARC,MARCAnalysisResearchCorporation,260SheridanAve.,PaloAlto,California,94306,
USA.
256
ECSSEHB3224A
24March2010
[16] ANSYS,SwansonAnalysisSystems,Inc.,P.O.Box65,JohnsonRoad,Houston,Pennsylvania,
15342,USA.
[17] ABAQUS,Hibbitt,Karlsson&Sorensen,Inc.,100MedwayStr.Providence,RhodesIsland,
02906,USA.
[18] Pilkey,W.,Saczalski,K.andSchaeffer,H.,eds.,StructuralMechanicsComputerPrograms,
Surveys,AssessmentsandAvailability,UniversityofVirginiaPress,Charlottesvill,VA.,1974.
[19] Noor,A.K.,Belytschko,T.andSimo,J.C.,eds.,AnalyticalandComputationalModelsof
Shells,CEDASME,3,1989.
257
ECSSEHB3224A
24March2010
15
Columns, Beams, Arches and Rings
15.1 Introduction
Some Aspects of column buckling are already outlined in Chapter 5.2. In this chapter the buckling
load of straight bars subjected to central axial load (column) and bending (beam) is considered in
greaterdetail.Thecompletedifferentialequationforbothinternalaction(beamcolumn)isderived.
Inplanebucklingofcurvedelementssuchasarchesandringsandtheparametricinstabilityofbarsis
alsoillustrated.
15.2 Columns
15.2.1 Overview
Failure of columns under axial compression is generally due to global instability or local instability
(seeFigure151).
Globalinstabilitymaybeelasticorinelastic.Itisrelatedtobucklingmodeswherethecrosssections
aretranslated(flexuralinstability),rotated(torsionalinstability)orboth(combinedflexuraltorsional
instability)butnotdistortedintheirownplanes.
Localinstabilityorcompressioncripplingisdefinedasaninelasticdistortion,withouttranslationor
rotation, of the cross sections. The average stress on the section at the failure load is called the
crippling stress. Compression crippling involves elastic or inelastic buckling of portions of the
columnsectionalongwithinelasticaxialcompressionofstifferpartssuchascornersandintersections
ofsectioncomposedofstraightthinelements(e.g.CSection,Tsection).
However,thedistinctionbetweenthetwogeneralmechanismislargelytheoreticalandoftenglobal
andlocalinstabilityoccursimultaneously.Intherangeoflowandintermediateslendernessratiosthe
interplayamongdifferentbucklingmodesishighlydependentuponthecolumncrosssectionaswell
asmaterialproperties.Figure152illustrates,fordifferentpossiblebucklingmechanisms,thevalueof
criticalstress cr versustheslendernessratio L '/ , L ' beingtheeffectivecolumnlengthand the
radius of gyration with respect to the centroidal axis about which buckling takes place (for cases
wherethebucklingmodeisnotevidenttheminimumradiusofgyrationshouldbeused).
258
ECSSEHB3224A
24March2010
Figure151:Modesofcolumnfailure
Figure152:Criticalbucklingstresscurvesforseveraltypesofcolumnfailure
Theeffectivelength L ' isdefinedasthetheoreticallengthofasimplysupportedcolumnthatwould

buckleatthesamecriticalloadastheactualcolumnandisrelatedtothecolumnboundaryconditions.
Intherangeoflongcolumns,bucklingoccursaspredictedbyEuler.Iftheslendernessratioisthatofa
short or intermediate column different buckling mechanism can describe the failure phenomenon,
namely:
a. Inelasticflexuralinstability(EulerEngessercurve)
b. Compressioncrippling
c. Combinedflexuralandlocalinstability(JohnsonEulerparabola)
d. TorsionalInstability
e. Combinedtorsionalflexuralinstability
259
ECSSEHB3224A
24March2010
15.2.2 Long Column (elastic buckling)
A column with large slenderness ratio fails through lack of stiffness due to flexural instability.
Accordingly to classical theory due to Euler, the column remains straight as it is loaded in axial
compressionuntilthecriticalload, PcE ,atwhichbucklingoccursisreached.Thecriticalloadisgiven
bytheEulerformula
2 EI
PcE = 151
L2
TheEulerbucklingloadshouldbeinterpretedasanupperlimittothestrengthofthecolumnsince
otherbucklingmechanismcouldariseatlowerloads.Itreferstoaninitiallyperfectuniformstraight
columnwithnoeccentricityintheaxialload.
The critical stress is the stress corresponding to the average stress over the cross section at the
bucklingload
PcE 2 E
cE = =
A L 2 152

Theeffectivecolumnlengthcanbeexpressedas
L
L = 153
c
Where L the actual column length and c the fixity coefficient dependent on boundary conditions.
FixitycoefficientsforpossibleperfectendrestraintsarereportedinTable151.Figure153showsthe
relatedbucklingmodes.
Table151:Fixitycoefficientsfordifferentendrestraint
Boundaryconditions c L
1
Bothendssimplysupported L
1/ 4
Oneendfixed,theotherfree 2L
4
Bothendsfixed L/2
Oneendfixed,theothersimply (4.493/)2 0.699L
supported
260
ECSSEHB3224A
24March2010
Figure153:Bucklingmodesfordifferentendrestraints
Eq.152alsoappliestomorecomplexcasesinvolvingelasticrestrains,intermediatesupports,stepped
ortaperedcolumns,distributedaxialload.Foreachsettingthepertinentfixitycoefficientshouldbe
adopted. The fixity coefficient for various elastically restrained columns is given in Figure 154 and
Figure155.,whereasFigure156referstothecaseofasteppedcolumn.Valuesofthefixitycoefficient
forseveralothercasesareavailablein[1].
Figure154:Fixitycoefficientforasimplysupportedcolumnwithelasticrotational
restraints
261
ECSSEHB3224A
24March2010
Figure155:Fixitycoefficientforasimplysupportedcolumnwithanintermediate
elasticsupport
Figure156:Fixitycoefficientforasimplysupportedsteppedcolumn
15.2.3 Inelastic flexural instability

The assumption of elastic material at stability loss is only valid for slender columns. For shorter
columns, buckling will occur at a stress level above the proportional limit p of the stressstrain
262
ECSSEHB3224A
24March2010
relationship. Indeed, for a column that is not very slender, it is possible to exceed the proportional
limitatsomepointsofthecrosssectionbeforethecolumnbuckles.Thistypeofbucklingisreferredto
asinelasticbuckling.ManyattemptshavebeenmadetoobtainasuitablemodificationofEq.152for
inelasticbuckling[2].Itisnowwidelyacceptedthatarationalmodificationisgivenbythesocalled
tangentmodulusformulaproposedbyEngesserin1889:
Pct 2 Et
ct = =
A L 2 154

where the tangent modulus Et is the slope of the stressstrain curve at the stress ct as shown in
Figure157.
Figure157:Idealizedstressstraincurve
The tangent modulus theory was initially criticized by some authors and subsequently apparently
improvedandcorrectedbyEngesser(1895)whoproposedthereducedmodulustheoryordoublemodulus
theory.Thedifferencebetweenthetwocompetingformulationconsistsofthevalueoftheaxialforceat
theonsetofbucklingwhenthecolumnstartstobend.Tangentmodulustheoryassumesthatthereis
an increase of the axial force P along with the bending moment M , resulting in an overall
increaseinaxialstrainacrossthesectionandassuringnostrainreversalanywhereinthecrosssection
(see Figure 158a). On the contrary, in the reduced modulus theory the axial force is assumed to
remainconstantduringbuckling;consequently,thebendingdeformationatbucklingwillcausestrain
reversalontheconvexsideofthecolumn(seeFigure158b).Sincetwoelasticmoduli, E and Et are
necessary to define the momentcurvature relationship of the cross section, the namedouble modulus
theorywasused.Theexpressionfor ct accordingtoreducedmodulustheoryisstillgivenbyEq.154
withtheonlydifferencethatthereducedmodulus Er ,functionof E and Et ,shouldbeintroducedin
placeofthetangentmodulus Et .Unlikethetangentmodulus,thereducedmodulusisafunctionof
bothmaterialpropertiesand geometryofthecrosssection.Forexampleforrectangularandidealized
Isection (i.e., Isection of equal flange areas connected by a web of negligible thickness), one gets,
respectively:
263
ECSSEHB3224A
24March2010
4 EEt 2 EEt
Er = ; Er =
( ) E + Et
2 155
E + Et
Inanycasethereducedmoduluswillbesmallerthantheelasticmodulus.
Fromtheclassicalinstabilityconceptthereducedmodulustheoryiscorrectsinceitisthecounterpart
of the Euler theory in the elastic buckling range. However, many experiments have shown that
columns tested in laboratory usually buckled at loads just slightly above the tangent modulus load.
Shanley [3], explained that the critical stress of a straight inelastic column lies between the tangent
modulus load and the reduced modulus load. Hence, the former represents a lower bound and the
latteranupperboundtothestrengthofaconcentricallyloaded,perfectlystraightcolumn.Becauseof
unavoidable imperfections, the strength of real columns usually falls closer to the tangent modulus
loadthantothereducedmodulusload.Becauseofthisandoftheeaseofusewithwhichthetangent
modulusloadcanbeobtained,itisusuallyadoptedinpracticetorepresenttheultimatestrengthofa
centrallyloadedrealcolumn.
The solution of Eq. 154 involves successive approximations since to plot ct we have to know Et ,
whichinturnisafunctionof ct .Theproblemcanbesolvedbyconstructingtangentmoduluscurve
for the given material, in which Et is plotted against . In that plot, Eq. 154 is a straight line
throughtheoriginwithaslopeof / ( L '/ ) .Thebucklingstressisobtainedastheintersectionof
2 2
thetangentmoduluscurveandthestraightline(seeFigure159).
Figure158:a)Tangentmodulustheoryandb)reducedmodulustheory
Oftenforaluminumalloy,stainlesssteelandcarbonsteelsheet,theRambergOsgoodequation[4]:
264
ECSSEHB3224A
24March2010
3
n 1
= 1 + ; 156 (a)
E 7 0.7

17
ln
n = 1+ 7
156 (b)

ln 0.7
0.85
is adopted to reproduce analytically the stressstrain curve [4] by interpolating the experimental
results.Thethreeparametersneededare E , 0.7 and 0.85 ,i.e.theYoungmodulusandtwosecant
moduli yield strength equal to the ordinate of theintersection with the experimental curve ofaline
throughtheoriginhavingaslopeequalto 0.7E and 0.85E (seeFigure1510).
ct
L
2
Et = f ()

1
E Et
Figure159:Determinationofthebucklingstressfromtangentmoduluscurve
Figure1510:ParameterdeterminationoftheRambergOsgoodstressstraincurve
265
ECSSEHB3224A
24March2010
Thetangentmoduluscurveisthengivenbythefollowingequation
d E
Et = = n 1

d 3n 157
1+
7 0.7
anditisplottedinFigure1511inanondimensionalform.
Figure1511:TangentmoduluscurveasafunctionoftheRambergOsgood
parameters
The RamberOsgood stressstrain curve can be easily exploited to construct stability column curves,
bymakinguseofEq.157andbydividingbothsides ofthe tangentmodulusformulaby 0.7 which
gives
1 0.7 L 1
=
E 158
cr 3n cr
n 1
1 +
0.7 7 0.7

NondimensionalstabilitycolumncurvesareshowninFigure1512.Toobtainthecriticalstressfora
givencolumnfirstofallthethreematerialparameters E , 0.7 and 0.85 aredetermined,sothatthe
ct
leftsideofEq.158isevaluated.Enterthatvalueonhorizontalaxisandread fromtheproper n
0.7
curve(selectedbyEq.156b),finally,compute ct .
These curves can be directly used for buckling design of columns of a given material with cross
sectionsnotsubjectedtolocalbucklingorcripplingbyestablishingasafetymarginandacutstress.A
typicaldesigncurve(usuallynamedEulerEngessercurve)isreportedinFigure1513.
266
ECSSEHB3224A
24March2010
Figure1512:NondimensionalcolumncurvesasafunctionoftheRamberOsgood
parameters[5]
Figure1513:TypicalEulerEngesserdesigncurveforagivenmaterial
15.2.4 Compression crippling

L'
Whentheslendernessratio is<20,thethinwalledcolumnwhichconsistsofflatplatesbucklesby

localinstability,inwhichthecrosssectionisdistortedandthebucklelengthisoftheorderofthecross
sectionaldimensions.Thistypeofinstabilityfailureismorecloselyrelatedtoplatebehaviorthanto
classicalcolumnbuckling.Thethinwallsofthecolumnbucklewhilethecornersandintersectionsof
267
ECSSEHB3224A
24March2010
thesectionwhichconstitutestifferpartitionofthecolumn,cancontinuetobeloadedbeingrestrained
against lateral movement. When also in these areas the stress reaches a threshold value, the section
loses its ability to support additional axial load and fails. Figure 1514 shows the crosssectional
distortion occurring over a thinwalled element and the corresponding stress distribution over the
crosssection.

Figure1514:(a)Distortionofathinwalledcolumnbeforecripplingand(b)stress
distributionoverthecrosssection
Thecripplingstress cc isanaveragestress;itiscalculatedasifthestresswereuniformlydistributed
overthecrosssection.Atfailure,themaximumstress, max reachedinthecornersorintersectionsis
always above the compression yield stress, although cc may by considerably less of that value.
Unlike elastic buckling, crippling induces deformations on the member that do not disappear when
thememberisunloaded.Themaximumstrengthofthecolumnwithregardtocripplingisessentially
afunctionofthecrosssectiongeometryandmaterialproperties.Cripplingappliestoextremelyshort
columns; the stress at which crippling failure occurs is essentially independent of the length of the
column
Thenonlinearbehaviorassociatedwithcompressioncripplingmakestheoreticalpredictionof cc for
arbitrary cross section a hard task. Different semi empirical methods have been proposed for
satisfactoryevaluationofthecripplingstressbydividingthesectioninsimplersubelements[6],[7].A
systematic semi empirical investigation has been made by Gerard [8] who extended a procedure
devotedtoflatplatesbucklingtothecripplingstressevaluation.Afterextensivetestdataanalysisand
interpolationformanydifferentmaterialsandcrosssectionsheproposedthefollowingformula:
m
cc gt E
= n 159
y btot y

where t isthethicknessofaformedcolumnorthemeanwallthicknessofanextrudedcolumn:
t =
bt i i
1510
btot
Here btot = bi is the developed length of the cross section with bi the width of plate elements
evaluatedforextrudedandformedsectionaccordingtoFigure1515.
268
ECSSEHB3224A
24March2010
(a) (b)
Figure1515:Crosssectionoftypicalextruded(a)andformed(b)column
The parameters , m and n are constants determined from test data and analogies with flat plate
buckling, whose value is reported in Table 152for typical thin wall cross section. Let us divide the
thin walls into two categories: flanges, which have a free edge and webs, which are supported by
adjoiningplate elementsonbothedges.Theparameter g isthenumberofthewebsplusthenumber
of the flanges that would exist after cutting each web in two parts. Figure 1516 shows how g is
determinedforsomecrosssections.
The compression yield strength, adequately reduced according to section shape and fabrication
process,isusedasthecripplingstresscutoff.ThecutoffstressislistedinTable152.
The results of Gerard method agree fairly well with experimental results. Some correction to the
parameter could be introduced for formed section where severe strain hardening occurs at the
cornersduringfabricationprocesses[9].Gerardformulacanalsobeemployedforsectionswithlips
andbulbs(Figure1517)which,ifcorrectlydimensioned,simulateasimplysupportededgecondition
fortheadjacentplateelementwhichcanbeassimilatedtoawebinsteadofaflange.Ifthebulborlipis
toosmall,itwillbucklebelowtheflangeadjacentplatewhichitissupposedtostabilize.Gerardhad
derived the curves in Figure 1518 to determine the minimum bulb or lip sizethat allow producing
web behavior of the outstanding plate. The figure gives also the estimated critical strain cr ,
according to plate theory, of the outstanding leg when the proportion of the bulb or lip is not
satisfactoryforwebaction.
269
ECSSEHB3224A
24March2010
Table152:ConstantsforGerardmethod
Section g m n Cutoff
stress
Extrudedangle 2 0.56 0.85 2 0.8 y
Extruded 12 0.56 0.85 2 0.75 y
rectangulartube
Formed 0.55 0.85 2 0.75 y
multicorner
section
ExtrudedT 3 0.67 0.40 2 0.8 y
Extruded 4 0.67 0.40 2 0.8 y
cruciform
ExtrudedH 7 0.67 0.40 2 0.8 y
ExtrudedC 5 0.96 0.75 3 2 3 t / bw y
ExtrudedZ 5 0.96 0.75 3 2 3 t / bw y
Figure1516:Valueof g fortypicalsections[8]
270
ECSSEHB3224A
24March2010
Itisimportanttonotethatacolumnmadeofthinlegallhavingthesamelocalbucklinglengthsoto
reach their buckling stresses simultaneously, will fail at a lower load than a similar column with
element with different buckle lengths. In fact in the former case none of the elements can provide
rotationaledgerestrainttoitsadjoiningelements,andalltheelementsbehaveasiftheyweresimply
supportedalongthelineswheretheyjoinotherelements.Thishappensforequalleggedangle,Tand
cruciformsections,squaretubeofconstantthickness.Whenthisoccursanalyticaltreatmentispossible
andsimplesolutionscanbederivedfollowingplatebucklingtheory[10].
Figure1517:Typicallipsandbulbs
Figure1518:Minimumbulb(a)andlip(b)dimensiontoassurewebbehaviorof
theoutstandingplate[11]
15.2.5 Combined flexural and local instability

Crippling stress, at small slenderness ratios, and the Euler stress, at large slenderness ratio, are
confirmed by test results. For intermediate slenderness ratio these approaches usually overestimate
thebucklingload.Thisisbecauseintheintermediaterange,thefailureisacombinationofprimary
and secondary modes. A primary instability mode in which, the column bends and twists
simultaneouslyoccursintheintermediaterangeforsectionswithalowtorsionalrigidity.Thebuckle
lengthisoftheorderofthecolumnheight;andevenifoutofplanewarpingispossibleasthecolumn
twists, the crosssectional shape does not change. For doubly symmetric section, in which the shear
271
ECSSEHB3224A
24March2010
centerandthecentroidcoincide,thepuretorsionalbucklingmayoccur.Thinwalledopensectionsare
particularlysusceptibleofthistypeofinstability,andbecausethesesectionsarealsosubjecttolocal
instability, failure may occur by a combination of bending, twisting and local buckling. Interaction
betweendifferentbucklingmodeswhichoccursatintermediateslendernessratioisnotamenableofa
general analytical treatment. As a result, empirical equationsare generally used for this region. The
JohnsonEulerparabolaisfrequentlyusedfrom L '/ = 0 toatransitionalslendernessratiowherethe
parabolajoinstheEulercurve.Whileitisempirical,the JohnsonEulerequationwillyieldreasonable
strengthpredictionformostcommontypesof structural sectionswhenusedinconjunctionwiththe
column cross section crippling strength determined in the previous section. The JohnsonEuler
solutionhasbeencommonlypresentedasafamilyofcurvesasshowninFigure1519.Iftheparabolas
aretangenttotheEulercurveat c = 0.5 cc andhavethevertex, c = cc ,at L '/ = 0 ,then,their
equationreads:
( L / ) 2
cr = cc 1 cc 2 1511
4 E
L'
Andthetransitionalslendernessratio wheretheparabolajoinstheEulercurveis:
tr
L 2E
= 1512
tr cc
cr
cc cc
2
L L

tr
Figure1519:JohnsonEulercurve
15.2.6 Torsional instability

15.2.6.1 Overview
Intheprevioussections,itwasassumedthatthecolumnwastorsionallystable;i.e.,thecolumnwould
eitherfailbybendinginaplaneofsymmetryofthecrosssection,bycrippling,orbyacombinationof
crippling and bending.However, if thetorsional rigidity of the section is verylow, the column will
buckleeitherbytwistingorbyacombinationofbendingandtwisting.Thisoccursforabarofthin
walledopencrosssectionwhichcanbucklebytwistingatloadswellbelowtheEulerload.Frequently,
272
ECSSEHB3224A
24March2010
thinwalledopensectionsarenotdoublysymmetric,thencentroidandshearcenterdonotcoincide
andtherefore,torsionandflexureinteract.
Pure torsional instability is an unstable rotation of a columns crosssections about the longitudinal
axis of the column (see Figure 1520); flexural bending of the axis of rotation does not occur during
pure torsional instability. In general the buckling load pertinent to pure torsional instability is
different from flexural instability loads for a particular column. A column is apt to pure torsional
bucklingifithaseitheradoublysymmetricorapointsymmetricopencrosssection.Ifacolumnhas
just one or no symmetry axes the torsional instability mode is coupled with flexural modes of
instability.Acolumnofdoublysymmetricclosedcrosssectionisnotpronetotorsionalinstability.
15.2.6.2 Cross section with two axes of symmetry or point symmetry

Whenthecrosssectionhastwoaxesofsymmetryorispointsymmetric(seeFigure1521),theshear
centerandthecentroidwillcoincide.Inthiscase,thepurelytorsionalbucklingstressisgivenby[12]:
Pc 1 2 ECW
c = = 2
+ GJ 1513
A I 0 LT
Where
I 0 = Polarmomentofinertiawithrespecttocentroid
Cw = warpingconstantofsection
LT = effectivetorsionallengthofthecolumn
G = shearmodulusofelasticity
J = SaintVernanttorsionconstant
Theeffectivetorsionallengthisthedistancebetweenpairsofpointsalongthebuckledcolumnwhere
the internal resisting moment is equal to zero. The effective torsional length is analogous to the
effectiveflexurallengths,andmaybeevaluatedbyanexaminationofthecolumnsendrestraintswith
respect to twisting and warping. If the ends are free to warp out of the crosssection plane but
restrainedagainstcolumnaxesrotation,than LT = L .Incaseendsectioncannotwarp, LT = L / 2 .In
practice,mostcolumnswillhaveaneffectivetorsionallengththatisbetweenthesetwoextremes.
Forcrosssectionswithtwoaxesofsymmetry,thereisnointeractionbetweenbucklingmodesandthe
columnfailseitherinpurebendingorinpuretwisting.
ThusthecriticalvaluesoftheaxialloadistheminimumbetweenEulerbucklingload,Eq.151,and
puretorsionalbucklingloadEq.1513.Theeffectivebucklingloaddependsontheshapeofthecross
sectionandthelengthofthemember.Ingeneral,puretorsionalbucklingwilloccurforacolumnwith
wideflangesandshortlengths.
273
ECSSEHB3224A
24March2010
Figure1520:Torsionalbucklingofacruciformcolumn
Figure1521:Possiblecolumnbucklingmodes
274
ECSSEHB3224A
24March2010
15.2.6.3 Cross section with one axis of symmetry
Ifthecrosssectionhasoneaxisofsymmetry,saythexaxis,thebucklingloadisgivenby[12]
( P PcEy ) r02 ( P PcEx ) ( P Pc ) P 2 x02 = 0 1514
where
2 EI x 2 EI y
PcEx = ; PcEy = 1515
L2 L2
r0 = I 0 / A is the polar radius of gyration and x0 is the distance between the shear center and the
centroid.Thebucklingload Pcr ofthecolumnwillbethesmallestsolutionofEq.1514;i.e.oneofthe
followingvalues:
1
(P + PcEx ) 4kPc PcEx
2
Pc1 = PcEy ; Pc 2 = Pc + PcEx c 1516
2k
where
2
x
k = 1 0 1517
r0
Therefore, a singly symmetrical open section such as an angle or channel can buckle either of two
modes: by flexural buckling ( Pc = Pc1 ) or torsionalflexural buckling ( Pc = Pc 2 ).Which of thesetwo
occursdependsonthedimensionandshapeofthegivensection.If I y > I x ,then,flexuralbucklingis
ruled out and only torsionalflexural instability can occur. Column bucklingregions, visualizing the
domainsrelatedtodifferentbucklingmechanisms,canbederivedforeachsectionasafunctionofits
geometricalproperties[13].
15.2.6.4 Cross section with no axis of symmetry

Inthecaseofacolumnofthinwalledopencrosssectionwithalackofsymmetry,bucklingoccursby
acombinationoftorsionandbending.Purelyflexuralorpurelytorsionalbucklingcannotoccur.The
equationgoverningtheinstabilityproblem[12]is:
r02 ( P PcEy )( P PcEx )( P Pc ) P 2 y02 ( P PcEx ) P 2 x02 ( P PcEy ) = 0 1518
Thesolutionofthiscubicequationyieldthreeroots Pc1 , Pc 2 and Pc 3 ,thesmallestwillbethecritical

load Pc .
15.3 Lateral buckling of beams

Abeammaybuckleoutoftheplaneofloadingunderthecircumstancethatitsstiffnessinthatplaneis
much higher than lateral and torsional stiffness and adequate lateral restraints are not present (see
Figure1522).Inthecaseofaperfectlystraightbeam,therearenooutofplanedeformationsuntilthe
applied load reaches a critical value, at which point the beam buckles by deflecting laterally and
twisting.Thelateraldeflectionandtwistingareinterdependent.Whenthebeamdeflectslaterally,the
induced moment exerts a component torque about the deflected longitudinal axis which causes the
beam to twist. Such buckling behavior has been referred to as flexuraltorsional buckling or simply
lateralbuckling.
275
ECSSEHB3224A
24March2010
Flexuraltorsional buckling influences the design of laterally unrestrained beams in much the same
waythatflexuralbucklinginfluencesthedesignofcolumns.Thusthebendingstrengthwillnowbea
functionofthebeamsslenderness.
Mx
Mx

Figure1522:Lateralbucklingofarectangularbeam
For the simplest case of doubly symmetric beams under uniform moment, the critical buckling
momentisgivenby:

M cr = EI y GJ 1 + W 2 1519
KL
where
ECw
W= 1520
KL GJ
ThesecondsquarerootinEq.1519representstheeffectofwarpingtorsionalstiffnessand K isthe
effectivelengthfactor;itsvaluesdependonrestraintconditions.Forasimplysupportedbeam, K = 1 ,
for a cantilever K = 2 , for clamped ends, K = 0.5 . For rectangular solid or boxed beams Cw is
negligible,andthen W maybetakenaszero.Eq.1519assumesthatthelateraltorsionalbucklingis
not influenced by inplane deflection. This hypothesis is justified when the flexural rigidity EI x , is
muchlargerthantheflexuralrigidity EI y ,sothattheinplanedeflectionwillbenegligiblecompared
withthatoftheoutofplanedeflection.Ifbothrigiditiesareofthesameorderofmagnitude,theeffect
ofbendinginthe verticalyzplanemaybeimportantandshouldbeconsideredincalculating M cr .
An approximate solution that includes the effect of inplane deflection is given by Kirby and
Nethercot[14]as
EI y GJ
M cr = 1+W 2 1521
KL Ir
where
276
ECSSEHB3224A
24March2010
Iy
Ir = 1 1522
Ix
Note that if I y = I x , I r becomes zero then according to Eq. 1521 M cr becomes infinite. One can
conclude that the lateral torsional buckling of beams is possible only if the cross section possesses
differentbendingstiffnessesinthetwoprincipalplanesandtheappliedloadsactintheplaneofthe
weakaxis.
Asaresult,lateraltorsionalbucklingwillneveroccurincircularcrosssectionsorsquareboxsections
inwhichallthecomponentplateshavethesamethicknesses.
Several factors affect elastic lateral buckling considerably, such as: non uniform bending moment
relatedtodifferentpatternsofspanloading,endrestraints,shapeofthecrosssections,lateralbracing
system,distancefromtheshearcentertothepointofapplicationoftransverseload.Acomprehensive
analysisoflateralbucklingofbeamswhichaccountsforthesematterscanbefoundin[15].
15.4 The beam-column
15.4.1 Overview
Inthischapterasupplementtotheusualbeamcolumnbucklingtableswillbepresentedwithspecial
referencetosomeanalysismethodsandtotheinfluencesencounteredinthelightweightconstructions
oftheaerospaceindustry.Inthissectionthecrosssectionofthebarisassumedtoremainplaneandto
keep its original geometrical form. Some other influences will be investigated, namely shear
deformation,pretensioningandelasticsupport.Thesuggestedanalysismethodswillbepresentedby
usingselectedapplications.
15.4.2 The "complete" differential equation of the beam-column

Here,theabovementionedinfluences(sheardeformation,pretensioningandelasticsupport)willbe
takenintoaccount,especiallycommentingthedifferentmathematicalmodelsusuallyencounteredin
thetechnicalliterature,oneofthemexcludingthepureshearbucklingofbeams.
Also some different methods will be shown for the analysis of critical loads. Such methods are: the
variationalcalculus,thesolutionofthedifferentialequation,transfermatrices,pseudoforcemethods,
dynamiccriteria(especiallyfornonconservativesystems),etc.
Thesimplestdifferentialequation(DE)isthatofastraightshearrigidbeamunderalineload p
EIwVI = p 1523
In Figure 1523 the sign conventions, the equilibrium as well as the deformations are explained.
IV
Hereby w is the displacement of the crosssection, w its fourth order derivative, and EI the
bendingstiffness.
However,thisequationneitherconsiderssheardeformationnorsupplementarytermsduetoelastic
foundation and secondorder theory. In the following, the DE will be stepwise completed.
Nevertheless the assumptions will be retained that the crosssection remains plane and keeps its
geometricalshape.Here,justlinearelasticmaterialbehaviourwillbeconsidered.
The most general beam, the DE of which is utilized for the computation of critical loads is the pre
tensioned beam on an elastic foundation. Especially the last item is very useful, amongst others, to
277
ECSSEHB3224A
24March2010
explainthebehaviourofshellsandplatesunderloading.Figure1523representstheequilibriumof
the pretensioned beam on an elastic foundation. The pretensioning force N (positive in tension)
keeps its direction duringthe deformation of the beam. The shear force Q is the original one acting
perpendicularlytothestraightreferenceaxis.
Theeffectiveshearforcewhichinducestheshearstressesatthefaceofthecrosssectionreads
Qeff = Q + NwI 1524
And is to be considered when calculating the bending moment according to dM = Qeff dx from
whichfollows
dM
= Qeff = Q + Nw I 1525
dx
Figure1523:Thepretensionedbeamonelasticfoundation,signconventionstatic
systemandequilibriumonthedeformedelement
Theequilibriumoftheverticalforcesdelivers
p* = dQeff / dx = p + kw NwII = Qeff I 1526
InsertingEq.1526intoEq.1523yields
EIwIV NwII + kw = p 1527
Further it is to be considered, that the deformation w consists of a bending wb and a shear

contribution ws
278
ECSSEHB3224A
24March2010
w = wb + ws 1528
I
The slope w of the midline is affected by the bending moment and the shear force. The bending
deformation only rotates the face of the crosssection (changes the curvature) whereas the cross
sectionremainsperpendiculartothecurvedmidline.Duringthesheardeformationthecrosssection
ofthebeamelementremainsparalleltotheoriginalposition.Thereby,theresultingslopebecomes
w I = wbI + wsI 1529
Therelationbetweenloadsanddeformationisderived,as
wsI = Qeff / (GAS ) 1530
( As = A / ,with beingtheshapeshearcorrectionfactor)sothat
w I = wbI + wsI = wbI (Q + Nw I ) / (GAS ) 1531
Andfinally
wbI Q / GAS
wI = 1532
1 + N / GAS 1 + N / GAS
Thesecondderivativeofthebendingdeformationis
wbII = M / ( EI ) 1533
Inordertoestablishthedifferentialequationthesecondandfourthorderderivativesareneeded:
forbending
wbII = M / ( EI ) , wbIV = QeffI / ( EI ) 1534
forshear
wSII = QeffI / (GAS ) , wSIV = QeffIII / (GAS ) 1535
This leads to a formulation of the full DE after inserting the above equations into Eq. 1527 and
resolvingfor w anditsderivatives
N kEI II EI
EIw IV (1 + ) (N + ) w + kw = p + p II ( ) 1536
GAS GAS GAS
The solution of the differential Eq.1536 is the sum of the homogeneous solution plus a particular
integral. As Eq. 1536 is an ordinary DE with constant coefficients as a solution function for the
homogeneouspart:
w = Ci exp( i x / L) 1537
can be employed. By substituting this solution into Eq.1536, the socalled characteristic equation is
obtained(assuming p = 0 ),
N / EI + k / GAs k / EI
4 2 + = 0 1538
1 + N / GAs 1 + GAs
Once the complete solution is found the integration constants can be determined based on the
boundaryconditions.Unfortunatelythissolutionisonlyvalidforauniformbeamwithconstant EI
and As G alongthebeamandforparticularboundaryconditions.
279
ECSSEHB3224A
24March2010
Ingeneralapplication,amatrixformulationismoreflexibleandnumericallyadvantageous.Forthis
purpose,thesocalledtransfermatrixoftheuniformbeamwillbeestablished.
Deformationandloadingofacrosssectionisentirelydefinedbythestatevector
v = ( w, wbI , M ( EI ) , Q ( EI ))T 1539
Thefirstderivativeofwhichreads
v I = Av 1540
In this equation is termed the differential matrix, the components of which are still ascertained,
andwhichmaybeformulatedusingtheEqs.1525,1526and1532inmatrixnotationas
I
w w
I I
wb wb
= A with
M M
Q Q
1 EI / GAS
0 0
(1 + N / GAS ) (1 + N / GAS ) 1541

0 0 1 0
A= N / EI 1

0 0
(1 + N / GAS ) (1 + N / GAS )

k 0 0 0
EI
Writing the inhomogeneous differential equation 1536 in vectorial form one can solve it with the
solutionfunction1537alsowritteninvectorialform.Then,itfollowswith
v = c exp(x / L) 1542
theequation
Ic exp(x / L ) = Ac exp(x / L ) 1543
wherein isthediagonalunitmatrix.Forthissystemasolutionexistsonlyifthedeterminant
A I = 0 1544
becomeszerowhichmeans
1 EI / GAS
0
(1 + N / GAS ) (1 + N / GAS )
0 1 0
N / EI = 0 1545
0
(1 + N / GAS ) (1 + N / GAS )
k
0 0
EI
TheresultisidenticaltothecharacteristicequationEq.1538.
Ontheotherside,thedifferentialmatrix leads,afteraPiccarditeration,tothetransfermatrix U ,
whichrepresentstheconnectionbetweenthestatevectors, v i+1 and vi ,atthecrosssectionfacesof
280
ECSSEHB3224A
24March2010
thesinglebeamsection(seeFigure1524).Byamatrixmultiplicationoftheassociatedmatrixofeach
ofthesegmentwise,uniform,juxtaposedsequenceofsinglebeamsegments
N ( Li , j A i , j ) n
Ui, j = I + 1546
n =1 n!
the connection between the endfaces of thesegmented complete beamis achieved.Here Li , j is the
lengthoftheindividualbeamsegment,whichisassumedtohaveconstantproperties.
Introducingtheboundaryconditionsthebucklingloadwithassociatedmodescanbedetermined.
Figure1524:Transferofstatevectorfromthefacei1tothefaceiofthebeam
element
Two different models can be applied in the beam transfer matrix method. One is called Sandwich
Modelling(similartothemodellingpresentedofthesheardeformationofasandwich)andtheother
Domino Modelling (here the shear deformation is described by a stack of dominos). Figure 1525
depictsthemaindifferences.
Figure1525:a)Sandwichmodellingandb)dominomodelling
a. Sandwichmodeling
Thedifferentialmatrix A isstillthatgivenbyEq.1541
b. Dominomodelling
Thesheardeformationreads
281
ECSSEHB3224A
24March2010
wSI = Q * /(GAS ) = (Q + Nwb I ) / (GAS ) 1547

Insertedintothederivedsetofequationsdeliversthetwoequations
Q + NwbI N Q
w =w +w =w
I I
b
I
S
I
b= wbI (1 ) 1548
GAS GAS GAS
N N
M I = Q + Nw I = Q (1 ) + N (1 ) wb I 1549
GAS GAS
and eventually the set of relationships is expressed in a short and concise manner by the
differentialmatrix
I
w w
I I
wb wb
= A with
M M
Q Q
N EI
0 (1 ) 0
GAs GAs 1550

0 0 1 0
A=
0 N N N
(1 ) 0 (1 )
EI GAs GAs

k 0 0 0
EI
15.4.3 Stability under axial load

Theaboveindicatedanalysismethodcanbeappliedfordifferentloadconditionsaswellasboundary
conditions:
arbitrary boundary conditions for columns can be considered e.g. using transfermatrices,
wherearbitraryboundaryconditionsincludingelasticclampingcanbeintroduced
isostaticandredundantsystems,oncemore,transfermatrixmethodsofferdifferentadvantages
forthistask,alsointhecaseofsectorwisevaryingcrosssectionproperties
Southwellplotlikestabilityanalysisisaniterativeprocedurebasedonmultipledeformation
analyses for several pretensions, accelerating the calculations by Southwells method. This
procedure isalso suitablefor the estimation of thestability in the case of plastic behaviourof
thematerials,providedthecurveisnearlybilinear.
axisymmetric buckling of shells as beams on elastic foundation considering supplementary
supports(e.g.solidrockethousingwiththefillingaselasticsupport),andarbitraryboundary
conditions
bucklingofsandwichfacesheets(lateralandaxialsupport),onthickorthincores
torsional instability, torsionbending buckling and lateral instability under axial compression
andbendingmoment,investigatedforclosedandopenprofileswithreducedtorsionalstiffness
follower forces as nonconservative systems, where classic investigationmethods fail for
stabilityprediction,heretheapplicationofdynamicprocedureswillbeexplained
282
ECSSEHB3224A
24March2010
buckling of pipes under internal pressure (depending on boundary conditions) and/or liquid
velocityviacentrifugeorCoriolisforcesasconservativeornonconservativesystems
bucklingunderthermalstresses,analysingtheinstabilitylevelandthesubsequentequilibrium
position
general beamcolumn theory, and, as an important example: discontinuity stresses in shells;
comparisonfirstandsecondordertheory
plasticityeffectsestimation(Shanley/Engesser/Krmnmethod)ornumericintegration
Asanexampleletusconsideracolumnwithsimplysupportedendsunderaxialload.Thefunction
w = w0 sin(n x / L) 1551
isapossiblesolutionwhichfulfilsthesimplysupportedboundaryconditions.
For determining the critical loads the two approaches Sandwich and Domino Modelling are
employed,again.
a. SandwichModelling
Inserting above solution function into Eqs. 1536 and 1538 the critical axial compressive load
becomes
(n / L) 4 + k (n / L) 2 / (GAs ) + k / EI
N crsandwich = 1552
(n / L) 4 / (GAs ) + (n / L) 2 / EI
ThisgivesforthevariouscasestheEulercolumnloads:
1. Classical:
EI 2
AS G = , k = 0, n = 1 N crsandwich = 1553
L2
2. withsheardeformation:
EI 2 EI 2
sandwich
N cr = / 1 + 2
1554
S1
L2 GAS L
3. rigidinbending,elasticinshear:
EI = , k = 0, n = 1 N crsandwich
S2
= GAS 1555
4. rigidinshear,onelasticfoundation:
2 n 2 k
GAs = , k 0 N crsandwich = EI 2 2
k
L 2
n 1556
L2
Withaminimumat k / EI = ( n / L) fromwhichfollows
4
N crsandwich
k min
= 2 kEI 1557
b. Dominomodeling
Inserting the deformation approach Eq. 1532 into the Eqs. 1548 and 1549 and imposing
det[A]=0yields,fork=0.
283
ECSSEHB3224A
24March2010
EI 2
N crdo min o = 0.5GAs (1 + 1 + 4 ) 1558
GAS L2
Figure 1526 shows the evolution of the critical column load for both models as functions of
x = EI 2 / ( GAs L2 ) . In the case of very low shear stiffness, x becomes infinite and as limit
valuesturnout
N crsandwich

= GAS 1559
EI 2
N crdo min

o
= 0.5GAS (1 + 2 ) 1560
GAS L2
Figure1526:SandwichmodellingandDominomodellingoftheshear
deformationcriticalcolumnloads
It can be seen that the domino modeling excludes a purely shearinduced buckling. However, this
statementiscontradictorywithexperiencesoncolumnshavingverylowshearstiffness.Timoshenko
[16] means that the dominomodeling is rather more accurate, anyhow he prefers the sandwich
modellingaswellasthesaferone.
In the case of the buckling of a column representing a helical spring Timoshenkos choice is the
domino modelling, where it remains to investigate, whether the low extensional stiffness is not
dominantforthebucklingofthehelicalspring.Anevaluationofrelevanttestresultswouldbeuseful.
15.5 In-plane buckling of rings and arches
15.5.1 Overview
Ringsandarcheswhenloadedintheirplanearesubjecttobuckling.Themodeofbucklingdepends
onthetypeofcrosssection,boundaryconditionandshapeofthecurvedelement.Inthefollowingwe
assumethatthebehaviorofthematerialislinearelasticandthatthethicknessismuchsmallerthan
theradiusofcurvature.Fromthebucklingpointofview,theanalyticaltreatmentofhigharchesand
rings,inwhichthecenterlinecanbeassumedextensionlessatinstability,isdifferentfromthatofflat
284
ECSSEHB3224A
24March2010
arches,inwhichaxialstrainneedstobeconsidered.Flatarchesexhibitsnapthroughinstability(also
called oilcanning), i.e. a visible and sudden jump from one equilibrium configuration to another
specular symmetric one. This contrasts with high arches behavior which usually experiences
asymmetrical buckling. For a complete theoretical and experimental investigation dealing with the
bucklingofringsandarchestheinterestedreadershouldconsult[12],[17]and[18].
15.5.2 The thin circular ring

On the contrary of perfect columns, the perfect thin ring under lateral loading undergoes
displacementspriortolossofstability.Letusassumethatplanesremainplaneafterdeformation,the
normalstotheneutralaxisareinextensional,andtheringisthin.
Figure1527:Ringgeometryandsignconvention[17]
FollowingthegeometryandsignconventiononFigure1527andtheanalyticaltreatmentin[17],the
strainatanymaterialpointatdistance z fromtheneutralsurfaceisgivenby
= 0 + z 1561
where denotestheextensionalstrainofmaterialpointsontheneutralsurfaceand thechangein

0
curvature.Forsmallengineeringextensionalstrainandmoderatelysmallrotationsonegets:
285
ECSSEHB3224A
24March2010
w v' ( w ' v ) '

0 = + +
R R 2R2 1562
w '' v'
=
R2
being w() and v() the displacement components of the neutral surface material points and the
primes denote derivative with respect to . Consider the thin circular ring to be loaded by a
uniformlydistributedloadarounditscircumferencewithcomponents pr and p intheradialand
tangentialdirections,respectively.Theequilibriumequationsforsuchloadcanbe derivedusingthe
stationarytotalpotentialenergycriteriontoget[17]:
NR + v' N ( w ' v ) M ''+ pr R 2 = 0

1563
RN '+ N ( w ' v ) M '+ p R 2 = 0
In Eq. 1563 N = EA and M = EI denote the axial stress resultant and bending moment,
0
respectively.
Ifauniformradialpressureloading p ispresent,then pr = 0 , p = p andtheprebucklingsolution

ischaracterizedbyuniformradialcontraction(orexpansion)with:
pR 2
wp = , v = 0 , N = pR , M = 0 , pr = p , p = 0
p p p p p
1564
EA
Thecriticalload,incaseofuniformpressureloading,isthesmallestload pcr forwhichequilibriumis
possible in adjacent non circular configuration. To investigate the possible existence of adjacent
equilibriumpositiononeintroducessmallincrements(denotedbystarredquantities)aroundthepre
bucklingsolution:
pcr R 2
w=w +w =
p *
+ w* , v = v p + v* = v* , N = N p + N * = pcr R + N *
EA 1565
M = M + M = M , pr = pcr + pr , p = p
p p * * * *
Since the increments can be taken as small as we wish, N << pcr R and the linearized kinematic
*
relationcanbeusedfortheinternalforces:
EA * EI
N* =
R
( w + v* ) , M * = 2 ( w* '' v* ')
R
1566
SubstitutingEqs.1565and1566intoEqs.1563resultsinthebucklingequations
EI *
EA ( w* + v* ' ) + pcr R ( w* '' v* ')
2 (
w '''' v* ''' ) + pr* R 2 = 0
R 1567
EI *
EA ( w '+ v '') + pcr ( w ' v ) 2 ( w ''' v '') + p R = 0
* * * * * * * 2
R
* *
Inordertosolvethepreviousequations,thebehavioroftheincrementalloadcomponents pr and pe
during the buckling process should be investigated. There are three possibilities concerning the
behavioroftheload,namedcaseI,IIandIII(illustratedinFigure1527).IncaseI,itisassumedthat
theloadremainsnormaltothesurfaceastheringdeforms(liveload).IncaseII,itisassumedthat
theloadremainsparalleltoitsoriginaldirections.Finally,incaseIIIitisassumedthatthepressure
remainscentrallydirected(deadload).Forthethreecasesonegets:
286
ECSSEHB3224A
24March2010
pcr * *
Case I: pr* = 0, p* =
R
( w v ')
Case II: pr* = 0, p* = 0 1568
v*
Case III: pr* = 0, p* = pcr
R
Then,Eqs.1567reducestotheeigenvalueproblem:
R2
( w* + v* ') + ( w* '' v* ') ( w* '''' v* ''' ) = 0

2 0 ( Case I ) 1569
R *
( w '+ v '') ( w ''' v '' ) + w ' v
* * * * *
( Case II ) = 0
w* ' ( Case III )
where istheradiusofgyrationofthecrosssectionalgeometryand
R3
= pcr 1570
EI
* *
The continuity conditions for the complete ring require that w and v be periodic with their
derivatives,thusthefollowingsolutioncanbeassumed
w* = An cos n , v* = Bn sin n 1571
n is a positive integer. The characteristic equation following

Where An and Bn are constants and
fromthehomogeneouslinearequationsin An and Bn hasthefollowingsolutions:
Case I: = (n 2 1)
Case II: = n2
1572
(n 1)2 2
Case III: = 2
(n 2)
Thecriticalvalueistheminimum asafunctionofintegervaluesof n .Since n = 1 correspondstoa
rigidbodymotionwhichissupposedtobeavoidedbyconstraints,thesmallesteigenvalueisseento
correspondto n = 2 (ovalizationofthering):
EI
Case I: cr = 3, pcr = 3
R3
EI
Case II: cr = 4, pcr = 4 3 1573
R
EI
Case III: cr = 4.5, pcr = 4.5
R3
CaseIisthebestmodelforthecaseoffluidpressure.CaseIIIdescribestheproblemofaringwhichis
loadedbycloselyspacedradialcablesconvergingtothecenterofthering.CaseIIisnotamenableof
practicalapplications.
287
ECSSEHB3224A
24March2010
15.5.3 High circular arch
The buckling behavior of high circular arches under normal pressure is still governed by Eq.1567,
under the assumption that first the arch is uniformly contracted and then it buckles with support
immovable,asshowninFigure1528.
Figure1528:Geometryofpinnedcirculararch[17]
Ifthearchissimplysupportedatbothends,theboundaryconditionsat = aregivenby
EI * EA *
2 (
w* = 0, M * = w '' v* ' ) = 0, N * = ( w + v* ' ) = 0 1574
R R
whicharesatisfiedbythefollowingsolution
n n
w* = An sin , v* = Bn cos 1575

Thecriticalconditioncorrespondsto n = 1 andisgivenby:
EI 2
Case I: pcr = 3 1
R
2
EI
Case II: pcr = 1576
R3
2
EI ( / ) 1
2
Case III: pcr = 3

R ( / )2 2
Thesolutionforthecaseofclampedends,duetoE.L.Nicolaiandpresentedin[16],isgivenby:
EI 2
pcr = k 1
R3
1577
Where k isthesolutionofthefollowingequation:
288
ECSSEHB3224A
24March2010
k tan cot k = 1 1578

These results are notapplicable to shallow arches. Buckling of shallowarchesis treated in [12],[17]
and[18].
15.6 References
[1] Wang,C.M.,Wang,C.Y.andReddy,J.N.,ExactSolutionsforBucklingofstructural
memebers,CRCPress,BocaRaton,FL,2005
[2] Hoff,N.J.,BucklingandStability,J.RoyalAero,Soc.,Vol.58,AeroReprintNo.123,1954.
[3] Shanley,F.R.,InelasticColumnTheory,J.Aero.Sci.,Vol.14,1947,pp.261267.
[4] Ramberg,W.andOsgood,W.R.,DescriptionofStressStrainCurvesbyThreeParameters,
NACATechNote902,1943.
[5] Cozzone,F.P.andMelcon,M.A.,NondimensionalBucklingCurves:TheirDevelopmentand
Application,J.Aero.Sci.,Vol.13(10),1946,pp.511517.
[6] Needham,R.A.,TheUltimateStrengthofAluminumAlloyFormedStructuralShapesin
Compression,J.Aero.Sci.,Vol.21(4),1954,pp.217229.
[7] Crockett,H.B.,PredictingStiffenerandStiffenedPanelCripplingStresses,J.Aero.Sci.,Vol.
9(13),1942,pp.501509.
[8] Gerard,G.,HandbookofStructuralStability,partV:CompressiveStrengthofFlatStiffened
Panels,NACATechNote3785,1957.
[9] Gerard,G.,TheCripplingStrengthofCompressionElement,J.Aero.Sci.,Vol.25(1),1958,
pp.3752.
[10] Rivello,M.,TheoryandAnalysisofFlightStructures,McGrawHill,1969.
[11] Becker,H.,HandbookofStructuralstability,partII:BucklingofCompositeselements,NACA
TechNote3782,1957.
[12] Brush,D.O.andAlmroth,B.O.,BucklingofBars,Plates,andShells,McGrawHill,NewYork,
1975.
[13] NASAAstronauticalStructuresManual,Vol.II,NASATMX73306,1975.
[14] Kirby,P.A.andNethercot,D.A.,Kirby,DesignforStructuralStability,GranadaPublishing,
Suffolk,1979.
[15] Chen,W.F.andLui,E.M.,StructuralStability:TheoryandImplementation,Elsevier,New
York,1987.
[16] Timoshenko,S.P.andGere,J,TheoryofElasticStability,McGrawHill,NewYork,1961.
[17] Simitses,G.J.andHodges,D.H.,Fundamentalsofstructuralstability,Elsevier,NewYork,
2006.
[18] Singer,J.,Arbocz,J.andWeller,T.,BucklingExperiments.ExperimentMethodsinBucklingof
ThinWalledStructuresVol.I,Wiley,NewYork,1998.
[19] J.P.DenHartog.Mechanicalvibrations.McGrawHill1940
[20] V.V.Bolotin.Thedynamicstabilityofelasticsystems.HoldenDay1964
289
ECSSEHB3224A
24March2010
16
Flat and curved panels
16.1 Overview
Thischapterpresentsthecharacteristicsofthebucklingbehaviourofunstiffenedandstiffenedpanels.
Section16.2treatsflatpanels,whileSection16.3dealswithcurvedpanels.
16.2 Flat Plates
16.2.1 Introduction
Important characteristics of plates have been presented in Section 5.3. For plates, the thickness
dimension h is small as compared with characteristic inplane dimensions, such as the length a and
width b in the case of a rectangular plate ( h << a , h << b ). Figure 161 shows a rectangular plate
subjectedtoinplanecompressiveloading.Inthischapter,emphasiswilllieonrectangularplates.The
loading cases that will be considered are inplane compressive loading in the x or ydirection, and
shear loading. The characteristics of plate buckling analysis will be mainly presented for isotropic
material,butthebasicsfortheanalysisoflaminatedplateswillalsobetreated.Postbucklinganalysis
of isotropic plates has been discussed in Section 5.3. The postbuckling loadcarrying capability is
essentialinthedesignofstiffenedpanelsandwillbetreatedinSection16.2.6.
Figure161:Rectangularplateundercompressiveloadinginthexdirection
16.2.2 Basic Equations for Isotropic Plates

Thebasicequationsforplatebucklinganalysisarethenonlinearequilibriumequations,thenonlinear
straindisplacement relations, and the constitutive equations. The stability equations will be used in
ordertoobtainexpressionsforthebucklingload.
290
ECSSEHB3224A
24March2010
Figure162:Stressandmomentresultants
Using the stress and moment resultants defined in Figure 162, the Von Krmn type nonlinear
equilibriumequationscanbewrittenas(seeRef.[1])
N x , x + N xy , y = 0 161a
N xy , x + N y , y = 0 161b
M x , xx + 2 M xy , xy + M y , yy + N x w, xx +2 N xy w, xy + N y w, yy = p 161c
Thecorrespondingnonlinearkinematicrelationsreferringtothemidsurfacequantitiesare
x = u , x + 12 w,2x x = w, xx
y = v, y + 12 w,2y y = w, yy 162
xy = u , y + v, x + w, x w, y xy = w, xy
Theconstitutiveequationsforanisotropicplateare:
N x = C ( x + y ) M x = D ( x + y )
N y = C ( y + x ) M y = D ( y + x ) 163
1
N xy = C xy M xy = D (1 v ) xy
2
Eh Eh3
and C = istheextensionalstiffnessand D = thebendingstiffnessoftheplate.
(1 v 2 ) 12(1 2 )
Introducing the appropriate constitutive and kinematic relations for the moment intensities in Eq.
161creducesthisequationtotheform
D 4 w ( N x w, xx +2 N xy w, xy + N y w, yy ) = p
where w w, xxxx +2w, xxyy + w, yyyy

4
Thustheequilibriumequationsforaplateunderinplaneedgeloadingcanbewrittenas
N x , x + N xy , y = 0 164a
N xy , x + N y , y = 0 164b
D 4 w ( N x w, xx +2 N xy w, wy + N y w, yy ) = p 164c
291
ECSSEHB3224A
24March2010
Nextthestabilityequationsarederivedmakinguseoftheadjacentequilibriumcriterion.Assuming
that
u u0 + u1
v v 0 + v1 165
w w0 + w1
where the incremental displacement ( u1 , v1 , w1 ) is arbitrarily small and, tentatively, ( u0 , v0 , w0 )
and ( u , v, w ) are any two adjacent equilibrium configurations. Substituting into Eqs. 164ac and
neglectinghigherordertermsoneobtains
N x1, x + N xy1, y = 0 166a
N xy1, x + N y1, y = 0 166b
D 4 w1 ( N x 0 w1, xx + 2 N xy 0 w1, xy + N y 0 w1, yy ) = 0 166c
where
(
N x 0 = C u0, x + vv 0,y ) N x1 = C ( u1, x + vv1,y )
N y0 = C (v 0,y + vu0, y ) (
N y1 = C v1,y + vu1, y ) 167
(1 v ) C (1 v ) C
N xy 0 =
2
(u 0, y + v0,x ) N xy1 =
2
(u
1, y + v1,x )
Equations166acarethestabilityequationsfortheplatesubjectedtoinplaneloading.NoticethatEq.
166cisuncoupledfromEquations166aand166b.Thisuncouplinggreatlysimplifiestheanalysisof
particularcasesinthefollowingsections.Thecorrespondingequationsfortheshellsarecoupled.
16.2.3 Applications for the Stability Equations

16.2.3.1 Overview
Equation166cappliesforallpossibleinplaneedgeloads.Inthemostgeneralcasesthecoefficients
N x 0 , N xy 0 and N y 0 are functions of the coordinate variables x and y. The following applications,
however,arelimitedtocasesinwhichthesecoefficientsareconstant.
16.2.3.2 Plate Simply Supported on Four Edges

As a first example we consider a flat plate simply supported on four edges and subjected to an in
planecompressiveload Px uniformlydistributedalongtheedgesx=0andx=a,asshowninFigure
161.Fromanequilibriumanalysisoftheplate
Px
N x0 = ; N xy 0 = N y 0 = 0 168
b
IntroductionintoEq.166csimplifiesthatexpressiontotheform
Px
D 4 w1 + w1, xx = 0 169
b
Tosimplifythenotation,thesubscripts1areomittedfromtheincrementalquantities ( w1 ,etc)inthe
remainderofSection16.2.3.
292
ECSSEHB3224A
24March2010
Thesimplesupportboundaryconditionsmaybewritten
w = w, xx = 0 on x = 0 and x = a
1610
w = w, yy = 0 on y = 0 and y = b
Equation169isaconstantcoefficientequation.Asolutionoftheform
m x n y
w = C1 sin sin m, n = 1, 2, 3,.... 1611
a b
whereC1isaconstant,isseentosatisfyboththedifferentialequationandtheboundaryconditions.
IntroductionintoEq.169andregroupingyields
m 4 m n n Px m
2 2 4 2
D + 2 + = 0 1612
a a b b b a
ThediscretevaluesofPxforwhichEq.169hasnontrivialsolutionsthereforeare
2
Px a m n
2 2 2
= D +
1613
b m a b
Thecriticalloadcorrespondstothesmallesteigenvalue.Forallvaluesofa,bthesmallesteigenvalueis
givenbyn=1.Accordingly
2
Px a m 1
2 2 2
= +
m = 1, 2, 3,...
D 1614
b m a b
Equation1614canbeexpressedintheform
D
Px = kc 2 1615
b
where
2
mb a
kc = + 1616
a mb
Thecoefficientkcisseentobeafunctionoftheaspectratioa/bandthewavelengthparameterm.Fora
givena/b,thevalueofmcanbechosenbytrialtoyieldthesmallesteigenvalue.Theresultsareshown
ascurveCinFigure163.AsketchoftheplateinbentequilibriumconfigurationisshowninFigure
164.
293
ECSSEHB3224A
24March2010

Figure163:Influenceofboundaryconditionsonthebucklingcoefficientsof
platessubjectedtoinplanecompressiveloading(from[2])
Figure164:Buckledshapeofplatesubjectedtoinplanecompressiveloading
16.2.3.3 Other Boundary Conditions

ThesimpleformofthesolutioninEq.1611isnotsuitableforotherboundaryconditions.Asamore
generalexample,letusconsidertheloadPxappliedtoaplatethatissimplysupportedontheloaded
edgesx=0andx=abuthasother,asyetunspecifiedboundaryconditionsontheunloadededgesy=0
andy=b.
294
ECSSEHB3224A
24March2010
Thesmallestcriticalloadforsuchboundaryconditionsevidentlycorrespondstofreeedgesony=0and
y=b.SuchaplateissimplyacolumnwhosebendingstiffnessEIisreplacedbyDb.ThenfromEulers
equation
EI 2 2 Db
Px = m 2 2 2
onegets Px = m 1617
a a2
Suchaplateiscalledawidecolumn.Thesmallesteigenvaluesareseentocorrespondtom=1.Inthe
more general case, the differential equation 169 and the boundary conditions on x=0 and x=a are
satisfiedbysolutionsoftheform
m x
w = f ( y ) sin m = 1, 2, 3,... 1618
a
IntroductionofEq.1618intoEq.169reducesthelattertotheordinarydifferentialequation
m d f m Px m
2 4 2
d4 f 2
2 + f =0 1619
dy 4 a dy
2
a Db a
Thisisaconstantcoefficientequationwhichalwaysadmitsexponentialsolutions.Thecharacteristic
equationassociatedwithEq.1619isseentobe
m 2 m Px m
2 4 2
2
4
+ =0 1620
a a Db a
TherootsofEq.1620are
1
m m Px 2
= 1621
a a Db
ButfromEq.1617
1
Px 2 m
= forawidecolumn
Db a
and
1
Px 2 m
> forallotherboundaryconditionsony=0andy=b
Db a
Consequently,forallothercases,therootsofEq.1620canbewritten
= , - ,
& , -
& 1622
where and arerealandpositiveandaregivenby

1
m 2 m Px 2
= +
a b Db
1623
1
m m 2
Px 2
= +
a b Db
Inthisnotation,thesolutionofEq.1619canbewritten
295
ECSSEHB3224A
24March2010
f = C1e y + C2 e y + C3 cos y + C4 sin y 1624
whereC1,C2,C3,C4areconstantstobedeterminedfromthefourboundaryconditionsony=0andy=b.
Asaparticularexampleconsideraplatethatissimplysupportedontheedgey=0andfreeony=b.The
equationsfortheboundaryconditionony=0arew=My=0yielding
w = w, yy = 0 1625

Ony=btheequationsfortheboundaryconditionsare Vy = Qy + M yx = 0 and M y = 0 yielding
x
w, yyy + ( 2 v ) w, xxy = 0 and w, yy + vw, xx = 0 1626
UsingEquations1624and1625oneobtains
C3 = 0 and C1 = C2 1627
ThenEq.1624canbewritteninthemoreconvenientform
f = A sinh y + B sin y 1628
whereAandBarenewconstants.IntroductionintoEquations1626gives
2 2 2
m 2 2 2
m
( 2 v ) 2 cosh b A + ( 2 v ) 2 cos b B = 0
a a
1629
2 vm 2 2 2 vm 2 2

2 sinh b A
+
2 sin b B = 0
a a
ForanontrivialsolutionthedeterminantofthecoefficientsofAandBinthishomogeneousequation
system is equal to zero. Setting the determinant equal to zero and rearrangement yields an implicit
Pb
expression for the nondimensional load parameter x in terms of the wavelength parameter m,
D
Poissonsratio v andaspectratioa/b.Calculationsshowthatforallvaluesofa/btheminimumload
occursform=1.Resultscanbeexpressedintheform
2
D
Px = kc 1630
b
wherekcisanondimensionalbucklingcoefficientforcompressiveloading.Theresultsareshownas
curveEinFigure163.
16.2.3.4 Shear Loading

ForaplatesubjectedtoauniformlydistributedshearloadingasshowninFigure165,Equation166c
reducesto
4
D w 2 N xy 0 w, xy = 0 1631
This expression, like that for uniform compression loading in Eq. 169, is a constantcoefficient
equation. Its simple appearance is deceptive, however. One term of the equation contains even
orderedderivativeswithrespecttoeachofthecoordinatevariables,andtheothertermoddordered
derivatives.Consequently,adeflectionfunctionoftheformofthatinEq.1611isNOTasolution.
296
ECSSEHB3224A
24March2010

Figure165:Platesubjectedtoinplaneshearloading
ExactsolutionsofEq.1631areavailableonlyforthecaseofaninfinitelylongstrip.Suchasolutionis
givenbySouthwellandSkan[3].Theyassumedasolutionoftheform
w = f ( y ) eikx / b 1632
wherekisalongitudinalwavelengthparameterandbistheplatewidth.IntroductionintoEq.1631
andrearrangementgivesanordinarydifferentialequation
2
d 2 f k 2 ik df
2 f 2 N xy 0 =0 1633
dy b Db dy
Ifthefunction f ( y ) isthenassumedtobeoftheform
f ( y ) = C ei y / b 1634
where C isaconstant,thensubstitutionintoEq.1633yieldsthefollowingcharacteristicequation
k
4 + 2k 2 2 + 2 N xy 0 + k4 = 0 1635
D
Foreachvalueofk,thisequationhasfourroots,whichmaybedesignated 1 , 2 , 3 , 4 .Thenthe
solutionoftheordinarydifferentialequation1633canbewritten
f = C1ei1 y / b + C2 ei2 y / b + C4 ei3 y / b + C4 ei4 y / b 1636
andthesolutionforwisoftheform
( )
w = C1ei1 y / b + C2 ei2 y / b + C4 ei3 y / b + C4 ei4 y / b eikx / b 1637
where C1, C2, C3, C4 are arbitrary constants. Introduction of this expression for w into four
homogeneousboundaryconditionequationsleadstoadeterminantalequationfromwhichthecritical
values of Nxy0 may be determinedasin Section16.2.3.3.SouthwellandSkan[3] carried out such an
analysisforsimplysupportededges(w=w,yy=0ony=0,b)andforclampededges(w=w,y=0ony=0,b).
Theirresultscanbeexpressedintheform
2D
N xyo = ks 1638
b2
where ks is a nondimensional shear buckling coefficient. In this notation their results for simply
supported and clamped edges, respectively, are ks=5.35 and ks=8.98. Corresponding solutions for
infinitelylongstripswithelasticallyrestrainededgesarealsoavailable(seeRef.[2]).
297
ECSSEHB3224A
24March2010
Forplatesoffinitedimensions,availablesolutionsofEq.1631utilizenumericalmethods.Resultsof
suchanalysesforplateswithsimplysupportededgesandwithclampededgesareshowninFigure
166(whichisFig.22ofRef.[2])

Figure166:Criticalvaluesofshearstressforplatessubjectedtoinplaneshear
loading(from[2])
16.2.3.5 Combined Loading

Consideraplatesubjectedtoinplanecompressiveloadingintwodirections,asshowninFigure167.

Figure167:Platesubjectedtoinplanecompressionintwodirections
Fromanequilibriumanalysisoftheplateintheflatform
Px Py
N x0 = ; N xy 0 = 0; N y0 = 1639
b a
IntroductionintoEq.166cyields
298
ECSSEHB3224A
24March2010
Px P
D 4 w + w, xx + y w, yy = 0 1640
b a
Equation 1640 is a homogeneous equation, but there are two independent load parameters. The
equationcanbechangedtoasingleparameterequationsimplybyletting
Py Px
=R 1641
a b
where R is a nondimensional constant. The resulting equation then can be solved for a series of
selectedvaluesofR.IntroductionintoEq.1640andrearranginggivesforthestabilityequation
Px
D 4 w +
b
( w, xx + Rw, yy ) = 0 1642
Forsimplicity,weagaintreatonlysimplysupportededges;thentheboundaryconditionsare
w = w, xx = 0 on x = 0 and x = a
w = w, yy = 0 on y = 0 and y = b
andsolutionsthatsatisfytheboundaryconditionsareoftheform
m x n y
w = C1 sin sin m, n = 1, 2, 3,... 1643
a b
Introducingthisexpressionintothestabilityequation1642andrearrangementgives
2D
Px = kcc 1644
b
where
2
( mb / a )2 + n 2
kcc = 1645
( mb / a ) + Rn2
2
ForgivenvaluesoftheloadratioRandplateaspectratioa/b,thevaluesofmandncanbechosenby
trialtogivethesmallesteigenvaluePcr.
Forasquareplate,forexample,a/b=1.ThenforR=1,0,1,respectively,Eq.1645givesthevalues
R m n kcc
1 1 1 2
0 1 1 4
1 2 1 8.33
Negative values of R of course signify tensile loading in the ydirection. As would be expected on
intuitivegrounds,theadditionofatensileloadinthetransversedirectionisseentohaveastabilizing
influence.
Results of stability analyses for combined loading frequently are presented in terms of socalled
interactioncurves.SuchcurvesforthepresentanalysisareshowninFigure168forthecasea/b=1.
Thecoordinatesinthegraphare x / xcr and y / ycr ,where
Px Py
x = ; y = 1646
bh ah
299
ECSSEHB3224A
24March2010
and xcr and ycr arethecriticalvaluesof x and y ,wheneachisactingalone.FromEqs.1641

and1646
y
R= 1647
x
Thesolidlinesintheinteractionplotrepresentminimumeigenvaluesandthedashedonesrepresent
higher eigenvalues. Negative values of the parameter y / ycr represent tensile loading in the y
direction. Interaction curves for an extensive variety of combinations of compression, bending, and
shearloadingaregiveninRef.[2]ingraphicalform.

Figure168:Interactioncurvesforplatesubjectedtoinplanecompressionintwo
directions
16.2.4 Stability Equations for Orthotropic Plates

The isotropic plate equations are readily generalized to apply to orthotropic construction as well.
Examplesofplatewallconstructionthatcanbetreatedasorthotropicincludecorrugatedsheets,fibre
reinforcedplasticplates,andplateswithcloselyspacedstiffeners.Fororthotropicconstruction,only
the constitutive equations in the foregoing analysis of isotropic plates need to be replaced; the
kinematicandequilibriumrelationshipsareindependentoftheconstitutionoftheplatematerial.Let
usconsideraplatereinforcedbycloselyspacedstiffenersthatareparalleltotheedgesoftheplate.For
simplicity,thestiffenersareassumedtobesymmetricalrelativetotheplatemiddleplane,asshownin
Figure169;i.e.theinfluenceofstiffenereccentricityisnotconsidered.(Foramoregeneralanalysisof
orthotropic construction that takes into account the influence of stiffener eccentricity see Reference
[4].)
300
ECSSEHB3224A
24March2010

Figure169:Integrallystiffenedplate
For symmetrical stiffeners the isotropic constitutive relations of Eqs. 163 are replaced by the more
generalrelations
N x = C11 x + C12 y M x = C44 x + C45 y
N y = C12 x + C22 y M y = C45 x + C55 y 1648
N xy = C33 xy M xy = C66 xy
where the Cijs are constants called the stiffness parameters (see Eq. 72). From Figure 169 the
extensionalstiffnessparametersC11andC22foruniformlyspacedstiffenersmadeofthesamematerial
astheplatecanbeseentobe
EAx EAy
C11 = + C22 = + 1649
dx dy
Eh
where = =extensionalstiffnessoftheplateitself
2
1 v
Ax , AY =crosssectionalareasofstiffenersparalleltothexandydirections,respectively
d x , d y =correspondingstiffenerspacings.
Because the stiffeners are discrete, there is no Poisson effect between them, and C12 = v as in the
isotropiccase.Thecorrespondingbendingandtwistingstiffnessparametersareseentobe
EI x EI y
C44 = D + C55 = D +
dx dy
1650
1 GJ x GJ y
C45 = vD C66 = (1 v ) D + +
2 dx dy

3
Eh
where D =
( )
=bendingstiffnessoftheplateitself
2
12 1 v
Ix, I y =stiffenermomentofinertiarelativetotheplatemiddleplane
G =shearmodulus
Jx, J y =torsionalconstant
Foraplatestiffenedinonly,say,thexdirection, Ay = I y = J y = 0 .
301
ECSSEHB3224A
24March2010
Substituting the orthotropic constitutive equations 1648 into the outofplane equilibrium equation
161cyields
C44 , xx + C45 ( x , yy + y , xx ) + 2C66 xy , xy + C55 y , yy

1651
( N x w, xx + 2 N xy w, xy + N y w, yy ) = p
IntroducingthekinematicrelationsfromEqs.162,thenlinearizationusingtheadjacentequilibrium
criterionnowyields
C44 w1, xxxx + 2 ( C45 + C66 ) w1, xxyy + C55 w1, yyyy

( N x 0 w1, xx + 2 N xy 0 w1, xy + N y 0 w1, yy ) = 0
1652
As an example let us consider on orthotropic plate that is simply supported on all four edges and
subjected to a uniformly distributed compressive load Px = N x 0b . Then N xy 0 = N y 0 = 0 and Eq.
1652simplifiestotheexpression
Px
C44 w1, xxxx + 2 ( C45 + C66 ) w1, xxyy + C55 w1, yyyy + w1, xx = 0 1653
b
Thisisaconstantcoefficientequation.Asolutionoftheform
m x n y
w1 = C1 sin sin m, n = 1, 2, 3,... 1654
a b
isagainseentosatisfythedifferentialequationandtheboundaryconditionsidentically.Introduction
ofthisexpressionintoEq.1653andrearrangementyieldstheexpression
Px a
2
m 4 2
m n
2
n
4

= C44 + 2 ( C45 + C66 ) + C55
b m a a b b
1655

where m and n are positive integers. Again, the smallest values of Px is seen to correspond to n=1.
Accordingly
Px a
2
m 4 m
2
1
4

= 44 C + 2 ( 45 C + C66 )
+ C55 1656
b m a ab b
As with isotropic plates, the values of the wavelength parameter m is chosen by trial to give the
smallestvalueofPx.Forisotropiccase,Eq.1656reducestoEq.1613.
Introducingthebucklingstressparameter
x x hb 2 x hb 2
= =
2 2 C55 2 D22
thebucklingstressforuniaxialloadingforvariousstiffnessratiosC44/C55=D11/D22isshowninFigure
1610.whilethebucklingstressforbiaxialloadingforvariousstiffnessratiosisshowninFigure1611.
Extensivenumericalresultsfororthotropicplatessubjectedtoinplanecompressionandshearloading
are givenin graphical form in Ref. [6]. Also included are resultsfor plates stiffened by one, two, or
threediscretestiffeners,andacomprehensivelistofreferences.
302
ECSSEHB3224A
24March2010
Figure1610:Bucklingstressofsimplysupportedorthotropicrectangularplate
underuniaxialloadingforvariousstiffnessratios D11 / D22 = C44 / C55 (from[5]).
Figure1611:Bucklingstressofsimplysupportedorthotropicrectangularplate
underbiaxialloadingforvariousstiffnessratios D11 / D22 = C44 / C55 (from[5]).
16.2.5 Stability Equations for Sandwich Plates

Sandwichplatesconsistoftwothinloadbearingsheetsseparatedbyalightweightlowstiffnesscore,
asillustratedinFigure1612.Thefacesheetsusuallyaremadeofmetaloroffibrereinforcedplastic.
Common core materials include foamed plastic, lightweight metallic honeycomb, and lightweight
corrugated sheet. The core serves to increase the overall bending resistance of the composite cross
sectionbykeepingtheloadbearingareasatarelativelylargedistancefromthemiddleplaneofthe
sandwich plate. The present analysis of sandwich plates follows in some respect the analysis of
orthotropicsandwichplatesinRef.[7].Onlyisotropicconstructionistreatedhere,forsimplicity.
303
ECSSEHB3224A
24March2010

Figure1612:Sandwichplate
One of the fundamental approximations of thinplate theory is that lines normal to the undeformed
middle surface of the plate remain straight and normal during deformation. Because of the low
stiffness of the lightweight core material, this approximation cannot be retained in sandwichplate
analysis;transverseshearingstrainsaretoolargetobeneglected.
The necessary modifications are introduced in the analysis as follows. In sandwich construction,
rotationsand curvatures are caused not only by themoment intensities M x , M y , M xy , but also by
the transverse shearing force intensities Qx , Q y . For isotropic sandwiches the rotations xQ , yQ
dueto Qx and Qy respectively,are(seeRef.[1])
Qx Qy
xQ = ; yQ = 1657
CQ CQ
where CQ , in Newton per mm, is a transverse shearing stiffness parameter analogous to the
extensionalstiffnessparameter .Themagnitudeof CQ canbedeterminedexperimentally,oritcan
becalculatedintermsoftheelasticconstantsEand v ofthefaceandcorematerial(seeRef.[8],for
example).
Theassociatedcurvatures xQ , yQ andaveragetwist xyQ correspondingto xQ , yQ are
xQ =
Qx , x
; yQ =
Qy , y
; xyQ =
(Q x, y + Qy , x )
1658
CQ CQ 2CQ
FromEqs.162thetotalcurvaturesandtwistaregivenbytherelations
x = w, xx ; y = w, yy ; xy = w, xy
Then the corresponding curvatures and twist x , M , yM , xyM due to the moments alone are
obtainedbysubtractingfromthetotalcurvaturesandtwisttheportionsduetoshear,asfollows:
Qx , x Qy , y
xM = w, xx ; yM = w, yy ;
CQ CQ

1659
Q + Qy , x
xyM = w, xy x , y
2 C
Q
304
ECSSEHB3224A
24March2010
Consequently, for sandwich construction, the constitutive relations for moments in Eqs. 163 can be
replacedbythefollowingexpressions
Q Qy , y
M x = D w, xx x , x + v w, yy
CQ CQ
Q Qx , x
M y = D w, yy y , y + v w, xx 1660

CQ CQ

1
M xy = (1 v ) D w, xy
2CQ
( Qx , y + Qy , x )

For CQ these equations reduce to the earlier expressions for homogeneous plates.
Correspondingconstitutive relationsfororthotropicsandwichplatesaregiveninRef.[7],Eqs.12.For
theisotropiccase,thepresentequationsandthoseinRef.[7]arethesame.
The kinematic and equilibrium relations in Eqs. 162 and 161 remain applicable for sandwich
construction.Those equations plus theconstitutive relations in Eqs.1660lead to stabilityequations
analogoustoEqs.166,inwhichthedisplacementcomponent w1 isuncoupledfromthecomponents
u1 , v1 .Theequationgoverning w1 canbewritteninthefinalform
D 2
D 4 w1 1 ( N x 0 w1, xx + 2 N xy 0 w1, xy + N y 0 w1, yy ) = 0 1661
C
Q
For CQ the equation reduces to the corresponding expression for homogeneous plates in Eq.
166c. As an example we consider a sandwich plate whose faces are subjected to a uniformly
distributedload Px = N x 0b ,asshowninFigure1612.ThenEq.1661specializestotheform
Px D 2
D 4 w1 + 1 w1, xx = 0 1662
b C
Q
Equation 1662 is a constant coefficient equation. If the plate is simply supported on all four edges,
thentheboundaryconditionsare
w = w, xx = 0 on x = 0, a
1663
w = w, yy = 0 on y = 0, b
andasolutionoftheform
m x n y
w = C1 sin sin m, n = 1, 2, 3,... 1664
a b
where C1 is a constant is seen to satisfy both the differential equation and the boundary conditions.
Introducingthisexpressionintothestabilityequation1662andrearrangementyields
2
D ( m / a ) + ( n / b )
2 2
2
Px a
= 1665
b m 1 + ( D / CQ ) ( m / a ) + ( n / b )2
2 2

wherem,narepositiveintegers.Again,theminimumPxisseentocorrespondton=1.introductionof
thatvalueandrearrangementgives
305
ECSSEHB3224A
24March2010
2 D ( mb / a + a / mb )
2
Px = 1666
b 1 + r 1 + ( mb / a )2

whererisadimensionlessshearstiffnessparameterdefinedbytherelation
2D
r= 1667
b 2CQ
Equations1666canbewritteninthealternativeform
2D
Px = K 1668
b
wherethenondimensionalbucklingcoefficientKisdefinedbytheexpression
( mb / a + a / mb )
2
K= 1669
1 + r 1 + ( mb / a )
2

For given values of the plate aspect ratio and the shear stiffness parameter r, the value of the
wavelength parameter m can be chosen by trial to give the smallest value of Px. For r , this
equationreducestothecorrespondingexpressionforhomogeneousplatesinEq.1616.
In sandwich construction it is commonly assumed that the bending stiffness of the core and the
bendingstiffnessoftheindividualfacesrelativetotheirowncentroidscanbeneglected.Then,forthe
2
sandwichplatecrosssections,themomentofintertiaperunitwidthisseentobe 2h f hc + h f ( ) ,
2

where h f and hc denotethefaceandcorethicknesses,respectively(seealsoFigure1612).Thenthe
bendingstiffnessparameterDforthesandwichplatebecomes
E f h f ( hc + h f )
2
D= 1670
2
where E f is Youngs modulus for the face materials; the influence of the anticlastic bending is
neglected.Forahomogeneouscore,furthermore
CQ = Gc hc 1671
whereGcistheshearingmodulusofthecorematerial.
16.2.5.1 Effect of Transverse Shear deformation

In analyzing laminated plates with classical laminated plate theory, the Kirchhoff hypothesis is
assumed. This assumption neglects the transverse shear deformations, an effect that becomes
importantastheplatethicknessincreasesrelativetotheotherdimensionsoftheplate.Inthiscasethe
flexibilityaddedtotheplane,duetotheshear deformations,canbeofmoreimportanceinlaminated
platesthaninisotropicones.
InFigure1613theresultsofastudy(Whitney[9])ontheinfluenceofsheardeformationsonbuckling
of a ( 45) angleply graphiteepoxy symmetrical laminate, uniaxially loaded and having all edges
simply supported, are compared with those corresponding to classical plate theory. This figure
depicts the considerable decrease in buckling stress due to shear deformations in relatively thick
plates.
306
ECSSEHB3224A
24March2010

a)Influenceofplatethicknessratioonbucklingsimplysupportededges
b)Comparisonofexactelasticity,classicalplatetheoryandsheardeformationtheory
solutions
Figure1613:Sheardeformationeffectsforan45oangleplysquareplate[9]
307
ECSSEHB3224A
24March2010
16.2.6 Failure of Plates
The equilibrium paths for an inially perfect plate subjected to inplane compression are shown in
Figure 1614. Also included in Figure 1614 are corresponding curves for a slightly imperfect plate
[10].Twoimportantresultsareimmediatelyapparentfromthediagrams:
a. Buckling of real (and therefore imperfect) plates is so gradual that it is difficult to decide at
preciselywhatloadthebucklingcanbesaidtooccur.Thereforecomparisonsoftheoreticaland
experimentalvaluesforplatecriticalloadsentailanelementofarbitrariness.
b. Inanycase,theplatecontinuestoacceptadditionalloadafterbuckling.ThereforetheloadPcr
fortheplate,unlikethatforthecolumn,doesnotrepresentitsultimatestrength.
Theseconclusions,basedonFigure1614forinplanecompression,applyforotherkindsofinplane
loadingaswell.

Figure1614:Equilibriumpathsforinitiallyperfectandimperfectplatessubjected
subjectedtoinplanecompression
Thefactthatplateswithsupportededgescanacceptadditionalloadafterbucklingwasdiscoveredin
the late 1920s through experimental studies made in connection with the structural design of
airplanes.In1929Wagner[11]establishedacriterionforthepostbucklingstrengthofashearweb,i.e.
athinplatesupportedonallfouredgesandsubjectedtoinplaneshearloading(seeSection16.2.3.4).
The prebuckling state of stress in a shear web consists of compressive stresses on diagonal lines
orientedat45ototheplateedges,andtensilestressesatrightanglestothecompressivestresses.The
compressivestressescausebucklestoforminthewebalongdiagonallines.Beforebucklingthetensile
andcompressivestressesareequalinmagnitude,butafterbucklingthetensilestressesarelarger.The
inbalanceinforcesiscarriedbytheedgesupports.Wagnerassumedinhisapproximateanalysisthat
forathinwebthecompressivestressescouldbeneglectedentirelyinthebuckledconfiguration.Such
models are sometimes called diagonal tension beams. A more refined method of analysis for shear
webs and extensive test results were reported by Kuhn [12]. For efficient design the postbuckling
strengthofplatesshouldbetakenintoaccount.Asanexampleoftheprocedure,letusconsideraplate
subjected to a uniformly distributed inplane compressive load Px, as shown in Figure 1615a. The
appliedloadPxisrelatedtothestress x bytheequation
b
Px = h x dy 1672
0
whereh andbareplatethicknessandwidth,respectively.For Px Pcr ,thestressisuniformacross

theplatewidth,asillustratedbythelines11and22inFigure1615b.Then Px = hb x .Inparticular,
for Px = Pcr
308
ECSSEHB3224A
24March2010
Pcr = hb cr 1673
For Px > Pcr , on the other hand, the stress near the plate edges y = 0,b is larger than that near the
center,becauseofthestabilizinginfluenceoftheedgesupports.Insuchcasesthestressdistributionis
nonuniform,asillustratedbuthecurves33and44inFigure1615b.
Figure1615:Stressdistributioninplatebeforeandafterbuckling
For configurations within the scope of the intermediate class of deformations, the postbuckling
distributionofstresscanbedeterminedbynumericalsolutionofthenonlinearequilibriumequations
164.Fordesignpurposesitisconvenienttoexpresstheresultsoftheanalysisintermsofaneffective
width over which the stress is considered to be uniform, as illustrated in Figure 1615c. From the
figure
Px = hbeff max 1674
where max is the maximum stress at the plate edges y = 0,b. A widely used approximation
expressionfor beff is(seeRef.[10],Eq.7)
309
ECSSEHB3224A
24March2010
1
2
beff = b cr 1675
max
where cr istheclassicalcriticalstressforthegivenboundaryconditions.Equation1675iscalledthe
von Krmn effectivewidth formula. For max = cr , 2 cr and 4 cr , Equations 1673, 1674 and
1675give Px = Pcr , 1.4 Pcr and 2.0 Pcr .
Themaximumallowablestress max attheplateedgesisgovernedbytheyieldstrengthoftheplate
material or by the buckling strength of the members supporting the plate at its edges. The value of
cr inEq.1675isgivenbyEq.1673,where Pcr isdeterminedfromEq.1615orEq.1630,depending
ontheplateboundaryconditions.
Similar design procedures have been developed for plates subjected to inplane bending or shear
loading. A comprehensive discussion of the ultimate strength of plates in bending, in shear, and in
combinedbendingandshearisgiveninRef.[13],Chapter5.
Occasionally, the initial buckling load Pcr itself is used as a conservative estimate of the ultimate
strengthoftheplate.
Ifyieldingofthematerialoccursbeforetheplatebuckles,thecriticalloadofcourseissmallerthanthe
value given by an elastic stability analysis. Because the state of stress in biaxial, inelastic stability
analysisismorecomplicatedforplatesthanforcolumns.Adiscussionoftheinfluenceofplasticityon
thebucklingandpostbucklingbehaviourofplatesisgiveninReference[14].
16.3 Curved Panels
16.3.1 Introduction
While plate structures possess excellent capabilities in resisting inplane forces, their performance
whenloadedintheflexuraldirectionisgenerallymuchlower.Examplesofthisphenomenonappear
intherelativelypoorbendingresistanceandthebucklingsensitivitywhensubjectedtocompressive
stresses.Apossiblesolutiontoincreasethelateralrigidityofplatestructuresistoprovidethemwith
stiffeningelements,suchasthestringersandframesthatarepresentinaircraftfuselagestructures.
The thinwalled plate segments in a stiffened panel are still prone to buckling phenomena when
subjectedtoaxialcompressiveloads,shearingloads,orcombinationsthereof.Generally,however,the
structureisstillabletocarryadditionalloadingbeyondthepointofinitialbucklingoftheindividual
skinsegments,whichareboundbythestringersandframes(Figure1616illustrateslocalbuckling).
Structuresconformingtothisparticularfeaturearesaidtopossesspostbucklingstrengthbeforeactual
failure of the structure takes place. If, however, the reinforcing elements are too weak to effectively
supporttheskin,theytendtobucklealongwiththeskin.Thisphenomenoniscalledoverallbuckling
andisessentiallyafailuremodeofthestructure.
The calculation of the failure load of a stiffened panels as far as stability aspects are concerned
generallycomprisesthefollowingsteps:
a. Determination of the initial buckling load (also referred to as the critical load), which is
generally done by means of linear (bifurcation) buckling analysis. The deformed shape of the
panelatthecriticalbucklingloadthecriticalmodeiseithercharacterisedbywavepatterns
havingnodallinesatthelocationofthestringersandframes,orbybucklingwavesthatextend
310
ECSSEHB3224A
24March2010
overthestringersand/orframes.Thefirsttypeofdeformationmodeiscalledlocalbucklingand
is thus confined to the individual skin segments, the second type is called overall or global
buckling and shows flexural displacements of the stiffening elements. While in the latter case
thestructurelosesitsloadcarryingcapacity(whichisequaltofailure),thestructuregenerally
still possesses strength beyond the critical load when the buckling mode is local. This post
bucklingstrengthisobtainedfromthestiffeningelements,whichtakeoverthe(majority)ofthe
additionalappliedloadafterinitialbuckling.
b. Postbuckling stress analysis in case the condition of local buckling is fulfilled. Since the skin
segments tend to lose (most of) their loadcarrying capability in the postcritical range, the
reinforcedelementswilltakeoverthemajorityoftheadditionalloading.
c. Checkingofthepostbucklingstressesintheindividualpanelelementsagainstfailurecriteria
orallowables.

Figure1616:Illustrationoflocalbucklinginafuselagepanelloadedinpureshear
Thissectionislogicallydividedintothesubgroupscompressionloadingandshearloading.Foreach
loadingtype,thecorrespondingdesignmethodologyforpanelswillbeoutlinedconcisely.Ingeneral,
theabovesteps2and3arecombinedbecauseofthesemiempiricalnatureofmostpracticaldesign
methodologies.Therefore,foreachrespectiveloadingsituationthediscussionoftheanalysismethod
willbedividedintoapartcoveringinitialbuckling,andapartdealingwithpostbucklingandfailure
ofthepanel.
16.3.2 Compression loading

16.3.2.1 Overview
Fortheskinstringercombination(stiffenedpanel)undercompressionseveraltypesofinstabilitycan
occur, including Euler buckling (flexural instability), torsional buckling, local skin buckling, local
stringerbuckling,interrivetbucklingandwrinkling(forcedcrippling),seeforexample[15].InFigure
1617bucklingmodesofopensectionstiffenedpanelsareshown.Ingeneral,bucklingoftheskindoes
notdirectlyresultinpanelfailure.Thepanelisabletocarryadditionalloaduptothestressatwhich
the stringer begins to fail. As the load is increased after the stress in the skin has reached the skin
buckling stress, the skin adjacent to the stringer will carry additional stress because of the support
given by thestringers. Topredict the failure loadofstiffened panelsloaded beyond initial buckling
semiempirical methods have been used in aircraft industry, such as the EulerJohnson method in
combination with the effective width concept. In this approach, the panel failure is assumed to
correspond to that of a thinwalled column consisting of the stringer and an effective width of the
skin.
311
ECSSEHB3224A
24March2010

Figure1617:Bucklingmodesofopensectionstiffenedpanel(fromWilliamsand
Stein[16])
16.3.2.2 Critical Buckling Load

Variousanalysismethods,rangingfromclosedformsolutionstosophisticatedlargecomputercodes,
areavailable topredicttheinitialbucklingloadsofflatandcurvedpanelsundercompressionand/or
shearloading.
To obtain the buckling load of a curved panel (see Figure 1618) loaded in axial compression and
supported along the two generators and along the two circular edges normal to the axis of the
cylindricalpanelTimoshenko[17]assumedthefollowingexpressionsforthebucklingdisplacement

m x n
u = A cos sin
L
m x n
v = B sin cos 1676
L
m x n
w = C sin sin
L

Figure1618:Axiallycompressedcurvedpanel
312
ECSSEHB3224A
24March2010
Itcanbeseenthattheradialdisplacementswandthebendingmomentsbecomezeroalongtheedges
ofthecurvedpanelat x = 0 and x = L andfor = 0 and = ,asisrequiredforsimplysupported
edges.
Substituting expressions 1676 into the three stability equations in terms of u, v and w one obtains
three homogeneous linear equations. The equations for determining the critical value of the
compressivestressisobtainedbyequatingtozerothedeterminantoftheseequationsyielding
( N x )cr E
cr = = 1677
h 1 v2
where
2 2
( )
2
c h 1 m + n m2
ch
= + 2
= cmn 1678
3 R 2

m2
(
m2 + n2 3 R
)
and
2
Rh 2 2 2 2 Rh 1
m2 = m 2 ; n = 2 n ; n = n
2c L
2
; c = 3 1 v
2c R
2
( )
Ifthecentralangle isnotsmallandthelengthLisofthesameorderofmagnitudeas R ,thenone
canexpectthecurvedpaneltobuckleintoalargenumberofcircumferentialandlongitudinalwaves.
Thesmallestvalueofthecriticalbucklingstressoccurswhen
(
2 2
)
2
1 m + n m2
= 1
"c = + 1679
2
( )
2
mn
m2 m2 + n2

yielding
E ch E h E h
cr = = 3 (1 v 2 ) =
1 v 3 R 3 (1 v )

3 (1 v 2 ) R
2 2 1680
R

whichisthecriticalstressforaxisymmetricbucklingofathincircularcylindricalshell.
Iftheangle isverysmall,theconditionsofbucklingofthecurvedpanelwillapproachthoseofa
longitudinally compressed rectangular plate. The critical value of the compressive stress is obtained
bytaking n = 1 inEq.1678.
AstheradiusRofthecurvedpanelbecomeslargerandlargeronecanneglectthesecondterminEq.
1678yielding

( ) = c h
2
2
c h 1 m + n
2 2
n2
= m + 1681
3 R 2 m2 6 R m

Thesmallestvalueofthisexpressionoccurswhen
L
= R and n = 1 1682
m
313
ECSSEHB3224A
24March2010
thatis,whenthelengthofthelongitudinalhalfwavesisequaltothewidthofthecurvedpanel.Then
Eq.1681reducesto
2
h2 m 2 h2
= 2 = 1683
12 L 3 ( R )2
FinallyfromEq.1677onegets
2 Eh 2
cr =
3 (1 v 2 ) ( R )
2 1684

ThisisthesamevalueasthecriticalstressthatweobtainedforlongrectangularplatesinEq.1615.
Further,Timoshenko[17]hasshownthatif
R 2 h2
R 2 4
12 (1 v 2 )
1685

that is, if the circumferential dimension of the curved panel is at least equal to twice the halfwave
lengthforaxisymmetricbucklingoftheshell,thenoneobtainsforthecriticalstressthevaluegivenby
Eq.1680.
This indicates that for any value of the width R of the curved panel, which satisfies the condition
specifiedbyEq.1685,onecanfindfromEq.1679for n = 1 alengthL/moflongitudinalhalfwaves
suchthatthecriticalstressbecomesequaltothecriticalstressforaxisymmetricbuckling.Thisvalue
shouldbeusedforthedesignofcurvedpanelsuniformlycompressedalongthegenerators.
IfthecircumferentialdimensionofthecurvedpanelissmallerthatwhatisrequiredbyEq.1685then
Eq.1679becomesaminimumif
2
R 2
m = 2 1686
L

This means that for a considerable length L such a narrow curved panel, like a long narrow
compressedplate,willsubdivideduringbucklingintosquares.FromEq.1679themagnitudeofthe
criticalstressinsuchcaseis
2 Eh 2 E 2
cr = +
3 (1 v 2 ) ( R )
1687
2
4 2

Thefirsttermontherightsidegivesthestresscalculatedasforaflatplateandthesecondtermgives
theincreaseofthecriticalstressduetothecurvatureoftheplate.
16.3.3 Curved Panels under Shear

The problem of buckling of curved sheet panels under pure shear is of practical interest in aircraft
structures.Usingthesamegeneralmethodsasintheprecedingarticles,thecriticalvaluesofshearing
stress cr at which bucklingbegins werecalculated for long and wide curved panels [18], [19]. The
resultsarepresentedinFigure1619andFigure1620.
314
ECSSEHB3224A
24March2010
16.3.4 Curved Panels under Combined Loading
The problem of combined shear and axial compression was solved by Kromm [20]. The results
obtainedaregivenbycurvesinFigure1621,inwhichabcissasarevaluesof
b 4
4
= 12 (1 v 2 ) 1688
R

where R is the radius of the curved panel, h is its thickness and b = R is the width of the curved
panel. Each curve corresponds to a definite value of the ratio of the compressive stress x to the
criticalbucklingstress cr calculatedfromEq.1680.Negativevaluesoftheratio x / cr ,indicated
inthefigure,correspondtoaxialtension.Thecurve x / cr = 0 correspondstothecaseofpureshear,
discussedinSection16.3.3.Itisassumedinthiscalculationthatthelongitudinaledgesofthecurved
panelaresimplysupportedandthecircumferentialnormalstress vanishesalongtheseedges.
Thepointsofintersectionofthecurveswiththehorizontalaxes 4 (seeEq.1688)givethevaluesof
forwhichbucklingisproducedbythecompressivestresses x actingalone ( = 0) .Thepointsof
intersectionofthecurveswiththeverticalaxesgivethecriticalshearingstress cr foralongflatstrip
forvariousamountsofthelongitudinaltensilestress x .

Figure1619:Shearbucklingcoefficientforlongsimplysupportedcurvedplates
(from[19]).
315
ECSSEHB3224A
24March2010

Figure1620:Shearbucklingcoefficientforwidesimplysupportedcurvedplates
(from[19]).

Figure1621:Curvedpanelsundercombinedshearloadandaxialload(from[17]).
16.4 References
[1] Brush,DonO.andAlmroth,BOO.,BucklingofBars,PlatesandShells,ISBN0070327580,
McGrawHill,Inc.,NewYork,1975
[2] Gerard,G.,andH.Becker,HandbookofStructuralStabilityPartI,BucklingofFlatPlates,
NACATN3781,NationalAdvisoryCommitteeforAeronautics,July1957.
316
ECSSEHB3224A
24March2010
[3] Southwell,R.V.andSkan,S.W.,OntheStabilityunderShearingForcesofaFlatElasticStrip,
Proc.Roy.Soc.London,ser.A,vol.105,pp.582607,1924
[4] Baruch,M.andSinger,J.,EffectofEccentricityofStiffenersontheGeneralInstabilityof
StiffenedCylindricalShellsunderHydrostaticPressure,J.Mech.Eng.Sci.,Vol.5,pp.2327,
1963
[5] Leissa,A.W.,BucklingofLaminatedCompositePlatesandShellPanels,ReportAFWALTR
853069,AirForceWrightAeronauticalLaboratories,June1985
[6] Becker,H.,HandbookofStructuralStabilityPartII,BucklingofCompositeElements,
NACATN3782,NationalAdvisoryCommitteeforAeronautics,1957
[7] Libove,C.andBatdorf,S.B.,AGeneralSmallDeflectionTheoryforFlatSandwichPlates,
NACAReport899,1948
[8] Libove,C.andHubka,R.E.,ElasticConstantsforCorrugatedCoreSandwichPlates,NACA
TM2289,1951
[9] Whitney,J.M.,TheEffectofTransverseShearDeformationontheBendingofLaminated
Plates,JournalofCompositeMaterials,pp.534547,July1969
[10] VanderNeut,A.,PostbucklingBehaviorofStructures,NATOAGARDReport60,1956
[11] Wagner,H.,EbeneBlechwandtrgermitsehrdnnemStegblech,Z.Flugtech.
Motorluftschiffahrt,Vol.20,pp.200,227,256,279und306,1929
[12] Kuhn,P.,InvestigationsontheIncompleteDevelopedPlaneDiagonalTension,NACAReport
697,1940
[13] Johnston,B.G.(ED.),GuidetoDesignCriteriaforMetalCompressionMembers,2nded.,
ColumnResearchCouncil,Wiley,NewYork,1966
[14] Hutchinson,J.W.,PlasticBuckling,in:AdvancesinAppliedMechanics,Vol.14,Editedby
C.S.Yih,AcademicPress,NewYork,1974
[15] Niu,M.C.,Airframestressanalysisandsizing,Hongkong:ConmilitPress,1997
[16] Williams,J.G.andStein,M.,BucklingBehaviorandStructuralEfficiencyofOpenSection
StiffenedCompositeCompressionPanels,AIAAJournal,Vol.14,No.11,pp.16181626,1976
[17] Timoshenko,S.P.andGere,J.M.,TheoryofElasticStability,McGrawHillBookCompany,
Inc.,NewYorkTorontoLondon,1961
[18] Batdorf,S.B.,Schildcrout,MurryandStein,Manuel,CriticalShearStressofLongPlateswith
TransverseCurvature,NACATN1346,1947
[19] Batdorf,S.B.,Stein,ManuelandSchildcrout,Murry,CriticalShearStressofCurved
RectangularPanels,NACATN1348,1947
[20] Kromm,A.,JahrbuchdeutscherLuftfahrtForschung,p.1832,1940
317
ECSSEHB3224A
24March2010
17
Closed shells
17.1 Introduction
Thinwalled shells are frequently used structural elements in such diverse applications as cooling
towers, legs of offshore bore islands, aircraft fuselages or as the main load carrying elements of
aerospace launch vehicles. The popularity of shells is due to the fact that they are efficient load
carryingstructures.However,unfortunately,oftentheyarepronetocatastrophicelasticinstabilities.
Thusathoroughunderstandingofthestabilitybehaviourofthinwalledshellsisaprerequisiteforall
thosewhoemploythem.
17.2 Circular cylindrical shells

Circular cylindrical shells are treated separately because their stability equations are much simpler
thanthoseofshellsofgeneralshape,andthuscanbeusedveryconvenientlytoillustratethedifferent
types of instabilities that may occur. In the present analysis the relatively simple Donnell type shell
theory will be employed. These equations give accurate results for cylindrical shells whose
displacement components in the deformed configuration are rapidly varying functions of the
circumferentialcoordinate.ForthesignconventionusedseeFigure171.
Figure171:Circularcylindricalshellsymbolsandsignconvention.
318
ECSSEHB3224A
24March2010
In the age of computerized shell stability the interest in using the Donnell type shell equations has
practically disappeared. However, their relative simplicity makes them ideally suited for rapid
approximate analytical developments and hence also for the following introductory analytical
examinationofshellstability.
TheDonnellequationsarebasedonthefollowingmiddlesurfacekinematicrelations
1
x = u,x + 2x x = w,x x = x,x
2
w 1
y = u,y + + 2y y = w,y y = y,y 171
R 2
1
xy = u,y + v,x + x y xy = ( + y,x )
2 x,y
Comparing the circumferential rotation y with the one used for the circular ring subjected to
uniformexternalpressure
v
= w,y 172
R
oneseesthataccordingtoDonnellsapproximationthecircumferentialdisplacementcomponent v is
neglected relative to the gradient of the normal displacement component in the circumferential
direction w,y .
Employing the stationary potential energy criterion, the following set of nonlinear governing
equationsarederivedin[1]forisotropiccircularcylindricalshells.
Nx,x + Nxy,y = 0 173
Nxy,x + Ny,y = 0 174
1
Mx,xx + (Mxy + Myx ),xy + My,yy Ny + Nx w,xx +2Nxy w,xy +Ny w,yy = p 175
R
Threeequationsinthreevariables,thedisplacementsu,v,wmaybeobtainedbyintroductionofthe
isotropicconstitutiveequations
Nx = C( x + y ) Mx = D( x + y )
Ny = C( y + x ) My = D( y + x )
176
(1 ) Mxy + Myx
Nxy = C xy = D(1 ) xy
2 2
andthekinematicrelationsfromequation171intoequations173,174and175.Theextensionaland
thebendingstiffnessparametersare,respectively, C = Eh /(1 2 ) and D = Eh3 /12(1 2 ) .
Asimplersetoftwoequationsintwovariableswandfcanbederivedasfollows.Noticethatifone
definesanAirystressfunctionfsuchthat
Nx = f,yy , Ny = f ,xx , Nxy = f,xy 177
thentheinplaneequilibriumequations173and174areidenticallysatisfied.Theremainingoutof
planeequilibriumequation175andthecompatibilityequation
1
x,yy + y,xx xy,xy = w,2xy w,xx w,yy + w,xx 178
R
yielduponsubstitutionandregrouping
319
ECSSEHB3224A
24March2010
1
D 4 w + f, (f,yy w,xx 2f,xy w,xy + f,xx w,yy ) = p 179
R xx
1
4 f Eh(w,2xy w,xx w,yy + w, ) = 0 1710
R xx
Where
4 ( ) = ( ),xxxx +2( ),xxyy +( ),yyyy 1711
TheseequationswerefirstpresentedbyDonnellasthreeequationsin1933[2].
When talking about buckling of thinwalled shells one distinguishes between collapse at the
maximumpointofaloaddeflectioncurveandbifurcationbuckling.Thusifoneemploysthegeneral
nonlinearanalysisgovernedbyequations1791711,theaxiallycompressedperfectisotropicshell
initially deforms axisymmetrically along the path OA (see Figure 143) until a maximum (or limit)
load A is reached at point A. However, in this case there exist many bifurcation points along the
fundamentalpathbetweenOandA.Hence,oncethelowestbifurcationload c isreached,theinitial
failure of the perfect structure will be characterized by a rapidly growing asymmetric deformation
alongthepathBDwithadecreasingaxialload .Noticethatinthiscase,the(axisymmetric)collapse
loadoftheperfectstructure A isofnoengineeringsignificance.
Thelinearizedstabilityequationsforthedeterminationofthecriticalload c atthebifurcationpoint
canbederivedbytheapplicationoftheadjacentequilibriumcriterion.Toinvestigatetheexistenceof
adjacentequilibriumconfigurationsoneassumesthatthetwovariablesw,faregivenby

w = w 0 + w, f = f0 + f 1712
where w 0 , f0 represent the prebuckling solutions along the fundamental path and w, f represent
smallperturbationsatbuckling.Directsubstitutionoftheseexpressionsintoequations179 and1710
and deletion of squares and products of the perturbation quantities, yields a set of nonlinear
governing equations for the prebuckling quantities w 0 , f0 which are identical in form to equations
1791711,andasetoflinearizedstabilityequationsgoverningtheperturbationquantities w, f
1
D 4 w
+ LNL (w 0 , f) = 0
f, L (f , w) 1713
R xx NL 0
Eh
4f xx +EhLNL (w 0 , w)
w, =0 1714
R
where
LNL (S, T) = S,yy T,xx 2S,xy T,xy +S,xx T,yy 1715
17.2.1 Level-1 solutions for isotropic shells

17.2.1.1 Overview
Usingequations1791711toevaluatetheprebucklingequilibriumstate,inthefollowingthecritical
buckling load for various loading conditions will be calculated with the help of the appropriate
linearizedstabilityequations.
17.2.1.2 Axial Compression

Firstconsiderthestabilityofacylindricalshellthatissimplysupportedatitsendsandsubjectedtoa
uniformly distributed axial compressive load P. Under this loading the prebuckling deformation of
320
ECSSEHB3224A
24March2010
theshellisaxisymmetricasshowninFigure172.Thecriticalload Pc isthelowestaxialloadatwhich
theaxisymmetricequilibriumstateceasestobestable.
Assumingthattheshellissufficientlylongsothattheeffectofbendingoftheshellwallclosetothe
ends can be neglected, then the prebuckling state can be approximated by the following membrane
state
P Eh2
Nx0 = = , Ny0 = Nxy0 = 0, w0 = constant 1716
2R cR
where
= Nx / Ncl ,with Ncl = Eh2 / cR and c = 3 (1 2 ) 1717
Figure172:Axiallycompressedcylinder.
ThustheaxisymmetricforminFigure172aineffectisreplacedbythatinFigure172b.Noticethat
this membrane state satisfies the nonlinear governing equations of the prebuckling path, equations
1791711, identically and reduces the linearized stability equations, equations17131715 to the
followingsetofconstantcoefficientequations
1 Eh2
D 4 w
+ f,xx + xx = 0
w, 1718
R cR
Eh
4f xx = 0
w, 1719
R
xx = 0 at x = 0,L theseequations
= w,
ForconstantDandsimplysupportedboundaryconditions w
admitseparablesolutionsoftheform
x y x y
w = A sin m cos n , f = B sin m cos n 1720
L R L R
leadingtoastandardeigenvalueproblemwiththeeigenvalues
2)2
1 (m + n
2 2
m
c,mn = + 1721
2 2 2 + 2)2
m ( m n
andtheeigenfunctions
2
x y Eh3 m x y
W = h sinm cos n , f = sinm cos n 1722
L R 2c (2 + 2)2 L R
m n
321
ECSSEHB3224A
24March2010
where
2 2
2 Rh Rh 1
m = m2 ; n2 = n2 1723
2c L 2c R
Noticethattheeigenvalues c,mn dependnotonlyonthegeometricparametersbutalsoontheaxial

andcircumferentialwavenumbersmandn.
For cylinders of intermediate length, a close estimate of the smallest eigenvalue may be obtained
2
directlybyanalyticalminimizationof c,mn withrespecttothequantity mn = (m + n2 )2 / m
2
in
equation1721.Differentiationleadstotheresultthat c,mn isaminimumfor
2 + 2 )2
( m
mn = n = 1
2 1724
m
Thus all mode shapes which satisfy equation 1724 have the same (lowest) eigenvalue of c = 1 .
Regroupingequation1724onegetsthewellknownKoitercircle[3]
2
m + n2 m = 0 1725
whichisthelocusofafamilyofmodesbelongingtothelowesteigenvalue
c
c = = 1 1726
cl
Noticethatequation1726isnormalizedby
Eh E h
c l = =
cR R 1727
3(1 2 )
thecriticalbucklingstressforaxiallycompressedcircularcylindricalshells,derivedshortlyafterthe
turnofthecenturyindependentlyfromeachotherbyLorenz[4],Timoshenko[5]andSouthwell[6].
For short cylinders, because m and n are integers the analytical minimization to arrive at equation
1724 is inadmissible. In such cases equation 1721 is evaluated repeatedly for different values of m
andninatrialanderrorproceduretodeterminethecriticalload.Ifthecylinderissoshortthat
2
R R
L > 2c h 1728

thenduringbucklingonlyahalfwaveintheaxialdirectionwillbeformedandthesmallestvalueof
equation1721isobtainedfor n = 0 .Thus

2
1 Rh 1
c = c = c,m0 = +
2
1729
cl 2 2c L Rh
2c L

Bytakingthelengthofthecylindershorterandshorter,thesecondterminequation1729becomes
smallerandsmallerincomparisonwiththefirstterm.Thus,byneglectingitoneobtains
2
c Rh
c = = 1730
cl 4c L
or
322
ECSSEHB3224A
24March2010
2
2E h
c = 1731
2
12(1 ) L
whichisEulersformulaforawidecolumn,i.e.aflatplatethatissimplysupportedattheloaded
edgesandfreealongtheunloadededges.
A very long cylinder can buckle as an Euler column with undeformed crosssection (m = n = 1) . The
Donnell formulation used does not yield the correct result for this case as can be seen from Figure
173. Comparing these results with the values displayed in Figure 173, which are based on Loves
theory (equation. (i) on p. 464 of [7]), one sees that Donnells approach also yields somewhat
inaccurate results for moderately long cylinders. Notice that the results of Loves theory show the
properlimitingbehaviourforverylongshells.
TheEulerbucklingloadofverylongthinwalledcylinderscanbeobtainedbysetting I = R 3h and
A = 2 Rh intheappropriatecolumnequationyielding
2
P EI E R
c = c = 2 = 2 1732
A AL2 2L
Figure173:ComparisonofbucklingloadpredictionsusingDonnellandLove
typetheories(fromSinger,ArboczandWeller[8])
17.2.1.3 Combined External Pressure and Axial Compression

If the shell is simply supported at its ends then under the simultaneous action of uniform lateral
pressureandaxialcompressiontheprebucklingdeformationoftheshellisaxisymmetricasshownin
Figure174.Thecriticalpressure p c isdefinedasthelowestpressureatwhichtheaxisymmetricform
losesitsstability.
Againitisassumed,forsimplicity,thattheshellissufficientlylongsothattheprebucklingstatecan
beapproximatedbythefollowingmembranestate
Eh2 Eh2
Nx0 = ; Ny0 = peR = pe ; Nxy0 = 0 ; w0 = constant 1733
cR cR
Noticethatthus,ineffect,theaxisymmetricforminFigure174aisreplacedbythatinFigure174b.It
canbeverifiedthatthismembranestatesatisfiesthenonlineargoverningequationsoftheprebuckling
path, equations 1791711, identically (whereby p = pe ) and reduces the linearized stability
equations17131715tothefollowingsetofconstantcoefficientequations
323
ECSSEHB3224A
24March2010
1 Eh2 Eh2
D 4 w
+ f,xx +pe yy +
w, xx = 0
w, 1734
R cR cR
Eh
4f xx = 0
w, 1735
R
Figure174:Cylindersubjectedtouniformexternallateralpressure
Theboundaryconditionsandtheseparablesolutionsarethesameasfortheprecedingexample.The
useofequations1720leadstoastandardeigenvalueproblemwitheigenvalues
2 + 2 )2 2 2
1 (m n m
= + pe n 1736
2 m2 2 + 2 )2
( m 2
m
n
Asingleparametereigenvaluecanbeobtainedintroducingtherelation
= Rpe 1737
where R is a nondimensional constant. Notice that if R = 0 the pressure acts only on the lateral
surface,whereasif R = 1 / 2 thenthepressurecontributesalsotoaxialcompressionthroughtheend
plates, forming the socalled hydrostatic pressure case. With the help of this expression the
eigenvaluescanbewrittenas
2
m (2 + 2 )2 m2
pc,mn = m n +
2 + 2 )2
1738
2 + 2 )
2(R 2 (
m n m m n
The eigenfunctions are the same as for the preceding example (see equations 1722). Considering
equation1738,adistincteigenvaluecorrespondstoeachpairofvaluesmandnanditisseenthatthe
smallest eigenvalue corresponds in every case to m = 1 . For particular values of L/R and R/h, the n
correspondingtothesmallesteigenvaluecanbedeterminedbytrialanderror.
Numericalresultsbasedonequation1738areshowninFigure175.Fromthesecurves,calculatedfor
differentR/hratios,itisseenthatforshortertubesthecriticalexternalpressure p c increasesrapidly
as the ratio L/R decreases. On the other hand for long tubes, for L/R>50 say, the critical external
pressuredoesnotdependonthelength.Itsvaluecanalsobededucedfromthe criticalpressurefora
ring subjected to external fluid pressure, as follows. Recalling that the compressive force per unit
length Ny acting on the elemental ring of unit width is equal to p c R , then the critical pressure
becomes
EI
Nyc = pcR = 3 1739
R2
IfonenowreplacesEby E /(1 2 ) andsets I = h3 /12 ,thenequation1739yields
324
ECSSEHB3224A
24March2010
3
E h
pc = 1740
2
4(1 ) R
thecriticalbucklingpressureforlongtubessubjectedtouniformexternalpressure.
It also becomes apparent from the results displayed in Figure 175 that for n = 4 or less there is a
noticeabledifferencebetweenthepredictionsofequation1738,whichisbasedon Donnellstheory,
andtheresultsoftheLovetheoryof[7](equation(d)onp.496).For n = 2 ,aswellknown,theDonnell
valuesareabout33percenttoohigh.
Figure175:Bucklingdiagramsforuniformexternallateralpressure(fromSinger,
ArboczandWeller[8]).
17.2.1.4 Combined Torsion and Axial Compression

Assuming, for simplicity, that the shell is sufficiently long, then the prebuckling solution under the
simultaneous action of axial compression and torsion can be approximated by the following
membranestate
Eh2 Mt Eh2
Nx0 = ; Ny0 = 0 ; Nxy0 = = ; w 0 = constant 1741
cR 2R 2 cR
where Mt istheappliedtorsionalmoment.
Directsubstitutionshowsthatthismembranestatesatisfiesthenonlineargoverningequationsofthe
prebuckling path, equations 1791711 identically and that the linearized stability equations,
325
ECSSEHB3224A
24March2010
equations17131715,arereducedtothefollowingsetofconstantcoefficientequations
1 Eh2 Eh2
D 4 w
+ f,xx 2 xy +
w, xx = 0
w, 1742
R cR cR
Eh
4f xx = 0
w, 1743
R
Notice that these equations differ markedly from the previously derived stability equations (see
equations 17181719 and 17341735) in that in the outofplane equilibrium equation one
encountersbothoddandevenderivativesof w withrespecttothesameindependentvariable.This
indicatesthatonecannolongersatisfythestabilityequationsbyusingseparablesolutionsintheform
of simple products of sines and cosines. Physically this means that there are no generators which
remainstraightduringbucklingandwhichformasystemofstraightnodallinesforabuckledsurface.
Under torsional loading the buckling deformation of a cylindrical shell consists of a number of
circumferentialwavesthatspiralaroundthecylinderfromoneendtotheother.Ifonenowassumes
thatthebucklingmodeisrepresentedby
N x y N x y
w = h Cmn sin m sin n + h Dmn sin m cos n 1744
m =1 L R m =1 L R
xx = 0 at x = 0,L ,thenan
= w,
anexpressionthatsatisfiessimplysupportedboundaryconditions w
approximatesolutionofthelinearizedstabilityequationscanbeobtainedasfollows.
First, the compatibility equation 1735 is solved exactly for the stress function f in terms of the
assumedradialdisplacement w .Sinceitisassumedthattheshellissufficientlylongsothattheeffect
ofbendingoftheshellwall closetotheendscanbeneglected,onlyaparticularsolutionofequation
1735needstobeconsidered.Secondly,theequationofequilibrium1734issolvedapproximatelyby
substituting therein f and w and thenapplying Galerkins procedure. Carrying out the steps yield
for a given number of circumferential full waves n the following homogeneous system of two
simultaneousalgebraicequations
1
M C + NmjD jn = 0 ,m=1,2,3,N 1745
m mn
1
NmjC jn + M D =0 1746
m mn
where

Mm = m 1747
4 c,mn
N N
jm jm
NmjD jn = 2 2
j m D jn

; NmjC jn = 2 2
j m C jn
1748
j =1 m j j =1 m j
Nxy xy
= = 1749
h c l cl
2 + 2 )2 3
1 (m n m
m
c,mn = + 1750
4 mn 2 2 2
n (m + n ) 2
n
= 1 if j m = odd integer
jm 1751
= 0 otherwise
326
ECSSEHB3224A
24March2010
Using matrix notation equations 17451747 can be put into the form of a standard eigenvalue
problem
A B X% = 0 1752
whichcanbesolvedroutinelyonadigitalcomputer.Sincethestructurebucklesattheloweststressat
whichinstabilitycanoccur,foragivenshell isminimizedwithrespecttothecircumferentialwave
numbern.Thisisdonebytruncatingthedeterminantofthecoefficientsofequation1752andfinding
thelowesteigenvaluebymatrixiteration.Thesizeofthedeterminantisincreaseduntiltheeigenvalue
convergestothedesiredaccuracy(say,fivesignificantfigures).
Results for R / h = 1000 and different L/R ratios are displayed in Figure 176. As can be seen, for
shortershellsthecriticalnormalizedtorqueparameter c increasesrapidlyastheratioL/Rdecreases.
Notice also, that by taking the radius of the cylinder larger and larger, while keeping its length
constant,thelowerboundfestooncurvefor c approachesthecriticalshearloadofaninfinitelylong
stripwithsimplysupportededgesobtainedbySouthwellandSkan[9]
2D
c = 5.35 1753
L2h
whereagain D = Eh3 /12(1 2 ) .
Limiting results for large values of L/R, when the shell will buckle with two full waves in the
circumferentialdirection,havebeenderived,forexample,in[1]usingDonnellstheoryyielding
3/2
E h
c = 0.272 R 1754
2
(1 ) 3 / 4
andin[7]usingaLovetypetheoryyielding
3/2
E h
c = 0.236 1755
(1 2 )3 / 4 R
Onceagain,asnotedearlier,for n = 2 Donnellsequationsareinaccurate.
AlsoshowninFigure176aresolutionsbasedonthefollowingbucklingmode
x y
w = hCmn sin m n 1756
L R
where Cmn isaconstantandm,nareintegers.Equation1756satisfiestherequirementofperiodicity

inthecircumferentialcoordinate,butdoesnotsatisfyanyofthecommonlyusedboundaryconditions
at the cylinder ends. Consequently, this simple expression may only be used for sufficiently long
cylinders,whoseendconditionshavelittleinfluenceonthemagnitudeofthecriticalload.
327
ECSSEHB3224A
24March2010
Figure176:Bucklingdiagramsforcylinderssubjectedtotorsion(fromSinger,
ArboczandWeller[8])
Using the Donnell type theory and proceeding as outlined earlier one obtains a particularly simple
solution of the linearized stability equations 17421743 with the following expression for the
eigenvalues
2 + 2 )2 3
1 (m n m
m
c,mn = +
2 + 2 )2 2
1757
4 mn ( n
n m n
where c,mn = xy / cl . Notice that the eigenvalues of c,mn depend not only on the geometric
parameters and the specified axial load = / cl , but also on the axial and circumferential wave
numbersmandn.
Timoshenkousedexpressionssimilartotheonegivenbyequation1756tosolveLovetypestability
equationsin[7].Hissolutioncurveagreeswellfor n 4 fullwavesinthecircumferentialdirection
withtheonebasedonDonnellsstabilityequations.However,forlargevaluesofL/Rwhentheshells
bucklewithtwofullwavesinthecircumferentialdirection,onecanobservethewellknownfactthat
Donnells equations yield about 10 percent higher values than solutions based on the more accurate
Lovetypetheory.
ThefirstinvestigationofbucklingofcylindricalshellsundertorsionisduetoE.Schwerin[10].Fora
completereviewofthetorsionproblemtheinterestedreadershouldconsultYamakisbook[11].
17.2.1.5 Combined Bending and Axial Compression

Ifacylindricalshellisrelativelyshortandtheshelledgesareheldcircular,thenthecircumferential
flatteningofthecylindercrosssectioncausedbythebendingmomentcanbeneglected.Inthiscase
328
ECSSEHB3224A
24March2010
theprebucklingstateunderanexternalloadconsistingofcombinedbendingandaxialcompression
canbeapproximatedquiteaccuratelybythefollowingmembranestate
Eh2 y
Nx0 = R a + R b cos R ; Ny0 = Nxy0 = 0 ; w 0 = constant 1758
cR
where
= N0 = Nb P M0 Eh2
R a ; R b ; N0 = ; Nb = ; Ncl = ; c = 3(1 2 ) 1759
Ncl Ncl 2R R 2 cR
Notice that this membrane state does not satisfy rigorously the nonlinear equations governing the
prebucklingstate,equations1791711.However,becauseofitssimplicityithasbeenwidelyusedin
the literature (see [12], [13] and [14]) to obtain approximate solutions. The linearized stability
equations1713and1714thenreducetothefollowingsetofvariablecoefficientequations
1 Eh2 y
D 4 w
+ f,xx + xx = 0
R a + R b cos R w, 1760
R cR
Eh
4f ,xx = 0
w 1761
R
Ifonenowassumesthatthebucklingmodeisrepresentedby
x N y
w = h sin m Cmn cos n 1762
L n =1 R

xx = 0 at x = 0,L ,then an
= w,
anexpressionthatsatisfiessimply supportedboundaryconditions w
approximatesolutionofthelinearizedstabilityequationscanbeobtainedasfollows.
First the compatibility equation 1761 is solved exactly for the stress function f in terms of the
assumedradialdisplacement w .Hereitisassumedthattheeffectofbendingoftheshellwallcloseto
theendscanbeneglected.Thusonlyaparticularsolutionofequation1761needstobeconsidered.
Secondly,theequationofequilibrium1760issolvedapproximatelybysubstitutingtherein f and w
andthenapplyingGalerkinsprocedure.Carryingoutthestepsyieldsforagivennumberofaxialhalf
wavesmthefollowinghomogeneoussystemofalgebraicequations
1
( c,mn R a ) Cmn R b (1 + 1n 0n ) Cm,n 1 + (1 0n ) Cm,n +1 = 0 1763
2
with
n = 1,2,3,...N
where
2 + 2 )2 2
1 (m n m
c,mn = +
2 + 2 )
1764
2 2 (
m m n
2 2
2 = m2 Rh Rh 1
m ; n2 = n2 ; c = 3(1 2 ) 1765
2c L 2c R
and 1n , 0n , 0N areKroneckerdeltas.
Usingmatrixnotationequation1763canbeputintothefollowingform
329
ECSSEHB3224A
24March2010
1 C
c,m0 R a R
2 b m0

1 1 Cm1
2 Rb c,m1 R a R
2 b

1 1 C
R c,m2 R R
2 b a
2 b m2 = 0 1766

..... ..... ..... .....
1 1
R
c,mN 1 R R .....
2 b a
2 b

1
R b
2
c,mN R a
CmN
ThistridiagonalmatrixeigenvalueproblemcanbesolvedveryconvenientlybyarecursiveGaussian
eliminationschemeoriginallyderivedbyPotters[15]andusedlaterextensivelybytheHarvardgroup
underBudiansky[16].Eitherthenormalizedaxialloadparameter R a ( = ) orthenormalizedbending
moment parameter R b can be chosen as the eigenvalue, whereby in each case the other load
parameterhasa specifiedfixedvalue.Sincetheshellbucklesattheloweststressatwhichinstability
canoccur,theeigenvaluechosenisminimizedwithrespecttotheaxialhalfwavenumberm.Thisis
done by truncating the size of the matrix in equation 1766 for a given value of m and finding the
lowest eigenvalue by matrix iteration. The size of the determinant is increased until the eigenvalue
convergestothedesiredaccuracy(say,fivesignificantfigures).
Numerical results for R / h = 100 and different ratios are displayed in Figure 177 for pure bending
= 0) .Noticethatforshortershellsthenormalizedbendingstressratio R increasesrapidlyasthe
(R a b
ratioL/Rdecreases.Further,whereasforshortershells (L / R < 0.5, say)forcertainL/Rratios R b may
varynoticeably,forlongershells (L / R > 1.0, say)thecriticalnormalizedbendingstressratio R b can
besetequaltothelowerofthefestooncurvesofabout1.014.
Thus for pure bending the maximum critical bending stress is only slightly higher than the critical
stressforaxialcompressiononly.
Here it should be mentioned that the statement made in [7] on p. 483 about the maximum critical
stressforbendingaloneis1.3timesthecriticalstressforpurecompressiondoesnotholdingeneral.It
is only true for the particular set of geometric and material properties used by Flgge [13] for his
habilitationpaper.
Finally to check whether it is safe to neglect the effect of ovalization of the circular crosssection
caused by the applied bending moment one can use the results of [17] here reproduced in part in
Figure178.TheauthorsofthispaperusedthetwodimensionalfinitedifferencecodeSTAGS[18]to
calculatethecollapsebendingmomentwhiletakingtheeffectsofboundaryconditionsandgeometric
nonlinearities in the prebuckling state into account. As can be seen from Figure 178 for shells of
moderatelength (L / R < 3, say)theresultsobtainedwithequation1766arequiteaccurate.
330
ECSSEHB3224A
24March2010
Figure177:Variationofbucklingstressratiowithcylinderlength.
Figure178:Comparisonofcollapsemomentsofcylindersunderpurebending
with classicalresults(from[17]).
17.2.2 Level-1 solutions for orthotropic shells

17.2.2.1 Overview
Orthotropic cylindrical shells are widely used in structural applications. Examples of shellwall
construction that can be treated as orthotropic include cylindrical sheets stiffened by closely spaced
circular rings and/or longitudinal stringers, fibrereinforced shells and corrugatedskin construction.
For a shellwall construction that is not symmetrical relative to the shell middle surface, there is a
coupling between extensional forces and curvature change and between bending moments and
extensional strain. To account for this coupling effect the generalized constitutive equations for
anisotropicshellsofRef.[20]areused,whichread
331
ECSSEHB3224A
24March2010
N A A12 A16 B B12 B16

x 11 x 11 x
Ny = A12 A 22 A 26 y + B12 B 22 B26 y 1767

Nxy A16 A 26 A 66 xy B16 B 26 B66 xy
M B B12 B16 D D12 D16

x 11 x 11 x
My = B12 B22 B 26 y + D12 D22 D26 y 1768

Mxy B16 B26 B 66 xy D16 D26 D66 xy
The corresponding sign convention is shown in Figure 171. In shorthand notation these equations
canbewrittenas
N = A + B 1769
M = B + D 1770
Invertingthefirstconstitutiveequation1769oneobtains
= A * N + B * 1771
where
A * = A 1 B * = A 1 B 1772

Substitutingtheinvertedfirstconstitutiveequation1771intothesecondconstitutiveequation1770
yields
M = C * N + D * 1773
where
1
C * = B A * = B A 1774
1
D * = D B A B 1775
Noticethatsince A issymmetric,thereforealso A * issymmetric.Furthersince B issymmetric,

T
therefore C * = B * .Finallysince D issymmetric,thereforealso D * issymmetric.
Theequilibriumandcompatibilityconditionsareindependentofthemateriallaw,thereforethesetof
nonlineargoverningequationsforanisotropic/orthotropiccircularcylindricalshellsarealsogivenby
equations173,174and175.
A simpler set of two equations in the two variables w and f can be derived by introducing an Airy
stressfunctionfsuchthat
Nx = f,yy , Ny = f,xx , Nxy = f,xy 1776
thentheinplaneequilibriumequations173and174areidenticallysatisfied.Theremainingoutof
plane equilibrium equation 175 and the compatibility equation 178 yield upon substituting the
appropriate form of the anisotropic constitutive equations (equations 1773) into the outofplane
equilibrium equation 175 and equations 1771 into the compatibility equation 178 and regrouping
thefollowingtwononlineargoverningequations
1
LB* (f) + LD* (w) = f, +L (f, w) + p 1777
R xx NL
1 1
L A* (f) LB* (w) = w,xx LNL (w, w) 1778
R 2
332
ECSSEHB3224A
24March2010
where
L A* ( ) = A22 ( ),xxxx 2A26 ( ),xxxy +(2A12

+ A ) ( ),
66

xxyy 2A16 ( ),xyyy + A11 ( ),yyyy 1779
LB* ( ) = B21 ( ),xxxx +(2B26 B61 ) ( ),xxxy +(B11

+ B22 2B*66 ) ( ),xxyy
1780
B )( ),
+ (2B16
62 xyyy +B12 ( ),yyyy
( ),
LD* ( ) = D11 xxxx +4D16 ( ),xxxy +2(D12 + 2DA66 ) ( ),xxyy +4D26 ( ),xyyy +D22 ( ),yyyy 1781
andthefollowingnonlinearoperator
LNL (S, T) = S,xx T,yy 2S,xy T,xy +S,yy T,xx 1782
Smearedstiffenertheory,inwhichthestiffenersaresmeared,ordistributed,overthewholeshell
inamannerthattakesintoaccounttheeccentricityofstiffeners,hasbeenfoundtobeasatisfactory
approachforcloselystiffenedshellsthatfailbygeneralinstability(see[21]and[22]).
Asanillustrationletusconsiderathinwalledcylindricalshell,reinforcedbycloselyspacedcircular
ringsattachedtotheinsideoftheshellskinandwithlongitudinalstringersattachedtotheoutside,as
showninFigure179.
Figure179:Shellwallconstructionforstiffenedcylinder.
Ifthestiffenersandtheskinaremadeofthesamematerial,thesemiinvertedstiffnessparametersare
given by the following expressions expressed in terms of parameters introduced by Baruch [23] in
1964
1 = 1 Hxx = Hxx
A22 = (1 2 )(1 + s ) 1783
Eh Eh

+ A =
2A12 66
1 + 1 + = 1 Hxy = Hxy
2(1 2 ) 1784
Eh (1 2 ) Eh

= 1 = 1 Hyy = Hyy
A11 (1 2 ) (1 + r ) 1785
Eh Eh
h
B21 =
s = Q = Q xx 1786
2c xx
2B26 B61 = 0 1787

B11 (1 + r ) s + (1 + s )r = h Q xy = Q xy
+ B22 2B66 = 1788
2c
B = 0
2B16 1789
62
r = h
B12 = Q = Q yy 1790
2c yy
333
ECSSEHB3224A
24March2010
= D 1 +
D11 (1 + r ) s s = D D xx = D xx
0s 1791
=0
D16 1792

+ 2D ) = D 2 + + +
2(D12 ( s r + r s ) = D D xy = D xy
66 ts tr 1793
D26 = 0 1794
D22 = D 1 + 0r
(1 + s )r r = D D yy = D yy 1795

Baruchs[23]parametersaredefinedasfollows
As Ar A
s = (1 2 ) r = (1 2 ) s = (1 2 ) s es
ds h dr h ds h
Ar EA s EAr
r = (1 2 ) e s = e r = e
dr h r dsD s dr D r
1796
EI EI GJ
0s = s 0r = r ts = s
dsD dr D dsD
GJr Eh3 = 1
tr = D=
dr D 12(1 2 ) (1 + s ) (1 + r ) 2
Thesubscriptssandrrefertostringersandrings,respectively.
Is and Ir aremomentsofinertiaofthestiffenercrosssectionsrelativetotheshellmiddlesurface; Js
and Jr are the torsion constants of the stiffener crosssections; and the eccentricities e s and er
representthedistancefromtheshellmiddlesurfacetothecentroidofthestiffenercrosssection.Itis
assumed that the effects of moments of inertia about the zaxis, warping and the product of inertia
maybeneglected.Thecouplingparameters s , r and s , r arepositiveforoutsidestiffenersand
negativeforinsidestiffeners.
Thusfororthotropicshellsthenonlineargoverningequations1777and1778become
1
L Q (f ) + LD (w) = f, +L (f, w) + p 1797
R xx NL
1 1
LH (f) L Q (w) = w, L (w, w) 1798
R xx 2 NL
where
LH ( ) = Hxx ( ),xxxx +Hxy ( ),xxyy +Hyy ( ),yyyy 1799
L Q ( ) = Q xx ( ),xxxx +Qxy ( ),xxyy +Qyy ( ),yyyy 17100
LD ( ) = Dxx ( ),xxxx +Dxy ( ),xxyy +Dyy ( ),yyyy 17101
LNL (S, T) = S,xx T,yy 2S,xy T,xy +S,yy T,xx 17102
To derive the linearized stability equations for the determination of the critical load c at the
bifurcationpointonecanusetheadjacentequilibriumcriterion.Assumingthatthetwovariables w,f
aregivenby

w = wo + w , f = fo + f 17103
334
ECSSEHB3224A
24March2010
where w o , fo represent the prebuckling solutions along the fundamental path and w, f represent
small perturbations at buckling. Direct substitution of these expressions into equations 1797 and
1798 and deletion of squares and products of the perturbation quantities, yields a set of nonlinear
governing equations for the prebuckling quantities w o , fo which are identical in form to equations
1797and1798andasetoflinearizedstabilityequationsgoverningtheperturbationquantities
+ L (w) 1 w )
L Q (f) D = f,xx +LNL (fo , w)
+ LNL (f, o 17104
R
L (w 1
LH (f) Q ) = w,

L (w , w) 17105
R xx NL o
Using equations 1797 and 1798 to evaluate the prebuckling equilibrium state, in the following the
criticalbucklingloadforvariousloadingconditionswillbecalculatedwiththehelpoftheappropriate
linearizedstabilityequations.
17.2.2.2 Axial Compression

Considering the stability of an axially compressed cylindrical shell that is simply supported at its
ends,theprebucklingdeformationisaxisymmetricasshowninFigure172.Thecriticalload Pc isthe
lowestaxialloadatwhichtheaxisymmetricequilibriumstateceasestobestable.
Assumingthattheshellissufficientlylongsothattheeffectofbendingoftheshellwallclosetothe
ends can be neglected, then the prebuckling state can be approximated by the following membrane
state
P Eh2
Nxo = = , Nyo = Nxyo = 0 , w o = constant 17106
2R cR
where = Nx / Ncl with Ncl = Eh2 / cR and c = 3(1 2 ) .
Noticethatthismembranestatesatisfiesthenonlineargoverningequationsoftheprebucklingpath,
equations1797and1798identicallyandreducesthelinearizedstabilityequations17104and17105
tothefollowingsetofconstantcoefficientequations
+ L (w) 1 Eh2
L Q (f) D = f,xx w,xx 17107
R cR
L (w) 1
LH (f) Q = w,
17108
R xx
Theseequationsadmitseparablesolutionsoftheform
sin m x cos n y
w = A sin m x cos n y
, f = B 17109
L R L R
SubstitutionandregroupingyieldsaStandardEigenvalueProblem
2m2 2 A
Q,m,n m
D,m,n =0
2 17110
( Q,m,n m ) H,m,n B
withtheeigenvalues
2 2
1 D,m,n ( Q,m,n m )
mn = + 17111
2 2 2

m m H,m,n
andtheeigenvectors
335
ECSSEHB3224A
24March2010
x y
w = h sin m cos n 17112
L R
2
Eh3 ( Q,m,n m ) x y
f = sin m cos n 17113
2c H,m,n L R
Definitionofthestiffenerparametersusedisasfollows
4 + D 2 2 + D 4
D,m,n = D xx m xy m n yy n 17114
4 + H 2 2 + H 4
H,m,n = Hxx m xy m n yy n 17115
4 + Q 2 2 + Q 4
Q,m,n = Q xx m xy m n yy n 17116
where
4c 2 2c
Dij = Dij ; Hij = EhHij ; Qij = Q ; c = 3(1 2 ) 17117
Eh3 h ij
withthenormalizedwavenumberparameters
2 2
2 = m2 Rh Rh 1
m ; n2 = n2 17118
2c L 2c R
To find the critical (lowest) buckling load c using equation 17111 one carries out a search over
integer valued axial halfwave numbers m and integer valued circumferential full wave numbers n.
Theresultsofthesearchforthecritical(lowest)bucklingload c canbedisplayedinacontourmap
as shown in Figure 1710 for the ring stiffened shell AR1 [24] and in Figure 1711 for the stringer
stiffenedshellAS2[24].
Inordertoprovideaquickoverviewofthedistributionoftheeigenvalues,thevaluesdisplayedinthe
contour plots are renormalized such that the lowest eigenvalue is identically equal to one. Thus in
Figure1710fortheringstiffenedshellAR1thefollowingrenormalizedeigenvaluesareplotted
mn
m
c = 17119
1.09938
where 17,0 = 1.09938 is the lowest eigenvalue of the ring stiffened shell AR1; whereas in Figure
1711forthestringerstiffenedshellAS2thefollowingrenormalizedeigenvaluesaredisplayed
mn
m
c = 17120
1.44589
where 1,10 = 1.44589 isthelowesteigenvalueofthestringerstiffenedshellAS2.
Noticethatthedistributionoftheeigenvaluesisdistinctlydifferent.FortheringstiffenedshellAR1,
ascanbeseeninFigure1710,thelowesteigenvaluehasashortwaveaxisymmetricbucklingmode,
thatis m = 17, n = 0 ;whereasforthestringerstiffenedshellAS2,ascanbeseeninFigure1711,the
lowesteigenvaluehasalongwaveasymmetricmode,thatis m = 1, n = 10 .
336
ECSSEHB3224A
24March2010
Figure1710:DistributionofbucklingloadsbasedonLevel1membrane
prebucklinganalysisringstiffenedshellAR1[24]underaxialcompression
Figure1711:DistributionofbucklingloadsbasedonLevel1membrane
prebucklinganalysisstringerstiffenedshellAS2[24]underaxialcompression
337
ECSSEHB3224A
24March2010
17.2.2.3 Combined External Pressure and Axial Compression
If the shell is simply supported at its end then under the simultaneous action of uniform lateral
pressureandaxialcompressiontheprebucklingdeformationoftheshellisaxisymmetricasshownin
Figure174.
It is assumed for simplicity that the shell is sufficiently long so that the prebuckling state can be
approximatedbythefollowingmembranestate
Eh2 Eh2
Nxo = ; Nyo = pe , Nxyo = 0 , w o = hW + hWpe = constant 17121
cR cR
h Eh
where = Nx / Ncl ; pe = P / cl ; cl = ; Ncl = hcl and c = 3(1 2 )
R cR
It can easily be verified that this membrane state satisfies the nonlinear governing equations of the
prebucklingpath,equations1797and1798identically(whereby p = pe )andreducesthelinearized
stabilityequations,equations17104and17105tothefollowingsetofconstantcoefficientequations
+ L (w) 1 Eh2 Eh2

L Q (f) D = f,xx xx pe
w, yy
w, 17122
R cR cR
L (w) 1
LH (f) Q =
w, 17123
R xx
Theseequationsadmitseparablesolutionsoftheform
sin m x cos n y
w = A sin m x cos n y
, f = B 17124
L R L R
UponsubstitutingandregroupingoneobtainsthefollowingStandardEigenvalueProblem
2 + 2 )
2pe (R 2 ) A
( Q,m,n m %
D,m,n m n = 0
2 17125
( Q,m,n m ) H,m,n B

Notice that the nondimensional constant R = / pe has been introduced in order to obtain a single
parameterproblem.
If R = 0 theeigenvaluesare
2 )2
( Q,m,n m
1
pmn = pe = +
2 D,m,n
17126
2n H,m,n
criticalbucklingpressureforuniformexternallateralpressure
whereas,if R = 0.5
2 )2
( Q,m,n m
1
pmn = pe = +
2 D,m,n
17127
2
2(0.5m + n ) H,m,n
criticalbucklingpressureforhydrostaticpressure.
Thecorrespondingeigenvectorsaregivenbyequations17112and17113.
338
ECSSEHB3224A
24March2010
17.3 References
[1] Brush,D.O.andAlmroth,B.O.,BucklingofBars,PlatesandShells,McGrawHillBook
Company,NewYork,1975.
[2] Donnell,L.H.,StabilityofThinWalledTubesunderTorsion,NACAReportNo.479,1933.
[3] Koiter,W.T.,OntheStabilityofElasticEquilibrium,Ph.D.thesis(inDutch),THDelft,H.T.
Paris,Amsterdam(1945).Englishtranslations:NASATTF10,833p.(1967)andAFFDLTR70
25,(1970).
[4] Lorentz,R.,AchensymmetrischeVerzerrungenindnnwandigenHohlzylinder,Zeitschrift
desVereinesDeutscherIngenieure,Vol.52,1908,pp.17061713.
[5] Timoshenko,S.P.,EinigeStabilittsproblemederElastizittstheorie,Zeitschriftfr
MathematikundPhysik,Vol.58,1910,pp.337385.
[6] Southwell,R.V.,OntheGeneralTheoryofElasticStability,PhilosophicalTransactionofthe
RoyalSocietyofLondon,Vol.213,SeriesA,1914,pp.187244.
[7] Timoshenko,S.P.andGere,J.M.,TheoryofElasticStability,McGrawHill,NewYork,1961.
[8] Singer,J.,Arbocz,J.andWeller,T.,BucklingExperimentsExperimentalMethodsinBuckling
ofThinWalledStructuresVol.1,JohnWiley&Sons,ChichesterWeinheimNewYork
BrisbaneSingaporeToronto,1998.
[9] Southwell,R.V.andSkan,S.W.,OntheStabilityunderShearingForcesofaFlatElasticStrip,
Proc.Roy.Soc.Londen,SerieA,Vol.105,1924,pp.582607.
[10] Schwerin,E.,DieTorsionsStabilittdesdnnwandigenRohres,Z.Angew.Math.Mech.
(ZAMM),Vol.5,1925,pp.235243.
[11] Yamaki,N.,ElasticStabilityofCircularCylindricalShells,NorthHollandSeriesinApplied
MathematicsandMechanics,Amsterdam,1985.
[12] Hedgepeth,J.M.andHall,D.B.,StabilityofStiffenedCylinders,AIAAJournal,Vol.3,No.12,
December1965,pp.22752286.
[13] Flgge,W.,DieStabilittderKreiszylinderschale,IngenieurArchiv,Vol.3,1932,pp.463506.
[14] Seide,P.andWeingarten,V.I.,OntheBucklingofCircularCylindricalShellsunderPure
Bending,ASMEJournalofAppliedMechanics,Vol.28,No.1,March1961,pp.112116.
[15] Potters,M.L.,AMatrixMethodfortheSolutionofaLinearSecondOrderDifferenceEquation
inTwoVariables,ReportM.R.19,MathematicalZentrum,Amsterdam,1955.
[16] Budiansky,B.andRadkowski,P.P.,NumericalAnalysisofUnsymmetricalBendingofShellsof
Revolution,AIAAJournal,Vol.1,No.8,August1963,pp.18331842.
[17] Stephens,W.B.,Starnes,J.H.andAlmroth,B.O.,CollapseofLongCylindricalShellsunder
CombinedBendingandPressureLoads,AIAAJournal,Vol.13,No.1,January1975,pp.2025.
[18] Almroth,B.O.,Brogan,F.A.,Miller,E.,Zele,F.andPeterson,H.T.,CollapseAnalysisforShells
ofGeneralShape:UsersManualfortheSTAGSAComputerCode,ReportAFFDLTR718,
AirForceFlightDynamicsLab,WrightPattersonAFB,1973.
[19] Brazier,L.G.,OntheFlexureofThinCylindricalShellsandOtherThinSections,Proc.Royal
SocietyofLondon,SeriesA,Vol.116,1926,pp.104114.
[20] Jones,R.M.,MechanicsofCompositeMaterials,Taylor&FrancisInc.,Philadelphia,PA,1999.
[21] Baruch,M.andSinger,J.,TheEffectofEccentricityofStiffenersontheGeneralInstabilityof
StiffenedCylindricalShellsunderHydrostaticPressure,JournalofMechanicalEngineering
Sciences(England),Vol.5,No.1,March1963,pp.2327.
[22] Singer,J.,Baruch,M.andHarari,O.,OntheStabilityofEccentricallyStiffenedCylindrical
ShellsunderAxialCompression,InternationalJournalofSolidsandStructures,Vol.3,1967,
pp.445470.
339
ECSSEHB3224A
24March2010
[23] Baruch,M.,EquilibriumandStabilityEquationsforStiffenedShells,IsraelJournalof
Technology,Vol.2,No.1,1964,pp.117124.
[24] Singer,J.,Arbocz,J.andBabcock,C.D.Jr.,BucklingofImperfectStiffenedCylindricalShells
underAxialCompression,AIAAJournal,Vol.9,No.1,January1971,pp.6875.
340
ECSSEHB3224A
24March2010
Part 3: Structures
341
ECSSEHB3224A
24March2010
18
Buckling design, analysis and testing of
large structures
18.1 Design
18.1.1 General
Today,aeronauticalandspacestructuredesignis abalancebetweenmassand safety(reliability)(see
alsosections8.3and8.5.1).Anessentialgoalinthedesignofthestructureistobeaslightaspossible,
and this requirement is to be respected from the beginning of the development process. Every
componentintheidealcaseisdesignedwiththeobjectivetowithstandthedesignloadsduringits
wholelifewithmarginswhichcanbeclosetozero(butremainingpositive).Theseloadsarecausedby
transport, handling and other steps during its ground life, and inflight loads. All these loads are
allocated with their factor of safety. Thus, critical locations of the structure are designed so close to
their failure state that the appearance of original imperfections or flaws above their defined
acceptancelimits,aswellasanyincreaseofloadduetoanevolutionofthespecificationmayleadtoa
violation of the reliability target defined by the project with respect, e.g., to the demonstration of
strength or to a required deformation limit. This is particularly true for a failure mode such as the
globalbucklingofastructure.Alltheseconstraintsareconsideredinthecontextofloweringcost.
18.1.2 Design aspects

Thedimensioningofastructureisalongandcomplexprocess(seechapter8forglobalconsiderations
on buckling design): it combines the different external and internal loadings acting on the structure
(sources:static,dynamic,thermal,environmental, residualstresses)consideringthedifferentground
andflightphases.Specialconcernisontheeffectoflocalloadingsduetoattachmentsandjoints.The
difficultyofthestructuraldimensioningalso lies inthesequenceofallthe tasks:duringafirststep,
the loading is separated into elementary cases and in a second step, these elementary cases are
combined taking into account the loading history. This is particularly true when different kinds of
loadingsactonthesamestructure.Aclassicalexampleistocombinecorrectlythemechanical,thermal
andlocalloads.
Fromtheverystartofthedimensioningprocess,itisthedesignerstasktodevelopastructurewhich
meets the stiffness, loads (static and dynamic), environmental and mass requirements. Taking into
accountthelessonslearnedfromearlierlaunchvehicles,thepreliminarydimensioningstepleadstoa
designclosetothefinaldesignofthestructure.Thisimportantfirststepdeterminesthegenerallayout
(e.g.frame,stiffener,box,etc.)andleadstoafirstimpressionofthefuturedesignandtoafirstchoice
of among others the materials and the wall thicknesses. Once this first design is obtained, the
resistanceagainstlocalandglobalbucklingofthestructurecanbeassessed.
With regard to the global buckling, the first calculations can be straightforward. They can be
performed with the application of simple formulas (theories of Timoshenko or Bruhn). This can be
understood as the Level1 calculations as described in chapter 11. For instance, the first buckling
calculations on the ARIANE 5 Front Skirt were Euler calculations on stiffened panels. In the same
342
ECSSEHB3224A
24March2010
trend,fortheARIANE4and5maintanks,allthebucklingcalculationswerecarriedoutanalytically
withtheNASASP8007formulae.Someofthemwereimprovedwithfeedbackfromexperience.
More recently, for the last generation of launchers (ARIANE 5 ECA), the increasing computer
capabilities allowed the development of large finite element models to study the buckling
phenomenon.ItcanbeconsideredastheLevel3calculationsinthehierarchicalapproachdescribedin
chapter 11. For more information on modelling aspects, refer to chapter 10. At first, a series of
calculationwithdifferentsetsofloadsareperformedinorderto:
a. understandthebehaviourofthestructurewithrespecttobuckling,
b. defineareferenceconfigurationwithasetofimperfections.
The imperfections deal with the real geometry of the structure including the manufacturing defects
andtolerances(seechapter9),andtherealbehaviourofmaterial(seechapter7).Theseareinputdata
fornonlinearcalculations.Comparedtotheanalyticalcalculations,thisapproachshowsanincreasing
marginofsafetywithregardtobuckling.Allthesedesignaspects(modelprecision,loadpath)willof
coursebeconsolidatedbyspecifictests(elementarytestsorlargestructuretests)(seechapter12).
18.1.3 Sub-structuring of large structures

Foralargestructureasalauncher,thedetailedfiniteelementmodelcanbetoolargetobeeasytouse,
tosuchanextentthattheapplicationofthesubstructuringtechniqueisnecessary.Forinstance,itis
not possible to perform a calculation at the level of the launcher. In addition, various contractors
design specific sections of the launcher and provide finite element substructures for the complete
mathematicalmodel(thismodelbeingdedicatedtodynamicanalysis).
Forthesereasons,alargestructurelikealauncherissectionedintosubstructuresandalltheanalyses
arecarriedoutatthelevelofthesesubstructures.Figure181visualizeshowthelauncherARIANE5
has been substructured and how the loadings are applied for a structure like the front skirt at the
interfacebetweenthemaincryogenictankandtheupperstage.
ModernfastFEAcodesallowtakingintoaccountrealisticboundaryconditionsbyincludingthefinite
elementmodelsoftheadjacentstructures.Ofcourseitincreasesthesizeofthecomputationalmodel.
In fact, the way to include the adjacent structure is always a compromise between the size of the
adjacentfiniteelementmodelsandthewilltosimulatesomerealisticboundaryconditionsinstiffness.
Asaresultofthisstackingofstructures,thefirstbucklingmodecanappearonanadjacentstructure
andnotonthestudiedone.Inthiscasethedesignercanlocallymodifythestructuralstiffnessofthe
adjacentpartsinordertomakethebucklingmodeappearfirstinthestudiedstructure.Inanycase,
the load path should not be changed by these modifications. This kind of process is based on
engineeringexperience.
To improve the precision of the results in the critical areas where buckling is supposed to appear,
somedetailedmodelscanbedirectlyintroducedintotheglobalmodel.
343
ECSSEHB3224A
24March2010
Figure181:Substructuringofalargestructure
18.2 Analysis and testing
18.2.1 Introduction
The assessment of the results, the methodology for the correlation and the validation of the
mathematicalmodelandthepitfallstobeavoidedhavebeenalreadymentionedinchapter11butwill
berevisitedthroughthedifferentexamplespresentedinchapter19.
Thefollowingparagraphfocusesonthemainpointsrelatedtothelogicofanalysisandtest.
18.2.2 Loading processing

Before going more in details, it is necessary to differentiate the various types of mechanical loads
actingonastructure(seealsoconsiderationsdevelopedinsection8.2).
The loading to which each structure or each substructure is submitted can be classified in two
categories:
a. the destabilising loads, which create a stress field decreasing the apparent stiffness of the
structure leading to its buckling. For instance, for a launcher tank, destabilising loads are
externalpressureorcompressionloads.
b. thestabilisingloads,whichinthecaseofthebucklingcreateafieldofconstraintsincreasingthe
apparent stiffness of the structure. In the case of a cylindrical shell, they can be the internal
pressureorthetensionloads.
344
ECSSEHB3224A
24March2010
Aloadingcombinationwhichdoesnottakeintoaccountthedistinctionbetweenthesetwo typesof
loadsleadstotheapplicationofthesamesafetyfactortobothloads(destabilisingandstabilising)and
thereforeleadstoincreasingtheseloadssimultaneouslyuntilthestructurebreaksdown.Thiswayto
considertheloadsiscorrectifitisassumedthatthestabilisingloadsarestatisticallydependentonthe
destabilising loads, which is a priori not always the case. In this last case, the method is optimistic
withregardtothecriticalbucklingloadandoverestimatesit.
If the two types of loads are statistically independent, the approach to be followed is to apply the
safety factor to the destabilising loads and to increase them until collapse. Stabilising loads are
maintainedattheirspecifiedlevel(limitloads).
Moreover,thespecifiedloadsareingeneralmaximumloadingsandmorepreciselythelimitloads,i.e.
having a probability of 99% to not being exceeded. In this assumption, it is clear that to consider a
maximum value for the stabilising loads also leads to over estimate the critical buckling load, since
theseloadstendtorelievethestructure.Thus,itisnecessarytohaveminimumvaluesforstabilising
loads.Iftheyarenotdetermined,thestabilityanalysiswillbecarriedoutwithoutconsideringthem,
thereforeinaconservativeway.
18.2.3 Flow chart for analysis and testing (logic)

18.2.3.1 Overview
The dimensioning process starts with the selection of the dimensioning load cases (DLC). This is a
selection amongst all loadings acting on a structure in order to retain only the most severe
combinations of loading which drive the design: static and quasistatic mechanical loads, thermal
loads. Cyclic loads for fatigue are not addressed here. These loads are time dependent and the
selectionofthedimensioningcasesshouldtakeintoaccountthehistoryoftheloading.
Twomainanalysesareconventionallyperformed,dependingonthecomplexityoftheloading.Ifthe
setofloadingisrathersimple,theselectionofthedimensioningcaseisbasedonthescanningofall
possible combination of the elementary loads and the extraction of the combinations leading to the
lowest margins in the structure. If the loading is more complex, the selection of the dimensioning
casesisrelatedtoabatchofsimplifiedfluxes.Thesefluxesaregenerallythemaximumaxialfluxesat
themaininterfacesofthestructure,whicharecalledenvelopecases.
Figure 182 illustrates how analysis and testing are logically linked. It repeats in a slightly different
waytheFigure181.
Fordimensioningloadcasesselection,twomaincasestudiescanbediscriminatedandarepresented
here below: the case study where the loading is rather simple (section 18.2.3.2) and the case study
wheretheloadingismorecomplex(section18.2.3.3).
345
ECSSEHB3224A
24March2010
Figure182:Flowchartforanalysisandtesting
18.2.3.2 Simple Case Study - Stable Operational Temperature and Pressure

ThiskindofloadingistheoneusedtodimensionthemaintanksoftheARIANElauncher.Here,the
selection of the DLC for buckling means the selection of the most unfavourable load combination
whichisrelevantforbuckling.Itinvolvesconsideringtimedependency.
A theoretical value for collapse is identified by an analytical calculation which takes into account a
knockdownfactortocoverdifferenteffectswhicharenotconsideredintheanalyticalmodel,mainly
suchas:
geometricaldefects(differencebetweenidealshapeandrealshapeorgeometry),
manufacturingdefects.
Accordingtothemarginofsafetylevelobtainedwiththeaboveanalysis,twopossibilitiesoccur:
a. the margin of safety is very comfortable (conventionally higher than 70%): the analytical
methodisconsideredassufficient.
b. the margin of safety is lower than the threshold above: the designer has to perform
sophisticated computations by improved mathematical models with the purpose of idealising
the actual structure more adequately. This can be achieved by careful attention to geometric
imperfections, by improvement of the material law, by considering of residual stresses and
misalignments (e.g., the misalignment of the neutral fibre, which is generated by the junction
betweenframeandcurrentskinorbytheweldjunction).Atthisstep,ifthefinalbucklingsafety
marginispooratestisrecommended.
346
ECSSEHB3224A
24March2010
18.2.3.3 Case Study of Complex Type - Different Types of Loadings Act Together
To makeeasier the explanation, two examples are proposed to illustrate the procedure dedicated to
the selectionof the DLCrelevantfor buckling. During the selection process, it can be noted that for
suchcomplexcasestudies:
a. thecombinationofthedifferenttypesofloadingshouldrespectthetimeconsistency,
b. then,thenumberofdimensioningloadcasestobeconsideredincreasesrapidly,
c. then,envelopecasesaresystematicallybuiltinordertoreducetheanalysisandtheassociated
models.
18.3 Examples to illustrate the use of the Flow Chart
18.3.1 Overview
This paragraph displays two examples on the ARIANE 5launcher in orderto shown the use of the
flow chart with respect to the selection process of dimensioning load cases, the main step of the
bucklinganalysisandthetestneeds.TheapplicationcasesaretheVulcain2nozzleandthecylindrical
structureatthefrontofthemaincryogenictank.
18.3.2 Vulcain 2 Nozzle

18.3.2.1 Description
Thefirstexampleisanozzleengineunderaveryseverefieldofloadings.Figure183describesthe
structureandtheloadingofthenozzlecone.
Figure183:Loadingsandcoordinatesystemofthenozzlecone
347
ECSSEHB3224A
24March2010
18.3.2.2 Loading types
Therearetwomaintypesofloadingsactingonthenozzle:
a. externalloadsincludingboththermalandmechanicalloads,suchas:
1. mechanicalloads(externalandinternal),
2. aerodynamicstationaryloads(axial,transverse,bendingandovalisation),
3. externalpressure,
4. aerodynamicinstationaryloads(buffetingloads),
5. sinusdynamicloads,
6. gimballingloads,
7. thermalloads(externalandinternal).
b. operationalorinternalloadswithbothtemperatureandinternalpressure.
18.3.2.3 Load cases

Thecombinationoftheseloadingsrevealstwoessentialdimensioningloadcases:
a. radialbuckling:maximumovalisationloadsaredimensioning,
b. axialbuckling:minimumlongitudinalaccelerationisdimensioning
NOTE Thetwomajorstaticfailuremodeofthenozzlearedesignatedasaxial
buckling and radial buckling. In fact, they are both elastoplastic
collapsefailuremodesandnotelastic(orEuler)buckling.
18.3.2.4 Additional variables

Inadditionitistobeaccountedfor:
a. geometricspecifications(internaltubes,frames,stiffeners),
b. geometricaldefects(weldedjoints),
c. ovalisationdefects.
18.3.2.5 Flight phases

Moredetailedanalysisenablestodefinetwoflightphasesforwhichtheloadsregardingaxialbuckling
arethehighest.Thesephasesare:
a. buffetingphase(between30and80s),
b. EPCphase(duringthemainstageflightphase).ThesineloadsusedfortheEPCphasearean
envelope that includes the EAP separation loads, the EPC phase itself loads and the engine
shutdownloads.
During the startup and ignition phases, the atmospheric pressure generates a reverse thrust which
does not lead to axial buckling but the associated blast wave generates radial buckling. The load
history with a simultaneous application of loadings is an important point for any kind of buckling
analysis for the nozzle. Therefore, when computing up to the critical axial buckling load, normally
onlytheexternalloadshavebeenincremented.Thermalloads,ovalisationloadsandunloadingloads
arekeptconstant.Theloadpathsequenceneedstofollowtherealconditionsduringtheflight:atfirst
thermalloads,thenthrust,andfinallyovalisation.
348
ECSSEHB3224A
24March2010
18.3.2.6 Study cases
Fortheradialbuckling,twocaseswerestudied:
a. first case: thrust decreases, drag forces and aerodynamic stationary bending moment remains
constant,ovalisationloadsareloading
b. second case: thrust is constant; drag forces and aerodynamic stationary bending moment are
loading,ovalisationloadsareloading
NOTE ThemainriskliesinthepossibilitynottofindallthedesigndrivingDLL.
Thiscouldbecatastrophic.
The above selection is performed in agreement with the rules established in the ARIANE General
specificationfordesignandtestofstructure[A5SG1X10ASAI(5/12)].
18.3.2.7 Factors of safety

According to this specification, the factors of safety j used in nominal case for radial and axial
bucklingare:
a. j=1,appliedonthethermalloadsandontheunloadingloadslikethelongitudinalacceleration,
theaerostationaryforcesandthedragforces,
b. j=1.25,appliedontheloadswhichcanleadtobuckling,liketheaerodynamicforces,thethrust
andtheaerostationarybendingmoment,
c. j=1.25,appliedontheovalisation.
18.3.2.8 Discussion
In order to partially validate all these calculations at least a buckling test should be performed.
However, due to test restrictions for cost limitations, an experimental simulation of the nominal
flightconditionsisgenerallynotpossible.Themaindifferencesare:
a. thetestisperformedatroomtemperature,
b. nothrustisapplied,sothethermalconditionsontheinnersidearenotthecorrectones,
c. atruncatednozzleisusedforthetest.
Therefore, the test load case is a fictitious load case that does not take the real thermal field into
account.Asthisfictitiousloadcaseisbasedonthemaximumflux,onthebendingmomentatthetop
of the nozzle and on cold conditions, the desired flight case conditions are not met. In order to
performcorrectlythistestitisthedesignerstasktoprepareananalyticaltestpredictionwiththetest
conditions.Afterthetestacorrelationofthemathematicalmodelisexpected.
Figure184illustratesthetestconfigurationchosenforthenozzlebucklingtest.
349
ECSSEHB3224A
24March2010
Figure184:Loadingsandcoordinatesystemofthenozzlecone
18.3.3 ARIANE 5 Main Cylindrical Structure

18.3.3.1 Description
Thesecondexampleisacylindricalstructurewithadiameterof5.4meter.Itissubjectedtoextreme
mechanicalloadsandatemperaturegradientbetweeninnerandoutersurface.
In the predimensioning phase, stability studies based on linear eigenvalue analyses revealed low
margins of safety (MOS). For increasing the low MOS and for a better examination of the post
bucklingbehaviouramoredetailedglobalnonlinearanalysisbecomesnecessary.
Therefore,themeshwasrefinedinspecificareasinordertoenableamappingofthemostimportant
low order modes. The necessary information on the convergence of the nonlinear analysis was
deliveredbyameshconvergencecheckwithintheFEMcodeused,whichwasABAQUS.
18.3.3.2 Analysis strategy

Thestrategyusedforthisbucklinganalysisfordeterminingthedimensioningloadcases(DLC)isas
follows:
a. Atfirstthefactorsofsafety(FOS)areprovided:
1. j=1.0,forpressureloads
2. j=1.0,forthetemperaturefield
3. j=1.25,foraerodynamic,globalandlocalmechanicalloads.
b. Five kinds of loadings are applied on the structure in a certain sequence and regarding the
concurrency:
350
ECSSEHB3224A
24March2010
1. maximumaxialflux,
2. pressureinthetank,
3. temperaturefield,
4. airpressure,
5. globalandlocalmechanicalloads.
18.3.3.3 Discussion
Atfirst,linearstabilityanalysisisemployedfortheselectionoftheDLC.Accordingtotheresults,the
conclusionshavebeendrawnthattheobtainedlowesteigenmodescanbededicatedtolocalbuckling
modes.Somenegativesafetymarginsappearonthestructure.
AccordingtotheseunacceptableMOSlevel,nonlinearstabilityanalysesarenecessary.Thegoalofthese
analyses is to study the capabilities of the structure to transfer the applied loads with the local
eigenmodesinducedbythethermalandmechanicalfieldspresentinthestructure.Inotherwords,the
resulthastobeanassessmentofthepostbucklingbehaviourofthestructure.Thedimensioningload
casesusedforthiscalculationarethoseselectedinthelinearanalysis.Thestrategyofloadingconsists
in applying consecutively the different components of loadings mentioned above. In nonlinear
analysis,followingtheloadingsequenceisprimordial.
Twotypesofimperfectionshavebeentakenintoaccount:geometricaldefectsandresidualstresses.In
addition,thematerialnonlinearityhasbeenimplemented.Fromthebeginningoftheanalysis,these
hardwarespecificationsnonlinearityhavebeenconsideredinthecalculations.Theyarenotdetailed
herebutithastobenoticedthatthistypeofcalculationcanrapidlybeverycostly,especiallywhen
modellingdefectswhichcandrasticallyincreasethesizeofthemodelsandthecomputingtimes.This
typeofanalysisrequestsaverycarefulpreparationofthecalculations.
18.4 Validation of the Mathematical Model

Thefinalstepofthequalificationprocessisthecorrelationofthenumericalmodels(seealsosection
11.4.7forgeneralconsiderations).Thisstageoftheanalysisisessential,butthetimetobedevotedto
thistaskisoftenunderestimated.Itisgloballypresentedintheschemeherebelow(Figure185).
The part of the logic which is now under consideration is the one surrounded by the blue oval in
Figure 185. The correlation task consists at first in comparing the test predictions to the measured
dataandcheckingthemismatches,andsecond,ifthemismatchesareconsideredasnotacceptable,to
adaptthemodelaccordingtoseveralsteps.Amongothers,themainonesare:
a. refiningmeshesgloballyorlocally,byusingvolumetricelements
b. being fully representative of the fine geometry (all discontinuities like stiffeners, connections,
holes)
Thecorrelationconcernsalsothefailuremodewhenmodelsareabletoreproduceitwithacceptable
calculationtimes.
Criteria of acceptation of the good matching between predictions and test results are commonly
agreedbythedesignresponsibleandthecustomer.Conventionally,intheARIANEcommunity,itis
asked15%ofmismatchondisplacementsand20%onstresses.
In the case where the test is not considered as sufficiently representative of the set of loadings
inducing buckling because of the complexity, especially where it demands implementation of
thermomechanicalfieldsthecorrelationislimitedtoasimplifiedsetofloadscomparedtotheones
reallyactinginflight.Thesimplificationcanalsoleadtoadistortedfailuremodeintestcomparedto
351
ECSSEHB3224A
24March2010
theoneexpectedinflight.Thus,sincethecorrelationispartial,itisaskedtomakethecorrelationtasks
by different entities, with their own tools and methodologies. This crosschecking is assumed to
compensatethelackofcompletionoftheverificationtest.
Oncethecorrelationisconsideredasacceptable(correlationbycomparisonwiththetestorbycross
checking), the numerical model is considered as validated and useful for the production life of the
structure.
The return experience on the launcher structures shows that, whatever the care brought to the
representativeness of the test, the full reality of loads and boundary conditions cannot be
implementedbecauseofthecomplexitytoperformtestsonsuchlargestructuresandtheconsequent
costisoftennotaffordableforadevelopment.Thus,evenwithanavailabletest,thequestionofthe
confidenceinthemarginsofsafetywithrespecttobucklingisalwayspresent.
Thedifferentexamplesgiveninthefollowingsection18.5willfocusonthattopic.
Figure185:Logicofmodelcorrelation
352
ECSSEHB3224A
24March2010
18.5 Overview of Buckling Analysis Performed on the

ARIANE Launchers
As an introduction to the following paragraph, the Table 181 has the objective to present in a
syntheticwaythedifferentmethodologiesusedfordimensioningtobucklingthemainstructureson
theARIANE5launcher.TheFigure186recallsthearchitectureofthelauncherARIANE5inorderto
localisethestructuresmentionedinthetable.
Figure186:ARIANE5ECAlauncherconfiguration
353
ECSSEHB3224A
24March2010
Table181:OverviewofbucklinganalysisonARIANE5structures
Structure FailureMode Analysis Knockdown Model Buckling
Factor Margin
UpperSkirt Axisym.axial Linear 0,365 3D(Nastran) 2%
buckling Nonlinearwithdefects NO 3D(Nastran) 25%
Elongated Axisym.axial Linear 0,484 3D(Nastran) 20%
LowerSkirt buckling Nonlinearwithdefects NO 3D(Nastran) 22%
RLH2ESCA Axisym.axial NASASP8007(CSP YESin NO 4,92%
cylinder buckling method) formulae
UpperYring Circumferential Linearwithvariationsof NO Axisymetric 390%
ESCA buckling thecircumferentialwave (STAGSC)
LowerYring Circumferential Linearwithvariationsof NO Axisymetric high
ESCA buckling thecircumferentialwave (STAGSC)
ISS Axisym.axial Linear 0,65 3D(Nastran) 29%
buckling
InnerDome Snapthrough Nonlinearwithdefects NO 3D(Nastran) 28%
buckling
RLH2EPC Axisym.axial NASASP8007(CSP YESin NO 16%
JAVE Axisym.axial Linear 0,71 3D(Nastran) 2%
buckling Nonlinear NO 90(Abaqus) 38%
RLOXEPC Axisym.axial NASASP8007(CSP YESin NO 12%
ITSRingECS Axisym.axial Linear NO 3D(Nastran) 38%(fail
A buckling safe)
ITSStrutESC Axisym.axial Analytical(EulerJohnson) NO NO high
A buckling
Equipment Axisym.axial Linear 0,65 3D(Nastran) 19%
Bay buckling
Fairing Axisym.axial Linear 0,65 3D180 69%
buckling (Nastran)
LowerYring Axisym.axial Nonlinearwithdefects NO 3D(Nastran) 56%
EPC buckling
Axisym.axial Linear NO 3D(Nastran) 68%
buckling
UpperYring Axisym.axial Linearwithvariationsof NO Quasi 202%
EPC buckling thecircumferentialwave axisymetric1
(Nastran)
ACU Axisym.axial Linear 0,33 3D(Nastran) 34%
buckling
SYLDA5 Axisym.axial Linearwithvariationsof NO Axisymetric 70%
buckling thecircumferentialwave (Bosor4)
354
ECSSEHB3224A
24March2010
18.6 Abbreviated Terms

ThefollowingabbreviatedtermsaredefinedandusedwithinthisChapter:
ACU AdaptateurChaurgeUtilsePayloadAdapter
DLC DimensioningLoadCases
EPC EtagePrincipalCryogniqueMainCryogenicTank
ESCA EtageSuprieurCryogniqueversionAUpperCryogenic
StageversionA
ISS InterStageStructure
ITS InterTankStrut
JAVE JupeAVantEquipeFrontSkirt
RLH2 RservoirHydrogneLiquideLiquidHydrogenTank
RLOx RsevoirOxygneLiquideLiquidOxygenTank
SYLDA SYstmedeLancementDoubleARIANEDualLaunch
Structure
355
ECSSEHB3224A
24March2010
19
LH2-Tank ARIANE 5
19.1 Overview
ThischapterpresentstheverificationoftheLH2TANKofARIANE5withreferencetothecollapse
dimensioningcase.
Foreachsubstructuretheboundaryconditions,thedimensioningloadcase,themethodappliedtothe
collapse,(imperfections)arepresented.
This methodology is in agreement with the dimensioning rules applied on ARIANE community
especiallytheapplicationofthedimensioningspecificationA5SG1X10ASAI[1].
Wetryinthisparagraphtoexplainthedifferentmethodsusedonthisstageforthebucklinganalysis.
AccordingtothesubstructurewecanfindanalyticalmethodsuchastheapplicationoftheNASASP
8007[2],linearcalculationandnonlinearcalculationtakingintoaccounttheimperfections.
The buckling is studied for each substructure. But the reader will see along this paragraph that the
bucklingisnotrupturemodeofonesubstructurebutratherarupturemodeofthewholeofthestage.
19.2 Presentation of the structure

The aim of this chapter is to present all the calculations dealing with the buckling analysis on the
hydrogentankoftheESCAARIANE5stage.
This stage is under ASTRIUMST responsibility. The structure has been developed and qualified by
CRYOSPACEandMTA(exManTechnologies).
The structure localization and a schematic view of its constitution are presented here below (Figure
191).
356
ECSSEHB3224A
24March2010
Figure191:LH2tanklocationinARIANE5ECAandconstitutiveelements
TheESCLH2tankisbeingdevelopedforARIANE5cryogenicUpperStageESCA.Thetanksupplies
liquid hydrogen to the engine and transfers thrust from main stage EPC and ESC to the upper
composite.
19.3 Sizing rules

Thebucklingorcollapsephenomenaarehighlysensitivetoallkindsofimperfections(materials,wall
thickness,shape,boundaryconditions).
Inapredimensioningphase thiscanbetakenintoaccountbyperformingalinearbucklinganalysis
using the nominal structure and applying a certain knockdown factor (KDF) based on statistical
evaluationofexperiments,asmentionedinsection4.3andpresentedinthesummarytable181.
Theresultofageometricalandmaterialnonlinearcollapseanalysisofastructurerepresentstheload
carryingcapacityofthestructuremodelledwithESCrelatedstabilityanalyses.Thismeans,thatifa
geometricalperfectstructureismodelled,thecollapseanalysisleadstotheloadcarryingcapacityof
theperfectstructure.TheFEMmodelhastorepresenttherealstructureinthebestpossibleway.
If geometricimperfections(shape, wall thickness)are includedinthe model, the adjacent structures
arereflectingtherealflexibilitiesandmaterialdeviationsaretakenintoaccount,theanalysisvalidates
thestructuralresistancewithouttheapplicationofaKDF.
Deviationsinthematerialspropertiescanbetakenintoaccountbyapplyingstrainstresscurveswith
theminimumvaluesforyieldandrupture(seeFigure192).For linearanalysispathofstressstrain
curve, the mean Youngs modulus is applied to map the correct elastic behaviour. During the
development phase, shape deviations are not known. Artificial shape deviations can be taken into
account such as eigenmodes stemming from a linear buckling analysis either isolated or linearly
combined.
357
ECSSEHB3224A
24March2010
Figure192:Typicalstressstraincurves
For the linear buckling analysis, the nominal thickness has been used to ensure the correct elastic
behaviourofthestructure.
Forthenonlinearcollapseanalysis,theminimalthicknesshasbeenusedtotakethicknessvariations
asimperfectionsintoaccount.
The buckling analysis for each component of the LH2 tank is described below following these
differentpoints:
a. Modelusedfortheanalysis
b. Boundaryconditions
c. Failuremode
d. Analysismethod
e. Loadcases
Ingeneral,forthebucklingsizingoftheupperskirt,thecylinderandtheElongatedLowerSkirt(ELS),
thegeneralloadshavebeenincreasedby15%(whichisaprojectmargin).
19.4 Upper skirt
19.4.1 Model used for the analysis

The upper skirt of the LH2 tank is designed as an orthotropic stiffened shell with a skin of 3 mm
nominalthicknessand210longitudinalstringersof3mmnominalthickness.Duetotheshortlength
oftheupperskirtof179mmcomparedtotheradiusof2705mm,acollapseanalysisregardingthis
part would not make any sense the buckling behaviour of a very short cylindrical shell is
comparabletotheplatebucklingphenomenon.Thus,theFEMmodelfornonlinearcollapseanalysis
oftheupperskirtconsistsoftheupperskirt,theupperYring,thecylinderandoneadjacentstructure
representingthevehicleequipmentsbay(whichisthestructurelocatedjustabovetheLH2tank).Due
358
ECSSEHB3224A
24March2010
to the symmetric behaviour of the cylinder between the two Yrings, only half of its length is
modelled.Theupperdomeisalsotakenintoaccountinordertoapplyapredeformationstatecaused
bytheinternaltankpressure.Forsymmetricreason,onlya180modelisused.Allstructuralpartsare
representedbyshellelements.

The upper boundary conditions of the upper skirt is represented by the adjacent structure
representingthevehicleequipmentsbay,wheretheradialandcircumferentialrotations,relatedtoa
cylindrical coordinate system are fixed at the upper boundary. The load path eccentricity is also
introducedintotheanalyticalmodel.
To avoid buckling mode or collapse in the area of the cylinder, an axial boundary condition (i.e.
blockeddisplacementandrotation1and4)isappliedatthelowerinterfaceoftheupperYring(see
Figure193).
Localization of the
boundary conditions
Figure193:FEMconfigurationandboundaryconditions.
19.4.3 Expected failure mode

Duetotheaxialfluxloads,theUpperskirtisabucklingsensitivestructure.
19.4.4 Analysis method

Forthepreliminarydesign,alinearbucklinganalysis(eigenmodescalculation)hasbeenperformed.
Theeffectofinitialimperfection(geometryandmaterial)aswellasartificialboundaryconditionshas
beencoveredbyaKDF.
TheKDFisdefinedinaccordancetotheNASASP8007[2]usingthemethodofSEIDE:
359
ECSSEHB3224A
24March2010
191

ThedetailsoftheKDFcalculationareshowninTable191.
Table191:CalculationoftheKDF(KnockdownFactor)

Fortheupperskirt,theformulaleadsto:
Seide _ ort = 0.365 192
The final design has beenjustified bya nonlinear collapseanalysis taking scaled eigenmodes froma
linearbucklinganalysisasinitialimperfectionsintoaccount.

The dimensioning case of the Upper skirt occurs during the EPC flight. The main contributor is the
generalloads.Moreprecisely,theloadscontributingtothedimensioningloadcasearethefollowing:
a. General loads the compression loads represent the buckling load case. This load case takes
intoaccountanaxialandbendingfluxonthemainsectionandalsoacompressivefluxcoming
fromtheEAP.Thevalueofthisfluxismoreorless300N/mm
b. Thermalconditions:athermalgradientisappliedtotheUpperSkirtanalysismodel(seeFigure
194);
c. InternalpressureoftheLH2tankinducesdeformationsintothestructureandthusitisapplied
to the wetted areas. It is then checked, whether the maximum or the minimum pressure in
combination with the thermal predeformation leads to higher stresses respectively lower
bucklingorcollapseloadscausedbythegeneralloads.
d. Local loads due to the cut outs and different supports have been applied in that direction
producingthehigheststressesincombinationwiththepreviousloads
360
ECSSEHB3224A
24March2010
Figure194:Thermalconditionsappliedonthemodel.
19.4.6 Results
Themarginsofsafetyarecalculatedasfollow:
Eigenvalue lim KDF
MOS = 1 193
lim J ult
Themarginsofsafetyarepositivewithalinearanalysis.
Allthesafetymarginscalculatedwithanonlinearanalysishavebeenfoundtobehigherthanthose
calculatedbythelinearmethod.
19.5 Y-rings
19.5.1 Model used for the analysis

ThemodelusedisbasedonaquasiaxisymmetricmodeloftheLH2tank(0.1degreesegmentofthe
whole tank with symmetry conditions of the edges). The Yring is modelled by solid elements to
representthestressstateofthesestructuralpartsinthemorerealisticway.

ThecylindricalpartoftheLH2TankisembeddedintotwodummystructuresabovetheUpperSkirt
and below the lower Yring. The upper dummy structure is fixed in axial translation and
circumferentialrotation.Thelowerdummyisfixedincircumferentialrotation.
For the Upper Yring, the boundary conditions are clamped at the top of the Upper Skirt and
symmetryconditionsatthetopoftheUpperDomeandthebottomofthecylinder.
FortheLowerYring,theboundaryconditionsareclampedatthebottomoftheELSandsymmetry
conditionsatthetopofthecylinder.
19.5.3 Expected failure modes

TheYringsaredesignedbasedonstrengthandfracturemechanic.Neverthelessthecylindricalparts
of the Yrings can contribute to the buckling of theadjacent structures(namely the UpperSkirt and
theELS)whicharesensitivetobuckling.Thus,Yringswillbecheckedwithrespecttobuckling,and
theanalysisisperformedconsideringalsotheadjacentstructures.
361
ECSSEHB3224A
24March2010
19.5.4 Analysis method for the buckling
ThetwoYringsoftheLH2Tankareanalysedtounderstandtheirbucklingbehaviourunderinternal
pressureusingthecomputercodeASTRA.
TheUpperYringismodelledincludingpartoftheLH2Tank,theUpperDomeandtheUpperSkirt.
TheLowerYringismodelledincludingthecylindricalpartoftheLH2Tank,theLowerDome,and
theURinginasimplifiedshape,thelowerSkirtandtheELS.

As it is already mentioned in the previous paragraph on expected failure modes, the Yrings are
sensitivetothebucklingphenomena.Accordingthispoint,themajorcontributoroftheloadingisthe
internal pressure. For this case the value of the internal pressure is 3.45 bars. Since the internal
pressureleadingtothisbucklingphenomenonisassumedtosmoothouttheimperfections,noKDFis
included.
19.5.6 Results
The critical pressure for the Upper Yring amounts to Pcrit = 12.84 bars. The initial and deformed
configurationispresentedbelow(Figure195).
Figure195:InitialanddeformedbuckledshapefortheupperYringwiththe
schematicboundaryconditions.
The critical pressure for the Lower Yring amounts to Pcrit = 135.3 bars. The initial and deformed
configurationsarepresentedbelow(Figure196).
362
ECSSEHB3224A
24March2010
Figure196:InitialanddeformedbuckledshapeforthelowerYringwiththe
schematicboundaryconditions.
Thesafetymargins(seeEquation193)withreferencetothebucklingloadcaseareverycomfortable,
superiorto100%bothfortheupperandlowerYring
ItisclearthatthebucklingisnotadimensioningloadcaseforthesepartsoftheLH2tank.
19.6 Cylinder
19.6.1 Overview
Forthecylinderpart,twotypesofanalysisweredone:
a. ApredimensioningusingtheNASASP8007method[2],
b. ThispredimensioningisconfirmedbyanaxisymmetricFEManalysis.
Inthefollowing,thetwomethodsarepresented.
19.6.2 Buckling analysis of the pressurized cylinder using the NASA SP-8007
Forthebucklinganalysis,thecylindricalpartoftheLH2Tankistreatedasawettedareabytheliquid
hydrogen(temperature20K)underinternalpressure(minimalvalueof2.255bars)andcompressive
load(throughaxialfluxandbendingmoment).
FortheevaluationofthebucklingloadandtheMOSagainstbucklingthemethoddescribedinNASA
SP8007 [2] is used. Within this analysis loop the factor of 1.3, which was applied formerly on the
bendingload,iswithdrawnduetotheexaminationsperformedbytheindustrialpartnersresultingin
minimumrequiredmarginsincaseofinteractingaxialandbendingloads.
Inaddition,totakeintoaccounttheimperfections,theKDFaccordingtoAlmroth[3]appliedformally
totheaxialfluxportionisreplacedbythelessconservativeKDFaccordingtoSeideasrecommended
inNASASP8007[2].
1 axi bend
MOS = 1 Kf = +
Kf crit _ axi bend _ crit 194

363
ECSSEHB3224A
24March2010
with
t axi pR
crit _ axi = E + + 195
R 3(1 ) 2
t bend
crit _ bend = E + + 0.8 pR 196
R 3(1 )
Axialload:
1 R

axi = 1 0.901(1 e 16 t
) 197
Bendingload:
1 R

bend = 1 0.731(1 e 16 t
) 198
Theimperfectionsensitivityisreducedbytheinternalpressurethrough:
3p R
( ) 199
= 0.24(1 e E t )0.75
Forinformation,withanaxialfluxaround500N/mm,thesafetymarginsarepositiveonthecylinder
forthebucklingmode.
ThefinalsafetymarginsarepresentedinTable192:
Table192:Finalsafetymargins
crit _ axi = 598 N / mm crit _ bend = 838 N / mm MOS = 17.5%

bend
Required MOS MOS req = 5% + 15% MOS = 12.5%
axi + bend
Final MOS MOS fin = MOS MOS req MOS = 4.9%
19.6.3 Axisymmetric FEM analysis

19.6.3.1 Overview
TheprojectrequiresthatthispredimensioningwillbeconfirmedbyaFEmodelanalysis.TheNASA
SP8007 is based on tests and consequently on a KDF approach taking mismatches, deviations in
geometry, boundary conditions or material properties into account. A confirmation by FEM would
requirea360modeltotaketheeffectofacircumferentialwavenumberoftheeigenmodeaswellas
theapplicationofthebendingfluxintoaccount.
19.6.3.2 Model used for this analysis

A 1.7 model is used with a solid mesh for the Yrings to give the first hint on the linear buckling
analysisofthetankcylinderapplyingtheFEMmethod.Theweldsmismatchesaretakenintoaccount
byaKDFofSeideasappliedwithintheNASASP8007[2].
19.6.3.3 Boundary conditions

TheboundaryconditionsarerepresentedbytheYringsandtheadjacentskirts.
364
ECSSEHB3224A
24March2010
19.6.3.4 Dimensioning load cases
Threedifferentloadcasesareconsideredforthisanalysis,onegroundloadcase,oneflightloadcase
(with a temperature gradient applied on the skirt) and one artificial load case (with a uniform
temperatureappliedonthewholestructure).
19.6.3.5 Results
UsingEq.101fromChapter10,thecalculatedmarginofsafetyisdefinedasfollows:

FEM axi +
MOS = 0.605
1 1910
pR
J p J ult limit
2
Where
andaxidefinedintheNASASP8007[2]
Jp asprojectmarginof15%
Jult asultimatesafetyfactorof1.25.
TheminimumMOSisobtainedforthegroundcaseandthevalueissuperiorto40%.
The FEM calculation gives a safety margin higher than the safety margin found by the analytical
method.
19.7 Elongated lower skirt (ELS)
19.7.1 Overview
As performed on the cylinder activities, two types of analysis were performed on the ELS (Figure
197):
a. AlinearanalysisapplyingaKDFapproach
b. Anonlinearcollapseanalysiswithimperfectionsensitivitystudy
Thisparagraphpresentsthetwomethodsusedonthecylinder.
H18 location
Figure197:ElongatedLowerSkirt
365
ECSSEHB3224A
24March2010
19.7.2 Linear analysis applying a KDF approach
19.7.2.1 Overview
TheFEMmodelusedforthelinearbucklinganalysis(Eigenvalueanalysis) isaglobalmodeltaking
intoaccountELS,dummyofISS,theLowerYring,theLowerdomeandtheCylinder(Figure198).
Figure198:FEMusedforELSbucklinganalysis
Theappliedloadsarethefollowing:
a. Internalpressure,
b. Twocases:fortemperature
1. forthepredeformedcalculation,agradientisappliedontheELS,
2. forthecalculationwithoutpredeformation,thehighestoccurringtemperatureisapplied,
c. Generalloads:maximumcompressiveloads.
TwoKDFareusedaccordingtoSeide[4]andKoiter[5]:
19.7.2.2 SEIDE
1

R
29.8
4
DxDy
ExEy

KDFseide = 1 0.901 1 e 1911

Where
Dx,Dy bendingstiffnessinlauncheraxisandcircumferentialdirection,
Ex,Ey extensionalstiffnessinlauncheraxisandcircumferentialdirection.
19.7.2.3 KOITER
w0 2
KDFkoiter = . (1 KDFkoiter )
27 (1 )
1912
t
With w0: flux eccentricity caused by the pre deformation due to the temperature and the internal
pressure.
Eigenvalue lim KDF

ThemarginofsafetyisdefinedasmentionedinEquation193: MOS = 1
lim J ult
366
ECSSEHB3224A
24March2010
Thecalculatedmarginofsafetyiscontainedbetween20%forthepredeformedanalysiswithaKDF
SEIDEapproachand0%forthenonpredeformedanalysiswiththeKDFKOITERapproach.
InordertoachievearealisticanalysisoftheELS,anonlinearcollapseanalysiswasperformed.
19.7.3 Nonlinear collapse analysis with imperfection sensitivity study

19.7.3.1 Overview
The purpose of the analysis is to determine criticalcollapse fluxes in nonlinear analysis considering
theinfluenceof:
a. pressureandtemperatureloads,
b. anonlinear,temperaturedependentmateriallaw,
c. geometricnonlinearity.
Collapse is not an isolated problem for a substructure of the whole LH2Tank structure, this
phenomenon appearsat system level. It is necessary to carry outthe collapseanalysis on the whole
LH2Tanksystemtoconsidertheboundaryconditionsasrealisticaspossible.Accordingtothefactof
relativelylargeradialdeformationcausedbytheloadcasetemperatureandtankpressure,thereexists
nobifurcationproblemforthefurtherfluxpathinanordinarysense.Thesepressureandtemperature
induceddeformationsoftheshellareprincipallyofbendingcharacter.Foracylinderwithcutouts,
preliminary analysis showed that the linear buckling analysis underestimates the critical loads. In
suchcalculations,itisnotpossibletodistinguishlocalorglobalbucklingmode.
Thegeometricaltolerancesonthediameterandthethicknessofthewallaretakenintoaccount.The
deformationcausedbytheloadcases,temperatureandpressureareabsolutevalues.
19.7.3.2 Modes calculations: determination of the different local and global modes
FortheELS,thewalledcylindersarehighlyimperfectionsensitiveduetothecylindricalcompression
loadcases.Thus,acompleteimperfectionstudyhasbeenperformedfortheELS.
As a standard method for nonlinear collapse analysis eigenmodes or linear combinations of them is
appliedontheundeformedstructure.Thispredeformedstructureisusedasinitialconfigurationfor
the nonlinear analysis. This analysis was performed with MSC/NASTRAN 70.7. The eigenvalues and
eigenmodesarepresentedinTable193.
Allthemodesexceptmode18canbeconsideredaslocal.
Figure199illustratessomeofthecorrespondingeigenmodes
The adaptation of FEmesh has been performed in order to ensure the correct representation of the
eigenmodesthataremainlylocal,intheneighbourhoodofthecutouts.
367
ECSSEHB3224A
24March2010
Table193:EigenvaluesobtainedontheELS
368
ECSSEHB3224A
24March2010
mode 1 mode 3
mode 4 mode 7
Figure199:EigenmodesobtainedontheELS(part1).
369
ECSSEHB3224A
24March2010
mode 8 mode 9
Figure199:EigenmodesobtainedontheELS(part2).
19.7.3.3 Combinations of the modes for considering realistic imperfections

The main problem is to find the most unfavourable geometrical imperfection, which provides the
lowestcollapseload.
Threedifferentimperfectionshavebeenconsidered:
a. Imperfection n1 is a linear combination of the modes 1, 3, 4, 8. This combination covers the
wholecircumferenceandissymmetricaltotheH18/2interface,andthusitisaglobalmode.
b. Imperfection n2 is mode 18. This combination covers the whole circumference and can be
consideredasaglobalmode.
c. Imperfectionn3isalinearcombinationofmodes1,4,7,8,9,10.Thiscombinationcoversthe
wholecircumferenceandisantisymmetricaltotheH18/2interface.
TheH18/2interfacehasbeenusedasacriterionbecauseitseemsthattheinitialfailureisintheregion
ofH18/2andH19/2.
19.7.3.4 Amplitude imperfection examination

Thecalculationsoftheminimumloadfactortakesintoaccountasensitivityanalysisontheamplitude
of these imperfections. It is necessary to normalize the amplitude. For instance, for ELS, the ratio
betweentheamplitudeoftheimperfectionandthetotalheightoftheskinandthestringerisequalto
1.Inthiscasethemaximalamplitudeoftheimperfectionisequalto35mm.
370
ECSSEHB3224A
24March2010
19.7.3.5 FEM Model, material and boundary conditions
Themodelusedforthesecalculationscontainsaround69000elementsand60000nodes.Quadrilateral
fournodesshellelementsareused.ThemeshingisfinefortheELSandrelativelycoarsefortheother
structures. The boundary conditions concept is to exclude failure modes in the adjacent structures.
Thestructureisrestrainedatthebottomintheradialdirection.Theaxialdirectionisrestrainedmuch
closertotheELS(Figure1910).
The material behaviour is elasticplastic with isotropic hardening. The properties are temperature
dependent. The stressstrain curves are given for the temperature 20K and 293K. For all other
temperaturesthematerialpropertiesarelinearlyinterpolated.
Figure1910:Boundaryconditions
19.7.3.6 Loads
Thesequenceofloadsapplicationisthefollowing:
a. Temperature,
b. Pressureload,
c. Themostcriticalloads.
Theloadsappliedaretemperatureload,apressureof3,3barsand2,6barsandthegeneralloads.
Twoloadcasesaredefinedandrecalledhereafter:
371
ECSSEHB3224A
24March2010
Table194:Summaryoftheloadcasesconsidered
LoadCase1(LC1)
Temp293K Temp20K OtherTemp Pressure Flux
Loadstep1 ISS,VEB UpperDome, ELS,LowerY 3,3bar None
dummy, ring UpperDome,
LowerDome,
UpperSkirt
Inner LowerDome,
Bulkhead, InnerBulkhead,
LowerYring, Cylinder
Cylinder
Loadstep2 Remaining UpperDome, ELS,LowerY 3,3bar YES
structure, ring UpperDome,
LowerDome,
Upperskirt LowerDome,
Inner
Cylinder

LoadCase2(LC2)
Temp293K Temp20K OtherTemp Pressure Flux
Loadstep1 ISS,VEB UpperDome, ELS,LowerY 2,6bar None
dummy, ring UpperDome,
LowerDome,
UpperSkirt
Inner LowerDome,
Cylinder
Loadstep2 Remaining UpperDome, ELS,LowerY 2,6bar YES
structure, ring UpperDome,
LowerDome,
Upperskirt LowerDome,
Inner
Cylinder
19.7.3.7 Analysis procedure

Inthisnonlinearproceduretheloadsareappliedinincrementsandthedisplacementsandthestate
variablesarecomputedforeveryloadstep.Thestiffnessofthestructureiscorrectedaftereveryload
increment as well as the geometry of the structure. The structure fails when the stiffness matrix
becomesnonpositivedefinite.
372
ECSSEHB3224A
24March2010
Theindividualloadsareappliedtothestructureincrementally.Firstthetemperatureinseveralsteps,
then the pressure and the flux load is applied in several subcases. The direction of the flux load is
updatedwiththedeformation.
19.7.3.8 Nonlinear buckling calculation and results

Theresultsofthisthesenonlinearcollapseanalysesarethefollowing.
Although the boundary conditions are set in manner that the collapse of adjacent structures is
precluded as much as possible, the collapse does not occur in ELS itself but the buckling mode is a
global one. The geometrical imperfections have been dominated by the axisymmetrical pre
deformationduetothetemperatureandtankpressure.
Thefirstcalculationisacollapsedue totheloadcase1(LC1)withoutgeometricalimperfection.The
loadfactorassociatedisaround3(seeFigure1911).
Figure1911:CollapseduetoLC1withoutgeometricalimperfection
The other calculations are to combine the load case 1 (LC1) and the imperfection defined in the
paragraph19.7.3.3.
Thelowestloadfactorisfoundbyacombinationoftheloadcase1(LC1)andtheimperfectionn3
(definedinparagraph19.7.3.3)withamplitudeof35mm.Theloadfactorobtainedislowerthan2.The
resultingcollapsemodeisshowninFigure1912.
373
ECSSEHB3224A
24March2010
Figure1912:CollapseduetoLC1withtheimperfectionn3withamplitudeof35mm
19.8 Test Campaign
19.8.1 Qualification test plan

Following all the analysis activities, a complete qualification test plan was established. In this
qualification test plan, in addition to the traditional checks of flexibilities and constraints, coupled
testsweredefinedtocheckthebucklingofthedifferentpartsconstitutingthetank.
Thesecoupledtestsarethefollowing:
a. UpperskirtandUpperYringbucklingqualification,
b. ELSandLowerYringbucklingqualification,
c. Cylinderbucklingverificationtest.
These tests are performed on a complete tank with adjacent structures which simulate the Vehicle
EquipmentBay(VEB)attheupperinterfaceandInterTankStruts(ITS)atthelowerinterface(Figure
1913).
374
ECSSEHB3224A
24March2010

Figure1913:TestsetupfortheARIANE5ECALH2tank
19.8.2 Test results

19.8.2.1 Upper skirt and Upper Y-ring buckling qualification test
Forthistest,theloadintroductionconfigurationispresentedinthefollowingsketch(Figure1914).
TheloadingsequenceisasinFigure1915.
Loading by pressure in tank and longitudinal

fluxes at the upper interface
Figure1914:LoadintroductionconfigurationfortheupperskirtandtheupperYring
375
ECSSEHB3224A
24March2010
Figure1915:LoadingsequencefortheupperskirtandtheupperYring
During the test, the Upper skirt and the Upper Yring substructure have seen load levels slightly
higherthantheonesspecified.Nobucklingwasidentifieduptothequalificationlevel(J=1,25).
There is a relative good correlation between the prediction and the behaviour of the qualified sub
structure strain gages. A higher bending moment (w.r.t. the prediction) is observed in the cylinder,
which is mainly due to the behaviour under pressure loading. During the correlation activities, the
behaviourhasbeenexplainedbyintroducinginthemodeltheglobalshapedefectontheYaxisand
thelocalmismatchduetothewelding.
19.8.2.2 ELS and Lower Y-ring buckling qualification test

Forthistest,theloadintroductionconfigurationispresentedinthefollowingsketch(Figure1916).
Loadingbypressureintank,longitudinalfluxesatthe
upperinterfaceandloadscomingfromtheLOXtank
Figure1916:LoadintroductionconfigurationfortheELSandthelowerYring
376
ECSSEHB3224A
24March2010
TheloadingsequenceisinFigure1917.
Figure1917:LoadingsequencefortheELSandthelowerYring
Duringthetest,allstraingaugesontheELSreturnedtozero.
An ovalisation of the ELS has been noticed through the analysis of the LVDT radial response. This
ovalizationwasduetothepresenceofcutoutsalongthecircumferenceoftheELS.
The qualification levels have be reached and even slightly exceeded. No evidence of buckling was
foundneitherinstraingaugesnorinLVDT.
19.8.2.3 Cylinder buckling verification test

Threebucklingtestshavebeenperformedonthecylinder.Thedifferencebetweenthethreetestsisthe
level of the pressure and the loads associated. For these tests, the loadintroduction configuration is
presentedinthefollowingsketch(seeFigure1918).
Loadingbypressureintankandlongitudinalfluxesatthe
upperinterfaceappliedwithacircumferentialdistribution
Figure1918:Loadintroductionconfigurationforthecylinder
377
ECSSEHB3224A
24March2010
For the cylinder buckling test configuration, a laser profiler is used to scan the most loaded fibre
meridian profile. A good correlation was shown between the laser profiler and the LVDT
measurements.
Thetestmodelhassustainedthequalificationloads.Noevidenceofbucklingcouldbefoundonthe
straingauges,LVDTnorlaserprofileswhencomparingtheirresponsesbeforeandafterthetests.
19.9 Conclusion
All LH2 tank substructures prone to buckling were directly qualified by a static test. All the test
resultswereinaccordancewiththetestsprediction,intherangeoftheacceptableuncertainties.
Itisinterestingtonoticethatspecificmeasurementmeanscanbeimplemented,asthelaserprofileron
themostloadedfibreofthecylinder,inordertogivemoreaccuratemeasurements.
19.10 References
[1] Anonym,StructureDesign,DimensioningandTestSpecifications,A5SG1X10ASAI,
Aerospatiale,Espace&Defence,Issue5,July1996
[2] Anonym,BucklingofThinwalledCircularCylinders,NASASP8007,Aug.1968
[3] Almroth,B.O.,Burns,A.B.andPittner,E.V.,DesignCriteriaforAxiallyLoadedCylindrical
Shells,J.Spacecraft&Rockets,Vol.7,No.6,pp.714720,1970
[4] Weingarten,V.I.,Morgan,E.J.undSeide,P.,Elasticstabilityofthinwalledcylindricaland
conicalshellsunderaxialcompression,AIAAJournal,3(3),500505,1965
underaxialcompression,Koninkl.Ned.Akad.Wetenschap.Proc.B66,pp.265279,1963

EAP EtagePrimaire
ELS ElongatedLowerSkirt
EPC EtagePrimaireCryognique
ESC EtageSecundaireCryognique
FEM FiniteElementModel
ISS InterStageStructure
KDF KnockdownFactor
LVDT LinearVoltageDisplacmentTransducer
MOS MarginofSafety
VEB VehicleEquipmentBay
378
ECSSEHB3224A
24March2010
20
Buckling of the Inner Dome of the
ARIANE 5 Upper Stage ESC-A
20.1 Overview
20.1.1 General
Theliquidhydrogentank(LH2tank)ofthecryogenicupperstage(ESCA)oftheARIANE5hasan
unconventionaldesignforapressurizedpropellanttank,seeFigure201.ThesocalledInnerDomeof
theLH2tankissubjectedtocompressiveloadsincaseofinternalpressure.ThereforetheInnerDome
shellisdesignedcarefullyagainstbuckling.
Figure201:ThecryogenicupperstageESCAofARIANE5
TheInnerDomeneedsanorthotropicstiffeningwhichmakesthispartheavyandmoreexpensivein
manufacturing.Buttheoverallstagedesignunderthegivenconstraintsturnedouttobeanoptimum
withsuchadesign.Theconstraintsandrequirementshavebeen:
379
ECSSEHB3224A
24March2010
Theoveralllengthofthestageshouldbeminimized
The liquid oxygen tank (LOX tank) design including engine and equipped thrust frame is a
carryoverdesignfromthepreviousARIANE4upperstage
DuetocommunalitytheLH2tankmanufacturinglineisbasedonthetankmanufacturingline
ofthelowerstage(EPC).Thereforetheouterdiameterofthecylindricalparts(5.4m)andthe
sphericalshellradii(3m)oftheupperandlowerdomeareidentical.
The Inner Dome of the LH2 tank is a spherical shell with a radius of 2000 mm. The maximum
operating pressure together with the hydrostatic pressure (inertia loads) leads to a limit pressure of
3.45bar(deltapressure).Takingintoaccountasafetyfactorof1.25theInnerDomehastowithstand
againstbucklinganultimatepressureof4.31barswithoutfailure.
For simplicity other loads (temperature distribution, local loads) are not considered in this article.
Alsootherdesigncriteria(e.g.damagetolerance)arenotregardedhere.Thisarticleconcentrateson
bucklingonly.
ThematerialoftheInnerDomeisAl2219.Makinguseofthecryogenictemperatureofhydrogen(20
Kelvin)thematerialpropertiesarehere:
Youngsmodulus: E=80000N/mm2
Poissonratio: =0.3
YieldStress: 0.2=400N/mm2
Density: =2800kg/m3
Fororthotropicstiffeningagainstbucklingdifferentintegrallystiffenedtypesofdesign(machinedout
ofthickplates)arefeasible,e.g.Wafflegriddesignin0/90orin+/45orIsogriddesign(seeFigure
202)
b
s =d + t
d t
Figure202:Wafflegrid(0/90quadratic)andIsogridDesign
The article describes how to find an optimum design withstanding the 4.31 bar ultimate load. The
application and usefulness of the so called Three Level Method is described. The Three Level
Method(seechapter11)usesforthedifferentlevel:
Level1:Analyticalsolution
Level2:Semianalyticalsolution
Level3:3Dnumericalanalysis,e.g.FiniteElementMethod
380
ECSSEHB3224A
24March2010
Particularlyhighlightedistheinfluenceofimperfectionsonthebucklingloadandthetreatmentofthe
imperfectionsinthedifferentdevelopmentphases.
Figure203:TheinnerdomeoftheLH2tankoftheARIANE5upperstageESCA
inwafflegriddesign
20.1.2 Buckling Modes:

Both,thewafflegridandtheisogridshell,mayfailaccordingtofollowingbucklingmodes:
a. SkinBuckling
The skin between the ribs buckles. Due to the almost harmless postbuckling behaviour of
plates the load capability of the structure can be still provided (to be proven). Anyhow, for a
pressurizedpropellanttankskinbucklingisnotallowed.
b. RibBuckling
Thestiffeningribbuckles.Thislocalbucklingisnotallowedbecausethismodeinitiatesgeneral
instability.
c. GeneralInstability(GlobalBuckling)
This is a global buckling mode of the shell where the stiffeners and the skin buckle together.
Thisbucklingmodewillleadtocollapseandhastobeavoided.
381
ECSSEHB3224A
24March2010
20.2 Buckling investigation levels
20.2.1 Level-1 investigations

Level1investigationsarebasedonsimplifiedanalyticalsolutions.Theanalyticalsolutionsforgeneral
instability(globalbuckling)arebasedonfollowingassumptions:
The shell is axisymmetric, the displacements depending on the circumferential coordinate are
harmonicfunctions
The orthotropic wall stiffness can be regarded as smeared out and is constant in meridional
direction
Theboundaryconditionsaresimplified(e.g.simplysupported)
Thedisplacementsdependingonthemeridionalcoordinatearesupposedtobesinusoidal.
The influence of imperfections on the theoretical buckling results achieved by the analysis can be
takenintoaccountbyKnockDownFactors(KDFs).TheseKDFscanbetakenfromstatisticevaluated
empiricaldata(comparisonoftheoreticalbucklingvalueswithtestdata).Awellknownapproachfor
orthotropiccylindricalshellsunderaxialcompressionisAlmrothsformula.Duetotheanalogyofthe
isotropiccylindricalshellsunderaxialcompressionwiththesphericalshellunderexternalpressure(if
theradiusofthecylinderisequaltheradiusofthesphere)andtheirequalsensitivitytoimperfections
(seeRef.[1])itcanbeassumedthatitisthecasefororthotropicshellstoo.
AccordingtoAlmroth[1]theknockdownfactor KDF isgivenwith99%that KDF N cl < N exp by
0.54
R
KDF = 6.48 201
se
Almrothsformulaforequivalentthickness se is:

D11* + D22*
se = 5.46 202
A11 A122 / A22

B112
D = D11
*
11 203
A11

B222
D = D22
*
22 204
A22

and Aij, Bij, Dij are the ABD stiffness coefficients for orthotropic wall construction (see Sections 7.2.2
and17.2.2)
The KDFs taken from statistic evaluation of test data do not distinguish between individual
imperfections.TheKDFscorrecttheinfluenceofimperfections(geometricalimperfectionsorresidual
stresses)andsimplificationsintheanalysismethod(boundaryconditions,prebucklingdeformations)
bystatisticalenvelopingoverallkindofshellsandmanufacturingprocesses.
Alternatively Koiters approach [2] can be used which includes in the analysis a sinusoidal
axisymmetric geometrical imperfection with amplitude wo By this way the sensitivity of a
axisymmetricharmonicimperfectiononthebucklingresultcanbeevaluated.
382
ECSSEHB3224A
24March2010
Koitersapproachwasderivedforisotropiccylindricalshellsunderaxialcompression.Makinguseof
the analogy to spherical shells under external pressures, the KDF can be calculated according to
equation205.
w0 2
(1 KDF )
2
KDF =
27 (1 v )
s 2 205

ForthefirstdesignAlmrothsformulahasbeenapplied.
Thestiffenedshellhastowithstandboth,globalbucklingandlocalbucklingoftheribsandtheskin.
Tofindanoptimumpreliminarydesign(byparametricoptimization)itisrecommendedtostartwith
theskinbuckling.Theskinbetweentheribscanberegardedasaplatewhichissimplysupportedat
allsides,whichisaconservativeassumption.
Theequationsforplatebuckling(seechapter16)foranisotropicmaterialwithPoissonratio0.3are:
Forwafflegrid(quadratic)withcompressivestressinbothdirections( x = y = ):
2

t
crit = 1.82 E 206
h

Forisogridwithcompressivestressinbothdirections( x = y = ):
2

t
crit = 3.39 E 207
h

Orwitha=h/cos(30)
2

t
crit = 4.52 E 208
a

The ultimate membrane fluxes in the spherical shell are equal in meridional and circumferential
direction.Withpult=4.31barthecriticalfluxesare:

1
N xult = N yult = N ult = pult R = 0.5 0.431 N / mm 2 2000 mm = 431 N / mm 209
2

Withthedimensionlessparameter

bd
= 2010
ht

Itispossibletoformulatethestresswhichtakestheribcrosssectionintoaccount:

Nx
x =
t (1 + )
2011

383
ECSSEHB3224A
24March2010

Ny
y = 2012
t (1 + )

Atthebeginningofthedimensioningtheribcrosssectionisnotknown.Thereforethenecessaryskin
thickness between the ribs needs to be evaluated for different parameter alpha. In the following a
procedurefortheparametricoptimizationisgivenindetailforthewafflegriddesignwithquadratic
pattern.Theisogriddesignoptimizationisanalogous.
Forwafflegridthenecessaryskinthicknessagainstskinbucklingisaccordingtoequation206.
1/ 3
N ult h 2
tcrit = 2013
1.81E (1 + )
With a given rib height d the minimum thickness against rib buckling can be determined with the
formula for plate buckling in case of a long strip (simply supported and free, longitudinal
compression):

N ult
bcrit =d 2014
t (1 + ) 0.45E

Ontheotherhandtheribthicknessbneedstobecompatibletothegiven,therefore(seeequation
2010):
ht
b= 2015
d
The four free parameters in the optimization (h, d, t, b) can be reduced by taking into account
geometricandmanufacturingconstraints.
Afirstconstraintisthenumberofpocketswhichisrelatedtotheparameterh.Asimpleestimationof
thenumberofpocketsispossiblebydividingthesurfaceareaofthesphere(halfsphericalball)bythe
areaofonepocket.
Halfsphericalballsurface
A = 2R 2 = 2 3.14 2000 2 mm 2 = 25.1 E 6 mm 2 2016
Wafflegrid(quadratic)
n pockets = A / h 2 2017
Isogrid

A
n pockets = 2 2 cos ( 30 ) = 1.73 A / h
o 2
2018
h

A 2
= = 2.31A / h 2
h cos ( 30o )

2 2019

384
ECSSEHB3224A
24March2010
Thenumberofpocketsfordifferenth,orarespectively,is:
Table201:Numberofpocketsfordifferenth
Wafflegrid Isogrid
h[mm] pocket a[mm] pocket
number number
50 10040 50 23192
60 6972 60 16106
70 5122 70 11833
80 3922 80 9059
90 3099 90 7158
100 2510 100 5798
110 2074 110 4792
120 1743 120 4026
Here is obvious that an isogrid design means significantly more pockets (factor 2.31). The mass
advantageofanisogriddesignneedstobebalancedwiththehighermanufacturingcosts.
Preciselythenumberofpocketsforasphericalshellismoreduetogeometricreasons.Forexample,
the width of the pockets of the wafflegrid in circumferential direction will be getting smaller and
smaller along the meridional coordinate in direction to the pole. If the width in circumferential
directionattheequatorish,atanelongationof60itisonlythehalf(h/2).Therefore,inpractisethere
willbeanewarrangementofthepocketsat60.
Twopocketswillbecombinedtoonepocketwhichstartsagainwiththewidthh.Attheelongationof
75.52thewidthincircumferentialdirectionisagainh/2andtwopocketswillbecombinedoncemore.
The meridian can be therefore divided into three sectors (see Figure 204) with different number of
pocketspercircumference.
Figure204:Threesectorswithdifferentnumberofpocketsincircumference
Taking into account these three sectors and selecting a number of pockets in circumference at the
equatorwhichcanbedividedbytwoandfour,amoreprecisenumberofpocketscanbeachievedfor
thewafflegrid.Thepocketswillnotkeepaquadraticformifthedeltaelongationiskeptconstant.The
consequence is that the smeared wall stiffness in meridional direction is getting stiffer towards the
pole(forconstantribcrosssection)inasector,butthesmearedcircumferentialstiffnessisnotaffected
(itremainsconstant).
385
ECSSEHB3224A
24March2010
Alternatively the length of the pockets in meridional direction can be adapted in order to keep the
quadraticformofthepockets,butthiswouldleadtosignificantmorepocketsandisnotconsidered
here.
Table202:Numberofpocketsfordifferentconfigurations
h=78.5mm Numberofpocketsin Numberofpocketsin

5120Pockets circumferential meridionaldirection
direction
Sector1(060) 160 27
Sector2(6075.52) 80 7
Sector3(75.5290) 40 6

direction
Sector1(060) 144 24
Sector2(6075.52) 72 6
Sector3(75.5290) 36 6

direction
Sector1(060) 120 20
Sector2(6075.52) 60 5
Sector3(75.5290) 30 5
Taking into account realistic pocket numbers only a few variation of the parameter h have to be
carriedoutfortheoptimization(e.g.hereonlyh=78.5mm,h=87.3mmandh=104.7mm).
Afurthermanufacturingconstraintisthethicknessofthebaseplate.Herethefinaltotalthicknesshas
beenlimitedto:
s = t + d = 25 mm 2020
Thereforetheparameterd(ribheight)dependsonlyontheskinthicknesst(d=25mmt).Smaller
values for s make no sense, due to the loss of bending stiffness which is driven by the rib height.
Higherplatethicknesswouldbegoodformassoptimization.
Additionalmanufacturingconstraintstobeconsideredintheoptimizationprocessare:
AminimumthicknessoftheskinandtheribthicknessformachiningandforNonDestructive
Inspection(NDI)ofinitialcracksfordamagetolerance
Theriskofmanufacturinganomalies(e.g.machiningoflocalunderthickness)
386
ECSSEHB3224A
24March2010
With these restrictions the parametric investigations in order to find the optimum are only to be
carriedoutfor3valuesofh,incombinationwithvariousvaluesof.
Firsttheskinthicknessaccordingtoequation2013iscalculated.
Thentheribheightiscalculated:d=25mmt
Thenthethicknessoftheribisdeterminedresultingfromthegivenalpha:
b = ht / d 2021
Theminimumribthicknessforribbucklingisdeterminedaccordingtoequation2014andthen
it is compared with the rib thickness resulting from equation 2021. If the thickness resulting
fromequation2021islowerthantheminimumthicknessforribbucklingthisisnotafeasible
design.Iftheribthicknessresultingfromequation2021isgreaterthantheminimumthickness
forribbucklingthisdesignisapotentialsolution.
Allpotentialsolutionsareanalysedforgeneralinstability(globalbuckling)takingintoaccount
theKDF.WhenthecriticalglobalbucklingincludingtheKDFmeetstheultimatepressure,the
optimumforthegivenparameterhhasbeenfound.
Bythiswaytheoptimumtandbforeachhisdetermined(t&bopt)
Theoptimumdesignforalightweightstructureisfoundbycomparingthemass(searchforthe
minimummass)ofthedifferentoptimums(t&bopt)
Table 203 shows the parametric results for different h. In the Table 203, the evaluation of the rib
thickness according equation 2021 is called balpha and the minimum thickness for rib buckling is
calledbmin.
Themassoftheorthotropicstiffenedshellcanbequicklyestimatedbythesmearedthicknessofthe
ribs
Wafflegrid:msimple=At(1+2)
Isogrid:msimple=At(1+3)
Theexactnetmass(mnet)canbecalculatedbyconsideringtheribmasspreciselyinthethreesectors.
Table203:Parametricresultsfordifferenth
h alpha t d=s-t b-alpha bmin m-simple pcrittheo KDF pcrit
[mm] [mm] [mm] [mm] [mm] [kg] [N/mm2] [N/mm2]
78,54 0,1 2,56 22,44 0,89 1,46 215,70 no feasible design (rib buckling)
78,54 0,165 2,51 22,49 1,44 1,44 234,54 0,76 0,49 0,37
78,54 0,2 2,48 22,52 1,73 1,43 244,46 0,82 0,5 0,41
78,54 0,22 2,47 22,53 1,89 1,42 250,06 0,85 0,51 0,44
87,27 0,131 2,72 22,28 1,39 1,39 241,12 0,74 0,46 0,34
87,27 0,19 2,67 22,33 1,98 1,37 259,23 0,86 0,5 0,43
87,27 0,2 2,66 22,34 2,08 1,37 262,26 0,88 0,5 0,44
104,72 0,088 3,11 21,89 1,31 1,30 257,02 0,71 0,42 0,3
104,72 0,15 3,05 21,95 2,18 1,28 278,92 0,89 0,48 0,42
104,72 0,16 3,04 21,96 2,32 1,28 282,39 0,91 0,48 0,44
Theoptimumdesignparameters(t&bopt)forthewafflegridarehighlightedinTable203.
387
ECSSEHB3224A
24March2010
From the Level1 optimization, it turned out that the isogrid design is the lighter design but with
muchmorepocketsandincreasedmanufacturingcostsandrisks.
Thewafflegriddesignwasselectedandthepreliminarydesignwasrefinedandinvestigatedindetail
withLevel2andLevel3methods.
20.2.2 Level-2 investigations

The Level2 investigations are done with semianalytical tools which suppose axisymmetry of the
structure(e.g.BOSORorASTRAcode).Theorthotropicwallstiffnessisregardedassmeared.
TheherepresentedinvestigationsfromRef.[3]havebeendonewiththeASTRAcode.Thecodecan
takeintoaccount:
Therealisticboundaryconditions(byadjacentstructures)
Meridionalvariablestiffness(stepwise)
Axisymmetricgeometricalimperfections
AxisymmetricNonlinearities(prebucklingdeformation,plasticity)
Bythiswayamorerealisticviewabouttheinfluenceofimperfectionsispossiblewithoutstartingthe
timeconsuming3DFiniteElementcodes.
The sensitivity study of the influence of geometrical imperfections on the buckling load takes into
accountalotofpossibleimperfectionshapes.
It is recommended by Arbocz [4] to perform investigations for the lowest mode of the perfect shell,
someneighbouringmodesandeventuallysomecombinationsofdifferentbucklingmodes.Ofcourse,
other types of shapes are possible, e.g. polynom functions (see Galletly, Ref. [5]). For Level1
investigationsoneislimitedinthesimulationofaxisymmetricimperfectionshapes.
Importantaretherealisticimperfectionshapescausedbythemanufacturing(learnedbyexperience).
Koga and Hoff [6] introduced for the spherical shell two type of imperfection, type1 a dimple like
shapedescribedbycosinefunctionandtype2describedbyaderivationofthecurvatureradius(see
Figure205).Iftheseimperfectionsarelocatedatthepole,anaxisymmetricmodellingcanbeused.
388
ECSSEHB3224A
24March2010
Figure205:ImperfectionShapesofType1andType2accordingKogaandHoff
Koga and Hoff studied the influence of this type of imperfection for isotropic spherical shells and
investigateddifferentparametersofsize(angle)andimperfectionamplitudewo.
Introducingadimensionlessparameter,whichisdefinedforisotropicshellsaccordingtoequation
2022 and for orthotropic shells according to equation 2023, a comprehensive overview of Knock
Downofperfectshellispossible.
(
ISO = 4 12 1 2 ) R
s
2022
(
ORT = 4 12 1 2 x)
t R2
Ix
2023

where
tx=smearedthicknessinthemeridionaldirection
= t (1 + )
Ix=bendingstiffnessinthemeridionaldirection[6]
bd t
3 2 2
3 t
= 1 + 1 + + 2
12 h 1 + d d
ForisotropicshellsKogaandHoff[6]foundout,thatfortype1theworstcaseimperfectionsize(angle
)isnear=4andfortype2near=3,independentfromtheimperfectionamplitude.Fororthotropic
stiffenedshellsitissimilar,theminimaarebetween2and4.
TheresultoftheimperfectionsimulationfortheisogriddesignisgiveninFigure206andFigure207.
Thecarriedoutanalysisconsideredgeometricalnonlinearity.
389
ECSSEHB3224A
24March2010
Figure206:Isogriddesignwithsimulationofimperfectiontype1(nonlinear
analysis)
Figure207:Isogriddesignwithsimulationofimperfectiontype2(nonlinear
analysis)
TheminimaaresummarizedinFigure208andcomparedtoclassicalresultsofisotropicshells(Koiter
[2] and Galletly [5]). The theoretical buckling pressure is here the Level1 formula according to the
Isogrid Handbook [7]. The predicted Kock Down for the perfect shell results mainly from pre
buckling deformations, which are consideredin the ASTRA code. It can be seen that for the isogrid
design,thesameimperfectionsensitivitycanbeobservedasfortheisotropicshell,iftheimperfection
amplitude wo is related to the total thickness s (and related to the skin thickness t for the isotropic
shell).
390
ECSSEHB3224A
24March2010
Figure208:Imperfectionsensitivityofisogriddesigncomparedtoclassical
isotropicshell
Thefinalchosendesignwasawafflegriddesign.Thesensitivityofthefinalchosenwafflegriddesign,
includingdesigndetailslikeanunstiffenedpolecap,isshowninFigure209.
Figure209:Imperfectionsensitivityofthefinalchosenwafflegriddesignwithan
unstiffenedpolecap
20.2.3 Level-3 investigation

Adetailed3DFiniteElementanalysiscanincludenonaxisymmetricstructuraldetailsandcanreflect
therealstiffeninginsteadofsmearingtheribs.
IntheLevel2investigationstheimperfectionsimulationwasdoneforaxisymmetricimperfection.But
theimperfectionscanbepresentatarbitrarylocations,sothatanonaxisymmetricsimulationshould
be carried out. Additional to type1 and type2 imperfection around the pole, different locations of
type1andtype2havebeeninvestigatedwithadetailedFiniteElementmodel.
ThemeshoftheFiniteElementmodelisshowninFigure2010.Itreflectsthefinalchosenwafflegrid
designwithanonstiffenedpolecap(seeFigure2011).Figure2012showsthemodelledimperfection
oftype1atanelongationof45(stronglyscaled).Detailsofthemeshwithgeometricimperfectionscan
beseeninFigure2013andFigure2014(stronglyscaled).
391
ECSSEHB3224A
24March2010
Figure2010:FiniteElementmeshofthefinalchosenwafflegriddesign
Figure2011:FiniteElementmeshofthefinalchosenwafflegriddesignwithdetail
ofpolecap
392
ECSSEHB3224A
24March2010
Figure2012:Meshincludingatype1imperfectionatanelongationof45(strongly
scaled)
Figure2013:Meshoftype1imperfection(stronglyscaled)
Figure2014:Meshoftype2imperfection(stronglyscaled)
393
ECSSEHB3224A
24March2010
Figure2015showsthebucklingmodeoftheimperfectshell(type1atelongation60).
Figure2015:Bucklingmodeofimperfectshell(type1at60elongation)
The result of the nonlinear buckling analysis including an imperfection of type1 and type2 with a
magnitudeofw0/s=0.2ispresentedinFigure2016andFigure2017below.AKnockDownof0.5is
herepossible.
0,70
60 from Pole
45 from Pole
0,65
Pcri/Pcri-Theo
0 from Pole
0,60
0,55
0,50
1,50 2,00 2,50 3,00 3,50

Figure2016:Imperfectionsimulationwithtype1forw0/s=0.2
0,65
0,60 60 from Pole

Pcri/Pcri-Theo
45 from Pole
0,55 0 from Pole
0,50
0,45
1,50 1,75 2,00 2,25 2,50

Figure2017:Imperfectionsimulationwithtype2forw0/s=0.2
394
ECSSEHB3224A
24March2010
20.3 Conclusion
The Level1 investigations are used to find a preliminary optimum design. Imperfections are taken
into account by empirical KnockDown Factors. Simulations of the influence of imperfections are
possiblewithLevel2andLevel3methods.
TheseinvestigationsarerefinedwithLevel2methods,considering
therealisticboundaryconditions(adjacentstructures)
differentmeridionalstiffness(stepwiseconstant)
axisymmetricgeometricimperfections
influenceofplasticityandnonlinearityofprebucklingdeformation
Thedetailed3DFiniteElementModel(aLevel3method)isfinallyusedtoinclude
nonaxisymmetricstructuraldetails
nonaxisymmetricimperfections
nonaxisymmetricnonlinearities
Bythisapproachalightweightdesigncanbefoundefficientlywithsufficientconfidence.
The investigations with Level 1, 2, and 3 methods serve to find a design which includes expectable
imperfections. The real size of imperfections can only be experienced in the applied manufacturing
process. It turned out by measurement, that type 1 imperfections at the pole are realistic and
sometimesexceedtheexpectedlimitofw0/s=0.2(inthatcasearepairisinitiated).Furtheritturned
out by measurement, that the residual stresses due to the forming process in the dome have a
magnitudeof50N/mm2,whichisquitehighcomparedtothestresslevelof120N/mm2duetointernal
pressure.
Thegeometricimperfectioncanbecontrolledeasilyintheproduction,butitremainstheuncertainty
of residual stresses. Therefore it was decided to perform for each produced dome a proof test for
bucklingwithapressurelevelfactor1.05toreducetheriskoffailureinflight.
20.4 References
[1] Almroth,B.O.,Burns,A.B.andPittner,E.V.,Designcriteriaforaxiallyloadedcylindrical
shells,JournalofSpacecraft&Rockets,Vol.7,No6,pp.714720,1970.
underaxialcompression,Koninkl.Ned.Akad.Wetenschap,Proc.B66,pp.265279,1963.
[3] ry,H.,Reimerdes,H.G.,Schmid,T.,Rittweger,A.AndGmezGarcia,J.,Imperfection
sensitivityofanorthotropicshpericalshellunderexternalpressure,Int.JournalNonLinear
Mechanics,Vol.37,pp.669686,2002.
[4] Arbocz,J.,PotierFerry,M.,Singer,J.andTvergaard,V.,BucklingandPostBuckling,Lecture
NotesinPhysicsNo.288,SpringerVerlag,Berlin,Heidelberg,1987,pp.83142.
[5] Galletty,G.D.,Blachut,J.andKruzelecki,J.,Plasticbucklingofimperfecthemisphericalshells
subjectedtoexternalpressure,Proc.Inst.Mech.Eng.201,1987.
[6] Koga,T.andHoff,N.J.,Theaxisymmetricbucklingofinitiallyimperfectcompletespherical
shells,Int.JournalSolids&Structures,Vol.3,pp.679697,1969.
[7] Meyer,R.R.,IsogridDesignHandbook,McDonnellDouglasAstronauticsCompanyWest,
MDCG4295A,PreparedforNASAMarshallSpaceFlightCenter,ContractNAS828619,1973.
395
ECSSEHB3224A
24March2010
21
Non-Linear Stability Analysis of ARIANE 5
Front Skirt (JAVE)
Attention: it is recommended to view and print this chapter in colour for

optimumresolution.
21.1 Overview
This chapter presents the results of the nonlinear analyses, which have been performed for the
prediction of the response of the EPC front skirt (JAVEC) due to the action of specified load cases.
These are defined in the dimensioning load case document, which is specified by ASTRIUM and
deliveredasapplicabledocumenttoMTAerospace.Thecriticalityofanumberofloadcasesonthe
response of the front skirt isinvestigated bylinearbucklinganalyses. These resultsareused for the
selectionofdimensioningflightloadcaseswhichareappliedforfurtherdetailedanalyses
Previous analysis loops, which have been performed in the past, are characterised by simplified
modellingoftheadjacentstructures.Thesimplestmodelusestherebyretraintconditionsatthelower
and upper front skirt interfaces. In detailed invetigation it has then been verified that the adjacent
structures have a major influence on the structural behaviour of the front skirt. Therefore, it was a
majortargettoimprovetheconfidencelevelbyamorerealisticmodellingoftheboundaryconditions
(geometric,force,thermal).
Thecurrentanalysis,whichisassociatedwiththeATV(AutomatedTransferVehicle)configuration,
uses accurate models of the adjacent structures. These include the realistic model of the RIEtank
below the front skirt interfaces (STN28) as well as the VEB above the upper front skirt interface
(STN25),respectively.
Thustheoverallmodelconsistsofthefollowingsubmodels.
thelowerboundarystructureRIE(LOXTank).
theEPCfrontskirt(JAVEC)
theupperboundarystructurecomposedoftheVEBATVandthestructureabove
In Section 21.2 the description of the structure and the introduction of its structural components is
presented.
In Section 21.3 the mathematical model (geometry, material, loads, geometric boundary conditions
and imperfections) is introduced. Additionally the definition of the required load proportionality
factor(LPF)applicableonultimateloadconditions,whichisspecifiedasrepresentativedimensioning
variable,isgiven.
396
ECSSEHB3224A
24March2010
In Section 21.4 the solution method including an overview about the applied discretisation is
presented.
Section 21.5 deals with the analysis results. They are linked with the solution of specific problems
arisinginstabilityanalysis,i.e.infindingandtraversingthecriticalpoints(limitpointsorbifurcation
points) closest to the operational range. It presents the results of the eigenvalue analyses, the post
buckling analysis on basis of imperfect reference configurations and verifies the predicted load
proportionalityfactor.
Section21.6presentsashortsummaryandtheconclusion.
21.2 Description of the Structure

Figure211showsthelocationoftheFrontSkirt(JAVE)intheARIANE5launcher.Themainpurpose
oftheJAVEistodistributethethrustanddynamicloadsoftheboosterstagetothecentralstagesand
tocarrythethrustoftheVulcainIIenginethroughthemaincryogenicstagetotheupperstage.
Figure211:PositionoftheFrontSkirt(JAVE)withinARIANE5Launcher
TheJAVEincludingthedefinitionofitsmajorstructuralcomponents(below)isillustratedinFigure
212.
(1) BLIS (BoosterLoadIntroductionStructure)
(2) RLRS (RadialLoadReactionStructure)
(3) ALRS (AxialLoadReactionStructure)
(4) MISS (MinorInterfaceSupportStructure)
397
ECSSEHB3224A
24March2010
(2)RLRS
(4)MISS
(3)ALRS
(1)BLIS
Figure212:JAVEwithmajorStructuralComponentsBLIS(1),RLRS(2),ALRS(3)
andMISS(4)
The basic structure is conceptually subdivided into independently designed and analyzed
substructures,depictedinFigure213.
398
ECSSEHB3224A
24March2010

Protected Naked Basic BoosterLoad

Flight Flight Structure Introduction DAAVFitting

Structure Structure Structure(BLIS)
ShearPlates

BackupRibs

AxialLoad
ReactionStructure CASEFaceRing
(ALRS)
OuterSkinPanels

RIEConn.Sheets

RadialLoad
MajorFrames
Reaction Structure
(RLRS)
LightFrames

LightRibs

HighRibs

InnerSkinPanels

Attachments
MinorInterface
SupportStructure
(MISS)
Protection Prosial

Mastic

Figure213:StructuralComponentsoftheJAVEflightunit
399
ECSSEHB3224A
24March2010
21.3 Mathematical Model
21.3.1 Launcher Coordinate System and Sign Convention

The relationship between the launcher coordinate system X, Y, Z used for the specification of the
generalloads,andtheJAVEcoordinatesystemU,V,SisshowninFigure214below:
Figure214:ARIANE5LauncherandJAVECoordinateSystemandSign
Convention
21.3.2 Geometry
The geometry of the mathematical model contains the JAVE and its substructures as well as the
adjacentstructuresoftheJAVE(RIETankCentralStage,UpperStructureVEB),showninFigure215
below.
400
ECSSEHB3224A
24March2010
UpperStructure
JAVE
RIETank
Figure215:GeometricalModelforJAVEStabilityAnalysisincludingadjacent
Structures
21.3.3 Material Definition

Table 211 shows the materials breakdown of the JAVE structural components, which are
characterised by a nonlinear material behaviour with von MISES yield criterion and isotropic
hardening.
Table211:StructuralComponentsandcorrespondingMaterialParametersforthe
DefinitionofJAVENonlinearMaterialBehaviour
StructuralComponent Rp0.2[MPa] Rm[MPa] EModul[MPa] Poisson[]
LightFramesforged 248 413 73100 0.32
MachinedRibs 340 430 71000 0.36
OuterSkin
MajorFrames 340 430 72000 0.33
LightFramesmachined
HighRibs
435 505 71800 0.33
LightRibs
The hardening behaviour is approximated by the RAMBERGOSGOOD law. For the adjacent
structures(RIETankandUpperStructureVEB)alinearelasticmaterialbehaviourispostulated.
401
ECSSEHB3224A
24March2010
21.3.4 Loading
21.3.4.1 Load Case Description
The overall load events which are applied for the investigation of the structural response are
composedof
RIEPressure
Temperature
AerodynamicPressure
MechanicalLoad+FrictionMoment
Theyarecharacterisedbyalteringloadintensitiesduringtheboosterflightduration.Theexistenceof
potentiallycriticalloadstatesendswiththeseparationoftheboosters.
In the following the characteristic of one singular load event is introduced. In this regard it is
mentionedthatafullanalysisloopoftheJAVEconsistsoftheinvestigationofseveralnumberofload
events.Thefollowingfiguresshowtherebyonesuchloadevent.
TheheterogeneityofthetemperaturefieldattheJAVE,whichislinkedto aspecifictimestepduring
flight,isindicatedinFigure216.
Figure216:ThermalEventduringFlightappliedforJAVEStabilityAnalysis
Theresponseduetotheactionofthemechanicalloads(boosterthrust,windandgust)andthefriction
momentattheDAAVfittingsisextractedbytheinvestigationofthefluxdistributionalongthelower
andupperboundaries(=stations;STN28andSTN25)oftheJAVE.Itisagainrelatedtoacertaintime
stepduringtheflight,incompliancewiththetemperaturefieldgivenabove.Theaxialfluxvariation
402
ECSSEHB3224A
24March2010
along the circumference, summarising the partitions from all load events, is shown below. Thereby,
Figure 217 corresponds to the lower and Figure 218 to the upper boundary of the JAVE,
respectively.
Figure217:VerticalFluxDistributionalongSTN28duetoAllLoadEvents
Figure218:VerticalFluxDistributionalongSTN25duetoAllLoadEvents
403
ECSSEHB3224A
24March2010
21.3.4.2 Load Sequence
The sequence of the application of the singular load steps, introduced in the previous section, is
appliedasfollows:
STEP1 RIEPressure
STEP2 Temperature
STEP3 MechanicalLoads+Friction
ThenotationSTEPisrelatedtotheABAQUSterminology.
21.3.4.3 Load Application

Theloadsareappliedontothestructureonbasisofaspecifiedstrategy.Thereby,thecapableloadof
thestructureisafunctionof:
Lcapable = f ( A1 * RIE _ press * jult + A2 * Temp + A3 * AERO _ press * jult +
211
+ A4 * FRICT * jult * j sys + A5 * ( MECH + STN 2) * jult * j sys )
With
AiLoadcombinationfactor
jultultimatesafetyfactor
jsyssystemmargin
TheLoadProportionalityFactor(LPF)isevaluatedwithrespecttotheliveloadtermsFrictionand
MechanicalLoads,thus:
Lcapable = P0 + LPF P1 212
P0 = state( A1 * RIE _ press * jult + A2 * Temp + A3 * AERO _ press * jult ) 213

DeadLoad
P1 = A4 * FRICT * jult * j sys + A5 * ( MECH + STN 2) * j ult * j sys 214
LiveLoad
21.3.5 Geometric Boundary Conditions

The geometric boundary conditions (=essential boundary conditions) are defined such that the
displacementsandrotationsofallnodeslocatedalongthecircumferenceofthetopsectionattheVEB
arerestraint.
21.3.6 Imperfections
21.3.6.1 Overview
Twokindsofimperfectionsaretakenintoaccount:
GeometricImperfections
PhysicalImperfections
21.3.6.2 Geometric Imperfections

Geometricimperfectionsareappliedtoinvestigatethesensitivityofthestructuralresponseincaseof
deviations against perfect geometric conditions. The lower limits of the amplitudes are thereby
derivedfromfabricationtolerances.
404
ECSSEHB3224A
24March2010
Additionalimperfectionsareincorporatedinareaswithalargenumberofbifurcations.Thepurpose
istonudgethestructurealongapostbifurcationpath(duetonumericalimperfections).
Forthepurposeofeffectivelymeetingabovegoals,amplitudesof1.5mmhavebeendefinedforthe
incorporatedgeometricimperfections.
21.3.6.3 Physical Imperfections

Physical imperfections are associated with fabrication induced stresses. During panel forming self
equilibrating residual stresses occur in the material. These stresses have been measured during the
qualification programme and they are implemented as local initial compression stress fields in the
model.
21.4 Finite Element Analysis
21.4.1 Solution Method

Thesolutionmethodfortheanalysisoftheequilibriumresponseisperformedasfollows:
1. A nonlinear analysis with respect to perfect geometric conditions is performed. The
application of the loads is thereby in compliance with section 21.3.4.2 and 21.3.4.3. To
advance thesolution, theanalysis stages are broken into incremental steps. For the live
load a maximum increment size of 0.05 is chosen. The analysis terminates at that load
state where a critical point is predicted(limit or bifurcation point, indicated by negative
eigenvaluesinthemessagefile).
2. Eigenvalues at the last increment, which converged with positive eigenvalues, are
extracted. Thereby the null eigenvalue, which indicates the existence of two or more
equilibriumpaths(branches),isinvestigated.
3. A sufficiently large number of eigenmodes which are associated with the smallest
eigenvaluesareimplementedasgeometricimperfections(refertosection21.3.6.2)tothe
model.
4. Rerun of the nonlinear analysis with respect to the incorporated imperfect reference
configuration. Due to the predefinition of a stable and rising equilibrium path, the
analysisstopsbeyondthecriticalpointpredictedinstep1.
5. Proceedasintroducedinsteps2and3.
6. Rerun of the nonlinear analysis with respect to the updated imperfect reference
configuration.
Consequently,theideaistoperturbtheresidualequationinsuchawaythattheunderlyingregularity
intrinsictoabifurcationpointisdestroyed.Thus,thebifurcationpointistransformedtoalimitpoint.
The problem is turned into a problem with continuous response instead of bifurcation, which is
accomplished by the introduction of geometric and numerical imperfections. In the mathematical
literature,thisapproachisreferredtoasunfolding.
Thisproceduregivenaboveisrepeatedasoftenasitisverifiedthatthepredictedcriticalpointdoes
notcoincidewithafailurepoint.Thephenomenawhichareassociatedwiththelattercanbelocalor
globalinnature.Inthefirstcase(e.g.failureofanoncriticalstructuralcomponent)thestructuremay
regainfunctionalequilibriumafterdynamicallyjumpingtoanotherequilibriumpath.Inthelattercase
thefailureiscatastrophicordestructive,withtheconsequencethatthestructure isnolongerableto
regainthefunctionalrequirement.
405
ECSSEHB3224A
24March2010
21.4.2 Amendments to Extraction of Eigenvalues
The eigenvalues i are extracted at selected load levels on basis of nonlinear analyses runs. For that
purposetheloadeventsaredividedintothedeadloadvectorPN(equaltopreload;loadhistoryuntil
selected load state) and the live load vector QN (incremental load pattern). Thus the extracted
eigenvaluesreflecttheconditionofthesystemmatrixattheconsideredloadstate,differentfromthe
basestate.
Thecorrespondingcriticalbucklingloadisexpressedby:
P N + i Q N 215
whichisrelatedtotheexistenceoftwoormoreequilibriumpaths.
21.4.3 Discretisation
Anoverviewabouttheapplieddiscretisationisgivenbelow:
NUMBEROFELEMENTSIS 304730
NUMBEROFNODESIS 336470
NUMBEROFNODESDEFINEDBYTHEUSER 288816
NUMBEROFINTERNALNODESGENERATEDBYTHEPROGRAM 47654
TOTALNUMBEROFVARIABLESINTHEMODEL 1753332
Figure219showsasectionofthewholediscretisedmodel.
406
ECSSEHB3224A
24March2010
UpperStage
JAVE
RIETank
Figure219:SectionofDiscretisationoftheJAVEModel
21.5 Results Nonlinear Analyses of Load Case A151
21.5.1 Introduction and Chronology

Thepresentationoftheanalysesresultsinthefollowingsectionsisnotperformedinachronological
manner.Heretherepresentativeanalysisrunwhichjustifiesthecompliancewiththespecifiedsafety
requirementsisintroducedfirst(section21.5.2),followedbyanalysesrunsrepresentingintermediary
results,whichwerenecessaryfortheextractionofeigenvaluesatloadlevelswherecriticalpointsare
predicted(section21.5.3).
Thereby,theidentificationofthejobsisasfollows:
A151_NEWTON_Evolxx_Imp
The first characters define the load case name A151, followed by the nonlinear solution strategy
(NEWTON/RIKS). The evolution status of the job is identified by the extension Evolxx, where the
U U
final one Evol20 is representative with reference to the applied justification strategy introduced in
section 21.4.1. The final characters are linked to the underlying reference configuration
(perfect/imperfect).
407
ECSSEHB3224A
24March2010
21.5.2 Jobname A151_NEWTON_Evol20a_Imp
21.5.2.1 Overview
Inthissectiontherepresentativenonlinearanalysisrunisintroduced.Thereby,thefinalequilibrium
configuration,whichisassociatedwithacriticalpoint,ispredictedatLPF=1.63.
21.5.2.2 Basic Data

Accordingtosection21.3.4.2thesubcasesareappliedasinTable212.Thejobtimesummaryandthe
hardware configuration are as in Table 213. The corresponding description of the characteristic
convergence behaviour, including the cutbacks and number of negative eigenvalues within the
consideredsubcasesisshowninTable214.
Table212:Solutionstrategy
SubCases SubCaseDeclaration SolutionStrategy
STEP1 RIEPressure NEWTONRAPHSON
STEP2 TemperatureT114s NEWTONRAPHSON
STEP3 MechanicalLoads+FrictionMoment NEWTONRAPHSON
Table213:Jobnames
JobName CPUTime HostSystem
A151_NEWTON_Evol20a_Imp 16.5Hours(=Wallclock) Altix350,4CPUsused
Table214:NumberofnegativeeigenvaluesatINC/STEPtime
SubCases CutBacksatINC Nr.ofnegativeEigenvaluesatINC/STEPTime
STEP1 0 0
STEP2 /13/ 0
STEP3 /35/36/37/38/ 0
CutBack=ReductionofLoadIncrementSize
21.5.2.3 Status File

TheABAQUS/Standardstatusfile(*.sta)containsasingle80characterrecordforeachincrementandis
updated upon completion of each increment of an analysis. Therefore, it provides a monitor of the
progressofeachcurrentlyactiveanalysis(seeFigure2110).
The analysis run stopped at LPF=1.63. The criterion for the termination is that the required
incrementalstepsizeislessthanthespecifiedminimumrequired(1.0e5).Thesystemmatrixdoesnot
containnegativeeigenvaluesatanyconvergedincrementalongthewholeloadhistory.Consequently,
itisverifiedthattherearenostructuralcomponentswhichareloadedbeyondtheircriticalpoints.
TheequilibriumbifurcationappearingatloadstateLPF=1.63isinvestigatedinsection21.5.2.6.
408
ECSSEHB3224A
24March2010
Figure2110:ABAQUS/StandardStatusFilemonitoringtheProgressofNonlinear
Analyses(A151_NEWTON_Evol20a_Imp.sta)
409
ECSSEHB3224A
24March2010
21.5.2.4 Investigation of Equilibrium Configuration at LPF=1.63
In order to check the equilibrium state at the highest predicted load state (LPF=1.63) the axial flux
distributions at the upper and lower boundaries (STN25 and STN28) of the JAVE are investigated.
Thereby,theaxialfluxesextractedfromthenonlinearanalysisaresetincomparisontotheonefrom
the linear analysis. For the purpose of an adequate comparison the axial fluxes at LPF=1.63 are
normalisedtoLPF=1.0.
Figure2111andFigure2112showthecomparisonalongthecircumferentialdirectionsatSTN25and
STN28,respectively.
From the investigation above it can be derived that there is no essential change of the equilibrium
stateagainsttheoneofthereferenceconfiguration.
400
200
Vertical Flux [N/mm] 0

-180 -150 -120 -90 -60 -30 0 30 60 90 120 150 180
-200
-400
-600
-800
-1000
-1200
Angular Position []
Figure2111:ComparisonofAxialFluxDistributionsfromLinear(RedLine)and
NonLinearAnalysis(BlueLine)atSTN25
410
ECSSEHB3224A
24March2010
1400
1200
Vertical Flux [N/mm] 1000

800
600
400
200
0
-180 -150 -120 -90 -60 -30 0 30 60 90 120 150 180
-200
Angular Position []
Figure2112:ComparisonofAxialFluxDistributionsfromLinear(RedLine)and
NonLinearAnalysis(BlueLine)atSTN28
21.5.2.5 Investigation of Material Nonlinearity

In order to check the structural response on the level of the singular material point the equivalent
plasticstrainsareinvestigatedatthehighestpredictedloadlevel(LPF=1.63).Duetothefactthatthe
adjacent structures (RIETank, Upper Structure VEB) are solely defined on basis of a linear elastic
materialmodel,thisinvestigationisrestrictedtotheJAVE.
Figure2113showsthecontourplotoftheequivalentplasticstrains(PEEQ)withintheJAVE.Thereby,
the maximum PEEQ reaches an amount of 3.6[%], locally appearing at the outerskin close to the
DAAV.
Figure2114showsthislocationindetail.Therebyitisvisible,thatthemaximumPEEQappearsatthe
outerskinlocationatthedirectextensionoftheDAAVthrustrib.
From this investigation it can be stated that the prediction of the materials response at LPF=1.63 is
closely related to a failure point in the JAVE. Apart from local areas in the range of the DAAV the
wholeJAVECishowevercharacterisedbyamateriallylinearresponse.
411
ECSSEHB3224A
24March2010
Figure2113:EquivalentPlasticStrains(PEEQ)atJAVEduetomechanicalLoad
CaseA151(STEP3)atLPF=1.63
412
ECSSEHB3224A
24March2010
Figure2114:LocalAppearanceofmaximumEquivalentPlasticStrain
PEEQ=3.6[%]atOuterSkinSTN3abovetheDAAVatLPF=1.63
21.5.2.6 Investigation of Zero-Eigenvalue at LPF=1.63

The eigenvalue extraction at the predicted maximum mechanical load state (STEP3) at INC=38
(LPF=1.63)isperformedfortheinvestigationoftheappearingequilibriumbifurcation,associatedwith
thezeroeigenvalue.
ThecorrespondingrestartfileisA151_NEWTON_Evol20a_Impwherethesignificantjobdataareas
inTable215.
Table215:Jobdata
JobName RestartfromNonlinearRun CPUTime HostSystem
EigenI_M38_Evol20 A151_NEWTON_Evol20a_Imp 1Hours Altix350
STEP3/INC38(LPF=1.63) (=Wallclock) (4CPUsused)
LPF=LoadProportionalityFactor
Table 216 presents the smallest 5 eigenvalues and the description of the appearance of the
corresponding eigenmodes. Additionally, the load levels associated with the bifurcation points are
extrapolatedaccordingtosection21.4.2
413
ECSSEHB3224A
24March2010
Table216:SmallestEigenvaluesbasedonResultsofNonlinearJob
A151_NEWTON_Evol20a_Imp,extractedatLPF=1.63.
EigenI_M38_Evol20:EigenvalueExtractionatSTEP3/INC38(LPF=1.63)
Mode Eigenvalue Location BifurcationPoint
1 0.0011 VEB,Skin,ConeQ2/Q3 1.631
2 0.1492 VEB,Skin,ConeQ2/Q3 1.779
3 0.2922 VEB,Skin,ConeQ2/Q3 1.338
4 0.3989 VEB,Skin,ConeQ2/Q3 1.231
5 0.4435 VEB,Skin,ConeQ2/Q3 2.074
In Table 216 it is visible that all investigated eigenmodes appear within the cone of the upper
structure VEB between the quadrants Q2/Q3. The characteristic shape of the eigenmode associated
with the zeroeigenvalue is illustrated in Figure 2115. Thereby, it is visible that the affected area
coversasectionoftheconicalstructureandoneadjacentshearwall.Theoccurrenceofthenegative
eigenvaluesassociatedwithmodes3and4refertoloadreversal.
Figure2115:EigenmodeassociatedwithEigenvalue1(1=0.0011)ofEigenvalue
B B
ExtractionEigenI_M38_Evol20atSTEP3/INC38(LPF=1.63)
414
ECSSEHB3224A
24March2010
21.5.3 Generation of Final Imperfect Reference Configuration
21.5.3.1 Overview
The final imperfect reference configuration is established on basis of selected, magnified and
superimposedeigenmodeswhichareextractedatdifferentloadlevels.Theidentificationistherebyas
follows:
EigenI_Mxx_Evolxx
Thefirstcharactersrefertotheeigenvalueanalysis,followedbythelinktoaPerfect/Imperfectreference
U U U U
configuration. The next character references the subcase (Mechanical/Thermal) and the increment
U U
number(xx).Thelastcharacterstringidentifiesthestatusofthecorrespondingnonlinearanalysisrun
(Evolxx).
Theselectionoftheeigenvalueextractions,usedfortheestablishmentofthefinalimperfectreference
configurationisasshowninTable217:
Table217:ConfigurationofEigenvalueExtractionsandcorrespondingRestart
FilesfortheDefinitionoftheFinalImperfectReferenceConfiguration
Jobname:A151_NEWTON_Evol20a_Imp
Jobname LoadLevel Locations
EigenvalueExtraction Tmech(LPF)
B B RestartfromNLJob Eigenmodes
EigenP_M14 0.281 A151_NEWTON VEB
EigenI_M28 0.866 A151_NEWTON_Evol3_Imp JAVE
EigenI_M18_Evol16 0.664 A151_NEWTON_Evol16b_Imp VEB
EigenI_M44_Evol17 1.44 A151_NEWTON_Evol17a_Imp VEB
According to the notation given above it is visible that all the eigenmodes are extracted within the
mechanicalstep.Additionally,thelevelofthemechanicalloadstate(=deadload)isincluded.Thelast
column allocates the appearance of the smallest eigenmodes to the main structural components
(VEB/JAVE).
In the following subsections the eigenvalue extraction and the introduction of the eigenmodes are
presented.
21.5.3.2 EigenP_M14 Eigenvalue Extraction at LPF=0.281

The first eigenvalue extraction is performed within the mechanical load state (STEP3) at INC=14
(LPF=0.281).ThecorrespondingrestartfileisA151_NEWTON,whichwasterminatedatLPF=0.33.
ThesignificantjobdataareasinTable218:
Table218:Jobdata
EigenP_M14 A151_NEWTON 2Hours Altix350
415
ECSSEHB3224A
24March2010
extrapolated.
A151_NEWTON,extractedatLPF=0.281
EigenP_M14:EigenvalueExtractionatSTEP3/INC14(LPF=0.281)
Mode Eigenvalue Location BifurcationPoint
1 0.181 TankPlatform,VEB 0.462
InTable219itisvisiblethatallinvestigatedeigenmodesappearwithintheupperstructureVEB.These
areas can further be reduced to locations along the ring structure of the tankplatform. The
characteristicofonerepresentativeeigenmodeisillustratedinFigure2117.Thereby,itisvisiblethat
theaffectedareacoversthecircumferenceofonecircularopening.
Theoccurrenceofthenegativeeigenvaluesassociatedwithmodes2,5and10refertoloadreversal.
416
ECSSEHB3224A
24March2010
ExtractionEigenP_M14atSTEP3/INC14(LPF=0.281)
For the definition of an imperfect reference state the following eigenmodes are selected and
introduced:
Table2110:ModesforestablishmentofImperfectReferenceConfiguration
JobName ModesforestablishmentofImperfectReferenceConfiguration
EigenP_M14 Modes14,20,3,10,15,7,13,17(magnifiedtoAmplitude1.5mm)
21.5.3.3 EigenI_M28 Eigenvalue Extraction at LPF=0.866

The second eigenvalue extraction is performed within the mechanical load state (STEP3) at INC=28
(LPF=0.866).ThecorrespondingrestartfileisA151_NEWTON_Evol3_Imp,whichwasterminatedat
LPF=0.866.
Thesignificantjobdataareasfollows:
Table2111:Jobdata
EigenI_M28 A151_NEWTON_Evol3_Imp 8Hours Altix350
417
ECSSEHB3224A
24March2010

extrapolated.
A151_NEWTON_Evol3_Imp,extractedatLPF=0.866
EigenI_M28:EigenvalueExtractionatSTEP3/INC28(LPF=0.866)
Mode Eigenvalue Location(OS=OuterSkin) BifurcationPoint
1 0.0002 JAVE,OS,Q1,VPanel 0.866
5 0.152 JAVE,OS,Q1,UVPanel 0.714
In Table 2112 it is visible that the first eigenmodes 110 appear at quadrant Q1 outerskin locations
withintheJAVE.Theeigenmodeofthezeroeigenvalue(1=0.0002)occursattheVpanel,illustratedin
B B
Figure2118.
Theoccurrenceofthenegativeeigenvalueassociatedwithmode5referstoloadreversal.
418
ECSSEHB3224A
24March2010
Figure2118:EigenmodeassociatedwithZeroEigenvalue1(1=0.0002)of B B
EigenvalueExtractionEigenI_M28atSTEP3/INC14(LPF=0.866)
introduced:
EigenI_M28 Modes1,5,36,41,43,42,21,49,35,14,23,33,24,48,38,34,20,47,40
(magnifiedtoAmplitude1.5mm)
21.5.3.4 EigenI_M18_Evol16 - Eigenvalue Extraction at LPF=0.664

The third eigenvalue extraction is performed within the mechanical load state (STEP3) at INC=18
(LPF=0.664).ThecorrespondingrestartfileisA151_NEWTON_Evol16b_Imp,whichwasterminated
atLPF=0.664.
419
ECSSEHB3224A
24March2010
Table2114:Jobdata
EigenI_M18_Evol16 A151_NEWTON_Evol16b_Imp 5Hours Altix350

extrapolated.
A151_NEWTON_Evol16b_Imp,extractedatLPF=0.664
1 0.0042 VEB(EndofCone) 0.668
3 0.655 JAVE,OS,Q3,VNPanel B B 0.009
5 0.704 JAVE,OS,Q2,UNVPanel
B B 0.040
6 0.707 JAVE,OS,Q4,UVNPanel B B 0.043
7 0.713 JAVE,OS,Q3,UNVNPanel
B B B B 0.049
B B B B 0.062
B B B B 0.097
InTable2115itisvisiblethatthefirsteigenmodes110appearattheVEBaswellasattheouterskin
within the JAVE. The eigenmode of the zeroeigenvalue (1=0.0042) occurs at the end of the conical
B B
structurewithintheVEB,illustratedinFigure2119.Thesamecharacteristicisfoundforeigenmode2,
locatedattheoppositeside,Figure2120.
In this regard, it is explicitly mentioned that the eigenvalues associated with the JAVE modes are
negativeandthereforetheyarerelatedtoloadreversal.
420
ECSSEHB3224A
24March2010
Figure2119:EigenmodeassociatedwithZeroEigenvalue1(1=0.0042)of
B B
EigenvalueExtractionEigenI_M18_Evol16atSTEP3/INC18(LPF=0.664)
B B
ExtractionEigenI_M18_Evol16atSTEP3/INC18(LPF=0.664)
421
ECSSEHB3224A
24March2010
introduced:
EigenI_M18_Evol16 Modes1,2(magnifiedtoAmplitude1.5mm)
21.5.3.5 EigenI_M44_Evol17 Eigenvalue Extraction at LPF=1.44

The fourth eigenvalue extraction is performed within the mechanical load state (STEP3) at INC=44
(LPF=1.44).ThecorrespondingrestartfileisA151_NEWTON_Evol17a_Imp,whichwasterminatedat
LPF=1.44.
Table2117:Jobdata
EigenI_M44_Evol17 A151_NEWTON_Evol17a_Imp 5Hours Altix350
extrapolated.
A151_NEWTON_Evol17a_Imp,extractedatLPF=1.44.
1 0.0002 VEB,Skin,ConeQ4 1.440
5 0.3580 JAVE,OS,Q3,UnPanel B B 1.798
6 0.3935 JAVE,OS,Q2,UnVPanel
B B 1.047
7 0.4285 JAVE,Frame7,0 1.869
9 0.4952 JAVE,Frame7,0 1.935
10 0.5062 VEB,Skin,ConeQ4 0.934
422
ECSSEHB3224A
24March2010
InTable2118itisvisiblethatthefirsteigenmodes110appearattheVEBaswellasattheouterskin
within the JAVE. The eigenmode of the zeroeigenvalue (1=0.0002) occurs at the end of the conical
B B
structurewithintheVEB,illustratedinFigure2121.
In this regard it is explicitly mentioned that the eigenvalues associated with the JAVE modes are
negativeandthereforetheyarerelatedtoloadreversal.
Figure2121:EigenmodeassociatedwithZeroEigenvalue1(1=0.0002)of B B
EigenvalueExtractionEigenI_M44_Evo17atSTEP3/INC44(LPF=1.44)
According to Table 2118, it is visible that the next equilibrium bifurcation is characterised by
eigenmode3appearingagainwithintheconicalstructureQ4.Thecorrespondingbifurcationpointis
predictedtobeatLPF=1.75.Thedifferencetoeigenmode1,whichisrepresentingthezeroeigenvalue,
isestablishedbyaslightrotationabouttheverticalaxis.Thereforeitcanbestatedthateigenmode1is
reproducedatahigherloadstatebyasmallchangeofthecircumferentialpositionoftheamplitude.
For the definition of an imperfect reference state only eigenmode 1 is selected and introduced as
follows:
EigenI_M44_Evol17 Mode1(magnifiedtoAmplitude1.5mm)
423
ECSSEHB3224A
24March2010
21.5.3.6 Summary of Establishment of Imperfect Reference Configuration
InTable2120,theselectionofthoseeigenvalueextractionsaresummarisedwhicharedefinedforthe
establishment of the final imperfect reference configuration applied to job
A151_NEWTON_Evol20a_Imp(referencetosection21.5.2).
Table2120:SummaryofEigenvalueExtractionsforEstablishmentofFinal
ImperfectReferenceConfigurationofjobA151_NEWTON_Evol20a_Imp
EigenvalueExtraction atLPF SelectedModes Location
EigenP_M14 0.281 14,20,3,10,15,7,13,17 TankPlatform,VEB
1,5,36,41,43,42,21,49,35,14,
EigenI_M28 0.866 OuterSkin,JAVE
23,33,24,48,38,34,20,47,40
EigenI_M18_Evol16 0.664 1,2 EndofCone,VEB
EigenI_M44_Evol17 1.44 1 EndofCone,VEB
SuperpositionandMagnificationFinalImperfectReferenceConfiguration
21.6 Summary and Conclusion
21.6.1 Summary
Sections21.1and21.2introducetheARIANE5Frontskirt(JAVE)anditsadjacentstructures.
Section21.3presentsthecharacteristicsofthemathematicalmodel,includingthedefinitionof
MaterialModels
Loadings
GeometricandPhysicalImperfections
AnalysisTechnique
Section21.4definesthesolutionmethodandthediscretisation.
Section 21.5 discusses the analysis results. Thereby, the compliance with the safety requirements is
introducedforoneanalysisrunindetail.
21.6.2 Conclusion
ItisjustifiedthatthepredictedcriticalloadstateLPF=1.63,iscloselyrelatedtoafailurepointinthe
JAVE.Thisisderivedbytheinvestigationoftheappearanceoflocalplasticityattheouterskinarea,
whichconvergescloselytotherangeofthedesignallowablelimits.
Apart from these local areas closely spaced to the DAAV the whole JAVEC is characterised by a
materiallylinearresponse.
Therefore it is justified that the postulated safety requirement LPF 1.50 is met. In this regard it is
explicitly mentioned that this requirement is only related to the justification against buckling
(requirementofgeneralspecificationSG110).Additionallyperformedanalyses,whichinvestigatethe
limitingconditionscausedbyconnections(e.g.rivets,bolts)arenotpresentedinthischapter.Theyare
justifiedonbasisofgeneratedsubmodels.AccordingtothestipulatedsafetyrequirementLPF1.50
424
ECSSEHB3224A
24March2010
on the ultimate load conditions it is however ensured that failure by instability would appear
distinctively after appearance of local failure caused by the violation of strength requirements (e.g.
rivetfailure).Thisstatementhasbeenvalidatedbyqualificationtests.
Thepredictionoftheloadproportionalityfactorreferstoanimperfectreferenceconfiguration,which
is established on basis of selected, magnified and superimposed eigenmodes. Investigations with
respect to different imperfection amplitudes show no significant influence on the predicted LPF.
Therefore,thefrontskirtcanbeclassifiedasimperfectioninsensitiveagainsttheinvestigatedcritical
loadcase.

ALRS AxialLoadReactionStructure
BLIS BoosterLoadIntroductionStructure
DAAV ThrustRib
LPF LoadProportionalityFactor
MISS MinorInterfaceSupportStructure
PEEQ EquivalentPlasticStrain
RIE TankCentralStage
RLRS RadialLoadReactionStructure
STN25 Upperboundary
STN28 LowerBoundary
VEB VehicleEquipmentBay
425
ECSSEHB3224A
24March2010
22
Buckling Analysis and Qualification Static
Load Testing of VEGA Interstage 1/2
Structure
22.1 Overview
As part of the VEGA program, Dutch Space B.V. is responsible for the design, development and
manufacture of the Interstage 1/2, a cone shaped structure joining the 1st Stage P80 SRM to the 2nd
P P P P
StageZ23(Zefiro)SRMandincorporatingtheseparationsystemforthe1stStageseparation.
P P
TheoverallheightoftheVegaInterstage1/2is2.1m,theTopinterfaceringdiameteris1.9mandthe
Bottominterfaceringdiameter3.0m.TheInterstage1/2isconstructedfromaluminiumanddesigned
andbuiltapplyingconventionaltechnologies.
ThedesignoftheInterstage1/2isdrivenbyarequirementforhighoverallstiffness.Inordertomeet
the stiffness requirements, a construction with relatively thick panel skins, 6.3 mm is required.
Doublers,thickness3.2mmareappliedaroundcutoutsandextendbeyondtheedgesofthepanelsto
form the panel interconnection joints. A monocoque aluminium construction is selected for its
simplicityandbecauseitequallymeetstherequirementsforhighstiffness,albeitthatstrengthisthen
compromisedthroughalowerbucklingstability.
ManylargeopeningsexistintheInterstage1/2onaccountoftheseparationretrorocketsandaccess
provisions for prelaunch integration and servicing activities. These openings severely disrupt load
paths through the primary loadcarrying structure, so that buckling strength cannot be accurately
predicted using classical design formulae. Detailed geometrical nonlinear analyses using Finite
Elementmodellingtechniquesarethereforeapplied.
ThischapterpresentsthedetailsofthebucklinganalysesandstaticstrengthtestingoftheInterstage
1/2 for the Vega Launch Vehicle. The Interstage 1/2 is constructed as monocoque and is buckling
critical,hencetheneedtoobtainaccuratepredictionsforbucklingstrength.
To guarantee a reliable prediction of the critical buckling load the expected manufacturing and
assembly imperfections, and the general sensitivity of the type of structure for buckling failure are
includedintheanalysis.
These effects can be incorporated in the, largely empirically based, knockdown factors. The
literature provides us with many examples of analyses and correlations with test data, from which
estimatesforknockdownfactorscanbederived.Valuesquotedgenerallyrangebetween0.33and0.65,
dependingonthedetailofanalysesandtestsandestimatedbucklingsensitivityofthestructure.For
the development of the Interstage 1/2 as monocoque structure with complex load paths it is
consideredessentialthatanaccurateanalysisofbucklingstabilitybeperformedandthatitssensitivity
forinitialimperfectionsbewellunderstood.
426
ECSSEHB3224A
24March2010
Twoapproachesareappliedforevaluationofthestructuressensitivitytoinitialimperfections.Inthe
firstapproach,aknockdownfactorisderivedbyperformingasensitivityanalysisusingtheSRA200
program[1].Theanalysis,basedonKoitersasymptotictheory,calculatestheKoiterconstantsaand
b,fromwhichtheknockdownfactorisderived.TwoconfigurationsofInterstage1/2areevaluated
applying this approach, namely the monocoque construction and a version with blade stiffeners. In
thesecondapproachusingMSC.Nastrannoknockdownfactorisusedbutinsteadtheimperfections
areapplieddirectlyintheanalysismodelasinitialdisplacements.Thedisplacementsaremodelledas
the buckled form scaled to half the cone skin thickness. The scaled buckled forms of the first four
modes,obtainedfromnonlinearbucklinganalyses,areappliedseparatelyasinitialdisplacementsina
seriesofgeometricalnonlinearbucklinganalyses.
Static load tests were carried out on a qualification hardware model of the Interstage 1/2 at TNO
facilitiesinDelft.Astructuralmathematicalmodelofthecompletetestsetupwasdevelopedinorder
derivethetestloadcorrectionfactorsandtoprovidefinalcorrelationofthecriticalbucklingloadand
mode shape. As strength qualification, the structure was subjected to an offset compression load
which induces an axial force and bending moment on the structure, such that the maximum flux is
createdacrosstheweakestsectionoftheInterstage1/2.Twoorientationsofthestructureweretested
inthiswayandnofailureswereencountereduptoqualificationtestloadlevels.Later,inaseparate
RuptureTestprogram,thestructurewasloadedtofailureinordertoestablishtheactualmarginof
safetywithrespecttobuckling.
22.2 Interstage 1/2 Assembly

The Interstage 1/2 is an allaluminium structure assembled from rolled panels and three main rings
machined from forgings. Two of the rings comprise the bolted ring interfaces with the LV adjacent
structures,P80SRMandZ23SRM,thethirdringisthesocalledseparationringseparatingforward
FwdandAftpartsoftheInterstage1/2.Theseparationringfurtheraccomodatesthepyrocutting
cordsystemwhichisfiredfor1stStageseparation,cuttingthroughafrangiblesectioninthering.At1st
P P P P
Stageseparationthestructureiscutthroughatthefrangiblesectionbytheactivationofapyrocutting
corddeviceafterwhich,six(6)RetroRockets,positionedintheAftPartoftheInterstage1/2,arefired
toensurethatthe1stStageissafelymanoeuvredawayfromthe2ndStage.
P P P P
The coneshaped Interstage structure is further stabilized against buckling by four(4) internal
ringframes(threeintheAftPartandoneintheFwdPart).Theringframesarepositionedtocoincide
withtheupperandloweredgesofcutoutsinordertoreinforcetheopenings.
Cutoutsareprovidedintheconepanels;six(6)ofthese,intheAftPart,arerequiredforinstallationof
the Retro Rockets, while four(4) additional openings are provided to allow access to internal
equipmentsforintegrationactivities.3.2mmthickaluminiumdoublersarerivetedtotheconepanel
skinasreinforcementaroundthecutoutsbutalsotoprovidelapjointsasmeanstointerconnectthe
panels. The cutouts are further reinforced by longitudinal Zshaped stiffeners bolted along the
meridionaledgesoftheopenings.Twelve(12)stringersarethusprovidedintheAftPart,reinforcing
the six(6)Retro Rocket openingsandfour(4)stringers are provided in the Fwd Part reinforcing two
accessopenings.
427
ECSSEHB3224A
24March2010
Figure221:VegaLaunchVehicleshowingpositionofInterstage1/2
Figure222:GeometricalmodelofInterstage1/2
22.3 Imperfection Sensitivity Analyses
22.3.1 Classical Buckling Load

Theclassicalbucklingloadofatruncatedunstiffenedperfectconicalstructuralmemberloadedwitha
uniformrunningloadalongtheedgewiththesmallestradiusisaccordingtoRef.[2]:
Pcrit = 2 Eh2 cos 2 c / 3 (1 2 ) 221
where c is the semivertex angle, h is the wallthickness, E is the Youngs modulus and is the
Poissonratio.
428
ECSSEHB3224A
24March2010
Thesocalledknockdownfactor 1 isgenerallyappliedtotheresultfromequation221totake
into account the imperfection sensitivity of thin walled conical structures. Multiplied by this knock
downfactor,thebucklingloadthenrepresentsasafebutoftenveryconservativeestimate.
The estimate for knock down factor obtained from [2] is = 0.33 . This applies for truncated
unstiffened perfect conesunderuniform applied loadsand with semivertex angle 10 c 75 .
o o
From[3],however,analternativeestimatefortheknockdownfactor canbeobtainedapplyingthe
expression:
0.83 1
= for 212
1 t 222
1 + 0.01
t
0.70 1
= for > 212
1 t 223
0.1 + 0.01
t
Rmin
where 1 = .
cos c
TheInterstageconesemivertexangle c is14.17o,smallestradius Rmin

P P = 0.975 mandthicknesst
1
is6.35mm,fromwhich = 158.4 .Thecorrespodingknockdownfactorapplyingequation222is
t
= 0.516 .
22.3.2 Asymptotic Post Buckling Theory

The asymptotic post buckling theory of W.T. Koiter [4] is applied to investigate the imperfection
sensitivityoftheVEGAInterstage1/2structure.Theasymptoticpostbucklingapproach,illustratedin
Figure223,consistsofthefollowingsteps:
a. Determinethelowestbifurcationpoint c ontheequilibriumpath(seeFigure223a),
b. Determine the sensitivity to initial geometric imperfections of the limit point ( s ) of the
maximumloadcarryingcapacityofthestructure,seeFigure223b.
429
ECSSEHB3224A
24March2010
Figure223:Loaddeflectioncurvesshowinglimitandbifurcationpoints.
a)generalnonlinearanalysis,b)asymptoticanalysis
Wehavetosolvethefollowingequationtoobtaintheratio s / c [5]
(1 s / c ) + a 2 + b 3 = ( s / c ) 224
The structure is imperfectionsensitive when a < 0 or when the combination a = 0 and b < 0 [5]
holds.Weinvestigatethecombination a = 0 and b < 0 forthecaseofaconicalshell.
ThecalculationofknockdownfactorsapplyingKoitersasymptotictheoryisdescribedin[4].Theso
calledKoiterspostbucklingcoefficientsaandbarecalculatedusinganadvancedversionoftheSRA
programme[6]writtenbyGeraldA.Cohen.
InRef.[6]thefollowingasymptoticexpansionisinvestigated
( s / c 1) = a 2 + b 3 (s / c 1) 225
Equation225isanextensionofequation224.
InRef.[6]expressionsareprovidedtocalculatetheconstantsa,b, and .
For a symmetric branching point where a = 0 and b < 0 the asymptotic relationship between the
bucklingloadofanimperfectshell s andtheimperfectionamplitude isgivenby

(1 ) ( ) ( )
3 1
3
s / c + 3b 2 2 2 1 s / c 1 = 0
2 226
2
where = /t
InordertoaccountforthenormalizationusedinSRAinthefollowingtheimperfectionamplitude
isnormalizedasfollows:
t
= 227
C
430
ECSSEHB3224A
24March2010
Where is the physical RMS normalized imperfection and the factor C is calculated by the SRA
program.
In Figure 224 for given / ratios, where and
are the first and second imperfection form
factors(seeRef.[7]forfurtherdetails),onecanreadofftheunknown s / c ratiosasafunctionof
givenvaluesoftheparameter ( b ) 2 2 1/ 2
.
Figure224:Criticalloadsofimperfectionsensitivestructures:[Plotsof S C vs
( b )
1
2
2 2
forvariousratiosof ]
22.3.3 Analyses Applying SRA Programs

22.3.3.1 Overview
TheimperfectionsensitivityanalysesareperformedapplyingthefollowingseriesofSRAprograms:
SRA 200 computes the nonlinear large deflection stress and displacement response to axisymmetric
torsionless loads. The nonlinear response is computed by an iterative process based on Newtons
method
SRA201isusedtodeterminetheasymmetric(harmonic)bucklingmodesofaxisymmetrictorsionless
prebuckling states. Geometrically speaking, the method used consists of seeking bifurcation of
fictitiousequilibriumstatesonthetangenttononlinearloaddeformationcurveataloadlevel .
SRA 202 is used to determine the initial post buckling behaviour and imperfection sensitivity of
unique harmonic bifurcation buckling modes of axisymmetric torsionless prebuckling states. The
programisbasedonKoitersfirstorderimperfectiontheorywhichpredictsthebucklingloadknock
431
ECSSEHB3224A
24March2010
down s / c due to small imperfections in terms of the second post buckling coefficient b and
imperfectionformfactors and .TheknockdownfactorcanbeobtainedfromFigure224.
Two configurations of Interstage 1/2 are analysed, namely the monocoque and a stringerstiffened
version.(Thestringerstiffenedversionisanalysedastradeforanalternativedesignconceptforwhich
thepanelswouldbemanufacturedbyshotpeening).
22.3.3.2 SRA Analysis Models

For the SRA analyses of the stringerstiffened structure, the Interstage 1/2 is assumed to have 72
longitudinalstringersintheAftPartand60stringersintheFwdPart.Theoveralldimensionsofthe
monocoqueandstringerstiffenedstructurearethesameandareasfollows:
a. BottomradiusR=1490mm
b. TopradiusR=927mm
c. MeridionallengthL=2211mm
The main interface rings and intermediate ring frames are modelled by their representative areas,
secondmomentsofareaandoffsetsfromthegenerator.
TheconeissimplysupportedatitsbaseandradiallysupportedattheTopI/FRing.
ElasticitymodulusisE=70GPaandPoissonsratio = 0.3
A skin thickness t=6.35 mm is applied for the monocoque version but, for the stringerstiffened
version,theskinthicknessisreducedto5.4mm,compensatingforthematerialintheaddedstringers,
thusobtainingthesamecrosssectionalarea.ThestringersinbothFwdandAftPartsareassumedto
haveaheightof27mmandwidth5mm.
22.3.3.3 SRA Analysis Results

Analysesareperformedforanaxiallineloadof N = 1.5E 6 N/mappliedattheTopring.
TheresultsofthetwosetsofanalysesaresummarizedintheTable221.
Table221:SRAanalysisresultsformonocoqueandstringerstiffenedIS1/2
conestructures
Parameter ValuecalculatedbySRA
Monocoque Stringer
stiffened
B C B
1.1902 1.429
NC 1.785*106N/m
P P 2.144*106N/m
P P
b 11242 2123
0.1519 0.2031
0.2149 0.1850
C 0.3520 0.2815
Havingdeterminedthevaluesofthecoefficients,wecansolveequation226.
432
ECSSEHB3224A
24March2010
The knock down factors s / c (KDF) are presented in Figure 224 as function of the parameter
( b )
1
where isafunctionoftheRMSimperfection (seeEq.227).
2 2 2
( = 100 %meansthattheRMSimperfectionisequaltotheconeskinthickness).
The imperfection sensitivity trends are shown in Figure 225 for both the monocoque and stringer
stiffenedshells.Thestringerstiffenedconicalshellshowsgenerallylowerknockdownfactorsdueto
imperfections. Furthermore, the classical buckling load is approximately 20% higher for a stringer
stiffenedcone.
Figure225:Comparisonimperfectionsensitivitymonocoqueandstringer
stiffenedshell
22.4 Finite Element Analysis
22.4.1 Structural Mathematical Model

ThestructuralmathematicalmodelisaverydetailedandhighlyrepresentativeMSC.Nastranmodel,
representingallstructuralelementsoftheInterstage1/2.PanelsaremodelledapplyingQUAD4shell
elements,andthethreemainringsarerepresentedbyHEXAsolidelements;thustheoffsetatthering
connectionjointisrealisticallymodelledforbucklinganalysis.
TheringframesaremodelledasBEAMelementswithappropriatesectionproperties.
BEAMelementsarealsousedtomodelthestiffenersrunningalongtheedgesoftheopenings.
Equipmentsandretrorocketunitsareincludedinthestructuralmathematicalmodelasconcentrated
masses(CONM2),sothatlocalinertialoadscanbecombinedwiththemainthrustloadsinanalysesof
thestructure.
433
ECSSEHB3224A
24March2010
Table222presentsthemaximumtotalnumberofGridpointsandElementsinthemodeloftheVega
IS1/2.
Table222:QuantitiesofelementsandGRIDSusedinFEM
Typeof TotalNumber
element ofmeshpoints
used
GRID 63669
QUAD4 31262
CONM2 110
BEAM 736
HEXA 16558
22.4.2 Buckling analyses of flight configuration

Linearized and nonlinear buckling analyses are performed in MSC.Nastran applying the detailed
structural mathematical model. The boundary conditions represent the flight configuration, i.e. the
local flexibilities of the structures adjacent to Top and Bottom rings are modelled by including
portionsoftheZ23SRMandP80SRMStagesintheoverallmodel.
The SOL 106 solution method of MSC.Nastran is applied for the nonlinear buckling analysis.
MSC.Nastran provides SOL 106 as a structured solution sequence for nonlinear static analysis,
whichfacilitatesrestartsfromintermediateanalysisresults.
This sequence provides an incremental procedure (conventional NewtonRaphsons method) and
pathfollowingprocedures(ArcLengthmethods).Thenonlinearequationsaresolvedbycontinuation
methods,alsoknownasincrementaliterativemethodsorpathfollowingmethods.Thesemethodsare
designedtocomputetheloaddeformationpathsfromthegoverning(discretised)equations.
For the analyses, the worst combination of flight loads is applied. This consists of the main thrust,
shearandbendingmomentonthestructure,supplementedbylocalinertialoadsonequipmentsand
aerodynamicloadsonprotuberances.Asurfluxcomponentcorrespondingto15%ofthemaximum
compressivefluxontheTopI/Fflangeisalsoincluded,allowingforpeakloadsfromtheZ23adjacent
structure.
Thebucklinganalysisisperformedintwomainssteps.
A nonlinear buckling analysis is first performed applying an envelope load case comprising a
uniform compressive load simulating the maximum flux at the Top Ring I/F. The purpose of this
analysisistwofold:firstly,weuseittoidentifytheweakestsideorsectionofthestructuresothatthe
loadingdirectioncanbedeterminedforsubsequentanalysesapplyingaworstcasecombinationof
axial, shear loads and bending moment. The second purpose, or usage, is the determination of the
bucklingmodeshapestobeappliedasassumptionsfortheshapeofinitialimperfectionsinthecone
structure.AsummaryoftheresultsisshowninTable223.
Having established which side of the Interstage 1/2 is most sensitive for buckling failure, the worst
loadscombinationissetupsothatthemaximumshearloadandbendingmomentproducemaximum
compressive flux in that side of the structure. The 15% surflux supplement, equipment inertia and
aerodynamic pressure loads are added to complete the load set which is then applied in a second
434
ECSSEHB3224A
24March2010
series of buckling analyses. In this series, initial imperfections are simulated based on the buckling
modeshapescomputedfromthefirstanalysiswithuniformaxialload.
Theflightlimitlevelsofthemainaxial,shearloadsandbendingmomentareappliedasinitialloadsin
thebucklinganalyses(seealsoFigure226).TheseloadsareappliedatthecentreoftheTopRingI/F
flangewhichisnotaphysicalpointonthestructurebutwhichisconnectedtotheflangeat360points
around its circumference by RBE2 rigid body elements. The grid point at the centre of the ring
representstheindependentnodeandtheconnectionpointsaroundtheringarethedependentnodes.
Loadsapplied at the independent grid point are thus distributed along the edge of the top I/F ring.
The radial translational DOF of the dependent nodes is not restrained, so the Top ring is free to
deforminradialdirection.
Theseriesofbucklinganalysesapplyingflightloadsandinitialimperfectionsconsistsoffiveanalysis
runs. In four of these runs a buckling failure mode derived from the first series of runs with
envelopedloadandscaledtoamaximumdeformationofahalfskinthicknessisappliedasinitial
imperfection. The scaling of the buckling form to a half skin thickness, 3.15 mm represents a
conservative estimate justified by manufacturing and assembly tolerances of actual qualification
hardware.Inafifthandfinalrun,thefourbucklingmodesarecombinedasimperfection,againscaled
toahalfskinthickness.
TheresultsofthebucklinganalysesapplyingworstcasecombinedloadingaregiveninTable224.
Inthefirstcolumnthemodeshapes(fromthefirstanalysis)usedasbasisforinitialimperfectionare
listed. The second column presents the reserve factor of the calculated critical buckling load as a
multiplieroftheappliedflightlimitloadlevels.
Table223:Modeshapesobtainedfrombucklinganalysisapplyinguniformaxial
load(modeshapesappliedasinitialimperfectionsinseriesofanalyseswith
combinedworstcaseloading)
Mode#. Buckledform
435
ECSSEHB3224A
24March2010
Figure226:FEmodelofIS1/2andadjacentstructuresshowingloadapplicationof
axialthrustloadandbendingmoment
Table224:ResultsofNastrangeometricalnonlinearbucklinganalysesapplying
SOL106
Imperfection Reservefactor(1st Buckledform
P P
modeshape failuremode)
#
1 1.381

2 1.549

3 1.644

4 1.392

Combined
1.550
(1+2+3+4)

436
ECSSEHB3224A
24March2010
22.5 Test Campaign
22.5.1 Overview
StatictestingoftheIS1/2wassplitintotwoseparatecampaigns,usingthesametestjigandtestfacility
at TNOCentre for Mechanical and Maritime Structures Delft, The Netherlands. The first test
conducted in April 2006 was the qualification test of a qualification model of the IS1/2. After
correlating the FE model to the test data, testing was resumed in August 2006, this time with the
purposeoftestingthestructuretofinalfailure(seeFigure227).
22.5.2 Test Jig

The test jig is constructed in steel, including upper and lower adapter cylinders, simulating the
adjacentstructures.Constructionofthetestjigallowsapplicationofthemainthrustloadasanaxial
loadalongthelauncheraxisbutalsoasloadoffsetfromthelauncheraxis.Theoffsetfromlauncher
axis is calculated to provide the required axial thrust and bending moment simulating flight
qualification conditions. For characterization of overall stiffness properties, the IS1/2 was tested for
pure axial load and eccentric loading for two orientations of the test article, one of which
correspondedwiththeworstloadingdirectionforbucklingstability.
SincetheanalysesidentifiedalikelihoodofbucklingfailureoccurringinregionsadjacenttotheRetro
Rocketopenings,provisionswereincludedinthetestsetuptoapplyanadditionalloadononeofthe
Retro rockets, effectively taking into account the inertia load of this item and the drag and
aerodynamicpressureloadsontheRetroRocketfairing.
The test article is loaded by hydraulic jacks pullingon a centralcolumn positionedinternally in the
testsetup;theloadistransferredintothetestarticlethrougha150mmthickcircularsteelplatefixed
onthetopboundarycylinder.Anadditionalhydraulicjackinsidethetestsetupappliesaloadona
retro rocket assembly to simulate local equipment inertia and fairing pressure loads at one of the
openings.TheextraloadisappliedinthedirectionoftheRetroRocketthrustvectorwithacomponent
forceactingnormaltotheskinsurface.
22.5.3 Test Correction Factors

Aqualificationtestisthefinalstepintheverificationprocessofastructureundermissionconditions.
The qualification covers conditions such as elevated temperatures during flight when material
propertiesaredegradedbutalsoprovidevalidationforminimummanufacturingtolerancesandthe
differencesbetweenactualflightandtestjigboundaryconditions.Particularlyforastructurewhichis
bucklingcritical, the boundary conditions can have a major impact on the critical load and failure
mode.Thereforeextraanalysesarecarriedoutonthestructureinthetestconfigurationtodetermine
socalledcorrectionfactorsforapplicationtothetestloads.
The requirements for derivation of correction factors are specified in Vega project documentation,
followingthemethodologyusedforthedevelopmentoftheARIANE5launcher.
Thecorrectedtestloadsareexpressedbytheequation:
PQ = Plim j C 228
where: PQ isappliedtestload(qualificationlevel)
Plim isappliedlimitload
437
ECSSEHB3224A
24March2010
jC = ( j K min K adj + K T )
1
229
K K
jC isthecorrectedsafetyfactor(seeRef.[8]forfurtherdetails),applyingj=1.25forultimatefailure
duetogeneralbuckling.
Theconstituentfactors K min and K T coverthicknesstoleranceandtemperaturegradienteffectsand

K and K arematerialallowablecorrections.Theremainingfactor K adj representsthecorrection
thatisusedtoaccountforthedifferencebetweenflightandtestboundaryconditionsandisobtained
bycomparingtheresultsofanalysesperformedfortheIS1/2underflightandtestsetupconditions.If
K adj 1.0 thenthisfactor,liketheothers,alsoleadstoanincreaseintheappliedtestloadrequiredto
qualifythestructure.
22.5.4 FE Analyses of Test Set-up

The FE model of the test setup includes a detailed representation of the top and bottom boundary
cylindersandthebasestructure.
The buckling analyses to determine K adj were limited to linear bifurcation analyses applying the
NastranSOL105solutionmethodanddisregardinginitialimperfections.Althoughthisapproachcan
beexpectedtoyieldoptimisticresults,itisconsideredadequatetocharacterizethedifferencebetween
testandflightboundaryconditions.
Figure227:Interstage1/2qualificationmodelintestsetup
Analysisofthetestarticleconsiderstheapplicationofthemaineccentriccompressionloadcombined
with the load on the Retro Rocket, while the analysis of the flight configuration includes the
equipment inertia loads, fairing pressure loads and surflux contributions. Results of the analyses,
expressedasreservefactorsoftheappliedlimitload,areasfollows:
Forflightconfiguration, Pcrit = 2.25 Plim 2210
Fortestconfiguration, Pcrit = 2.51 Plim 2211
The test jig factor, K adj is the ratio of the buckling critical loads calculated for the test and flight
configurationsoftheIS1/2,i.e:
438
ECSSEHB3224A
24March2010
Pcrit , test 2.51

K adj = = = 1.12 2212
Pcrit , flight 2.25
Since general buckling is an ultimate failure condition, the safety factor j =1.25 is applied. From
equation 229, applying all the correction factor contributions K min , K adj , KT , K and K , the
correctedsafetyfactor jc isfoundtobe1.53,i.e.qualificationultimatetestload, T
PQ = Plim 1.53 = 1743 1.53 = 2667 kN 2213
22.5.5 Static Test results

TheInterstage1/2wasqualificationtestedapplyingthecorrectedultimatetestload PQ of2667kNat
anoffset645mmfromthelauncherXaxis.
In the rupture test subsequently performed in August 2006, the Interstage 1/2 was tested to the
collapseloadof3034kN.ThefailuremodewasgeneralbucklingintheFwdPart,extendingtotheAft
Part in the region just below the separation ring. The structure was inspected after failure with the
followingobservations:
a. IntermediateringinFwdPartseverelydistorted(buckled)atthree(3)locations,
b. Separationringdeformedatfrangiblesection,convexform,
c. FwdandAftpanelsclosetoseparationringdeformedintocircumferentialwaveform.
The buckled panel deformations are mapped by contour lines drawn on the hardware, as shown in
Figure228.
Figure228:DeformedInterstage1/2structureafterfailureduetogeneral
buckling.(Eachcontourlinerepresents0.5mmdeformation)
The failure mode corresponds well with the 4th buckling mode predicted by analyses of the flight
structure.Thismodedoesnotgivethelowestbucklingloadbut,ascanbeseenfromTable224,the
correspondingcriticalbucklingload,reservefactor1.392isveryclosetothelowestvaluecalculated
(mode1withreservefactor1.381).
439
ECSSEHB3224A
24March2010
No failure or permanent deformation was identified in the region adjacent to the retrorocket
openings,theareawherefirstbucklingwaspredicted.
22.5.6 Correlation between test and FE analyses

Nonlinear buckling analyses applying the FE model of the test setup and the initial imperfections
modeledasamixofthefirstfourbucklingmodesscaledtoahalfskinthicknessresultedinabuckling
criticalload, Pcrit = 2.5 Plim ,i.e.4357kN(seeFigure229).Thisisanoverestimationofthecollapse
load,3034kN.TheFEmodelwasthereforemodifiedtorepresenttheneutralsurfaceoffsetofthecone
panels and the analyses repeated. Results then obtained for nonlinear analyses with and without
initialimperfectionsmodelledasamixofthefirstfourbucklingmodesareasfollows:
Withimperfections: Pcrit = 2.0 Plim = 3486 kN

Withoutimperfections: Pcrit = 2.14 Plim = 3730 kN
Comparing the above results, we see that the modelling of the panel neutral panel offsets has a
significanteffect,leadingtoa20%lower,moreaccurateestimateforthebucklingload,however,the
analysisresultisstillapproximately15%higherthanthetestresult.Anexplanationforthiscouldbe
that the modelling of the imperfections by buckling modes is too optimistic. Comparing the test
collapse load, 3034 kN with the analysis result, 3730 kN obtained for the analysis done without the
initialimperfections,weseethattheeffectiveknockdownfactoris0.81.
Figure229:AnalysisresultforgeneralbucklingofIS1/2intestconfiguration
(correlatedFEmodel)
22.6 Conclusions
The sensitivity of the IS1/2 monocoque cone has been analysed applying Koiters asymptotic post
buckling theory in calculations performed in the SRA programs. Results are compared with an
equivalent stringerstiffened structure and both configurations show moderate sensitivity to
imperfectionsalthoughastringerstiffenedstructureisfoundtobethelesssensitiveofthetwo.Onthe
440
ECSSEHB3224A
24March2010
basisoftheSRAanalysesandassuminganRMSimperfectionof50%oftheskinthickness,aknock
downfactorofapproximately0.7canbejustifiedfortheIS1/2orsimilartypeofstructure.
IndetailedMSC.NastranFEanalysesoftheIS1/2applyingthegeometricalnonlinearsolutionmethod
SOL106,initialimperfectionswereintroducedasbucklingmodesscaledtoahalfskinthickness.The
analysesofbothflightandtestconfigurationsidentifiedthebucklingmodeandlocationintheFwd
Partwherethestructurefinallyfailedduringtherupturetest.
The Interstage 1/2 is manufactured from relatively thick panels and this led to an overestimation of
bucklingstrengthbecausetheneutralsurfaceoffsetwasnotadequatelyrepresentedintheFEmodel.
Duringcorrelationofthetestsetupmodel,theoffsetswereintroducedintotheFEmodelresultingin
abetterpredictionbutstillanoverestimationofthebucklingstrength.
The remaining discrepancy of approximately 15% between test and analysis suggests that the
representationofinitialimperfectionsasbucklingmodesscaledtohalfskinthicknessmayhavebeen
too optimistic. Comparison of the results of the analyses of the test setup with and without initial
imperfections shows an effective knockdown factor of 0.93 while the test demonstrates that the
knockdownfactorshouldnotexceed0.81.
22.7 References
[1] Cohen,G.A.,ComputerAnalysisofAsymmetricBucklingofRingStiffenedOrthotropicshells
ofRevolution,AIAAJournal,Vol.8,No1,January1968,pp.141149.
P P
[2] Anonym.,BucklingofThinWalledTruncatedCones,
NASASP8019,September1968
[3] A.Spagnoli,e.a.,BucklingDesignofStringerStiffenedConicalShellsinCompression,
JournalofStructuralEngineering,January1999,pages4048
[4] W.T.Koiter,TheStabilityofElasticEquilibrium,TechnicalReportAFFDLTR7025,1970
[5] B.Budiansky,DynamicStabilityofStructures,editorG.Hermann,chapterDynamicBuckling
ofElasticStructures:CriteriaandEstimates,PergamonPress,1967
[6] Cohen,G.A.,ComputerProgramforAnalysisofImperfectionSensitivityofRingStiffened
ShellsofRevolution,NASACR1801,1971
[7] Arbocz,J.andHol,J.M.A.M.,ANILISAComputationalModuleforKoitersImperfection
SensitivityTheory,ReportLR582,FacultyofAerospaceEngineering,TUDelft,The
Netherlands,January1989.
[8] Anonym.,ARIANE5Program.Structuraldesign,dimensioning&testspecification,A5SG1
XASAI,2001.
441
ECSSEHB3224A
24March2010
22.8 Abbreviated Terms and Symbols

Thefollowingabbreviatedtermsandsymbolsaredefinedandusedwithinthischapter:
a Firstpostbucklingcoefficient
b Secondpostbucklingcoefficient
E Youngsmodulus
R Radius
h Wallthickness
Imperfectionformfactor
Imperfectionformfactor
RMSImperfectionnormalizedwithrespecttothewall
thickness
Knockdownfactor
c Classical(critical)normalizedbucklingload
s Imperfectbucklingload
Imperfectionamplitude
Poissonsratio
c Semivertexangleofcone
Pcrit Criticalbucklingload
DOF DegreeOfFreedom
FEM FiniteElementModel
IS1/2 VegaInterstage1/2
SRM SolidRocketMotor
TNO NetherlandsOrganisationforAppliedScientificResearch
442
ECSSEHB3224A
24March2010
23
Stability analysis of the 3rd-stage skirts of
the ELDO-A
23.1 Overview
The ELDOA or EUROPA1 was a threestage launch vehicle developed by several European
countriesunderthecoordinationoftheEuropeanLaunchvehicleDevelopmentOrganization.Itnever
U U U U U U U U
succeededandafteranumberofmissionfailurestheprogramwascancelledin1972,givinglater
freewayforthesuccessfulARIANEprogram.
However, the technical effort of all participants of the ELDO program was not in vain. It was an
excellent training period, forming a European technical team with sufficient knowhow for future
activitiesinspaceprogramsandinthemanagementofthem.
TheELDOAlaunchvehicle(Figure231)wasathreestagerocketwithatotalmassofabout110tons.
Thefirststage,builtinGreatBritain,wasaderivativeoftheUSATLASmissile,havingaspropellants
kerosin/LOXandamassofabout94tons.Thesecondstage,aFrenchcontribution,wasthesuccessor
oftheVERONIQUEwithUDMH/N2O4 aspropellantsandamassof12tons.Thethirdstage,builtin
B B B B
WestGermany,hadnopredecessor.Itwasafullynewdesignwithdifferent,sometimescontradictory
design constraints. One of them propellant mass versus external diameter led to a spherical tank
(diameter1.63m)inwhichacommonbulkheadseparatedthetwopropellants.Thetankwasmounted
inacylindricalshell(diameter2.0m)forminga3.8mlong,nonpressurizedskirtwithamassof3.7
tons.
Main concern of this Large Example is the design and the analysis of the cylindrical shell above as
wellasthestructuralqualificationtestsservingtoverifythedesign.
23.2 Structural description and design aspects

Due to kinetic heating the very first thermal analyses predicted skin temperatures up to 400C.
Therefore,aluminiumalloyswereexcludedasmaterialsfortheexternalstructure.Inconsequence,the
externalloadcarryingshellwasbuiltupofthreeparts,seeFigure231:
Upperpart:Ti13.11.3titaniumalloy
Foreseen,toattaintheorbittogetherwiththepayload
Middlepart:AISI301hard,transversal,stainlesssteel
Foreseen,tobejettisonedsoonaftertheseparationfromthe2ndstageP P
Lowerpart:AISI301hard,transversal
Same design as middle part. Fixed at bottom crosssection to the 2nd stage, using radial bolts,
P P
eachfittedindividuallyduringassembly.Thetopcrosssectionofthelowerpartwasprepared
443
ECSSEHB3224A
24March2010
forthestageseparationhavingafittedplaneintersectiontothemiddlepartandaconnection
toitvia12axial,explosivebolts.
All three parts (skirts) were manufactured from thin metallic foils forming an orthotropic stiffened
shell. The shell consists of an axial, semicircular corrugation (see Figure 232) of a thin cylindrical
shearskin,stiffenedbystiffringswithclosedhatprofilecrosssection,whicharebuiltupfromthin
foilsaswell.Thepitchoftheringsis100mm.
All these thinfoiled elements are joined together by electrical pointwelding and/or linewelding.
Insideoftheringframesinstalledarestiffeningbulkheadsinordertopreventthecrosssectionfrom
shapedeformation.
Figure231:TheELDOA(orEUROPA1)launchvehicleanditsthirdstage
444
ECSSEHB3224A
24March2010
Ringstiffenerpitchis100mm
Figure232:Designoftheexternalshellstructureofthethirdstage.
23.3 Input data
23.3.1 Material
The characteristics of the materials used as specified by the ASME can be seen in Table 231. A
specialandastonishingcharacteristicofthestainlesssteelfoilsisthattheyieldstrengthintransversal
directionistwiceofthatintherollingdirection.Becausethewidthoftherollswaslessthanthelength
ofthemiddlepartanoverlappingbecamenecessaryforthefabrication.
445
ECSSEHB3224A
24March2010
Table231:ELDOA,materialpropertiesoftheorthotropicshell
Sheet Rm R p 0.2 E 0.7 0.85 fr

material MPa MPa MPa MPa MPa %
AISI301 812 747 159800 747 668 8.6
Ti13.11.3 980 850 91000 13.7
Table232:MaterialpropertiesfromBruhn[1]
Temp. Ftu Ec F0.85
Temp e Fcy F0.7
Material Exp. n
Hr.
F %
ksi
ksi 10 6 psi ksi
ksi
AISi301Hard
Sheet: RT 25 125 80 27.0 73 63 6.9
1/2
transversecompression RT 25 125 43 26.0 28.2 23 5.2
longitudinalcompression
AISi3011/2Hard
Sheet: RT 15 150 118 27.0 116.5 105 9.2
transversecompression 400 15 118 108.5 23.2 108.5 97 8.6
600 110 107.5 20.9 108.5 96.5 8.2
longitudin.compression
1/2 1000 86 85 16.2 94.5 83.5 8.0

RT 150 58 26.0 48 37 4.4

400 118 53.3 22.4 45.6 36 4.7

600 110 52.8 20.1 44 31 3.5

1000 86 45.2 15.6 40 30.5 4.3

Thestressstraincurvewasalsomeasuredinhouse(seeFigure233).However,foralltheanalysesthe
valuesspecifiedintheTable232wereapplied.
446
ECSSEHB3224A
24March2010
Figure233:Stressstraincurvesofthematerialsused(measuredinhouse)
23.3.2 Loadings and dimensioning load cases

Threedimensioningloadcases(LC)havebeendefined[2]:
LC1: Launch preparation and ignition/release, taking into account Krmn vortex shedding
excitations
LC2: Jet stream wind loads in 11 km altitude, taking into account mission profile, axial
acceleration,drift,gusts,andvehicleaerodynamics
LC3: End of 2nd stage flight, taking into account the axial acceleration, a small lateral
P P
acceleration, and the high skin temperature due to kinetic heating (significant aerodynamic
loadingsdonotexistatthisaltitude).
ItwasfoundthatforshellstabilityLC2isthemostrelevantloadcase[3].InLC2,inadditiontothe
bendingmomentalsointhiscasealleviatingwarpingloadsareconsideredwhicharegeneratedon
the relatively long nonpressurized cylindrical shell through the crossflowinduced ovalizing
aerodynamic loading. The maxima of the unit loads of the axial fluxes are plotted for the generator
location ( = ) in Figure 234. They were reduced later to a reasonable level. The windinduced
loads were actually checked prior to each launch by prelaunch load checks based on local
atmosphericwindmeasurementsviameteorologicalballoons.
447
ECSSEHB3224A
24March2010
LC2Flightin11kmaltitude,time66sec, x& =1.63Mach,1000kgpayload:Considered

areaxialacceleration,windandgust,drift,bendingmoment(static+dynamic)and
ovalizing(warping)stresses.
a)Distributionofaxialfluxcomponentsoveraxiallengthoflauncher
b)DistributionofaxialfluxcomponentsoverhoopatcrosssectionAA
Figure234:Axialfluxesinthegenerators = 0 and = .Firstestimates.
448
ECSSEHB3224A
24March2010
LC3 is mainly relevant for the stresses in the heated skin. For this purpose, beside the mechanical
stressesthermalstressesareconsidered,too,seeFigure235.Whencalculatingtheresultingstressat
designultimateloadlevel(linearanalysispermitted),
ult ,Re s = DLL jult + ther 231
Thermalstressesarenotfactorized.Theyarecomputedfromtheequation
ther 24.8 T1 + 12.5 T2 [ kp / cm 2 ] 232
LC3Flightattheendofthe2ndstagepropulsion
(1 kp/cm 2
= 9.807 N/cm 2 = 0.098 MPa )
Figure235:Evaluationofthethermalstresses
23.3.3 Analysis of material behaviour

23.3.3.1 Overview
InordertoperforminelasticanalysestheRamberg/Osgood(R/O)relationship[4]isapplied,however
atfirst,inanapplicationformwhichusesthesecantyieldstressasreferencestress s .Theadvantage
of this type of equation is that, once the value of s has been determined, only a single material
dependent parameter exists, termed the R/O exponent n , thus making graphic representations
simpler, because less curves are required. Usually 0.7 = s is used, because for many metallic
449
ECSSEHB3224A
24March2010
materials 0.7 approximates the widely used 0.2 % yields strength Rc 0.2 . Especially in buckling the
curvesareoftenmadenondimensionalwith 0.7 .
1 / ( n 1)
3 R c 0 ,2
0.7 = R c 0.2 233
7 0. 002 E
ThedoublydimensionlessR/Ostressstrainrelationshipbecomes,
n
E 3
= + . 234
0. 7 0 .7 7 0.7

withtheR/Oexponent,
ln(17 7)
n =1+ 235
ln( 0.7 0.85 )
which are derived from test data and may be found for various lightweight materials in the MIL
HDBK5[4].Valuesofnarebasedonaverage(typical)R/Ocurves.Andfinally,
E 0.7 = 0.7 E , E 0.85 = 0.85 E . 236
For the inelastic analyses, also the tangent modulus and the secant modulus are needed. They are
obtained,seeFigure236,
Tangent modulus as the instantaneous slope of the stressstrain curve (for comparison the
classicalR/Orelationshipneededlaterisaddedhere)
d E E
E
tan = = n 1
= n 1

d E 3 237
1 + 0.002 n 1 + n
Rc 0.2 R c 0.2

7 0.7
Secantmodulusistheinstantaneousratioofstresstostrain(seeFigure236,leftdiagram)
E E
E
sec = = n 1
= n 1

E 3 238
1 + 0.002 1+
Rc 0.2 R c 0.2 7 0.7

Offurtherinterestforstabilitydesignisthestrainformula
st = / Esec E tan 239
Undertheconditionofuniaxialloading,theabsolutevalueoftheratiolateralstraintoaxialstrainin
theelasticregion(Poissonsratio)is = el .Enteringtheplasticrange,thevalueof increasesupto
0.5whileincreasingstress.Forstabilityanalyses(compression,shear)ofmetallicmaterialsitisusually
appliedintheplasticregionas
Esec
= 0.5 (0.5 el ) 2310
E
Another R/O relationship is necessary for analysis when the R/O exponent is not known. Here, the
classicalR/Orelationshipisused(seeChapter7):
U U
n

= + 0.002 , 2311
E Rc.2
450
ECSSEHB3224A
24March2010
R0.2
1.2 + 0.002 , 2312
E
R
n = ln ( Aus / 0.2 ) /(ln m ) , 2313
R
0.2
Inprinciple,theEqs.2311through2313arevalidfortensionandcompression.However,forthemost
t
often ductile behaving lightweight materials a tensile strength R m together with the associated
permanent strain Aus at Rmt Rm , to be inserted in %, exists only, but not the compressive
equivalent. Therefore in the compressive case, the R/O exponent has to be determined via Eq.235.
This is sufficient for the classical stability analyses which practically are bound to small permanent
strainsortothelevel Rc 0.2 respectively.
Figure236:StressstraincurvedescriptionbyRambergOsgood
23.3.3.2 Application to ELDO-A materials:

AISI 301 hard, transversal, input data: Rm = 812MPa, Rc 0.2 = 747MPa,
0.7 = 747 MPa , 0.85 = 668 MPa , E = 159800MPa , n = 8.6
TitaniumalloyTi13.11.3,inputdata: R m = 980MPa , Rc 0.2 = 850MPa, E = 91000MPa , n = 13.7
yieldsthecurvesshowninFigure237.
a)AISI301hard,transversal, b)TitaniumalloyTi13.11.3
Figure237:RambergOsgoodmappedstressstraincurvewithvisualizationofthe
functions ( ) , tan = / Etan , st = / Esec Etan
451
ECSSEHB3224A
24March2010
23.4 Stability failure modes and load-carrying capacity of

the shell
Figure238showsthefailuremodesofastiffenedcompressionloadedorthotropicshell:
Generalinstability:
With eigenmodes including both, ring deformation and stringer deformation. This stability
failure mode is analyzed with smeared stiffness; plasticity is considered. Rings (frames) are
modelledlikesprings.
Panelinstability:
Buckling of the stringers as elasticplastic columns with rings keeping their form. For this
analysistheEulerJohnsonconceptisapplied
Cripplingaslocalfailureofthestringerprofile:
Estimationbyempiricalmethods
Whendimensioning,itisdesirabletohaveaconsiderabledistancebetweenthethreestabilityfailure
mode levels. Otherwise, the three modes come closer to another due to the fact that joint failure
probabilitycomestoact.
Usually,thecripplinglevelischosenasfirst.Withthisvalue,panelinstabilitycanbesizedonstiffness
and ring pitch, applying the Shanley formula [5]. Dimensioning for general instability (global
buckling) based on smeared stiffness values and appropriate knockdown factors (KDFs) is
expectedtoshowsufficientmarginsofsafety(MoS)withrespecttopanelinstability.
So, the layout of the orthotropic ELDO shells followed the following sequence: beam buckling
crippling globalbuckling,aspointedoutintheparagraphsbefore.
a)Generalinstabilitywithstringerandringdeformation
b)Panelinstabilitywithbucklingofthestringersbetweentherings(ringsremaincircular)
c)Cripplingaslocalfailureofthestringers
Figure238:Failuremodesofastiffenedcylindricalorthotropicshellunderaxial
load
452
ECSSEHB3224A
24March2010
23.5 Local failure of stringer profiles under compression

elastic local buckling and elastic-plastic crippling
23.5.1 AISI steel

The crippling strength of the semicircular corrugation can be estimated to be equal to the elasto
plasticbucklingstrengthofthecorrespondingthinwalledisotropiccylindricalshellwithusualshape
imperfections.Inordertotakeintoaccounttheinfluenceoftheshapeimperfectionscatterstatistical
KDFsareestablishedbyseveralauthors,seealsoChapter9.Twoofthemaregivenhere,bothdeliver
valuesforp=99%survivalprobability(reliability),visualizedinFigure239:
Almroth[6]foraerospacequality(cutoffat = 1 ):
= 6.48 /( R / t ) 0.54 , 2314
DASt013[7]forcivilengineeringstandardquality:
= 5.20 / 100 + R / t . 2315
Therebyitisassumedthatthesamplefullyreflectsthebasicpopulation(samestandarddeviationand
mean value). In other words, the confidence level when transferring from the sample to the basic
populationis100%.
Figure239:KDFsuggestionsfromAlmroth, ( ) ,andDAST013, ( )
Taking the fact into account that the fabrication method (electric resistance pointwelding and line
welding)doesnotleadtolowimperfectionsthemoreconservativeKDF istaken.
Usingtheclassicalshellstabilityformula
critD = 0.605 E t KDF / R 2316
withtheactualdimensionsofthesteelcorrugation
R = 6 mm , t = 0.15 mm , E = 159800 MPa , t1 = 0.15 mm , t2 = 0.1 mm

andwiththeKDFfromtheDASt013of KDF = = 0.439
thecriticalbucklingstressbecomes
critD = 0.00665 E = 1063 MPa

andthecriticalelasticstrain
critD = 0.00665 = 0.665% .
453
ECSSEHB3224A
24March2010
The corresponding critical plastic stress can be obtained from the curve st of Figure 237a as
cr . pl = 655 MPa .
Theeffectiveaxiallysmearedthicknessreads
t x = t1 / 2 + t 2 / 3 . 2317
Hereinitisrespectedthattheverythinskin t2 = 0.1 mm worksfarbeyonditscriticalstress.Thisis

considered,referringtoMarguerreseffectivewidthformula,bytakingonly1/3t2. B B
Aftermultiplicationwith t x theeffectiveplasticcripplingflux
ncripp = ncr , plD = t x cr , pl = 0.269 655 N / mm = 176 N / mm 2318
canbecomputed.
Inordertoremovetheuncertainty,cripplingtestshavebeenperformedwithappropriatespecimens
anddecisivedesignvaluesevaluatedfromthesample.Thestatisticalscatterofthesamplewaspartly
generatedduetothehighlynonuniformstressdistributionfacedintheperformedtests(seesection
23.8).
Thefixedcripplingfluxesarefinally
Specimenswithoutoverlapping: ncripp ,test = 150 N / mm (p=50%),or145(p=99%)
Specimenswithoverlapping: ncripp ,test = 145 N / mm .(p=50%),or140(p=99%).
23.5.2 Titanium
Withtheactualdimensionsofthetitaniumcorrugation
R = 6 mm , E = 91000 MPa , t1 = 0.20 mm , t2 = 0.20 mm

andwiththeKDFfromtheDASt013of KDF = = 0.439
thecriticalbucklingstressbecomes
critD = critD E = 0.00981 E = 716 MPa .

The corresponding critical plastic stress can be obtained from the curve st of Figure 237 as
cr . pl = 716 MPa .
Withtheeffectiveaxiallysmearedthickness
tx = t1 / 2 + t2 / 3 = 0.38 mm , 2319
aftermultiplicationwith t x ,theeffectiveplasticcripplingflux
ncripp = ncr , plD = t x cr , pl = 0.38 716 N / mm = 272 N / mm 2320
canbecomputed.
Thedesigndecisivevalue,however,istakenfromthetestresults.So,thefixedcripplingfluxisfinally
(seeFigure2314)for
specimenswithoutoverlapping: ncripp ,test = 210 N / mm ( p = 50% )

specimenswithoutoverlapping: ncripp , test = 175 N / mm ( p = 99.9% ) .
454
ECSSEHB3224A
24March2010
23.6 Panel instability: Elasto-plastic buckling of the

stiffened shell between the ring frames
23.6.1 Assumptions:
Thefollowingisassumed:
Thestiffenedshellcanbemodelledasacolumn
Theringframesremaincircular
TheEulerJohnsonmethodcanbeused.
Theslendernessofthecolumnisdefinedaslengthoverradiusofgyration
= Lr / rx 2321
Thecriticalelasticplasticcolumnstressis
1 cripp
22
col = cripp 2322

4 2 E
withthecripplingstressasthevalueat =0
col = cripp . 2323
Eq.2322isaparabolajoiningtheEulerhyperbolaat 0.5 cripp .
TransferringEq.2322withthecorrespondingelasticstrains
col = col / E , cripp = cripp / E 2324
resultsin
col = cripp cripp 2 2 /(4 2 ) 2325
whichisplottedinFigure2310.
23.6.2 Application to AISI steel corrugation:

For the steel corrugation the crippling stress is the value found in the sample test. This value is
evaluatedfromasamplecomposedoftencripplingtestspecimenresults.Aslevelsfor p = 50% and
p = 99% (ofthepopulation)areobtained
( ncripp , cripp )50% = (150 N / mm, 557 MPa ) ,
( ncripp , cripp )99% = (125 N / mm, 464 MPa ) .
455
ECSSEHB3224A
24March2010
Figure2310:ParametricEulerJohnsondiagramforstrains
Indetail,theanalysiswiththeaveragevalue(50%)willbeshown:
For cripp = 557 MPa anEulerJohnsonstresscurveisplottedinFigure2311.
Then,withthedata
l = 100 mm , rx = 4.43 mm , red = 22.6 , E = 159800 MPa , t x = 0.269 mm

TheintersectionoftheEulerJohnsonstresscurvewith red delivers
col = 532.6 and ncol = 0.269 mm 532.6 MPa = 143 N / mm .
Steelcorrugation(AISI301hard,transversal)
BoundaryconditionsSS/SS(simplysupported)
Figure2311:EulerJohnsoncolumnstresscurve ( ) fortheELDOA,
Fortheanalysisaboveitwasassumed,thattheringframesatbothendsofthecorrugatedpanelskeep
theircircularformandensuringbythistheboundaryconditionsSS/SS,hereconservative.Inorderto
fulfilthisconditiontherequiredinplanebendingstiffnesscanbeestimatedviaShanleysformula[5]
Nult R 4
( I E ) Shanley = I E . 2326
1273 LR
Withtheinputdata
R = 1000 mm , LR = 100 mm , E = 159800 MPa

andfromTable233
Nult = ncol = 143 N / mm or N ult = ncol = 120 N / mm .
456
ECSSEHB3224A
24March2010
thisleadsto
I Shanley = 7029 mm3 orI Shanley = 5900 mm3 .

Inordertokeepthegeneralinstabilitylevelhigherthanthepanelbuckling( columnbuckling)one
choosesthevalueof
I Shanley = 6312 mm3 .
23.6.3 Application to titanium corrugation:

FortheTitaniumcorrugationthesameanalysesareanalogouslyperformedwiththeresultsdepicted
inTable233.
Table233:EulerJohnsonbucklingfluxesandstressesattwoprobabilitylevels
Probability ncripp cripp col ncol
ofsurvival N/mm MPa MPa N/mm

50% 150 557 532 143
AISI301
99% 125 464 447 120
50% 210 552 510 193
Ti13.11.3
99% 175 460 431 163
23.7 General instability of the stiffened shell
23.7.1 Assumptions
Infiniteshelllength
Smearedstiffnessofthecorrugatedshell
Uniformaxialload
Lineartheoryapplicable
Chessboardbucklingpattern.
23.7.2 Methods
Approximate calculation of the linearly elastic general instability of orthotropic shells with
eccentricstiffenersbythelineartheoryofvanderNeuts[8]orbyEq.17111.Determinationof
theclassicalcriticalbucklingload.
ImperfectionsensitivityconsideredbyAlmrothsKDF ( R / t ) ,basedonanequivalent R / t
PlasticityconsideredbytheEngesserKrmnprocedure[9],basedonthecriticalstrainconcept
andonthe ST ( ) curve,seeEq.239andtheFigure237andFigure2311.
457
ECSSEHB3224A
24March2010
23.7.3 Van der Neut method
Thecriticalbucklingloadoforthotropicshellsisinthecaseof
Ringbuckling:
ncrit , ring = 2 E rx t /(ax a ) 2327
Chessboardbuckling:
ncrit ,chess = min ncrit ,ring 2328
Thefactor min inthecaseofchessboardbucklingistheminimumofthefollowingfunction
(Y + A + 1 / Y ) (Y + B + C / Y ) + F 2 F
(Y ) = .
Y + A + 1/ Y 2329

containingthelaterdescribedparameter Y .Eq.2329involvesthealsolaterdescribedparametersA
throughF.Asathumbrule,theminimumof (Y ) isaround Y C ,however, ( C ) > min (is
notonthesafeside).
Inordertoobtainacharacteristicbucklingload(orevenadesignbucklingload)thecriticalbuckling
loadisreducedbyaKDF ,whichcanbetakenfrom
99% = 6.48 /( R / t ) 0.54 , 90% = 8.76 /( R / t ) 0.54 ,

2330
50% = 11.86 /( R / t ) 0.54
with
tx
( R / t ) = 0.428 R 2331
I x + I
usingthemomentsofinertiadefinedbelow.
The application of the van der Neut method to derive min is depicted hereafter in detail. As
parametersareapplied:
a =outerradiusofring, b =stringerspacing,c=ringspacing,
t=skinthickness, t x , t :=smearedthicknessin x , (ring)direction,
t x = t + Ax / b , t = A / c ,
Ax ,( A )=crosssectionareaofstringer(ring)incl.effectivewidthofskin,
I x , ( I ) = smeared moment of inertia of: stringer (ring) incl. effective width of
skin,
I x , I x =smearedtorsionalmomentofinertia,
I x = I x a /(b a x ) , I = I / c ,
rx = I x a /(b ax ) , r = I / c , rx = I x a /(b ax ) , r x = I x / c ,
E t t x a x G rx ax 2 + r x a ax
A= , B =
G t 2 a E a 2 rx t ax 3 /(t x a a 2 )
458
ECSSEHB3224A
24March2010
r a 3 t x (a ax ) a + (a a ) a l 2 t x a a 2
C= , F = , Y = ( )
rx a 3 t rx a 2 / t x lx t a x
3
l 2 t x a x a 2
and Y =( ) 2
with l x , l aslengthofhalfwaves.
lx t a x
Figure2312:Generalinstabilityoftheorthotropicshell
23.7.4 Numerical analysis of the steel corrugation shell

23.7.4.1 Input data and results:
Ineachparticularcaseatfirsttheparametersarecalculated:
( a0 , a, ax , a ) := (1000,1003.6,1005.3,980 ) mm ,
( t1 , t2 , l ) := ( 0.15, 0.10,100 ) mm , ( t , tx , t ) := ( 0.160, 0.269, 0.238) mm ,
( I p, x , I p, x , I p, x , I p, ) := ( 5.57,38.27,5.28, 63.12 ) mm3 ,and
( A, B, C , F ) := ( 4.15,3.51,14.51,5.06) .
Figure2313:Minimumsearchof (Y ) , min = 0.688
459
ECSSEHB3224A
24March2010
Then for Y several values are taken and (Y ) is plotted, see Figure 2313. The minimum ordinate
valueisthenfoundas
(Y1 , min ) := (5.17, 0.686) .
Forcontrol,usingasanapproximation Y = C = 14.51 = 3.8 theminimumsearchofFigure2313

deliversmin = 0.75 > 0.686
23.8 Structural testing
23.8.1 Sample tests with the corrugated skin

Inordertoevaluatethelocalloadcarryingcapacityofthecorrugatedshellcripplingsampletestshave
beenperformedwithbothconfigurations.Thedescriptionandtheevaluationofthetestsisshownin
Figure2314.

Figure2314:Cripplingtestswithsteelandtitaniumsamplesofthecorrugated
shell
Thestraingagemeasurementsdepictquiteanonuniformstressdistribution,probablyduetoanon
uniformloadapplication.Thisfactmayexplaintherelativelylargescatteroftheresults.
23.8.2 Static structural tests with the corrugated orthotropic shell

Threeofthenumerousshelltestsarereportedhere,seeTable234.Theexecutionofthefairingtestsat
LehmwerderisshowninFigure2315.
460
ECSSEHB3224A
24March2010
Table234:Schemefororientationonexampleminimumvaluesforbuckling
FoS,specialfactors,andtestfactorsforpressure
(valuesarefixedineachproject)
Objective Maximum Local
LoadCase
Qualificationof LocalFlux MoS
Interfacecoldtest LC2:
1 separationplane 119N/mm 0.13
(atHamburg) Maximumbending
2ndand3rd LC3:Max.axial
mechanical+
2 stageswarmtest acceleration+ 124N/mm 0.41
heating
(atLesGatines) heating
Fairing+2ndand
LC2:Most
3rdstagescold
3 mechanical complete 138N/mm 0.56
test(at
configuration
Lehmwerder)
FortheLC3theMoSiscalculatedconsideringthemaximumdesignflux N x = 88 N / mm inFigure
124 N / mm
234aas MoS = 1 = 0.41 .
88 N / mm
461
ECSSEHB3224A
24March2010
Figure2315:Testsetup,fairingtestsatLehmwerder
23.9 General conclusions

Inthecaseoforthotropicshellsthecriticalloadlevelforchessboardbucklingloadmostoftenis
lowerthanthatwhichcausesringbuckling
Stiffenersoutsideoftheshearcarryingshellaremoreefficient(uptoafactorofmorethan2).
VanderNeutsmethoddoesnotconsidertheinfluenceoftheboundaryconditions
The empirical formulas from Gerard [10] and from the NASAHandbook [11] give unrealistic
lowvaluesforstringerprofiles.Anyhow,theyarenotapplicabletocurvedflanges.
Themodelsarevalidatedinprinciple.
462
ECSSEHB3224A
24March2010
23.10 References
[1] Bruhn,E.F.,AnalysisandDesignofFlightVehicleStructures,TRISTATEOFFSETCompany,
Cincinati,Ohio,1965.
[2] Lucke,H.E.,DieStrukturder3.StufeundihreFertigungsentwicklungELDO/CECLESTech.Rev.
1,93114GauthierVillars,Paris,1966.
[3] ry,H.,StatischeunddynamischeProblemeder3.StufederELDOTrgerrakete.Vortrag
anlsslichderTagungderDeutschenRaketengesellschaftinHamburg,1963
[4] Anonym.,MetallicMaterialsandElementsforAerospaceVehicleStructures,MILHDBK5,
[5] Shanley,F.R.,Simplifiedanalysisofgeneralinstabilityofstiffenedshellsinpurebending,
JournalofAeronauticalSciences,Vol.16,pp.590592,October1949.
[6] Almroth,B.O.,Burns,A.B.andPittner,E.V.,Designcriteriaforaxiallyloadedcylindricalshells,
JournalofSpacecraft&Rockets,Vol.7,No6,pp.714720,1970.
P P
[7] Anonym.,BeulsicherheitsnachweisefrSchalen,DASTRichtlinie013,DeutscherAusschussfr
Stahlbau,1980.
[8] VanderNeut,A.,TheGeneralInstabilityofStiffenedCylindricalShellsunderAxial
Compression,Report5.314,NationaalLuchtvaartlaboratorium,TheNetherlands,1946.
[9] Timoshenko,S.P.andGere,J.M.,TheoryofElasticStability,McGrawHill,NewYork,
p.176,1961.
[10] Gerard,G.,TheCripplingStrengthofCompressionElements,JournalAeronauticalSciences,
Vol.25,January1958.
[11] Anonym.,NASAAstronauticalStructuresManual,GeorgeC.MarshallSpaceFlightCenter,
FORM454,RevisedOct.1967.
463

ECSS HB 32 24A Buckling of Structures

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

ECSS HB 32 24A Buckling of Structures

Transféré par

Droits d'auteur :

Formats disponibles

ECSS-E-HB-32-24A

Change log .................................................................................................................3

3 Terms, definitions and abbreviated terms..........................................................20

Part 1: General .........................................................................................................23

5 Types of instability and failure behaviour of typical structural elements .......64

6 Analysis Methods .................................................................................................87

8 Design to stability and its verification by analysis ..........................................130

10 Modelling aspects of numerical analysis .......................................................168

11 Strategy for hierarchical high fidelity analysis applied to stability

12 Buckling experimental methods and design verification by tests ...............218

Part 2: Structural elements with examples..........................................................243

14 Implementation of Stability Methods ..............................................................244

15 Columns, Beams, Arches and Rings ..............................................................258

16 Flat and curved panels.....................................................................................290

17 Closed shells ....................................................................................................318

Part 3: Structures ..................................................................................................341

18 Buckling design, analysis and testing of large structures ...........................342

19 LH2-Tank ARIANE 5..........................................................................................356

20 Buckling of the Inner Dome of the ARIANE 5 Upper Stage ESC-A...............379

21 Non-Linear Stability Analysis of ARIANE 5 Front Skirt (JAVE) ....................396

22 Buckling Analysis and Qualification Static Load Testing of VEGA

23 Stability analysis of the 3rd-stage skirts of the ELDO-A...............................443

3.1 Terms from other documents

3.2 Terms specific to the present document

3.2.2 buckling mode or shape (with reference to buckling)

3.2.3 buckling modes or shapes (with reference to buckling analysis)

3.2.4 buckling resistance

3.2.5 design buckling load

3.2.6 global buckling

3.2.7 local buckling

3.2.8 margin of safety (MOS)

3.3 Abbreviated terms

4.1 Instability of structures: concepts and definitions

4.2 Stability of columns, beams and plates

4.2.1 Buckling of beams

Experimentalevidenceindicatesthatforaslendernessration L A I > 80 Eulersformulasarequite

4.2.2 Lateral torsional buckling

4.3 History of shell buckling

4.3.3 Postbuckling of the perfect shell

4.3.4 Buckling and postbuckling behaviour of imperfect shells

Authors Literature Year Imperfectionpattern Amplitude

Authors Literature Year Imperfectionpattern Amplitude

4.4 Design load for thin-walled isotropic shells

4.4.2 Empirical knock-down factors

4.4.3 The Southwell method

4.4.4 Numerical analysis with realistic geometrical imperfections

4.4.5 Numerical analysis with worst geometrical imperfection

4.4.6 Numerical analysis with stimulating imperfections

4.4.7 Physics based design load

Source:[136] [136],[61] [139],[137] [137],[75]

4.4.8 Overview of different approaches for modelling imperfections

4.5 Influence of non-traditional imperfections

(a) (b) (c) (d)

4.6 External pressure and torsion on cylindrical shells

4.7 Dynamic buckling

4.8 Stiffened shells

4.8.2 Imperfection sensitivity

4.8.3 Orthotropic shell approach

4.8.4 Eccentricity of stringers

4.8.5 Boundary conditions

4.8.6 Load cases

4.9 Plastic buckling

4.10 Composite shells

800 w0/t = 0,2