Vous êtes sur la page 1sur 685


Proceedings of
SDSS’Rio 2010
International Colloquium
Stability and Ductility of Steel Structures
First published in Brazil in 2010 by Federal University of Rio de Janeiro and State University of Rio de Janeiro

All rights reserved. No part of this book may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, recording or by any information storage and retrieval
system, without permission in writing from the publisher.

International Colloquium on Stability and Ductility of Steel Structures, SDSS’Rio 2010

ISBN: 978-85-285-0137-7

Graphic Art Production: Angela Jaconniani and Lucia Lopes

Printed and Bound by: J. Sholna Reproduções Gráficas Ltda.

Federal University of Rio de Janeiro and State University of Rio de Janeiro
Telephone: (+55 21) 2562-8474 and (+55 21) 2334-0469
E-mail: batista@coc.ufrj.br ; pvellasco@globo.com and luciano@eng.uerj.br

Editors: Eduardo de M. Batista, Pedro C. G. da S. Vellasco and Luciano R. O. de Lima

08 - 10 SEPTEMBER 2010

Federal University of Rio de Janeiro, UFRJ
State University of Rio de Janeiro, UERJ
Structural Stability Research Council, SSRC

Eduardo de M. Batista
Pedro C. G. da S. Vellasco
Luciano R. O. de Lima


S.A.L. Andrade - Brazil J.R. Liew - Singapore

R. Azoubel - Brazil L.R.O. Lima - Brazil
E.M. Batista - Brazil J. Lindner - Germany
R. Battista - Brazil M. Mahendran - Australia
E. Bayo - Spain R. Maquoi - Belgium
D. Beg - Slovenia F. Mazzolani - Italy
F. Bijlaard - Netherlands E. Mirambel - Spain
R. Bjorhovde - USA J.R. Muzeau - France
M.A. Bradford - Australia D.A. Nethercot - UK
I. Burges - UK L.C. Neves - Portugal
D. Camotim - Portugal J. Packer - Canada
P.J.S. Cruz - Portugal G. Parke - UK M. Pfeil - Brazil
J.B. Davison - UK R. Plank - UK
R. Driver - Canada A. Plumier - Belgium
D. Dubina - Romania K.J. Rasmussen - Australia
L. Dunai - Hungary P.V. Real - Portugal
W.S. Easterling - USA J.P.C. Rodrigues - Portugal
R.H. Fakury - Brazil B. Schafer - USA
J.M. Franssen - Belgium J.G.S. Silva - Brazil
T. Galambos - USA L.S. Silva - Portugal
P.B. Gonçalves - Brazil N. Silvestre - Portugal
R. Greiner - Austria M. Skaloud - Czech Rep.
G. Hancock - Australia H. Snijder - Netherlands
J.-P. Jaspart - Belgium R. Souza - Brazil
V. Kodur - USA B. Uy - Australia
U. Kuhlmann - Germany P.C.G.S. Vellasco - Brazil
R. LaBoube - USA A. Wadee - UK
D. Lam - UK F. Wald - Czech Rep.
A. Landesmann - Brazil B. Young - Hong Kong
R. Landolfo - Italy R. Ziemian - USA


Alexandre Landesmann - UFRJ Maximiliano Malite - USP / SC

Arlene Maria Sarmanho Freitas - UFOP Michele Schubert Pfeil - UFRJ
Eduardo de Miranda Batista - UFRJ Pedro Colmar G. da Silva Vellasco - UERJ
Francisco Carlos Rodrigues - UFMG Remo Magalhães de Souza - UFPA
José Guilherme S. da Silva - UERJ Ronaldo Carvalho Battista - UFRJ
Leandro Palermo Junior - UNICAMP Sebastião A. L. de Andrade - PUC-RJ/UERJ
Luciano Rodrigues Ornelas de Lima - UERJ Zacarias M. Chamberlain Pravia - FUPF
Luciano Mendes Bezerra - UNB




How all of it started - some reminiscences about the first international colloquia
in the stability series 3
M. Škaloud

Some issues for column stability criteria 9
Reidar Bjorhovde
The effect of edge support on tensile membrane action of composite slabs in fire 21
Anthony K. Abu and Ian W. Burgess
Latest developments in the GBT analysis of thin-walled steel structures 33
Dinar Camotim, Cilmar Basaglia, Rui Bebiano, Rodrigo Gonçalves and Nuno Silvestre
Dual-steel frames for multistory buildings in seismic areas 59
D. Dubina
Advanced stability analysis of regular stiffened plates and complex plated elements 81
László G. Vigh and László Dunai
Direct strength design of cold-formed sections for shear and combined actions 101
Cao Hung Pham and Gregory J Hancock
Manufacturing specifications for hollow sections in 2010 115
Jeffrey A. Packer
Numerical study on stainless steel beam-columns with transverse loading 123
N. Lopes, P. Vila Real and L. Simões da Silva


Improved cross frame connection details for steel bridges with skewed supports 133
Craig Quadrato, Anthony Battistini, Todd A. Helwig, Karl Frank
and Michael Engelhardt
Flange thirkness transitions of bridge griders-buckling behaviour in global bending 141
A. Lechner, A. Taras and R. Greiner
A visco-elastic sandwich solution for orthotropic decks of steel bridges 149
Ronaldo C. Battista, Emerson F. dos Santos, Raimundo Vasconcelos
and Michèle S. Pfeil


Elegance and economy - a new viaduct over the river Llobregat 157
Peter Tanner, Juan L. Bellod and David Sanz
Design of beam-to-beam butt-plate joints in composite bridges 165
A. Lachal, S.S. Kaing and S. Guezouli

Experimental analysis of composite connections using slab made by precast joist
with lattice and bricks 175
William Bessa, Roberto M. Gonçalves, Carlo A. Castiglioni and Luis Calado
Response of end-plate joints under combined forces 183
N. Baldassino, A. Bignardi and R. Zandonini
Influence of member components on the structural performance of beam-to-column
joints of pitched roof portal frames with class 3 and 4 sections 191
I. Mircea Cristutiu, Dan Dubina
Application of Eurocode 3 to steel connections with four bolts per horizontal row 199
J.-F. Demonceau, K. Weynand, J.-P. Jaspart and C. Müller
A new hybrid testing procedure of the low cycle fatigue behaviour
for structural elements and connections 207
Carlo Andrea Castiglioni, Alberto Drei and Roberto Goncalves
Proposal of a three-dimensional semi-rigid composite joint: tests and finite
element models 215
Beatriz Gil, Rufino Goñi and Eduardo Bayo
Strength and ductility of bolted T-stub macro-components under monotonic
and cyclic loading 223
Nicolae Muntean, Daniel Grecea, Adrian Dogariu and Dan Dubina
Prediction of the cyclic behaviour of moment resistant beam-to-column
joints of composite structural elements 231
Nauzika Kovács, László Dunai and Luís Calado
Numerical modeling of joint ductility in steel and steel-concrete
composite frames 239
Leslaw Kwasniewski and Marian Gizejowski
Influence of chord axial loading on the stiffness and resistance of welded “T”
joints of SHS members 247
Rui M. M. P. de Matos, Luís F. Costa-Neves and Luciano R. O. de Lima
Experimental studies of behaviour of composite beam-column flush end plate
connections subjected seismic loading 255
Olivia Mirza and Brian Uy
Modelling connections of moment resisting steel frames for seismic analysis 263
L. Mota, A. T. da Silva, C. Rebelo, L. Simões da Silva and Luciano R. O. de Lima
Influence of local plastic buckling of joint on carrying capacity of a thin-walled truss 271
H. Pasternak, G. Kubieniec and V. Bachmann


Numerical analysis of endplate beam-to-column joints under bending and axial force 279
Monique C. Rodrigues, Luciano R. O. de Lima, Sebastião A. L. de Andrade,
Pedro C. G. da S. Vellasco and José G. S. da Silva
Loss of preload in bolted connections due to embedding and self loosening 287
Roland Friede and Jörg Lange
Plastic resistance of L-stubs joints subjected to tensile forces 295
M.Couchaux, I.Ryan and M.Hjiaj
Composite beam modelling at real scale including beam-to-beam joint 303
S. Guezouli and A. Lachal
Resistance of laser made t RHS joints under compression load 311
Jerzy K. Szlendak
Cold-formed steel and concrete composite beams: study of beam-to-column
connection and region of hogging bending 319
Mairal R. and Malite M.
Shear bolted connections: numerical model for a ductile component,
the plate-bolt in bearing 327
J. Henriques, L. Ly, J.-P. Jaspart and L. Simões da Silva


Considerations on the design, analysis and performances of eccentrically braced
composite frames under seismic action 337
Hervé Degée, Nicolas Lebrun and André Plumier
Effect of the loading modelling, human heel impact and structural damping
on the dynamic response of composite footbriges 345
José Guilherme S. da Silva, Francisco J. da C. P. Soeiro,
Pedro C. G. da S. Vellasco, Luciano R. O. de Lima and Nelson L. de A. Lima
Structural damage assessment using the differential evolution and the ant colony
optimization techniques 353
Genasil F. dos Santos, José Guilherme S. da Silva and Francisco J. da C. P. Soeiro
Free and forced nonlinear vibrations of steel frames with semi-rigid connections 361
Andréa R.D. Silva, Ricardo A.M. Silveira, Alexandre S. Galvão and
Paulo B. Gonçalves
Static and dynamic behavior of lens-type shear panel dampers for
highway bridge bearing 369
Tatsumasa Takaku, Feng Chen, Takashi Harada, Masayuki Ishiyama,
Nobuhiro Yamazaki, Tetsuhiko Aoki and Yuhshi Fukumoto
Elasto-plastic buckling behavior of H-shaped beam with large depth-thickness
ratio under cyclic loading 377
Tao Wang and Kikuo Ikarashi
Analytical studies of a full-scale steel building shaken to collapse 385
Keh-Chyuan Tsai, Yi-Jer Yu and Yuan-Tao Weng


Robustness of steel and composite buildings under impact loading 393

Ludivine Comeliau, Jean-François Demonceau and Jean-Pierre Jaspart
Design of steel frames of dissipative shear walls 401
C. Neagu, F. Dinu and D. Dubina
Effects of wind on a 3 dimensional steel structure for the central
corridor roof (central spine) at TCS campus at Siruseri, Chennai (India) - a case study 409
T. S. Gururaj and Nagaraja M. Thontalapura
Influence of system uncertainties on structural damage identification
through ambient vibrations of steel structures 417
Leandro Fadel Miguel, Letícia Fadel Miguel, Jorge D. Riera, Marta G. Amani
and Raúl O. Curadelli
Non-linear dynamic analysis of stayed steel columns 423
Ricardo R. de Araujo, José G. S. da Silva, Pedro C. G. da S. Vellasco,
Sebastião A. L. de Andrade, Luciano R. O. de Lima and Luis A. P. Simões da Silva


Interaction diagrams for design of concrete-filled tubular columns under fire 433
Rodrigo B. Caldas, João Batista M. Sousa Jr. and Ricardo H. Fakury
Ductility of simple steel connections in fire 441
J. Buick Davison, Ian W. Burgess, Roger J. Plank, Hongxia Yu and Ying Hu
Methodology for reliability-based design of steel members exposed to fire 449
Shahid Iqbal and Ronald S. Harichandran
Capacity reduction and fire load factors for steel columns exposed to fire 457
Shahid Iqbal and Ronald S. Harichandran
Stability of steel columns subjected to fire 465
Markus Knobloch, Diego Somaini, Jacqueline Pauli and Mario Fontana
Fire behavior of concrete-filled steel circular hollow section columns with
massive steel core 473
Martin Neuenschwander, Markus Knobloch and Mario Fontana
Buckling of concrete filled steel hollow columns in case of fire 481
Tiago A. C. Pires, João P. C. Rodrigues and J. Jéfferson Rêgo Silva
Buckling of steel and composite steel and concrete columns in case of fire 489
Antonio M. Correia and João Paulo C. Rodrigues
A coupled fluid-thermal-mechanical analysis of composite structures
under fire conditions 497
Julio Cesar Gonçalves da Silva and Alexandre Landesmann
Behaviour of composite floor beam with web openings at high temperatures 505
V.Y. Bernice Wong, Ian W. Burgess and Roger J. Plank
To shear resistance of castellated beam exposed to fire 513
F. Wald, A. Pelouchová, J. Chlouba and M. Strejč ek


Numerical analysis of steel columns considering the walls on fire condition 521
Jonas B. Dorr, Jorge M. Neto and Maximiliano Malite
On the strength and DSM design of cold-formed steel columns failing
distortionally under fire conditions 529
Alexandre Landesmann, Dinar Camotim



The collapse load in submarine pipelines under compressive load and internal pressure 539
Luciano M. Bezerra and Ramon S. Y. C. Silva
Stability analysis of 3D frames using a mixed co-rotational formulation 547
Rabe Alsafadie, Mohammed Hjiaj and Jean-Marc Battini
Plastic collapse mechanisms in compressed elliptical hollow sections 555
Aimar Insausti and Leroy Gardner
Collapse of a steel structure as a result of local buckling 563
Heiko Merle and Jörg Lange
Strength and ductility of steel beams with flange holes 571
K.S. Sivakumaran, P. Arasaratnam and M. Tait
Residual stress measurements in roller bent he 100b sections 579
R.C. Spoorenberg, H.H. Snijder and J.C.D. Hoenderkamp
Lateral torsional buckling of space structures with I-beams - structural behavior
and calculation 587
Richard Stroetmann
Analytical derivation of a generalized-slenderness formula for in-plane beam-column
design and comparison with interaction-concept formulae 595
Andreas Taras and Richard Greiner
Load bearing capacity of bracing members with almost centric joints 603
Harald Unterweger
Finite element modeling of angle bracing member behavior in experimentally
tested sub-frame specimens 611
Anna M. Barszcz, Marian A. Gizejowski and Wael A. Salah Khalil
Influence of splices on the stability behaviour of columns and frames 619
Pedro D. Simão, Ana M. Girão Coelho and Frans S. K. Bijlaard
Further results on the application of the extrapolation techniques 627
Tadeh Zirakian
Elastic lateral-distortional buckling of singly symmetric i-beams:
the 2005 AISC specification 635
Tadeh Zirakian and Jian Zhang


System stability design criteria for aluminum structures 641

Ronald D. Ziemian and J. Randolph Kissell
Advanced nonlinear investigations of a 50 m span frame case study:
the steel structure of the ice rink, city of Targus-Mureş, Romania 649
Zsolt Nagy and I. Mircea Cristutiu
Local post-buckling behaviour of elliptical tubes 657
Nuno Silvestre and Leroy Gardner
Plastic design of stainless steel structures 665
Marios Theofanous and Leroy Gardner
Inelastic behaviour of partially restrained steel frames 673
Renata G. L. da Silva and Armando C. C. Lavall
Robust design – alternate load path method as design strategy 681
Lars Roelle and Ulrike Kuhlmann
Use of eigenvalue analysis for different levels of stability design 689
Jozsef Szalai
The non-destructive measurement of residual stresses in stainless steel roll
formed sections 697
Rachel B. Cruise and Anna M. Paradowska
Influence of flange-to-web connection on the patch load resistance of I beams 705
László G. Vigh
Further studies on the lateral-torsional buckling of steel web-tapered beam-columns:
analytical studies 713
Gabriel A Jimenez
Lateral buckling of continuous steel beams with hinges 721
Peter Osterrieder, Stefan Richter and Matthias Friedrich


The interaction behaviour of steel plated structures under transverse loading,
bending moment and shear force 731
Benjamin Braun and Ulrike Kuhlmann
Hybrid steel plate girders subjected to patch loading 739
Rolando Chacon, Enrique Mirambell and Esther Real
Numerical and experimental research in tapered steel plate girders
subjected to shear 747
E. Real, A. Bedynek and E. Mirambell
Shear strength of steel plates with reinforced opening 755
Bo Chen and K.S. Sivakumaran
Crack initiation under static loads including the influence of residiual welding stresses 763
Michael Volz and Helmut Saal


The fatigue and serviceability limit states of the webs of steel girders subjected
to repeated loading 771
M. Škaloud and M. Zörnerová
Imperfections in steel plated structures and their impact on ultimate strength 779
J.Kala, M.Škaloud, J.Melcher and Z.Kala
Imperfection sensitivity analysis of longitudinally stiffened plated girders subjected
to bending-shear interaction 787
F. Sinur and D. Beg

9. SHELLS 795
Influence of geometry on the dynamic buckling and bifurcations of cylindrical shells 797
Zenon N. del Prado and Paulo B. Gonçalves
Buckling of a shallow rectangular bimetallic shell subjected to outer loads
and temperature 805
M. Jakomin and F. Kosel
A geometry based method for the stability analysis of plates 813
Hesham Ahmed, John Durodola and Robert G. Beale
Critical loads and stability of an open elastic-plastic cylindrical shell with the core
of variable stiffness 821
Jerzy Zielnica


Ductile response of composite steel and concrete frames 831
Luigi Di Sarno
Stability and ductility of castellated composite beams subjected
to hogging bending 839
Marian A. Gizejowski and Wael A. Salah Khalil
Shear connection in steel and concrete composite trusses 847
Josef Machacek and Martin Cudejko
Steel and concrete composite building structures – an economical approach 855
Catarina Costa, Luís F. Costa-Neves and Luciano R. O. de Lima
Behavior of steel-concrete composite beams with flexible shear connectors 863
Gilson Queiroz, Francisco C. Rodrigues, Sebastião S. Pereira, Michèle S. Pfeil,
Cláudia G. Oliveira and Luciene A. C. da Mata
Influence of the friction at the support in the longitudinal shear strength
of composite slab 871
A. C. C. Lavall, R. S. Costa and F. C. Rodrigues
A parametric analysis of composite beams with t-perfobond shear connectors 879
Juliana da C. Vianna, Sebastião A. L. de Andrade, Pedro C. G. da S. Vellasco
and Luís F. da C. Neves


FE modelling of slender concrete-filled stainless steel tubular columns

under axial compression 887
Zhong Tao, Brian Uy and Lin-Hai Han
Buckling resistance of steel-concrete columns composed of high-strength materials 895
Marcela Karmazínová and Jindrich J. Melcher


Effects of distortion on the shear stiffness of rack structures 905
Sambasiva R. Sajja, Robert G. Beale and Michael H.R. Godley
Buckling, post-buckling, collapse and design of two-span cold-formed steel beams 913
Cilmar Basaglia and Dinar Camotim
Experimental and numerical investigation of polyurethane sandwich panels 921
I. Vayas, M.-E. Dasiou and X. A. Lignos
Imperfections’ sensitivity analysis of pitched roof cold-formed steel portal frames 929
Dan Dubina, Viorel Ungureanu, Zsolt Nagy, Luis Nunes and Paul Pernes
Ultimate limit strength of perforated cold-formed steels sections 937
Andrei Crisan, Viorel Ungureanu and Dan Dubina
On the use of cold-formed thin walled members for vertical addition
of existing masonry buildings 945
G. Di Lorenzo, A. Formisano, R. Landolfo, F. M. Mazzolani and G. Terracciano
Theoretical analysis of perforated rack columns 953
Arlene M. S. Freitas, Marcílio S. R. Freitas and Flávio T. Souza
Optimization of cold-formed steel channel using the direct strenght method
and finite strip method 961
Gladimir de Campos Grigoletti, Ignacio Iturrioz, Gustavo Mezzomo
and Zacarias Martin Chamberlain Pravia
Experimental investigation of high strength cold-formed supacee sections in shear 969
Cao Hung Pham and Gregory J. Hancock
Effective design of cold-formed thin-walled channel beams with bent edges of flanges 977
Ewa Magnucka-Blandzi and Krzysztof Magnucki
Lateral buckling of steel sigma-cross-section beams with web holes 985
Jindrich J. Melcher and Marcela Karmazínová
Load-bearing capacity of perforated trapezoidal sheeting 993
Thomas Misiek and Helmut Saal
Trapezoidal sheeting made of stainless steel – two amendments
to complete the design codes 1001
Thomas Misiek, Helmut Krüger, Karsten Kathage and Thomas Ummenhofer
Pull-trough resistance of tensile-loaded screw-fastenings of thin-walled sheeting
and sandwich panels 1009
Thomas Misiek, Saskia Käpplein and Karsten Kathage


The cold work of forming effect in steel structural members 1017

Tian Gao and Cristopher D. Moen
Finite element analysis of high strength cold-formed supacee sections in shear 1025
Cao Hung Pham and Gregory J. Hancock
Classification of stability failure modes of sandwich panels under compression loading:
global and local buckling, crippling at support line 1033
Saskia Käpplein and Thomas Ummenhofer
Strength estimation of end failures in corrugated steel shear diaphragms 1043
Nobutaka Shimizu and Kikuo Ikarashi
Plate buckling according to Eurocode 3. Comparison of the effective width method
and the reduced stress method 1051
Jose M. Simon-Talero and Ana Caballero
DSM design of lipped channel columns undergoing local/distortional/
global mode interaction 1061
Nuno Silvestre, Pedro B. Dinis, Dinar Camotim and Eduardo M. Batista
Bracing stiffness and strength in sheathed cold-formed steel stud walls 1069
Luiz C. M. Vieira Jr. and Benjamin W. Schafer
Shear behaviour of trapezoidal sheeting without shear panel construction 1077
J. Lindner and F. Seidel
Thin beam static stability analysis by an improved numerical method 1085
A. Khelil
Investigation of the test method for distortional buckling of compressed
pallet rack members 1093
Miquel Casafont, Magdalena Pastor, Francesc Roure and Teoman Peköz
Behaviour of expanded metal panels under shear loading 1101
Phung Ngoc Dung and André Plumier
Analysis of contact buckling in built-up cold-formed steel beams assembled
by laser welding 1109
F. Portioli, O. Mammana, G. Di Lorenzo and R. Landolfo
Cross-sectional stability of structural steel 1117
Mina S. Seif and Benjamin W. Schafer
Influence of holes on the behaviour of cold-formed steel sections under compression 1125
M.M. Pastor, M. Casafont, F. Roure, J. Bonada and J. Noguera
Applications of pure and combined buckling mode calculation of thin-walled
members using the finite element method 1133
Gustavo P. Mezzomo, Ignacio Iturrioz and Gladimir de C. Grigoletti
Post-buckling behaviour and strength of angle columns 1141
Pedro B. Dinis, Dinar Camotim and Nuno Silvestre
Experimental study on cold-formed steel lipped channel columns undergoing
local-distortional-global interaction 1151
Eliane S. Santos, Eduardo M. Batista and Dinar Camotim



Stability bracing requirements of trusses 1161
Rangsan Wongjeeraphat and Todd A. Helwig
Theoretical and experimental analysis of steel space-truss with stamped connection 1169
Cleirton A. S. Freitas, Luciano M. Bezerra and Ramon S. Y. Silva
Real scale experimental analyses of circular hollow section
multi-planar steel trusses 1177
Rodrigo C. Vieira, João A. V. Requena, Newton de O. P. Junior
and Afonso H. M. de Araújo
Natural period of steel chimneys 1185
Aleksander Kozlowski, Andrzej Wojnar and Leonard Ziemianski
Dynamic response of conical and spherical shell structures subjected to blast pressure 1193
Tomasz Kubiak, Zbigniew Kolakowski, Katarzyna Kowal-Michalska,
Radoslaw Mania and Jacek Swiniarski


Volume 1 537
Volume 2 1201

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010




* Department of Civil Engineering, University of Brasilia, 70910-900 – Brasilia, DF - Brazil

e-mails: lmbz@unb.br, ramon@unb.br

Keywords: Pipeline Collapse Load, Pressurized Pipelines, Petroleum Pipelines.

Abstract. In off-shore plataform, petroleum from the oil well may have to be heated up so that its density
decrease, making easier the pumping of petrolium along pipelines. Due to temperature increase, such
pipelines may be under thermal dilatation and, consequently, under high compressive thermal loading.
There is a great difficult in finding the collapse load of such submarine pipeline. An analytical method is
presented in this paper for the determination of the collapse load of pressurized pipelines extended over
large free spans. The collapse load is determined from a closed solution equation. Results of the
presented formulation are compared with sophisticated finite element analyses. For the determination of
the collapse load of pressurized freespan pipelines under compression, non-linear finite element analysis
requires a lot of computer processing while the present formulation takes practically no time to assess a
good approximation for the collapse load.


Luciano M. Bezerra and Ramon S. Y. C. Silva

Submarine pipelines are often laid on relatively rough sea-bottom terrains and, consequently, may be
supported by soil only intermittently, without intermediate support. Such spans are identified as
“freespans”. The scope of this paper is to predict the behavior of freespan pipelines under compressive
loads originating from effects such as temperature differentials. The paper deals exclusively with free
span pipelines under compressive load combined with internal pressure. The compressive load, P, is
assumed to be applied at the ends of the pipe segment and to be collinear with a line through the end
supports of the pipe segment. Consequently, the load is considered to act along the chord connecting the
two ends of the freespan segment, without change in the load direction. The collapse mechanics of a
segment of a free span pipeline (FSP) under compressive load is not necessarily the same as for a buried
pipeline (BP). Adequate support around a BP may prevent it from buckling globally. Assuming a FSP
under compression deforms as shown in Fig.-1, the collapse mode of a FSP under compression, depends
upon the length of the free span, and will be different than for local wrinkle formation typically observed
in short segments of BPs. For short free span lengths, the collapse mode of the FSP might be similar to
the local wrinkle formation mode observed in BPs. For long free spans, the collapse mode might be
comparable to the global buckling collapse mode observed in a structural column.


Assuming small deformation theory, a long FSP, if ideally straight, elastic, and isotropic, loaded
along the central axis, should behave like any long structural member under compression. The first model
that comes to our mind is the buckling of the Euler’s column. For all practical purposes, the prescribed
Euler’s collapse load in Eq.(1) for a pinned-end column is an upper limit of compressive loading for an
ideal FSP.

Figure-1: Freespan pipeline under compressive load and initial imperfection δ 0.

S 2 EI
PE (1)
To determine a more realistic behavior of FSPs under compressive load it is necessary to admit the
existence of initial imperfections, the possibility of inelastic behavior, and the mobilization of fully plastic
moment capacity of the pipe section. Let us consider the effect of initial imperfection and plastic
deformations, using the mechanical model shown in Fig.-2 [6]. Subsequently, it is possible to examine for
the effects of initial imperfection and inelastic material behavior on the buckling behavior. The
mathematical model consists of two rigid arms pinned together at the span center-line at C. On the ends
(A and B) they are pinned too, as in Fig.-3. A vertical spring, with stiffness K, is attached at C. Applying
an increasing horizontal axial force P at point A through the centroidal cross-sectional axis, with G = 0,
will make P reaches its critical load, P cr.
At this critical load, when buckling takes oplace, the model forms a mechanism in which point C
displaces laterally through a distance G, and the arm rotates D - see Fig.-2b. Prior to instability, the force
in the spring, Ps = 0. As soon as the instability takes place, P s=Kδ - with K being the spring constant.
From moment equilibrium of the arm from A to C, about C, for Fig.-2b, we can write for small angles

Luciano M. Bezerra and Ramon S. Y. C. Silva

§ PS ·§ L · Pcr L
Pcr G ¨ 2 ¸¨ 2 ¸ or (2)
© ¹© ¹ Ps 4G
In Fig.-2a, to simulate the Euler’s buckling load, the lateral defection G will take place abruptly when the
critical load reaches the Euler’s load. Substituting KG for Ps in Eq. (2) and equating Pcr = PE yelds

Pcr PE or K (3)
4 L

Figure-2: (a) An elementary buckling model and Figure-3: The mechanical model with initial
(b) free-body diagram. imperfection.

Up to now, the pre-buckling shape is a straight line, however, by now let’s consider the existence of an
initial imperfection Go z 0 in Fig.-3. Note that small initial imperfections will be amplified by the axial
force. The model of the FSP with an initial imperfection Go is in Fig.-3a. Go exists initially, for P = 0 and
Ps = 0. For P z 0, the incremental displacement at the centerline increases by the amount G due to the
rotation of the arms – see Fig.-3a. The total displacement, due to the arm rotation, E, becomes Gtot = G0 + G
- Fig.-3b. The moment equilibrium of arm A-C, about C, for Fig.-3b may be expressed by Eq.(4).

§ PS ·§ L · § Ps L ·
PG tot ¨ 2 ¸ ¨ 2 ¸ or G tot ¨ 4P ¸ and Gtot = Go + G (4)
© ¹© ¹ © ¹
As PS = KG and δ = δtot – δ0, and using Eq.(3) for K; Eq.(4) can be transformed into the following

§ Ps L · § KLG · § 4PE LG · § PE · § 1  G0 ·
G tot ¨ 4P ¸ ¨ ¸ ¨ 4P ¸ ¨ P ¸ G tot  G o or P ¨ ¸ PE (5)
© ¹ © 4P ¹ © ¹ © ¹ © G tot ¹
The model for the FSP with an initial Go will result in an increase in bending moment at the center
of the span as P and Gtot increase. However, P in Eq. (5) indicates that the compressive load for the model
of the imperfect column (or FSP) will never reach the Euler’s load, PE, but approaches PE asymptotically.
In addition, the maximum bending moment that can arise at the central section cannot exceed that
associated with the fully plastic condition for the pipe. At the central section M = PGtot and the collapse
load for can be determined by the load P that produces the moment which, when combined with the axial
effects, mobilizes the fully plastic capacity of the pipe section ( M Tpc ). To compute the full plastic
capacity of the pipe it is necessary a yield criterion.
Pressurized pipes are subjected to hoop and longitudinal stresses due to axial forces and transverse
bending moments acting on the pipe cross section. For a thin-walled pipe, the hoop stress is considered
constant and stresses other than hoop and longitudinal may be neglected. The longitudinal stress Vl and
the hoop stress VT are identified as the principal stresses V1 and V2, respectively. Using the Von-Mises-

Luciano M. Bezerra and Ramon S. Y. C. Silva

Hencky yield criterion (with Vy as the uniaxial yield strength) the maximum (and minimum) longitudinal
stresses that the fully-plastic pipe cross section can sustain on the cross section may be calculated as
§ Vl · § 2VT · § 3 ·§ V ·
¨ ¸ ¨ ¸ r 1 ¨ ¸¨ T ¸ (6)
¨V y ¸ ¨ Vy ¸ © 4 ¹ ¨© V y ¸
© ¹ © ¹ ¹
Eq.(6) is also valid for the ultimate stress (Vu) in the place of the yielding stress (Vy). Eq.(6) also
identifies the two values of Vl that produce yielding for a specified VT. One corresponds to a compressive
stress (Vl = Vc), and the other, to a tensile stress (Vl = Vt) - Fig 4. These values represent the maximum
longitudinal compressive and maximum longitudinal tensile stresses that can be developed on the extreme
fibers of the pipe cross-section for the given VT. If VT = 0, then, for yielding, Vl=Vc=Vt=Vy. If VT z 0, then,
the longitudinal stress Vl required to origin yield in tension is Vl = Vt which is different than that required
in compression (Vl = Vc) (See Fig.4). Naming [ = VT /Vl and K = Vl /Vy, the Von-Mises-Hencky yield
criterion is shown in Fig.-4 [4, 5]. From Fig.-4 and Eq.(6), the extreme values for Vl and VT. For the
determination of the fully-plastic capacity of the pipe section, we will assume that the stress-strain curve
shows a well defined yield-stress plateau. The yield stress is an important engineering property in order to
establish limits on the longitudinal and hoop stresses. The hoop stress VT is given by

VT (7)
The longitudinal stress acting on the pipe cross-section will depend on the axial force P and the
bending moment. The limiting combinations of axial force and bending moment that develop the fully
plastic capacity of the pipe section can be presented on an interaction diagram due to [2, 3]. In the
following Section, the equations for the fully plastic moment capacity of the FSP pipe section will be


For a pipe, Fig.-5 shows the fully plastic stress distribution, accounting for the effects of stresses Vt
and Vc [2]. As the pipe is under compressive load P applied at the pipe ends, the applied force is
concentrically distributed on the pipe end sections with area Ao giving rise to an equivalent longitudinal
uniform stress V = P/A0 at points A and B of Fig.-3, therefore

ªS R  r 2 º
A0 2
¬ ¼
The stresses on the pipe section at the point of maximum moment are in Fig.-5 which is a fully plastic
condition. At such a point, at the center of the span, we have a combination of stress from bending
moment plus stress from axial loading. However, at the ends of the FSP (see Fig.-1 and 3); the force P
acts in concert with the transverse force of Ps/2, and the combination of these loads must be equilibrated
by the stress distribution of Fig.-5 at the centerline of the span. Therefore, at the point of maximum
moment, the resultant longitudinal force given by the difference between the tensile force F t = σtAt and
compressive force Fc = σcAc, in Fig.-5, must be in equilibrium with the external applied force P at the
ends of the FSP. The areas Ao, At, and Ac in Fig.-5 can be expressed as

§\ · 2 2 ª§ 2S \ · º
S R 2  r 2 , At ¨ ¸ R  r and Ac R  r 2 «¨ ¸»
2 (9)
©2¹ ¬«© 2 ¹ ¼»

Luciano M. Bezerra and Ramon S. Y. C. Silva

From Eqs. (8) and (9) we can write the following longitudinal equilibrium equation

ª 2S \ º 2 §\ ·
P VS R 2  r 2 Fc  Ft V c R2  r 2 « »  V t ( R  r )¨ ¸
¬ 2 ¼ ©2¹
The angle \ can be calculated as a function of the stresses V, Vt, and Vc of Eq.(10).

ª § V  V · º
\ 2S «¨¨ c ¸¸ » (11)
«¬© V c  V t ¹ »¼
A search for P (that causes the stress distribution depicted in Fig.-5) is the same as a search for the
equivalent stress V = P/Ao at the end of the FSP. The arms yt & yc of the respective forces Ft and Fc,
such forces are at the centroids of the areas At and Ac in Fig.-5 - can be calculated as

ª§ § \ · ·º ª§ § \ · ·º
«¨ 4 R  r sin ¨ ¸ ¸ » «¨ 4 R  r sin ¨ ¸ ¸ »
3 3 3 3

yt «¨ © 2 ¹ ¸ » and y «¨ © 2 ¹ ¸»
«¨ 3\ R  r ¸ » «¨ 3 2S \ R  r 2 ¸ »
2 2 c 2
«¨ ¸» «¨ ¸»
¬© ¹¼ ¬© ¹¼
Knowing At, Ac in Eq.(9); yt , yc in Eq.(12); and Vc and Vt in Eq.(6); the maximum plastic resisting
moment MTpc can be determined due to the load P (or stress σ) at the ends of the FSP. MTpc is in
equilibrium with the moment caused by the external force P and the eccentricity Gtot of Fig.-3 and Eq.(4),
M T pc Fc yc  Ft yt V c Ac yc  V t At yt PG tot V Ao G tot (13)

Using Eqs.(9), and (12) in Eq.(13), the expression for the maximum plastic bending moment is

§ 2S \ · 2 4 R3  r 3 §\ ·
M T pc Vc ¨
© 2 ¹
¸ R  r 2
3 2S \ R 2  r 2
sin ¨ ¸ 
§\ · 4 R3  r 3 §\ ·
 V t ¨ ¸ R2  r 2 sin ¨ ¸
© ¹
2 3\ R  r
2 2

Substituting into Eq.(14) the expression for the angle< from Eq.(11), we arrive at the following
simplified version of Eq.(14) which is an expression for the maximum moment capacity for the FSP

§2· ª§ S V c  V · º
M T pc ¨ ¸ V t  V c R  r sin «¨¨ ¸¸ »
3 3 (15)
©3¹ ¬«© V c  V t ¹ ¼»


The limiting fully plastic moment for the FSP as expressed in Eq. (15) is an upper bound on the
moment that can be developed before a plastic collapse buckling mechanism occurs. For this mechanism
to occur we note that MTpc is a function of: (a) the maximum allowable longitudinal stresses, Vt and Vc;
and (b) the equivalent applied stress σ (or load P, since V = P/A0) applied at the ends of the FSP. It is
assumed that a structure with an initial imperfection and under increasing applied compressive load will
deform until its fully plastic moment capacity is developed. The expression for maximum moment

Luciano M. Bezerra and Ramon S. Y. C. Silva

capacity in Eq.(15) shows that for an increase in V, there will be a decrease in MTpc at center span. The
formulation contained, herein, is based upon the argument that, to find the compressive collapse stress V
of a FSP, the effect of out-of-straightness must be taken into account. In reality, every structure has
imperfections in geometry; but long structures like FSP laid on rough terrains, are more susceptible. The
initial imperfection G0 is taken into account in Eq.(5). Such equation represents the behavior of the FSP in
the elastic range until the fully plastic stress distribution of Fig.-5 is developed giving rise to Eq.(13).
Note that Eq.(5), expressed in terms of Euler’s critical stress VE = PE/A0, can give an expression for Gtot as

P ª § G0 · º § PE · ª § Go ·º
V «1  ¨ ¸ » ¨ ¸ or V «1  ¨ ¸ » V E and, G tot G o ª¬1  V V E º¼ (16)
Ao ¬« © G tot ¹ ¼» © Ao ¹ ¬« © G tot ¹ ¼»
Once yielding has fully developed, put MTpc from Eq. (15) into Eq.(13) to get an expression for Gtot as

§ M T pc · ° ª 2 V t  V c R  r º ½
­ ° ª S V c  V º
3 3

V ¨¨ ¸¸ or G tot ®« » ¾ sin « »
© AoG tot ° «¬ 3S R  r V »¼ ¿ ° «¬ V t  V c »¼
2 2
¹ ¯
Finally, by equating the right hand sides of Eq.(16) and Eq.(17), we arrive at the following
transcendental equation for the determination of the collapse stress V, which will be designated as V

° ª 2 V t  V c R  r º ½
­ · ª S V c  V º
3 3
°§ V
« » ¾¨ 1  »  VG o
® ¸ sin « 0
° «¬
3 S R 2
 r 2
»¼ ¿° © E V ¹ ¬« V c  V t ¼»

Developing Eq.(18) into a Taylor series and keeping only two terms of this series, one obtains:

ª R2  r 2 º
V ²  CV  D 0 where C V E  V c  1,5G oV E « 3 3 » and D V EV c (19)
«¬ R  r »¼

Figure 4: Von-Mises-Hencky Yelding Criterion Figure 5: Idealized fully plastic stress distribution

The solution for the collapse stress V in Eq.(19) takes into consideration: (a) the geometric
properties of the pipe section; (b) the initial imperfection for the particular FSP; (c) an upper bound limit
represented by the Euler’s buckling load; (d) the fully-plastic stress distribution, (Fig.-5); (e) the fully-
plastic capacity depends on both the plasticity criterion and the hoop stress, which is a function of the
applied internal pressure; (f) long structures, with initial imperfection, never reach the Euler’s Load
(which is an upper bound limit); (g) the Euler’s load (or stress) which is a function of the modulus of

Luciano M. Bezerra and Ramon S. Y. C. Silva

elasticity, pipe cross-section properties, and pipe length; and (h) the consideration of initial imperfection
that is essential as it triggers the limiting fully-plastic moment capacity mechanism.


For obtaining collapse load for FSPs as a function of L i/D ratios; consider a typical pipeline for
petroleum transportation with a range of free spans L i. The material and cross section properties of the
pipeline are: (a) E= 200000MPa, (b) Vy =448MPa, (c) Vu =531MPa, (d) D = 323.85 mm, (e) t = 19.05
mm, (f) d = 285.75 mm, (g) R = D/2 (161.925 mm), and r =d/2 (142.875 mm), and na internal pressure p
= 10.2Mpa. The range of free span ratios (or Li/D) are shown in Table 1. For each FSP length L i, an initial
imperfection Goi is assumed. In this paper, Goi is taken as the transversal deformation of the FSP such that
the extreme fibers of the pipe cross section are just reaching the onset of yielding. Any other value of Goi
could be arbitrarily used. For simplification, and just to calculate an initial imperfection, it was assumed
that on the onset of yielding Vt=Vc=Vy. Each Li determines different Euler’s load PE and stress VE. The
collapse loads of such FSPs without internal pressure are readily obtained and reported in Table-1. The
analytical solutions are compared to Finite Element Analyses using ABAQUS [1]. It is also noticed that
the analytical results reported in Table-1 consider the ultimate stress Vu in the place of yielding stress Vy
into Eq.(6) - in the ABAQUS runs and results the ultimate stress is reached.

Table-1: Comparison with FEM results

FSP Collapse(kN)
Euler’s Load & Stress
With Internal Pressure
δ0i(mm) Euler’s Euler’s ABAQUS Error(%)
Load Stress FE with w.r.t.
PE(kN) σE(Mpa) Pressure ABAQUS

0 0.00 f f 7871.194 NA NA
4 2.902 2.50E+05 1.37E+04 6694.619 8041.080 -16.74
6 6.529 1.11E+05 6.09E+03 6546.097 7324.820 -10.63
8 11.607 6.25E+04 3.43E+03 6348.670 6828.900 -7.03
12 26.119 2.78E+04 1.52E+03 5844.567 5913.680 -1.17
15 40.815 1.78E+04 9.75E+02 5408.596 NA NA
20 72.577 1.00E+04 5.48E+02 4660.695 4027.00 15.74

This paper presented a mathematical formulation regarding the investigation of compressive
collapse loads of pressurized FSPs. A strategy for obtaining collapse loads as a function of the span
length, initial imperfection, and fully plastic stress capacity has been presented and discussed. Examples
of collapse loads, for pressurized FSPs with a variety of lengths and initial imperfections, were compared
to the sophisticated FE results from the ABAQUS program. The numerical tests show that the proposed
analytical formulation represents a good approximation to freespan solutions. Instead of yield stress, the
analytical solutions were almost coincident with the collapse results generated by ABAQUS FE analyses.
Each complex nonlinear FE run in ABAQUS took approximately 5 hours of CPU on a SUN workstation.
Finally, it is noted that the scope of the present formulation is not to propose a method to substitute
precise FEM modeling and analyses, but to provide an easy, faster and practical way for a first assessment
of compressive collapse loads of pressurized FSPs for the petroleum industry.

Luciano M. Bezerra and Ramon S. Y. C. Silva

Figure 9: Comparison of analytical and FEM results for FSP

[1] ABAQUS. (2000). Standard User’s Manual. Version 6.3. Hibbitt, Karlsson and Sorense, USA.
[2] DOREY, A.B.. (2001). Critical Buckling Strains for Energy Pipelines. PhD Thesis, University of
Alberta, Edmonton, AB, Canada.
[3] MOHAREB, M., D. W. MURRAY. (1999). Mobilization of Fully Plastic Moment Capacity for
Pressurized Pipes. Journal of Offshore Mechanics and Arctic Eng., ASME. vol. 121. p. 237-241.
[4] POPOV, E. (1998), Eng. Mechanics of Solids. Prentice Hall. Englewood Cliffs, New Jersey, USA.
[5] HOFFMAN AND SACHS, (1953), Theory of Plasticity, McGraw-Hill Book Inc. New York, USA.
[6] SHANLEY, F. R. (1957). Strength of Materials. McGraw-Hill Book Company Inc. New York,

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Rabe Alsafadie*, Mohammed Hjiaj* and Jean-Marc Battini**

* Structural Engineering Research Group/LGCGM, INSA de Rennes, 20 avenue des Buttes de Coësmes
35043 Rennes Cedex France
e-mails: Rabe.Alsafadie@insa-rennes.fr, Mohammed.Hjiaj@insa-rennes.fr
** Department of Civil and Architectural Engineering, KTH, Royal Institute of Technology, SE-10044
Stockholm, Sweden
e-mail: Jean-Marc.Battini@sth.kth.se

Keywords: Geometrically nonlinear, 3D beams, corotational formulation, mixed finite element analysis,
arbitrary cross-sections, elasto-plastic material behavior, Hellinger-Reissner functional.

Abstract. The corotational technique is adopted for the analysis of 3D beams. The technique applies to a
two-noded element a coordinate system which continuously translates and rotates with the element. In
this way, the rigid body motion is separated out from the deformational motion. Then, a mixed
formulation is adopted for the derivation of the local element tangent stiffness matrix and nodal forces.
The mixed finite element formulation is based on an incremental form of the two-field Hellinger-Reissner
variational principle to permit elasto-plastic material behavior. The proposed element can be used to
analyze the nonlinear buckling and postbuckling of 3D beams. The mixed formulation solution is
compared against the results obtained from a corotational displacement-based formulation having the
same beam kinematics. The superiority of the mixed formulation is clearly demonstrated.

In recent literature, there have been notable contributions to improve the accuracy and efficiency of
displacement-based finite elements. This approach has the limitation in elasto-plasticity since the
approximations of the axial strains and curvatures are constrained by the element's assumed displacement
fields. Nonetheless, these curvatures can vary in a highly nonlinear fashion along the length of an elasto-
plastic structural member. For example, Izzuddin and Smith [1] found that a large number of
displacement-based beam finite elements are typically required to represent elasto-plasticity behavior
accurately. In the mixed formulation, both internal forces and displacements are interpolated
independently. This formulation addresses the fundamental limitation of conventional displacement-
based elements: the inability of simple displacement polynomials to represent the highly nonlinear
distribution of the curvatures along the member lengths due to general distributed yielding.

The corotational approach has been recently adopted by several authors to handle the geometric
nonlinearity in 3D displacement-based beam models (Alsafadie et al. [2], [3], Battini and Pacoste [4], [5],
Crisfield and Moita [6]). This paper extends the works of Battini on corotational beam elements by
applying the two-field Hellinger-Reissner variational principle for the development of a mixed local
formulation. The corotational approach is employed to handle the geometric nonlinearity, where, in the
corotational frame, the element rigid body motion has been removed and the formulations focus solely on
the element deformational degrees of freedom.

R. Alsafadie et al.


The central idea in the corotational formulation for a two-noded 3D beam is to introduce a local
coordinate system which continuously rotates and translates with the element. Then, local deformational
displacements d l are defined by extracting the rigid body movements from the global displacements d g .
The local displacements are expressed as functions of the global ones, i.e.

d l = d l (d g ) (1)

Then, d l is used to compute the internal force vector fl and tangent stiffness matrix K l in the local
frame. The transformation matrix B between the local and global displacements is defined by

δ dl = Bδ d g (2)

and is obtained by differentiation of (1). The expression of the internal force vector in global coordinates
f g and the tangent stiffness matrix K g in global coordinates can be obtained by equating the internal
virtual work in both the global and local systems, i.e.

f g = BT f l , K g = BT K l B + ∂ (BT fl ) / ∂d g (3)

Relations (1), (2) and transformations (3) are explained in details in [4].


In this section, the internal force vector fl and tangent stiffness matrix K l of a mixed local element
formulation based on the kinematics assumption of the Bernoulli beam theory are derived.
3.1 Kinematics and local displacements interpolations


ry R u
az P

rx G II
rz x

Figure 1: Local beam configuration.

Let x0P ( x, y, z ) denote the position vector of an arbitrary point P in the initial configuration and let
x P ( x, y, z ) denote the position vector of P in the current configuration (see figure 1).

x0P ( x, y, z ) = x G0 ( x) + y ry + z rz
x P ( x, y, z ) = x G ( x) + y a y ( x) + z a z ( x) + α ( x) ω ( y , z ) a x ( x)

where x0G and xG denotes the position vectors of the centroid G in the initial and current onfigurations,
respectively. In the case of thin-walled open cross-sections, the normalized warping displacement is
expressed as the product of the warping parameter α ( x ) and the warping function ω ( y , z ) . To handle in
a convenient way nonsymmetric cross-sections with distinct shear center and centroid, the warping
function ω is defined according to Saint-Venant torsion theory and refers to the centroid G, [7]:

R. Alsafadie et al.

ω ( y, z ) = ω − yc z + zc y (5)

and ω refers to the shear center defined by its coordinates yc , zc . The orthonormal triad a i , i = ( x, y , z )
which specifies the orientation of the current cross-section, is given by

a i = R ri , i = ( x, y , z )
The rotation defined by the matrix R can be considered as the sum of two bending rotations and a
twist rotation, and given by (cf. [8]).

ª1 −v, x − w, xϑx − w, x + v, xϑx º

« »
R = « v, x 1 −ϑx » (7)
« w, x ϑx 1 »
¬ ¼

where v , w and ϑx are the transverse displacements and the twist rotation of the cross-section centroid
relative to the local coordinates system, respectively. Introducing the local rotation matrix defined in (7)
into(4), the displacement vector can be evaluated as

U = u − y (v, x + w, x ϑx ) − z ( w, x − v, x ϑx ) + ωα
V = v − z ϑx (8)
W = w + y ϑx

To obtain the strain vector the following assumptions are adopted: the nonlinear shear strain
components generated by warping are omitted since warping effects are rationally taken into account in a
linearized way only, the warping deformations are proportional to the variation of the torsional angle
(Vlasov assumption), an average value of the axial strain is taken in order to avoid membrane locking and
finally the nonlinear terms in the expressions of the bending curvatures and are neglected. With these
modifications, the following strain expressions are obtained:

1 Io 2
ε xx = ε av − y κ z + z κ y + (r 2 − ) κ x + ω κ x, x
2 A
2 ε xy = (ω, y − z ) κ x (9)
2 ε xz = (ω, y − z ) κ x

with κ x = ϑx , x , κ y = − w, xx − ycϑx , xx , κ z = v, xx − zcϑx , xx

1 ª 1§ I ·º
L ³L «¬
and r 2 = ( z 2 + y 2 ) , ε av = u, x + ¨ v,2x + w,2x + o ϑx2, x ¸ » dx, I o = ³ r 2 dA
2© A ¹¼ A

Since the strain field in (9) is obtained from the local displacement field d l , therefore, all the
components of the strain vector deduced from d l will be designated with a superimposed hat and
combined as ݈ = (εˆxx 2 εˆxy 2 εˆxz ) . Based on the above expression for strain vector idealization, the strain at
any point in the cross-section of the beam element can be related to the cross-sectional generalized strain
vector eˆ = (ε av κ y − κ z κ x2 κ x , x κ x ) as
2 ݈ = A( y, z ) eˆ ( x) (10)
In the present formulation, the axial rotation ϑx is interpolated with shape functions based on the
closed-form solution of the torsional equilibrium equation for an elastic prismatic and geometrically
linear beam. Cubic Hermitian shape functions are chosen for the transverse displacement v and w of the

R. Alsafadie et al.

centroid of the cross-section relative to the local element axes. And finally, linear interpolation is adopted
for the axial elongation u of the local element. Thus, the variation in the cross-section deformation
ê can be written as δ eˆ = N eˆ δ dl . Hence, an infinitesimal change in strain vector can be written as
δ ݈ = A N eˆ δ d l (11)

3.2 Equilibrium and generalized stress interpolation functions

The generalized stress resultants vector S, which is work conjugate to the generalized strains ê, may
be expressed in vector form as S=( N My Mz B ȍ Tsv) where N is conjugate to İav, My and Mz are
conjugate to țy and țz, respectively. The bimoment B, Wagner stress resultant ȍ, and the uniform torque
Tsv are conjugate to țx,x, țx2/2 and țx, respectively. Within each element, the generalized stress resultant
internal force vector is approximated as

S = N S1 fS (12)

where fS =( N MIy MIz BI T ȍ MIIy MIIz BII) the corotational force degrees of freedom of the mixed
formulated element (where I: first node, II: second node and T a constant torque) and NS1 is the force
shape functions matrix satisfying the equilibrium equations.

ª1 0 0 0 0 0 0 0 0º
«w 1 − x / L 0 0 0 −x / L 0 0 0 »»
«v 0 x / L −1 0 0 0 x/ L 0 0»
NS1 = « » (13)
« 0 0 0 0 0 0 0 0 1»
«0 0 0 f BI 0 0 0 f BII 0»
« »
«¬ 0 0 0 f BI, x 1 0 0 f BII, x 0 »¼

sinh [ k ( L − x) ] sinh(k x) GJ
where f BI = − , f BII = and k =
sinh( k L) sinh( k L) E Iω
The resulting element will subsequently be termed as bmw3d element. It should be mentioned that
relation (12) includes P-į effects in the internal moment fields, based on the interpolated transverse
displacements. The variation of the generalized stress resultant internal force vector, may be expressed as

δ S = N S 2 δ dl + NS1 δ fS (14)

3.3 Hellinger-Reissner potential for beams

In the Hellinger-Reissner mixed formulation, both the displacement and the internal forces are
approximated by independent shape functions. This principle is applied to a beam element of length L loaded
by end forces only. This two-field variational principle yields two sets of nonlinear equations

EQ = ³ NTeˆ S dx + ³ NTS 2 (eˆ − e) dx − Flext = 0 (15)


EC = ³ NTS1 (eˆ − e) dx = 0 (16)


where EQ and EC are the element equilibrium and element strain-displacement compatibility equations,
respectively. A third equation, the cross-section equilibrium, may be expressed as

SQ = S Σ − S = 0 (17)

R. Alsafadie et al.

where SȈ is given by the nonlinear cross-section constitutive relation and represents a general function
that permits the computation of cross-section stress resultants for given cross-section deformations. The
linearization of the cross-section constitutive relation SȈ= SȈ(e) is obtained using the cross-section tangent
stiffness matrix k = ∂S Σ / ∂e . The cross-section tangent exibility matrix q is obtained by inverting the
cross-section tangent stiffness matrix: q=k-1. Furthermore, S is the interpolated generalized stresses acting
over a cross-section and defined by (12).
3.4 Linearization of the Hellinger-Reissner functional
The nonlinear system of equations EQ=0, EC=0 and SQ=0 may be solved using various combinations
of Newton iteration at the element, and cross-section levels. Since interelement compatibility is not
enforced for the generalized stress variables interpolation, the nonlinear discretized strain-displacement
compatibility equation EC=0 can be solved iteratively at the element level for every global equilibrium
iteration. Similarly, the nonlinear constitutive equation SQ=0 can be solved iteratively at the cross-section
level for every element level iteration. In the following Subsections, the consistent linearization of the
above nonlinear equations is presented. In the process of consistent linearization, it is important to
recognize the arguments of any given function.
3.4.1 Linearization of the Cross-section Constitutive Equation
By expanding SQ=0 about the current cross-section state while holding S constant, we can write

∂ SQ
S Qj +1 ≈ S Qj + Δe j Ÿ Δe j = −qS Qj = q (S − S Σj ) (18)

3.4.2 Linearization of the Element Compatibility Equation

The incremental form of the element compatibility condition EC=0 may also be derived by taking a
Taylor series expansion of EC=0 about the current state variables fS and dl
∂E ∂E
ECi +1 ≈ ECi + C Δd li + C ΔfSi = 0 (19)
∂d l ∂fS
Then, solving for ǻfiS, we obtain
ΔfSi = H11 ª¬(M κ + G1 − H12 ) Δd il + ECi º¼ (20)

3.4.3 Linearization of the Element Equilibrium Equation

Although the linearization of the discrete weak-form of the element equilibrium equation EQ=0
follows standard procedure [35, 36], this linearization is complicated by the presence of displacement-
dependent nonlinear interpolation functions for the generalized stresses. For the case at hand, the
consistent local tangent stiffness equations are obtained by expanding (15) for each of the state variables
dl, fS and Flext about the current state. Again, a Taylor series expansion of EQ=0 is written as follows

EQn +1 ≈ EQn + Δd ln + ΔfSn − ΔFlext , n = 0 (21)
∂d l ∂fS

This equation can be rewritten by substituting ΔfSn from (20), and solving for Δd ln . Hence,

K l Δd ln = Flext , n +1 − fl (22)

where K l is the local consistent tangent stiffness matrix given by

K l = K g + G T2 + G 2 − H 22 + (M κ + G1 − H12 )T H11 (M κ + G1 − H12 ) (23)

and fl is the local internal force vector

R. Alsafadie et al.

fl = G1T fSn + ³ NTS 2 (eˆ n − e n ) dx + (Mκ + G1 − H12 )T H11

−1 n
EC (24)

3.5 Nonlinear state determination algorithm

The nonlinear system of equations is iteratively solved using Newton's method using three imbricated
loops at different levels (structural level, element level and cross-section level). The local element strain-
displacement compatibility equation EC=0 is solved iteratively for every global equilibrium iteration.
Similarly, the local cross-section constitutive equation SQ=0 is solved iteratively at the cross-section
level for every element compatibility level iteration. Therefore, residuals at the cross-section equilibrium
and element compatibility levels are eliminated through iterations at each of these levels. In this
subsection, the superscripts n, i and j denote the iteration indices for the global structural equilibrium, the
element compatibility and the cross-section equilibrium levels, respectively. Before launching the
computer program (initialization), the local displacement vector for each finite element and the
generalized strains e at each Gauss point along the element length, need to be stored as zero vectors. Once
the local displacements are obtained, the state determination procedure, established to obtain the internal
force vector and tangent stiffness matrix in the local frame, starts as follows:
1. Evaluate the generalized strains ê compatible with the interpolated displacements d l
eˆ n = eˆ(d ln )
2. Evaluate the nodal force degrees of freedom fS . (iterate on ΔfSi )
EC , fSi +1 =: fSi + ΔfSi where EiC = ³ NTS1 (eˆ n − ei ) dx
−1 i
ΔfSi = H11

3. Evaluate the generalized stress resultant internal force vector S

S i +1 (fSi +1 ) = N S1 fSi +1
4. Cross-section equilibrium level : Consider for the first iteration at this level that e j =: ei , then,
evaluate the generalized strains e derived from the interpolated stress-resultant force fields
Δe j = q (S i +1 (fSi +1 ) − S Σj (e j )), e j +1 =: e j + Δe j where S Σj (e j ) = ³ A ı ( A e j ) dA

5. Repeat the above step until S i +1 − S Σj +1 ≤ tolerance , then consider ei +1 =: e j +1 upon convergence.
6. Repeat the above steps from 2 to 5 until EiC+1 ≤ tolerance , then consider fSn =: fSi +1 , e n =: ei +1 and
ECn =: ECi +1 upon convergence.
7. Calculate local element forces fl and tangent stiffness matrix K l


4.1 Cantilever with channel-section

Figure 2 contains the problem description. This example was first introduced by Gruttmann et al. [7]. The
beam is modeled using 4 bmw3d elements and the results are compared against 4 and 20 pbw3d
displacement-based beam elements. These meshes used 2 Gauss points per element length and 80
integration points within the cross-section. In Figure 2, the load versus the vertical displacement v of
point O at the cantilever tip is depicted, where the nonlinear response has been computed up to v = 200.
The results obtained with 4 bmw3d and 20 pbw3d elements are in very good agreement with those
presented by [7] based on shell elements. Furthermore, the results obtained with 4 pbw3d elements do not
agree well over a large extend of the computed load deflection curve. It can be observed that, in elasto-
plasticity, the number of mixed-based beam elements used to discretize the structure is considerably
reduced compared to the number of displacement-based beam elements needed to obtain the load-
displacement curve with the same accuracy. This problem demonstrates the capability of the mixed
formulation to satisfactorily predict the nonlinear behavior of beams with nonsymmetric cross-sections.

R. Alsafadie et al.

Figure 2: Cantilever with channel-section: data and results.

4.2 Right-angle frame

The right-angle frame, shown in Figure 3 is subjected to a concentrated out-of-plane load P acting at the
middle of the span of the horizontal member. In the first model, each member is modeled using 4 bmw3d
elements with 3 Gauss integration points along the element length. The square beam cross-section is meshed into
a grid of 64 integration points. Four and 20 pbw3d elements per member are also used for the second and third
models, respectively, with the same elemnt and cross-section discretization. Nonlinear analysis is also performed
with FineLg [9], using 20 corotational two-noded spatial beam elements . The load versus the out-of-plane
displacement of point O curves are depicted in Figure 3 for all models. The comparison between the models
shows a very good agreement between the results obtained with the mixed model and those obtained with
FineLg and with 20 pbw3d displacement-based elements per member.

Figure 3: Right-angle frame: data and results.

This paper proposed an efficient local mixed finite element formulation for the analysis of 3D
Bernoulli beams with small strains and large displacements and rotations. The corotational technique
proposed in [4] is employed here. The local strains are derived based on a consistent second-order
linearization of the fully geometrically nonlinear Bernoulli beam theory. A 3D, geometric-nonlinear,
elasto-plastic local beam element based on the incremental form of the two-field Hellinger-Reissner

R. Alsafadie et al.

functional has been presented. This element is targeted particularly for the analysis of thin-walled beams
with generic open cross-section where the centroid and shear center of the cross-section are not
necessarily coincident. Several numerical examples have demonstrated the superiority of the mixed
formulation over displacement-based one: the use of mixed formulation leads to a considerable reduction
in the number of elements needed to perform the analysis with the same accuracy.

[1] Izzuddin B.A., Smith D.L. “Large-displacement analysis of elasto-plastic thin-walled frames. I:
Formulation and implementation”. Journal of Structural Engineering (ASCE), 122(8), 905-914,
[2] Alsafadie R., Battini J.-M., Somja H., Hjiaj M. “Local formulation for elastoplastic corotational
thin-walled beams based on higher-order curvature terms”. Finite Elements in Analysis and
Design, submitted.
[3] Alsafadie R., Battini J.-M., Hjiaj M. “Efficient local formulation for elasto-plastic corotational
thin-walled beams”. Communications in Numerical Methods in Engineering, in press.
[4] Battini J.-M., Pacoste C. “Co-rotational beam elements with warping effects in instability
problems”. Computer Methods in Applied Mechanics and Engineering, 191(17), 1755-1789,
[5] Battini J.-M., Pacoste C. “Plastic instability of beam structures using co-rotational elements”.
Computer Methods in Applied Mechanics and Engineering, 191(51), 5811-5831, 2002.
[6] Crisfield M.A., Moita G.F. “A unified corotational framework for solids, shells and beams”.
International journal of Solids and Structures, 33(20-22), 2969-2992, 1996.
[7] Gruttmann F., Sauer R., Wagner W. “Theory and numerics of three-dimensional beams with
elastoplastic material behavior”. International Journal for Numerical Methods in Engineering,
48(12), 1675-1702, 2000.
[8] Van Erp G.M., Menken C.M., Veldpaus F.E. “The nonlinear flexural-torsional behavior of
straight slender elastic beams with arbitrary cross-sections”. Thin-Walled Structures, 6(5), 385-
404, 1988.
[9] FineLg User's Manual. V9.0. Greisch Info S.A. - Department ArGEnCo - Liege University
(ULg), 2005.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Aimar Insausti* and Leroy Gardner*

* Department of Civil and Environmental Engineering, Imperial College London, London SW7 2AZ
e-mails: a.insausti@imperial.ac.uk, leroy.gardner@imperial.ac.uk

Keywords: Compression; Elliptical hollow sections; Local buckling; Plastic mechanism; Steel structures.

Abstract. The plastic collapse response of structural steel elliptical hollow section (EHS) profiles in
compression is examined in this paper. As an initial step, a parametric study to identify the factors that
determine which plastic mechanisms would arise has been carried out using finite element (FE) results
from the current work and experimental data from the literature. Following this, an analytical model to
describe the “split flip disc” plastic collapse mechanism in compressed EHS is derived. The parameters
controlling the shape and size of the plastic hinges have been investigated and found to be of key
importance; hence, special care has been taken in their definition. Finally, the analytically derived load–
displacement curves have been compared with FE results. The comparisons have revealed good
agreement, confirming the ability of the developed analytical models to predict the plastic collapse
response of elliptical tubes.

Hot-finished structural steel elliptical hollow sections (EHS) have recently been introduced to the
construction sector. These structural elements can offer greater structural efficiency than circular hollow
sections (CHS) when subjected to bending or combined loading, or when used as columns with
intermediate restraint about the weaker axis, since they posses different major and minor axis flexural
properties. Despite recent investigations involving the testing, numerical modelling and development of
design rules for EHS, a number of aspects of their structural response remain unexplored. In particular,
the behaviour of elliptical profiles in the post ultimate region has not yet been examined.
The aim of the present work is to develop an analytical model to predict the load–deformation
response of EHS under pure compression. To this end, rigid-plastic theory has been applied to EHS based
on the method presented by Murray for plates [1]. As an initial step, a parametric study was carried out to
identify the most common local plastic collapse mechanisms arising in EHS. The study involved finite
element (FE) modelling and the analysis of existing test data [2]. Whilst a number of failure modes were
identified, an inward plastic collapse mechanism of the form shown in Fig. 1 was the most prominent. An
analytical description of this collapse mechanism is therefore the focus of the present study. The key
parameters controlling the size and shape of the hinge lines in the plastic mechanism were carefully
examined and their influence on the overall load–deformation response was assessed. Finally,
comparisons between the analytical model and the results obtained from the FE models are presented.

The intermediate response of elliptical tubes between that of flat plates and circular shells has been
previously identified in terms of elastic buckling [3]. In anticipation of an analogous scenario for plastic
collapse, previous studies on rigid-plastic failure mechanisms in flat plates and circular shells are initially
reviewed. For rigid-plastic mechanisms in flat plates, pioneering work was carried out by Murray [1],
who introduced a number of different failure modes and derived corresponding load–deformation curves.
Among the common plate-like patterns identified, the so called “flip disc” mechanism was presented in

Aimar Insausti and Leroy Gardner

detail. This mechanism is formed by two curved hinge lines, one of them folding outwards and the other
inwards, creating a disc shape plate within the hinge lines that flips around the horizontal mid axis.
Research on compressed CHS has identified two main local plastic failure modes - the “elephant
foot” and “Yoshimura” mechanisms. The elephant foot is an axisymmetrical mode with outwards
deformation that creates a concertina shape in the surface, while the Yoshimura pattern includes several
sequential folding lines that deform the cross-section in a non axisymetrical way. The boundaries that
define the occurrence of each plastic failure mode were studied experimentally by Andrews et al. [4].
Both failure modes as well as a mixed mode were identified in the experimental work and a classification
chart to predict their occurrence was developed. Later, the available experimental data on compressed
CHS was expanded by Guillow et al. [5] carrying out more tests over a wider cross-section slenderness
range. As a consequence, a revised classification chart was developed. For the elephant foot mode, load-
displacement curves were derived by Grzebieta [6] using the plastic mechanism approach. Further
experimental results on circular tubes specifically proportioned to develop axisymmetric failure modes
were presented by Gupta and Velmurugan [7] while Johnson et al. [8] studied the Yoshimura type
mechanism by means of tests on circular PVC tubes.
The key difference between the elements previously studied (i.e. flat plates and CHS) and elliptical
hollow sections lies in the continuously varying curvature brought about by the following geometrical
definition where the symbols are defined in Fig. 1.
z 2 y2
 1 (1)
a 2 b2
With the recent introduction of hot-finished EHS into the construction sector, heightened interest in
the structural behaviour of elliptical profiles, as well as the need to develop design guidance, have
emerged. Structural performance data have been generated on elliptical sections in compression [2, 9]
and bending [10]. The result of the 25 compression tests (stub column tests) given in [2], together with
numerically generated results, have been used in the present study for the development and validation of
the analytical model. Based on the results of compression and bending tests [2, 10] and an analysis of the
elastic buckling of EHS, a cross-section slenderness parameter was derived for the purpose of cross-
section classification [11]. The slenderness parameter was based on an equivalent diameter De, which
allowed the classification of EHS to be made on the basis of the CHS slenderness limits. For the pure
compression case, De = 2a2/b, which corresponds to the point in the section with maximum radius of
curvature, 2a and 2b being the larger and smaller outer dimensions respectively of the EHS, as shown in
Fig. 1. This point of the section was identified by Kempner [12] as suitable for use with the classical
formula for CHS in determining elastic buckling stresses for EHS. Later, more precise expressions for
determining the equivalent diameter were proposed [3, 13], and an alternative approach to EHS
classification, based on an equivalent rectangular hollow section has also been investigated [9].


In this section, the use of FE analysis to examine the behaviour of stocky EHS under compression is
described. These models have already been validated against a total of 25 compression tests in [2]. As
well as being used to generate a series of load–deflection curves that will be used to validate the
analytical model presented later, the FE models were used to identify the different plastic collapse
mechanisms arising for different geometries, making it possible to focus the research on the more
common plastic collapse modes. The FE models also provided useful information about the size and
shape of the mechanism throughout the deformation process.
All models were developed using the nonlinear FE software ABAQUS. The elements designated as
S4R in ABAQUS were employed throughout the modelling. The cross-sectional dimensions of the
modelled elliptical sections was kept constant at 150×75 mm, while the thicknesses used were 4, 5, 6.3
and 8 mm, covering a range of cross-section slenderness values. The above thickness values are also
representative of commercially available profiles and consistent with those previously studied

Aimar Insausti and Leroy Gardner

experimentally. The member length was fixed at 300 mm, which was sufficiently short to ensure no
global buckling, and all sections were subjected to concentric compression. All models were assigned
rigid-plastic material properties without strain-hardening or residual stresses to allow direct comparison
with the analytical model developed herein.
Fixed boundary conditions were applied to the ends of the columns, with all degrees of freedom,
other than vertical displacement at the loaded end, restrained. Initial geometric imperfections were
introduced in the shape of elastic buckling modes obtained from eigenvalue analyses. Three modes were
considered, in which the number of half sine waves along the stub column length was either odd or even,
the imperfection was either positive or negative (positive being inwards at the mid-height of the stub
column) and the imperfection was either symmetrical or asymmetrical about the mid-height. Three
imperfection amplitudes were also considered: t, t/10, t/100, where t is the section thickness.


According to plastic theory, the number of possible plastic mechanisms in a thin-walled structure is
unlimited; however, some plastic failure patterns are more commonly repeated when the element is
loaded in a specific way. The present section discusses the trend of the elliptical profiles to follow
specific failure mechanisms with reference to the FE and test results. Hence, the more common local
plastic failure modes in EHS can be identified and focussed upon in the analytical study.

Figure 1. (a) Split flip disc failure pattern observed in tests and (b) illustration of the mechanism.
Variation of the initial geometrical imperfections described in the previous section trigger four
different plastic failure patterns, two of them akin to plate-like behaviour and the other two akin to shell-
like (CHS) behaviour. Within the plate failure modes, one of them is similar to the flip disc mechanism
proposed by Murray [1] for flat plates, while the other is a variation of it. This variation includes an extra
straight hinge line in the middle of the mechanism splitting the disc into two half parts as shown in Fig. 1.
This plastic mechanism is referred to herein as the split flip disc (SFD) mechanism. The shell-like plastic

Aimar Insausti and Leroy Gardner

collapse mechanisms observed in the EHS are the elephant foot and the Yoshimura patterns. All four
mechanisms are shown in Fig. 2.
The specific failure mode that a given EHS profile would succumb to was found to be influenced by
the shape and amplitude of the initial geometric imperfection, as well as the slenderness (De/t, where
De = 2a2/b) of the cross-section. The FE models showed that plate-like failure modes dominate the
profile’s behaviour for smaller (and more practical) levels of initial imperfection, and that, overall, the
split flip disc mechanism was the most common.
The experimental work presented in [2] has also been used to identify the plastic failure mechanisms
in compressed EHS. The tests exhibited three of the four modes revealed in the numerical study: the flip
disc, the split flip disc and the elephant foot modes. However, the elephant foot was present only in four
out of 25 tests. The flip disc and the split flip disc modes appeared in the remaining 21 tests in no clear
pattern; overall, the split flip disc mechanism appeared more frequently. Hence, the most commonly
arising mode in both the experimental and numerical studies was the split flip disc mode; consequently,
development of an analytical description of this plastic failure mechanism is the focus of this paper.

(a) Flip disc (b) Split flip disc (c) Elephant foot (d) Yoshimura
Figure 2. Plastic collapse mechanisms obtained from FE models for EHS.

In this section, an analytical model for the split flip disc failure mode arising in compressed EHS is
developed based on the plastic theory for thin-walled structures presented by Murray [1]. As stated by
Murray, assuming that the material stress-strain curve is a step function with a step height between tensile
and compressive yielding of 2fy, where fy is the material yield strength, a cross-section’s load-carrying
capacity can be derived as a function of the displacement from equilibrium, based on an assumed plastic
collapse mechanism. Rigid-plastic material behaviour is therefore used, which neglects strain hardening,
and assumes that all deformation is localised along the hinge lines with no deformation elsewhere.
The moment capacity of a plastic hinge, based on a rectangular element of width b and thickness t is:
fy b t2
M pl (2)
And the reduced plastic moment M'pl in the presence of an axial load N may be shown to be:
§ N2 ·
Mcpl M pl ¨1  2 ¸ (3)
¨ Ny ¸
© ¹
where N is the axial load and Ny = fybt is the yield load in the element. Eq. 3 applies when the hinge
line is perpendicular to the direction of the thrust. However, for an inclined hinge line, the reduced plastic
moment is given by Eq. 4, where ȕ is the angle between the line perpendicular to the thrust and the
studied inclined hinge line.
Mcplc Mcpl sec2 E (4)

Aimar Insausti and Leroy Gardner

In the present work, both parabolic and elliptical functions were considered for the description of the
curved hinge lines in the plastic mechanism. However, the parabolic hinge lines were found to more
accurately replicate the actual load-displacement response of EHS, thus, this shape is used throughout the
present derivation. A parabola can be defined using two parameters that fix the height Xh and the width
Sh of the curve on the surface of the EHS. Hence, the values of Xh and Sh define the mechanism along the
development of the plastic hinge (Eq. 5). Fig. 1 summarises the notation used in the present section to
define the modelled failure mechanism, as well as the angle ȕ used in Eq. 4.
§ s2 ·
x (s) X h ¨¨1  2 ¸¸ (5)
© Sh ¹
With reference to Fig. 1, the load transmitted by the complete cross-section can be obtained in terms
of the load inside the plastic hinges (Nin) plus the load outside the plastic hinges (Nout). Furthermore, by
using symmetry, only one quarter of the cross-section needs to be analysed with the result for the full
cross-section being factored accordingly. The contribution to the load-carrying capacity of the cross-
section from within the hinge lines and outside hinge lines are derived in the following two sub-sections.
5.1 Load-carrying contribution within the plastic hinges
Since the inclination of the hinge lines is variable around the cross-section, a differential strip of
material is analysed, as depicted in Fig. 1, with the following reduced plastic moment:
f y t 2 §¨ § dN · ·¸
M cplc ¨ ¸ 2
¨1  ¸ sec E dS
4 ¨ ¨ f y t ds ¸ ¸
© © ¹ ¹
in which dN is the load in the strip and dS is the strip width. Fig. 3(a) presents the free body diagram
of the material strip, that relates the lateral displacement of the strip ǻds to the applied load dN and the
reduced plastic moment. The bending moment diagram in the material strip is presented in Fig. 3(b). The
load borne by the strip dN can be related to the lateral displacement by considering equilibrium at a null
bending moment point. For the split flip disc mechanism, the point of zero bending moment lies between
the curved hinge line and the straight hinge line that splits the flip disc into two equal parts. Denoting ǻ'
the straight distance between the null bending moment point and the undeformed position (see Fig. 3(a)),
this distance can be related to ǻdS through the magnitudes of M'pl and M''pl, and consequently as a
function of the plastic hinge inclination ȕ:
M cplc sec 2 E
'c ' dS ' dS (7)
M cpl  M cplc 1  sec 2 E

Figure 3. (a) Free body diagram and (b) bending moment diagram of the material strip.
Having obtained ǻ', the load borne by the strip dN may be related to the reduced plastic moment:
dN 'c M cplc M cpl sec 2 E (8)
Merging Eq. 7 and Eq. 8, introducing the reduced plastic moment in the material strip (Eq. 6) and
rearranging, we obtain.

Aimar Insausti and Leroy Gardner

§ 2 ·
¨ § 2 ' dS · 2 ' dS ¸
dN f y t ¨ ¨¨ 2
¸  1  2 ¸dS (9)
¨ © (1  sec E ) t ¹ (1  sec E ) t ¸
© ¹
The lateral displacement in the strip ǻdS can be related to the maximum lateral displacement in the
hinge ǻ through Eq. 10, where x(s) defines the parabolic shape of the curved hinge line (Eq. 5).
x (s ) § s2 ·
' dS ' ' ¨¨1  2 ¸¸ (10)
Xh © Sh ¹
From Eq. 9 and 10, and following some manipulation, we obtain the load carried by each strip as:
§ 2 ·
¨ § S2h (S2h  s 2 ) ' · S2h (S2h  s 2 ) ' ¸
dN f y t ¨ ¨ 4
2 2 ¸  1  (S4  2 X 2 s 2 ) t ¸dS
¸ (11)
¨ © (Sh  2 X h s ) t ¹ h h ¸
© ¹
Eq. 11 can not be integrated explicitly; hence, in the present work Simpson’s rule has been employed
to obtain the load–lateral displacement curve, as advised in [1]. The load in the strip was evaluated at
s = 0, s = Sh/2 and s = Sh, leading to the following relationship between load within the plastic hinges Nin
(for one quarter of the section) and lateral displacement ǻ for the split flip disc mechanism:
f y t Sh § ' 6S2h ' '2 9S4h '2 ·
N in ¨1     1  4 1¸ (12)
¨ 2 2 2 2 2 2 2 ¸
6 t ( 2Sh  X h ) t t 4( 2Sh  X h ) t
© ¹
5.2 Load-carrying contribution outside the plastic hinges
The area of the cross-section outside the plastic hinges is considered to be working at the yield stress.
Hence, the load-carrying contribution from outside the plastic hinges is proportional to the arc length
outside the hinge Sout, and is given for one quarter of the section as:
N out f y t Sout (13)
The total load carried by the full cross-section is obtained from Eq. 12 and Eq. 13 as:
N ( ' ) 4 N out  N in (14)

5.3 Governing parameters

Since the load-carrying capacity of the section depends on the shape of the plastic mechanism, Xh
and Sh must be defined as a function of ǻ in order to determine the final load–lateral displacement curve.
At this point, information obtained from the FE models and the tests has been used to monitor the value
of both parameters throughout the deformation process. The height of the parabola Xh has been observed
to remain almost constant during deformation; its value may therefore be defined simply as a function of
the cross-section dimensions. However, the width of the parabola Sh, has been seen to increase as lateral
displacement increases. Furthermore, the FE models showed that Sh is not directly proportional to the
lateral displacement, but grows more rapidly at the beginning of the deformation process and tends
towards a final value. Hence, both Xh and Sh have been defined by means of the cross-section dimensions
in the present work, Xh being constant throughout the deformation and Sh being a function of ǻ.
As stated in [3], the longitudinal wavelength for elastic buckling of CHS is a function of the radius r
and the thickness t. Clearly the elastic buckling wavelength is influential in the definition of the size of
the plastic collapse mechanism, particularly at the early stages of the deformation process. Hence, the
measured values of Xh from the FE models and the experiments have been plotted against (Det)0.5 in Fig.
4. The data may be seen to follow an approximately linear trend, and hence Eq. 15 was obtained by least
squares regression, ensuring that the line passes through the origin, and used in the analytical model.
Xh 1.22 De t (15)

Aimar Insausti and Leroy Gardner


Xh (mm)
60 ABAQUS data
40 Tests data
20 Eq. 15
0 20 40 60 80 100 120
(Det)0.5 (mm)
Figure 4. Xh values from the FE models and the tests.
The FE models showed that the Sh parameter increased more rapidly at the beginning of the plastic
deformation than when the mechanism was fully developed. Hence, it was chosen to represent Sh with a
rational expression of the form given by Eq. 16 that tends to Sh,a, and where C is a constant.
Sh Sh,a (16)
Based on observations of the test failure patterns, it was found that the maximum extent of the plastic
mechanism was approximately 75% of the way around the quarter perimeter of the section – i.e.
Sh,a = 0.75P/4 = 3P/16, where P is the perimeter of the ellipse. Furthermore, from the FE results, it was
found that C = t provided a good approximation of the progression of Sh towards its asymptotic value.
Hence, Eq. 17 was established:
3P § ' ·
Sh ¨ ¸ (17)
16 © '  t ¹


The analytical equations developed throughout Section 5 are validated in this section by reference to
the results of the FE study. Load-lateral displacement curves for two cases are shown in Fig. 5 – EHS
150×75×4 and EHS 150×75×8, representing the extremes of slenderness in currently available elliptical
steel profiles. An initial geometric amplitude of t/100 was used in the FE models, since this had being
found to provide the best agreement with test results [2]. The comparisons reveal good agreement
between the results of the analytical model and of the FE simulation. For the sections investigated, a
maximum deviation between the two load-lateral displacement curves of 6.7% was observed.

1.0 Analytical model 1.0 Analytical model

0.8 FE model 0.8 FE model

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0 10 20 30 40 0 10 20 30 40
ǻ (mm) ǻ (mm)
(a) t = 4 mm (b) t = 8 mm
Figure 5. Load-lateral displacement comparisons for 150×75 EHS with (a) t = 4 mm and (b) t = 8 mm.

Aimar Insausti and Leroy Gardner

Local plastic collapse mechanisms in compressed EHS have been examined in this study. Four
collapse mechanisms were identified, two of which were akin to plate-like behaviour and two to shell-
like behaviour. A numerical study, coupled with examination of existing test data, revealed that the so-
called split flip disc mechanism arose most frequently. Hence an analytical model to describe the load–
lateral displacement response of EHS under pure compression following this failure pattern was derived.
Simple expressions, in terms of section geometry, to determine the key parameters required to fully
describe the shape of the collapse mechanism were developed. Comparisons between the analytical
model and FE model revealed good agreement over a range of cross-section slenderness, with a
maximum discrepancy of 6.7%. It is concluded that the derived analytical model provides an accurate
means of predicting the load-lateral displacement response of a compressed EHS undergoing local plastic
collapse in the split flip disc mechanism.

The authors would like to acknowledge the Basque Government (Department of Education,
Universities and Research) for the financial support given under the overseas post-doctoral development
scheme in 2009 and 2010.

[1] Murray NW. Introduction to the theory of thin-walled structures. Oxford University Press 1984.
ISBN 0-19-856186-5.
[2] Chan TM and Gardner L. Compressive resistance of hot-rolled elliptical hollow sections.
Engineering Structures 2008;30(2),522–532.
[3] Ruiz-Teran A and Gardner L. Elastic buckling of elliptical tubes. Thin-Walled Structures
[4] Andrews KRF, England GL and Ghani E. Classification of the axial collapse of cylindrical tubes
under quasi-static loading. International Journal of Mechanics and Science 1983;25(9):687-696.
[5] Guillow SR, Lu G and Grzebieta RH. Quasi-static axial compression of thin-walled circular
aluminium tubes. International Journal of Mechanical Sciences 2001;43(9):2103–2123.
[6] Grzebieta RH. An alternative method for determining the behaviour of round stocky tubes
subjected to an axial crush load. Thin-Walled Structures 1990;9(1-4):61-89.
[7] Gupta NK and Velmurugan R. An analysis of axi-symmetric axial collapse of round tubes. Thin-
Walled Structures 1995;22(4):261-274.
[8] Johnson W, Soden PD and Al-Hassani STS. Inextensional collapse of thin-walled tubes under axial
compression. Journal of Strain Analysis 1977;12(4):317-330.
[9] Zhao XL and Packer JA 2009. Tests and design of concrete-filled elliptical hollow section stub
columns. Thin-Walled Structures, 47(6-7): 617-628.
[10] Chan TM and Gardner L. Bending strength of hot-rolled elliptical hollow sections. Journal of
Constructional Steel Research 2008;64(9):971-986.
[11] Gardner L and Chan TM. Cross-section classification of elliptical hollow sections. Steel and
Composites Structures 2007;7(3):185-200.
[12] Kemper J. Some results on buckling and postbuckling of cylindrical shells. Collected papers on
instability of shell structures. NASA TND-1510, Dec. 1962:173-186. Polytechnic Inst. Brooklyn.
[13] Silvestre N. Buckling behaviour of elliptical cylindrical shells and tubes under compression.
International Journal of Solids and Structures 2008;45(16):4427-4447.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Heiko Merle, Jörg Lange

Institut for Steel Structures and Material Mechanics, Technische Universität Darmstadt, Germany
merle@stahlbau.tu-darmstadt.de, lange@stahlbau.tu-darmstadt.de

Keywords: stability, buckling, local joint collapse,

Abstract. The collapse of a steel structure required a detailed analysis of the reasons for the catastrophe.
Several parts of the whole process of the design and construction had to be assessed. The results of this
analysis are presented. Furthermore it will be shown how the design process, the autonomous checking
procedure of the structural design, the workmanship, the construction on the erection side, the material
characteristics as well as human failure affect the failure.

Trusses with hinge joints experience only tension and compression forces. Compression forces in
combination with hinge joints lead to a structural design using the well known Euler’s cases especially
the Euler case II. By designing the hinge joints the mounting of the joints has to be considered. This asks
for a long connection area. Due to the compression, the joints have to be as compact as possible. These
two oppositional requirements have to be considered in the design and the construction. The analysis of a
collapse of a truss within the construction of a coal power plant showed that the failure of non-compact
joints led to a catastrophe. The non-compact joints increased the buckling length severely beyond the
Euler case II. The collapse of the joints led to a global collapse of the truss.


Power plants have as main load-bearing structure a boiler supporting steelwork. This steelwork bears
the steel boiler on its inner side. Several working platforms, power piping and the technical installations
are attached to the supporting steelwork. During the erection the boiler supporting steelwork has to be
built first. Subsequent the boiler itself as well as the platforms and installations will be fit in. Therefore
several auxiliary steel structures are needed.
The building described in this paper contains auxiliary platforms up to the top of the main structure.
These platforms were mounted in segments and held by the boiler supporting steelwork at several levels.
One auxiliary structure was seated at level +78.00 m and rose up to the level of +145.00 m. At level
78.00 m the structure was held by a truss. The boiler supporting steelwork bore this truss. Figure 1 shows
the truss with a height of 4740 mm and a length of 36500 mm. This girder was held out of its plane by
several supports. Therefore buckling out of the plane can be neglected.
The girder’s cross sections were H-Beams of different sizes. The tension forced bottom chord was a
hot rolled H-Beam of 700 mm height and 300 mm width with a steel grade of S 355 (yield strength = 360
N/mm²). The upper chord was a welded H-Beam of 2100 mm height and 400 mm width with a steel grade
of S 235(yield strength = 240 N/mm²). This member was part of the final structure and therefore its
design had to take care of additional requirements. Because of the large slenderness of the upper chord it

Heiko Merle, Jörg Lange

was stiffened with many braces. The trusses outer diagonals were hollow sections with a height and width
of 350 mm and a thickness of 30 mm with a steel grade of S 355.

Level + 82.740m

Level + 78.000m

6208 3559 3560 4923 4923 3560 3559 6208


Figure 1: Truss at level +78.00m

A portal frame transferred the loads from the sections standing above into the truss. This structure
may be reduced to the single girder with vertical loads from the portal frame. A static analysis of the
upper auxiliary steel structure as a two-dimensional framework led to four main vertical loads for the
girder. The loading points were the outer joints of the upper chord and the outer diagonal. These loads
were design loads with specific partial safety factors. Figure 2 shows the loads and loading points.

Fd = 1514kN Fd = 1070kN Fd = 1070kN Fd = 1509kN

Level + 82.740m

Level + 78.000m

Figure 2: Truss and design loads

The inner diagonals as well as the vertical struts were designed for small forces with max
Nd = 130 kN. The bottom chord got a tension force of Nd = 3300 kN. The upper chord received a
compression force of Nd = 2500 kN. Due to its large dimension buckling was eliminated. At last the outer
diagonals were analyzed. They receive a compression load of Nd = 4160 kN. A comparison of the original
structural design and a new and autonomous structural design after the collapse led to the same
compression and tension forces. All members were strong enough to carry the applied loads.


The structural design of the outer diagonals is governed by buckling according to Euler’s case II.
Figure 3 shows that this Euler case relies on non-sway hinge joints at both ends. By using the national
code the ideal buckling load as well as the ultimate load can be analyzed.

Heiko Merle, Jörg Lange

Figure 3: Euler’s cases of buckling taken from [1]

π2 π2 (1)
N ki ,z = ⋅ EI = ⋅ 21000kN / cm 2 ⋅ 66112cm 4 = 22442kN

By using the German code DIN 18800 [2] the effective slenderness can be calculated with the help of
the buckling curve c.

N pl 384cm 2 ⋅ 36kN / cm 2 (2)

λ= = = 0.785 Ÿ κ = 0.68
N ki 22442kN

The hollow sections ultimate load is defined by:

N ultimate = κ ⋅ N pl,d = 0.68 ⋅12557 kN = 8539kN ≥ N d = 4160kN (3)

The ultimate load is twice as big as the compression force of the outer diagonal. Therefore a collapse
of the outer diagonal according to Euler’s case II is not expected.


1 0

M27 - 10.9

M27 - 10.9

Plate 30 x 295; S355 A-A
58 Plate 30 x 350; S355
HEB 700; S355 0






Figure 4: Construction of the outer diagonal and its joints

Heiko Merle, Jörg Lange

By analyzing the workshop drawings it can be assessed that the support conditions are not as assumed
in Euler’s case II. The top and bottom flanges of the hollow section were cut out at the end of the beam to
enable the boltability. Figure 4 shows an overview of the outer diagonal and its joints. Special attention
shall be given to the 704 resp. 705 mm of the compression strut that are without top and bottom flange.

The hollow section is connected with the joint plate by bolts. The joint plate is connected to the chord
by a welding seam. Therefore the modeling of the outer diagonal to the Euler’s case is incorrect. The
slenderness of the joint plate and its length lead to the model of a segmented beam with different
stiffnesses. Rotation springs as well as translation springs have to be used at the end of the segments.
Whereas the Euler’ case II has an elastic buckling mode the segmented beam has a combination of an
elastic buckling and a rigid body motion (fig. 8). The stiffness of the springs as well as the relative
stiffness of the equivalent end segments built by the joint plates are difficult to define. Therefore the
whole truss was modeled as a three-dimensional finite element system.
5.1 Finite element method – elastic eigenvalue analysis
By using the finite element software SOFiSTiK the top as well as the bottom chord, the outer
diagonals, and the joints of the girder were modeled with shell elements. The stiffeners of the upper chord
were modeled too, to avoid local buckling effects. The inner diagonals and the vertical compression struts
were modeled using beam elements. A very fine discretization of the model was used to get results as
close to reality as possible. The model was supported as shown in figure 1. To avoid a horizontal
displacement of the system a support in the point of symmetry was modeled.
The material characteristics were ideal elastic and ideal plastic functions with a yield strength of
fy = 240 N/mm² (S235) resp. fy = 360 N/mm² (S355) and a modulus of elasticity of E = 210000 N/mm².
The loads were as shown in figure 2. Figure 5 gives a view of the finite element model.

Figure 5: General finite element model

Figure 6 (left) shows the connection between the outer diagonal and the bottom chord whereas figure
6 (right) shows the connection between the diagonal and the upper chord.

Upper joint
Bottom joint

Figure 6: Connection between diagonal and chords in detail

By using the finite element model and the loads an elastic eigenvalue analysis was done. 10% of the
design loads were applied to the system. Thirty eigenvalues with the associated buckling modes were

Heiko Merle, Jörg Lange

calculated. The eigenvalues can be divided into two major groups. Firstly there are many values with
local buckling figures of the slender web of the upper chord. These values will not be pursued. The other
group is made up of eigenvalues with buckling modes containing both outer diagonals.
The result of this analysis is shown in figure 7. Eigenvalues 1 to 3 are buckling modes of the outer
diagonals. The first and second mode is a rigid body motion of the diagonal with a sideways buckling of
the bottom joint plates. There is a symmetric rigid body motion of both diagonals in eigenvalue 1 whereas
in eigenvalue 2 an asymmetric rigid body motion can be observed. The third eigenvalue describes a rigid
body motion of the diagonal with a sideways buckling of the top joint plates. The first eigenvalues are
close together between the load factor 13.1 and 14.1. By using 10% of the design load, the ideal buckling
load can be calculated.

Eigenvalue 1: Eigenvalue 2: Eigenvalue 3: Eigenvalue 27:

load factor 13.1 load factor 13.7 load factor 14.1 load factor 59.1
Figure 7: Results of the eigenvalue analysis: First, second and third eigenvalue with rigid body motion
and 27th eigenvalue with buckling mode of Euler’s case II

Eigenvalue 1: Nki,1 = 5370kN Eigenvalue 2: Nki,2 = 5600kN Eigenvalue 3: Nki,3 = 5780kN

Figure 7 shows the 27th eigenvalue and the buckling mode analogous to Euler’s case II. This buckling
mode has an ideal buckling load of Nki,Euler = 24200 kN with the load factor of 59.1. Compared with
formula (1) there is an explicit accordance between the two ideal buckling loads. This shows the usability
of the finite element model. Altogether the first three eigenvalues show a rigid body motion. The failure
mode is not the buckling of the outer diagonal but rather the buckling of the joint plates. The joint plates
act like a spring at the end of the diagonal. Schmidt et al. [3] showed an equivalent system that is given in
fig. 8.
Joint plates
Outer diagonal
Euler's case two Rigid body motion

Equivalent rotation spring

Rigid body motion

Equivalent translation spring

Figure 8: Equivalent spring system

5.2 Finite element method –plastic and nonlinear ultimate load analysis
To get the ultimate load of the finite element model a plastic and nonlinear load analysis was used.
Therefore the ideal elastic and ideal plastic function of the material characteristics was used. To include
the nonlinear secondary order theory the first eigenvalues buckling mode was fit in the model as its
imperfection. To take a realistic value the imperfection was set to 1/500 of the length of the joint. Then an

Heiko Merle, Jörg Lange

ultimate load iteration was started. Within load steps the design load was raised to the ultimate load of the
Design load factor

Ultimate load factor
Load factor



0,2 Sideways deformration of the upper joint plates

Sideways deformation of the bottom joint plates

0 2 4 6 8 10 12 14 16 18 20
Sidways deformation [mm]

Figure 9: Sideways deformation of the joints in the plastic and nonlinear load analysis

Figure 9 shows the joint plates sideways deformation of the connection between the chords and the
diagonals which are shown in figure 6. The plates get a clear nonlinear sideways deformation. At the load
factor 0.69 the deformations will be infinite and the outer diagonals as well as the joint plates fail. At this
load factor the system analysis is not convergent and the whole system collapse at 69% of the design
loads. The deformation figure in the last convergent load step is affine to the first buckling mode of the
first eigenvalue.


After the collapse the scrap was analyzed. Figure 10 shows in the upper left photography the upper
part of the diagonal with its connection to the upper chord. The joint plates have a sideways deformation
affine to the finite element solution. The upper right photography shows the diagonal with its joint plates
to the bottom chord as well as the bottom chord. Again the sideways deformation of the joint plates is
clearly visible. The bottom photography shows the bended bottom chord, the joint plates and the straight

Figure 10: Debris after the collapse

Heiko Merle, Jörg Lange

As the finite element system showed, the connections between the outer diagonals and the chords fail
first. In this moment the whole system is still stable. The inner forces of the truss without the diagonals
create bending moments in the bottom chord. These bending moments create the extreme bending
deformation shown in figure 10. Due to the large deformations the whole girder slips off the support and
the system collapses.

In the process of design and construction of steelwork structures there is a straight progression.
Firstly the structural analysis gives the loads and forces as well as profiles and defined materials.
Subsequently the detail design with workshop drawings and detail drawings follows.
The truss was firstly designed as a framework. All structural elements and their material definitions
were determined by using the theory of stability respectively the elastic second order theory. This
structural analysis was done by the structural engineer. After this the design engineer determined the
structural details. He has to be aware of problems like the boltability and the erection. The design
engineer receives the forces and the stresses from the structural analysis. The load flow in the details had
to be analyzed. The discrepancy between a compact design of the connection and montage aspects as well
as the boltability had to solved. Therefore the structure had to fulfill structural restraints as well as the
static requirements. After the detail design was finished the structure was fabricated.

Figure 11: Progression of design and construction

There was no feedback loop between the engineers. Figure 11 shows that after the construction of the
details the structural engineer should have checked the design. The outer diagonals in coherence with the
new designed details had to be analyzed again. There the engineer had to notice that an important
alteration of the support conditions of the outer diagonal happened. An updated structural analysis should
have been done. This might have led to the recognition of the structures ultimate load and failure
criterion. The design engineer would have had to redesign the details to fulfill the requirements of the
structural design. This kind of feedback loop will take more time but maybe the collapse of the girder
could have been avoided.

Heiko Merle, Jörg Lange

[1] Petersen C., “Stahlbau”. Vieweg , Braunschweig , 1993
[2] DIN 18800-2 “Steel structures – Part 2: Stability – Buckling of bars and skeletal structures”,
[3] Schmidt H. et al., “Ein ungewöhnliches Stabilitätsproblem verursacht Schadensfall – An
uncommon stability problem causes failure”, Stahlbau 77 (12), 862-869, 2008
[4] Unterweger, H.; Ofner, R, “Traglast von Verbandsstäben aus Hohlprofilen mit quasi-zentrischem
Knotenblechanschluss – Load bearing capacity of bracing members with almost centric joints“,
Stahlbau 78 (6), 425-436, 2009

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



K.S. Sivakumaran, P. Arasaratnam, and M. Tait

Department of Civil Engineering, McMaster University, Hamilton, ON, CANADA, L8S 4L7
e-mail: siva@mcmaster.ca

Key words: Experimental, Steel beams, Flange holes, Ductility, Steel standards

Abstract: This paper presents an experimental investigation, involving twenty five steel beam specimens,
on the effects of flange holes on the flexural behaviour of steel I-beams. This study used ASTM A992
grade steel beams. Circular holes of various diameters, ranging form 0% to 48% of the gross flange area
are under consideration Based on the experimental results, this research study recommends a design
approach analogous to the axial tension member provisions as per the current CAN/CSA-S16.01
standard [1]. The comparison of the proposed procedure with the 15% exemption rule as per current
steel standard S16.01 [1] demonstrated that the current code provision is unnecessarily conservative for
steel grades such as A992 steel. On the other hand, the current code provision may not be adequate for
higher strength steels such as HSLA 80 steel, ASTM A913 Gr: 60 and HPS-485W having the minimum
yield-to-ultimate strength ratio of more than 0.85.

Flange holes are frequently made in structural steel construction, primarily for bolting purposes. The
influence of flange holes on the flexural behaviour of beam members has been the focus of debate for
many years. Early North American design codes allowed a designer to place holes in flanges up to 15% of
the gross area of the tension flange without penalty. If more than 15% of the gross flange area is removed,
the amount of area exceeding 15% would be deducted in calculating the section properties and typically,
only the yield moment could be used rather than the plastic moment. This provision was based on the
study by Lilly and Carpenter [2] on riveted plate girders made of ASTM A7 steel having the yield-to-
ultimate strength (Fy/Fu) ratio of about 0.5. However, in 1989, the Allowable Stress Design version of
specification [3] adopted a new provision that altered the use of the 15% exemption rule in this subject
matter. This specification introduced for the first time a mathematical formula based on the ratio of the
fracture strength of net area (AfnFu) and the yield strength of gross area (AfgFy) of the tension flange to
ignore the effects of holes. The present AISC-Load and Resistance Factor Design version [4] of the
specification also follows the same procedure as specified in the 1989-AISC specification [3] to ignore
the effect of flange holes.
The present trend in steel construction industry is to use higher strength steels with better structural
performance over traditionally used ASTM A36 steel. These high strength steels have the specified yield-
to-ultimate strength ratio ranging from 0.75 to a code permitted maximum of 0.85. In some instances,
steels such as HPS-485W, HSLA 80 steel and ASTM A913 Gr: 60 exhibit yield-to-ultimate strength
values of more than 0.85[5]. Nevertheless, the comparisons of corresponding various international code
provisions indicate that the 15% exemption rule which is currently in use as per the clause 14.1 of the
current Canadian Steel Design Code [1] is more restrictive for steel grades having the yield-to-ultimate
strength of less than 0.85, whereas it is inadequate and inappropriate for the high strength steels with the
minimum yield-to-ultimate strength of more than 0.85.

K.S.Sivakumaran et al.


The objectives of the research program presented in this paper were: (i) to investigate the effects of
flange holes and flange fastener holes on the strength and rotation capacity of steel I-beams made of
ASTM A992 steel, (ii) to assess the applicability of the 15% exemption rule used in the clause 14.1 of the
current steel code provision [1] along with various other international steel code provisions dealing with
the proportioning of flexural members with flange holes (or fastener holes) and (iii) to provided
recommendations on the modification of the current CSA code provision[1].
2.1 Test Specimens
The test program considered twenty five full scale beam specimens (Beam section W200X42) each
having a nominal length of 3050 mm. All beam specimens were from the same production batch thus the
material characteristics discrepancy would be minimal. The tests were divided into four series as follows:
Series-1: This series involved the beam tests with solid flanges. Four beams under consideration were
named as A100-1, A100-2, A100-3 and A100-4, wherein ‘A100’ denotes that the area of tension flange of
100% and the number that follows denotes the test-number.
Series-2: The beam tests of this series contain holes in tension flanges only. Seven different
configurations with the net flange area-to-gross flange area (Afn/Afg) ratio between 90% and 50% were
considered. The beam specimens tested in this category were named as A90-1, A85-1, A80-1, A75-1,
A70-1, A60-1 and A50-1. For example, here, A70 indicates that (Afn/Afg) ratio is 70%. In addition, tests
on the beam specimens A75, A70 and A60 were repeated due to the fact that such beam specimens
exhibited dominant failure modes varying from a mixed type of local compression flange buckle followed
by net-section fracture to a definite net-section fracture.
Series-3: This series included beam specimens with holes in both tension and compression flanges. It
includes four tests A85-B-1, A75-B-1, A70-B-1 and A60-B-1, where ‘B’ denotes both flanges. The
purpose of this test was to investigate the flexural behaviour when holes exist in both flanges.
Series-4: This series included beam specimens with flange holes in both flanges, and with fasteners
placed in these holes. Standard size of high strength ASTM A490 fasteners, leaving a clearance of
approximately 2 mm between the perimeter of hole and the outer surface of the fastener were inserted into
the holes of beam specimens: A85-F-1, A75-F-1, A70-F-1 and A60-F-1, where “F” represents fasteners.
The fasteners were tightened by a hand wrench to a specific level. The purpose of this type test was to
investigate the role of fasteners in resisting the flexural stresses in compression flanges.
2.1.1 Mechanical Characteristics
Six standard tension coupon tests involving 3-flange coupons and 3-web coupons were conducted. All
coupons, except the web coupons obtained closer to the flange-web junction, exhibited a sharp yield point
followed by a yield plateau. However, the web coupons obtained closer to the flange-web junction
exhibited no sharp yield point, and showed higher yield and ultimate strengths, and lower ductility
compared to other tension coupons tested in this research program. This can be attributed to the fact that
higher stresses exerted at the corner of rolled sections during the course of rolling process and faster
cooling following rolling due to the smaller web thickness. The average measured elastic modulus of such
coupons was of 215GPa. The yield strength of each coupon was established by the method of 0.2% strain
offset, though the flange and middle-web coupons exhibited a shaper yield point. The average measured
yield and ultimate strength of flange coupons were of 409 MPa and 531 MPa, respectively resulting in the
yield-to-ultimate strength ratio of 0.77.

2.2 Testing of Beams

Test Setup: Figure 1 shows a photographic image of the overall test setup. Each beam specimen was
simply supported at its ends and was subjected to two point loads applied at a distance of 750 mm apart
leaving a shear span of approximately 1075 mm on either side of the mid-span of the test beam. The test
arrangements allowed for large end rotations and vertical displacements that might occur during the test.

K.S.Sivakumaran et al.

Figure 1: Overall View of Test Setup Figure 2: Bracing System

Bracing System: Figure 2 shows a close-up view of the bracing system used in this test program. The
triangular bracing system consisted of 3 members whose sizes are shown. A solid cold-rolled bar of 25
mm diameter was welded to the vertical member to function as a knife edge guide. The whole assembly
was firmly fastened to the laboratory test floor. Prior to applying the test load, the bracing frames were
adjusted such as to touch the flange tips of the test specimen and then tightened to the test floor.
Loading System: As could be seen in Figure 1, the loading system consisted of a 500 mm stroke
actuator combined with a commercially available load cell of 900 kN capacity and a 500 mm stroke string
transducer attached between the load cell and the outer perimeter of the actuator. Since this is a
displacement controlled loading system, it also included a controller, function generator, power supply
and a servo valve. The loading system was positioned upside down and loaded from above at mid-span.
Instrumentation: In order to determine the rotation of the beam specimen, potentiometers were placed
at each ends of the beam. The differential readings between a pair of potentiometers and a pair of LVDTs
and the corresponding vertical distances between them were used to calculate the beam end rotations. The
deferential reading between a pair of vertical
potentiometers and the corresponding
horizontal distances between them were used to
establish the rotations at load points. The mid-
span deflection of the test beam was measured
using potentiometer S.P-3. The vertical
deflections at the quarter points of the test
beams were also measured using
potentiometers. High elongation capacity strain
gages were also used in some of the beam tests.
Additional instruments such as LVDT-3 and
LVDT-4 were used to monitor the out-of-plane
movements of the compression flanges with Figure 3: Instrumentation of Test Beam
respect to the tension flange at the center of the
mid-span. These instruments detected the initiation of local buckling at the center span of the test beam.
Test Procedure: Once the instruments were properly attached to the specimen, it was preloaded using
the displacement control loading system. The applied preloading was within the elastic range. The beam
specimen was then unloaded and instruments were reset. Once this preload protocol is completed, which
was to ensure proper seating of the test beam within the loading frame, actual loading began. The test
beams were subjected to increasing displacements until failure. The loading rate of 0.025 mm/sec was
maintained throughout the test. The beam test was considered complete when the load versus mid-span
deflection curve reached below the plastic load again on the unloading branch. However, in the case of
beam tests where the failure of the specimen occurred as a result of net-section fracture the test was
terminated as soon as a sudden drop in loading was noticed.

K.S.Sivakumaran et al.

Some of the results reported herein include normalized quantities of (1) load versus mid-span
deflection, (2) moment versus load point rotation and (3) moment versus beam end rotation. The rotation
is the average of the rotations measured underneath the two load points. Table 1 presents the measured
peak moments Mm associated with each test and the theoretical gross-section plastic moment MP of each
test specimen. Theoretical MP considers the openings and the resulting neutral axis shift. Table 1 also
provides the percentage reduction in strength as compared to the solid beams (see column 7).
Table 1: Comparison of Experimental Peak Moments with Theoretical Plastic Gross-Section Moment

Type [Afn Mm Mave %difference
Beam [AfnFu/ MPave rage
of /Afg] (Test) (Test) compared to
ID AfgFy] (kNm) Mm/
Test (%) (kNm) (kNm) solid section
(2) (4) (8) MP
(1) (3) (5) (6) (7)
A100-1 100 1.30 215
Series-1 A100-2 100 1.30 214
214 0.0 176 1.22
A100-3 100 1.30 214
A100-4 100 1.30 214
A90-1 91 1.18 214 214 0.0 176 1.22
A85-1 85 1.10 216 216 0.9 176 1.23
A80-1 79 1.03 212 212 0.9 174 1.22
Series-2 A75-1 74 0.96 210 209 2.3 175 1.19
A70-12 74
71 00.92
96 206
204 204 4.7 176 1.16
A60-12 71
62 00.81
92 205
197 195 8.9 174 1.12
A60 2 62 0 81 194
A50-1 52 0.67 178 178 16.8 172 1.03
A85-B-1 86 1.17 210 210 1.8 178 1.18
Series-3 A75-B-1 74 0.96 200 200 6.5 176 1.14
A70-B-1 70 0.91 197 197 7.9 176 1.12
A60-B-1 63 0.82 192 192 10.3 179 1.07
A85-F-1 85 1.10 212 212 0.9 175 1.21
A75-F-1 74 0.96 210 210 1.9 174 1.21
Series-4 A70-F-1 70 0.91 207 207 3.3 177 1.17

A60-F-1 62 0.81 194 194 9.3 175 1.10

Series-1: Solid Beam Tests: The maximum moment carrying capacity of solid beams A100-1, A100-
2, A100-3 and A100-4 were 215 kN.m, 214 kN.m, 214 kN.m and 214 kN.m, respectively. However, the
corresponding measured average load point rotations corresponding to peak moment were of 0.0938,
0.0972, 0.0949 and 0.0878 radians, respectively, resulting in the maximum deviation from the average
measured rotation (0.0934 radians) of approximately 6%. The normalized moment (M/MP) versus the
normalized load point rotation (Ө/ӨP) relationship for each solid beam was established. The moment
versus load point rotation relationship was in close agreement up to the peak moment, even though slight
variations were observed perhaps due to the inherent variability associated with the presence of residual
stresses and initial geometric imperfections. Two different rotation capacities such as Ry (a measure of

K.S.Sivakumaran et al.

available rotation capacity corresponding to the plastic moment MP obtained on the unloading branch)
and Rm (rotation capacity at peak moment) were established in this research program. The average
measured Ry and Rm of the solid beam specimens were 23.5 and 13.1, respectively. The failure of the
solid beam was due to local flange buckling of the compression flange which was followed by lateral
torsional buckling in the critical span region.
Series-2: Beams Having Holes in Tension Flange Only: Figure 4 shows the normalized moment,
M/MP versus the normalized average load point rotation, Ԧ/ԦP for the beam specimens with holes in
tension flanges only. In order to illustrate how the flexural behaviour of steel member could be influenced
due to the presence of holes in the tension flanges, the moment-rotation response of a solid beam (A100-
3) is also shown in the same figure. Figure 5 shows a close up view of failure pattern of the beam
specimen (A60-3) failed as a result of net-section fracture through the holes in tension flange.

Figure 4: Normalized Moment Versus Figure 5: Failure Pattern of beam with holes in
Normalized Load Point Rotation Tension Flange
From figure 4, it can be noted that the rotation capacities of the flexural members were reduced even
when the holes removed was small, say approximately 10% (A90-1). However, it can be observed that the
strength of the flexural members was not significantly impacted provided the nominal net-section fracture
strength was greater than nominal gross-section yield strength (AfnFu•AfgFy). This ratio is given in Table
1- Column 4. The percentage reduction in strength (Table 1- Column 7) increased as the AfnFu/AfgFy ratio
became lower than 1.0. Thus, for beam specimens having the AfnFu/AfgFy ratio of 0.96 (26% flange holes
of gross flange area) and 0.92 (29% flange holes of gross flange area), the percentage reductions in the
average maximum load were of 2.3% and 4.7%, respectively, compared to that of solid beams. These
specimens having the AfnFu/AfgFy of 0.96 and 0.92 eventually failed by net section fracture which
occurred after visible local bulking of the compression flanges in the uniform moment region which can
be seen in Figure 5. However, for beam specimens having the AfnFu/AfgFy ratio of 0.81 (38% holes of
gross flange area) and 0.68 (48% holes of gross flange area), which were well below 1.0, the percentage
reductions in the average maximum applied load were of 8.9% and 16.8%, respectively. Such beam
specimens failed by net-section fracture, prior to local bucking of compression flange. The reduction in
the moment capacity of beam specimens, having the AfnFu/AfgFy>1.0, was not substantial, although a
slight reduction did occur with increasing hole size. Based on these results, suppose it is presumed that
any strength reductions within ±5% range can be ignored from a design stand point, then the tension
flange holes of up to 29% of the gross flange area can be safely ignored in beams made of ASTM A992
steel having yield-to-ultimate strength ratio of 0.77. Table 1- Column 9 gives the ratio of test moment to
theoretical moment resistance. For series-2 specimens, since the Mm/MP for all specimens were more than
1.0, it can be concluded that the tension flange rupture did not occur prior to the attainment of the gross-
section plastic moment, when the holes removed was from 9% to 48%.

K.S.Sivakumaran et al.

Series-3: Beam Having Holes in Both Flanges: As presented in Table 1, the percentage decrease in
the moment capacity of beam specimens with holes in both flanges (Series-3) having the AfnFu/AfgFy
ratio of 1.17, 0.96, 0.91 and 0.82, compared to the solid beams, were 1.8%, 6.5%, 7.9% and 10.3%,
respectively. As expected, the flexural behaviour of beam specimens in terms of strength and rotation
capacity was considerably influenced as holes were present in both flanges. The percentage decrease in
the maximum moment capacity of beam specimens having the A fnFu/AfgFy ratio of 1.17 (A85-B-1), 0.96
(A75-B-1), 0.91 (A70-B-1) and 0.82 (A60-B-1), compared to the corresponding beam specimens having
holes in the tension flanges only (Series-2) were of 2.7%, 4.2%, 3.2% and 1.4%, respectively. The beam
specimen having the AfnFu/AfgFy ratio of 0.96 (26% holes of gross flange area), which is closer to 1.0,
failed due to local buckling of the compression flange whereas, similar beam specimen (A fnFu/AfgFy
=0.96) having holes in the tension flange only failed due to net-section fracture of the tension flange,
which occurred after noticeable local buckling of the compression flange. This can be attributed to the
fact that the compression flange was weakened locally due to the presence of holes which resulted in
early yielding of the locally buckled compression flange. However, the beam specimens having the
AfnFu/AfgFy ratio of 0.91 and 0.82 failed due to tension fracture.
Series-4: Beam Tests Having Holes With Fasteners in Both Flanges: These tests were somewhat
similar to Series-3 tests, in that both set of beams had holes in both flanges, except that fasteners were
present in the holes in the current set of beams. For beam specimens having the AfnFu/AfgFy ratio of 1.10
(A85-F-1), 0.96 (A75-F-1), 0.91(A70-F-1) and 0.82 (A60-F-1), the percentage reduction in the maximum
moment capacity in compared to similar solid beam specimens, were of 0.9%, 1.9%, 3.3% and 10.3%,
respectively. The moment capacities of beam specimens were greatly improved when the holes in the
compression flanges were filled with the standard size of fasteners. The percentage improvement in the
maximum moment capacity of beam specimens having the AfnFu/AfgFy ratio of 1.10, 0.96, 0.91 and 0.81
when compared to the similar beam sections having holes in both flanges were of approximately 50%,
71%, 58%, and 10%, respectively. This clearly indicated that the presence of fasteners within the holes in
the compression flanges improved the moment resistance of beams with flange holes.


Largest experimentally measured moment (Mm), the calculated gross-section plastic moment (MP=
ZgFy) and the calculated net-section fracture moment (Mfn= ZnFu) of each beam specimens were
established. Note that in calculating the plastic section modulus of net-section, Zn, the neutral axis of the
net-section was presumed to be shifting from the neutral axis of the gross-section to that of the net-section
for beam specimens having holes in tension flanges only. Also, similar procedure was adopted for beam
specimens having the fastener holes in both flanges although, the strain measurements at the middle of the
web indicated that the movement of the neutral axis was not detected. Nevertheless, the consideration of
the position of neutral axis shifting from the gross-section to net-section would yield a lower moment
capacity [conservative design approach]. Comparing the gross-section plastic moment (MP) and the net-
section fracture moment (Mfn), the MP/Mfn ratio increased with increasing AfnFu/AfgFy ratios. This
suggested the proposed design approach which is analogous to an axial tension member provision. That is
the gross-section plastic moment capacity and net-section fracture moment should be checked and the
lesser of two could be used as a design moment.
However, a detail analysis of the experimental results [Not given here] indicated that the (MP/Mfn)
ratio was less than 0.85 when the AfnFu/AfgFy ratio was greater than 1.0 for beam specimens having holes
in either tension flanges only or fastener holes in both flanges. Moreover, in such cases, the beam
specimens eventually failed due to local buckling of the compression flange preceded by lateral torsion
buckling in the critical span region [ductile failure]. On the other hand, the M P/Mfn ratio was greater than
0.85 when the AfnFu/AfgFy ratio was reduced to below 1.0. The failure of beam specimens in this case was
mainly due to local buckling of the compression flange in the critical span (mid-span) region which was
eventually followed by net-section fracture in the tension flange [brittle failure]. However, for the beam
specimens having holes in both flanges and having the AfnFu/AfgFy ratio of greater than 0.95, the MP/Mfn

K.S.Sivakumaran et al.

ratio was less than 1.0. Also, the failure of beam specimens in such cases was mainly due to local
buckling of the compression flange preceded by lateral torsional buckling in the critical span. On the
other hand, as the AfnFu/AfgFy ratio became less than 0.95 the MP/Mfn ratio was increased to more than
1.0. The failure of beam specimens in such cases was mainly due to net-section fracture of the tension
flange. By considering all the scenarios tested in this research program, a factor of 0.85 can be considered
as an optimum upper bound that should be used to multiply the theoretical net-section fracture moment
(Mfn). Therefore, this research study suggests a design check, which is analogous to the tension member
provision as per the current CAN/CSA-S16.01 [1] standard,; (a) The gross-section shall be designed for
the gross-section plastic moment capacity, MP (ZgFy) (or lower if compression flange or web limit states
control) (b) Calculate the factored net-section fracture moment, Mfnf =0.85ZnFu. If MP ≤ Mfnf, the effects
of holes (or fastener holes) shall be ignored and the flexural member shall be designed for its gross-
section plastic moment as usually followed in the design solid beams. Otherwise, design the member to
carry the factored net-section fracture moment.
Overall, the design moments calculated as per the proposed design approach resulted in higher design
moments than that permitted by the current code provisions for flexural members having either flange
holes or flange fastener holes. The 15% exemption rule, which is still in use as per the current CAN/CSA-
S16.01 (Clause 14.1) code provision [1], is conservative for currently used structural steels which often
possess a yield-to-ultimate strength ratio of less than 0.85. Note that the design moment as per the
proposed design procedure in this investigation has a reduction factor of approximately 0.85 as compared
to the maximum measured moments associated with the net-section observed in this investigation.
Moreover, the suggested design method was analogous to the tension member provision as per the current
CAN/CSA-S16.01 (Clause 13.2) code provision [1] eliminating unnecessary ambiguity in regards to the
design of flexural members having holes (or fastener holes) in tension flanges. That is, the clause: 14.1 of
the current CAN/CSA-S16.01 standard [1] treats the effects of holes and the effects fastener holes in
different manner, in which when holes occur in flanges a theoretical net-section calculations shall be
followed whereas, when fastener holes in beams is considered, the 15% exemption rule would be applied.
However, the proposed method in this investigation follows a unified approach, in which the effects of
holes or fastener holes that may present in flanges of a flexural member or a tension member would be
treated in an identical manner. In addition, the proposed method as opposed to the current CSA code
provision [1] takes into account the material characteristics in terms of yield-to-ultimate strength ratio.

The following points summarize the main observations of this research program:
[a] Experiments considered ASTM A992 steel with the measured yield-to-ultimate strength ratio of
0.77 beams having flange holes as high as 48% of the gross area of the tension flange. Though tension
flange rupture was observed in some cases, the peak moments in all of the beams were higher than the
gross-section plastic moment (MP) for the beam [b]The strain measurements indicated that no great
deviation occurred with regards to the position of the neutral axis of the gross cross-section when holes
when holes were made in the tension flange only (or fastener holes occurred in both flanges) [c]The strain
measurements made in the vicinity of hole region of beam specimens A75-3 and A75-2, in which holes
existed in tension flanges only, were about 1.2% and 2%, respectively when the beam members reached
the gross-section plastic moment, MP. This yielded a conclusion that the flexural members with holes in
tension flanges only require a strain in the range of 6-10 times the yield strain (0.2%) for the ASTM A992
steel as has been already verified by Dexter et al.[5] who performed flexural tests made of HPS 480W
steel grade. [d]When holes were present in the tension flange only, and for the cases of fastener holes in
both the tension flange and the compression flange, the failure of flexural members having the
AfnFu/AfgFy≥1.0 was primarily due to lateral torsional buckling which was eventually followed by local
buckling in the critical span (mid-span) region. It was noted in such cases that the gross-section plastic
moment-to-the net-section fracture moment (MP/Mfn) ratio was less than 0.85. [e] The design moment
calculation as per the proposed design method was quite comparable with the present AISC code

K.S.Sivakumaran et al.

provision [4]. However, beyond a threshold value, depending on the yield-to-ultimate strength ratio, the
proposed method allowed higher moments on net-sections than the presently used design code provisions.
It should also be noted that the proposed design moments are lower than the experimentally measured
maximum moments on the net-section. Therefore, the design moments as per the proposed design method
would be safe. [f] The ratio of the nominal net-section fracture strength (AfnFu)-to-the gross-section yield
strength (AfgFy) did not seem to be as of a significant parameter for flexural members as it is for the
tension members in determining the required strength since the flexural member (A50-1) having the
AfnFu/AfgFu ratio as low as 0.67 attained the maximum net-section moment which is more than the gross-
section plastic moment. However, this parameter seemed to significantly influence the available total
rotational capacity of flexural members having flange holes and fastener holes. [g] All beam specimens
tested in this investigation attained more than the required rotation capacity of 3 before the onset of local
buckling However, the required rotation capacity for non-seismic applications as per the current AISC
specification [4] is greater than or equal to 7-9. In this investigation, beam specimens having the
AfnFu/AfgFu ≥ 1.0 exhibited substantial inelastic rotation capacity beyond the maximum load and were
able to reach the gross-section plastic moment on the unloading branch. Thus, beam specimens with holes
in the tension flanges only and fastener holes in both flanges satisfying A fnFu/AfgFu≥1.0 exhibited a total
available rotation capacity, Ry of more than 9. If the condition was violated, the beam specimens failed
primarily due to a rupture of tension flange through the flange holes which occurred before the flexural
members reached the gross-section plastic moment again on the unloading branch. That is, for flexural
members having the AfnFu/AfgFu<1.0 in the tension flanges, the inelastic deformation beyond the ultimate
load was substantially reduced. However, the beam specimens with holes in both flanges satisfying the
AfnFu/AfgFu≥0.95 exhibited a total available rotation capacity, Ry of more than 9. It should be noted that
the available rotation capacities would substantially vary depending on many parameters, such as the
cross-sectional geometry of the beam specimens, bracing locations (closer bracing will result in higher
rotation ductility), material strain hardening, local instabilities associated with flange and/or web
buckling, presence of initial geometric imperfections, etc. Thus, the generalization of available rotation
ductility from a certain type of flexural test is not reasonable. Further details of this investigation are
available in the thesis by Arasaratnam [6].

[1] CSA (2003), Limit States Design of Steel Structures, CAN/CSA-S16-01, Canadian Standard
Association, ON, Canada.
[2] Lilly, S.B., and Carpenter, S. T., (1940), Effective Moment of Inertia of a Riveted Plate Girder,
Transactions, American Society of Civil Engineers, Paper No. 2089, pp.1462-1517.
[3] AISC (1989). Allowable Stress Design for Structural Steel Buildings,9th Edition, American
Institute of Steel Construction, Inc., Chicago, Illinois, USA.
[4] AISC (2005). Load and Resistance Factor Design Specification for Structural Steel Buildings, 4th
Edition, American Institute of Steel Construction, Inc., Chicago, Illinois, USA.
[5] Dexter, R.J. Alttstadt, A. and Gardner, C.A. (2002). Strength and Ductility of HPS70W Tension
Members and Tension Flanges With Holes, Research Report, University of Minnesota,
Minneapolis, MN, 55455-0116, USA.
[6] Arasaratnam, P. (2008), “Effects of Flange Holes on Flexural Behavior of Steel Beams”, Ph.D.
Thesis, McMaster University, Hamilton, Ontario, Canada, p.xxv, p. 350.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



R.C. Spoorenberg* **, H.H. Snijder** and J.C.D. Hoenderkamp**

* Materials innovation institute M2i, Delft, the Netherlands

** Eindhoven University of Technology
h.h.snijder@tue.nl, j.c.d.hoenderkamp@tue.nl

Keywords: Residual stresses, roller bending process, plastic deformation.

Abstract. This paper presents residual stress distributions in roller bent wide flange HE 100B sections
obtained from experiments. The wide flange steel beams are curved at ambient temperature by means of
the roller bending process, which alters the initial residual stress pattern. Curved wide flange steel
sections are frequently used in large span structures like roofings and bridges. Their geometry and
loading often make these arches susceptible to instability phenomena’s. An accurate knowledge of the
residual stresses is therefore necessary. An experimental program was set up to investigate the residual
stresses in roller bent sections. Measurements were carried out on straight and curved sections. It was
found that the residual stresses in roller bent sections differ significantly from those in their straight

A structural steel section contains residual stresses which are a result of the manufacturing process.
Residual stresses are in general of primary importance for structures susceptible to loss of stability, since
the presence of these stresses causes early yielding and thereby a reduction in load carrying capacity.
Extensive research has been carried out on the measurements of residual stresses in straight hot rolled
sections, [1], [2], [3] and [4] amongst others. Results were summarized and published in various forms
[5], [6] and [7]. It was found that residual stresses in straight hot-rolled sections are characterized by
compressive stresses in the flange tips and tension in the web-flange junction. The web can be either
under compressive or tensile residual stress depending on the shape of the cross-section.
Cold forming structural steel at ambient temperature changes the residual stress pattern due to plastic
strains. This characteristic of cold forming was first observed in [8] where the residual stress distribution
in cold-formed circular hollow sections was investigated. Residual stresses in press-braked plates were
measured in [9], [10]. The residual stresses in bent sheet metals were reported in [11].
Theoretical models are available to obtain a residual stress distribution in cold-bent sections. A first
solution was proposed in [12] for bars under uniform bending based on elastic perfectly-plastic material
behavior. The basic theory is illustrated in Figure 1, where Į=ratio between the plastic and elastic section
modulus or shape factor, fy=yield stress, h=height of the cross section and R=radius of the circular arch.
When a stress free bar or plate is plastically bent into a specific radius, a plastic stress distribution
emerges (Figure 1(b)). After releasing the acting moments, an elastic release or springback of the member
takes place, thereby imposing an elastic stress distribution on the already present loading stresses (Figure
1(c)). The result is a stress distribution, which is a summation of the loading and unloading stresses
(Figure 1(d)) thereby assuming a uniaxial stress condition. The theoretical model has found widespread
application in structural analysis of curved steel, as reported in [13].

R.C. Spoorenberg et al.

Figure 1 Simplified theory on residual stresses due to cold bending.

This theoretical model has been used to find residual stresses in roller bent sections. However the
model is questionable since the true bending process features a complex interaction between rollers and
beam which cannot be represented by a uniaxial loading-unloading procedure. At Eindhoven University
of Technology an experimental and numerical investigation has been started to study the roller bending
process and its effects on wide flange steel beams. This project includes a variety of wide flange sections
and different steel grades. Some of the experimental results have been published in [14].
A short description of the roller bending process is presented in section 2. The experimental program
and test setup are given in section 3. In section 4 the results of the residual stress measurements as taken
from both straight reference sections and roller bent sections are presented. Section 5 discusses the results
of the measurements and the paper ends with conclusions in section 6.


Roller bending is a manufacturing technique whereby a straight wide flange section is shaped into a
curved one by feeding it through a roller bending machine. The roller bending machine can be equipped
with either three or four rollers. This study is confined to the first type. Because of the three rollers’
pyramid arrangement, roller bending is sometimes referred to as ‘pyramid rolling’ (see Figure 2).

Figure 2 Roller bending process

A straight member is placed in the machine (Figure 2(a)) and successive rolling and movement of the
rolls induces permanent curvature of the sections (Figures 2(b) and (c)). The top flange and bottom flange
are elongated and shortened respectively. The roller bending process is mainly featured by continuously
changing bending-type deformations as the section moves through the rollers. During the bending process
all three rollers are driven and automatic speed compensation is implemented for the difference in surface
speeds between the inner and outer circumference of the section being rolled. Depending on the machine
type, sections with a height of up to 1 m can be bent. Several additional passes are required until the
designated radius is attained. Steel members can be bent the easy way or the hard way which involves
bending about its weak axis or about its strong axis respectively. When bent the hard way, a small roller

R.C. Spoorenberg et al.

is placed at the inside of the top flange to prevent web crippling. This investigation is limited to wide
flange sections bent about the strong axis. Although complex curvatures are possible, this study is
confined to beams bent into a circular shape. Due to placement requirements within the bending machine,
it is impossible to impose a permanent curvature along the complete length of the beam. The ends remain
straight and have to be considered as waste material (Figure 2(c)). A more extensive description of the
roller bending process is reported in [15].


3.1 Experimental Plan

The complete experimental program is presented in Table 1. It comprises commercially available hot-
rolled wide flange steel sections bent into different radii and with different steel grades. A full overview
on the HE 100B profiles which will be presented in this paper is given in Table 2.

Table 1: Total experimental plan

Section Steel Grade Bending Radius [mm]
HE 100A1 HE 100B S235, S355 1910, 2546, 3820
HE 360B1 S235, S355 8000
IPE 3601 S235, S355 4500, 8000
Published in [14]

The initial residual stress distribution was determined from straight reference sections from which the
curved ones were made. Tensile tests were carried out on coupons taken from the flange of the straight
and curved members in order to assess the increase in yield stress and ultimate tensile stress as a result of
the roller bending process (Figure 3).

600 600

500 500
stress [N/mm ]

stress [N/mm ]


300 300

200 200

100 100
S235 S355 S235 S355
0 0
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
strain [-] strain [-]

Figure 3. Stress strain relationship for HE 100B with steel grade S235 and S355, straight reference
section (left) and roller bent section with a radius of 2546 mm (right).

R.C. Spoorenberg et al.

Table 2: Experimental plan

Specimen no. Steel grade Section Radius [mm] Measured yield stress of straight
reference sections [N/mm2]
1 1910 248
2 S235 2546
3 3820
HE 100B
4 1910 386
5 S355 2546
6 3820

3.2 Preparation
The sectioning method was used to measure residual stresses in roller bent steel arches. The test
specimen was saw cut from larger steel arches. Electrical strain gauges were applied to the surface of the
roller bent wide flange sections. For this investigation small (2 x 6 mm) electric strain gauges
manufactured by Tokyo Sokki Kankyujo Co. Ltd. were used. The arrangement of the strain gauges is
shown in Figure 4(a). To reduce end effects, the test area was a distance of 2.5 times the height of the
beam from the ends (Figure 4(b)). Only longitudinal stresses were measured.
A transverse saw cut and subsequent longitudinal saw cuts were made with an electrical band saw and
hand saw respectively. The influence of heat release from the electrical band saw cuts was suppressed by
supplying fluid coolant. Short-circuiting of the electrical strain gauges was prevented by covering the
gauges with a protective layer of paraffin. Strain release was recorded during the entire saw cutting
procedure. Strain measurements were recorded until approximately 30 minutes after the end of the
cutting. Strain measurements were converted into stress values by multiplying the strain by the Young’s
modulus as obtained from the tensile tests on the straight reference sections, thereby assuming elastic
release of the strains. Stress values on opposite sides of the flanges and webs were averaged.

Figure 4 Location of strain gauges

R.C. Spoorenberg et al.


4.1 Straight sections

The measured residual stresses of the straight reference sections are presented. The number of the
specimen is related to its roller bent counterparts (Figure 5). It can be seen that the stresses are usually
small and below 100 N/mm2. Similar observations for straight hot-rolled HE 100B sections were made in
200 50
100 25
0 0
-100 -50 0 50 -25 -50 0 50
-200 -50
50 50
1 Top fl. Top fl.
S235 S235
0 0

Bot. fl. Bot. fl.

-50 -50
200 50


100 25
0 0
-100 -50 0 50 -25 -50 0 50
-200 -50

50 50
25 25
0 0
-25 -50 0 50 -25 -50 0 50
-50 -50
50 50
Top fl. Top fl.
4 5-6
S355 S355
0 0

Bot. fl. Bot. fl.

-50 -50


25 25
0 0
-25 -50 0 50 -25 -50 0 50
-50 -50

Outside, left Inside, right Average

Figure 5 Hot rolled residual stresses in straight HE 100B reference sections in N/mm2

R.C. Spoorenberg et al.

4.2 Roller bent sections

A plot of the residual stresses of the roller bent specimens is shown in Figure 6. The top flange and
bottom flange are referred to as the elongated flange and compressed flange, respectively. The
theoretically obtained stresses for cold-bent sections as proposed by Timoshenko are shown alongside the
measured stresses by using the measured yield stress (Table 2) of the straight reference sections and the
shape factor on the basis of nominal section dimensions (Figure 1).

200 200 200

100 100 100
0 0 0
-100 -50 0 50 -100 -50 0 50 -100 -50 0 50
-200 -200 -200
50 50 50
Top fl. Top fl. Top fl.
1 2 3
S235 S235 S235
R=1910mm 0 R=2546mm 0 R=3820mm 0

Bot. fl. Bot. fl. Bot. fl.

-50 -50 -50
400 400


200 200 200

0 0 0
-50 0 50 -50 0 50 -50 0 50
-200 -200 -200

200 200 200

100 100 100
0 0 0
-100 -50 0 50 -100 -50 0 50 -100 -50 0 50
-200 -200 -200
50 50 50
Top fl. Top fl. Top fl.
4 5 6
S355 S355 S355
R=1910mm 0 R=2546mm 0 R=3820mm 0

Bot. fl. Bot. fl. Bot. fl.

-50 -50 -50
600 600 600



400 400 400

200 200 200
0 0 0
-200 -50 0 50 -200 -50 0 50 -200 -50 0 50

Outside, left Inside, right Average Theory

Figure 6 Residual stress distributions after roller bending of HE 100B sections in N/mm2.

R.C. Spoorenberg et al.

The residual stresses as shown in Figure 6 display the following characteristics. High tensile residual
stresses were observed at the web-to-flange junction in the bottom flange. The web of the roller bent
specimens is primarily under compression. Small residual stresses, either in compression or tension, were
found in the top flange. The results show a symmetrical stress pattern with respect to the minor axis of
The net effect of the roller bending process on the residual stresses can be observed by comparing
Figure 5 with Figure 6. It can be seen that the roller bending process results in a residual stress pattern
that is entirely different from the hot-rolled pattern prior to bending. In particular, the maximum residual
stress of the roller bent sections is much larger compared to the maximum residual stress of the straight
hot-rolled sections.
Bending a beam into a smaller radius requires an increase in cold work and hence an expected
increase in residual stress. However, it can be observed that the bending radius has small influence on the
residual stress distribution.
The effect of the steel grade on the residual stress distributions in roller bent sections for the HE 100B
series can be observed by comparing the results of specimens 1 to 3 (Figure 6) with those of specimens 4
to 6. The effect of the steel grade is significant since the residual stresses of the specimens 1 to 3 (S235)
are much smaller than those of specimens 4 to 6 (S355).
The measured residual stresses of the roller bent specimens are generally below the yield stress of the
straight material, although the yield stress was exceeded by the residual stress values in the bottom flange.
A higher residual stress than the yield stress can be expected as a result of cold working the material
during the roller bending process. The additional tensile tests on coupons taken from curved sections
support this theory as they showed an increased yield stress (Figure 3). Recent findings as reported in
[17] have shown a similar phenomenon for residual stresses in stainless steel sections.
The experimental results are quite different from the theoretically obtained residual stresses in cold-
bent sections as proposed in [12]. The theoretical residual stress distribution does not have a stress
gradient along the flange width, which was clearly observed in all experimental results. An antisymmetric
stress pattern with respect to the major axis of bending postulated by theory has not been measured.

This paper presents experimental results of residual stress measurements carried out on straight and
roller bent HE 100B sections. The results are part of a larger experimental and numerical study on
residual stresses due to roller bending of wide flange steel sections. The sectioning method in conjunction
with electrical strain gauges was employed to measure the residual stresses. The experimental results for
the roller bent specimens showed a significantly different residual stress distribution when compared to
the residual stress distributions in their straight counterparts. In the roller bent specimen, high
compressive and high tensile stresses were observed in the web and in the bottom flange respectively. The
observations clearly indicate that a hot rolled residual stress pattern is not applicable to a roller bent
specimen. Also it can be stated that the theoretical solution for residual stresses in cold-bent members as
proposed by Timoshenko is not suitable for roller bent wide flange steel sections. The large deviations
between the theory and the experiments show that the roller bending process of wide flange steel sections
cannot be simplified by a beam subjected to loading and unloading in uniform bending.

This research was carried out under the project number MC1.06262 in the framework of the Research
Program of the Materials innovation institute M2i (www.m2i.nl). The majority of the experiments were

R.C. Spoorenberg et al.

carried out by H.L.M Wijen and T.J. van de Loo in the Pieter van Musschenbroek Laboratory and their
help is greatly acknowledged. The specimens were supplied and bent by Deltastaal BV and Kersten
Europe BV respectively.

[1] Beedle L.S. and Tall L. "Basic Column Strength". Transactions of the ASCE, 127, 138-179, 1962.
[2] Mas E. and Massonet Ch. "Part prise par la belgique dans les recherches experimentales de la
convention europeenne des associations de la construction metallique sur le flambement centriques
des barres en acier doux". Acier-Stahl-Steel, 9, 393-400, 1966.
[3] Lay M.G. and Ward R. "Residual stresses in steel sections". Journal of the Australian Institute of
Steel Construction, 3(3), 2-21, 1969.
[4] Young B.W. "Residual stresses in hot rolled members". IABSE reports of the working commissions,
23, 25-38, 1975.
[5] ECCS. Manual on stability of steel structures - ECCS Committee 8 Stability. 1976.
[6] Ballio G. and Mazzolani F.M. Theory and Design of Steel Structures. Chapman & Hall, Oxford,
[7] Galambos T.V. Guide to Stability Design Criteria for Metal Structures. John Wiley & Sons inc.,
New York, 1998.
[8] Kato B. and Aoki H. "Residual Stresses in Cold-formed Tubes". Journal of Strain Analysis, 13(4),
193-204, 1978.
[9] Weng C.C. and White R.N. "Residual Stresses in Cold-bent Thick Steel Plates". Journal of
Structural Engineering ASCE, 116(1), 24-39, 1990.
[10] Weng C.C. and Pekoz T. "Residual Stresses in Cold-Formed Steel Members". Journal of Structural
Engineering ASCE, 116(6), 1611-1625, 1990.
[11] Tan Z., Li B., Persson B. "On Analysis and Measurement of Residual Stresses in the Bending of
Sheet Metals". International Journal of Mechanical Sciences, 36(5), 483-491, 1994.
[12] Timoshenko S.P. Strength of Materials. Part II Advanced Theory and Problems. D. Van Nostrand
Company, Inc., New York, 1940.
[13] King C. and Brown D. Design of Curved Steel. The Steel Construction Institute, Berkshire, 2001.
[14] Spoorenberg R.C., Snijder H.H., Hoenderkamp J.C.D. "Experimental investigation of residual
stresses in roller bent wide flange steel sections". Journal of Constructional Steel Research, 2010
doi: 10.1016/j.jcsr.2010.01.017.
[15] Bjorhovde R. "Cold Bending of Wide-Flange Shapes for Construction". Engineering Journal,
43(4), 271-286, 2006.
[16] Daddi I. and Mazzolani F.M. Determinazione sperimentale delle imperfezioni strutturali nei
profilati di acciaio. Universita degli studi di Napoli istituto di tecnica delle costruzioni, Bari, 1971.
[17] Cruise R.B. and Gardner L. "Residual stress analysis of structural stainless steel sections". Journal
of Constructional Steel Research, 64(3), 352-366, 2008.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Richard Stroetmann


Keywords: Lateral torsional buckling, space structures, couple effects, calculation methods.

Abstract. The structural behavior of space structures with ,-beams subject of lateral torsional buckling
can often be described only insufficiently by plane subsystems. This is caused e. g. by couple effects of
stabilizing and destabilizing beams or transmissions of rotations and displacements from cross girders at
the connecting points. By means of special finite elements, which are designed for ,-beams in space
structures, an efficient calculation will be possible. Beam systems with cross-connected structural mem-
bers, like cross girders or trapezoid sheets will allow a derivation of approximate solutions for standard
applications. The intended paper deals with calculation methods for space structures with ,-beams and
demonstrates structural behavior as well as applications on the basis of typical examples.

Methods for analyzing space structures with beams subject to lateral torsional buckling are mainly re-
stricted to plane systems. Usually, the effect of adjacent structural elements is taken into account by
definition of bearings, discrete and continuous translational or rotational restraints. In many cases these
simplifications describe the structural behavior of such beams with sufficient accuracy.
Problems may occur when adjacent members do not have a stabilizing, but a destabilizing effect. Of-
ten deformations (displacements, angular rotations) are transferred to the beams to be stabilized. In con-
sequence, bearings, discrete and continuous restraints cannot describe the stiffening effect entirely. The
following examples will document the difficulties mentioned above.
Figure 1 shows two transversely loaded beams that are connected by a lattice bar at the midspan. In
case of lateral torsional buckling with different loading conditions qz,1 and qz,2 stabilizing forces will be
transmitted by the joining member. The more loaded beam will be restrained by the less loaded beam.
The interaction can only be recorded considering the entire global system.

plan view section

qz,1 qz,2

y y

x z

Figure 1: Lateral torsional buckling of cross-connected ,-beams

Richard Stroetmann

plan view section

longitudinal girder


cross girder

y y

x z

Figure 2: Crossing ,-beams with uniformly distributed load

The system in Figure 2 shows a cross girder with uniformly distributed load that is supported by a
longitudinal girder. The deformation of both girders at the connecting point can be assumed to be identi-
cal. The angular rotation at the end of the cross girder and the vertical deflection of the longitudinal
girder result in a three-dimensional deformation shape. In the longitudinal girder bending moments Mz
and torsional action effects will occur. They depend on the load level, the stiffness ratio and the dimen-
sions of the system. Connections without stiffeners lead to additional distortions of the cross-section in
the area of the connecting point.
Figure 3 shows a roof structure consisting of purlins, trusses and a roof bracing. Lateral displace-
ments of the trusses and the roof bracing are linked by means of the purlins. On the one hand the bracing
has a stabilizing effect on the trusses. On the other hand deformation, caused e. g. by wind loads, is trans-
ferred into the trusses. This causes deflecting forces in the trusses that lead to additional loadings in the
bracing. The assumption that purlins connected to the roof bracing act as rigid supports for the top flange
of the trusses is inaccurate. Especially when diagonal bracings are realized as round steel tension bars and
the bracings have to span large distances the influence of the deformation may become important.
The preceding examples demonstrate that the structural behavior of space structures with I-beams
subject to lateral torsional buckling can often only insufficiently be described by a division into plane
subsystems. Although known there is a lack of practical calculation tools that consider the interaction of
the involved components with reasonable effort.
plan view



Figure 3: Roof structure with vertical and horizontal loads


Applying the finite element method the structural system is idealized by the arrangement of structural
finite elements. Suitable elements must be available to describe the fundamental mechanical properties of
the members and their connections. Multifarious structures can be modeled by continuum elements such

Richard Stroetmann

as membrane, solid or shell elements. Decisive disadvantages of the analysis of ,-beam structures subject
to lateral torsional buckling are:
x The input of the structural systems including the definition of load and boundary conditions as well
as geometrical imperfections such as initial bow or sway imperfections is very complex and time con-
x Generating the numerical model requires a comparatively large number of elements. The resulting
meshes have a large number of nodes. To calculate the response of the structure a large system of
equations needs to be solved.
x The analysis of the huge data output requires a graphical post processor. The transparency of results
is partially lost since action effects are only presented as stresses, bending moments and in-plane
forces per unit length.

In practice such calculations remain restricted to special cases, e. g. when manufacturing a large num-
ber of identical girders justifies these efforts. However, it is more efficient to use “macro-elements” with
modified mechanical properties that represent the essential properties of the components to be modeled.
The effort needed for modeling, calculation and analysis can be minimized by using such elements, so
that FEM calculations can more widely be used for applications as described here.
Within the framework of the research project [1] tools for the calculation of structural systems with ,-
beams subject to lateral torsional buckling were developed based on the finite element method. Among
other aspects the research aimed at providing a design tool for space structures and thus to remove limita-
tions to plane systems such as single span or continuous I-beams and I-columns. It was intended to find a
possibility to describe the impact of adjacent structural members directly by means of suitable elements
and not only indirectly through the definition of discrete or continuous restraints and bearings. In addi-
tion, it was important to eliminate the assumption that there no change in cross-section shape during
buckling, because some section designs with thin webs and support conditions of beam flanges require
that web distortions need to be taken into account. The concept was to limit the mechanical properties of
the elements that are essential for the modeling of members and connections. Effects such as transverse
shear or membrane strains of ,-beams were neglected. In a first step, the development of finite elements
was limited to linear-elastic material properties and linearized distortion-translation relationship (second
order analysis).
For the modeling of beams and columns with double-symmetric ,-sections a “macro-“ or “super ele-
ment” composed of a group of element types (figure 4) was developed. The structural behavior in the
plane of web is idealized by a beam element. Perpendicular to the plane of web the ,-section consists of
two flange members and a web plate.
ˆ uˆ x,
ˆ uˆ
internal forces and moments

ˆ vˆ ˆ wˆ

ˆ wˆ
element loads continuous restraints
qz,o cxx,o
qy,o cy,o tfl czz,o
mx,o s s

qz,m r, f r tw
s, f s cxx,u

qy,u cy,u tfl czz,u

mx,u s s

Figure 4: ,-Profile-Element – reference system, loads, bedding, internal forces and moments

Richard Stroetmann

The splitting of ,-sections into different elements has the advantage that web distortions can be de-
scribed with two-dimensional buckling shapes. The separation of the cross section eased the application
of eccentric forces and moments, the modeling of practical supporting conditions and the attachment of
further structural elements to the flanges of the beams and columns. A frequently discussed issue in the
beam theory is how to model geometric and/or static coupling conditions (keyword transmission of warp-
ing deformations) especially in cases when the beams are attached perpendicularly or in a random angle
to each other or if they consist of stepped cross sections. The modeling with this type of element elimi-
nates such kind of problems.
Compared to the modeling of ,-profiles with shell elements the advantage of the proposed concept is
that a relatively small number of elements is required for discretization. This leads to a significant reduc-
tion in computing time. Generating the model is very practicable and the numerical results are easier to
analyze and interpret. The influence of cross-section fillets in rolled sections that significantly increase
torsional stiffness can easily be defined by means of modifying the stiffness values. Modeling with solid
elements requires a large number of elements, thus further increasing the effort for calculation and
evaluation of the results.
Besides the ,-profile element additional elements were developed. A stiffening element can be used
for the modeling of stiffeners at load applications and of end plates; a trapezoidal profile element, a beam
element for the modeling of bracings and lattice bars as well as different spring elements for the consid-
eration of connection flexibility.


ˆ uˆ
x, ˆ uˆ
ˆ vˆ
y, ˆ vˆ
y, cxx
ˆ wˆ
z, ˆ wˆ
z, q

ˆ uˆ
ˆ uˆ
ˆ vˆ
y, ˆ wˆ
ˆ wˆ

Figure 5: Assembly of the finite elements for trapezoidal sheeting, connection springs and ,-beams



3.1 Introduction to and survey of calculation methods

In practice the level of utilization of the single beams in structural systems differs often significantly.
This is for example the case, when beams with various loads for practical reasons are designed with the
same cross-section. Due to varying live loads the level of utilization can be different at a certain point of
If different loaded beams are cross-connected, they act together at lateral torsional buckling. The less
stressed beams will restrain the more stressed ones (figure 1).
In steel structures often girders with cross connections are used. Trusses, for example, can be coupled
by purlins or secondary girders. Roof, wall and ceiling coverings provide a more or less continuous cou-

Richard Stroetmann

pling of beams or columns. Due to the structural detailing diaphragm actions cannot always be taken into
The following description will briefly present different methods of calculating buckling loads of lat-
eral coupled beam systems. Knowing the ideal buckling load a simplified verification of buckling resis-
tance, e.g. according to EN 1993-1-1 [3], is possible.
One option to perform the structural analysis of coupled beam systems is to derive stiffness matrices
of the beams separately (based on second order analysis) and use suitable coupling conditions to obtain
the matrix formulation for the overall structure. Rigid couplings can be described by kinematic con-
straints and semi-rigid ones by coupling matrices. Connections can be at discrete points or in closely
spaced intervals along the beams. The influence of constraining effects against twist rotations at the
connecting areas can be considered by discrete or continuous torsional restraints.
For simple structures approximation formulas and diagrams can be derived to determine the buckling
load of coupled beams. Moreover, making use of programs for the calculation of single-span and con-
tinuous beams the lateral torsional buckling load of rigid coupled beam systems may by determined
according to second order analysis of lateral torsional buckling and non-linear spring characteristics of
the beams. A detailed description of the methods is given in [4].
3.2 Approximation formulas for rigid coupled ,-beams
In case of simply supported beams with uniform distributed and single loads the assumption of buck-
ling shapes in the form of half sinus waves for twist rotations - and lateral displacements vM of shear
centre axis leads to an acceptable approximation for the buckling load. When the beams are rigidly cou-
pled, the kinematic constraints provide the transformation rules of the stiffness matrices. In case of two
coupled beams the degrees of freedom will be reduced from four to three. The assembly of beam stiffness
matrices to the global stiffness matrix and the derivation of the determinantal equation result in a charac-
teristic cubic polynomial with three eigenvalue. A closed-form solution is possible.
The approximation formulas given in figure 6 provide buckling load values that in most cases deviate
by less than 5 % compared to calculations using more significant buckling shapes. Systems with a high
stiffness of discrete and continuous torsional restraints show larger differences. If four, six or more

a0 2 ˜ c1 ˜ c2 ˜ ( f 2 ˜ c1  c2 )
a0  a1 ˜K  a2 ˜K2  a3 ˜K3 0 o KKi
a1 2 ˜ c1 ˜ c2 ˜ [ f ˜ ( g1  g3 )  g2  g4 ]  ( f ˜ c1 )2 ˜ ( g2  g4 )

qz a2 ( f 2 ˜ c1  c2 ) ˜ ( g12  g32 )  2 ˜ c1 ˜ [ f ˜ ( g1 ˜ g4  g2 ˜ g3 )  g2 ˜ g4 ]

a3 2 ˜ f ˜ g1 ˜ g3 ˜ ( g1  g3 )  g12 ˜ g4  g2 ˜ g32
qz ,i ˜ L2 Fz ,i ˜ L 1 2 2
M q ,i M F ,i rM y ³ ( y  z ) ˜ z ˜ dA  2 ˜ z M
8 4 Iy A
EIz ˜ S 4 ECM ˜ S 4 GIT ˜ S 2 L
L/2 L/2 c1 c2   c- ˜  C- f ZK  ZM
2 ˜ L3 2 ˜ L3 2˜ L 2
z Mq,1 S2 MF ,1 S2
g1  ˜(1 ) ˜(1 )
qz,1 qz,2 L 3 L 4
c- c-
Mq,1 ˜ rM y S2 qz ,1 ˜ L MF ,1 ˜ rM y S2
g2  ˜( 1  ) ˜ ( zq,1  zM )  ˜(1 )  Fz ,1 ˜ ( zF ,1  zM )
L 3 2 L 4

y S S Mq,2 S2 MF ,2 S2
ZK ZM g3  ˜( 1 ) ˜( 1 )
M M L 3 L 4

EA=f Mq,2 ˜ rM y S2 qz ,2 ˜ L MF ,2 ˜ rM y S2
Fz,1 Fz,2 g4  ˜(1 ) ˜ ( zq,2  zM )  ˜( 1 )  Fz ,2 ˜ ( zF ,2  zM )
L 3 2 L 4

Figure 6: Approximation formulas for calculation of cross-connected Beams

Richard Stroetmann

beams are rigidly coupled and respectively half of them have the same load intensity the buckling load of
these systems can obviously be calculated as well using the formulas given in figure 6.
3.3 Diagrams for calculation of lateral torsional buckling moments
Using the program PROFIL [2] diagrams were created for the determination of lateral torsional buck-
ling moments of girder systems with discrete and continuous rigid couplings (see figure 7). The diagrams
are valid for systems of simply supported rolled ,-beams. The double-symmetric cross sections of the
respective I-beams are identical. The application points of the uniformly distributed loads and the level of
couplings are placed to the centroid from the top flanges of the beams. At the determination of the dia-
grams web distortions were excluded. The application is as follows:
The number of coupled beams determines the type of diagram to be used. Curve parameters are given by
the moment distribution My and the load relation q2/q1. Depending on torsion coefficient F (see equation
(1)) the diagram provides the coefficient k to determine the lateral torsional buckling moment of the
whole system (equation (2)). The value My,Ki refers to the total maximum moment My, that means either
the span or support moment. Lateral torsional buckling moments of single beams j will be determined
with equation (3) by the relation of beam load and system load.
ǼI W (1)
Gǿ ȉ ˜ L2

k (2)
M y ,Ki ˜ GIT ˜ EI z

18 q2 26 q2
q1 q1
q1 q2
k q2 q1 q2
k -M q 1
24 +
-M q
- 1
16 + 5

L 22 L f
Mq = q·L² / 8 5 M q = q·L² / 8 0,2
0 0,1

18 0
12 0,5 1
1 5
0,5 f
0,25 0,1
0,1 f
16 0
10,1 0,1 0,2
10 14,5

8,7 12,8

8,1 12
8 11,7

7,4 10,9
-0,67· M q
- -
6,8 + 10
6,3 9,1
6 8,6 -0,67· M q
- -
5,5 +
5,1 7,5
Mq + 7,3 Mq +
4 6
0 0,1 0,2 0,3 0,4 0,5 0 0,1 0,2 0,3 0,4 0,5
a) F b) F

Figure 7: Coefficients k for the calculation of lateral torsional buckling moments

Richard Stroetmann

qj (3)
M y , j ,Ki M y ,Ki ˜

The coefficient k is nearly independent on section series and size. The diagrams in figure 7 are based
on minimum values of the IPE series. In good approximation the application is possible for other rolled
beams with doubly symmetric ,-section. In the diagrams for three coupled beams it has to be considered
that the coefficient k increases from load relationships of zero to one and thereafter k decreases again.
Under certain conditions the diagrams may also be used for determining lateral torsional buckling mo-
ments of systems of more than three beams. This requires that only two different values of transverse
load are present and that the relation between the number of beams with the same load is n1:n2 = 1:1 or

In order to estimate the buckling loads of discrete coupled beam systems an iterative determination
can be performed. The aim of such determination is to find the load level KKi at which the sum of stabiliz-
ing and destabilizing forces that are transferred to the coupling beams reach equilibrium. Figure 8 dem-
onstrates the principle by an example. Three differently loaded beams are rigidly connected at midspan to
the top flange. If the spring characteristic Cy for lateral displacement of the I-beams at the connecting
point is known from a previous calculation performed according to second order analysis the iteration can
be carried out by using the diagram presented.
Figure 9 shows a girder grillage consisting of purlins and trusses, respectively with same cross-
section, on which a uniformly distributed load of q=5.00 kN/m2 acts. Lateral torsional buckling of the
purlins is prevented by restraints, e. g. due to trapezoidal sheeting. Caused by the structural system, the
load transfer of the inner trusses is approximately 2.75 times higher than that of the outer trusses. By the
connection with the purlins the trusses are torsionally restrained and coupled in transverse direction at the
top flanges.

Cy [kN/m]

3000 3000
qz,1 = 2,0 qz,2 = 5,0 qz,3 = 10,0

q [kN/m]
(1) (2) 290 x 7 (3) 0 10 20 30 z 40
150 x 10
Cy  C y ,B 1
qz qz
Cy,B 1 2
K Ki = 2,44
1 y,v Beam 3
S 3
Cy,B Beam 2
K Beam 1
3000 3000
Figure 8: Determination of buckling load with spring characteristic (loads [kN/m], dimensions [mm])

Richard Stroetmann

The stability of the structural system was calculated by the program PROFIL [2] and various ap-
proximation methods as documented in [4]. Moreover, various effects were investigated by means of
different calculations. Disregarding the coupling effect to the outer trusses and the torsional restraints
from the purlins the buckling load factor of the inner trusses is KKi=0.39. With torsional restraints this
factor increases to KKi=1.24. Additionally, the consideration of the coupling effect to the outer trusses
results in a buckling load factor of KKi=1.71. In this particular case it is necessary to consider both effects,
to verify the structural safety of the inner trusses.

6.000 6.000 6.000 q=5,00 kN/m2


IPE600 IPE600

8 1,71
purlins IPE220
trusses IPE600

4 without 1,24
2 torsional
restraints with with
and couple torsional torsional
effects restraints restraints
6 but and
0,39 without couple
couple effects
1 2 3
safety factor of the ideal buckling load

Figure 9: Girder grillage with trusses and purlins – influence of different stabilization effects

The stability of space structures is often only insufficiently assessed by the analysis of plane subsys-
tems. Specific finite element formulations allow system analyses that include essential effects with regard
to the overall structural behavior. Besides the possibility to consider the transmission of deformation of
adjacent structural members, stabilizing forces of bracing systems can be directly determined. By restric-
tion to the effects that are essential for the structural behavior the efforts for modeling, calculation and
interpretation are minimized. In this way, the finite element method can economically be applied to space
structures in practice.

[1] Stroetmann, R., “Zur Stabilitätsberechnung räumlicher Tragsysteme mit I-Profilen nach der Me-
thode der finiten Elemente“, Veröffentlichung des Instituts für Stahlbau und Werkstoffmechanik
der Technischen Universität Darmstadt, Heft 61, 1999.
[2] Stroetmann, R., “PROFIL  FEM-Program for the structural analysis of space structures with I-
Profiles“, Fachgebiet Stahlbau und Werkstoffmechanik, TU-Darmstadt, 1999.
[3] EN 1993-1-1: “Eurocode 3 – Design of steel structures, Part 1-1 – General rules and rules for
buildings“, 2005.
[4] Stroetmann R., “Zur Stabilität von in Querrichtung gekoppelten Biegeträgern“, Der Stahlbau
69, Verlag Ernst & Sohn, Berlin, 391-408, 2000.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Andreas Taras*, Richard Greiner*

* Graz University of Technology – Institute for Steel Structures and Shell Structures
e-mails: taras@TUGraz.at, r.greiner@TUGraz.at

Keywords: Beam-Columns, In-Plane Buckling, Interaction Factors, Generalized Slenderness.

Abstract. This paper presents a new formulation for the design of beam-columns against in-plane
buckling that makes use of an -increasingly popular- generalized slenderness definition and an “overall”
formulation of the buckling reduction factor for combined load cases. Thereby, great care is placed on
accurately describing the specific behavior of each studied cross-sectional type. The result is a
“generalized slenderness” formulation that is as accurate, safe and mechanically consistent as the
familiar and thoroughly studied interaction-concept formulae.

Beam-columns are characterized by the presence of compressive axial forces N and bending
moments M. The resistance of a steel member against either N or M is commonly determined in design
codes by the use of buckling reduction factors F=f( O ), whereby the plot of the function F over the
normalized slenderness O is a so-called buckling curve. The (usually) detrimental effect of the combined
action of N and M is taken into account in design codes by formulae that are based on one of the
following two concepts, see figure 1:
- The interaction concept, found e.g. in clause 6.3.3 of Eurocode 3 [1], makes use of the
information contained in the utilizations nFB and mLT of the buckling checks for flexural
buckling under N alone and LT-buckling under M alone. The combined effect of N and M is
then taken into account by an interaction factor k.
N ("overall slenderness"; "general method"; ...)
My N N
My N N

N N geometric residual
imperfections stresses
OGS h/b>1.2
O oF OLT o FLT 0.3 fy - + -
Ncr Mcr Rpl e0
Rcr h/b<1.2
NRk Npl ˜ F MRk Mpl ˜ FLT
0.5 fy


"mLT " Rk Rpl ˜ FGS FGS ?
N Rpl
M k˜ d 1.0 Rk
NRk / JM1 MRk / JM1 t 1.0
Figure 1: Concepts for beam-column design

Andreas Taras and Richard Greiner

- In the second group of concepts, a generalized definition of the (normalized) slenderness is used;
they are therefore called generalized slenderness concepts in the following. Specifically, these
concepts encompass the “overall load case method” used for the design of plates and shells (see
e.g. [2]) and the so-called “general method” for the design of beam-columns of clause 6.3.4 in
Eurocode 3. These methods have in common that they consider total utilizations for the
combined case as basis for the calculation of the normalized slenderness and of the (overall)
buckling reduction factor. As is illustrated in figure 1, the slenderness O GS is defined in a
generalized form as the square root of the total load proportionality factor LPF for the plastic
collapse load (LPFMNA=Rpl) over the pertinent buckling eigenvalue LPFLBA=Rcr. The buckling
strength is then defined as follows:
FGS ˜ R pl
Rb,d t 1.0 (1)
Thereby, Rb,d is the design buckling resistance (in terms of maximum LPF) of the component or
structure against the studied buckling mode for a given load combination.
Even though the current debate over these two concepts might seem to indicate otherwise, the
concepts are best thought of as two different forms of representation of the same information, with no
basis for attributing an (inexistent) higher degree of mechanical consistency to any of the two. In the case
of the interaction concept as found in the Eurocode, mechanical accuracy and safety/reliability have been
ensured by extensive theoretical, numerical and statistical studies, summarized in [3]. On the other hand,
one could argue that the “generalized slenderness” formulation according to (1) is more “consistent” with
the design checks for the single load cases, in the sense that it also implicitly contains the buckling
checks used for only N or only M. (In the case of the “general method” this is only true for M, since for
N it is based on Fy.Npl instead of on Npl alone). However, the reduction factor FGS must account for the
exact same effects as the interaction factor k. As is indicated by the question mark in figure 1, the values
to be adopted for FGS are not clear and still up for debate, with a common opinion being that they must be
studied and calibrated by means of GMNIA calculations, see e.g. [4].
The following figure 2 illustrates factors FGS obtained from such GMNIA calculations for the in-
plane buckling behavior of a beam-column under N+M with uniform moment diagram and different
values of the ratio K0=(M/Mpl)/(N/Npl)=m0/n0. Two different definitions of F and O are used. Figure 2a
makes use of Fip and O ip, which are based on the definitions of Rpl and Rcr valid for the overall load case
N+M and in-plane buckling behavior.
1.0 1.0
IPE 500 K0=m0/n0 GMNIA
1 1
2 M K0=0.25 2
Oip My/N=const. Oy
0.8 0.8

0.3 fy
=Nb/Npl [-]

- + - 0.6 K =1.00
F =R /R [-]

b,ip pl

0.4 0.4 K0=2.00
y, K 0


K0=1.00 K0=4.00
0.2 K0=4.00 0.2

0.0 0.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
O ip a) Oy b)

Figure 2: GMNIA buckling reduction factors Fip (a) and Fy,K0 (b) for in-plane buckling of an IPE 500.

Andreas Taras and Richard Greiner

In figure 2b, the familiar definition of the in-plane flexural buckling slenderness O y is used, as well
as a representation of the buckling strength for the combined, proportional load case solely in terms of
the achievable axial load Nb. Both types of representation yield the same curve for the case of K0=0.0,
representing the (imperfect) column, and result in distinctly different curves for varying values of K0.
As was mentioned above, any application of a generalized slenderness concept requires a
definition/formulation of the buckling reduction factor FGS that reproduces the same type of information
contained in an interaction factor k, thereby achieving high accuracy when compared to more
sophisticated GMNIA calculations. This paper presents an analytical formulation for the buckling
reduction factors Fip and Fy,K0 (i.e. FGS for the in-plane buckling phenomenon of beam-columns) that
fulfils this requirement.


This section presents the mentioned proposal for a “generalized slenderness concept” formulation for
in-plane beam-column buckling design, focusing on compact class 1 or 2 sections. The full derivation is
too lengthy to be included in this paper; thus, the reader is referred to the original source in [5]. However,
the basic concepts of the derivation are briefly discussed in the following, making reference to figure 3.
The first and essential step of the derivation consists of determining the generalized slenderness for the
in-plane buckling case under N+M. Thereby, it is found convenient to do so on the basis of a
linearization -with i=1, 2, … linear segments- of the cross-sectional interaction curve, see figure 3a. By
introducing the parameters kni and kmi, the cross-sectional resistance of a section in terms of obtainable
values of m=M/Mpl and n=N/Npl can be described as follows, for the applicable segment i:
kni ˜ n  kmi ˜ m d 1.0 (2)
Table 1 contains coefficients kni and kmi that were derived for a variety of double-symmetric sections.
Thereby, the factors KSCi indicate the ratios K=m/n at which the applicable linearized segment of the N/M
cross-sectional interaction curve changes; e.g., if K<KSC1  o kni=kn1; kmi=km1.
1 2
Rpl Rcr 1/ §¨ n0 ˜ O y ·¸
n0 ˜ k n1  K0 ˜ km1 © ¹

Oip Oy ˜
kn1  K0 ˜ km1
1.6 1.0
failure criterion:
1.4 kni * ˜F y,K0  k mi * ˜mII 1.0
1.2 1/kn1*
strong-axis buckling =1.0
1.0 N+My
K0=m0/n0 0.6
n=N/Npl [-]

1/kn2 1/kn2*

0.8 mII

0.4 m0 Oip

m0 1/kn1=1
0.4 F y,K0
1/km1 0.2
K0 n0 1/km2*
0.2 Ksc1
n0 1/km2=1
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 0.0 0.2 0.4 0.6 0.8 1.0
m=M/M pl [-] a) m=M/Mpl [-] b)

Figure 3: Linearization of the cross-sectional interaction and generalized slenderness definition (a);
definition of a second-order failure criterion (b).

Andreas Taras and Richard Greiner

Table 1: Ccoefficients used for the description of the cross-sectional N+M interaction behaviour.
Type of section, loading, underlying
# Parameters of the N-M interaction linearization J
residual stress distributions

I-section, strong
1 axis buckling 0.8 0.8
- + - kn1=1.0; km1= 1  0.5 ˜ a t 0.75 ; KSC1
N+My 1  0.8 ˜ k m1
0.3-0.5 fy
Rectangular - -0.2
kn2= ; km2=1.0; KSC2 f
2 hollow section 1  0.8 ˜ k m1 0.4
RHS, N+My -
+ + +0.5 fy

Circular Hollow kn1=1.0 ; km1=0.74; KSC1=0.8/(1-0.8 km1)=1.95

3 Section CHS, + + 0.6
- - kn2=0.2/(1-0.8 km1) =0.49; km2=1.0; KSC 2 f
N+M 0.15 fy
kn1=1.0; km1=(1-a)/1.45; KSC1=0.8/(1-0.8 km1)
- + -

I-section, weak
4 axis buckling kn2=0.8/(1.81-a) ; km2=(1-a)/(1-0.55a); KSC2=1/a 0.6
0.3-0.5 fy
N+Mz kn3=0.0; km3=1.0; KSC3 f
A  2 ˜ b ˜ tf Aw

Since the “generalized slenderness” concept operates with load amplification factors R, a reference
load level m0/n0 is proportionally increased in the following, meaning that the ratio K0=m0/n0=K0 is kept
constant. Taking this into account, the following expressions can be found for the cross-sectional (plastic)
amplification factor Rpl and the buckling eigenvalue Rcr pertaining to the in-plane mode:
1 K0
R pl (3)
n0 ˜ c0 m0 ˜ c0
Rcr 2
n0 ˜ O j
With c0 k ni  K0 ˜ k mi ; O j N pl / N cr, j ; j = axis y or z, depending on the case. The
generalized in-plane buckling slenderness can now be written as
O ip O j / c0 (5)
The next step consists of a definition of a (second-order) failure criterion. The basic concept behind
the adopted criterion is illustrated in figure 3b: the buckling load of the member is reached when at one
cross-section the following condition is fulfilled:
kni * ˜  kmi * ˜ 1.0 (5)
N pl M pl
Thereby, MII is the total, second-order bending moment in the critical cross-section at failure, while
kni* and kmi* are factors derived –once again- from a linearization of the cross-sectional interaction
diagram (see kni and kmi), but taking into account the transition from the applicability of the plastic and
elastic cross-sectional interaction curve with increasing slenderness. This transitional behavior, discussed
e.g. in [3] and [5], is caused by the detrimental effect of extreme-fiber (e.g. flange) yielding on the
obtainable buckling strength observed in tests or realistic GMNIA calculations. Accordingly, the
following expressions (6) and (7) for kni* and kmi* reproduce a transition from the values kni and kmi valid
for the plastic cross-sectional capacity to the values of kni*=1.0 and kmi*=Wpl/Wel=w applicable for the
elastic cross-sectional resistance at higher slenderness.

Andreas Taras and Richard Greiner

kni* kni  ( 1  kni ) ˜ U ˜ O ip ˜ CmS d 1 (6)

kmi* kmi  ( w  kmi ) ˜ U ˜ O ip ˜ CmS d w (7)
The factor J in (6) and (7) is a numerically calibrated, section-specific value that accounts for the
sensitivity to extreme-fiber yielding and is given in table 1 for the different studied types of cross-section.
CmS is the equivalent, sinusoidal moment coefficient for in-plane buckling.
For an imperfect beam-column with sinusoidal geometric imperfection of amplitude e0 and an in-
plane bending moment diagram, the total, second-order bending moment MII in equation (5) can be
calculated as follows:
CmS ˜ M  N ˜ e0 ˜
1  N / N cr

The use of (8) in (5) leads to

kni * ˜
 kmi * ˜
CmS ˜ M  N ˜ e0

1.0 (9)
N pl M pl 1  N / N cr
The next step, explained in detail in [5], consists of replacing the geometric imperfection amplitude
e0 by the generalized, Ayrton-Perry imperfection amplitude that leads to the EC3 column buckling
curves for the studied cross-section and the case where m0 = K0 = 0.0.
M pl
N pl ˜ kmi *
˜Kimp ; Kimp D ˜ O  0.2 (10)

The introduction of the normalized terms Fj,K0=N/Npl and O j N pl / N cr , as well as some

simplifying and re-writing, then leads to the following expression:
F j,K 0

kni * ˜F j ,K 0  kmi * ˜CmS ˜K0  Kimp ˜ 2
1.0 (11)
1  F j,K 0 ˜ O j
Expression (11) is mathematically equivalent to the Ayrton-Perry type failure criterion used by
Maquoi & Rondal [6] to establish and calibrate the well-known column buckling formulae in the
Eurocode; this statement is actually only true if the equivalent sinusoidal moment coefficient CmS is
expressed independently of the level of the (yet unknown) axial force N at failure. In [5], it is shown how
this can be achieved by a very accurate approximation, thereby simply replacing all terms N/Ncr in
published expressions for CmS by O j ²/(c0+ O j ²). By doing so, (11) can be solved explicitly for Fj,K0 ,
leading to the following, familiar-looking expression:
F j,K 0 d 1.0 (12)
) ip  ) ip 2  kni * ˜O j
1 § 2
with ) ip ˜ ¨ kni * Ktot  O j ·¸ ; Ktot kmi * ˜CmS ˜K0  Kimp (13)
2 © ¹
Expressions (12) and (13) allow for a calculation of buckling reduction factors Fj,K0 as exemplified in
figure 2b. The “overall”, in-plane buckling reduction factor Fip ,shown in figure 2a, can also be be
expressed equivalently, as a function of Oip , by considering (5) and the following relationship:
Rb,ip F y,K 0 / no
Fip F y,K 0 ˜ c0 (14)
R pl 1 / n0 ˜ c0
This finally leads to the explicit buckling design formula for in-plane beam-column buckling using
the “overall” reduction factor Fip:
Fip d 1.0 (15)
2 2
) ip  ) ip  kni * ˜O ip ˜ c0

Andreas Taras and Richard Greiner


In this section, the developed expressions (15) and (12) are compared to results of GMNIA
calculations, whereby the latter use the common assumptions for geometry and imperfections already
used by Beer & Schulz [7] during their elaboration of the numerical & theoretical foundation of the
current Eurocode column buckling curves. This ensures consistency with this essential benchmark case.
The first type of comparison is shown in figure 4. In this figure, the evaluations of (15) and (12) are
shown in the same type of representation already used for figure 2. Figure 4a shows the “overall”
reduction factor Fip, while figure 4b illustrates the reduction factor Fy.K0 for the obtainable axial load, both
for a circular hollow section, proportional loading and different values of K0. The figure shows that the
differences between numerical and analytical curves are very small, and approximately equal for the pure
column buckling case (K0=0.0) and the actual beam-column cases (K00.0). It can be shown that the type
of representation used in figure 4a convergences to a lower-bound curve, while the value of Fy,K0 in
fugure 4b of course tends towards zero with increasing values of K0. Both phenomena are very well
represented by the proposed formulation.
1.0 1.0
K0=0.00 K0=m0/n0 GMNIA
1 CHS178/10 1
2 M 2 EQU
Oip Oy
0.8 0.8

0.15 fy
=Nb/Npl [-]

0.6 K0=1.00
F =R /R [-]

b,ip pl

K0=0.00 K0=2.00
0.4 0.4
y, K 0


K0=2.00 K0=4.00
0.2 0.2

0.0 0.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
a) Oy d) b)
O ip

Figure 4: Comparison of newly developed analytical and numerical GMNIA buckling reduction factors
Fip (a) and Fy,K0 (b) for a CHS section.
A different type of comparison is shown in figure 5. In this figure, a spectrum of obtainable
combinations of the external loads N+M is plotted for two sections loaded by N + My or Mz, whereby
three different member lengths and a uniform moment diagram are considered. Figures a) and b) compare
the GMNIA results with the current Eurocode “interaction concept” formulae of clause 6.3.3, see [1] and
[3]. Figures c) and d) compare the numerical results with the evaluation of the newly developed “overall”
expression (15). The accuracy of both the two interaction-concept formulae found in the Eurocode
(Annex A and B) and the new formula can generally be said to be excellent.
Some advantages of the new formula can be mentioned here: the fact that expression (15) builds upon
an accurate (linearized) description of the actual, section-specific cross-sectional N+M interaction
diagram causes it to be “automatically” accurate at very low slenderness, and to lead to a consistent and
“correct” transition from the buckling to the (plastic) cross-sectional design check. The latter point is not
the case for the two sets of Eurocode interaction factors, which make use of approximations of the cross-
sectional interaction for the buckling check and usually don’t lead to the exact interaction at zero
slenderness. Another advantage of (15) appears in cases with non-equal end-moments, where failure can
be dominated by the cross-sectional check at one of the ends instead of by proper buckling. In the
interaction concept, this must be checked specifically, while (15) simply “includes” this check by being
limited by 1.0.

Andreas Taras and Richard Greiner

1.0 1.0 N/M

N N Oz=0.5 HEA 500z
M M IPE 240 z CS
M era Oz=1.0 M
M K0=m0/n0
CS c tion

- + -
-in Oz=1.5
0.8 Oy=0.5 0.8
t er
ac Oy=1.0
n Oy=1.5 0.3 fy
- + - 0.3 fy
0.6 0.6
n=N/Npl [-]

n=N/Npl [-]
0.4 0.4

0.2 0.2
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
m=M/Mpl [-] a) m=M/Mpl [-] b)
1.0 1.0 N/M
IPE 240 z C Oz=0.5 HEA 500z
N/ line S-in M
M M ariz tera Oz=1.0
lin C atio ctio

- + -
ea S- n n Oz=1.5
0.8 riz inte Oy=0.5 0.8
ion rac Oy=1.0
n Oy=1.5 0.3 fy
- + - 0.3 fy
0.6 0.6
n=N/Npl [-]

n=N/Npl [-]

0.4 0.4

0.2 0.2
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
m=M/Mpl [-] c) m=M/Mpl [-] d)

Figure 5: N+M buckling interaction curves according to Annex A & B of Eurocode 3 (a-b) and the
analytical formulation (c-d), compared with GMNIA results

In the final figure 6, the numerical GMNIA results, code formulae and the evaluation results of (15)
are compared on the level of interaction factors kj, computed so that they fulfill the following equation:
nj  kj 1.0 ; nj (16)
M pl F j ˜ N pl
The top three diagrams in figure 6 compare the values of kj (kyy or kzz) as defined in Annex A and B
of the Eurocode with GMNIA results, while in the bottom three diagrams GMNIA results are compared
with (iteratively determined) results of (15) that fulfill (16). Again, the proposed formulation follows the
GMNIA values of kj quite well, especially qualitatively. The curves obtained from (15) appear to have a
similar course as the ones of the EC3- Annex A formulae, but with some advantages in accuracy
particularly in the case of the circular cross-section, for which the cross-sectional interaction is poorly
represented by the Eurocode formulae. It should be noted that the accuracy of the kj values gives a rather
misleading representation of the accuracy of the formulation itself, particularly for higher values of ny or
nz. Even errors of some 20-30% in terms of kj only lead to total errors of only a few percentage points at
values of nj beyond 0.5. In this sense, it is a welcome observation that the accuracy of the proposed
formulation in terms of kj is highest for low values of nj, and mostly conservative in all other cases.

Andreas Taras and Richard Greiner

2.0 2.5 2.5

IPE 240 CHS178/10 HEA 500z

- + -
2.0 2.0

kzz [-]
ky [-]
1.5 0.3 fy
- + -
ky [-]

0.15 fy
0.3 fy 1.5 1.5

GMNIA 1.0 1.0
ny=0.2, 0.4, 0.6, 0.8 EC3-A ny=0.2, 0.4, 0.6, 0.8 nz=0.2, 0.4, 0.6, 0.8
0.5 0.5 0.5
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Oy a) Oy b) Oz c)
2.0 2.5 2.5
IPE 240 CHS178/10 HEA 500z

- + -
2.0 2.0
ky [-]


kz [-]
0.3 fy
- + -
ky [-]

0.15 fy
0.3 fy 1.5 1.5

kz [-]
1.0 1.0
ny=0.2, 0.4, 0.6, 0.8 EQU ny=0.2, 0.4, 0.6, 0.8 nz=0.2, 0.4, 0.6, 0.8

0.5 0.5 0.5

0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Oy Oy Oz

Figure 6: Comparison of interaction factors kj; top: Eurocode Annex A & B; bottom: new proposal.
The design proposal in this paper combines the advantages of the “interaction” and “generalized
slenderness” concepts for the case of in-plane beam-column buckling. The comparison with numerical
results and current Eurocode rules (with many more examples given in [5]) showed the new proposal to
have a consistent level of accuracy and safety. The proposal could serve as a procedural blue-print for the
expansion of “generalized slenderness” concepts to other member-buckling cases.

[1] EN 1993-1-1, Eurocode 3. Design of steel structures. General rules and rules for buildings, CEN,
Brussels, 2005.
[2] Rotter, J.M., Shell Buckling and Collapse Analysis for Structural Design: The New Framework of
the European Standard, Festschrift for Prof. Calladine, Cambridge, 2002.
[3] Boissonade, N., Greiner, R., Jaspart, J.P., Lindner, J., Rules for Member Stability in EN 1993-1-1,
Background documentation and design guidelines, ECCS TC 8 – Stability, Brussels, 2006.
[4] Greiner, R., Taras, A., On the variety of buckling curves, Proc. of “Stability and Ductility of Steel
Structures”, Lisbon (PT), Sept. 6-9, 1101-1108, 2006.
[5] Taras, A., Contributions to the Development of Consistent Stability Design Rules for Steel
Members, PhD Thesis, Graz University of Technology, 2010.
[6] Rondal, J., Maquoi, R., Formulations d’Ayrton-Perry pour le Flambement des Barres Métalliques,
Construction Metallique, 4, 41-53, 1979
[7] Beer, H., Schulz, G., Bases Théoriques des Courbes Européennes de Flambement, Construction
Métallique, 3, 37-57, 1970.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Harald Unterweger*

* Institute of Steel Structures, Graz University of Technology, Austria

e-mail: h.unterweger@tugraz.at

Keywords: Connections, Stability.

Abstract. In building constructions for bracing members often hollow sections are used with slotted
gusset plates at the ends. These plates are attached to non-stiffened plates of the adjacent construction. In
practice sometimes a nearly centered joint is designed, by arranging the slotted gusset plate with an
eccentricity of half the plate thickness to the member axis, so that the member eccentricity is minimised.
In the paper the load bearing behaviour of such members under compression and tension is discussed
based on numerical analysis with nonlinear FE – models including imperfections. The geometric
parameters and boundary conditions are varied in such a way that practical cases are covered and that
the typical load bearing behaviour can be seen. Based on these results an engineering model for the
design in practice is represented.

Bracing members and truss members are often designed with hollow sections and slotted gusset plates
at the ends. These plates are attached to non-stiffened plates of the adjacent construction, using welds or
bolts. In Fig. 1a some typical joints of this type are represented, including also joints with concrete
foundation. In Fig 1b the representative and idealized joint configuration for these joints - limited to
rectangular hollow sections (RHS) - is shown. The vertical plate (KB2) is restrained in axis I. Here, the
two different border cases related to boundary conditions are considered: - pinned (BC1) or, – fixed
(BC2, e.g. “concrete joint” in Fig. 1a). The slotted gusset plate of the member is joined by a fillet weld
(a1), passed around. An alternative solution with two bolt rows is possible, in accordance with the as-
sumptions of the numerical analysis (fixed connection between plate KB1 and KB2 along their borders).
A special feature of the studied joint is the eccentric position of the slotted gusset plate with an
eccentricity of half the plate thickness tKB1, as shown in Fig. 1b. In doing so, the eccentricity of the RHS -
member for the buckling check – relevant is member buckling out of plane (about the z – axis) - is
minimised (e* = 0,5·tKB2). The bending moment along the member-length is constant.
Based on the minimised eccentricity for the RHS - member, the opinion of practitioners is that only a
buckling check for the member under axial load is necessary to get the load carrying capacity of the
member. The results of this study will show that this approach would lead to high overestimations of the
load carrying capacity, especially for low slenderness ratios of the member. The reason for this, are high
bending moments in the gusset plate out of plane.
The loading of the RHS - member in this study is restricted to axial forces with bending moments
only due to eccentricities of the joints. This paper summarizes the results in [1].


In the following the load bearing behaviour of the member in compression, influenced by the specific
type of joint, is represented.

Harald Unterweger

First of all the finite element (FE) – model and the executed nonlinear numerical analysis are
presented. Afterwards the studied band width of the varied geometric parameters is summed up. At the
end the results of the numerical nonlinear calculations for an example of a rectangular hollow section are
presented, for different member slenderness and boundary condition.

Figure 1: Studied RHS - member joints: a.) different types in practice, b.) geometry and restraint
conditions of the studied representative joint.

2.1 FE model and calculation procedure

The numerical FE – model, based on the Software ABAQUS [2], consists of continuum (solid) and
beam elements and is represented in Fig. 2a. The linear continuum elements were used within the joint
and for the following parts of the hollow section (over a length of about 0,5 m). For the vertical gusset
plates eight elements over the thickness were considered, because the localized plastification in this
region affects highly the load bearing behaviour of the member. The continuation of the RHS - member
was modeled with linear beam elements only to the section at midspan, because symmetric or
antisymmetric boundary conditions there were sufficient to capture the real member behaviour. The
boundary conditions at the end of the gusset plate (axis I) were chosen adjusted on the two studied
configurations, pinned or fixed out of plane (see Fig. 1b). The two gusset plates were joined using contact
elements along the axis of the weld in between. The sealing plates were omitted.

a.) b.)

Figure 2: a.) FE – model of the joint, b.) relevant first eigenmode, L0 = 2 m, for a pinned (above) and
fixed gusset plate (below).

Harald Unterweger

In the study a squared, hot finished, RHS - profile with 100 / 100 / 5 mm was used, leading to gusset
plate dimensions of 250 / 130 mm (KB 1) and 100 / 330 mm (KB 2). The calculations were done for total
member lengths L0 = 2, 3, 4, 5, 6, 7 and 8 m. The corner radii of the hollow section were omitted, leading
to an area A0 = 1900 mm2 and a radius of gyration iz,0 = 38,84 mm.
In the calculations an ideal elastic – perfectly plastic material behaviour was considered with a
characteristic yield strength of fy = 235 N / mm2. A modulus of elasticity E = 210000 N/mm2 and a
Poisson ratio of Q = 0,3 were used.
First of all an LBA – analysis (linear buckling analysis) was made, leading to the capacity NLBA of
the member. Based on these results, on the one hand the “real” buckling lengths of the members were
determined (using the formula for the Euler buckling load for the RHS - member section). Due to the
limited bending stiffness of the gusset plates, the buckling length Lcr,0 of the idealized RHS - member
within the end – restraint in axis I (Lcr,0 = L0 for BC 1, Lcr,0 = 0,5 L0 for BC 2) is too small.
Afterwards the results are either based on the idealised slenderness തതതതതߣ௭ǡ଴ (Equ.1), with Lcr = L0 , or on
the slenderness based on the LBA – analyses തതതതതതߣ௅஻஺ (Equ. 2).
f y L0 1 L0 1
O z ,0 ˜ ˜ (1)
V cr ,0 iz ,0 S ˜ E f y iz ,0 93,9

N pl ,0 A0 ˜ f y
O LBA (2)

On the other hand the eigenmodes of the LBA – analyses, scaled to a maximum value of wmax= L0 /
1000 were used for a GMNIA – analyses (geometric and material nonlinear analyses with imperfections).
This was done with care, considering different eigenmodes (not only the one for the minimum ideal
buckling load), to get a minimum for the load carrying capacity NR of the member. In Fig. 2b the relevant
eigenmodes for a member with small length L0 is shown for the two different boundary conditions.
Additional GMNA – calculations were used to check the GMNIA – results. Residual stresses were
ignored, because they affect the buckling capacity of RHS – members not significantly (e.g. [4]).
Also for cold formed RHS - members the presented results mainly are valid, only for high slenderness
ratios – where the overall buckling of the member is relevant – the appropriate buckling curve should be
used (e.g. curve c instead of a, using Eurocode 3 [3]).
2.2 Studied joint parameters
The numerical study was limited to rolled RHS – members. The joint geometry is restricted to the
dimensions of Fig. 1b. Very important is the slotted length Ls • 1,5·h in the RHS – member. Otherwise
sometimes significant smaller load bearing capacities would occur. The overlapping length of the two
gusset plates was fixed with 0,75·h. The distance between member end and restraint axis I is limited to
L1 = 1,25·h.
The thickness of the two gusset plates was varied in such a way that the area ratio AKB1 / A0 = 0,8 to
1,4 and tKB2 = (0,5 to 1,0 )·tKB1.
2.3 Results of the nonlinear calculations
The results of the nonlinear calculations for different member length (i.e. different member
slenderness) and gusset plate thicknesses are presented in Fig. 3, based on an effective width of beff =
3,3·h = 330 mm of gusset plate KB2. The load carrying capacity NR is related to the section capacity of
the RHS - member Npl,0 = A0·fy =19,0·23,5 = 446,5 kN. The slenderness ratio തതതതതത
ߣ௅஻஺ (Equ. 2) is based on
the results of the LBA–analysis - that means based on the “real” buckling length.
In Fig. 3a the overall carrying behaviour is shown, based on the GMNA – results, without geometric
imperfections. The effect of these geometric imperfections is quantified in Fig. 3b, where the results with
and without imperfections are visible. It can be seen that the reduction of load carrying capacity is
comparatively small.

Haraldd Unterweger

Inn Fig. 3 also thhe buckling currve a, relevant for the bucklinng load capacity y of a hot finishhed, RHS –
membber under onlyy axial load in Eurocode
E 3 [3]] is plotted (dotted line). It caan be seen that the special
featurre of the studieed joint – minimmising of the member
m eccentriicity – is only usable
u for high slenderness
ratioss and thick gussset plates. For example
e with a gusset plate thiickness of tKB1 = tKB2 = 12 mm m, leading to
a ratiio AKB1 / A0 = 0,82 a dramatiic reduction of the load carryiing capacity occcurs in case off the pinned
gusseet plate KB2, also
a for very low w slenderness ratios
r (NR § 0,16·Npl,0 ). A fix xed gusset platee, however,
increases the load carrying capacitty significantly,, but also for veery low slenderrness ratios thee capacity is
far beelow the sectioon capacity of thhe RHS - mem mber (NR § 0,544·Npl,0 ). Not beefore the plate thickness
t is
increased significanntly (t KB1 = tKB22 = 20 mm, leaading to a ratio AKB1 / A0 = 1,37) and the gussset plate is
fixedd (BC2 in Fig. 1b) nearly about 80 % of thhe section capacity Npl,0 for small slendernness ratio is

curve a
curve a

slenderneess slendern
Fiigure 3: a.) GM
MNA - results deepending on thee slenderness raatio, b.) GMNA - results in com
with GMMNIA - results, for pinned (RB1=BC1) and fixxed (RB2=BC2 2) gusset plates.

The significant reduction of thhe load carryingg capacity – alsso related to the buckling capacity of the
RHS––member (see Fig.F 3) – is caussed by the locall bending momeents in the gussset plate, particuularly at the
end oof the RHS – member.
m This caan be seen in Fiig. 4 for a very short member with the thin gusset
g plates
mentiioned before. In
I Fig 4a the pinned
p and in Fig.
F 4b the fixeed ended gusseet plate can be seen at the
mate limit state. The gusset plaate section at thhe member endd reaches its seection capacity under axial
forcee and bending moment.

Figgure 4: Short meember (L0 = 2 m)

m at ultimate liimit state (GMN NA-analysis) with
w tKB1 = tKB2 = 12 mm;
a.) pinned gussset plate – NGM
MNA / Npl,0 = 0,1
173, b.) fixed guusset plate – NGMNA
G / Npl,0 = 0,543.

Harald Unterweger

Summing up, the load carrying capacity of the RHS - member with the specific joint configuration of
Fig. 1b has an upper limit – also for very low slenderness ratios – which primarily is influenced by the
gusset plate thickness tKB1 and the boundary condition of the gusset plate (pinned or fixed). The influence
of the effective width beff of the gusset plate is significant smaller.
These correlations are presented in Table 1. The tabulated load carrying capacities are calculated with
the engineering model presented in chapter 4.

Table 1: Ratios of maximum compression capacities N1,Rd / Npl,0 of the member –

influence of gusset plate thickness tKB1 and boundary condition (tKB2 = 12 mm).
boundary tKB1 = 12 tKB1 = 20 t KB1 = 25 t KB1 = 30
condition A KB1/A0=0,82 A KB1/A0=1,37 A KB1/A0=1,71 A KB1/A0=2,05
BC 1, beff = 3,3h 0,172 0,373 0,506 0,642
BC 2, beff = 1,6h 0,351 0,469 0,576 0,695
BC 2, beff = 3,3h 0,458 0,545 0,631 0,736


The load bearing behaviour of the member in tension was also studied. Now the tension axial force
reduces the bending moment in the relevant gusset plate section at the member end significantly (2nd
order effect). The influence of the boundary condition (BC1 or BC2) disappears nearly complete and for
the studied gusset plate thicknesses the following member capacities in tension were calculated (L0 = 2 m,
beff = 330 mm):
- tKB1 = tKB1 = 12 mm: NGMNA / Npl,0 = 0,87
- tKB1 = tKB1 = 20 mm: NGMNA / Npl,0 = 0,94
- tKB1 = 20, tKB2 = 12 mm: NGMNA / Npl,0 = 0,94


Based on the results of the numerical study, an engineering model was developed to calculate the
compression load-bearing capacity of the RHS - member with the specific joint configuration of Fig 1b. It
includes the following variations of the relevant parameter: - pinned (BC1) or fixed (BC2) end of the
gusset plate, - beff = 1,6·h to 3,3·h, - a “free” length of gusset plate L1 also longer than 1,25·h, - varying
thicknesses, but tKB1 • tKB2 .
The engineering model includes on the one hand a conventional member buckling check with the
specifications given in chapter 4.1 – relevant for high slenderness ratios – leading to N2,Rd.
On the other hand the calculated capacity N2,Rd is limited by an upper limit N1,Rd – independent of the
member slenderness – characterising the plateau of the load bearing capacity for small and medium
slenderness (see Fig. 3). The resulting compression load bearing capacity NRd is the minimum of both
As stated in Eurocode 3 [3], a partial safety factor Jf = 1,0 is considered, leading to the design yield
strength fyd = fy and the load capacity NRd.
4.1 Buckling member capacity N2,Rd
Although the “real” buckling length LLBA of the member is influenced by the smaller bending stiffness
of the gusset plates, the following simplifications are possible. For the buckling check about the z – axis
only the member section is relevant (A0, iz0) and the relevant buckling curves of the international codes
can be used (e.g. for hot finished, RHS – profiles, curve a in Eurocode 3 [3]).

Harald Unterweger

For pinned gusset plates (BC1) simplified LLBA = L0 , as long as L1 < 1,5·h. In addition the constant
moment M = N2,Rd·e* = N2,Rd·(tKB2 / 2) (see Fig. 1b) should be used in the buckling check. For higher
slenderness ratios the influence of the moment decreases and it can be omitted.
For fixed (BC2) gusset plates the eccentricity e* can be omitted, if LLBA = L0 is used. This
simplification leads to conservative results, mainly for high thicknesses tKB1 (see Table 2). Otherwise the
moment should be included (LLBA§ 0,85·L0 , as long as N2,Rd is relevant for design).
4.2 Upper limit for the member capacity – N1,Rd
The engineering model for the upper limit capacity N1,Rd of the member is defined by the load
carrying capacity of the gusset plate – section at the end of the member (axis II), considering 2nd order
effects. The model is summed up in Fig. 5.
The load bearing capacity N1,Rd is based on the full utilization of the plastic section capacity, due to
axial force and bending moment. As defined in Eurocode 3 [3], the acting axial force is considered by a
reduced moment capacity MN,Rd, given in Equ. 3 (fyd is the design yield strength).
§ § N · · hKB ˜ t 2 KB1 § N 21, Rd ·
M N , Rd M pl , KB1 ˜ 1  ¨ 1, Rd
¸ ¸ ˜ f yd ˜ ¨1  2 ¸ (3)
¨ ¨ N pl , KB1 ¸ ¸ 4 ¨ h KB ˜ t 2 KB1 ˜ f 2 yd ¸
© © ¹ ¹ © ¹
To calculate N1,Rd an iterative approach is necessary, until Equ. 4 is fulfilled.
M II d M N , Rd (4)
The acting bending moment MII in the gusset plate, depends on the axial force N1,Rd and the actual
boundary condition. For pinned gusset plates (BC1) the full eccentricity e0 is relevant in section II (see
Fig. 5), leading to Equ. 5. The 2nd order effect, is covered by the factor fII in form of a so called
“Dischingerfaktor”, including the Euler buckling load Ncr,BC1 for the gusset plate (Equ. 6), based on the
relevant buckling length lcr = 2·L1.

§t t · 1
M II N1, Rd ˜ e0 ˜ f II N1, Rd ˜ ¨ KB1 KB 2 ¸ ˜ (5)
© 2 N
¹ 1  1, Rd
N cr , BC1

S 2 ˜ E ˜ I z , KB1 S 2 ˜ E ˜ hKB ˜ t 3 KB1

N cr , BC1 (6)
l 2cr 48 ˜ L21
For fixed gusset plates (BC2) the bending moment due to the full eccentricity e0 is reduced, because
also section I gets a part of this moment, leading to Equ. 7.
M I  M II N1, Rd ˜ e0 (7)
For the two parts in Equ. 7 the bending stiffness of the two gusset plates is relevant (identical bending
deformations). This gives: MI / MII = Iz,KB2 / Iz,KB1 , where Ii are the moment of inertias. Based on Equ. (7),
finally we get MII using Equ. (8b). But now the Euler buckling load Ncr,BC2 (Equ. 9) is based on the
reduced buckling length lcr = L1 (see Fig. 5).

1 §t t · 1
M II N1, Rd ˜ f M ˜ e0 ˜ f II N1, Rd ˜ ˜ ¨ KB1 KB 2 ¸ ˜ (8a)
I z , KB 2 © 2 ¹ 1  N1, Rd
I z , KB1 N cr , BC 2

1 §t t · 1
M II N1, Rd ˜ ˜ ¨ KB1 KB 2 ¸ ˜ (8b)
beff ˜ t 3 KB 2 © 2 ¹ 1  N1, Rd
hKB ˜ t 3 KB1 N cr , BC 2

Harald Unterweger

S 2 ˜ E ˜ I z , KB1 S 2 ˜ E ˜ hKB ˜ t 3 KB1

N cr , BC 2 2
l cr 12 ˜ L21

Figure 5: Engineering model for the determination of the compression load capacity N1,Rd,;
a.) actual situation, b.) determination of N1,Rd for BC1 (above) and BC2 (below).

4.3 Accuracy of the engineering model

The accuracy of the engineering model is presented in Table 2, in comparison of the results for
different slenderness, boundary conditions and gusset plate variants with the results of the GMNIA –
For the load bearing capacity N2,Rd, relevant for higher member slenderness, the buckling check was
based on Eurocode 3 [3] with buckling curve a. Always the simplified buckling length Lcr = L0 was used.
Only for the pinned gusset plate the moment due to the eccentricity (M=N2,Rd·e*) was considered thereby.
This simplified approach would lead to very conservative results for fixed gusset plates (BC2).
Only for very thin gusset plates with pinned ends the engineering model for the upper load bearing
capacity N1,Rd (relevant for small and medium member slenderness) is on the unsafe side. But due to the
very limited capacities in those cases (N1,Rd § 0,2·Npl,0) these variants are not relevant in practice.


Based on the results of the numerical study, also for RHS - members in tension an engineering model
was developed. Now the 2nd order effects significantly reduce the maximum bending moments in the
relevant gusset plate section (in axis II, as in compression). The tension load bearing capacity Nt,Rd is
calculated, by using Equ. 4, based on MN,Rd in Equ. 3 (replace N1,Rd by Nt,Rd), and the reduced bending
moment MII in Equ. 10 (with tKB1 in mm). The latter is developed for a gusset plate thickness of tKB1 = 20
mm and the moment is approximately proportional to the bending stiffness, but nearly not influenced by
the boundary conditions of the gusset plate (pinned or fixed).

Harald Unterweger

t KB1 § t KB1 ·
M II Nt , Rd ˜ et N t , Rd ˜ ˜¨ ¸ (10)
5,14 © 20 ¹

Table 2: Compression load capacity ratios NRd / Npl,0 (RHS - profile 100/100/5, beff = 330mm), with
imperfections as well as based on an engineering model for 3 variants:
- tKB1 = tKB2 = 12 mm (V1), - tKB1 = tKB2 = 20 mm (V2), - tKB1 = 20 / tKB2 = 12 mm (V3).
boun- തതതത
slenderness ߣ ௭଴
dary calcu- 0,548 1,097 1,645 2,194
cond. lation
V1 V2 V3 V1 V2 V3 V1 V2 V3 V1 V2 V3
BC 1 GMNIA 0,161 0,309 0,367 0,163 0,302 0,357 0,160 0,272 0,295 0,148 0,177 0,183
Eng. 0,172 0,309 0,373 0,172 0,309 0,373 0,172 0,276 0,291 0,172 0,172 0,178
model +6,8% +0,0% +1,6% +5,5% +2,3% +4,5% +7,5% +1,5% -1,4% +16% -2,8% -2,7%
BC 2 GMNIA 0,526 0,825 0,643 0,525 0,654 0,643 0,361 0,425 0,424 0,219 0,286 0,275
Eng. 0,458 0,786 0,545 0,458 - 0,598 0,545 - 0,318 - 0,318 0,318 0,188 0,188 0,188
model -13% -4,7% -15% 13% -8,6% 15% 12% -25% -25% -14% -34% -31%

The almost centric joint in Fig. 1 only for RHS - members in tension gives a load bearing capacity
comparable with the section capacity Npl,0 (about 90 % of Npl,0 ). However, thick gusset plates are
necessary, leading to an area ratio of AKB1 / A0 > 1,25.
Also for such thick gusset plates with pinned ends the load bearing capacity in compression is limited
to about 50 % of the section capacity Npl,0 , independent of the member slenderness. The reason for that
is the high additional bending moment in the gusset plate at the member end (axis II in Fig. 5).
In case of a fixed gusset plate a maximum of about 70÷80 % of the section capacity Npl,0 is available.
Only for very high member slenderness (ߣ തതതതത
௭ǡ଴ > 1,5 / 1,0 for pinned / fixed gusset plates) the buckling
check of the member is relevant for design.
It should be noted that the studied joint configuration of Fig. 1 is nearly not usable, if fatigue loads are
relevant (stress cycles due to axial force). The reason for this, are very high stress peaks near the welds,
relevant for the fatigue check.

[1] Unterweger, H. and Ofner, R., “Traglast von Verbandsstäben aus Hohlprofilen mit quasi-
zentrischem Knotenblechanschluss“, Stahlbau, 78(6), 425-436, 2009.
[2] ABAQUS, Software package, Version 6.7, 2007.
[3] Eurocode 3, European Standard, Design of steel structures – Part 1-1: General rules and rules for
buildings, 2006.
[4] Wilkinson, T., “The plastic behaviour of cold – formed rectangular hollow sections”, Doctoral
Thesis, University of Sydney, 1999.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Anna M. Barszcz*, Marian A. Gizejowski* and Wael A. Salah Khalil**

* Department of Building Structures, Warsaw University of Technology, Warsaw, Poland

e-mails: A.Barszcz@il.pw.edu.pl, M.Gizejowski@il.pw.edu.pl
** Department of Civil Engineering, Al-Azhar University, Cairo, Egypt
e-mail: waelcivil@hotmail.com

Keywords: Steel Frame, Angle Bracing, Bolted Connection, Welded Connection, Numerical Modeling.

Abstract. This paper summarizes experimental and numerical investigations on the behavior of steel sub-
frame specimens braced with use of a diagonal angle member. The laboratory tests were designed to
investigate the effect of bracing member end connections (bolted asymmetric and welded symmetric) and
the effect of bracing member slenderness on the frame behavior in the whole range of frame load-
displacement characteristics. An advanced finite element numerical model is developed with use of
commercial ABAQUS code. Elastic buckling modes are evaluated. Displacement controlled Riks
geometrically and materially nonlinear analysis is carried out for the reproduction of the behavior of
specimens tested in laboratory. Numerical frame load-displacement characteristics are compared with
experimental ones. A simple analytical model of the compression member behavior developed elsewhere
is compared with experimental angle brace characteristics. Practical recommendations are formulated.

Steel truss bracing systems are commonly used for the enhancement of sway performance of
structural frames in multi-storey buildings. Vertical truss bracings are composed of diagonals made of
rolled profiles, single or compound. In typical braced frames of moderate height, diagonals are made of
single angles connected through bolted asymmetric joints or welded symmetric joints to continuous
columns, or columns and beams, of the primary load bearing frame.
The behavior of structural members is usually examined experimentally as isolated elements; see
Gizejowski et al. [1]. This type of experiments is helpful in predicting the effect of member slenderness
on the buckling strength but is not adequate to model the member performance as an element of structural
systems, especially with reference to the influence of real end conditions of bracing members on their
buckling strength and force-deformation characteristic. The behavior of members acting integrally as
elements of braced frame structures has been less investigated.
Analytical models, one based on tangent modulus theory and the second - on the evaluation of overall
member force-deformation characteristic, have been developed by Barszcz and Gizejowski in [2] for the
prediction of buckling strength according to Eurocode 3. Model based on the evaluation of overall
member force-deformation characteristic for the assessment of load-displacement characteristic of more
complex structures was presented in [3]. The above mentioned analytical model of the member behavior
is further verified in this paper for angle bracing members tested as components of sub-frame specimens.
A summary of experimental work concerned with the behavior of angle member as an element of
braced sub-frame portal specimens is presented hereafter. The experimental load-displacement behavior
of tested specimens is compared with the FE results obtained with use of commercial ABAQUS code.
Experimental force-deformation characteristics of the brace angle are compared with the analytical model

Anna M. Barszcz et al.

dealt with in [2] and [3]. Conclusions directed towards the practical application of developed analytical
model for modeling of the behavior of bolted and welded angle braces are drawn.

Tests were designed to examine experimentally the behavior of two sets of braced sub-frames that are
presented in figure 1. Figure 1a illustrates the general layout of sub-frame specimens BL with the angle
brace jointed to gusset plates with use of high-strength bolts though close tolerance holes, and figure 1b –
specimens WL with the angle brace welded to gusset plates. Details of both types of the connection are
given in figure 2a and b, respectively.

a) b)

Figure 1: Geometry of tested sub-frame specimens; a) BL specimens, b) WL specimens.

a) b)

Figure 2: Details of angle bracing connection; a) specimens BL, b) specimens WL.

Each set of tested specimens consists of three subsets characterized by different beam length L and
the diagonal distance Ld. Notation for each subset of specimens is given in table 1. In each subset of BL
and WL specimens, three specimens were tested.

Table 1: Description of tested specimen subsets.

Characteristic BL 1320 BL 1520 BL 1925
subset parameter WL 1320 WL 1520 WL 1925
L 1320 1520 1925
Ld 1405 1595 1985

Tests were conducted for sub-frames mounted in an upside-down position. In the upper left node, the
bracing member axis coincided with the left column axis and the node was held in position but allowed
for the in-plane rotation. The same boundary conditions were applied to the upper right node of the right
column end. The upper left node was loaded with a horizontal load F and the upper right node was
subjected to a reactive force. The beam of sub-frame specimens was restrained in the out-of-plane
direction in order to ensure that the frame deflects primarily in-plane. The incremental displacement

Anna M. Barszcz et al.

controlled loading program, corresponding to the horizontal load F, was applied. Details of test rig,
testing procedure and measurement devices were described in [4].


Numerical investigations are conducted with use of commercial software ABAQUS. All the rolled
double tee and angle section walls are modeled with use of thin shell four node finite elements S4R5 from
the ABAQUS library. This element is only suitable for thin elements with small strain using the thin shell
theory, however, large displacements are allowed for. The S4R5 elements are significantly less expensive
since they use the reduced integration rule (Gauss integration). They are also cost-effective for large
models with small strain and have good hourglass control. The aspect ratio of the mesh was kept close to
1.0 throughout.

a) BL 1320: Fcr=203,1 kN d) WL 1320: Fcr=211,9 kN

b) BL 1520: Fcr=178,0 kN e) WL 1520: Fcr=202,8 kN

c) BL 1925: Fcr=129,4 kN f) WL 1925: Fcr=171,2 kN

Figure 3: Frame critical loads and buckling modes from numerical simulations.

The multi point constraints option (MPCs) available in ABAQUS code is used to model the bolts in
the braced sub-frame specimens BL. For each bolt MPC, rigid beam like constrains are created at
matching nodes of two mid-surfaces of the angle brace leg and the gusset plate around the bolt shank

Anna M. Barszcz et al.

circumference approximated by a square in order to simplify capturing the behavior of two plate elements
jointed with high-strength close tolerance bolts. This helps also to avoid early analysis termination in case
of excessive plastic deformations in concentrated bearing zones, between the bolt shank and connecting
plate elements. Such a modeling technique seems to be accurate from engineering point of view.
The angle brace welded to gusset plates is modeled with connecting the angle brace to gusset plate
directly via common nodes along the intersection line between the angle brace and the gusset plate. It is
strongly believed that this modeling technique is quite accurate for the welded joint as long as no fracture
is developed in the area of weld material or its neighborhood.
In order to estimate the elastic failure loads and to find the sensitivity of tested specimens to buckling
effects, eigenproblems are solved. Critical loads Fcr and buckling modes are evaluated. Figure 3 illustrates
the buckling modes and the values of critical loads. It is observed that for the short-beam specimens a
distortional form of bracing member buckling governs while for the longest beam specimens - an overall
buckling mode is detected. It is clear that buckling is associated primarily with the bracing member
deformations and rather a negligible contribution of the frame beam and column deformations to the
buckling profile is observed.

a) WL 1320: Fult=124,6 kN

b) WL 1520: Fult=124,4 kN

c) WL 1925: Fult=115,3 kN

Figure 4: Frame deformation of welded brace specimens from tests and numerical simulations.

Finally, ultimate loads and deflected profiles are evaluated from fully nonlinear Riks analysis using
ABAQUS code. Figure 4 gives the values of the frame ultimate loads Fult and illustrates the deformed
profiles of inelastic WL sub-frames at failure, comparing them with those recorded during tests. One can

Anna M. Barszcz et al.

observe that the deflected frame profile is characterized by localized plastic deformations of the lower
node of the right column. Since bracing members in these specimen subsets are connected concentrically,
their deflected profile is associated with buckling deformations. The maximum coordinate of buckling
profile shifts from the lower joint for a shorter brace length to the mid-length for a longer brace length.
Figure 5 gives the values of the frame ultimate loads Fult and illustrates deformed profiles of inelastic
BL sub-frames at failure, comparing them with those recorded during tests. One can observe that the
deflected frame profile of BL specimens is similar to that of WL sub-frames. Since bracing members in
all BL subsets of tested specimens are connected eccentrically, their deflected profile is associated with
bending and torsion, without distinguished buckling in-plane deformations.

a) BL 1320: Fult=113,1 kN

b) BL 1520: Fult=109,3 kN

c) BL 1925: Fult=105,8 kN

Figure 5: Frame deformation of bolted brace specimens from tests and numerical simulations.

The most stressed joint in all the tests appeared to be the right lower frame joint at which three
elements are connected – the frame beam and column, and the bracing member. The detailed deformation
of this joint obtained from laboratory tests and from numerical modeling is shown in figure 6.

Figure 6: Details of WL 1320 column-beam-bracing joint deformations from tests and simulations.

Anna M. Barszcz et al.

Frame load-displacements characteristics from laboratory tests and from numerical modeling are
presented in figure 7. All six frame experimental characteristics are evaluated as average curves from
three tests of each subset. Since numerical modeling is carried out for initially perfect specimens, the
accuracy of computer simulations is of a different degree depending on the sensitivity to imperfections
and type of bracing member connections.

a) BL 1320 d) WL 1320

b) BL 1520 e) WL 1520

c) BL 1925 f) WL 1925

Figure 7: Frame load-displacement characteristics from tests and numerical simulations.

The following observations are made:

a) For frame specimens BL with eccentric connections of the bracing member, computer simulations
lead to a higher initial stiffness and a noticeably lower ultimate loads as well as a lower placement of the

Anna M. Barszcz et al.

post-limit branch of equilibrium path if compared with those obtained experimentally. Effect of geometric
imperfections of the bracing member does not play important role in this case since the member is under
combined bending, torsion and compression from the beginning of loading process.
b) For frame specimens WL with concentric connections of the bracing member, computer
simulations lead to a very close estimation of the initial stiffness but a higher ultimate loads and a lower
placement of the post-limit branch of equilibrium path if compared with those obtained experimentally.
Higher values of the ultimate load from computer simulations can be attributed to the effect of geometric
imperfections of the bracing member that is not accounted for in analysis. Since the bracing member is
concentrically connected with regard to out-of-plane deformations, it is predominantly axially loaded
almost up to the attainment of the frame buckling strength. A sharp drop in the value of applied load is
observed in the post-limit range. A more close estimation could be expected if the effect of geometric
imperfections is accounted for in computer simulations.


a) BL 1320 d) WL 1320

b) BL 1520 e) WL 1520

c) BL 1925 f) WL 1925

Figure 8: Bracing member force-deformation characteristics from tests and analytical model.

Anna M. Barszcz et al.

Finally, the behavior of bracing member assessed on the basis of test results is compared with simple
modeling of force-deformation characteristic developed in [2]. Force-deformation characteristics
presented in figure 8 are constructed in terms of the member axial force and member generalized strain
(shortening due to compression and bending divided by the length). Results of three experimental curves
are plotted for each subset of tested frame specimens. The upper bound of analytical solution is
constructed for the effective length factor ȝ=0.5 while the lower bound it corresponds to the effective
length factor ȝ=1.0.
The following observations are made:
a) The evaluation of welded connection bracing member behavior is reasonable enough using the
upper bound analytical solution. It indicates that that the force-deformation characteristic of the welded
type of member connection may be evaluated with the effective length factor equal 0.5.
b) The bolted connection bracing member behavior can not be evaluated so accurately using the cited
analytical formulation. The values of experimental initial stiffness are lower than those from analytical
solution for both values of the effective length factor. The experimental buckling strength is placed
between those corresponding to two extreme values of the effective length factor, i.e. for two different
types of boundary conditions assumed for the connection of brace member. This clearly indicates that the
strut model developed in [2] has to be refined in case of bolted connections in order to account for the
effect of connection eccentricity.

This paper presents experimental, numerical and analytical results of investigations aiming at the
development of a simple and yet reliable model for the angle member behavior in bracing systems of
structural frame structures. Welded brace angle force-deformation characteristics may be evaluated using
directly the analytical formulation presented in [2] for the effective length factor equal 0.5. Bolted brace
angle force-deformation characteristics can not be evaluated so accurately using the formulation
presented in [2]. The refinement is needed with regard to the effect of connection eccentricity and
inclusion of corresponding bending deformations in the evaluation of member generalized strain. This
would affect predominantly the region of pre-buckling deformations and the level of buckling strength,
resulting in lowering of the member force-deformation curve and bringing the analytical curves closer to
the experimental ones. A more accurate model would therefore be developed and the effective length
factor suggested accordingly.
The development of a refined model of the angle strut behavior in case of bolted connections is being
underway, and better prediction of the bolted brace force-deformation characteristic is expected.

[1] Gizejowski M.A., Barszcz A.M., Foster J.D.G., Uziak J., Kanyeto O.J., “Experimental
investigations of the behaviour of angle struts”, Proc. of ICMS2006 XIth International Conference
on Metal Structures, M. A. Gizejowski, A. Kozlowski, L. Sleczka and J. Ziolko (eds.), Taylor &
Francis, London / Leiden / New York / Philadelphia / Singapore, 152-153, 2006.
[2] Barszcz, A.M., Gizejowski, M.A. “An equivalent stiffness approach for modelling the behaviour of
compression members according to Eurocode 3”. Journal of Constructional Steel Research, 63(1),
55-70, 2007.
[3] Gizejowski M.A., Barszcz A.M., “Advanced analysis of inelastic steel truss and frame structures: a
unified approach”, Proc. of SDSS2006 International Colloquium on Stability and Ductility of Steel
Structures, D. Camotim, N. Silvestre and P.B. Dinis (eds.), IST Press, Lisbon, 431-438, 2006.
[4] Barszcz A.M., “Modelling and experimental investigations of the behaviour of angle bracing strut
in steel frames”, Proc. of Local Seminar of IASS Polish Chapter on Lightweight Structures in Civil
Engineering: Contemporary Problems, J. Obrebski (ed), Micro-Publisher, Warsaw, 106-113, 2007.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Pedro D. Simão*, **, Ana M. Girão Coelho*, *** and Frans S. K. Bijlaard***

* Institute of Computers and Systems Engineering of Coimbra, Portugal (INESC-Coimbra)

e-mail: pedro@dec.uc.pt, a.m.girao@clix.pt
** Department of Civil Engineering, University of Coimbra, Portugal
*** Department of Structural and Building Engineering, Delft University of Technology, The Netherlands
e-mail: f.s.k.bijlaard@tudelft.nl

Keywords: Column Stability, GBT, Lagrange Multipliers, Rayleigh-Ritz method, Spliced Columns.

Abstract. The paper presents a study on the influence of splice connections on the stability behaviour of
compressed steel columns. The column is modelled as two independent prismatic parts connected by a
rotational spring at the splice location and rotational and extensional springs at the column ends to
represent the effect of the adjacent structure. The general behaviour is characterized using a polynomial
Rayleigh-Ritz approximation substituted into the potential energy function, in combination with the La-
grange’s method of undetermined multipliers, and based on this model the critical load is found. The
load-carrying capacity is analysed with respect to the following variables: (i) location and rotational
stiffness of the splice, (ii) change in the column section serial size and (iii) column end-restraints stiffness
coefficients. A nonlinear regression model is developed to predict simple relationships between the criti-
cal load and the relevant column characteristics.

In structural engineering practice and due to manufacturing, transportation and/or handling restraints,
individual steel elements are usually fabricated with a maximum length of 12 meters. During erection of
a steel frame and where the element length is insufficient, splices are provided to form a single and
longer element. Designers often use the splices for changing cross-sections, in view of a more economical
and rational design.
In steelwork construction, column splices are located at a convenient distance for erection and con-
struction above floor beam level and have to be designed (i) to join lengths in line, (ii) to transmit forces
and moments between the connected member parts and (iii) to maintain continuity of strength and stiff-
ness through the splice to safeguard the robustness of the structure [1]. Column splices are usually disre-
garded in determining the distribution of moments and forces in the structure and when the design of the
columns itself is being considered, assuming that the splice is providing full continuity in stiffness and
strength of the column. This practice is questionable as the splices most times do not provide this conti-
nuity. So, the splices may adversely affect the overall frame behaviour, from a stiffness and strength
point of view.
Previous research pertaining to the load-carrying capacity of spliced columns includes investigations
by Lindner [2], Snijder and Hoenderkamp [3] and Girão Coelho et al. [4,5,6]. Lindner [2] carried out
experimental and numerical tests on different column splice types and highlighted the existence of eccen-
tricities at the splice. An adjusted buckling curve for columns having contact splices at column mid-
height was later proposed Lindner [2]. Snijder and Hoenderkamp [3] conducted a series of experimental
tests to analyse the influence of end plate splices on the load-carrying capacity of slender columns. These

Pedro D. Simão et al.

tests were used to make design recommendations for column splices. Girão Coelho and co-authors [4,5]
further extended this work to produce a relatively simple yet reasonably accurate engineering method for
predicting the critical behaviour of spliced columns in steel frames. This paper is a follow-up study to
this research.
The current work presents a generalized energy formulation of a framed spliced column in sway and
non-sway frames (Figure 1). For analytical modelling, a framed column is represented by means of ex-
tensional and rotational restraints at the ends A and B. The splice is modelled as a rotational spring at
point C. The potential energy functional of this system uses a Rayleigh-Ritz approximation of the rele-
vant deformation modes of the column. This formulation is presented together with the method of La-
grange multipliers to deal with the constraints at the column splice. Elastic buckling analysis is carried
out to find the critical load of the system. The concept of end fixity factor C [7] is successfully applied
and the significance of this factor in simplifying the analysis of results is emphasized. Simple relation-
ships between C and the relevant characteristics of the column and splice are derived to a point where the
critical load can be readily determined by hand or by computer.



LII Member II:


LI Member I:

z, w

Figure 1: The framed spliced column system.


2.1 Bending and axial strain energy

The deformation of a prismatic member under the action of loads is characterized by axial elongation
(mode 1) and bending deformations (mode 2). In the context of a simplified Generalized Beam Theory
(GBT) strategy [8,9], the displacement functions are assumed to be as follows:
ª1  z1 º
2 « »
Axial displacements: u x, y, z ¦ u y, z u f x
i i
u and u 1
» u
u ¼º «# # » (1)
i 1
« »
«¬1  zn »¼
ª0 1º
« »
>w @
Transverse displacements: w x, y, z ¦ w y, z u f x
i i
w and w 1 2
w « # #» (2)
i 1 « »
«¬0 1»¼

Pedro D. Simão et al.

whereby zj is the distance between point j and the neutral axis, iu(y, z) and iw(y, z) are pre-established
modal displacement patterns defined along the member cross-section, and ifu(x) and ifw(x) are modal
amplitude functions for warping and transverse displacements, respectively. For any mode of deforma-
tion the amplitude functions for axial and transverse displacements are related in the form [8,9]:
d ª¬ i f w x º¼ i
fu x f wc x i f c (3)
The extensional strain of a column segment of length dx is readily defined as:
2 2
§ wu · § ww ·
Hx¨1  wx ¸  ¨ wx ¸  1 (4)
© ¹ © ¹
Expansion in Taylor series, neglecting higher-order terms, yields the following kinematic relation:
wu 1 § ww · 1 2 2 k l k l
Hx |  ¨
wx 2 © wx ¹ k 1

¸ ¦ u f cc  ¦¦ w w f c f c
k k

2k 1l 1

From Hooke’s law (constitutive relation), the longitudinal stress is then given by:
ª 2 1 2 2 º

V x EH x E «¦ i u i f cc  ¦¦ i w j w i f c j f c »
¬i 1 ¼
The internal strain energy of the member, Um, is then equal to [5]:
1 EA 1 2 EI 2 2 EA 2 21 EA 2 4
Um ³ V x H x d:
2 2 ³
f cc d x 
2 ³
f cc d x 
2 ³ f c f cc d x 
8 ³L
fc dx (7)
: L L L

where ȍ denotes the member’s volume, A the cross sectional area and I the moment of inertia.
2.2 Strain energy stored in the springs
The energy stored in the linear springs (Figure 1) is given by the following expressions:
1 E I II 2
Rotational spring at end A: U șa
K șa T a2
kșa 2 f Ic
x 0

1 E I II
Rotational spring at end B: U șb K șb T b2 kșb 2 f IIc (9)
2 2L x LII

1 E I II
Rotational spring at splice: U șc K șc T c2 kșc 2 f IIc  2 f Ic (10)
2 2L x 0 x L I

1 E I II
Extensional spring at end B: U 'b K 'b ' b2 k 'b
f II (11)
2 2 L3 x LII

where KT and K' are rotational and extensional spring constants, respectively, and k are spring coeffi-
cients that are defined in non-dimensional form.
2.3 Work done by load
The final component of energy to be identified is the work done by the load. For a centrally loaded
column, the potential energy of the external loading is given by:
3  N Ed u 1 f IIc (12)

2.4 Potential energy functional

The total potential energy of the complete structure is a summation of Um (for each individual mem-
ber I and II), UTa, UTb, UTc, U'b minus 3:
V U m,I  U m,II  U șa  U șb  U șc  U 'b  3 (13)
This functional is subjected to the following kinematic constraints that ensure continuity at the splice:

Pedro D. Simão et al.

§ ·
uI,C  uII,C 0 œ 1 i
f Ic
¨¨ G1 a j , N Ed 0 ¸¸
x LI
 1 f IIc
x 0
© ¹
§ ·
wI,C  wII,C 0 œ 2
x LI
 f II
x 0
0 ¨ G2 a j , N Ed 0 ¸
¨ ¸
© ¹
We now wish to find a stationary value of a functional subjected to some subsidiary conditions or con-
straints Gk(iaj,NEd). The problem is easily tackled by using the approach proposed by Lagrange [10]. The
technique is to form a modified potential energy expression:

V i a j , Ok , N Ed
V i a j , N Ed  ¦ Ok Gk i a j , N Ed
k 1

where Ok are the Lagrange multipliers. For the spliced column, the modified potential energy functional
may be written as:
V U m  U șa  U șb  U șc  U 'b  3  O1 fc
I x L
 1 f IIc
x 0 O f 2
I x L
 1 f II
x 0 (16)

2.5 Critical buckling load

Eq. (16) is a functional representing the total potential energy of the physical system. The advantage
of this method lies in the fact that the problem with constraints can be treated in exactly the same manner
as though it was free. Thus, for the system to satisfy equilibrium V has to be stationary. The calculus of
variations is then used to find the stationary point of the functional. Exact solutions can be obtained using
the method of eigenvalue analysis. This is not a practical method to solve the characteristic equations of
the differential equations. Approximate methods such as the Rayleigh-Ritz method seem a very attractive
alternative to an otherwise complex problem. Essentially, in this method, the modes of deformation of the
system are defined by means of assumed displacement functions that satisfy the geometric boundary
conditions of the system. As a result, the total potential energy reduces from a functional to a gradient
potential function that depends on a finite set of discrete generalized coordinates iaj and Ok and the scaling
factor NEd. Thus, ordinary calculus can be used to obtain solutions directly.
In the context of the Rayleigh-Ritz method, the amplitude functions are approximated by polynomials
in the form:
f | ¦ i a j iM j (17)
j 1

These polynomials form a set of coordinate functions that satisfy the kinetic boundary conditions of the
problem and are orthonormal functions that enable fast convergence of the method [9]. The coordinate
functions are given by:
Member I:
M1,I 5 LI x 2 LI
M1,I 3 LI x
M 2,I
5 LI 3 x  4 x 2 LI
7 LI 6 x  20 x LI  15 x3 LI 2 2
9 LI 10 x  60 x LI  105 x LI  56 x 4 LI3
2 3 2

11 LI 15 x  140 x 2 LI  420 x3 LI 2  504 x 4 LI3  210 x5 LI 4
13 LI 21x  280 x 2 LI  1260 x3 LI 2  2520 x 4 LI 3  2310 x5 LI 4  792 x 6 LI5
15 LI 28 x  504 x LI  3150 x LI  9240 x LI  13860 x LI  10296 x LI  3003 x 7 LI 6
2 3 2 4 3 5 4 6 5

Pedro D. Simão et al.

Member II:
M 2,II 3 LII 1  x LII
M2,II 3 LII  LII  2 x
5 LII LII  6 x  6 x 2 LII
7 LII  LII  12 x  30 x 2 LII  20 x 3 LII 2
9 LII LII  20 x  90 x LII  140 x LII  70 x 4 LII 3
2 3 2

M6,II 2
11 LII  LII  30 x  210 x LII  560 x LII  630 x LII  252 x 5 LII 4 3 2 4 3

15 LII LII  42 x  420 x LII  1680 x LII  3150 x LII  2772 x LII  924 x 6 LII5
2 3 2 4 3 5 4

Equilibrium of the system is obtained by rendering stationary the total potential function with respect
to the generalized coordinates iaj and Ok. The solution that emerges from the unloaded state, the funda-
mental path (FP) is a function of NEd. A sliding set of incremental coordinates iqj and qk is then defined by
the following equations [11]:
a j i a j FP N Ed  i q j and Ok Ok FP N Ed  qk (20)
A new energy function W is now introduced [11]:

W i q j , qk , N Ed
V i a j FP  i q j , Ok FP  qk , N Ed (21)
A global numbering for coordinates qj and qk (ql) can now be adopted. The equilibrium and stability
conditions hold good for this transformed energy function W. In this new NEd – ql space, the fundamental
path is defined trivially by ql = 0.The critical points along the fundamental path are now those points that
render zero the determinant of the total potential energy Hessian matrix along the fundamental path:
H FP H FP,0  N Ed H FP,1 (22)
The relevant states of critical equilibrium are identified via a local linear eigenvalue equation
HFP q = 0, q representing the local eigenvector [11]. By substituting the forms in Eq. (22), we thus obtain
the critical state identity:
H FPq H FP,0  N Ed H FP,1 q 0 (23)
This analysis yields the critical buckling load of the spliced column, Ncr that can be expressed in terms of
an end fixity factor C [7]:
N cr CS 2 EI II L2 (24)

The purpose of this numerical study is to ascertain the effect of the following variables on the general
equilibrium response: (i) splice location (LI = DL), (ii) ratio between second moment of area of lower and
upper column members (E = II/III), (iii) splice rotational stiffness (kTc) and (iv) end-restraints stiffness
coefficients (kTa, kTb and k'b). Results are independent from the column length L. These properties are
varied parametrically as shown in Table 1.

Table 1: Parameters for regression analysis

Parameter Range of parameter selected
Splice location D 0.1 to 0.9, i = 0.2
Ratio between second moment of areaE 1 to 3, i = 0.5
Non-dimensional stiffness coefficients o0, 0.25, 0.5, 1, 3, 5, 7.5,10, 15, 20, 35, 50, 75, 100

Pedro D. Simão et al.

These results form a comprehensive analytical database. We can now generate a continuous function
that approximates the values of the end fixity factor within the domain of analyses and with a minimum
error. The analysis results are then used to develop a multiple regression model to approximate the end
fixity factor C { Cfit(Į, ȕ, kșa, kșb, kșc, kǻb) from the data in the database, by means of piecewise approxi-
mations. In developing the regression model, the relationship between the dependent variable and each
independent variable is studied separately, while all other independent variables are kept constant. The
dependent variable is approximated by a continuous function that is linear in terms of a set of regression
coefficients, which are determined by enforcing the method of least squares that minimizes the sum of
the squares of the residuals. Approximating (or coordinate) functions are then selected for each inde-
pendent variable. The multiple regression model is formed as the product of the individual coordinate

kTa = 0.01, kTb = 100, kǻb = 5 kTa = 100, kTb = 100, kǻb = 0.01
kTc = 0.01, 1, 5, 10, 20, 100 kTc = 0.01, 1, 5, 10, 20, 100

ȕ = 2, Į = 0.5, kǻb = 0.01 ȕ = 1, Į = 0.1, kǻb = 5

kTc = 0.01, 1, 5, 10, 20, 100 kTc = 0.01, 1, 5, 10, 20, 100

ȕ = 1, kTa = 100, kTb = 100 ȕ = 1, kTa = 0.01, kTb = 100

Į = 0.1, 0.3, 0.5, 0.7, 0.9 Į = 0.1, 0.3, 0.5, 0.7, 0.9
Figure 2: Variation of C with the relevant properties.

Pedro D. Simão et al.

Forming the regression model as a product allows the effect of each independent variable to be exam-
ined separately and facilitates the process of selecting suitable coordinate functions for the individual
independent variables. Some key results are illustrated graphically in Figure 2. The graphics suggest that:
1. The response C vs. D can be approximated by a quadratic function.
2. Typical C vs. Ebehaviour is characterized by a monotonic increasing function that can be gener-
ally approximated by a simple linear relationship.
3. The degree of rotational and extensional end restraint is an essential parameter for the computation
of C. The shape of the curves C vs. k (k { kTa, kTb), C vs. kTc and C vs. k'b is best described by an arctan-
gent function.
The end fixity factor is then predicted by means of an expression in the form:
ª § k ·º

Cfit D , E , k'b , kșa , kșb , kșc C1 1  C2D  C3D 2 1  C4 E «1  C5 arctan ¨¨ 'b ¸¸ » u
¬« © C6 ¹ ¼»
ª § kșa · º ª § kșb · º ª § kșc · º
u «1  C7 arctan ¨¨ ¸¸ » «1  C9 arctan ¨¨ C ¸¸ » «1  C11 arctan ¨ C ¸ »
¬« © C8 ¹ ¼» ¬« © 10 ¹ ¼» ¬« © 12 ¹ ¼»
where Ci are regression coefficients.
As expected in developing a predictive regression model, many models were tried, analysed and as-
sessed for accuracy and effectiveness. The final model presented here evolved out of several attempts to
develop conventional (nonlinear) regression models by means of simple mathematical functions. The
overall character of the response is well captured and the number of regression coefficients is kept small
in order to provide a compact procedure for the simplified method. The accuracy of the model is meas-
ured by means of the R-Squared value (R2). The R-Squared gives the fraction of the variation of the
response that is predicted by the model. A good model fit yields values of R-squared close to unity.
Nonlinear regression analysis is performed with the Mathematica software [12]. Regression coeffi-
cients are determined for the spliced column using piecewise approximations depending on the nature of
the segments that comprise the above relationship. The domain of analyses of the spring coefficients is
divided into three intervals: G1 for k  ]0,3], G2 for k  ]3,15] and G3 for k  ]15,100]. Table 2 sets out
the computed regression coefficients and values for the R-Squared factor are also given.


The paper has presented an application of the total potential energy method to the buckling behaviour
of a spliced column in sway and non-sway frames. This is a variational problem, i.e. the finding of thesta-
tionary point of a functional, with additional conditions at the splice that is solved by means of the La-
grange’s method of undetermined multipliers. The Rayleigh-Ritz procedure has been used to reduce this
variational problem with constraints to a mere differentiation that form a set of algebraic equations of
equilibrium. These equations are solved by using an algebraic manipulator >12@ and the critical load is
calculated. The buckling response was found to be particularly sensitive to the following variables: (i)
splice location, (ii) ratio between second moment of area of lower and upper column members, (iii) splice
rotational stiffness and (iv) end-restraints stiffness coefficients. The significance of each of these vari-
ables has been assessed. A parametric study was devised and the results were then used to develop re-
gression equations for predicting the critical load of the system via the concept of end fixity factor C.
The work outlined above affords some basis to produce design guidance on column splices. The au-
thors attempt to set up sound design criteria regarding the requirements for stiffness and strength of col-
umn splices. Experimental and numerical finite element studies focusing on the buckling response are
also necessary in order to validate the predictive expressions.
It should be noted that the investigated configuration was rather limited to a particular case. The deriva-
tion has been carried for two-dimensional frames and only uniaxial bending behaviour has been considered.
Future work will incorporate bi-axial bending and torsion effects in the design of splices. In addition to this,
the influence of splices on the overall stability behaviour of frames will also be investigated.

Pedro D. Simão et al.

Table 2: Parameter table.

kTa  G1 kTa  G1 kTa  G1 kTa  G2 kTa  G2 kTa  G2 kTa  G3 kTa  G3 kTa  G3
kTb  G1 kTb  G2 kTb  G3 kTb  G1 kTb  G2 kTb  G3 kTb  G1 kTb  G2 kTb  G3
C1 0.0188 0.0536 0.0621 0.0605 0.0890 0.1005 0.0685 0.0865 0.0995
C2 -1.0746 -0.8905 -0.5538 -0.9201 -0.6922 -0.4441 -0.6239 -0.3152 -0.0097
C3 1.6481 1.5441 1.2462 1.2899 1.1190 0.8795 1.0100 0.7906 0.4993
C4 0.2417 0.2390 0.2456 0.1874 0.1860 0.1934 0.2202 0.2304 0.2437
C5 8.5978 4.9043 4.7424 3.6457 2.7180 2.7107 3.2346 2.4620 2.4769
C6 8.1153 13.0053 16.9895 11.4866 15.2379 19.5321 16.3117 20.9021 26.3750
C7 0.2736 0.3294 0.3289 0.4854 0.6565 0.7675 0.5434 0.6613 0.7640
C8 1.5 1.5 1.5 3.0 3.0 3.0 5.0 5.0 5.0
C9 0.3438 0.4419 0.3667 0.4213 0.5721 0.3987 0.4304 0.6923 0.4497
C 10 1.5 3.0 5.0 1.5 3.0 5.0 1.5 3.0 5.0
C 11 1.5929 0.6725 0.6498 0.9489 0.5607 0.5687 0.8628 0.5690 0.5945
C 12 1.3821 1.6983 1.8891 1.5744 1.9524 2.3116 1.7536 2.1201 2.6502
R2 0.9043 0.8986 0.9115 0.9115 0.9311 0.9447 0.9057 0.9381 0.8874

[1] CEN (European Committee for Standardization), EN 1993-1-8 – Eurocode 3: Design of steel struc-
tures – Part 1-8: Design of joints, Brussels, 2005.
[2] Lindner, J., “Old and new solutions for contact splices in columns”. Journal of Constructional Steel
Research, 64, 833-844, 2008.
[3] Snijder, H.H., Hoenderkamp, J.C.D., “Influence of end plate splices on the load carrying capacity
of columns”. Journal of Constructional Steel Research, 64, 845-853, 2008.
[4] Girão Coelho A.M., Bijlaard F.S.K., “Requirements for the design of column splices”, Stevin Re-
port 6-08-3, Delft University of Technology, 2008.
[5] Girão Coelho A.M., Simão P.D., Bijlaard F.S.K., “Stability design criteria for steel column
splices”. Journal of Constructional Steel Research, 66, 1261-1277, 2010.
[6] Girão Coelho A.M., Bijlaard F.S.K., Simão P.D., “Stability design criteria for steel column splices
in non-sway frames”. Proceedings of the fourth international conference on structural engineering,
mechanics and computation SEMC 2010, (accepted for publication).
[7] Simitses, G.J., An introduction to the elastic stability of structures, Krieger Publishing Company,
Malabar, 1986.
[8] Schardt, R., Verallgemeinerte Technische Biegetheorie, Springer, Berlin-Heidelberg, Germany,
[9] Simão, P.D., Post-buckling bifurcational analysis of thin-walled prismatic members in the context
of the Generalized Beam Theory, Ph.D. thesis, University of Coimbra, Portugal, 2007
[10] Richards, T.H., Energy methods in stress analysis, Ellis Horwood, Chichester, UK, 1977.
[11] Thompson J.M.T., Hunt G.W., A General Theory of Elastic Stability, John Wiley & Sons, London,
UK, 1973.
[12] Mathematica 6, Wolfram Corp., Champaign, USA, 2007.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Tadeh Zirakian

Ph.D. Student, Department of Civil & Environmental Engineering

University of California, Los Angeles
E-mail: tzirakian@ucla.edu

Keywords: Buckling, Elasticity, Extrapolation Techniques.

Abstract. The experimental determination of critical buckling load of structures undergoing lateral
buckling has usually been accompanied by the application of certain and just a few deformation
characteristics such as lateral displacement and/or twist. This paper explores the possibility of
application of various deformation variables such as web transverse and longitudinal strains, vertical
deflection, and angles of twist of top and bottom flanges for experimental determination of the critical
buckling load of I-beams with different initial geometrical imperfections undergoing elastic lateral-
distortional buckling. After demonstrating the linear relationship between lateral displacement and the
various aforementioned deformation variables, the four Southwell, Massey, Modified, and Meck
extrapolation techniques are applied on these various deformation variables, and consequently
satisfactory estimates are acquired for the critical buckling loads.

The extrapolation or plotting techniques are experimental methods developed for determining
experimentally the critical buckling load of structures, without having to test them to failure. By plotting
the results of a structure test in a certain manner, it would be possible to determine the structure’s
buckling load.
Southwell [1] initially proposed a plotting method for a concentrically loaded pin-ended column with
a sinusoidal initial imperfection. Later on, Massey [2], Trahair [3], and Meck [4] successfully applied this
method and variations of it to predict buckling loads for beams.
A search of the literature shows that the extrapolation techniques have mostly been used on certain
and just a few deformation characteristics such as lateral displacement and/or twist, and also Mandal and
Calladine [5] demonstrated that lateral displacement tends to be proportional to rotation as deformations
increase in lateral-torsional buckling mode. In a recent research work reported by the author [6], it was
demonstrated that lateral displacement in I-beams undergoing lateral-distortional mode of buckling tends
to be directly coupled with the web transverse strains developed as a result of occurrence of web
distortion, and accordingly the application of the extrapolation techniques on the web transverse strains
yielded good predictions for the critical buckling load.
Based on findings of the previous studies, the possibility of application of various deformation
variables for experimental determination of the critical buckling load is explored in this paper. Initially,
the relationship between lateral displacement and the various considered deformation variables including
web transverse and longitudinal strains, vertical deflection, and angles of twist of top and bottom flanges
of I-beams with different initial geometrical imperfections is investigated, and subsequently the
applicability of the Southwell, Massey, Modified, and Meck Plots on the aforementioned deformation
variables is studied.

Tadeh Zirakian

Four finite element lateral-distortional buckling solutions are developed using the ABAQUS software
system [7]. All of the solutions are of simply-supported steel I-beams subjected to uniform bending
moment with identical cross-section dimensions hw=900 mm, bf=240 mm, tw=5 mm, tf=20 mm, and
L=7000 mm. The material properties adopted for the beams are E=200 GPa, v=0.3, Fy=345 MPa, and
G=0.385E. The beam cross-section components, i.e. flanges and the web, were modeled using a four-
node shell element S4R5. Finally, the details of various geometrical imperfection forms of the beams are
provided in Table 1.

Table 1: Details of various geometrical imperfection forms of the beams

Initial imperfections at center of beam
Beam number Initial imperfection form
Crookedness (mm) Twist (rad)
1 - - -
2 Half-sine wave along the length - 0.04363
20 (lateral displacement)
3 Half-sine wave along the length -
20 (vertical displacement)
20 (lateral displacement)
4 Half-sine wave along the length 0.04363
20 (vertical displacement)


Before applying the various deformation variables for experimental determination of the critical
buckling load, the relationship between lateral displacement and the considered deformation variables is
explored. It should be noted that the lateral displacement of the top flange is applied in this study.
Lateral displacement vs. web transverse strain: In this case, the relationship between lateral
displacement and web transverse strain representing web distortion and measured at midspan and
midheight is investigated. The acquired data are plotted straightforwardly as lateral displacement against
web transverse strain, as shown in Figure 1. The linear equations obtained using the least squares method
and the respective R-squared values are displayed in the figures. As seen in the figures, it is evident that
after the initial stages of loading the two deformation characteristics become proportional to each other.

Lateral displacement vs. web longitudinal strain: In addition to the web transverse strain, the
relationship between lateral displacement and web longitudinal strain is investigated as well. As seen in
Figure 2, lateral displacement is plotted against web longitudinal strain (measured at midspan and
midheight), and the obtained linear equations as well as the R-squared values are displayed on the chart.
It is clearly observed that after the initial loading stages, the data points align with the linear portion near
the latter loading stages and the direct coupling of the two deformation characteristics becomes evident.

Tadeh Zirakian

400 400

350 350

į L = -132388İ T + 310.46 300 į L = -126881İ T + 309.21 300

2 2
R = 0.9962 250 R = 0.9964
įL (mm)

įL (mm)
200 200

150 150

100 100

50 50

0 0
-0.0003 -0.0002 -0.0001 0 0.0001 0.0002 0.0003 -0.0004 -0.0002 0 0.0002 0.0004

(a) No imperfection (b) Initial twist

400 400

350 350

300 300
į L = -139884İ T + 282.27 į L = -127840İ T + 284.66
250 2 250
R = 0.9977 R = 0.9971
įL (mm)

įL (mm)
200 200

150 150

100 100

50 50

0 0
-0.0006 -0.0004 -0.0002 0 0.0002 0.0004 -0.0006 -0.0004 -0.0002 0 0.0002 0.0004

(c) Initial crookedness (d) Initial twist and crookedness

Figure 1: Plot of lateral displacement against web transverse strain
400 400

350 350

300 300
į L = 87525İ L + 253.42
į L = 92964İ L + 247.77
250 250 2
2 R = 0.9979
R = 0.9978
įL (mm)

įL (mm)

200 200

150 150

100 100

50 50

0 0
-0.0005 0 0.0005 0.001 0.0015 -0.0005 0 0.0005 0.001 0.0015

(a) No imperfection (b) Initial twist

400 400

350 350

300 300

250 250 į L = 93404İ L + 219.34

į L = 103760İ L + 206.15
įL (mm)
įL (mm)

200 2 200 R = 0.9983
R = 0.9987

150 150

100 100

50 50

0 0
-0.0005 0 0.0005 0.001 0.0015 -0.0005 0 0.0005 0.001 0.0015

(c) Initial crookedness (d) Initial twist and crookedness

Figure 2: Plot of lateral displacement against web longitudinal strain

Tadeh Zirakian

Lateral displacement vs. vertical deflection: The proportionality between the lateral displacement
and vertical or in-plane deflection at midspan and midheight of the analyzed I-beams undergoing lateral-
distortional buckling is investigated in here. The plots of lateral displacement against vertical deflection
are shown in Figure 3. The linear equations and R-squared values are displayed in the figures as well. To
a fair approximation, lateral displacement and vertical deflection seem to be proportional to each other
and the linearity range is comparatively large in this case.

400 400

350 350

300 300

250 250
į L = 3.0863į V + 113.9
įL (mm)

įL (mm)
200 2 200
R = 0.9979
į L = 4.1982į V + 15.085
150 150 2
R = 0.9807
100 100

50 50

0 0
0 20 40 60 80 0 20 40 60 80 100
į V (mm) į V (mm)

(a) No imperfection (b) Initial twist

400 400

350 350

300 300

250 250
įL (mm)
įL (mm)

200 200
į L = 4.41į V + 6.6499
150 į L = 4.8986į V - 3.5458 150
R = 0.9922
R 2 = 0.992
100 100

50 50

0 0
0 20 40 60 80 0 20 40 60 80 100
į V (mm) į V (mm)

(c) Initial crookedness (d) Initial twist and crookedness

Figure 3: Plot of lateral displacement against vertical deflection

Lateral displacement vs. angle of twist: Lastly, the relationship between the lateral displacement and
angles of twist of top and bottom flanges of the I-beams is examined. Unlike the lateral-torsional mode of
buckling, in lateral-distortional buckling mode top and bottom flanges have different angles of twist,
hence the two angles of twist are taken into consideration in this study. Plots of lateral displacement
against angles of twist of top and bottom flanges are made and shown in Figure 4. It is evident that lateral
displacement and angles of twist of the two flanges are directly coupled.

Tadeh Zirakian

400 400

350 į L = -375.69ș TF + 322.61 350 į L = -341.54ș TF + 320.67

R = 0.9954 R 2 = 0.9959
300 300

250 250
įL (mm)

įL (mm)
200 200

150 į L = 1377.3ș BF + 2.9492 150 į L = 1372.5ș BF + 11.653

R = 0.9993 R 2 = 0.9984
100 100

50 50

0 0
-0.1 0 0.1 0.2 0.3 -0.1 0 0.1 0.2 0.3
ș (rad.) ș (rad.)

(a) No imperfection (b) Initial twist

400 400

350 350

300 300
į L = -410.34ș TF + 278.89 į L = -361.52ș TF + 281.38
2 250 250
R = 0.9981 2
R = 0.9965
įL (mm)

įL (mm)

150 150
100 į L = 1287ș BF - 3.0457 100
R 2 = 0.9999 į L = 1292.1ș BF + 4.0579
50 2
R = 0.9996
-0.2 -0.1 0 0.1 0.2 0.3
-50 -0.2 -0.1 0 0.1 0.2 0.3
ș (rad.) ș (rad.)

(c) Initial crookedness (d) Initial twist and crookedness

Figure 4: Plot of lateral displacement against angle of twist

Based on the linear relationship between lateral displacement and the other considered deformation
variables, it seems logical to obtain straight lines by applying the extrapolation techniques on the various
deformation variables, and also acquire favorable predictions for the buckling load. To prove this, the
following five case studies are performed.
Lateral displacement: In this case, Southwell, Massey, and Modified Plots are applied on the lateral
displacement of the top flange at midspan. As an example, the Southwell, Massey, and Modified Plots for
beam 4 are shown in Figure 5. The extrapolated-to-ultimate failure moment ratios for the three methods
are also given in Table 2.

Tadeh Zirakian

0.35 8.00E-04 400000

0.3 7.00E-04 350000

6.00E-04 300000

5.00E-04 250000

įL / M2

įL * M
įL / M

4.00E-04 200000
3.00E-04 150000

2.00E-04 100000

0.05 1.00E-04 50000

0 0.00E+00 0
0 100 200 300 400 0 100 200 300 400 0 100 200 300 400
įL įL įL

(a) Southwell Plot (b) Massey Plot (c) Modified Plot

Figure 5: Southwell, Massey, and Modified Plots (beam 4)

Table 2: Comparison of the ultimate failure moments with the extrapolated moments for the case of
lateral displacement
Beam number MSouthwell/Mu MMassey/Mu MModified/Mu
1 1.038 0.992 1.025
2 1.069 1.352 1.051
3 1.137 1.174 1.189
4 1.165 1.474 1.232
Average Discrepancy (%) 9.07 18.44 10.50

In spite of some scatter in the results, it can be seen from the table that the extrapolated moments are
generally in good agreement with the ultimate failure moments. Moreover, as it is seen, the lowest
average discrepancy is found in the case of the Southwell Plot.
Web transverse strain: The four Southwell, Massey, Modified, and Meck Plot methods are applied
on the web transverse strains captured at the mid-height and midspan of the analyzed beams. The
extrapolated-to-ultimate failure moment ratios for the plotting methods are given in Table 3.

Table 3: Comparison of the ultimate failure moments with the extrapolated moments for the case of web
transverse strain
Beam number MSouthwell/Mu MMassey/Mu MModified/Mu MMeck/Mu a
1 1.038 0.992 0.995 1.040
2 0.950 1.022 1.003 1.033
3 1.010 1.017 1.002 1.009
4 1.036 1.042 1.008 0.992
Average Discrepancy (%) 3.29 2.15 0.45 2.18
MMeck is obtained as a result of the use of Meck Plot method on lateral displacement and web transverse

Tadeh Zirakian

As it is seen in Table 3, some scatter in the results is present in this case as well. In general, the
agreement between the extrapolated and the ultimate failure moments is satisfactory in all cases. Lastly, in
this case, the lowest average discrepancy is found in the case of the Modified Plot.
Web longitudinal strain: In this case, the use of Southwell, Massey, Modified, and Meck Plots on
the web longitudinal strains captured at the mid-height and midspan of the considered I-beams, is
investigated. The extrapolated-to-ultimate failure moment ratios for the four considered Plot methods are
presented in Table 4.

Table 4: Comparison of the ultimate failure moments with the extrapolated moments for the case of web
longitudinal strain
Beam number MSouthwell/Mu MMassey/Mu MModified/Mu MMeck/Mu a
1 1.038 0.992 1.044 1.014
2 1.069 1.022 1.051 1.092
3 1.010 1.087 1.044 1.018
4 1.036 1.114 1.059 1.091
Average Discrepancy (%) 3.64 5.28 4.73 4.96
MMeck is obtained as a result of the use of Meck Plot method on lateral displacement and web
longitudinal strain.

As can be seen from the results in Table 4, despite some scatter, the agreement between the
extrapolated and the ultimate failure moments is satisfactory. Furthermore, in this case, the lowest
average discrepancy is found in the case of the Southwell Plot.
Vertical deflection: The applicability of the extrapolation techniques on the beam midspan vertical
deflection is studied in this case. The extrapolated-to-ultimate failure moment ratios for the four applied
plotting methods are given in Table 5.

Table 5: Comparison of the ultimate failure moments with the extrapolated moments for the case of
vertical deflection
Beam number MSouthwell/Mu MMassey/Mu MModified/Mu MMeck/Mu a
1 1.038 1.072 1.025 1.022
2 1.069 1.209 1.040 1.104
3 1.137 1.174 1.126 1.010
4 1.165 1.318 1.154 1.077
Average Discrepancy (%) 9.07 15.74 7.72 4.92
MMeck is obtained as a result of the use of Meck Plot method on lateral displacement and vertical

In spite of the scatter in the results, it may be concluded that the extrapolated and the ultimate failure
moments are generally in good agreement in all cases. Also, the lowest average discrepancy in the present
case study is interestingly found in the case of the modified Meck Plot.
Angle of twist: In the last case study, the four considered extrapolation techniques are applied on the
angles of twist of both top and bottom flanges captured at midspan of the analyzed beams. The
extrapolated-to-ultimate failure moment ratios for the Southwell, Massey, Modified, and Meck Plot
methods are presented in Table 6.

Tadeh Zirakian

Table 6: Comparison of the ultimate failure moments with the extrapolated moments for the case of angle
of twist
Beam number Locationa MSouthwell/Mu MMassey/Mu MModified/Mu MMeck/Mu b
TF 1.038 0.992 1.001 0.948
BF 1.038 0.992 1.022 1.007
TF 0.950 1.022 1.002 1.037
BF 1.069 1.209 1.042 1.046
TF 1.010 1.017 1.007 0.956
BF 1.137 1.286 1.181 1.043
TF 1.036 1.042 1.011 1.025
BF 1.165 1.474 1.209 1.066
TF 3.29 2.15 0.50 3.92
Average Discrepancy (%)
BF 9.07 18.11 9.72 3.86
TF and BF stand for “top flange” and “bottom flange”, respectively.
MMeck is obtained as a result of the use of Meck Plot method on lateral displacement and angle of twist.
As it is seen in Table 6, despite some scatter in the results, the agreement between the extrapolated
and the ultimate failure moments is generally satisfactory in both cases. Finally, in this case study, the
lowest average discrepancies for the cases of angles of twist of top and bottom flanges are found in the
cases of Modified and Meck Plots, respectively.

Based on the key findings regarding the proportionality between lateral displacement and other
considered deformation variables including web transverse and longitudinal strains, vertical deflection,
and angles of twist of top and bottom flanges of the I-beams undergoing elastic lateral-distortional
buckling, the applicability of the Southwell, Massey, Modified, and Meck extrapolation techniques on the
various deformation variables was investigated in this paper and generally satisfactory and reliable results
were obtained. The results of this study may be considered as an indication of a great extension in the
application of the extrapolation techniques.

[1] Southwell, R.V. “On the analysis of experimental observations in the problems of elastic stability”.
Proc. of the Royal Philosophical Society of London, 135(A), 601, 1932.
[2] Massey, C. “Elastic and inelastic lateral instability of I-beams”. The Engineer, 216, 672-674, 1963.
[3] Trahair, N.S. “Deformations of geometrically imperfect beams”. Proc. of ASCE, Journal of the
Structural Division, 95(ST7), 1475-1496, 1969.
[4] Meck, H.R. “Experimental evaluation of lateral buckling loads”. Proc. of ASCE, Journal of the
Engineering Mechanics Division, 103, 331-337, 1977.
[5] Mandal, P. and Calladine, C.R. “Lateral-torsional buckling of beams and the Southwell Plot”.
International Journal of Mechanical Sciences, 44, 2557-2571, 2002.
[6] Zirakian, T. “Lateral-distortional buckling of I-beams and the extrapolation techniques”. Journal of
Constructional Steel Research, 64(1), 1-11, 2008.
[7] ABAQUS analysis user’s manual, Version 6.5. ABAQUS, Inc., 2005.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Tadeh Zirakian* and Jian Zhang

* Department of Civil & Environmental Engineering, University of California, Los Angeles

E-mail: tzirakian@ucla.edu

Keywords: Lateral-Distortional Buckling, Elasticity, Singly Symmetric I-Beams, AISC Code.

Abstract. It is clear from prior research studies that the web distortional flexibility can lead to a
substantial reduction relative to the beam theory lateral-torsional buckling resistance for I-sections with
stocky flanges and slender webs. Hence, the 2005 AISC Specification gives specific rules for controlling
the unconservative errors due to the neglect of web distortion effects. The accuracy of the 2005 AISC
code predictions in case of elastic lateral-distortional buckling of singly symmetric I-beams is
investigated in this paper through comparison with the accurate finite strip analysis distortional buckling
solutions as well as the theoretical predictions of two elastic distortional buckling design equations
proposed by other researchers. The code predictions in case of lateral-distortional buckling of slender-
web singly symmetric I-beams are found to be by and large conservative, and even overconservative in
some cases.

For the slender-web I-sections, the 2005 AISC Specification [1] bases the lateral-torsional buckling
resistance on Eq. (1), but the St. Venant torsional constant J is taken equal to zero.

CbS 2 E J (1)
Fcr 1  0.078 Lb rt 2
Lb rt 2 S x ho

In fact, the implicit use of J = 0 in Section F5 of the 2005 AISC Specification is intended to account
for the influence of web distortional flexibility on the lateral-torsional buckling resistance for slender-web
I-section members [2].
In addition to the destabilizing effect of web distortion in a slender-web singly symmetric I-beam
which results in lowering of the torsional rigidity of the beam, this may be coupled with the influence of
the Wagner effect to reduce significantly the buckling strength of the singly symmetric beam [3].
This paper focuses on distortional buckling of singly symmetric I-shaped flexural members with
slender webs, and evaluates the effectiveness of the 2005 AISC code rules by comparing the code
predictions with finite strip analysis (FSA) distortional buckling solutions developed using the finite strip
analysis software CUFSM [4] as well as the theoretical predictions of Bradford’s (Eq. (2)) [5] and Wang
et al.’s (Eq. (3)) [6] proposed distortional buckling design equations in the elastic range of structural

Tadeh Zirakian and Jian Zhang

Fcrd 490 t f b ft t f t w 1  0.560 b fc b ft (2)

Fcr E Fcr

J f<>)D  10 1  D @ (3)
1 d 1.0
Jo D[) 2

All of the I-beams in this study have compact flanges and slender webs in accordance with the
compact-flange and noncompact-web limits specified in the AISC Specification [1]. The cross-sectional
dimensions, lengths, and yield strengths of the I-beams considered for each case study are summarized in
Table 1.

Table 1: Summary of beam dimensions and yield strengths of the I-beams

ho tw bfc tfc bft tft Lb Fy
Case Section
(mm) (mm) (mm) (mm) (mm) (mm) (mm) (MPa)
750 4.5 varies 20 105 20 3,000 345
1 750 4.5 105 20 105 20 3,000 345
750 4.5 105 20 varies 20 3,000 345
625 4.0 90 15 110 15 varies 345
625 4.0 110 15 90 15 varies 345
890 varies 85 12 120 12 4,000 345
890 varies 120 12 85 12 4,000 345
670 4.0 60 varies 100 varies 3,000 345
670 4.0 100 varies 60 varies 3,000 345
800 4.0 70 20 110 20 3,500 varies
800 4.0 110 20 70 20 3,500 varies


Based on the findings of the previous studies, for a beam whose compression flange is the smaller
flange, the reductions in the elastic critical stress due to web distortion increase as the degree of
monosymmetry increases, while when the larger flange is the compression flange, the reductions in the
elastic critical stress decrease as the degree of monosymmetry increases. The formula for the coefficient
of monosymmetry (ȕx) for a general I-shaped singly symmetric beam is provided by Galambos [7]. The
elastic distortional buckling results are summarized in Table 2.

Tadeh Zirakian and Jian Zhang

Table 2: Distortional buckling results (Case 1)

Section Beam ȕx MnAISC/MnLTB MnFSA/MnLTB MnBradford/MnLTB MnWangetal./MnLTB
B1-1 -591.71 0.58 0.81 0.90 0.76
B1-2 -466.66 0.66 0.82 0.88 0.76
B1-3 -331.87 0.73 0.84 0.86 0.76
B1-4 -195.08 0.78 0.86 0.84 0.77
B1-5 -62.87 0.82 0.87 0.83 0.77
B1-6 0.00 0.83 0.88 0.82 0.77
B1-7 62.87 0.84 0.88 0.82 0.78
B1-8 195.08 0.85 0.89 0.84 0.78
B1-9 331.87 0.85 0.89 0.86 0.78
B1-10 466.66 0.86 0.90 0.90 0.79
B1-11 591.71 0.87 0.91 0.99 0.80

As it is seen in the table, the AISC code predictions seem to be remarkably conservative relative to
the FSA solutions particularly in sections as the section monosymmetry increases.


In this case, the code predictions are evaluated as a result of variation of length, while the cross-
sectional dimensions of the beams are all kept constant. The summary of the elastic distortional buckling
results is presented in Table 3.

Table 3: Distortional buckling results (Case 2)

Section Beam MnAISC/MnLTB MnFSA/MnLTB MnBradford/MnLTB MnWangetal./MnLTB
B2-1 3,000 0.81 0.90 0.90 0.87
B2-2 3,500 0.76 0.91 0.93 0.90
B2-3 4,000 0.72 0.92 0.94 0.91
B2-4 4,500 0.68 0.93 0.95 0.93
B2-5 5,000 0.65 0.94 0.96 0.94
B2-6 6,000 0.58 0.96 0.97 0.96
B2-7 8,000 0.47 0.99 0.98 1.00
B2-8 3,000 0.89 0.93 0.90 0.86
B2-9 3,500 0.85 0.93 0.92 0.88
B2-10 4,000 0.82 0.93 0.94 0.89
B2-11 4,500 0.79 0.93 0.95 0.90
B2-12 5,000 0.76 0.94 0.96 0.90
B2-13 6,000 0.70 0.94 0.97 0.91
B2-14 8,000 0.59 0.94 0.98 0.92

It is generally accepted that the distortional effects are smaller in longer beams. This fact is clearly
demonstrated by the FSA as well as the theoretical predictions of other two design equations, as shown in

Tadeh Zirakian and Jian Zhang

Table 3. However, the AISC code predictions demonstrate a distinct trend by providing reductions
increasing from 19% to 53% for sections, and 11% to 41% for sections, as the beam length increases.
It is quite obvious that the 2005 AISC code [1] equations provide remarkably conservative results relative
to the FSA and the other considered theoretical predictions especially in longer beams.


The elastic distortional buckling code predictions are assessed in this case as a result of variation of
web thickness, while the other beam dimensions are kept constant. The results of this case study are
tabulated in Table 4.

Table 4: Distortional buckling results (Case 3)

Section Beam ho/tw MnAISC/MnLTB MnFSA/MnLTB MnBradford/MnLTB MnWangetal./MnLTB
B3-1 296.67 0.66 0.91 0.96 0.90
B3-2 254.29 0.70 0.93 0.97 0.92
B3-3 222.50 0.73 0.95 0.98 0.94
B3-4 197.78 0.75 0.96 0.98 0.95
B3-5 178.00 0.78 0.97 0.98 0.96
B3-6 161.82 0.79 0.98 0.98 0.97
B3-7 148.33 0.80 0.99 0.99 0.98
B3-8 296.67 0.82 0.94 0.96 0.87
B3-9 254.29 0.85 0.95 0.97 0.89
B3-10 222.50 0.87 0.96 0.97 0.91
B3-11 197.78 0.88 0.96 0.98 0.92
B3-12 178.00 0.90 0.97 0.98 0.93
B3-13 161.82 0.91 0.97 0.98 0.94
B3-14 148.33 0.92 0.98 0.98 0.94

As it is seen in Table 4, the difference between the results of the distortional and lateral-torsional
solutions in both monosymmetry cases tends to increase as the web becomes more slender. However, the
code reductions in case of singly symmetric beams with smaller compression flange are relatively larger
than those of singly symmetric beams with larger compression flange. It is notable that both sets of
reductions are comparatively larger than the respective reductions of the FSA as well as the other
theoretical solutions.


The effects of web distortion may also vary as a result of variation of flange thickness in I-beams.
Hence, in this case, the accuracy of the code predictions is investigated for varying flange slenderness
ratios in singly symmetric I-beams. Table 5 summarizes the elastic distortional buckling results for both
orientations of the I-beam.

Tadeh Zirakian and Jian Zhang

Table 5: Distortional buckling results (Case 4)

Section Beam bfc/tfc MnAISC/MnLTB MnFSA/MnLTB MnBradford/MnLTB MnWangetal./MnLTB
B4-1 7.50 0.77 0.99 0.99 0.99
B4-2 6.00 0.74 0.97 0.98 0.97
B4-3 5.00 0.70 0.95 0.97 0.95
B4-4 4.29 0.66 0.91 0.95 0.91
B4-5 3.75 0.63 0.87 0.93 0.86
B4-6 3.33 0.59 0.82 0.90 0.80
B4-7 12.50 0.93 0.97 0.99 0.93
B4-8 10.00 0.93 0.97 0.99 0.91
B4-9 8.33 0.91 0.96 0.98 0.89
B4-10 7.14 0.90 0.94 0.98 0.87
B4-11 6.25 0.88 0.93 0.97 0.84
B4-12 5.56 0.86 0.91 0.96 0.81

From the table, it is found that the predictions of the AISC code equation are by and large below the
predictions of the FSA as well as the two proposed design equations, and the amount of conservatism of
the code predictions seems to be relatively high in sections with smaller compression flange.


The effect of variation of yield strength on distortional buckling of singly symmetric I-beams is
investigated in this study, which is believed to provide us with a better understanding of the implications
of web distortion as a result of variation of yield strength. A wide range of yield strengths, i.e. from 250
MPa to 690 MPa, are considered in this study, which are tabulated in Table 6. Distortional buckling
results of this case study are given in Table 6.

Table 6: Distortional buckling results (Case 5)

Section Beam MnAISC/MnLTB MnFSA/MnLTB MnBradford/MnLTB MnWangetal./MnLTB
B5-1 250 0.62 0.80 0.90 0.74
B5-2 290 0.61 0.80 0.90 0.74
B5-3 345 0.60 0.80 0.90 0.74
B5-4 415 0.59 0.80 0.90 0.74
B5-5 485 0.58 0.80 0.90 0.74
B5-6 550 0.57 0.80 0.90 0.74
B5-7 620 0.57 0.80 0.90 0.74
B5-8 690 0.56 0.80 0.90 0.74
B5-9 250 0.77 0.89 0.93 0.78
B5-10 290 0.85 0.89 0.93 0.78
B5-11 345 0.84 0.89 0.93 0.78
B5-12 415 0.83 0.89 0.93 0.78
B5-13 485 0.82 0.89 0.93 0.78
B5-14 550 0.82 0.89 0.93 0.78
B5-15 620 0.82 0.89 0.93 0.78
B5-16 690 0.81 0.89 0.93 0.78

Tadeh Zirakian and Jian Zhang

From the table, it is evident that the predictions of the FSA as well as the other proposed equations
are not affected by the variation of the yield strength in both and cases, while the reductions induced
by the AISC code equations are found to increase slightly in case of sections, and also initially decrease
and then increase gradually in case of sections, with the increasing of the yield strength. In any case,
the AISC code equations seem to yield conservative predictions relative to the FSA results, and the
conservatism in case of sections with smaller compression flange is considerably high.

The evaluation of effectiveness of the 2005 AISC code design rules in case of distortional buckling of
singly symmetric I-beams demonstrates that the 2005 AISC code equations generally provide
conservative strength estimates for elastic distortional buckling. Even the amount of this conservatism is
found to be relatively high in case of singly symmetric I-beams with smaller compression flange. This
indicates that the assumption of J = 0, which is used in Section F5 of the 2005 AISC Specification with
the aim of controlling the unconservative errors due to the neglect of web distortion effects, may not be
an appropriate approach to the problem, since it may impose economic burden in some cases.

[1] American Institute of Steel Construction. Specification for structural steel buildings, Chicago, IL,
AISC, 2005.
[2] American Institute of Steel Construction. Commentary on the specification for structural steel
buildings, Chicago, IL, AISC, 2005.
[3] Bradford, M.A. and Waters, S.W. “Distortional instability of fabricated monosymmetric I-beams”.
Computers & Structures, 29(4), 715-724, 1988.
[4] Schafer, B.W. CUFSM 3.12, Elastic buckling analysis of thin-walled members by finite strip
analysis, 2006. (http://www.ce.jhu.edu/bschafer/cufsm)
[5] Bradford, M.A. “Distortional buckling of monosymmetric I-beams”. Journal of Constructional
Steel Research, 5(2), 123-136, 1985.
[6] Wang, C.M., Chin, C.K., and Kitipornchai, S. “Parametric study on distortional buckling of
monosymmetric beam-columns”. Journal of Constructional Steel Research, 18(2), 89-110, 1991.
[7] Galambos, T.V., Guide to Stability Design Criteria for Metal Structures, 5th Ed., John Wiley &
Sons, New York, 1998.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010


Ronald D. Ziemian* and J. Randolph Kissell **

* Bucknell University, Lewisburg, PA, USA

e-mail: ziemian@bucknell.edu
** The TGB Partnership, Hillsborough, NC, USA
e-mail: randy.kissell@tgbpartnership.com

Keywords: Aluminum structures; Second-order effects, Stability, Direct analysis.

Abstract. The 2010 Aluminum Association Specification for Aluminum Structures has been significantly
revised to include more transparent stability provisions. Second-order effects, including P- and P-
moments, and factors known to accentuate these effects, such as geometric imperfections and member
inelasticity, will need to be considered in determining required strengths. This paper provides an
overview of these provisions and describes experimental and analytical studies that investigated their

Widely used in the US since its first publication in 1967, the Aluminum Association’s (AA)
Specification for Aluminum Structures [1] has always addressed the stability of individual structural
members. With regard to beams and columns, the Specification provides equations for determining the
strength of beams and columns that account for local buckling of elements such as flanges or webs, and
flexural, flexural-torsional, and lateral-torsional buckling of members. Prior to the 2010 Specification, a
moment-amplification factor was used to address the P- effect, which is the effect of axial load acting on
the deflected shape of a member between its ends, on the stability of beam-columns.
Although it addressed the stability of individual members, earlier editions of the Specification have
not directly considered the stability of structural systems as a whole. The Specification has never
required engineers to design for the P- effect, which is the effect of loads acting on the displaced
location of joints in a structure, and only in more recent editions of the Specification was system response
included through the use of the effective length concept. As a result, the strength of a structural system
designed by previous editions of the Specification can be significantly less than the strength of its weakest
With some collapses of aluminum structures attributed to system instability, the AA decided to
provide more comprehensive and transparent stability provisions in the 2010 edition of the Specification.
Recognizing that accurately determining the effective length of members is complicated by the wide
variety of non-orthogonal structural geometries used in aluminum structures, the AA has abandoned the
use of effective length. In an effort to be more consistent with other US design specifications, the AA
adopted stability provisions similar to those that appear in the 2010 American Institute of Steel
Construction’s (AISC) Specification for Structural Steel Buildings [2]. Because of differences in (1) the
stiffness and strength of steel and aluminum, in particular that the E/y ratio for steel is approximately
twice that of aluminum, and (2) the manufacturing processes of aluminum profiles and hot-rolled steel
sections, a study that includes experimental and analytical components was conducted to confirm the
adequacy of adopting the AISC provisions. A summary of this study is presented below.

Ronald D. Ziemian and J. Randolph Kissell


In Chapter C: Design for Stability, the AA Specification provides analysis requirements (calculation
of required strengths) and design requirements (calculation of available strengths) for the entire structural
system and for each of its components. With respect to the latter, available strengths shall be based on the
actual unbraced length of the member (i.e., an effective length factor of k = 1). Required strengths are to
be determined from an analysis that considers:
1. All member and connection deformation;
2. Second-order effects, including both P- and P- moments;
3. Geometric imperfections, such as frame out-of-plumbness and member out-of-straightness, that
reflect the tolerances permitted in contract documents;
4. Reductions in member stiffness due to:
a. inelasticity or partial yielding of members
b. uncertainty in defining the stiffness and strength of components.
As a means for reducing member stiffness due to inelasticity, the AA Specification requires that the
flexural stiffness of all members be reduced by , where

 = 1.0 for  0.5
 P  P Pr
 = 4  r   1 r  for 0.5 <  1.0
 Py  Py Py

in which Pr is the required axial compressive strength (i.e., axial force in member) and Py is the axial
yield or squash load (i.e., Py = Agy).
Because of uncertainty in the stiffness and strength of the structural system, there exists the potential
for larger deflections (due to less than expected lateral stiffness) and correspondingly increased second-
order moments. To account for this, a factor of 0.8 must be applied to all axial, flexural, and shear
stiffnesses in the structure. One simple method for achieving this is to employ 80% of the actual modulus
of elasticity in the analysis model.
In addition to the above analysis requirements, the Specification requires that all gravity loads must
be included in the analysis of lateral load-resisting systems. For example, additional lateral load may
result from columns that support only gravity loads but attain their sidesway stability by leaning on the
structure’s lateral system.


In general, the above analysis requirements should apply to any metal structure in which there is
significance in formulating the equations of equilibrium on the deformed, and perhaps partially yielded,
geometry of the structural system. Because the details in accounting for member inelasticity (-factor of
equation 1) and uncertainty in stiffness (stiffness reduction factor of 0.8) were specifically developed for
steel buildings, their applicability to aluminum structures deserves to be investigated.
A good starting point is to provide a background to the AISC approach. The AISC Specification has
long recognized the impact of residual stresses on the performance of hot rolled sections. Compressive
axial stresses on the order of 30 to 50 percent of the material yield strength can result from the steel
fabrication process and such stresses can obviously accentuate the partial yielding of a cross section as a
member reaches a strength limit state (e.g., inelastic flexural buckling of a column). Based on an
extensive calibration study [3], the AISC determined that the relatively simple parabolic expression
provided by equation 1, which was originally developed by Bleich [4], and the stiffness reduction factor
0.8 adequately account for the loss in bending stiffness of members subject to high axial compressive
loads. For frames with slender members, where the limit state is governed by elastic stability (i.e.,  = 1.0

Ronald D. Ziemian and J. Randolph Kissell

with P/Py < 0.5), the same 0.8 factor can be employed because it is approximately equal to the product of
the AISC resistance factor of 0.9 and the 0.877 reduction factor used in the AISC column curve to
account for member out-of-straightness.
In contrast, many aluminum sections are typically extruded and upon cooling are then pulled to
straighten or remove any initial out-of-straightness. This stretching process requires axially straining the
material beyond yield and as a result typically relieves the residual stresses that may have developed from
an uneven cooling process. On the other hand, some aluminum sections are fabricated by welding several
profiles together; a fabrication process that may result in significant residual stresses.
Differences in the stress-strain relationships for each material may also be a factor in determining the
appropriateness of adopting the AISC provisions. Hot-rolled steels typically have a fairly linear
constitutive relationship with a pronounced yield point. In contrast, the stress-stain relationships for most
aluminum alloys are inherently nonlinear and without pronounced yield points [5]; as a result, “yield” for
most aluminum alloys is defined by the stress at a 0.2% strain offset. Hence, the above reasons (e.g.,
absence of residual stresses) for not employing the parabolic form of equation 1 may be offset by the need
to model a nonlinear material.

As an initial study, a series of stub-column tests were performed according to the Structural Stability
Research Council’s Technical Memorandum No. 3: Stub-Column Test Procedure [6]. Using the
experimental setup shown in figure 1(a), three I-shape sections and three hollow-rectangular shapes were
compression tested. The dimensions for the I-shape are shown in figure 2 and the rectangular shape was
203.2  101.6 mm with a wall thickness of 12.7 mm. In all tests, the stub-columns were 6061-T6
aluminum of length 0.61 m. In each test, the compression force and longitudinal deformation at two
locations (mid-flange) over a gauge length of 254 mm were recorded. The force was converted to an
axial stress and the deformations were averaged and then converted to axial strain. Stress-strain plots
were then prepared for each of the specimens for the two shapes. For each shape, the stress-strain data
was averaged to produce a single curve.
An untested fourth specimen for each shape was sectioned and tensile coupons were machined from
material at two locations, including the flange tip and the flange-web intersection. Tensile tests were
performed on these coupons according to ASTM B557 [7]. The two resulting stress-strain curves for the
tension tests of each shape were averaged using the same procedure as that employed in the above
compression tests.

(a) (b)
Figure 1: Experimental setup and results of stub-column axial compressive tests.

Ronald D. Ziemian and J. Randolph Kissell

For each shape, the resulting tensile and compressive stress-strain curves were used to determine a
stiffness reduction relationship equivalent to that provided in equation 1. This was done by first
determining the stress 0.2% corresponding to a 0.2% offset strain and the modulus of elasticity Eo at low
stress values (taken as the slope of line “best-fit” to stress-strain curve between 0.150.2% and 0.650.2%).
The stress  normalized by 0.2% and the corresponding tangent modulus E (slope of stress-strain curve)
normalized by Eo are then plotted as shown in figure 1(b). Although this figure only includes the
resulting tension and stub-column curves for the I-shape, similar results were obtained for the hollow-
rectangular section.
A review of the three curves plotted in figure 1(b) indicate:
1. The stiffness relationship for tension confirms that the material has a significantly nonlinear
response. A material with a linear stress-strain relationship and a pronounced yield point
would closely match the grey bi-linear curve shown in the upper right of the figure.
2. In comparison to the tension response, the compression relationship indicates that the
material stiffness degrades noticeably faster with increased levels of axial compression.
Unfortunately, it is not clear how much of this difference can be attributed to the loading
direction on material response versus that which may be attributed to the existence of
residual stresses.
3. The stiffness reduction -factor defined by equation 1 is conservative (but in the authors’
opinion not necessarily overly-conservative) in estimating the loss of axial stiffness due to
increased levels of compressive force.

Using one of the frames appearing in the original AISC calibration studies mentioned above, a
computational study was performed to investigate the impact of adopting the AISC - and 0.8 factors
within the AA provisions. The symmetrical portal frame used in this study is shown in figure 2. Two
ratios of beam-to-column stiffness were considered, one of which included assuming rigid beams with
(EI/L)c/(EI/L)b = 0, and the other with moderately flexible beams (EI/L)c/(EI/L)b = 3. Using the cross-
section geometry of the bi-symmetrical I-shape used in the above experimental study, both major- and
minor-axis bending behavior of the columns was investigated. In all cases, members were assumed fully
braced out-of-plane.

Figure 2: Symmetrical portal frame used in computational study.

For each of the four models investigated, two nonlinear finite element programs were employed to
determine system strengths and obtain interaction curves for a wide range of resulting combinations of
axial force and bending in the columns.
A more refined and commercially available finite element program ADINA [8] was employed to
obtain a theoretically “exact” solution. Three-dimensional models of the I-shape were created using fully
integrated, 4-node shell elements (MITC4). The cross section was modeled with a mesh density of 10
elements across the flange width and 10 elements through web depth. The number of elements along the

Ronald D. Ziemian and J. Randolph Kissell

length of the member was varied to maintain an element aspect ratio of approximately one. A typical
model included approximately 6000 shell elements. All models considered both geometric (large
rotation/small strain) and material (multi-linear plasticity) nonlinear effects. A nonlinear stress-strain
response (figure 4) for aluminum was explicitly incorporated. Initial imperfections, including member
out-of-straightness and frame out-of-plumb, were included by distorting the original finite element mesh.
Because the focus of the study was to determine the impact of partially yielded compressive members on
system stability, beam elements were always modeled as elastic and column elements were permitted to
yield by using the constitutive relationship shown in figure 3.

Figure 3: Stress-strain relationship used in ADINA analyses.

Each ADINA analysis was performed until a strength limit state was detected. Such limit states were
typically defined by an instability resulting from a combination of yielding in the columns and second-
order effects.
The frame analysis software MASTAN2 [9] was employed to obtain results that meet the above AA
Specification’s stability requirements. MASTAN2 models second-order effects through the use of
element geometric stiffness matrices and an incremental solution scheme based on an updated Lagrangian
formulation [10]. Equation 1 is directly included in the analysis, which results in the flexural stiffness
being reduced according to the axial force in each element during every load increment.
Strength limit states for the MASTAN2 analyses were defined by the combination of axial force and
bending moment in the columns that just satisfied the AA Specification’s interaction equation:

Pr M r
+  1.0 (2)
Pc M c

where, Pr and Mr are the axial force and bending moment from the MASTAN2 analysis, Pc the design
compressive strength determined in accordance with the AA Specification’s column curve with kL = L,
and Mc the design strength determined in accordance with the AA Specification’s requirements for
flexure (which for this fully braced compact section column, Mr = bSy with b = 0.9 and S is the elastic
section modulus). Frame out-of-plumbness of H/500 was included in these analyses but member out-of-
straightness was not. The latter is assumed to be included in the AA Specification’s equation for column
strength Pc.

Using the same validation approach employed in the AISC studies, the AA Specification stability
requirements can be assessed by comparing P-M interaction plots of the limiting strengths from the AA-
MASTAN2 approach to the “exact” strength determined from sophisticated geometric and material
nonlinear ADINA analyses.

Ronald D. Ziemian and J. Randolph Kissell

Figure 4: Interaction curves for major-axis bending of column.

Results for major-axis and minor-axis bending cases are contained in Figures 4 and 5, respectively.
Two plots are presented in each figure, including one for rigid beams with (EI/L)c/(EI/L)b = 0 and one for
a moderate degree of flexibility in the beams with (EI/L)c/(EI/L)b = 3. Two sets of AA-MASTAN2 and
ADINA curves are provided in each plot.
The first set allows for a comparison of the ratios of the first-order moment in the column to its plastic
moment (WLc/Mp, with W and Lc defined in figure 2, and Mp = Zy where Z is the plastic section
modulus). The second set can be used to compare ratios of the total moment (including first- and second-
order effects) in the column to its plastic moment (Mc/Mp). Each point on the curves represents the results
of an analysis for specific combination of gravity load Q and lateral load W. In total, just under 50
separate ADINA and MASTAN2 analyses were performed in this study.
Based on Figures 4 and 5, several observations can be made:
1. For each analysis type (AA-MASTAN2 and ADINA), a comparison of the first-order moment
ratio WLc/Mp to the total moment ratio Mc/Mp at various values of P/Py indicates that second-
order effects are significant for this example. At loads as small as P/Py = 0.1, the second-order
moments are on the order of 10 to 15 percent larger than the first-order moments. As expected,
this moment amplification increases significantly for larger values of P/Py. The reason the
ADINA second-order effects are larger is because this analysis includes material nonlinear
behavior, which tends to reduce lateral stiffness and increase deflections.
2. By comparing the AA-MASTAN2 and ADINA total moment ratios Mc/Mp at various values of
P/Py, it is clear that the “exact” bending moment capacity of the column in the presence of any
amount of axial force always exceeds the moment capacity defined by the AA Specification’s

Ronald D. Ziemian and J. Randolph Kissell

Figure 5: Interaction curves for minor-axis bending of column.

beam-column interaction equation (see equation 2). A significant factor contributing to this is
that the AA Specification moment capacity of the member bSy is less than the actual (ADINA)
upper limit capacity of byZ. Another cause for this is that the AA Specification uses a single
linear interaction equation to represent the strength of beam-columns; in most other
specifications (e.g., AISC) a bilinear curve is used, which permits larger strengths at low- to
intermediate values of axial force, ranging from approximately P/Py = 0.1 to P/Py = 0.5.
3. The WLc/Mp curves also provide a direct indication of the ultimate strength of the frame
predicted by the AA-MASTAN2 and ADINA approaches. For example, the coordinate pair
(WLc/Mp, P/Py) = (0.2, 0.4) represents failure at gravity and lateral load combination of Q = 0.4Py
and W = 0.2Mp/Lc. In all major-axis bending cases, the strength predicted by the AA-MASTAN2
approach is less than the “exact” strength predicted by ADINA. This conservatism is repeated
for all minor-axis bending conditions with the exception of the high axial load case (P/Py > 0.5)
in the frame with a moderately flexible beam of (EI/L)c/(EI/L)b = 3. The over-predicted AA-
MASTAN2 strength, however, is quite small (see lower plot in figure 4). For a column-to-beam
stiffness of (EI/L)c/(EI/L)b = 3, a design method based on effective length would use an effective
length factor of approximately k = 2.5, where as the AA stability provisions permit the use of k =
4. The largest P/Py values observed in each AA-MASTAN2 case are between 0.6 and 0.7.
Substituting these values into equation 1 results in relatively inconsequential -factors of 0.96
and 0.84, respectively. Given that fairly stocky columns (L/r = 20 with r = I A ) were
investigated in this study, it should be noted that larger slenderness L/r values more common to

Ronald D. Ziemian and J. Randolph Kissell

design would result in smaller column strengths (i.e. lower P/Py values) and hence, even larger
(closer to 1.0) and less consequential -factors.

The new stability provisions that appear in the 2010 Aluminum Association’s Specification for
Aluminum Structures have been investigated by a study that contains experimental and analytical
components. Experimental stub-column tests were performed on I- and hollow-rectangular shapes in an
effort to better understand the reduction in axial stiffness in the presence of compressive force.
Computational analyses were then performed on a portal frame that is similar to one used in part to
validate the AISC stability provisions. In this work, various combinations of lateral and gravity loads
were examined as well as varying degrees of frame flexibility.
Insight from this study indicates that use of the AA stability provisions in conjunction with the AA
single linear interaction equation for designing beam-columns provides moderate to fairly conservative
results. The AA use of the same stiffness reduction factors  and 0.8 that appear in the AISC
Specification does not appear to be unreasonable, although it is unclear if the -factor is necessary.
The study presents several cases where the AA stability provisions are adequate for allowing the
routine use of an effective length factor of k = 1, even in cases where an effective length design method
requires using two to three times that value. Just as importantly, the research further justifies the need to
consider second-order effects in the design of aluminum structures.
It is recommended that additional studies be made to determine if the AA could avoid the use of a -
factor in future editions to their specification. Such studies should also explore cases that include built-up
sections, where the effects of welding may result in substantial residual stresses and may justify using the

The authors thank the Aluminum Association for their support of this research under grant number

[1] Aluminum Association, Specification for Aluminum Structures, Arlington, VA, 2010.
[2] American Institute of Steel Construction, Specification for Structural Steel Buildings, Chicago, IL,
[3] Surovek-Maleck, A., White, D.W. and Ziemian, R.D., Validation of the Direct Analysis Method,
Structural Engineering, Mechanics and Materials Report No. 35, School of Civil and Environmental
Engineering, Georgia Institute of Technology, Atlanta, GA, 2003.
[4] Bleich, F., Buckling Strength of Metal Structures, McGraw-Hill, New York, 1952.
[5] Kissell, J.R. and Ferry, R.L., Aluminum Structures: A Guide to Their Specifications and Design,
Wiley, Hoboken, NJ, 2002.
[6] Ziemian, R.D. (Ed.), Guide to Stability Design Criteria for Metal Structures, 6th edition, Wiley,
Hoboken, NJ, 2010.
[7] ASTM Standard B557, Standard Test Methods for Tension Testing Wrought and Cast Aluminum-
and Magnesium-Alloy Products, ASTM International, West Conshohocken, PA, 2006.
[8] ADINA, Theory Manual, ADINA Research and Development, Inc., Watertown, MA, 2009.
[9] MASTAN2, developed by R.D. Ziemian and W. McGuire, version 3.2, www.mastan2.com, 2009.
[10] McGuire, W., Gallagher, R.H., and Ziemian, R.D., Matrix Structural Analysis, Wiley, Hoboken,
NJ, 2000.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Zsolt Nagy*, I. Mircea Cristutiu**

* Technical University of Cluj-Napoca-400027, Romania

e-mail: zsolt.nagy@bmt.utcluj.ro

** “Politehnica” University of Timisoara-300224, Romania

e-mail: mircea.cristutiu@arh.upt.ro

Keywords: Nonlinear analysis, large span frame, hunched rafter, tapered column.

Abstract. The article describes the applied technological solutions to transform an existing ice rink into
an indoor arena in the city of Târgu Mureú, Romania. The new indoor arena will have a capacity of 1800
fixed seats. Using a large free span (50 m) structure will overcome the in situ technological constraints
due to the position of the existing building. There is limited access due to the fact that the ice rink is
situated between two buildings and the river Mureú which flows alongside the third side. Also, the
existing refrigeration system makes access impossible within the ice pad structure area. Taking into
account the above mentioned restrictions, the article describes the applied structural solutions which will
make the structural steel work erection possible. The structural solution using steel will ensure fast and
easy erection of the structural steel framework without causing damage to any of the existing buildings
and installations. The paper summarizes the results of the numerical study performed by the authors on
the frame structure. The frames were designed to withstand horizontal and vertical loads and also to
satisfy the ULS and SLS criteria. The frames have fixed base connections, tapered columns, hunched and
king-post truss rafters and a pitch roof angle of 30.


1.1 About the scope of works

To find a technological solution to transform an existing ice rink into an indoor arena in the city of
Târgu Mureú with a capacity of 1800 fixed seats imposed to use a large free span (50 meter) steel
structure. With the condition of a clear height of 9,00 m over the ice pad area, the geometrical dimensions
of the proposed building resulted 58,60x 67,00 x13.00 m (width x length x height) . The building on the
ground floor consists of the ice pad area - 1800 m2 (60 x 30 m) and the necessary annexes (public area,
offices and dressing rooms etc.) of 2200 m2. The scope of works included the following main
-To cover the existing ice rink in order to extend the usage lifetime;
-To have a capacity of 1800 fixed seats;
-To ensure 90 min fire rating of the steel structure;
-To ensure the specific internal micro climate.

Nagy Zsolt et al.

1.2 Constrains
Due to the destination and the particular position of the building - limited access due to the fact that
the ice rink is situated between two buildings and the river Mureú flows alongside the third side, there
were the following constrains:
-Access for erection only from one side, without the access on ice pad structure due to the
existing refrigeration system;
-To keep the existing buildings;
-To control the designed assemblies self weight, in order to facilitate the erection.
-Also the pressure of a short deadline acted as constraint. The site conditions and the proposed
architecture should be seen in Figure 1.

Figure 1. The site conditions and the architecture of the building (view from access side)


2.1 Steel structure

The primary load-bearing structure of the building uses a simple steel portal frame shape based on a
6.00 m grid (Figure 2), combined with a king post truss rafter. The clear span of the frame is 50 m, with
additional 4 m extension on both sides. The frames have fixed base connections, tapered columns,
hunched and king-post truss rafters and a pitch roof angle of 30. The rafter was extended over the two
lateral extensions and fixed at the top level of the columns from the extremities. In that way instead of
having a simple frame, we have transformed the rafter in a continuous beam, increasing both its strength
and stiffness. The supporting structure of the tribune is fixed to the frame column in the transverse plane,
increasing in that way the lateral stiffness of the whole transverse frame. In order to prevent lateral-
torsional buckling of the rafter, its lower flange was braced laterally to the roof purlins (see Figure 3
a&b). Supplementary lateral restraints were provided by means of longitudinal beams, stiffened together
by the roof bracings. At the mid span, king post truss was laterally restrained in order to prevent its lateral
displacement in case of horizontal actions (e.g. seismic action-see Figure 3a). All the assemblies
(excluding longitudinal beam, bracings) are made from welded steel sections. A structural steel with
S355 steel grade (fy=355 N/mm2) have been used.
For the first and second floor slab in situ reinforced concrete solution was applied. For the composite
action of steel and concrete, mechanically fixed shear studs have been used on floor beams. Precast
concrete elements were designed for the tribune. A central skylight cut out of the roof to bring daylight
down to the ice rink.
The 90 min fire resistance of the structural steel columns and 60 min for rafters and floor beams is
assured by intumescent coating of the steelwork.

Nagy Zsolt et al.

Figure 2. Characteristic section of the structure

a) lateral restraining of the king post truss b) rafter-to-column detail

Figure 3. Connections detail
2.2 The building envelope
The insulated building envelope makes it possible to control the indoor climate regardless of the
outdoor climate. In the case of this type of buildings, air tightness is a more important feature of the
envelope than thermal insulation. Large glazing of the facade has been avoided due to energy costs by
operating the facility. Windows are placed mainly on facility area, because the most optimized ice rink
can be done by a fully closed casing.
The wall cladding is made of 120 mm thick horizontal sandwich panels. On the roof 200 mm thick
rock wool insulation is laid down on the supporting trapezoidal steel profiles, waterproofing is assured by
a protective membrane.
The supporting structure of the facade is a steel framework of rectangular hollow sections.


3.1 Loading of the main structure

In order to evaluate the structural response, in the design process were considered the following loads
(characteristic values):
-Roof loads (EN1991-1-1): dead load + technological load qk = 1.0 kN / m2
-Live loads on floors (EN1991-1-1) uk= 5.0 kN / m2
-Snow loads on the roof according to CR 1-1-3-2005 (EN1991-1-3), s0,k=1.5 kN/m2
-Wind loads on building envelope according to NP-082-04 (EN1991-1-4), qref=0.4 kN/m2
-Fire loads of 120 MJ/m2
-Seismic action according to P100-2006 (EN1998-1), with peak ground acceleration ag=0.12g
and control period of seismic motion Tc=0.7 sec

Nagy Zsolt et al.

-Load combination for ultimate limit state (ULS) and serviceability limit state (SLS) according
to CR-0-2005 (EN 1990).
3.2 Design of the main structure-linear elastic analysis (LEA)
The design of the steel structure was performed following the Romanian code STAS 10108/0-78 [6].
For strength, stability and stiffness requirements of the structural elements the prescription of SR-
EN1993-1-1[4], SR-EN1993-1-8[5] and P100/2006 [3] were also used.
In the case of large spanned structures, the vertical deflection under gravitational loads represents one
of the major constraints in the design process. In order to keep under control the deformations of the
frames, fixed base connections, tapered columns and hunched king-post truss rafter solution were chosen
[6]. The rafters were extended on both sides over the annexes, increasing both the vertical and horizontal
stiffness of the frame. A suitable horizontal and vertical bracing system were provided in order to control
structural flexibility, eigen values and deflections of the main structure. Fly braces were provided at the
inner flange of the rafter in order to improve the flexural-torsional buckling resistance of these elements.
Having class 3 section of the structural elements, linear elastic structural analysis was performed,
using a seismic behavior factor of q=1 according to P100-2006 [3]. Even with q=1, the combinations of
actions for seismic design situations were not the dominant load combinations. The design checks of the
structural elements for ULS include persistent or transient design situations (fundamental combinations)
where snow loads play the key role.
For SLS design checks of the structural elements fundamental and exceptional load combinations
were used. Performing a dynamic 3D analysis of the structure, with the structural masses concentrated on
joints, first longitudinal eigen period of Tlong=0.588 sec and first transversal eigen period of Ttransv=0.448
sec were obtained (see Figure 4).

First longitudinal vibration mode T=0.588 sec First transversal vibration mode T=0.448 sec
Figure 4. Eigen vibration modes and periods

The maximum transversal and longitudinal sway displacement for SLS check under seismic loads
according to P100-2006 are:
d rSLS 0,005 ˜ h
d rSLS
,x 0.014 d ,a
112.5 mm (1)
Q ˜q 0.4 ˜ 1.0
d rSLS 0,005 ˜ h
d rSLS
,y 0.047 d ,a
137.5 mm (2)
Q ˜q 0.4 ˜ 1.0
The maximum vertical deflection of the rafter for SLS check under snow load is:
f 161.4mm d f a 166.7 mm (3)

Nagy Zsolt et al.

In order to have an overview about the real behavior of the structure, a finite element linear elastic
analysis (FEM) of the transverse frame has been performed with Ansys computer program. The elements
of the frame were modeled using shell finite elements (Shell 43- see Figure 5). The forces on the rafter
were applied as point loads (points where purlins are fixed on the frame). The connections between
structural elements rafter-to-column, beam-to-column, rafter-to-rafter, column base connections were
considered fully rigid. The results of the detailed linear-elastic analysis (LEA) confirmed the previously
evaluated ULS and SLS results. The recorded vertical displacement in case of FEM linear elastic analysis
was 152 mm (instead of 203 mm –linear elastic analysis). It might be emphasized that the resulted
structure is more rigid in case of FEM analysis, explained by the shift of the neutral axis along the
elements with variable cross sections (i.e. tapered column, and hunched rafter). Figure 6 shows the stress
distribution along the transverse frame, where we can observe the maximum stress concentration around
the joint of the king post rafter and the hunched frame rafter.

Figure 5. FEM model of the transverse frame

Figure 6. Stress distribution along the transverse frame under gravity load combinations-linear elastic

Nagy Zsolt et al.

The resulted maximum stress does not exceed 252 N/mm2. There were many concerns about the
stress distribution in the connection of the king post truss with the rafter. As it can be observed from
Figure 6 the distribution of the stresses does not exceed the maximum allowable yield limit.
3.3 Design of the main structure- non linear elastic-plastic analysis
In Figure 8, it is illustrated the way in which the initial bow (out of plane) imperfection is considered
in the nonlinear-elastic analysis (GMNIA). Three types of lateral restraints of the rafter were considered
separately in the analyses (see Figure 7 [7],[8]). Types 2 simulate the purlin/sheeting effect, when the
purlin can be connected with one or two bolts, respectively. Type 3 is the same with type 2 with an
additional fly brace. Type 1, the reference case, actually means no lateral restrains introduced by purlins
and side rails.

(a) restraint type 1 (b) restraint type 2 (c) restraint type 3

Figure 7. Types of lateral restraints considered in the analysis

a) global view b) lateral view

Figure 8. Manufacturing imperfections considered in non-linear elastic-plastic analysis [6]

Figure 9. Stress distribution along the transverse frame under gravity load combinations-non linear
elastic-plastic analysis

To simplify the computational model, in the analysis the lateral restraints were considered axially
rigid. The values of the applied imperfection is 167 mm (50 m span frame) for initial bow imperfection
(er), l/150 corresponding to curve c, for plastic analysis, according to clause 5.3.2 (3)-a) of EN1993-1.1
[1]. The material behavior was introduced by a bilinear elastic-perfectly plastic model, with a yielding
limit of 355 N/mm2. In Figure 10 are illustrated the capacity curves for different type of analysis. As it

Nagy Zsolt et al.

was expected the lateral restrains of the rafter played an important role in the total capacity of the main
frame. Also it must be emphasized that there is more than 25% structural capacity reserve.

Figure 10. Capacity curves for different type of analyses

The paper illustrates the successful application of the steel structure for a large span using a simple
portal frame shape, combined with a king post truss rafter. A wide range of design parameters are briefly
summarized. The paper emphasizes the whole design process, assisted by FE analysis - in order to
perform supplementary stability checks of the framed structure. Due to the unusual shape of the
transverse frame, there were many concerns about its real behavior under gravitational load, the most
important ones in this particular case. For this purpose a linear elastic analysis (LEA) followed by a
nonlinear elastic-plastic analysis (GMNIA) were performed in order to determine the real behavior of the
frames. From structural point of view a good agreement between 3D structural analysis and LEA-FEM
has been found. GMNIA analysis confirm at least 25% overstrenght of the structure by applying the
chosen structural solution and lateral restraints of the main rafter.
Even with behavior factor q=1, the combinations of actions for earthquake design situations were not
the dominant load combinations. In the design checks of the structural elements, gravity loads played the
key role.

Figure 11. The whole structural model and the actual stage of the building

Nagy Zsolt et al.


[1] EN 1993-1-1 Eurocode 3: Design of steel structures Part 1.1: General rules and rules for buildings;
[2] CR-0-2005: Cod de proiectare pentru bazele proiectarii structurilor in constructii (Design Code.
Basis of design. Romanian design code).
[3] P100-2006: Cod de proiectare seismica P100. Partea I-Prevederi de proiectare pentru cladiri
[4] SR-EN 1993-1-1: Eurocod 3: Proiectarea structurilor de oĠel Partea 1-1: Reguli generale úi reguli
pentru clădiri
[5] SR-EN1993-1-8:Eurocod 3: Proiectarea structurilor de oĠel. Partea 1-8: Proiectarea îmbinărilor
[6] H. C. Schulitz, W. Sobek, K. J. Habermann - Steel Construction Manual, Birkhauser Verlag 2000,
ISBN no. 3-7643-6181-6
[7] D. Dubina, I. M. Cristutiu, V. Ungureanu, Zs. Nagy: Stability and ductility performances of
light steel industrial building portal frames, 3-rd European Conference of Steel Structures,
Eurosteel 2002, Coimbra-Portugal, sept. 2002, pp 635-643 (2002);
[8] D. Dubina, I. M. Cristutiu: Buckling strength of pitched-roof portal frames of Class 3 and
Class 4 tapered sections, International Conference on Steel and Composite Structures - Eurosteel
2005, Maastricth-Holland, 7-11 june 2005, pp 635-643 (2005);

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010


Nuno Silvestre* and Leroy Gardner**

* Department of Civil Engineering, IST-ICIST, Technical University of Lisbon, Portugal

** Department of Civil and Environmental Engineering, Imperial College London, United Kingdom

Keywords: Buckling, Elliptical Hollow Sections, Oval Hollow Sections, Post-buckling, Steel Structures.

Abstract. The local post-buckling behaviour of elliptical hollow section (EHS) tubes under compression
is analysed in this paper. It is found that EHS tubes with low to moderate aspect ratios can support loads
up to their limit loads but are imperfection sensitive (shell-type behaviour), while EHS tubes with
moderate to high aspect ratios can carry loads higher than their limits loads (plate-type behaviour) and
are imperfection insensitive. For increasing EHS aspect ratio, it is found that the compressive stresses
accumulate near the zones of minimum radius of curvature while the zones of maximum radius of
curvature experience a relatively low compressive stress level. Thus, it is likely to apply the effective
width concept to EHS tubes with moderate to high aspect ratio.

EHS steel tubes are now available as hot-rolled structural products [1,2] and represent an interesting
solution for many visible applications in steel construction, particularly for glass facades. These shapes
are included in the new edition of EN 10210 [3] and are available in a standard range of dimensions. In
response to the emergence and commercial availability of EHS tubes, several recent investigations on
their buckling behaviour and strength have been published. Gardner and Chan [4] and Chan and Gardner
[5,6] assessed the non-linear behaviour of hot-rolled EHS tubes by means of experimental and numerical
analyses and proposed structural design rules. They found that the slenderness limits for pure
compression set out in EC3 for circular hollow section (CHS) classification can be safely adopted for
EHS, based on the equivalent diameter of the point of the EHS with maximum radius. Zhu and Wilkinson
[7] also performed shell finite element analyses to evaluate the buckling and post-buckling behaviour of
EHS in compression. Silvestre [8] developed a formulation of Generalised Beam Theory (GBT) to
analyse the elastic buckling behaviour of members with non-circular hollow sections (NCHS) and applied
it to study the behaviour of EHS shells and tubes under compression, particularly the variation of the
critical buckling stress with the member length and cross-section geometry. Ruiz-Teran and Gardner [9]
have also examined the buckling response of EHS tubes in compression and proposed analytical formulae
to accurately predict the critical stress. Thus, the main objective of this paper is to unveil the mechanics
of the elastic local post-buckling behaviour of EHS tubes and to explain in a detailed fashion the
transition between the shell-type (imperfection sensitive) behaviour of EHS tubes with low eccentricity
and the plate-type (imperfection insensitive) behaviour of EHS tubes with high eccentricity.

An in-depth study on the influence of the EHS aspect ratio on the variation and nature of the post-
buckling equilibrium path, ensuing stress distributions and imperfection sensitivity is presented. A
reference EHS stub column with length L = 300 mm, thickness t = 4 mm and fully fixed supports is

Nuno Silvestre and Leroy Gardner

considered. The EHS geometry is characterised by the major axis width 2a and minor axis width 2b,
which are considered here as the dimensions of the EHS mid-line (i.e., 2a + t and 2b + t are the outer
dimensions). Based on a commercially available [1,2] reference geometry (2a = 150 mm, 2b = 75 mm, a/b
= 2.0), five further EHS configurations were generated and studied. These were obtained by (i) keeping
the cross-section perimeter unaltered (P = 363 mm) and (ii) varying the aspect ratio a/b from 1.10 to 5.0.
The six EHS geometries considered in this paper are represented in figure 1 and are characterised by an
equal cross-section area A=1450 mm2 and thus an equal amount of steel in each column.
a/b=1.10 a/b=1.25 a/b=1.50 a/b=2.00 a/b=3.00 a/b=5.00

2a=121.0mm 2a=128.0mm 2a=137.4mm 2a=150.0mm 2a=163.0mm 2a=173.0mm

2b=110.0mm 2b=102.4mm 2b=91.6mm 2b=75.0mm 2b=54.4mm 2b=34.6mm

Figure 1: Selected EHS geometries and corresponding aspect ratio a/b.


The local post-buckling behaviour of EHS tubes under compression is investigated numerically using
the finite element code ABAQUS [10]. In order to analyse the local behaviour of a given thin-walled
member, one must adopt a two-dimensional model to discretise its mid-surface, a task that can be
adequately performed by means of 4-node isoparametric shell elements with reduced integration (S4R
elements in the ABAQUS nomenclature). In the case of the EHS tubes dealt with in this work,
discretisation of the cross-section into 36 finite elements was found to be sufficient − this corresponds
roughly to adopting 10 mm wide elements. A mesh size of 5 mm in the length direction was used, leading
to a total of 2160 elements and 2196 nodes. In order to ensure adequate modelling of the fixed end
support conditions, rigid plates were attached to the stub column end sections, thus preventing all local
and global displacements and rotations, including (i) rigid-body motions (with the exception of the axial
translation of the loaded end section), (ii) warping and (iii) in-plane deformation. These rigid end plates
were modelled by means of 3-node R3D3 finite elements (again ABAQUS nomenclature). The compressive
load was applied through the centroid of the axially free end section and, in order to obtain the load
versus axial shortening equilibrium path, the corresponding axial displacement was assessed by using an
ABAQUS command termed “MONITOR”.
a/b=1.10 a/b=1.25 a/b=1.50 a/b=2.00 a/b=3.00 a/b=5.00

σcr=7938 σcr=6843 σcr=5535 σcr=4004 σcr=2600 σcr=1508

Figure 2: Critical buckling modes and corresponding critical stresses σcr (in N/mm2)

Nuno Silvestre and Leroy Gardner

The material behaviour of the steel tube was assumed to be homogeneous, isotropic and linear elastic, which
is fully characterised by the values of Young’s modulus (E = 210000 N/mm2) and Poisson’s ratio (ν = 0.3).
Initial geometrical imperfections were included in the models in the form of the most relevant (critical) local
buckling mode shapes, incorporated into the tube initial geometry by means the initial conditions ABAQUS
command. Therefore, preliminary buckling (eigenvalue) analyses were carried out for all the six columns in order
to obtain the local buckling stress values and the corresponding buckling mode configurations, which are
depicted in figure 2. For the initial studies, a small imperfection amplitude equal to 0.1 mm (2.5% of EHS
thickness) was adopted, while imperfection sensitivity is investigated further in Section 6 of this paper. No
residual stresses were incorporated into the numerical analyses, since they were deemed to be of very low
magnitude in hot-finished elliptical tubes [5].


Before introducing the study on the influence of the EHS aspect ratio on the variation and nature of the
post-buckling equilibrium path and ensuing stress distributions, the variation of the local critical buckling
stress with the EHS aspect ratio a/b is first assessed. For the six different aspect ratios considered herein, the
values of the critical buckling stress σcr (N/mm2) are given in table 1. As expected, it may be seen that the
critical stress decreases with increasing aspect ratio, almost inversely. Having examined the elastic critical
buckling behaviour of EHS, subsequent studies of the elastic post-buckling behaviour were then performed.
The non-linear equilibrium paths (applied stress σ versus axial shortening u) obtained from the post-buckling
analyses are plotted in figure 3(a)). The same results are presented in a normalised format – critical stress
ratio σ/σcr versus critical strain ratio ε/εcr – in figure 4.
Table 1: Critical stresses and limit stresses obtained from FE models
a/b σcr (N/mm2) σlim (N/mm2) σlim/ҏσcr
1.10 7938 6810 0.86
1.25 6843 5988 0.88
1.50 5535 4600 0.83
2.00 4004 3555 0.89
3.00 2600 2130 0.82
5.00 1508 1202 0.80
Stress σ (N/mm2) σ/σcr
8000 1.0
7000 6810 N/mm
2 a/b=2.00
a/b=1.10 0.8
6000 5988 N/mm

4600 N/mm
2 0.6

3555 N/mm
3000 Imperfection amplitude:
1% thickness
2130 2.5% thickness
2000 10% thickness
0.2 25% thickness
50% thickness
1000 100% thickness
0 0.0
0 5 10 15 20 25 30 0.0 2.5 5.0 7.5 10.0 12.5 15.0
Axial shortening u (mm) Axial shortening u (mm)

Figure 3: (a) Applied stress σ vs. axial shortening u for several a/b ratios (imperfection = 2.5% thickness)
and (b) critical stress ratio σ/σcr vs. axial shortening u for a range of imperfections (a/b=2).

Nuno Silvestre and Leroy Gardner

σ /σcr
1.8 a/b=3.00
1.6 a/b=1.50
Limit point of the Bifurcation point of
1.2 imperfect tube the perfect tube




0.2 1
0 ε /εcr
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

Figure 4: Relationship between the critical stress (σ/σcr) and strain (ε/εcr) ratio for varying a/b.

From careful observation of figures 3(a) and 4 the following comments can be made:
(i) The six tubes possess an equal cross-section area and, therefore, the slope of the primary path is given
by E/L and is equal for all six tubes; the primary paths are also perfectly linear until the limit stress
σlim is reached. As expected, the limit stress σlim is lower than the corresponding critical stress σcr due
to the influence of the imperfections. The variation of limit stress σlim with aspect ratio is shown in
table 1. In fact, the ratio between the limit stress and the critical stress (σlim/σcr) shows little variation
with a/b and remains around 0.85 for the adopted imperfection amplitude of 2.5% of the EHS
thickness. This suggests that the imperfection sensitivity of EHS tubes is not strongly dependent on
the aspect ratio a/b of the cross-section, within the examined range of 1.10 < a/b < 5.00.
(ii) After reaching the limit stress σlim, the six non-linear paths show very distinct responses. The
descending branch (decrease of both applied stress and axial shortening) is much more pronounced
for tubes with lower aspect ratio a/b, as shown in the graph of σ/σcr versus ε/εcr given in figure 4.
(iii) Figure 3(a) shows that all the tubes are associated with post-critical curves that possess local minima, which
move horizontally towards the linear primary path as the aspect ratio increases. Beyond the local minima the
post-buckling paths (for a/b ≥ 1.10) exhibit a positive slope (increase in applied stress and axial shortening)
and are stable. It may be observed that the maximum slope of the ascending post-buckling branch increases
with aspect ratio a/b, this increase being more substantial for low to moderate aspect ratios (a/b ≤ 2) and less
pronounced for moderate to high aspect ratio (a/b ≥ 2). Furthermore, the slope of the ascending post-
buckling path can reach values up to 40% of the initial slope of the linear primary path; a value similar to that
for flat simply-supported plates with unrestrained edges.
(iv) From points (i) to (iii), it may be concluded that the maximum applied stress σmax that an elastic EHS tube
with low to moderate aspect ratio (a/b ≤ 1.5) can support is its limit stress σlim whereas, the maximum applied
stress σmax that an elastic EHS tube with moderate to high aspect ratio (a/b ≥ 2.0) can carry is higher than its
limit stress σlim (see figure 3(a)). The initially unstable post-buckling response exhibited by all six tubes
investigated (with an imperfection of 2.5% of the section thickness), means that snap-through behaviour is
experienced at the limit stress. However, figure 3(a) shows that the snap-through reduces with increasing
aspect ratio a/b. For instance, the very eccentric tube with a/b = 5.0 experiences, at the limit stress level
σlim=1202 N/mm2, a very small snap between u = 1.86 mm and u = 2.22 mm. Conversely, the moderately
eccentric tube with a/b = 2.0 experiences, at the limit stress level σlim=3555 N/mm2, a larger snap between u
= 4.98 mm and u = 12.85 mm.
The deformed configurations of the EHS tubes in the post-buckling regimes are shown in figure 5,
where deformation may be seen to be concentrated towards the mid-height of the specimens. Initial
geometrical imperfections were imposed with an inward deformation of the flatter region (i.e. maximum
local radius of curvature) of the EHS at mid-height (see figure 2). From figure 5, it may be seen that the

Nuno Silvestre and Leroy Gardner

post-buckling deformed configurations are characterised by a pronounced flattening of the cross-section

at its mid-height; this flattening is associated with high tensile normal stresses that develop in the
transverse (circumferential) direction.

a/b=1.10 a/b=1.25 a/b=1.50 a/b=2.00 a/b=3.00 a/b=5.00

Fig. 5: Deformed configurations of the EHS tubes in the post-buckling regime.


In order to explain the distinct post-critical behaviour exhibited by EHS tubes and its dependency on
the aspect ratio a/b, the normal stress distributions in the weakest (most deformed) zone of the tube, (i.e.,
at mid height) are examined. Figure 6 comprises six graphs (for the six different aspect ratios
investigated), each one showing the evolution of normal stress distributions σ(θ) for several load levels
(σ/σcr), i.e., for several points travelling along the equilibrium path. In the key to each figure, the values
of σ/σcr initially increase up to the limit stress, after which they decrease to the local minimum and then
increase again; compressive stresses are positive. The normal stress σ(θ) is measured along the mid-line
of the EHS and associated with membrane behaviour and θ is the angle to the centre of the EHS; (θ=0°
and180° correspond to the points of minimum radius of curvature rmin = b2/a, coincident with the EHS
major axis, and θ=90° and 270° correspond to the points of maximum radius of curvature rmax = a2/b,
coincident with the EHS minor axis). The solid curves correspond to the limit stress ratio σlim/σcr. To aid
comparison between the six diagrams, the normal stress σ(θ) axes have the same vertical scale. From
observation of figure 6 the following remarks can be made:
(i) For σ < σlim and independently of the aspect ratio a/b, the normal stress is essentially uniform along
the EHS mid-line (i.e. σ is almost independent of θ). Further, as also seen in table 1, the stress level
corresponding to σ = σlim decreases with increasing aspect ratio a/b. From this observation, it could be
interpreted that it is preferable to design EHS with lower aspect ratios – the CHS being the limit
configuration. However, as will be seen later, such an approach does not truly corresponds to
optimum EHS design.
(ii) For σ > σlim and independently of the aspect ratio a/b, the normal stress ceases to remain uniform
along the EHS mid-line (i.e. σ varies with θ). However, it should be highlighted that the non-linear
distribution of σ(θ) varies markedly with aspect ratio a/b. The zones of maximum EHS radius of
curvature (θ = 90° and 270°) are always in compression (σ(θ) > 0) but the normal stress decreases
continuously for increasing applied stress ratio (σ/σcr). The points of minimum EHS radius of curvature
(θ = 0° and 180°) may be in compression or in tension, and this fact has far reaching implications for
the stability of the post-buckling branches. Immediately after the peak (σlim), the normal stress in the
θ=0° and 180° regions decreases for all a/b values; this decrease being much more pronounced for
moderate to low aspect ratios (a/b ≤ 1.5) than for moderate to high aspect ratios (a/b ≥ 2.0).

Nuno Silvestre and Leroy Gardner

σ (N/mm2) σ (N/mm2)
25000 25000
σ/σcr σ/σcr
0.408 0.511
a/b=1.10 0.858 a/b=1.25 0.875
20000 0.807 20000 0.746
0.659 0.659
0.557 0.567
0.487 0.503
15000 0.436 15000 0.457
0.393 0.415
0.350 0.412
0.371 0.462
10000 10000

5000 5000

0 0
0 90 180 270 360 0 90 180 270 360
θ (°) θ (°)
-5000 -5000

σ (N/mm2) σ (N/mm2)
25000 25000
σ/σcr σ/σcr
a/b=1.50 0.508 a/b=2.00
20000 0.831 20000 0.888
0.681 0.670
0.584 0.589
0.508 0.532
15000 0.466 15000 0.514
0.485 0.568
0.542 0.693
0.723 0.859
10000 10000 1.220

5000 5000

0 0
0 90 180 270 360 0 90 180 270 360
θ (°) θ (°)
-5000 -5000

σ (N/mm ) 2
σ (N/mm2)
25000 25000

20000 a/b=3.00 σ/σcr 20000 σ/σcr
0.401 0.400
0.819 0.797
0.612 0.732
15000 0.658 0.862
1.300 1.760
1.630 2.510
2.020 3.370
10000 2.250 10000 3.840

5000 5000

0 0
0 90 180 270 360 0 90 180 270 360
θ (°) θ (°)
-5000 -5000

Figure 6: Evolution of normal stresses σ(θ) with σ/σcr for the six different aspect ratios a/b.

Nuno Silvestre and Leroy Gardner

For a/b = 1.10, the stress decreases and changes sign, with the points of minimum EHS radius of
curvature (θ = 0° and 180°) being in tension at the maximum axial shortening (black line). For 1.25 ≤
a/b ≤ 2.0, the stress decreases and changes sign, but then tends to increase again with increasing
displacement. For 3.0 ≤ a/b ≤ 5.0, the stress slightly decreases after the peak but then increases
significantly, and the points θ = 0° and 180° are never in tension. The points of maximum
compressive stress are not at the minimum EHS radius of curvature (θ = 0°, 180°) but are
immediately adjacent to these points at θ = 10-30°, 150-170°, 190-210° and 330-350°.
(iii) As noted above, non-linear distributions of σ(θ) vary markedly with aspect ratio a/b. For low to
moderate aspect ratios (a/b ≤ 1.5), the stress distribution in the zones of minimum EHS radius of
curvature (θ < 60°, 120° < θ < 240°, θ > 300°) is distinctly non-linear with θ, with compressive and
tensile stresses occurring over different ranges of θ. For increasing aspect ratio (a/b ≥ 1.5), the stress
distribution in the zones of minimum EHS radius of curvature (θ < 60°, 120° < θ < 240°, θ > 300°)
becomes more uniform. This may be seen most clearly for the EHS tube with a/b = 5, where the
compressive stresses are almost uniform inside the ranges θ < 60°, 120° < θ < 240° and θ > 300°, for
the maximum axial shortening (black line).
(iv) For EHS tubes with low aspect ratio a/b, the development of tension stresses in the zones of minimum
EHS radius of curvature lead to a “softening effect” in behaviour of the tube, since the average stress
along the EHS mid-line reduces as the tensile stresses grow. This is the reason for the almost horizontal
post-buckling branches following the initial drop in load (see figure 3(a)) exhibited by the EHS tubes
with a/b = 1.1 and 1.25; their post-buckling behaviour is similar to that exhibited by circular shells and
they do not possess any post-critical stiffness. Conversely, for EHS tubes with moderate to high
eccentricity (or aspect ratio a/b), the development of high compressive stresses in the zones of minimum
EHS radius of curvature lead to a “hardening effect” in behaviour of the tube, since the average stress
along the EHS mid-line now tends to increase as the compressive stresses grow. This is the reason for the
ascending and stable post-buckling branches (see figure 3(a)) exhibited by the of EHS tube with a/b =
1.5, 2.0, 3.0 and 5.0. Consequently, their post-buckling behaviour is closer to that exhibited by flat plates
and they do possess notable post-critical stiffness.
(v) On the basis of the above findings, it may be concluded that an approach based on the “effective
width concept”, widely used for the strength analysis of flat plates, may be adapted to the design of
EHS tubes with moderate to high aspect ratios. This procedure is outside of the scope of the present
paper, but is the subject of ongoing research.

In previous sections, a constant imperfection amplitude of 2.5% of the section thickness has been
adopted. In this section, the imperfection sensitivity of EHS tubes under compression is examined. Each
of the six tubes was analysed for six imperfection amplitudes: ξ = 0.04 mm (1% of the thickness), ξ =
0.10 mm (2.5% of the thickness), ξ = 0.4 mm (10% of the thickness), ξ = 1.0 mm (25% of the thickness),
ξ = 2.0 mm (50% of the thickness) and ξ = 4.0 mm (100% of the thickness). For the tube with a/b = 2,
figure 3(b) shows the equilibrium paths (critical stress ratio σ/σcr versus axial shortening u) obtained for
the several imperfection amplitudes. It is clear that, for imperfection amplitudes lower than 25% of the
thickness, the equilibrium paths possess a limit point, and the limit stress σlim decreases significantly with
increasing imperfection amplitude. Conversely, for imperfection amplitudes higher than 50% of the tube
thickness, (i) the equilibrium paths always ascend and (ii) there is no limit stress. This behavioural aspect
has far reaching implications for the imperfection sensitivity of EHS tubes: due to its stable and
ascending post-buckling branch, a moderately (or highly) eccentric EHS tube may or may not be
imperfection sensitive depending on the range of imperfections being considered. For a given aspect ratio a/b,
there is always a “bound imperfection amplitude” (ξb) that separates the ranges of imperfection amplitude where
the tube is imperfection sensitive (ξ < ξb) and insensitive (ξ > ξb). From figure 3(b), for the EHS tube with a/b =
2, this ξb value should lie between 25% and 50% of the thickness. Using a trial-and-error procedure, the exact
value of the “bound imperfection amplitude” was found to be ξb = 1.2mm = 30% of the thickness.

Nuno Silvestre and Leroy Gardner

The elastic local post-buckling behaviour of tubes with elliptical hollow sections (EHS) under
compression was analysed in this paper. The obtained numerical results were then presented and
analysed; the following conclusions are drawn:
(i) Non-linear equilibrium paths - The maximum applied stress that an elastic EHS tube with low to
moderate aspect ratio (a/b ≤ 1.5) can support is its limit stress σlim, while the maximum applied stress
that an elastic EHS tube with moderate to high aspect ratio (a/b ≥ 2.0) can carry is higher than its
limit stress σlim. It was observed that the slope of the ascending branch increases with aspect ratio a/b
and can reach values up to 40% of the initial slope of the linear primary path.
(ii) Normal stress distributions - For increasing aspect ratio a/b, the compressive stresses grow and
accumulate near the zones of minimum radius of curvature while the zones of maximum radius of
curvature possess an approximately uniform and relatively low compressive stress level. Therefore, it
is expected that an approach based on the “effective width concept” widely used for the strength
analysis of flat plates may be adapted to the design of EHS tubes with moderate to high aspect ratios.
(iii) Imperfection sensitivity - For a given aspect ratio a/b, there is a “bound imperfection amplitude” ξb
that separates the ranges of imperfection amplitude where the EHS tube is imperfection sensitive (ξ <
ξb) and insensitive (ξ > ξb). Moreover, it was shown that the imperfection sensitivity of EHS tubes
significantly drops for increasing aspect ratio a/b, ranging between shell-type behaviour (strongly
imperfection sensitive) and plate-type behaviour (imperfection insensitive).

[1] Corus. Celsius 355s Ovals. Internet: http://www.corusgroup.com, 2006.
[2] Interpipe – The Hollow Section Company. Elliptical Hollow Sections to S355 J2H, Internet:
http://www.interpipe.co.uk/, 2007.
[3] CEN – Comité Européen de Normalisation. EN 10210-2: Hot finished structural hollow sections of
non-alloy and fine grain steels – Part 2: Tolerances, dimensions and sectional properties, 2006.
[4] Gardner, L., Chan, T.M., “Cross-section classification of elliptical hollow sections”, Steel and Composite
Structures, 7(3), 185-200, 2007.
[5] Chan, T.M., Gardner, L., “Compressive resistance of hot-rolled elliptical hollow sections”, Engineering
Structures, 30(2), 522-532, 2008.
[6] Chan, T.M., Gardner, L., “Flexural buckling of elliptical hollow section columns”, Journal of
Structural Engineering-ASCE, 135(5), 546-557, 2009.
[7] Zhu, Y., Wilkinson, T., “Finite element analysis of structural steel elliptical hollow sections in
compression”, Research Report No R874, Centre for Advanced Structural Engineering, The University of
Sydney, 2007.
[8] Silvestre, N., “Buckling behaviour of elliptical cylindrical shells and tubes under compression”,
International Journal of Solids and Structures, 45(16), 4427-4447, 2008.
[9] Ruiz-Teran, A.M., Gardner, L., “Elastic buckling of elliptical tubes”, Thin-Walled Structures, 46(11),
1304-1318, 2008.
[10] DS Simulia Inc. ABAQUS Standard (version 6.7), 2007.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010


Marios Theofanous* and Leroy Gardner*

* Department of Civil and Environmental Engineering, Imperial College London, London SW7 2AZ
e-mails: marios.theofanous@imperial.ac.uk, leroy.gardner@imperial.ac.uk

Keywords: Stainless Steel Structures, Continuous Beams, Plastic Design, Continuous Strength Method.

Abstract. Despite the high material ductility of structural stainless steels and the existence of a Class 1
limit in the European structural stainless steel design code EN 1993-1-4 [1], plastic design is not
permitted for stainless steel structures, which leads to uneconomic design. The present paper investigates
the applicability of inelastic design procedures to indeterminate stainless steel structures. Five three-
point bending tests and ten two-span continuous beam tests on stainless steel square and rectangular
hollow sections are reported herein. Analysis of the results reveals that current design provisions are
overly conservative and significant moment redistribution and hence material savings can be achieved if
inelastic design procedures are followed at both cross-sectional level and system level.

The need for metallic structures to resist high loads that have a small probability of occurrence in an
economic way necessitates the exploitation of the inelastic range of the material’s stress-strain curve,
provided that they possess sufficient ductility. Modern structural design guidance specifies the extent to
which the exploitation of the material’s inelastic range is allowed, following the cross-section
classification procedure. The European structural design codes for stainless steel EN 1993-1-4 [1] and
carbon steel EN 1993-1-1 [2] specify four behavioural classes of cross-sections according to their
susceptibility to local buckling. Indeterminate structures employing carbon steel cross-sections classified
as Class 1 may be plastically designed. Despite the high material ductility of structural stainless steels [3]
and the existence of a Class 1 limit in [1], plastic design is not permitted for stainless steel structures,
which leads to uneconomic design.
In this paper the applicability of inelastic design procedures to stainless steel indeterminate structures
is investigated. Five three-point bending tests and ten two-span continuous beam tests on stainless steel
SHS and RHS are reported. The experimental response of both the simply supported beams and the
continuous beams is then compared with the predictions of EN 1993-1-4 [1]. Analysis of the results
reveals that current design provisions are overly conservative, since they do not account for material
strain-hardening and the significant moment redistribution (in the case of the continuous beams) taking
place before collapse occurs. Hence material savings can be achieved if inelastic design procedures are
followed at both cross-section level and system level. To this end, the continuous strength method
(CSM), outlined in [4]-[6], which allows for the actual material response at cross-sectional level, is
adapted to stainless steel indeterminate structures, resulting in more favourable strength predictions.

An experimental investigation into the structural response of stainless steel simple and continuous
beams has been carried out in the Structures Laboratory at Imperial College London. The employed
cross-sections were SHS and RHS in grade EN 1.4301/1.4307 stainless steel with nominal sizes of
50×50×3, 60×60×3, 100×100×3 and 60×40×3. The specimens were extracted from the same lengths as

Marios Theofanous and Leroy Gardner

the ones utilised in the experimental study reported in [7]. The tensile coupon test data reported in [7] are
utilised herein, as no further material coupon tests were conducted. The obtained tensile flat material
properties are shown in Table 1, where E0 is the Young’s modulus, V0.2 and V1.0 are the proof stresses at
0.2% and 1% offset strains, respectively, and n and n'0.2,1.0 are strain hardening exponents, utilised in the
two stage Ramberg-Osgood model [8]-[10]. The 0.2% proof stress ı0.2, obtained from tensile flat
coupons, is utilized to obtain the elastic and plastic moment resistances (Mel and Mpl respectively).

Table 1: Tensile flat material properties.

E ı0.2 ı1.0 ıu
Cross-section n n0.2,1.0
(N/mm2) (N/mm2) (N/mm2) (N/mm2)
SHS 50×50×3 198000 552 608 798 5.50 2.90
SHS 60×60×3 197730 483 546 745 5.25 2.90
SHS 100×100×3 201300 419 470 725 5.25 2.25
RHS 60×40×3 191690 538 592 753 5.00 3.50

Five three-point bending tests were initially performed, to provide fundamental flexural performance
data, which were utilised to assess the suitability of current design provisions codified in EN 1993-1-4
(2006). Subsequently ten two-span continuous beam tests (five-point bending) were conducted, which
enabled the study of stainless steel indeterminate structures and an assessment of the current codified
provisions. Performing both simply supported and continuous beam tests on the same cross-sections
enables the assessment of the effect of moment redistribution on ultimate capacity.
2.1 Simply supported beam tests
Five simply supported beam tests have been conducted in the three-point bending configuration. One
test was conducted for each of the three SHS employed, whilst two tests were conducted for the RHS
60×40×3 specimen, one about the major axis and one about the minor axis. All beams had a total length
of 1200 mm and were simply supported between rollers, which allowed axial displacement of the beams’
ends. The rollers were placed 50 mm inward from each beam end. Wooden blocks were placed within the
tubes at the loading point to prevent web crippling. The applied crosshead movement rate was 3 mm/min.
Prior to testing, measurements of the geometry of the specimens were taken, which are summarised
in Table 2, where the experimentally obtained ultimate moment Mu and the Mu/Mel and the Mu/Mpl ratios
are also included. In Table 2, B and D are the outside width and depth of the cross-section respectively, t
is the mean section thickness and ri is the internal corner radius. A typical failure mode, exhibiting local
buckling of the compression flange and the upper part of the web, is shown in Figure 1.

Table 2: Measured dimensions and test results from 3-point bending tests.
Axis of B D t ri Mu
Specimen Mu/Mel Mu/Mpl
bending (mm) (mm) (mm) (mm) (kNm)
SHS 50×50×3 Major 50.18 50.24 2.76 1.53 7.00 1.68 1.41
SHS 60×60×3 Major 60.37 60.63 2.79 3.50 8.74 1.62 1.36
SHS 100×100×3 Major 99.85 99.93 2.78 2.13 18.77 1.35 1.16
RHS 60×40×3-MA Major 40.00 60.11 2.75 1.88 7.99 1.84 1.49
RHS 60×40×3-MI Minor 60.10 39.95 2.75 1.88 5.69 1.66 1.41

Figure 1: Failure mode of the RHS 60×40×3-MA specimen.

Marios Theofanous and Leroy Gardner

2.2 Continuous beam tests

Ten continuous beam tests were conducted on the same section sizes employed for the simply
supported beam tests; two tests were conducted for each cross-section. As before, the RHS 60×40×3 was
tested about both its major and minor axes. All beams had a total length of 2400 mm and were resting on
three roller supports; the end rollers allowed free axial displacements, while the central roller was fixed
against axial displacement. The clear span distance between the roller supports was 1100 mm and a
further 100 mm were provided at each specimen end. The measured geometric properties are shown in
Table 3, where the symbols are as previously defined.

Table 3: Measured dimensions of continuous beam specimens.

Axis of B D t ri
Specimen Configuration
bending (mm) (mm) (mm) (mm)
SHS 50×50×3-1 Major 1/2 span 50.22 50.26 2.76 1.38
SHS 50×50×3-2 Major 1/3 span 50.28 50.23 2.76 1.69
SHS 60×60×3-1 Major 1/2 span 60.38 60.68 2.79 3.50
SHS 60×60×3-2 Major 1/2 span 60.36 60.66 2.79 3.50
SHS 100×100×3-1 Major 1/2 span 99.94 99.79 2.78 2.13
SHS 100×100×3-2 Major 1/2 span 99.87 99.85 2.78 2.13
RHS 60×40×3-MA-1 Major 1/2 span 40.05 60.14 2.75 1.88
RHS 60×40×3-MA-2 Major 1/2 span 39.90 60.12 2.75 1.88
RHS 60×40×3-MI-1 Minor 1/2 span 60.10 39.90 2.75 1.88
RHS 60×40×3-MI-2 Minor 1/3 span 60.15 39.90 2.75 1.88

All tests were displacement-controlled with a loading rate of 3mm/min in terms of vertical crosshead
movement. Two symmetrical loading configurations were employed to vary the required rotation
capacity and moment redistribution before collapse. In the first configuration, denoted ‘1/2 span’ in Table
3, the loads were applied at midspan, whilst in the second configuration, ‘denoted 1/3 span’, the loads
were applied at a distance equal to 366.7 mm (1/3 of the clear span length) from the central support. The
1/3 span configuration is shown in Figure 2, where the employed instrumentation is also depicted.
Wooden blocks were inserted at the supports and at the loading points of each specimen and the loads
and reactions were applied through a steel block of thickness 15 mm and width 30 mm, to prevent local
bearing failure.

Loading jack

Spreader beam

Beam specimen
Strain gauge LVDT5 200 LVDT6



Load Cell
100 733.3 366.7 366.7 733.3 100

Figure 2: Test configuration ‘1/3 span’ - loads applied at 366.7 mm from central support.

The employed instrumentation consisted of a load cell at the central support, eight LVDTs and six
strain gauges, as shown in Figure 3. The load cell was utilised to measure the reaction force at the central

Marios Theofanous and Leroy Gardner

support, which is necessary to determine the stress condition of each specimen, due to their static
indeterminacy. The strain gauges were affixed at the mid-width of the top and bottom flanges at a
distance of 60 mm from each loading point and from the central support point. Their readings verified
that no net axial load occurred in the specimens and hence the end rollers did not provide any axial
restraint. Six LVDTs were employed in pairs at the ends of the specimens and the central support, as
shown in Figure 2, to measure the end rotations and the rotation of the plastic hinge at the central support,
whilst two additional LVDTs were employed at the loading points to measure the vertical displacement.
The applied load and crosshead movement were also recorded. All readings were taken at 2 second
The key experimental results are summarised in Table 4, including the ultimate load Fu and the
plastic rotation at ultimate load normalised by the corresponding elastic rotation at ultimate load,
șpl,max/șel,max. The load corresponding to the formation of the first plastic hinge at the central support,
denoted Fh1, and the theoretical collapse load Fcoll are also included. The load Fh1 was determined based
on elastic calculations, whereas Fcoll was determined by classical plastic analysis procedures, assuming
rigid-plastic material (and moment-rotation) response. All specimens failed by developing three distinct
plastic hinges, one at the central support and one at each loading point. A typical failure mode for the 1/2
span arrangement is displayed in Figure 3.

Figure 3: Failure mode of SHS 50×50×3-1 - configuration: 1/2 span.

Table 4: Summary of test results from continuous beam tests.

Fu Fh1 Fcoll
Specimen Configuration șpl,max/șel,max
(kN) (kN) (kN)
SHS 50×50×3-1 1/2 span 80.24 48.3 54.35 0.95
SHS 50×50×3-2 1/3 span 98.87 48.8 67.67 1.35
SHS 60×60×3-1 1/2 span 97.08 62.2 70.00 0.70
SHS 60×60×3-2 1/2 span 92.47 62.2 69.94 0.79
SHS 100×100×3-1 1/2 span 173.86 156.3 175.83 0.45
SHS 100×100×3-2 1/2 span 172.21 156.3 175.89 0.20
RHS 60×40×3-MA-1 1/2 span 92.99 52.0 58.54 1.10
RHS 60×40×3-MA-2 1/2 span 91.92 51.9 58.37 1.10
RHS 60×40×3-MI-1 1/2 span 63.94 39.0 43.84 1.00
RHS 60×40×3-MI-2 1/3 span 77.57 39.5 54.84 1.70


In this section, the reported test data are analysed and discussed. Various design methods are outlined
and their accuracy is assessed on the basis of the test data. These include the design provisions specified
in EN 1993-1-4 [1], the continuous strength method [4]-[6] and conventional plastic design, assuming
rigid-plastic material behaviour. For the simply supported beams, discrepancies between the actual
resistance and code predictions are due to the effect of material nonlinearity (i.e. strain-hardening) at
cross-sectional level, whilst for the continuous beams (indeterminate structures), nonlinearity affects both
individual cross-sections, due to material strain-hardening, and the whole structure, due to statical
indeterminacy and the corresponding moment redistribution. A method for plastic design of steel
structures, which takes into account strain-hardening, was recently proposed [11] and its applicability to
stainless steel indeterminate structures is assessed herein.

Marios Theofanous and Leroy Gardner

3.1 European codified design predictions

No distinct difference in the treatment of Class 1 and Class 2 sections exists in EN 1993-1-4 [1], since
plastic design of stainless steel indeterminate structures is not currently allowed, despite the existence of a
Class 1 slenderness limit. On average, EN 1993-1-4 [1] underestimates the capacity of the three-point bending
specimens by 33% with a coefficient of variation (COV) of 8%. Improved results in terms of consistency are
obtained when the calculation is based on the revised slenderness limits and effective width formulae,
proposed in [6], as shown in Table 5. The continuous beams are treated similarly to the simply supported ones.
Hence failure is assumed to occur when the most heavily stressed cross-section reaches its codified resistance,
as determined through cross-section classification. The codified resistance is compared to the actual capacity in
Table 6, where the predictions based on the revised slenderness limits are also included. Measured material
properties and geometries have been used throughout the comparisons.

Table 5: Codified and proposed classification and effective width formulae for simply supported beams.
EN 1993-1-4 [1] Revised slenderness limits [6]
Class Mpred/Mu Class Mpred/Mu
SHS 50×50×3 1 0.71 1 0.71
SHS 60×60×3 1 0.73 1 0.73
SHS 100×100×3 4 0.65 4 0.68
RHS 60×40×3-MA 1 0.67 1 0.67
RHS 60×40×3-MI 3 0.60 1 0.71
MEAN 0.67 0.70
COV 0.08 0.04

Table 6: Codified and proposed classification and effective width formulae for continuous beams.

EN 1993-1-4 [1] Revised slenderness limits [6]

Class Fpred/Fu Class Fpred/Fu
SHS 50×50×3-1 1 0.60 1 0.60
SHS 50×50×3-2 1 0.49 1 0.49
SHS 60×60×3-1 1 0.64 1 0.64
SHS 60×60×3-2 1 0.67 1 0.67
SHS 100×100×3-1 4 0.68 4 0.71
SHS 100×100×3-2 4 0.68 4 0.72
RHS 60×40×3-MA-1 1 0.56 1 0.56
RHS 60×40×3-MA-2 1 0.56 1 0.56
RHS 60×40×3-MI-1 3 0.52 1 0.61
RHS 60×40×3-MI-2 3 0.43 1 0.51
MEAN 0.58 0.61
COV 0.15 0.13

3.2 Continuous strength method

The continuous strength method (CSM) explicitly accounts for material strain-hardening at cross-
sectional level [4]-[6]. Hence, more favourable ultimate capacity predictions can be achieved for both
simply supported and continuous beams if the cross-section failure is based on the CSM rather than on
cross-section classification, as shown in Table 7. As expected, the ultimate capacity of the simply

Marios Theofanous and Leroy Gardner

supported beams is very well-predicted and a low COV is observed. For the continuous beams, the CSM
gives more favourable strength predictions compared to the classification procedure, but failure to
account for moment redistribution results in excessive conservatism. Moreover, a relatively large COV is
observed, due to the dependency of the effect of moment redistribution on the cross-section slenderness.

Table 7: Assessment of the CSM for simply supported and continuous beams.
Simply supported beams Continuous beams
SHS 50×50×3-1 0.90 0.68
SHS 50×50×3-2 - 0.56
SHS 60×60×3-1 0.95 0.73
SHS 60×60×3-2 - 0.77
SHS 100×100×3-1 0.91 0.89
SHS 100×100×3-2 - 0.90
RHS 60×40×3-MA-1 0.87 0.64
RHS 60×40×3-MA-2 - 0.64
RHS 60×40×3-MI-1 0.87 0.67
RHS 60×40×3-MI-2 - 0.56
MEAN 0.90 0.71
COV 0.04 0.17

3.3 Conventional plastic analysis

Allowing for the effects of moment redistribution is the key feature of plastic analysis. Despite the
deviation of stainless steel’s material response from the assumed bilinear elastic, perfectly-plastic model,
application of plastic design to stainless steel indeterminate structures is attempted herein. The theoretical
collapse load Fcoll has been calculated for all continuous beam specimens and is given in Table 4. In
Table 8, the classification procedure codified in EN 1993-1-4 [1] and that proposed in [6] are once again
assessed; in this case the capacity of the specimens with Class 1 cross-sections is calculated by means of
plastic design, the resistance of the Class 3 beams is calculated using elastic design and for Class 4
beams, elastic design and effective section properties are used. The revised classification approach seems
to offer more consistent ultimate capacity predictions than the one codified in EN 1993-1-4 [1]. However
the embedded conservatism remains significant.

3.4 Continuous strength method for indeterminate structures

Both the CSM and plastic analysis offer significant improvements in terms both design efficiency
compared to the current design approach. However, plastic analysis seems superior to the CSM in terms
of consistency of the predictions. This is due to the fact that, when applying the CSM, the effect of
moment redistribution has been ignored, thereby reducing the failure of a structural assembly to the
failure of a single cross-section.
A method combining the merits of both is desirable, since both strain-hardening at cross-sectional
level and moment redistribution affect the structural response of stainless steel indeterminate structures.
Gardner and Wang [11] recently proposed a modification to the plastic analysis procedure currently
applied to carbon steel structures. The proposed method, called the CSM for indeterminate structures,
allows for moment redistribution in a similar fashion to traditional plastic analysis and for full
exploitation of material strain-hardening at the location of the first plastic hinge; strain-hardening at
subsequent hinges is partly accounted for.

Marios Theofanous and Leroy Gardner

Table 8: Assessment of codified and proposed classification for continuous beams allowing for plastic design.
EN 1993-1-4 [1] Revised slenderness limits [6]
Class Fpred/Fu Class Fpred/Fu
SHS 50×50×3-1 1 0.68 1 0.68
SHS 50×50×3-2 1 0.68 1 0.68
SHS 60×60×3-1 1 0.72 1 0.72
SHS 60×60×3-2 1 0.76 1 0.76
SHS 100×100×3-1 4 0.68 4 0.71
SHS 100×100×3-2 4 0.68 4 0.72
RHS 60×40×3-MA-1 1 0.63 1 0.63
RHS 60×40×3-MA-2 1 0.63 1 0.63
RHS 60×40×3-MI-1 3 0.52 1 0.69
RHS 60×40×3-MI-2 3 0.43 1 0.71
MEAN 0.64 0.69
COV 0.15 0.06

The novelty of the method lies in departing from the traditional rigid-plastic material response and
assuming that the ultimate moment capacity of the first plastic hinge can be fully exploited. In essence,
the method utilises the upper bound theorem of limit analysis and relies on the determination of a suitable
collapse mechanism. The moment capacity at the location of the plastic hinges is calculated by means of
the CSM; for the first plastic hinge the full deformation capacity is exploited, whilst for subsequent
plastic hinges, the deformation capacity is a fraction of the deformation capacity at the first hinge,
proportional to the plastic rotation ratio as determined from kinematics.

Table 9: Assessment of the CSM for indeterminate structures.

CSM for indeterminate
Specimen structures
Class Fpred/Fu
SHS 50×50×3-1 1 0.85
SHS 50×50×3-2 1 0.86
SHS 60×60×3-1 1 0.92
SHS 60×60×3-2 1 0.96
SHS 100×100×3-1 4 0.89
SHS 100×100×3-2 4 0.90
RHS 60×40×3-MA-1 1 0.80
RHS 60×40×3-MA-2 1 0.80
RHS 60×40×3-MI-1 1 0.84
RHS 60×40×3-MI-2 1 0.86
MEAN 0.87
COV 0.06

The accuracy of the CSM for indeterminate structures is assessed in Table 9, where all cross-sections
classified as Class 1 according to the revised slenderness limits proposed in [6] have been treated with
this method. The SHS 100×100×3 specimens, which have a slender (Class 4) cross-section, have been
treated with the conventional CSM; hence the effect of moment redistribution has not been considered for
these sections. Overall, significant enhancement in design efficiency and good agreement with the test

Marios Theofanous and Leroy Gardner

results is observed as evidenced by the low COV of 0.06. Further research into the topic is underway to
determine the slenderness range within which the proposed method can be safely applied.

An experimental study comprising five three-point bending tests and ten two-span continuous beam
tests (five-point bending) has been conducted and the conservatism embedded in the provisions for
stainless steel indeterminate structures codified in EN 1993-1-4 [1] has been highlighted. The application
of conventional plastic analysis to stainless steel indeterminate structures and the accuracy of the CSM
have been investigated. It was concluded that both material strain-hardening at cross-sectional level (at
the location of the plastic hinges) and moment redistribution occurring in indeterminate structures,
comprising sections with sufficient deformation capacity, are significant and should therefore be
accounted for in design. A recently proposed adaptation of the CSM for carbon steel indeterminate
structures [11] has been further investigated and applied to stainless steel indeterminate structures,
yielding excellent results for stocky cross-sections. Hence CSM for indeterminate structures emerges as a
promising design approach for stainless steel continuous beams.

[1] EN 1993-1-4. Eurocode 3: Design of steel structures - Part 1.4: General rules – Supplementary
rules for stainless steel. CEN, 2006.
[2] EN 1993-1-1. Eurocode 3: Design of steel structures - Part 1.1: General rules – General rules and
rules for buildings. CEN, 2005.
[3] Gardner, L. “The use of stainless steel in structures”. Progress in Structural Engineering and
Materials, 7 (2), 45-55, 2005.
[4] Gardner, L. and Ashraf, M. “Structural design for non-linear metallic materials”. Engineering
Structures, 28 (6), 926-934, 2006.
[5] Ashraf, M., Gardner, L. and Nethercot, D. A. “Structural stainless steel design: Resistance based on
deformation capacity”. Journal of Structural Engineering, ASCE, 134 (3), 402-411, 2008.
[6] Gardner, L. and Theofanous, M. “Discrete and continuous treatment of local buckling in stainless
steel elements”. Journal of Constructional Steel Research, 64 (11), 1207-1216, 2008.
[7] Nip, K.H., Gardner, L. and Elghazouli, A.Y. “Cyclic testing and numerical modelling of carbon
steel and stainless steel tubular bracing members”. Engineering Structures, 32 (2), 424-441, 2010.
[8] Mirambell, E. and Real, E. “On the calculation of deflections in structural stainless steel beams: an
experimental and numerical investigation”. Journal of Constructional Steel Research, 54 (1), 109-
133, 2000.
[9] Rasmussen, K.J.R. “Full-range stress-strain curves for stainless steel alloys”. Journal of
Constructional Steel Research, 59 (1), 47-61, 2003.
[10] Gardner, L. and Nethercot, D.A. “Experiments on stainless steel hollow sections - Part 1: Material
and cross-sectional behaviour”. Journal of Constructional Steel Research, 60 (9), 1291-1318, 2004.
[11] Gardner, L. and Wang, F. “Influence of strain hardening on the behaviour and design of steel
structures”. International Journal of Structural Stability and Dynamics, (submitted).

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010


Renata G. L. da Silva*, Armando C. C. Lavall*

* Departamento de Engenharia de Estruturas da Universidade Federal de Minas Gerais

e-mails: rglsilva@dees.ufmg.br, lavall@dees.ufmg.br

Keywords: Inelastic Behaviour, Partially Restrained Connections, Spread of Plasticity.

Abstract. The behaviour of beam-column connections for conventional analysis of a structure is

simplified to the two idealized extremes of either rigid-joint or pinned-joint behaviour. However most of
the connections used in steel frames actually exhibits semi-rigid deformation which influences the global
behaviour of structures. This paper presents the development of a finite element for use in second-order
inelastic analysis of partially and fully restrained planar steel frames. The finite element considers the
spread of plasticity within the cross section and along the member length, several residual stresses
distributions, shear deformation of members through the Timoshenko theory and P-G and P-' effects.
Nonlinear spring elements are used to include connections. The behaviour of connections is modelled
using multilinearized moment-rotation curves. A computer program associated with the finite element
model is developed for Advanced Analysis of planar steel frames. Numerical examples are presented.

Conventional analyses of steel frame structures are usually carried out under the assumption that the
beam-column connections are either fully rigid or ideally pinned. However, most of the connections used
in current practice are semi-rigid type whose behaviour lies between these two extreme cases. The
predicted response of the idealized structure may be quite unrealistic compared to that of the actual
structure if connection stiffness is ignored in the analysis and design procedures.
The semi-rigid connections have important function in structural steel design, because influence
substantially the moment distribution in beams and columns and negatively affect the stability of the
frame, since they increase the drift of the frame and cause a decrease in effective stiffness of the member.
So, the disregard of the actual behaviour of the connections can lead to unrealistic predictions of response
and resistance of structures.
The important attributes that affect the behaviour of semi-rigid steel frames structures are connection,
geometric and material nonlinearities. The connection nonlinearity is given by the nonlinear moment-
rotation relationship of semi-rigid connections. The geometric nonlinearity includes second-order effects
associated with the P-G and P-' effects and geometric imperfections. And finally, material nonlinearity
includes spread of yielding or plasticity associated with the influence of residual stresses. The realistic
modelling of a steel frame requires the use of these attributes if an accurate response is to be obtained.
One way to account for all these effects in semi-rigid frame design is through the use of an advanced
analysis. Advanced Analysis is a method that can sufficiently capture the limit state of strength and
stability of a structural system and its individual members, so that separate checks of the capacity of
members are not required. With technological advances in computational area has been possible to
employ advanced analysis techniques directly in the offices of engineering design.
During the past 20 years, researches efforts have been devoted to the development and validation of
several nonlinear inelastic analysis methods for steel frames with semi-rigid connections, as the studies
presented by [1]-[9]. The behaviour of semi-rigid connections has been progressively incorporated in

R. da Silva, A. Lavall

structural analyses, resulting in more realistic analysis of the response global of structures, allowing a
design accurate and certainly more economical.
This paper presents the development of a finite element for use in the second-order inelastic analysis
of partially (PR) and fully (FR) restrained planar steel frames. The finite element considers the spread of
plasticity within the cross section and along the member length, several residual stresses distributions,
shear deformation of members, P-G and P-' effects. Nonlinear spring elements are used to include
partially restrained connections. The behaviour of the connections is modelled using multilinearized
moment-rotation curves. The formulation considering Timoshenko theory and self equilibrated residual
stresses is based on updated Lagrangian formulation. The Corotacional technique is used to obtain the
element’s tangent stiffness matrix. A computer program associated with the finite element model is
developed. Numerical examples are presented and the results are compared with those previously
published by others researchers with the objective to validate the finite element model for the Advanced
Inelastic Analysis.


The knowledge of connections behavior between structural elements is essential for the analysis and
design of a structure. Efforts transmitted through the beam-column connections consist of axial force,
shear force, bending moment and torsion. The effect of axial and shear forces can be negligible when
their deformations are small compared to the rotational deformation of connections. The effect of torsion
is excluded of in-plane study. So in this work, only the effect of bending moment in the rotational
deformation of the connections will be considered.
The moment-rotation relationship, M-Tr, depends on the connection type. The rotational deformation
is expressed as a function of the moment in the connection. The angle Tr corresponds to the relative
rotation between beam and column at the connection.
Most experiments have shown that the curve M-Tr is nonlinear in the whole domain and for all
connections types. May be observed that a flexible connection has a smaller ultimate moment capacity
and a larger rotation, and vice versa for a rigid connection. The behaviour of a simple connection is
represented by Tr-axis with M=0 and the behaviour of a fully-rigid connection is represented by the M-
axis with Tr=0. All semi-rigid connections are represented by curves lying between these two extremes,
allowing some moment to be transferred and some rotation to occur in a connection.
Experimental works on connections have been performed, and a large body of moment-rotation data
has been collected, as researches of [10]-[11]. Using these databases, researchers have developed several
connection models. The main are: linear; bilinear, trilinear, multilinear; polynomial; b-spline; three-
parameter power and exponential models.
The multilinear model is proposed in this work to represent moment-rotation curves of partially
restrained connections. This model is simple and able to describe the M-Tr curve with higher precision
than the bi and trilinear models. The values of the pair bending moment and rotation are inserted directly
as input in the program and the stiffness values for each segment are automatically calculated for a given
connection. Unloading and reloading of the connection are assumed to follow the initial stiffness. A
representation with five linear segments of the moment-rotation curve is shown in figure 1.

Figure 1: Multilinear moment-rotation curve for partially restrained connections

R. da Silva, A. Lavall


This paper outlines the development of a finite element for use in the second-order inelastic analysis
of partially (PR) and fully (FR) restrained planar steel frames. This finite element is shown in figure 2.
The structural nodes have three degrees of freedom, namely, the displacements u and v along the x and y
axes, respectively, and the rotation T, positive when counter-clockwise. In the reference configuration,
the chord between elements nodes has the length lr. On the chord a local reference coordinate system
(xr,yr) is placed, with the origin at the center. The angle between the axis x and the chord is denoted by
Ir. At current configuration the chord between element nodes has length lc. A corotational coordinate
system (xc,yc) is defined on this chord, with the origin at the center, as indicated in the figure 2. The angle
between the axis x and the chord is now Ic while the angle between the chord and the axis of the bar is
denoted by D.

Figure 2: Finite element deformation

The natural and Cartesian degrees of freedom of the element are defined, respectively, by:

qDT ^q1 lc  l r ; q 2 D a ; q3 D b ` ; p ^u a ; va ;T a ; ub ; vb ;T b ` (1)

The relations between natural and Cartesian degrees of freedom are important and listed below:

­q1 lc  lr
® q2 D a T a  T c p3  M c  M r (2)
°q D T  T p6  Mc  M r
¯ 3 b b c

Longitudinal and shear deformations are, respectively:

du du dT
Hx  yr H x  yr D ' (3)
dx dx dx
du dv
J xy   T  D J (4)
dy dx

The virtual power theorem is used in the development of the finite-element stiffness:

³ V GH dVr  ³ W GJ dVr Pi Gpi (5)

Vr Vr

where dVr is the volume element in the reference configuration, ı the normal stress, W the shear stress, įİ
virtual longitudinal deformation and įJ virtual distortion of a fiber.
The virtual longitudinal deformation and virtual distortion are respectively:
GH H ,D qD ,iGpi ; GJ J ,D qD ,iGpi (6)

R. da Silva, A. Lavall

Therefore, the equilibrium equation of the element is given by:

§ ·
Pi ¨ ³ V H ,D dVr  ³ W J ,D dVr ¸qD ,i QD qD ,i
¨V ¸ (7)
© r Vr ¹

Considering an incremental formulation of equilibrium, differentiation of P at time can be given by:

dP wP dp dp
kt (8)
dt wp dt dt

where, kt is the tangential stiffness matrix of element in Cartesian coordinates. The components kij are
obtained through differentiation of Pi with respect to Cartesian coordinate’s pj:
kij qD ,i QD ,E qE , j  QD qD ,ij (9)

§ dV dW ·
QD ,E ³Vr ¨¨ H ,D H ,E  VH ,DE  J ,D J ,E  W J ,DE ¸¸ dVr (10)
© dH dJ ¹

The tangent stiffness matrix is given by:

§ § dV dW · · § ·
k ij qD ,i ¨¨ ³ ¨¨ H ,D
d H
H ,E  J ,D
d J

J ,E ¸¸dVr ¸¸ q E , j  qD ,i ¨¨ ³ V H ,DE  W J ,DE dVr ¸¸q E , j  QD qD ,ij (11)
© Vr © ¹ ¹ © Vr ¹

The first term of the equation represents the constitutive part, the second and third parts represent the
P-G and P-' effects, respectively.


The matrices obtained from the formulation were implemented in the program developed by [12].
The program, written in FORTRAN 90, employs Newton-Raphson method and the displacement control
to obtain the nonlinear equilibrium path and to allow the correct determination of the collapse load.
The program considers P-G and P-' effects, partially restrained connections, shear deformations
through the Timoshenko theory and spread of plasticity. The frame element, made up of layers, enables
to identify the plastic region through the cross section and along the member length and to consider any
kind of residual stresses distribution.
The connection behaviour is characterized by moment-rotation curve. Nonlinear spring elements are
used for an approximation of the actual connection behaviour. The spring elements have three degrees of
freedom, namely, the displacements u and v along the x and y axes, respectively, and the rotation T,
positive when counter-clockwise. Stiffness is given in terms of relative displacements. In this study, the
rotational stiffness KT is obtained by linearized curves of different types of connections available in the
literature. Unloading and reloading can be considered in any segment of the curve.


5.1 Portal Frame

This example aims to study the loading and unloading behaviour of the connections of a portal frame,
when it’s subjected to a lateral force after the total vertical loading to be applied in the structure, as

R. da Silva, A. Lavall

shown in figure 3. The uniformly distributed load is modelled as a set of equivalent nodal loads. The
frame is analyzed in second-order elastic theory. Beam and columns are made of profiles I. The area and
the moment of inertia of the beam are, respectively, equals to 43 cm2 and 2770 cm4. The area and the
moment of inertia of columns are equals to 33,4 cm2 and 1510 cm4, respectively. The lengths and
numbering of bars are also shown in figure 3. The Young’s modulus is 21000 kN/cm². Beam and
columns were modelled, respectively, with ten and four finite elements and the cross sections were
divided into twenty layers.
1 kN/cm
30 kN
2 3 4 5
400 cm
1 6
600 cm
Figure 3: Portal frame with semi-rigid connections

The top and seat angle with double web angle connections, C23 and C45, are identical at the ends of
the beam. The behaviour and data of the connections are presented by [11], considering the three-
parameter model. At this paper, the behaviour of the semi-rigid connections is represented by multilinear
curves, according figure 1, with data shown in table 1.

Table 1: Parameters connection

Segments (i) 1 2 3 4 5
Mi (kNcm) 2316 4632 6176 6948 7566
Tri (rad) 0,00097 0,00366 0,01046 0,02510 0,15745

Figure 4 shows the graphic of moment versus relative rotation for the windward connection (C23)
and leeward connection (C45), for all loading increments. It is observed that, when the frame is subjected
firstly to uniformly distributed load, the connections presented same behaviour. The bending moment is
equal to 7245 kNcm and the relative rotation is equal to 0,08880 rad, for total vertical loading. When
lateral load is applied, the windward connection (C23) unloads, showing a linear behaviour with slope
equal to the initial stiffness and the leeward connection (C45) continues load, ie, continues to rotate in the
same direction, with slope based on tangent stiffness, as shown in Figure 4. For the structure completely
loaded, the connection C45 presents moment equal to 7437 kNcm and relative rotation equal to
0,12982 rad. The connection C23, due to the unloading caused by the lateral force, presents moment
equal to 2395 kNcm and relative rotation of 0,08676 rad. It can be concluded that, the loading
characteristics are very different from the unloading characteristics of the connections. The connection
behaviour is very much affected by the history and direction of the loads applied sequentially.

Moment (kNcm)

4000 Windward
3000 connection
2000 Leeward
1000 connection
0,00 0,02 0,04 0,06 0,08 0,10 0,12 0,14
Relative Rotation (rad)
Figure 4: Behaviour of partially restrained connection under loading and unloading

R. da Silva, A. Lavall

5.2 Ten-story and two bay frame

Consider the ten-story and two-bay frame with endplate connections shown in figure 5. The frame
has been analyzed and designed previously by [13], for loads and dimensions shown in figure. The nodes
of structural model have the same semi-rigid connection type. The frame is investigated to demonstrate
the effect of semi-rigid connections on structural response up to failure. The second-order inelastic
analysis, by the plastic-zone method is performed. Two cases are analyzed: semi-rigid and fully-rigid
113,43 kN
56,71 kN
21,00 kN

41,19 kN

39,90 kN

38,61 kN

10 @ 457,2 cm = 4572 cm
37,28 kN

35,76 kN

33,27 kN

32,34 kN

29,40 kN

26,69 kN Beams: W 530x123

Columns: W 530x300

914,4 cm 914,4 cm

Figure 5: Ten-story and two-bay frame with endplate connections

The vertical and lateral loadings were considered incrementally in the numerical analysis until the
failure. The yielding strength and Young’s modulus of the steel are assumed equals to 25 kN/cm2 and
20000 kN/cm2, respectively, in the elastic-perfectly plastic behaviour. The initial geometric imperfections
and the residual stresses are not considered in the analysis. Beams and columns were modelled with four
finite elements in the structural model and the cross sections were divided into twenty layers. Three
endplate connections types are considered, whose parameters can be found in [13]. At this paper, the
connections behaviour is represented by multilinear curves with data shown in table 2, according
figure 1.

Table 2: Parameters connections

C1 Connection C2 Connection C3 Connection
Segments M (kNcm) Tr (rad) M (kNcm) Tr (rad) M (kNcm) Tr (rad)
1 19108 0,00092 24996 0,00085 36635 0,00070
2 57325 0,00703 49991 0,00284 73269 0,00245
3 76433 0,01490 74987 0,00688 109904 0,00613
4 95542 0,03083 99982 0,01496 146538 0,01375
5 107771 0,04979 124978 0,03177 183173 0,03015

The figure 6 shows the load-deflection behaviour of the frame with endplate connections, with
different rigidity and moment capacity, namely C1, C2 and C3 connections, until the failure. In graphics
of this figure, the abscissa axis represents the maximum lateral sway of the top of the frame and the

R. da Silva, A. Lavall

ordinate axis denotes the load level. It is observed that, the load originally proposed, according figure 5,
was gradually expanded until the strain of steel reached the limit of 21Hy. Results of the load-deflection
behaviour, obtained by the developed program, are compared with results of [13], obtained by a computer
program able of performing a second-order inelastic analysis of planar steel structures based on the
refined plastic hinge method. It can be noted that the results of the program, developed based on the
plastic-zone method, showed a good correlation with the results obtained by [13]. A difference less than
5% in the ultimate load between the analyses was obtained for all structural models.
Frame with C1 connection Frame with C2 connection Frame with C3 connection
2,5 (2,35) 2,5 (2,44) 2,60 (2,51)
2,0 (2,30) (2,35) (2,40)
2,0 2,08
Load Factor

Load Factor

Load Factor
1,5 1,5 1,56
1,0 1,0 1,04
Zhou Zhou Zhou
0,5 Program 0,5 Program 0,52 Program
0,0 0,0 0,00
0 15 30 45 60 75 90 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70
Lateral Deflection (cm) Lateral Deflection (cm) Lateral Deflection (cm)
Figure 6: Load-displacement behaviour at the top of the frame

Figure 7 shows, comparatively, the load-deflection behaviour at the top of frame, obtained by the
program developed, considering conventional rigid connections and the C1, C2 and C3 connections. The
results show that the frame with the C1 connection has larger deflection, resulting in the more flexible
structure between models analyzed. The frame with the C3 connection presents deflection and load factor
values very close to the conventional model with perfectly rigid connections. It can be concluded that the
properties of connections have significant influence on the strength, stiffness, and ductility of the frame.

Load Factor

1,2 Frame with C1 connection
0,9 Frame with C2 connection
0,6 Frame with C3 connection
0,3 Rigid Frame
0 10 20 30 40 50 60 70 80 90
Lateral Deflection (cm)
Figure 7: Load-displacement behaviour at the top of frame obtained by the program developed

It can be noted that, when connections in a frame become stiffer, the response of the semi-rigid frame
get close to the rigid frame. So, endplate connections can be regarded semi-rigid or rigid depending on
their rigidity. A frame with endplate connections can be regarded as a rigid frame if its connections are
rigid enough.

A computer program for Advanced Inelastic Analysis of partially (PR) and fully (FR) restrained
planar steel frames, considering the geometric, material and connections nonlinearities, is developed. The
finite element considers P-G and P-' effects, shear deformations of members through the Timoshenko

R. da Silva, A. Lavall

theory and spread of plasticity. The frame element, made up of layers, enables to identify the plastic
region through the cross section and along member length and to consider any kind of residual stresses
The method of advanced analysis showed to be very efficient in the analysis of the behaviour of steel
structures involving semi-rigid connections from the initial stage of loading until the final stage of
collapse. The results indicate that semi-rigid connections in steel structures have fundamental importance,
since greatly affect the behaviour of the structure. The developed method considering the nonlinear
behaviour of the connections through M-Tr multilinear curve showed to be suitable for these analyses.

The authors are grateful to FAPEMIG – Fundação de Amparo à Pesquisa do Estado de Minas Gerais,
for its support to carry out this research.

[1] Liew, J.Y.R., White, D.W., Chen, W. F. “Second-Order Refined Plastic Hinge Analysis of Frame
Design: Part I”. Journal of Structural Engineering, 119(11), 3196-3216, 1993.
[2] Chen, W.F. and Toma, S. Advanced Analysis of Steel Frames: Theory, Software, and Applications.
CRC-Press, Boca Raton, 1994.
[3] Kruger, T. S., van Rensburg, B. W. J., du Plessis, G. M. “Non-linear Analysis of Structural Steel
Frames” Journal of Constructional Steel Research, 34, 1995.
[4] Kim, S.E. and Chen, W.F. “Practical advanced analysis for semi-rigid frame design”. Journal
Engineering. Fourth Quarter, 129-141, 1996.
[5] Chen, W.F., Goto, Y., Liew, J.Y.R. Stability Design of Semi-Rigid Frames. John Wiley e Sons,
Inc., New York, 1996.
[6] Foley, C.M. and Vinnakota, S. “Inelastic Behaviour of Multistory Partially Restrained Steel
Frames. Part I”. Journal of Structural Engineering, 125(8), 854-861, 1999.
[7] Foley, C.M. & Vinnakota, S. “Inelastic Behaviour of Multistory Partially Restrained Steel Frames.
Part II”. Journal of Structural Engineering, 125(8), 862-869, 1999.
[8] Gizejowski, M.A., Barszcz, A.M., Branicki, C.J., Uzoegbo, H.C. “Review of analysis methods for
inelastic design of steel semi-continuous frames”. Journal of Constructional Steel Research, 62, 81-
92, 2006.
[9] Liu, Y., Xu, L., Grierson, D.E. “Compound-element modeling accounting for semi-rigid
connections and member plasticity”. Engineering Structures, 30, 1292–1307, 2008.
[10] Chen, W.F and Kishi, N. “Semi-Rigid Steel Beam-to-Column Connections: Data Base and
Modeling”. Journal of Structural Engineering, 115(7), 105-119, 1989.
[11] Kishi N., Ahmed, A., Yabuki, N., Chen, W.F. “Nonlinear Finite element Analysis of Top-and Seat-
Angle with Double Web-Angle Connections”. Journal of Structural Engineering and Mechanics,
12(2),201-214, 2001.
[12] Lavall, A.C.C. Uma Formulação Teórica Consistente para a Análise Não-linear de Pórticos Planos
pelo Método dos Elementos Finitos Considerando Barras com Imperfeições Iniciais e Tensões
Residuais nas Seções Transversais. Ph.D.Thesis. School of Civil Engineering, São Carlos, 1996.
[13] Zhou, F. Model-Based Simulation of Steel Frames with Endplate Connections. Ph.D. Thesis,
Department of Civil and Environmental Engineering, University of Cincinnati, Ohio, p. 133, 2005.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Lars Rölle* and Ulrike Kuhlmann*

* Institute for Structural Design, University of Stuttgart, Germany

e-mail: sekretariat@ke.uni-stuttgart.de

Keywords: Robustness, robust design, alternate load paths, ductility, composite joints, catenary action

Abstract. The paper highlights the ductility demand of beam-to-column connections in the frame of
progressive collapse assessment of steel or composite structures considering sudden column loss. For
progressive collapse mitigation different design strategies are available to increase the collapse
resistance of a building. A very effective design strategy in buildings is the provision of alternate load
paths. Besides strengthening the structural system alternate load paths could be also realized by allowing
change of bearing mechanism within the structural elements. Therefore the structural system has to
undergo large deformations resulting in high demands on ductility of members and joints. The ductile
joint configurations, presented within this paper, allow for redistribution of internal forces within the
structural system by enabling large deformations. So they are contributing to the redundancy of steel or
composite frame structures due to their beneficial properties concerning ductility supply, the possibility
to activate plastic reserves as well as energy absorption capacity.

Depending on the public or commercial relevance of a building today it is no longer sufficient for
engineers to consider only basic design criteria for planning of structural framework. Engineers are
increasingly required to consider progressive collapse mitigation as additional design criteria. For
building structures the design strategy of alternate load path is therefore quite effective. The alternate load
path method is realized for that matter by activating plastic system reserves and by transition from
flexural loading to membrane tensile action in the members and joints initiating of catenary action.
Therefore the joints have to be designed in detail and all single joint components have to be adjusted in
such a manner that under bending and tensile loading at each time of loading the weakest component has
to be always ductile. This is feasible with only small additional effort by using the inherent plastic
reserves of the material steel.


2.1 General
The definition of robustness as given in EN1991-1-7 [2] refers to limiting local failure to such an
extent that no disproportionate collapse occurs. Such a general statement of robustness is very close to the
definition of the concept of collapse resistance. Collapse resistance has to be provided to ensure the
mitigation of progressive collapse. Robustness as characteristic of the load-bearing structure is thereby of
special importance. A robust structure is at the same time collapse resistant [12].
Increasing the redundancy of the structure by well designed alternate load paths is advantageous if
local failure is accepted and limitation of the collapse of the remaining structure is required. Therefore the
structure has to be designed to be able to redistribute the loads from the damaged part into the undamaged
part by avoiding at the same time a propagation of the collapse disproportional to the initial failure.

Ulrike Kuhlmann and Lars Rölle

Particularly in cases in which collapsing parts cause impact loading on key elements. Designing for such
high impact loads is, in most cases, not possible. Such conditions are particularly found in structures of
primarily vertical alignment, such as buildings structures.
2.2 Redundancy as robustness measure
In order to be effective as robustness measure alternate load paths have to be designed sufficiently
strong to transfer the actual occurring forces, including e.g. over-strength effects. An alternate load path
may on the one hand be formed within the structural system (global level) by e.g. strengthening of
transfer girders or by bracing a full floor level to suspend the loads above the damaged part like an
outrigger, see Figure 1a. On the other hand redundancy can be achieved by allowing force redistribution
within a structural member (local level). Structural steel and composite buildings with inherent
sufficiently ductile material behavior allow large deformations when local failure occurs. Large
deformations result in large plastic strain rates of material which enables the activation of additional
plastic material reserves. So on local level the material steel has the capability to form plastic hinges
which all activating also plastic system reserves by redistributing. On global level the redundancy of steel
structures for progressive collapse mitigation may form alternate load paths. e.g. by activation of catenary
action in the horizontal members that means by transition from flexural loading to membrane tensile
acting in the members and joints, see Figure 1b. Therefore a highly ductile behavior of all structural
members combined with sufficient strength is necessary. In framed structures the joints are in general the
weakest link and therefore special focus is on the joint design to avoid a premature failure of the
connections during the procedure of force redistribution.

a) b)
Figure 1: Alternate load paths by a) strong transfer girders or b) catenary action

Steel and composite members benefit from the ductile material behavior of structural steel. So steel
has the capability to combine strength, ductility and energy absorption capacity which are basic properties
for designing robust and redundant buildings.
Plastic material reserves of steel depend on the distance between the level of the nominal values and the
actual values as well as on the ratio of fu/fy. In a structural robustness analysis the actual material
properties are of main interest. Information about actual material resistance of steel is e.g. available in the
probabilistic model code of the Joint Committee of Structural Safety [6].
2.4 Ductility demands for members and joints
For common steel profiles in structural engineering depending on the rotation capacity of the cross-
section diverse categories of ductility classes exist. So the capability of the cross-section to undergo
locally a total plastification i.e. to develop a plastic hinge and to assure additionally sufficient rotation
capacity without premature stability failure (class 1 cross-section) is ensured by slenderness limits of
cross-sectional parts. Therefore for plastic analysis of a steel structure including redistribution the
requirements according the various codes are to use only those cross-sections with sufficient moment
bearing capacity as well as rotation capacity. If rigid and full-strength joints are used the plastic hinges
are located in the beams. So the total required deformation and rotation capacity to activate the membrane
effect in the direct affected part of the structure has to be offered by the beam members. But full-strength
joints cause much additional effort and they are costly.
Ductility demands for joints are decisive for partial-strength joints which have less resistance than the
beams but also reduced fabrication costs compared to full-strength joints. Using partial-strength joint

Ulrike Kuhlmann and Lars Rölle

configurations the plastic hinges are developing initially in the joint which requires high rotation
capacities of the joints. Therefore a detailed joint design is necessary considering the interaction of all
joint components including over-strength effects to ensure that for the whole loading sequence of the joint
the decisive weakest component is always ductile, see also chapter 4.
For bolted connections there is interplay of hardening or over-strength effects and the various
deformation capacities of the single components. By ensuring that especially the components “endplate in
bending” and “column flange in bending” have a certain ductility additional membrane effects on local
level (in the T-Stub) may be activated leading to a further increase of the resistance provided there is
sufficient bearing capacity of the bolts.


In steel-concrete-composite structures the choice of the slab system not only influences the erection
time and building costs but also the redundancy of the global structure or the structural robustness as
characteristic of the structural system. Depending on the slab design, for exceptional load cases like
column loss either 3D-behavior or only 2D-behavior is available.

Figure 2: Composite frame under the event column loss with a) 2D-effect and b) 3D-effect

For framed composite structures without a continuous slab (floor system as single spans with only
minimum reinforcement) the single slabs are not transversely tied together. For this reason in case of
column loss the membrane action may only be activated in plane of the directly affected frame. So only
the composite main beam system is able to redistribute forces and offer alternate load paths.
The slab in this case is unable to activate additional membrane effects transversally to the frame
plane. So only 2D-behavior may be assumed within a large displacement analysis, see Figure 2a.
Having a continuous RC slab in the composite structure including a uniform amount of reinforcement
the slab is connecting the single frames transversally. Consequently for the event of column loss the slab
is not only contributing to the resistance of the composite beam in the plane of the directly affected frame
but also acting as a tie in transversal direction. The RC slab provides ties in two horizontal directions and
enables therefore 3D-behavior for the case of column loss (see Figure 2b). However, for effective opera-
tion the continuity and anchorage of the ties is obligatory. Activating membrane action in longitudinal
direction within the composite beams and joints and additionally in transversal direction within the RC
slab the redundancy of the composite frame is clearly increased in comparison to 2D systems.


4.1 General
For partial-strength joint solutions highly ductile joint behavior is especially important due to the fact
that the plastic hinge is located in the joint and all global deformations have to be realized mainly by joint
rotation/deformation. Therefore the joints are the decisive link in the structure and their resistance and

Ulrike Kuhlmann and Lars Rölle

deformability define the global redundancy of the structure. In comparison to nominally pinned joints
there are only small extra costs (material + labor) but much more redundancy of the structure, so by only
small additional efforts the effectiveness concerning progressive collapse mitigation is improved [10].
4.2 Design of the steel elements of the joint
For the design of pure steel joints or for the steel elements of composite joints the adjustment of the
single joint components is of high importance in order to design highly ductile joint configurations.
Therefore the parameters mainly influencing the joint behavior have to be treated with special care. Table
1 describes the qualitative degree of influence concerning the rotation capacity as well as the bearing
capacity of the various parameters investigated for the joint components. So small changes of some of
these parameters might positively influence the ductility of the joint in a significant way whereas at the
same time the bearing capacity is only decreasing marginally. The listed parameters are mainly
influencing the components “endplate in bending” and “column flange in bending” which are able to
activate additional local membrane effects under large deformations. So local additional bearing effects
may compensate the decrease of the joint bending capacity when reducing e.g. the endplate thickness or
the steel grade.
Table 1: Qualitative influences of main steel joint parameters
Parameter influence on rotation capacity Influence on bearing capacity
ratio dbolt/tendplate + -
bolt arrangement + -
steel grade endplate + -
ratio tendplate/tcolumn flange o n.n
+ disproportionate high o proportionate - little

4.3 Design of the concrete slab in a composite joint

By adding a reinforced concrete slab and shear connectors to the pure steel joint a composite joint is
obtained. To get also a highly ductile behavior for the composite joints the tension bar in the slab in the
hogging moment region should be designed with high deformation capacity. Thus the reinforcement
within the joint region should be able to undergo high plastic strains. As meshed reinforcement has a
negative influence on the deformation capacity only steel rods should be used. Furthermore the following
parameters are influencing significantly the available extension in the slab:
- class of reinforcement
- reinforcement ratio
- and arrangement of shear connectors
According to EN 1992-1-1 [3], Annex C there are three classes of reinforcement A, B and C in which
class C (seismic steel) is the most ductile one in terms of maximal available strain and high ratio of fu/fy.
For class C the ratio fu/fy is higher than 1.11, which is relevant for high available strains of the reinforced
slab under tension resulting in a high deformation capacity as visible in Figure 3a.

a) b)
Figure 3: a) Influence of ratio fu/fy for the available ultimate strain of a reinforced concrete bar in tension,
b) Influence of stud and reinforcement arrangement in the hogging moment zone of the slab [11].

Ulrike Kuhlmann and Lars Rölle

The reinforcement ratio is influencing the moment resistance and the deformation/rotation capacity of the
joint. By increasing the amount of reinforcement the deformation capacity is also increased. The reason is
that the reinforcement ratio is significantly influencing the steel stress Vsr1 of the rebar when a first crack
has formed. The ratio of Vsr1/fyk is important for the available plastic strain Hsmu of the reinforced concrete
slab under tension, where fyk is the yield strength of the rebars. For high deformation capacities of the slab
a higher reinforcement ratio is advantageous because a rising ratio causes steel stresses Vsr1 when a first
crack has formed which are well below the yield strength.
Beside the reinforcement ratio and class the arrangement of the shear studs in the hogging moment
region are influencing the deformation capacity of the slab. More precisely the distance of the first shear
stud to the column profile is decisive for the available expansion length of the rebars. By increasing the
distance of the first stud the length for activating plastic strain in the reinforcement is clearly increased
resulting in increase of deformation capacity. It is pointed out that also a discontinuous amount of
reinforcement within the “tension bar” in the joint region should be provided to profit from the modified
stud arrangement, see Figure 3b.
4.4 Over-strength effects
According to the basic design criteria (ULS + SLS) members and joints are designed assuming
nominal material values. This is justified by the present safety concept. However for large displacement
analysis considering only nominal values may lead to results which are non-conservative.
So aside of the plastic behavior of the material and the stability sensitivity of the sections which
dominate the ductility of the members the joint behavior is decisive. Composed of various components
the aim should be that only ductile components control the overall joint behavior. For this not only the
component behavior itself is of importance but the interplay of the various components considering also
possible over-strength effects play an important role.
Figure 4 gives the example of a joint composed of a ductile and a brittle component, e.g. the endplate
in bending acting together with bolts which usually fail in a brittle manner. The design according to the
nominal values of strength leads to a moment rotation curve of the joint also acting ductile, see case a).
However the actual values of strength may exceed the nominal values (over-strength effects) so that no
longer the ductile component dominates the failure load, but the brittle one, see case b). As a consequence
the overall behavior of the joint shows a very limited rotation capacity. Disregarding over-strength effects
the joint may lead to only limited ductility as shown and as consequence no redistribution of forces can
take place that means the structure has only reduced redundancy.

Figure 4: Influence of over-strength effects on the rotation capacity


5.1 General
The European RFCS research project Robustness [8] recently finished has carried out extensive
experimental and theoretical investigations on the behavior of steel-composite joints under biaxial
loading, especially concerning the joint ductility to create robust structures which are able for load
redistribution under exceptional loading and are insensitive to progressive collapse. It has demonstrated
that the former concept to strengthen the joints in order to achieve that the plastic hinges appear in the
beams is not a necessary condition for activation of catenary action in a frame structure for the design
strategy of alternate load path method, but that it is also possible to place the plastic hinges into the joints

Ulrike Kuhlmann and Lars Rölle

by designing partial-strength joints with sufficient ductility. Within a national research project [9]
parameters influencing the ductility of bolted beam-to-column connections were investigated.
Furthermore the influence of over-strength effects on the resistance and rotation capacity of the joint was
analyzed. In a current diploma thesis [7] a composite structure is analyzed for the event column loss.
5.2 Experimental investigations
Within the two mentioned research projects the joint deformability and ductility as well as the
combined bending and tensile resistance have been investigated.
The performed steel joint tests mainly aimed at the investigations of increasing the joint ductility by
varying different parameters. The main parameters influencing the deformability in the tension zone of
the joint are the ratio of the endplate thickness and the bolt diameter (under consideration of the individu-
al material strength) and the arrangement of the bolts depending on the distance to the web, see Table 1.
By decreasing the ratio of the bolt diameter and the endplate thickness the rotation capacity is in-
creased. By modifying the bolt arrangement particular by increasing the distance of the bolts to the beam
web and beam flange the rotation capacity is also increased. A test series in [9] examined the influence of
the steel strength and the simultaneous activation of the components endplate and column flange in
bending. The resulting moment-rotation curves are given in Figure 5a. By reducing the steel grade of the
endplate and the column flange the rotation capacity is also increased accompanied by only small
decrease of the resistance as, see e.g. test curves Z6 and Z3 in Figure 5a.

a) b)
Figure 5: a) Influence of the steel grade to the rotation capacity and b) measured M-N-interaction of the
composite joint tests

The objective of the composite joint tests was the determination of the simultaneous moment-tensile-
resistance within the joint. The tests simulated the loading procedure from pure bending state to a mixed
bending and tensile state up to a pure tensile state at the end. The tests were successfully following the
whole theoretical M-N-curve (as shown in Figure 5b). The failure of the joints always occurred under
mainly pure tensile exposure. From the results of the composite joint tests under combined bending and
tension exposure it can be concluded that having a highly ductile joint behavior due to well-advised
adjustment of the single components the transition from pure bending state up to a membrane state in the
joint is feasible. The design of the joint specimens considered already over-strength effects and the bolts
were intentionally oversized to exclude premature brittle failure of the connection. The results of the
joints tests have been also con-firmed by a substructure tests executed by the project partner ULg (Liége,
Belgium). Within this substructure test the activation of catenary action, after the event column loss
happened, was possible due to the highly ductile performance of the joints [8]. Failure was mainly
induced by the concrete slab: for the hogging moment joints by increased cracks and final rupture of the
reinforcement, for the sagging moment joints by crushing of the concrete. In addition a remarkable
residual resistance and ductility remained when the concrete slab had already failed.
5.3 Numerical investigations
The numerical simulations were executed by the Finite Element software ANSYS [1]. First
recalculations were made to verify the FE-Model at the tests results and afterwards parametrical studies

Ulrike Kuhlmann and Lars Rölle

followed to extend the range of parameters as well as investigate the actual influence of material over-
strength effects on the joint behavior. The influence of the material properties on the joint behavior
(ductility and bearing capacity) depending on stochastic distribution of the material strength was
investigated in a first step by considering various combinations of characteristic, see Figure 6a-c [9].
Numerical simulations were used due to the fact that the local membrane effect in the T-stub of the
components “endplate in bending” or “column flange in bending” is not yet implemented in the analytical
approach of the component method acc. to EN 1993-1-8 [1][4].

a) b) c)
Figure 6: Deformed shape of steel joint considering a) nominal material values and b) over-strength
effects and c) Moment–rotation curves for varied material strengths

In terms of resistance the over-strength effects usually cause an additional material reserve which can
be activated in the case of progressive collapse analyses. But considering connections where different
types of steel grade are assembled the over-strength effects may result in unintentional negative effects.
Particular limited ductility is the main phenomena as the distribution of the available joint rotation
depending on the statistical spread of the material properties in the diagram of Figure 6 shows.

Figure 7: Simulation of column loss in a composite frame

Within a current diploma thesis numerical simulations on global level at steel and composite
structures have been performed to analyze the collapse resistance of the structure as well as requirements
for the implemented partial-strength joint configurations. Another aspect is also to determine the
additional positive contribution of the continuous RC slab in two horizontal directions, see Figure 7. [7]
First results showed that such a composite beam-column structure is able to resist the event of a
column loss under the accidental load combination for about 70-80% utilization of ULS loading. The
identified requirements for the partial-strength joints concerning ductility and M-N-resistance are also
feasible and within the range of the available rotation capacity and strength determined by the
experimental investigation.

Ulrike Kuhlmann and Lars Rölle


As terrorist attacks become more and more frequent the demand of building safety has been raised.
Under such exceptional loading situations, the ability of a structure to survive largely depends on the
performance of key structural elements and their connections, preventing progressive collapse. But until
now, aside of some theoretical concepts there are only very few and/or insufficient recommendations in
the codes. Whereas heavy reinforced concrete buildings are generally regarded as safe, light steel framed
structures have to provide evidence of a sufficient robustness against impact or blast in order to be
However, former and ongoing research projects have shown that intelligent robust design concepts
such as the alternate load path method achieved by ductile joints solutions lead to advantages of steel and
composite structures. In comparison to RC structures, steel and composite structures combine the
characteristics high strength, ductility, great plastic reserves, high residual strength and energy
dissipation. Furthermore the own research activity showed that obviously intelligent and highly ductile
joint design is increasing the robustness of the structure. So by only small additional effort in joint de-sign
additional resistance for exceptional loadings such as column loss may be activated.
Part of the work presented here is carried out, as a joint research project by different European
partners, with a financial grant from the Research Fund for Coal and Steel (RFCS) of the European
Community. The authors like to thank their partners for co-operation and gratefully acknowledge the
financial support, also from the German funding of AiF.

[1] ANSYS User’s manual. ANSYS Mechanical Solutions Release 11.0. 2007.
[2] EN 1991-1-1: Eurocode 1 – Actions on structures – Part 1-7: General Actions - Accidental actions
due to impact and explosions, CEN, 2002
[3] EN 1992-1-1 Eurocode 2: Design of concrete structures - Part 1-1: General rules and rules for
buildings. CEN, 2005
[4] EN 1993-1-8: Eurocode 3. Design of Steel Structures – Part 1-8: Design of Joints, CEN, 2005
[5] GSA 2003: US General Services Administration: Progressive Collapse Analysis and Design
Guidelines for New Federal Office Buildings and Major Modernization Projects, GSA. 2003
[6] JCSS: Probabilistic Model Code: Part III Resistance Models – Steel. Joint Committee on Structural
Safety, 2001.
[7] Kleiner, A. Untersuchung der Robustheit und Kollapsresistenz von Stahl- und
Verbundrahmentragwerken – Numerische Simulation eines Stützenausfalls. Universität Stuttgart,
Mitteilung des Instituts für Konstruktion und Entwurf Nr. 2009-43X, 2009
[8] Kuhlmann, U. et al. (2008): Robust structures by joint ductility. Publishable Report. Contract-No.
RFS-CR-04046, 2008.
[9] Kuhlmann, U., Rölle, L. (2008): Duktilitätskriterien für typisierte Stirnplattenverbindungen,
Schlussbericht. DASt-Forschungsvorhaben im Auftrag der AiF. AiF-Vorhaben Nr. 14627,
Dezember 2008.
[10] Rölle, L.; Kuhlmann, U. (2009): Alternate load path method for robust design by ductile steel and
composite joints. Proceedings of IABSE Symposium Bangkok. September 2009
[11] Schäfer, M. (2005): Zum Rotationsnachweis teiltragfähiger Verbundknoten in verschieblichen
Verbundrahmen. Dissertation. Universität Stuttgart, Mitteilung des Instituts für Konstruktion und
Entwurf Nr. 2005-1, 2005.
[12] Starossek, U., Wolff, M.: Progressive Collapse - Design Strategies. Proceedings of the IABSE
Symposium Lisbon, 2005.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



József Szalai*

* ConSteel Solutions Ltd., Hungary

e-mail: szalaija@gmail.com

Keywords: stability design, eigenvalue analysis, buckling, slenderness.

Abstract. Stability analysis and design has always played a key role in the process of verification of steel
structures. The possible analysis methods and design procedures have long history with a plentiful
literature providing various proposals for the engineers. This paper concentrates on the use of different
types of eigenvalue analysis as a simple and powerful tool for stability design. Nowadays almost all the
engineering software products have some kind of eigenvalue analysis options so these tools are easily
available for the practicing engineers providing them a deeper look on the structural behavior. Various
types of application possibilities are reviewed and new methods are proposed supporting the most up-to-
date standard procedures of different levels from the isolated member design to the partial or global
structural stability design. The suitable theoretical (both mathematical and mechanical) background is
developed and the numerical procedure is implemented. The technique is applicable for a wide range of
structural types and stability problems making the automatic effective length calculation possible in
general without the use of any iterative process or tabulated values for certain cases. An application
example is presented showing the comprehensiveness of the methods, and special efficiency indicators
are presented in order to supply information about the adequacy of the applied design method.

There are two available methods for stability design of steel structures provided by the EN 1993-1-1
ƒ isolated member approach: Sections 6.3.1, 6.3.2, 6.3.3
ƒ general method: Section 6.3.4
The first one is valid for uniform members only and based on structural member isolation and
buckling mode separation. The main difficulties of this method arise from these two simplifications, the
member isolation is usually handled by applying suitable effective length factors – considering the
appropriate support and restraint conditions – while the mode separation is solved by special interaction
factors creating the connection between the pure loading and buckling cases. There are a great number of
papers on both topics including some theoretical investigations about the mechanical basics [2, 3] and
several proposals for the practical application [4, 5]. The EN 1993-1-1 regularizes only the calculation of
interaction factors the problems coming up from the member isolation are not dealt with in the standard.
In the general method these two simplifications are eliminated by examining a complete structural part
and calculating only one slenderness belonging to the real, compound loading and buckling situation.
Although in the recent version of EN 1993-1-1 there are several restrictions on the application field of
this method however on the other hand there are heavy research and development efforts on extending its
applicability [6, 7] and this method is expected to cover much larger area of practical problems then the
isolated member based conventional procedures. It is also important to note that in case of the general
method the calculation of the generalized structural slenderness requires more complicated analysis which
can usually provided only by application of some software package. It seems evident is that in both

József Szalai

methods the key question is the determination of the slenderness values and especially the calculation of
the appropriate elastic critical values (critical forces or critical load levels). This aspect is very poorly
treated in the structural standards and accordingly the practicing engineer is fully responsible for it.
Moreover the calculated elastic critical values are usually very important and have a significant influence
on the final result of stability design.
In this paper the possibilities of eigenvalue analysis for the calculation of elastic critical values are
examined from the point of view of standard stability design according to the recent version of EN 1993-
1-1. Different application methods are introduced adapted to the different design approaches, and special
indicator factors are developed highlighting the relevance of the used type of eigenvalue analysis. An
application example is presented to show the practical working of the different methods.


For a usual steel structure composed of beam-column elements the general loss of elastic stability can
be quite accurately described by bifurcation analysis. In a standard finite element environment this
problem can be formulated as a linear eigenvalue analysis with the following basic form:
K E  DKG U 0 (1)

where KE is the elastic stiffness matrix, KG is the second order geometric stiffness matrix, D is the
eigenvalue and U is the corresponding eigenvector. In the mechanical interpretation the eigenvalue
denotes the elastic critical load level and the eigenvector shows the eigenshape (eigenmode) or buckling
shape (buckling mode). It is important to note that the mechanical meaning and accuracy of the calculated
eigenmodes and eigenvalues highly depend on the definition of stiffness matrices. In this paper special
decomposition techniques are applied for the compiled geometric stiffness matrix – while the elastic
stiffness matrix is always formed on the complete structural model – to suitably calculate the elastic
critical load levels necessary for the certain design approaches. Upon these techniques the following
eigenvalue analysis are proposed:

ƒ Complete Eigenvalue Analysis (CEA) – the geometric stiffness matrix is compiled on the whole
structural model
ƒ Partial Eigenvalue Analysis (PEA) – the geometric stiffness matrix is compiled only on a
separated part of the structural model
ƒ Selected Eigenmode Analysis (SEA) – the geometric stiffness matrix is compiled only on
selected displacement degree-of-freedoms

The CEA is the mostly known and commonly used analysis technique, the resulted critical load levels
and corresponding buckling modes apply to the whole global structural model – even it is apparently
restricted to a part of it – and consider the compound loading case. In the further text we write the CEA in
the following form:

K E  D cr

K GC U cr
0 (2)

The CEA method is applicable for the critical load level calculation for the general stability design
If a part of the complete model is examined and intended to design for stability the PEA can be used.
In this method the structural model is divided into two parts: (P1) a relevant part and (P2) a remaining
part. Accordingly the complete geometric stiffness matrix can be decomposed:

K GC K GP1  K GP 2 (3)

One can obtain an eigenvalue solution for the relevant part by solving the following equation:

József Szalai

 D crP1 K GP1 U crP1 0 (4)

In a mechanical interpretation this calculation yields special buckling shapes which are induced by the
internal forces acting only in the members which are part of the relevant substructure while the initial
stiffness of the whole structure is considered (as a restraint condition for the examined substructure). In
that sense this solution is similar to the ones applying fictitious springs at the joining parts of the relevant
substructure modeling the restraints from the remainder of the whole structure. The PEA method also
provides global type solution for the real, combined buckling situation so it is also applicable for the
general stability design.
The third proposed method (SEA) is directly developed for the isolated member approach since it is
able to calculate the separated buckling modes. In this case the geometric stiffness matrix is compiled in
such a way that the rows and columns associated with the displacement degree-of-freedom necessary for
the relevant buckling mode are considered and the rest of the matrix is neglected. For example if the
subject of the analysis is the lateral buckling about the minor axis (axis ‘z’ according to EN 1993-1-1)
then those terms are kept only in the geometric stiffness which contain the second order compression
effect for the selected displacements (uy and Iz). Further if this buckling mode is required only on an
isolated member then this reduced compilation is done only on this element. Consequently the second
order effects of the system are concentrated so as to be able to experience only the relevant buckling
mode. Considering however the complete first order elastic stiffness matrix of the whole structure the
appropriate restraints (and accordingly the necessary effective lengths) of the isolated member can be
calculated quite accurately. The SEA method for the selected i-th buckling mode (for instance Ncr,z) is
written as follows:

 D crNcr,z K GNcr,z U crNcr,z 0 (5)


After introducing the proposed possibilities for the determination of elastic critical load levels by
applying certain eigenvalue analysis methods the next important issue is the selection of the most
appropriate method for the current structural problem examined. It is very important to detect the most
relevant buckling modes of the structure and the associated most proper design method. The introduced
eigenvalue analysis methods can yield various results and solution possibilities but it is the decision of
the engineer which method and buckling mode is the most relevant for the problem. In order to help this
decision special indicator factors are developed showing the relevance of the calculated buckling modes
(eigenmode relevance indicator – ERI). The ERIs are formulated on energy base, this approach was
introduced to the interpretation of stability calculations in [8], however, for different purposes; in this
paper the ERIs are developed so as to supply appropriate information about the described issues. The
basic formula of all the possible ERIs is the internal (and the corresponding external) energy induced by
the i-th eigenmode; this can be written using Eq. (2):
1 T 1
Ei U cr ,i K E U cr ,i D cr ,i U cr
,i K G U cr ,i (6)
2 2
The next sections present the ERIs in case of the different eigenvalue analysis methods based on Eq. (6).
3.1 CEA
When analyzing the complete structure for stability the following questions may arise regarding the
obtained eigenmodes:

(1) for a certain eigenmode what are the relevant members (relevant model portion)?
(2) for certain members (certain model portion) which is the most relevant eigenmode?

József Szalai

For the problem (1) let the energy be calculated for members or model portions (k denotes the relevant
member or model portion):
1 C k C
Eik U cr ,i K E U cr ,i (7)
In this case the elastic stiffness of the relevant model part is considered only. Obviously the sum of all the
energy values of model parts gives the total energy of the complete model, i.e. (having m number of
model parts):
Ei ¦ Eik (8)
k 1

Accordingly an ERI can be constructed so as to show the relevance of the separated model portions
considering the i-th eigenmode as a percentage:
C k C
Eik U cr ,i K E U cr ,i
ERI1ik 100 100 C C
Ei U cr ,i K E U cr ,i

Problem (2) is more complicated but also more significant, since in the case of a complex structural
model it is usual, that different eigenmodes describe the buckling behavior of distinct parts of the model.
For that reason a scaling procedure is necessary in order to select the appropriate eigenmode for the
stability design. To develop a proper scaling factor let us examine the basis of the stability design
approach of the EN 1993-1-1 which is the buckling curve based reduction factor. The mechanical model
for the buckling reduction factors is the Ayrton-Perry formula. In this model the failure is associated with
the load level at which the second order maximum elastic stress of the geometrically imperfect member
reaches the yield stress. Consequently the reduction factor depends mainly on the amplified imperfection
which has usually a shape equal to an appropriate eigenmode or a combination of them. The scaling
factor should therefore consider this effect to show the importance of the eigenmodes in accordance with
the mechanics of the buckling reduction factors. Firstly the basic amplitudes for the eigenmodes are
determined by normalizing using the elastic stiffness matrix:

1 N U crN , j K GC U crN , j D cr

Ei U cr ,i K E U crN ,i (10)
2 U crN ,i K GC U crN ,i D cr

As a result the greater the critical factor the less the amplitude of the eigenmode is, this is realistic when
considering the eigenmodes as geometric imperfection. It is known [6] that the geometrical imperfections
having the shape of an eigenmode cause the following additional amplified second order displacements:
U add U imp (11)
D cr  1

If the normalized eigenmodes of Eq. (10) are considered as imperfections in Eq. (11) then a further
scaling factor can be created by calculating the energy of this amplified imperfection in the k-th model
1 §¨ 1 · § 1 ·
E ik U crN ,i ¸ K Ek ¨ U crN ,i ¸ (12)
2 © D cr ,i  1 ¸ ¨ ¸
¹ © D cr ,i  1 ¹

József Szalai

Finally if it is considered that enough number (m) of eigenmodes is calculated (i.e. the last eigenvalue –
what is the highest elastic critical load level – is sufficiently high) then the following ERI can be
constructed showing the relative significance of the i-th eigenmode for the k-th model portion:
§ 1 · N k N
¨ ¸ U K U
Eik ¨ D  1 ¸ cr ,i E cr ,i
ERI2ik 100 100 © cr ,i ¹ (13)
m 2
m § · N k N
¦ Eik ¦ ¨ D  1 ¸ U cr,i K E U cr,i
¨ 1 ¸
i 1 © cr ,i ¹
i 1

3.2 PEA and SEA

In this case the determination of the relevant model portion is not examined since the preliminary
selection of model part (or isolated member) aims at concentrating the buckling mode to this part. What is
important however that how accurate the calculated partial buckling mode is compared to the possible
complete modes. Let us write the eigenvalue equation for the complete structure (Eq. (2)) using the
partial eigenmode and eigenvalue yielded by the solution of Eq. (4):

K E  D cr
P1 C
,i K G U cr ,i K E
 D crP1,i K GP1 U cr
,i  D cr ,i K G U cr ,i
P1 P 2 P1
D cr
P1 P 2 P1
,i K G U cr ,i (14)

Eq. (14) expresses an error force vector generated by the partial buckling mode on the remaining model
portion (P2). Obviously if this term is significant then the partial buckling mode is possibly irrelevant
which shows that the current model portion has no distinct buckling problem. On the other hand if this
force approaches zero then the buckling mode is dominant and accurate for the selected model portion.
To develop a straightforward indicator we use again an energy format for Eq. (14):

§ U P1 K P 2 U P1 · U crP1,i K GP1U crP1,i

ERI3iP1 100¨ 1  crP,1i GC Pcr1,i ¸ 100 (15)
¨ U cr ,i K G U cr ,i ¸ U crP1,i K GC U crP1,i
© ¹

In the following example the CEA method is used and the ERI1 and ERI2 factors are examined.




Figure 1: Model of the example (made in ConSteel [10]).

József Szalai

In Fig. 1 the structural model is illustrated, which consists of two columns (C1 and C2) of HEA200
and a beam (B1) of HEA220 loaded by a 30 kN/m line load. According to the support condition four
cases are evaluated (the beam is always simply supported) summarized by Table 1.

Table 1: Evaluated support conditions.

C1 C2
Case1 Pinned Pinned
Case2 Fixed Pinned
Case3 Pinned Fixed
Case4 Fixed Fixed

All the eigenvalue analysis are calculated by ConSteel software [9], [10] using a 7 degree-of-freedom
finite element model. Figs. 2-5 show the calculated eigenmodes and elastic critical factors for Case1-
Case4 – all buckling modes form some kind of out-of plane buckling –, and Tables 2-5 contain the ERI1
and ERI2 factors evaluated for the beam and the two columns (for ERI1 the columns and for ERI2 the
rows give the 100% value).

Dcr1 = 3,43 Dcr2 = 8,03 Dcr3 = 13,39 Dcr4 = 26,27

Figure 2: Case1 – first four eigenmode and eigenvalue.

Table 2: Case1 – ERI1 and ERI2 factors

Dcr1 Dcr2 Dcr3 Dcr4 Dcr1 Dcr2 Dcr3 Dcr4
B1 99,6 2,4 95,4 7,6 96,1 0,3 3,5 0,1
C1 0,1 97,5 4,4 12,9 0,5 97,1 1,4 1,0
C2 0,3 0,1 0,2 79,5 26,9 0,4 0,6 72,1

Dcr1 = 4,18 Dcr2 = 13,29 Dcr3 = 17,1 Dcr4 = 26,34

Figure 3: Case2 – first four eigenmode and eigenvalue.

Table 3: Case2 – ERI1 and ERI2 factors

Dcr1 Dcr2 Dcr3 Dcr4 Dcr1 Dcr2 Dcr3 Dcr4
B1 73,1 91,7 10,1 5,2 91,7 7,7 0,5 0,1
C1 26,7 8,2 89,6 0,2 86,8 1,8 11,4 0,0
C2 0,2 0,1 0,3 94,6 13,9 0,4 0,5 85,2

József Szalai

Dcr1 = 6,51 Dcr2 = 8,04 Dcr3 = 13,60 Dcr4 = 26,93

Figure 4: Case3 – first four eigenmode and eigenvalue.

Table 4: Case3 – ERI1 and ERI2 factors

Dcr1 Dcr2 Dcr3 Dcr4 Dcr1 Dcr2 Dcr3 Dcr4
B1 54,1 2,4 91,5 9,4 73,6 2,0 23,8 0,6
C1 0,0 97,5 4,5 90,4 0,1 92,3 1,3 6,3
C2 45,9 0,1 4,0 0,2 98,3 0,0 1,6 0,1

Dcr1 = 6,82 Dcr2 = 13,63 Dcr3 = 17,28 Dcr4 = 34,04

Figure 5: Case4 – first four eigenmode and eigenvalue.

Table 5: Case4 – ERI1 and ERI2 factors

Dcr1 Dcr2 Dcr3 Dcr4 Dcr1 Dcr2 Dcr3 Dcr4
B1 43,3 89,0 8,1 76,0 66,0 28,8 1,6 3,6
C1 20,0 5,4 89,9 19,8 60,1 3,4 34,6 1,9
C2 36,7 5,6 1,9 4,2 95,9 3,1 0,7 0,3

It can be seen that the first and dominant buckling mode is always the lateral-torsional buckling of the
B1 beam coupled with some form of flexural buckling of the columns. The corresponding critical load
factors increases from Case1 to Case4 denoting the significance of the applied additional restraints for the
supports of the columns however from the ERI factors it becomes clear that the contribution of the
columns to the certain buckling modes is significantly different. From the ERI1 values it can be
concluded that in the first buckling mode the B1 beam is always dominant and for the other modes the
dominant member is also highlighted in the tables. Looking at however the ERI2 values the followings
can be determined for a certain member:
ƒ which is the most relevant buckling mode;
ƒ which is the appropriate critical load factor for the calculation of member slenderness in the
stability design.
In Case1 the ERI2 values show the strong dominance of the member B1 in the first mode, C1 in the
second mode and C2 in the fourth mode. This is a consequence of the pure shape of the buckling modes,
since the columns have pinned supports. In Case2 the fix support of the column C1 has a considerable
restraining effect on the beam increasing the first critical load factor accordingly in this case the first
mode is more dominant for the column C1 than the third one which show an isolated buckling mode for
this member. The column C2 is not really effected by the additional restraint, so the dominant mode

József Szalai

remains the fourth one with almost the same critical load factor value. In Case3 the situation is quite the
same, the column C1 has the dominant isolated buckling mode with same critical load factor as in Case1,
and for column C2 the first mode is the most relevant. In Case4 from the ERI1 factors it is clear that the
complete model contributes to the first buckling mode and the ERI2 values are explicitly show that this
mode is the most relevant for all the members. Naturally the meaning of the ERI2 values can be refined
by increasing the number of calculated eigenmodes which is usually necessary in case of larger structural
models. This simple and straightforward example is intended to show the mechanical meaning of the
different indicator factors.

One of the most important issues in stability design which is out of the field of standard regulations is
the calculation of elastic critical forces or load levels for the determination of slenderness values. In this
paper several methods are presented for this problem using the eigenvalue analysis based approaches. For
different structural arrangements different types of buckling modes can be dominant and moreover the
modern structural standards provide several different possibilities for the stability design. To yield
appropriate slenderness for the different problems and design methods three approaches are proposed: the
complete eigenvalue analysis, the partial eigenvalue analysis and the selected eigenmode analysis. Further
supporting the selection of the relevant approach and eigenmode special indicator factors are developed.
In an application example the practical working with the proposed eigenvalue analysis approaches and
indicator factors are presented.

[1] European Standard, EuroCode 3. Design of Steel Structures – Part1-1: General rules and rules for
buildings, EN 1993-1-1, 2005.
[2] Boissonade, N., Jaspart, J.P., Muzeau, J.P., Villette, M., “Improvement of the interaction formulae
for beam columns in Eurocode 3”. Computers and Structures, 80, 2375-2385, 2002.
[3] Greiner, R., Lindner, J. “Interaction formulae for members subjected to bending and axial
compression in EUROCODE 3—the Method 2 approach”, Journal of Constructional Steel
Research, 62, 757-770, 2006.
[4] Serna, M.A., López, A., Puente, I., Yong, D.J. “Equvivalent uniform moment factors for lateral-
torsional buckling of steel members”, Journal of Constructional Steel Research, 62, 566-580, 2006.
[5] Dong-Ho C., Hoon Y., “Iterative system buckling analysis, considering a fictitious axial force to
determine effective length factors for multi-story frames”. Engineering Structures, 30, 560-570,
[6] Szalai, J., Papp, F. “On the theoretical background of the generalization of Ayrton-Perry type
resistance formulas”. Journal of Constructional Steel Research, 66(5), 670-679, 2010.
[7] Sedlacek, G., Müller, C. “Zur Vereinheitlichung der Stabilitatsregeln im Eurocode 3“. Stahlbau,
73, 733-744, 2004.
[8] Perelmuter, A.V., Slivker, V.I. “The problem of interpretations of the stability analysis results”,
European Conference of Computational Mechanics, Cracow, Poland, 26-29 June, 2001.
[9] Papp, F., Iványi, M. “ConSteel as the prototype of a CAD/CAM oriented program for concurrent
design of beam-column structrures”. Proc. 5th International Conference on Computational Steel
Structures Technology, Leuven, Belgium, 2000.
[10] Website: www.consteel.hu

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Rachel B. Cruise* and Anna M. Paradowska**

* The School of Architecture, The University of Sheffield, Sheffield, S10 1FL.

e-mail: R.Cruise@sheffield.ac.uk
** ISIS Facility, Rutherford Appleton Laboratory, Science and Technology Facility Council, Didcot, UK,
OX11 0QX.
e-mail: anna.paradowska@stfc.ac.uk

Keywords: Stainless steel structures, Residual stress, Roll forming, Neutron diffraction.

Abstract. During the manufacture of roll formed structural members the production and storage of sheet
materials, as well as their subsequent forming causes plastic deformation in varying degrees around the
resulting cross section. Plastic deformation causes both an increase in material strength in the section
material through cold working and it also affects the residual or internal stress distribution present
throughout the resulting structural section. Both the material strength and the residual stress distribution
influence the structural behavior of the cross section, therefore it is important to map both these
properties in order to achieve efficient structural design. Destructive techniques have commonly been
used to map residual stresses in structural sections. To achieve a high resolution of measurements these
techniques are extremely labor intensive and sensitive to the measurement technique and it is almost
impossible to measure the strain relaxation that occurs in three orthogonal components by this process.
Non-destructive residual stress measurements are relatively infrequently used for structural engineering
applications. The presented experimental program demonstrates the applicability of the non-destructive
technique of neutron diffraction for mapping residual stresses in structural members at four locations
through the thickness of a roll formed stainless steel section. The measurements were made using the
ENGIN-X instrument at the UK’s pulsed neutron source: ISIS in Oxford.

Cold formed structural sections are a comparatively novel type of structural section that started to be
more widely used for construction in the 1940s. They now comprise 15% and 13% of all new structures
in the housing market in the USA and Australia respectively [1] and expansion in the UK markets is
being actively encouraged for both environmental and economic reasons [2]. Since cold formed sections
are produced by plastically deforming metal sheets at room temperature the thickness of the sections can
be less than the minimum thickness required to retain the high temperatures essential to manufacture the
more conventional hot rolled sections. Hence cold formed sections which are used to carry light loads and
span short distances can be lighter and structurally more efficient than the hot rolled alternatives.
Roll formed sections are the most prevalent type of cold formed section. During the manufacture of
roll formed sections plastic deformation can occur at several stages and plastic deformation will have
been experienced to varying degrees around the resulting cross section. Through a process termed cold
working the regions of the section that have experienced plastic deformation exhibit an increase in
material strength and a decrease in ductility [3]-[4]. The resulting distribution of material strength around
roll formed sections has been mapped and used in structural design codes of both carbon steel and
stainless steel roll formed sections to increase the material efficiency [5]-[6]. However the plastic
deformation which causes the increase in material strength also influences the distribution of residual or

Rachel B. Cruise and Anna M. Paradowska

internal stresses. Residual stresses are stresses that exist within a structural member in its unloaded state
and their magnitude and distribution can affect the structural behaviour of structural cross sections.
Since hot rolled sections have been used in the construction industry for a much longer period of time
than roll formed sections there is a deeper understanding of their structural behaviour. Furthermore the
techniques for measuring factors that can influence structural behaviour, such as residual stresses, have
been developed to capture the significant aspects of the magnitudes and distributions observed in hot
rolled sections. However, these techniques have been shown to give an incomplete picture of the
magnitude and distribution of residual stresses in roll formed sections [7]-[9]. This paper will therefore
present data from a pilot study that used a non-destructive technique, namely neutron diffraction, to
measure residual stress distributions in a roll formed stainless steel box section to demonstrate the
potential of adopting this technique.


Whilst carbon steel is the most commonly used structural metal in the construction industry one
disadvantage of carbon steel is its potential to corrode. It has been only in relatively recent times that the
use of a non corrosive alternative such as stainless steel as a structural material has been explored. This is
principally because stainless steel as a material is more expensive than carbon steel. Despite its initial
expense stainless steel has been adopted for applications in exposed conditions such as bridges and
offshore structures because the cost saving associated with its ease of maintenance can outweigh the high
initial cost [10]. Due to the expense of stainless steel and the efficiency of the roll forming process most
structural sections currently available are roll formed sections.
There are three different microstructures of stainless steel: austenitic, ferritic and martensitic. The
most commonly used grade of stainless steel for structural applications is 1.4301 which has an austenitic
microstructure. Cold working of austenitic stainless steel causes a significantly larger increase in material
strength than in carbon steel. This offers a relatively larger increase in design efficiency which has clear
benefits to realizing stainless steel as a competitive structural material. However the co-existing residual
stresses can, depending on their magnitude and distribution and on the loading condition of the roll
formed section, have a negative effect on structural behavior by causing a loss of stiffness and early

Roll forming is a highly automated and therefore efficient production process. There are two types of
sheet material that are commonly used as the starting material for roll forming: hot and cold rolled sheet
material. Stainless steel can be rolled whilst hot to produce hot rolled sheet of a minimum thickness of
approximately 3mm (see stage 1 in figure 1). If thinner sheet material is required, since the sheet will be
too thin to retain the heat needed to allow for hot rolling to occur, the stainless steel is passed through
rollers whilst it is at room temperature, therefore plastically deforming the sheet to reduce its thickness.
This process produces cold rolled sheet material.
For reasons of efficient storage and to enable the sheet material to be used as the starting material of
this completely automated section forming route both hot and cold rolled sheet material are wound into
coils as shown in stage 2 of figure 1. To manufacture roll formed sections the coil material is unwound
(see stage 3 in figure 1) and then fed into shaping rollers which plastically deform the sheet material into
the required cross sectional shape. To roll form a box section the sheet is rolled into a circular cross
section, welded closed and then this tube is crushed into a rectangular cross section as shown in stages 4-
6 in figure 1.
Roll formed structural sections can therefore experience plastic deformation at three stages in their
manufacture. Firstly plastic deformation can be experienced in cold rolled sheet production, secondly
during the coiling and uncoiling of the sheet material and finally during the forming of the cross section.

Rachel B. Cruise and Anna M. Paradowska

1: Sheet 2: Sheet 3: Sheet 4: Forming a 5: Seam 6: Crushing into

production coiling uncoiling circular tube welding a box section

Figure 1: Manufacture of a roll formed box section.

In general cold working, or the increase of material strength through plastic deformation, can be
explained by considering the effect of plastic deformation on the ordered arrangement of atoms in a
metallic lattice. Plastic deformation can be described on the atomic scale as the movement of planes of
atoms with respect to one another in the metallic lattice. In carbon steel and stainless steel this causes an
increase in dislocations in the metallic lattice. The creation of more dislocations in the metallic lattice
increases the number of obstacles to planes of atoms moving. Therefore the cold worked material is
observed on the macro scale to increase its resistance to further plastic deformation and so exhibit an
increase in material strength [11].
For stainless steel with an austenitic microstructure an increase in dislocations is not the only
mechanism that can increase the material strength of the cold worked material. The arrangement of atoms
in an austenitic microstructure, prior to experiencing cold working, can be described by the unit cell
shown in figure 2a. When this unit cell is duplicated and stacked together the arrangement of atoms in the
metallic lattice of an austenitic microstructure is described. This particular unit cell is termed a Face
Centred Cubic (FCC) unit cell.

Top face: Top face:

Mid section: 2.8r Mid section: 2.3r

Base face: Base face:

a) Face Centred Cubic (FCC) unit cell of b) Body Centred Cubic (BCC) unit cell
the austenitic microstructure. of the martensitic microstructure.
Figure 2: Microstructures in cold worked austenitic stainless steel.

The austenitic microstructure of stainless steel grade 1.4301 is a metastable microstructure which
means that work done to the material through plastic deformation will cause the austenitic microstructure
to, in part, transform into a martensitic microstructure. The unit cell of the martensitic microstructure is
shown in figure 2b and it is termed a Body Centred Cubic (BCC) unit cell. The BCC unit cell is smaller
and has a higher ratio of volume of atom to volume of unit cell compared to the FCC unit cell. This ratio
is commonly termed the Atomic Packing Factor (APF) and it is an indication of the density of the unit
cell. In addition, unlike the FCC unit cell, the BCC unit cell has within its geometry no planes where the
atoms are as tightly packed together as possible. This is of significance because owing to the geometry of
these close packed planes they can easily slide past one another and the lack of these in the BCC unit cell
causes the martensitic microstructure to give the cold worked stainless steel its increase in strength and
reduction in ductility [12]. The relationship between the two microstructures, and therefore the
transformation that occurs during plastic deformation, is shown in figure 3 by identifying the atoms that

Rachel B. Cruise and Anna M. Paradowska

will create the unit cell of the martensitic microstructure within the metallic lattice of the austenitic
Atoms that will form the x
martensitic unit cell

50 mm Weld
6 mm
microstructure 100 mm
Figure 3: Transformation from an austenitic Figure 4: Test sample.
microstructure to a martensitic microstructure.

Uneven plastic deformation also creates residual or internal stress distributions that equilibrate over
the whole cross section. Residual stresses are defined at three different scales by the distance over which
they equilibrate. Type I residual stresses relate to the macro scale, where equilibrium is achieved over
distances that relate to the scale of the structural cross section. It is this type of residual stress that is
considered to have the greatest importance for structural behavior. Type II and type III residual stresses
relate to the micro scale. Type II residual stresses are defined as equilibrating over several metallic grains
(regions where the metallic lattice is continuous) and type III residual stresses are defined as equilibrating
within metallic grains [13].
Residual stresses ıx, ıy and ız act in three orthogonal directions; normal to the surface of the section,
transverse to the section length and along the section length, respectively. This coordinate system is
defined in figure 4. Because of the influence that residual stresses can have on structural behavior it is
important to measure the magnitude and distribution of residual stresses in cold formed sections and there
are two distinct types of techniques which have been employed to date: destructive and non-destructive.


Destructive techniques used to measure residual stresses all involve mechanically removing material
from the test sample in order to disturb the equilibrium of the residual stress distribution, thereby causing
a geometrical relaxation. This geometrical change can be measured in order to quantify the released
stress. Owing to the size of sample required for material to be mechanically removed the destructive
techniques are commonly used by structural engineers because the measurements are made over the
macro scale and therefore result in determining type I residual stresses.
Type I longitudinal residual stresses, ız that exist along the length of a structural member are
considered to be the most significant in determining a member’s structural behaviour and they have been
commonly quantified by a destructive technique termed sectioning. This destructive technique cuts the
cross section into strips, thereby disturbing the equilibrium of residual stresses as illustrated in figure 5.
The strain caused by geometrical relaxation on the surface of each sectioned strip can be measured once
each strip has re-established equilibrium and used to identify two types of residual stress. Uniform tensile
or compressive strain is used to identify the longitudinal membrane residual stress, ız,m and the curvature
of the sectioned material indicates a variation of stresses through the material thickness, which is
commonly assumed to be linear [14] and which is used to quantify the longitudinal bending residual
stress, ız,b. However this measurement technique has disadvantages when used to measure residual
stresses in roll formed sections.

Rachel B. Cruise and Anna M. Paradowska
Membrane residual stress, ız,m +
x İz,b
- İz,b Ȝ
Bending residual stress, ız,b =
d ș
Combined residual stress, ız ız

Figure 5: Residual stresses released Figure 6: Neutron diffraction.

during sectioning.

Using the sectioning technique, combined with an electrolytic technique to remove layers of section
material, longitudinal residual stresses, ız were measured at different depths through the thickness of a
cold formed carbon steel box section in [7]. From these measurements it was observed that the
longitudinal residual stress distribution through the thickness of the section was not a linear variation as
commonly assumed [14] and therefore that the membrane and bending residual stresses could not be the
only residual stresses to exist in the section. Through this study [7] a third residual stress component
termed the layering residual stress was identified, which is not released and therefore not measured during
sectioning, since it has no resultant axial force or moment. This unmeasured layering residual stress is
important to quantify to determine peak residual stresses in the section material. Analytical models that
map the coiling, uncoiling and cold forming of stainless steel and carbon steel sheet material into
structural sections [8]-[9] have also determined that the variation of longitudinal residual stresses, ız
through the thickness of a cold formed section does not conform to the assumed linear model.
Longitudinal residual stresses, ız in roll formed stainless steel sections were determined through the
sectioning technique in [15] where both a linear and a rectangular block through thickness distribution
were assumed to calculate the longitudinal bending residual stresses, ız,b. It was observed that for
sectioned material with a rectangular cross section, there was a difference of two thirds in the magnitude
of the bending stresses between the two assumed distributions. This study showed that assuming a linear
through thickness residual stress distribution can cause large errors in determining the longitudinal
residual stresses.
Furthermore residual strains released normal to the surface of the section, İx and transverse to the
length of the section, İy are not easily quantified in the same location as the longitudinal strain, İz through
the use of the sectioning technique. However all strain components contribute to the normal, transverse
and longitudinal residual stresses (ıx, ıy and ız respectively) through the three dimensional definition of
Hookes’ Law, given in equations 1-3. Where E is the Young’s modulus and Ȟ is Possion’s ratio.

σx =
(1 +ν )(1 − 2ν )
[ε x (1 −ν ) +ν (ε y + ε z )] (1)

σy =
(1 +ν )(1 − 2ν )
[ε (1 −ν ) +ν (ε
y x ]
+εz) (2)

σz =
[ (
ε (1 − ν ) + ν ε x + ε y
(1 + ν )(1 − 2ν ) z
)] (3)

Rachel B. Cruise and Anna M. Paradowska

Using sectioning to measure residual stresses has the disadvantage that the complete residual stress
distribution is not fully released and therefore not measured and the strains in the normal, transverse and
longitudinal directions are hard to measure simultaneously to correctly determine the corresponding
residual stresses. Also the method of removing material can affect the residual stress pattern through
plastic deformation and heating that might occur during mechanical interventions.


The alternative way to measure residual stress distributions is with a non-destructive technique where
it is the effect of the test sample on magnetic fields, X-rays or a neutron beam which is used to determine
residual stresses and no material need be removed from the test sample. The use of magnetic techniques
to measure residual stresses is not possible in this case since the austenitic microstructure of stainless
steel is nonmagnetic. However X-ray diffraction techniques have been used to measure through thickness
residual stress distributions in an austenitic stainless steel roll formed section [16] but the depth of
penetration was not sufficient, so electrolytic material removal was used to obtain measurements at
greater depths. Problems were also experienced making measurements by X-ray diffraction due to the
large size of metal grains in cold worked stainless steel [16]. Whilst neutron diffraction does not offer
such a fine resolution as is possible using X-ray diffraction this technique does offer a suitable
penetration depth for roll formed cross sections. In addition the larger volume over which the
measurements are made, compared to X-ray diffraction could reduce the potential problems associated
with diffraction measurements made in a large grain microstructure.

Axonometric Plan view:
North The south x The north
collimator collimator y collimator
South collects collects
collimator neutrons that neutrons that
determine determine
atomic atomic
Neutron spacings in the spacings in
source Apertures normal the transverse
Sample direction, x. direction, y.
a) Sample positioned to measure atomic spacing in the normal and transverse direction.
Axonometric Plan view:
view: North
The south The north
collimator x collimator
South z collects
collimator collects
neutrons that neutrons that
determine determine
Sample atomic atomic
Neutron spacings in the spacings in
source Apertures normal the
direction, x. longitudinal
source direction, z.
b) Sample positioned to measure atomic spacing in the normal and longitudinal direction.
Figure 7: Test setup.

Rachel B. Cruise and Anna M. Paradowska

The non-destructive technique of neutron diffraction uses the interaction of a neutron beam and the
specimen’s atomic structure, as governed by Bragg’s law, to measure the spacing between atomic planes,
d. Bragg’s law is given in equation 4 and the variables Ȝ, d and ș, are defined in figure 6.
nλ = 2d sin θ (4)
When n in equation 4 is an integer the diffracted neutrons interfere constructively. This causes the
collimators either side of the test sample to detect a peak of neutrons at atomic spacings characteristic of
the arrangement of atomic planes in the microstructure under observation. Just such neutron diffraction
measurements were performed during a three day pilot study using the ENGIN-X instrument at ISIS.
Through thickness residual stress distributions were measured in four locations A-D around a roll formed
austenitic stainless steel grade 1.4301 box section of dimensions 100×50×6 mm, as shown in figure 4. At
each location, A-D, seven diffraction measurements were made over a 2×2×2mm3 gauge volume at
intervals of 0.5 mm through the thickness of the section. Measurements were made with the test sample
held in two orientations in order to measure atomic spacings in three orthogonal directions, as illustrated
in figure 7. The atomic spacings measured at different locations in the test sample were compared with a
stress free atomic spacing measured in 2×2×2 mm3 cubes, cut using an Electric Discharge Machine from
locations A-D in the same cross section. The atomic strains held in the roll formed cross section were thus
determined and converted to residual stresses using equations 1-3 and material data obtained from tensile
coupon tests performed on material cut from locations A-D in the test sample.

Figure 8 shows a longitudinal through thickness residual stress distribution taken from location D
with vertical error bars and horizontal lines indicating the overlapping of each measurement.

Figure 8: Longitudinal residual stress profile from location D.

The atomic spacings measured at locations A-D in the test sample were characteristic of an austenitic
stainless steel microstructure with no detection of the presence of a martensitic microstructure. This could
be because the martensitic microstructure created during cold working is small or it could be very
localized and its presence was not detected due to the use of a large gauge volume.

Despite the significant overlap of the through thickness measurements this pilot study successfully
demonstrates that neutron diffraction can be used to obtain through thickness residual stress data to a
good accuracy and that the variation of longitudinal residual stresses through the thickness of roll formed
stainless steel sections is clearly not linear as conventionally assumed. Further measurements using a
smaller gauge volume are planned to allow the measurement of through thickness residual stress

Rachel B. Cruise and Anna M. Paradowska

distributions to a higher resolution and thereby reduce any smoothing effect of the overlapping
measurements and also increase the chance of detecting any martensite present.

The author would like to acknowledge the hard work and technical expertise of Dr. Shu Yan Zhang
and to thank the Science and Technology Facilities Council for their financial and technical support.

[1] Way A.G.J., Popo-Ola S.O., Biddle A.R. and Lawson R.M., Durability of Light Steel Framing in
Residential Building, Second Edition, SCI, Silwood Park, Report no: P262, 2009.
[2] National Audit Office, Using modern methods of construction to build homes more quickly and
efficiently, 2005.
[3] Karren K.W., “Corner properties of cold-formed steel shapes”, Journal of the Structural Division,
ASCE, 93(ST1), 401-432, 1967.
[4] Cruise R.B. and Gardner L., “Strength enhancements induced during cold forming of stainless steel
sections”, Journal of Constructional Steel Research, 64 (11), 1310-1316, 2008.
[5] EN 1993-1-4, Eurocode 3: Design of steel structures - Part 1.4: General rules. Supplementary
Rules for Stainless steels, European standard, CEN, 2006.
[6] EN 1993-1-3, Eurocode 3: Design of steel structures - Part 1.3: General rules. Supplementary
rules for cold-formed members and sheeting, European standard, CEN, 2006.
[7] Key P.W. and Hancock G.J., “A theoretical investigation of the column behaviour of cold-formed
square hollow sections”, Thin-Walled structures, 16(1-4), 31-64, 1993.
[8] Moen C.D., Igusa T. and Schafer B.W., “Prediction of residual stresses and strain in cold-formed
steel members”, Thin-Walled structures, 46(11), 1274-1289, 2008.
[9] Quach W.M., Residual stresses in cold formed steel sections and their effect on column behavior,
PhD Thesis, Dept. of Civil and Structural Engineering, Hong Kong Polytechnic University, 2005.
[10] Gardner L., Cruise R.B., Sok C.P., Krishnan K. and Ministro J., “Life cycle costing of metallic
structures”, Proceedings of the Institution of Civil Engineers, Engineering Sustainability,
160(ES4), 167-177, 2007.
[11] Edwards L. and Endean M. (eds.), Manufacturing with Materials, Butterworth-Heinemann Ltd.,
[12] Spencer K., Embury J.D., Conlon K.T., Véron M. and Bréchet Y., “Strengthening via the formation
of strain-induced martensite in stainless steels”, Materials Science and Engineering, A 387-389,
873-881, 2004.
[13] Withers P.J. and Bhadeshia H.K.D.H., “Residual stress: Part 2 - Nature and Origins”, Materials
Science and Technology, 17:4, 366-375, 2001.
[14] Schafer B.W. and Peköz T., “Computational modelling of cold-formed steel”, Journal of
Constructional Steel Research, 47(3), 193-210, 1998.
[15] Cruise R.B. and Gardner L., “Residual stress analysis of structural stainless steel sections”, Journal
of Constructional Steel Research, 64 (3), 352-366, 2008.
[16] Jandera M., Gardner L. and Machacek J., “Residual stresses in cold rolled stainless steel hollow
sections”, Journal of Constructional Steel Research. 64(11), 1255-1263, 2008.
[17] Webster G.A. and Wimpory R.C., “Non-destructive measurement of residual stress by neutron
diffraction”, Journal of Material Processing Technology, 117(3), 395-399, 2001.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



László G. Vigh*

* Budapest University of Technology and Economics, Department of Structural Engineering

e-mail: geri@vbt.bme.hu

Keywords: patch load, steel and aluminium I-beams, curved flange-to-web connection, non-linear
numerical simulation, interaction of bending and transverse load resistance, Eurocode.

Abstract. Current design methods for calculating transverse load resistance of web plates neglect the
effect of the flange-to-web connection. This simplification is on the conservative side and in deep plate
girders its influence is practically negligible. However, in case of hot-rolled or extruded profiles, the
fillet corner shaping through its geometry, rigidity and strength may highly increase the patch load
resistance. Using non-linear numerical analysis, the author completed a parametric study on simply
supported girders subjected to transverse load in order to study a) the effect of the connection and b) the
capacity in interaction of transverse load and bending. The study clearly confirms the beneficial
influence of the curved corners. To take this effect into account the author proposes a modification in the
Eurocode formulation. The results of the simulation prove the validity of the proposed method.

It is well known that transverse (or patch) load resistance of plate girders is influenced by the load
case (patch load, opposite patch load, end patch load), the web and flange material properties (i.e. yield
limits), the dimensions (a, hw) and thickness (tw) of the unstiffened web panel, and the flange dimensions
(bf, tf). Additionally, the loaded length (ss) has major importance in the resistance, too. These details are
typically considered in the design method as well, just like in current formulations of corresponding
Eurocodes, such as Eurocode 3 Part 1-5 for steel plated structures (EC3 [1]) and Eurocode 9 Part 1-1 for
aluminium structures (EC9 [2]).
However, effect of the connection between the flange and web, thus the effect of the fillet curved
corners is fully neglected. Even though this simplification is on the conservative side and reasonably
accurate for deep slender girders, this type of connection shaping may highly increase the resistance, as it
is confirmed in [3]. Firstly, – similarly to the calculation of local plate buckling under compression – one
may claim to consider the fillet reducing the web height. Secondly, it widens the effective loaded length
of the web. Thirdly, it may highly influence the plastic hinge capacity of the flanges; that is the base of
the mechanism solution model of patch loading originally recommended by Roberts et al [4].
Reviewing the development of the actual code formula, the reason why this effect is out of
consideration can be found. (Note that EC9 applies the same method as given in the steel standard EC3.)
According to Lagerqvist et al [5], the calibration of the semi-empirical design method was completed on
the basis of 388 test specimens made of steel, including 358 welded girders, 11 European and 19
American rolled beams. Most of these tested girders come with high, slender web and only few cases
represent rolled/extruded profiles with stocky webs and relatively large curved corners. Consequently,
the calibration is directly valid for slender webs only. Needless to say, welded connection that has
smaller extent than practically applied curved corners of rolled profiles results in much smaller influence
on the resistance. It can be also stated that the higher and more slender the web is, the less the effect of

László G. Vigh

the edge boundary condition is. Note that this effect can be much higher in case of aluminium, because
extruded profile usually comes with larger radius due to fabrication and material reasons.
The author completed a parametric numerical study on various steel and aluminium I-beams
configurations and the corner effect is quantified. Based on the results, the author proposes a simple
modification in the Eurocode method to account for this detail and its beneficial effect.


This section summarizes the current basic procedures for bending and transverse load resistance
calculations according to EC3 and EC9 (referred as standard method hereafter).
In this study, only compact, ductile profiles (classified as Class 1 sections) with I-shape cross-section
(Figure 1) are considered, i.e. local plate buckling due to axial stresses does not affect the static
behaviour of the girder. Accordingly, plastic bending resistance Mc,Rd is calculated. EC9 alternatively
allows to consider strain hardening through the application of Ramberg-Osgood law, [2]. Thus,
fy fo
steel: M c , Rd = W pl aluminium: M c , Rd = α M ,1Wel (1/a,b)
γM0 γ M1

where Wpl and Wel are the elastic and plastic section modulus, respectively; αM,1 stands for the
correction factor to account for the plastic overstrength including strain hardening; fy and fo are the
characteristic yield strength and the proof strength, respectively; while γM is the partial safety factor.
Both EC3 and EC9 prescribe exactly the same mechanism-solution based procedure for the
transverse load resistance calculation. As per EC9, for simple patch load case the method follows:

ª 2
t w3 § h · º t3
critical load: Fcr = 0.9k F E = 0.9 «6 + 2¨ w ¸ » E w (2)
hw «¬ © a ¹ »¼ hw

( )
dimensionless parameters: m1 = f of b f / ( f ow t w ); m2 = 0.02 hw / t f ( )
effective loaded length: l y = s s + 2t f 1 + m1 + m2 ≤ a ) (4)
l y t w f ow
slenderness: λF = (5)
reduction factor: χ F = 0.5 / λ F ≤ 1.0 (6)
f ow
transverse load resistance: FRd = χ F l y (7)
γ M1

where E is Young’s modulus; kF is the buckling coefficient; a is the length of the unstiffened web
panel (Figure 1); hw and tw are the web height and thickness; bf and tf are the flange width and thickness;
ss is the stiff bearing length; fof and fow are the proof strength of the flange and the web, respectively.
The interaction of design bending moment MEd and transverse loading FEd shall be checked
through the following interaction formula:
+ 0.8 Ed ≤ 1.4 (8)
FRd M c , Rd

For simplicity, this study does not deal with the complex interaction of shear, bending and transverse
load. (Note that influence of shear load on the patch load resistance is currently not covered by the basic
method of Eurocode. A useful method is discussed in [6].) Additionally note that the following results are
corresponding to the simple patch load case; the author did not deal with opposite and end patch loading.

László G. Vigh


3.1 Programme
The author completed a parametric study in order to quantify the effect of the curved corners on
patch load resistance of rolled steel and extruded aluminium girders. Simply supported girders (with I-
shape cross-section shown in Figure 1) subjected to transverse concentrated load at midspan is
considered. Varying parameters are the section geometry, the span a (= length of unstiffened web panel),
the loaded length ss and the radius r of the curved corner. The parametric study programme is
summarized in Table 1. One series of analysis was carried out assuming HEA sections made of steel
grade S235 and one with modified HEA sections made of a specific AlMgSi alloy. In the aluminium
case, profile modification – namely, change of flange width – aimed to obtain ductile sections.
Altogether, the analysis series include more than 150 cases.
The cross-sections are ductile (Class 1) in each case. Varying the span permits of analysing cases of
dominant bending failure, dominant web crippling or their interaction. Three different loaded lengths are
investigated: 0 mm, 50 mm and 200 mm, respectively. The basic radius for the different sections are
12 mm, 18 mm and 27 mm, respectively; these values are multiplied by 0, 1, 1.5 and 2.
For simplification, interaction with shear is not discussed here: cases where the influence of shear on
design bending resistance is larger than 5% are excluded.




a) cross-section b) shell-element model

Figure 1: Parameters and numerical model for the parametric study.

3.2 Numerical modelling technique

For the parametric study, geometrically and materially non-linear analysis is completed using
ANSYS [7]. Figure 1/b shows the shell-element geometrical model. The 4-node SHELL181 element is
suitable to model the curved corner: the element may have linearly varying thickness along its edges.
Bilinear approximation – illustrated in Figure 2/a – is applied in such way that the same joint section area
is provided and thus the resulting transverse plate rigidity is certainly not overestimated.
One-bow geometrical imperfection is applied in the web with a magnitude of h/200. Elastic-perfectly
plastic bilinear material model is adjusted to the steel cases: to overcome numerical problems a fictive
strain hardening with a tangent slope of E/10000 is applied. To the aluminium specimens the elastic-
hardening non-linear model shown in Figure 2/b is adjusted. It is assumed that the profiles are
manufactured as a whole and no welding is necessary. Thus, welding does not influence the material
behaviour of the aluminium specimens.
The numerical model is validated in [3]. Note that according to the Eurocodes, – beside the
standardized procedure – such numerical simulation can be alternatively applied for design purposes.

Table 1: Programme and results

"HEA 100" / "HEA 100-75" "HEA 200" / "HEA 200-100" "HEA 300" / "HEA 300-140"
h bf tf tw r ss a h bf tf tw r ss a h bf tf tw r ss a
Steel Alu Steel Alu Steel Alu
[mm] [k N] [mm] [k N] [mm] [k N]
150 112.9 101.5 500 204.0 158.8 1000 371.8 277.3
300 106.1 93.9 0 1200 180.0 131.6 0 2000 334.5 233.4
500 95.2 83.5 2000 149.0 103.6 3000 293.0 195.3
750 81.5 71.9 500 256.1 204.8 1000 441.0 333.2
300 139.3 129.9 0 50 1200 231.0 173.3 0 50 2000 402.0 286.8
50 500 123.8 118.6 2000 188.0 122.8 3000 350.6 230.2
750 104.1 91.7 500 391.0 346.3 1000 645.0 ---
300 211.0 209.4 200 1200 325.0 230.7 200 2000 598.8 386.9
200 500 147.0 154.8 2000 212.0 135.5 3000 435.0 262.6
750 124.0 112.4 200 500 245.0 200.7 300 1000 460.1 355.5
150 135.8 130.3 (St) 0 1200 212.0 163.8 (St) 0 2000 404.2 300.0
300 125.0 116.7 2000 173.8 124.2 3000 348.1 244.3
(St) 0 190 10 6.5 290 14 8.5
500 109.7 101.9 500 303.4 251.6 1000 537.9 419.0
96 8 5 750 93.0 84.4 100 18 50 1200 269.6 207.9 140 27 50 2000 482.0 355.7
150 189.9 184.2 (Al) 2000 205.2 139.7 (Al) 3000 409.8 270.9
(Al) 12 300 155.0 154.7 500 417.0 401.3 1000 691.0 622.3
500 134.0 135.3 200 1200 329.0 256.3 200 2000 641.0 441.4

750 110.9 99.5 2000 222.0 150.7 3000 463.0 293.9
300 223.0 246.8 500 282.2 240.1 1000 535.7 434.0
László G. Vigh

200 500 148.0 168.2 27 0 1200 243.0 194.5 40.5 0 2000 471.5 362.4
750 127.0 121.9 2000 197.0 142.4 3000 404.1 286.5
500 123.0 116.8 500 331.1 288.4 1000 626.5 529.7
18 0
750 103.2 94.6 36 0 1200 283.7 231.7 54 0 2000 558.0 437.4
150 187.2 190.7 2000 221.0 164.8 3000 465.0 338.1
300 159.0 163.2
24 0
500 137.7 134.7 h – total section height; bf – flange width; tf– flange thickness; tw – web thickness; r – radius of corner;
750 114.2 106.9 ss – loaded length; a – span (= length of unstiffened web panel); FRd,FEM – FEM load capacity

Table 2: Comparison to r = 0.

Steel Alu
% avg % % avg %
r 5 - 24 16 8 - 29 20
1.5 r 27 - 44 36 32 - 57 46
2r 40 - 67 54 49-91 74
László G. Vigh

350 E = 70 GPa; n = 49.5

fo = 249 MPa; fu = 268 MPa


Stress (MPa)
shell elements 200 E = 210 GPa; Et = 21 MPa
t2 fy = 235 MPa
with varying
tw thickness
100 AlMgSi
axis of shell
elements 0
0 0.05 0.1 0.15
Note: these are not body elements Strain (-)
a) approximation of curved corners b) material models
Figure 2: Modeling details.
3.3 Discussion of results

3.3.1 Results
The load capacities obtained by the analyses are tabulated in Table 1. Utilization factor for bending
moment is calculated as the ratio of the ultimate load FRD (or MRd) computed from the numerical analysis
to the design bending moment resistance Mc,Rd calculated in accordance with Eurocodes. Utilization
factor for transverse load resistance (FRD to Fpatch,RD) is similarly determined. The interaction results are
illustrated by the help of these parameters in Figure 3 where the EC interaction curve is also plotted.
3.3.2 Reliability of the results
The applied non-linear numerical analysis is an alternative design procedure allowed for by both
Eurocodes. In Figure 3/a, the interaction points corresponding to the reference case r = 0 lie in the
vicinity of the standard interaction curve, which in general confirms the validity of the results. On
average the deviation is within 10%. However, in many cases the numerical simulation gives
conservative result compared to the standard procedure. Different interpretations may be given for this
1) On the one hand, this may indicate that the numerical model is conservative, i.e. especially the
applied geometrical imperfection, or the way of joint discretization, etc. is conservative.
2) Reference comparison perhaps should be done to cases of normal radius, as the calibrated design
method may indirectly include some connection effect.
3) The fact that zero loaded length cannot be kept in experimental environment queries the
reliability of the standard method: it may overestimate the capacity for cases ss = 0. When
excluding these cases, only 5 of the points fall below the standard interaction curve.
As a consequence, further study is required to accurately evaluate the reliability of the alternative
design methods.
Despite the discussed uncertainties, it can be stated that the numerical results are in accordance with
the standardized method and the analysis with different radius and loaded length gives a solid base for the
following comparative study.
3.3.3 Effect of curved corners
The results confirm that the curved corner may highly influence the transverse load resistance as well
as the resistance in interaction with bending. Compared to the reference cases r = 0, even the
consideration of normal radius leads to notable increase in the capacity, as Figure 3/a,b and Table 2
prove. When using double radius, up to 67% and 91% increase can be achieved in case of steel and
aluminium, respectively.

László G. Vigh

Load vs. deflection curves and deformed shapes of Figure 4 well demonstrate the quantitative and
qualitative change in capacity and in nature of behaviour. For example, compare post-ultimate behaviour
of cases ss = 200 mm in Figure 4/a: in case of no radius web crippling dominates as indicated by the
sudden drop in the post-peak range, while the existence of curved corner leads to governing bending
failure. In the latter case, due to the ductile (Class 1) section, long yield plateau can develop, followed by
the capacity drop due to instability at the very end of the curve.
The larger influence in case of aluminium alloy can be explained by two reasons:
1) The selected aluminium profiles are more sensitive to web crippling than the steel ones, which is
also reflected by the reduction factor χF calculated in accordance with EC: it ranges within
0.9~1.0 for the steel and within 0.53~1.0 for the aluminium cases, respectively.
2) Strain hardening is considered in the aluminium calculations. Note that the manual calculation
also accounts for the strain hardening in the bending moment resistance formulation, but not in
the transverse load resistance.
The results thus promise that improved capacity values in the steel cases can possibly be achieved by
implementing advanced (more accurate and realistic) non-linear simulation.

2 EC 2.5
r= 1 r
r= 0
r= 1 r r = 1.5 r
1.6 2
r = 1.5 r
r= 2 r
1.4 r= 2 r
FRd / Fpatch,Rd

FRd / FRd,r=0

1.2 1.5

0.8 1


0.4 0.5


0 0
0 0.5 1 1.5 0 0.5 1 1.5
MRd / Mc,Rd MRd / Mc,Rd

a) M-F interaction b) corner effect

Figure 3: Interaction results and corner effect.



Load (kN)

400 ss = 0;
300 ss = 0; r = 0
r = 1.5r
ss = 0; r = 1
200 ss = 0; r = 1.5
ss = 0; r = 2
100 ss = 200; r = 0
ss = 200; r = 1
0 5 10 15
ss = 200;
Deflection (mm)
a) load vs. deflection curves b) deformed web at ultimate load
Figure 4: Typical simulation results – steel, HEA300, a = 2 m.

László G. Vigh


Based on the previous observations and parametric study results, the author proposes two simple
modifications in the standard design method in order to utilize the advantageous effect of the curved
connection configuration.
On the one hand, the curved configuration results in larger transverse plate bending rigidity of the
web (analogous to the bending capacity of haunched girder – column connection). Consequently, it is
supposed that – similarly to the calculation of the plate buckling due axial stresses – the clear web height
hw between the inner ends of the radius is used instead of the full web depth, as shown in Figure 5/a.



hw R
a) effective web height hw

tf n.a tf’

Mpl,1 = Mpl,2
b1) equal section modulus

tf tf’’

A1 A2
A1 = A2
b2) equal area method
b) effective flange thickness tf or tf’’ ’

Figure 5: Determination of modified section properties.

2 EC
1.8 radius neglected
eq. area
eq. section mod.
FRd / Fpatch,Rd






0 0.5 1 1.5
MRd / Mc,Rd

Figure 6: Comparison of methods.

László G. Vigh

On the other hand, the plastic hinge of the flanges occurring in the ultimate stage (plastic hinge
mechanism) can extend to the curved corner area. Thus, the curved corner can be considered as part of
the flange. Since the resistance against transverse loading is dominated by the flange plastic hinge, one
has to calculate the plastic bending capacity of a fictive section including the flange and the
accompanying corner area. Without rebuilding the existing design formula, this can be easily achieved by
introducing an effective flange thickness providing the same local plastic capacity (Figure 5/b1). This is
referred as equal section modulus method hereafter. As a simplification, the effective thickness can be
conservatively calculated by simply smearing the curved corner area to the flange (Figure 5/b2, referred
as equal area method). This latter method gives smaller effective thickness than the previous, more
accurate one; consequently, it is always on the safe side.
Introducing these two modifications into the design method, the interaction relation shown in
Figure 6 is obtained for the studied configurations. Regardless to the mentioned uncertainties (knife-edge
load, etc.), it is concluded that the proposed modification gives more accurate evaluation of the patch
load resistance.

Parametric study is completed on simply supported girders made of rolled steel or extruded
aluminium profiles, subjected to transverse load. Based on the results, the following conclusions are
Influence of the curved-corner web-to-flange joint on the transverse load resistance can be significant
in case of stocky webs; the increase in capacity may reach 60-90%.
To take this beneficial effect into account, the author proposed a simple modification in the current
Eurocode design method. The modified procedure utilizes the clear web height and effective flange
thickness; thus, more accurately representing the actual connection rigidity and plastic flange strength.
The proposed procedure well estimates the transverse load – bending interaction capacity computed
by non-linear numerical simulation.
Further study is needed on the relation of the basic standardized procedure and the numerical
simulation with respect to reliability. The research shall include study on the role of imperfection, knife-
edge loading case, material modelling.
The method should be validated to other load application cases, as well.
Interaction of transverse load, bending and shear is additionally subject to further research.


[1] EN 1993-1-5:2005 Eurocode 3 – Design of steel structures – Part 1-5: Plated structural elements,
Final Draft, CEN, Brussels, 2005.
[2] EN 1999-1-1:2007 Eurocode 9 – Design of aluminium structures – Part 1-1: General structural
rules, CEN, Brussels, 2007.
[3] Vigh, L.G., Virtual and real test based analysis and design of non-conventional thin-walled metal
structures, PhD dissertation, Budapest University of Technology and Economics, 2006.
[4] Roberts, T.M., “Slender plate girders subjected to edge loading”, Proc. Inst. Civil Engineers, Part
2, Vol. 71, 805-819, Sept. 1981.
[5] Lagerqvist, O. and Johansson, B., “Resistance of I-girders to Concentrated Loads”, Journal of
Construction Steel Research, 39 (2), 87-119, 1996.
[6] Kövesdi, B., Kuhlmann, U., Dunai, L., “Combined shear and patch loading of girders with
corrugated webs”, Periodica Polytechnica Civil Engineering, 2010. (submitted for publication)
[7] ANSYS Structural Analysis Guide, Online Documentation ANSYS Inc., 2005.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Gabriel A. Jiménez, Ph.D., P.E., S.E.

Principal/Managing Director, Walter P Moore and Associates, Houston, Texas, USA

e-mail: gjimenez@walterpmoore.com

Keywords: Tapered, Stability, Beams, Columns, Beam-columns, Steel Structures.

Abstract. The purpose of this paper is to present solutions for the elastic and inelastic lateral-torsional
buckling of steel web-tapered beam-columns using two different computational procedures and compare
these solutions with the AISC Specification for tapered members.

Web-tapered members are structural members commonly used in the typical one-story pre-engineered
building. Appreciable savings in materials and in the cost of structural framing can be assumed by the
use of elements having a tapering depth or flanges.
In the United States of America, the last specification that addressed tapered members is the 1999
American Institute of Steel Construction Specification [1] for web tapered members which was based on
a study performed in 1966. The contributors to the study were the Column Research Council, presently
known as the Structural Stability Research Council, and the Welding Research Council, under the
technical guidance of Lee et al. [3] at the University of New York at Buffalo. The general design
approach used in the 1999 Specification is to apply modification factors to convert the tapered members
into appropriately proportioned prismatic members so that the prismatic AISC equations may be applied.
From the practitioner’s point of view, the 1999 AISC design equations for tapered members represent the
use of existing basic formulas for prismatic members altered with the use of an additional factor.
Furthermore, the additional factor will give the designer an inherent feeling for the increase in strength
over a prismatic section.
At the same time, the “easy to use” 1999 AISC Specification is restricted to doubly symmetric I-
shaped sections. The reason for this limitation was the inability to uncouple the torsional and flexural
deformations due to varying locations of the shear center for singly-symmetric sections during Lee’s
study. The development was also limited to members with small tapering angles. According to Lee et al.
[5], Boley showed that the methods used by Lee and his colleagues to compute normal stresses are
reasonably accurate as long as the tapering angle is less than 15 degrees. For practical considerations, the
limiting tapering ratio has been further restricted to 6. Moreover, the development is limited to members
with flanges of an equal and constant area with webs that are not slender. However, what is of interest is
that the current practice in the low-rise metal building industry is the use of flanges of unequal area and
slender webs. Therefore, the 1999 AISC Specification does not appear to provide equations for web-
tapered I-shaped beam geometries of proportions that are consistent with what has been the industry
standard for metal buildings.
Jimenez et al.[5] and other researchers have performed new studies on the topic of inelastic stability
of tapered members and have shown that the 1999 AISC equations predict unconservative results when
determining the lateral-torsional buckling strength of tapered beams and beam-columns for certain

Gabriel A. Jiménez

slenderness values of typical tapered members. The current AISC Specification [2] does not explicitly
define the use of the AISC provisions for tapered members. In 2006 White et al. [6] performed a
prototype study on how to use the current AISC Specification to tapered members. The findings of their
prototype study appeared to generate reasonable solutions; however additional verifications with other
versions of the code as well as experimental results are needed.
The general behavior of a typical beam-column is illustrated in Figure 1, where the relationship
between the applied end-moment Mo and the resulting end-slope θ is shown for a wide-flange member
bent about its strong-axis, in which the length as well as the axial force P is assumed to remain constant
as the moment Mo is increased from zero to its maximum value and past the maximum moment into the
unloading zone.
The optimum performance of the beam-column is reached if failure is due to excessive bending in the
plane of the applied moment, and this behavior is represented by the upper branch of the curve in Figure
1. The corresponding maximum moment is Mo1max. If no lateral bracing is provided, failure will be due to
lateral-torsional buckling and the resulting moment is Mo2max represented by the lower branch of the
curve in Figure 1. The additional incremental moment represented by Mo2max beyond Mocr is small, and,
therefore, the bifurcation point is considered to reasonably determine the buckling limit to the beam-
column. The work described in this paper deals with the determination of the value of Mocr for web-
tapered beam-columns.

Figure 1: M-Θ curves for beam-columns


The differential equations governing the lateral-torsional buckling of tapered members subjected to
centroidal axial forces P and to end moments Mo and ρMo are given in Jimenez [4] and are repeated here
for convenience:

d 2v ª zº
B x (z ) + Pv − Mo« ρ + (1 − ρ ) » = 0 (1a)
dz 2
¬ L ¼

d 2u ª ­ z½ º
By + Pu − β «Mo ® ρ + (1 − ρ ) ¾ − P y o (z )» = 0 (1b)
¬ ¯ L¿
dz ¼

Gabriel A. Jiménez


The beam-column prescribed by the above differential equations is shown in Figure 2. It is subjected to
end bending moments Mo at z = L and ρMo at z = 0, where “z” is the coordinate axis along the
undeformed centroidal axis and “L” is the length of the member. The coefficient “ρ” is the ratio of the
end moments. The deformations of the shear center are: “u” in the x-direction, “v” in the y-direction and
the cross-section twists about the shear center an angle “β”. In Figure the smaller end will be denoted as
end A and the larger end as end B.

Figure 2: Loading condition. Figure 3: Stress-strain diagram.

The stress-strain diagram of the material is shown in Figure 3. The coefficients Bx(z), By, CT(z), Cw(z),
yo(z) and K( z) in the differential equations are defined as follows: Bx(z) is the bending stiffness about
the x-axis; By is the bending stiffness about the y-axis; CT(z) is the St. Venant’s torsional stiffness; Cw(z)
is the Warping stiffness; yo(z) is the distance between the centroid “C” and the shear center “S” in the

plane of symmetry; K( z) = ³ σs d A
where: σ = is the stress on any cross-sectional element dA
(positive in compression) and “s” is the distance of element dA from the shear center. These coefficients
vary with respect to the coordinate “z” to account for the non-uniform variation of the cross-section
properties along the length of the column. Also, when the beam-column is in the inelastic range the
coefficients will vary with the different patterns of the yielding.


Solutions for the elastic and inelastic lateral-torsional buckling of steel web tapered beam-columns
were computed using advanced analyses. The beam-column elements are subjected to an axial force and
to bending moments applied at both ends of the member. A computational procedure based on the finite
difference method using a direct discretization of the differential equations of lateral-torsional buckling
was utilized. The coefficients appearing in the finite difference equations are determined considering the
reductions of the flexural and torsional stiffnesses due to yielding in the inelastic range. The effects of
residual stresses are included. The resulting simultaneous equations are then set up to compute the
buckling determinant which yields the critical load.
The lateral-torsional buckling of tapered beam-columns is determined by using equations (1b) and
(1c) where the cross-section coefficients are variable with respect to “z”. The finite difference equations
corresponding to the equations (1b) and (1c) at each station by first-order central differences becomes:

Gabriel A. Jiménez

[ ] [
ui −1[By(i )] + ui Ph 2 − 2By(i ) + ui+1[By(i )] + βi λ (i )h 2 = 0 ] (2b)

[ ] ª 2Mo(1 − ρ )h 2 º
ui−1 − λ (i )h 2 + ui « [ ]
» + ui +1 λ (i )h + βi−2 [− C w (i )] +

¬« n ¼»

ª § K (i ) ·¸º ª ­ § K (i ) ·¸½°º
βi −1 «2C w (i ) + CT (i )h 2 ¨1 − » + βi +1 «− ®2C w (i ) + CT (i )h 2 ¨1 − »+
«¬ ¨
© CT (i ) ¹»
¸ « °̄ © CT (i ) ¹°¿»¼
¼ ¬

βi+ 2 [C w (i )] = 0

ª iº
where : λ (i ) = Py o (i ) − M (i ), M (i ) = Mo « ρ + (1 − ρ ) »
¬ n¼

The ends of the beam-column are allowed to rotate, the end sections are free to warp, and the ends of the
member are not permitted to twist or to translate. These boundary conditions can be written as follows:
uo = 0, un = 0, β-1 = -β1 , βn+1 = - βn-1 ,βo = 0,βn = 0.

This leads to a set of simultaneous algebraic equations in the lateral displacement u and the rotation β at a
number of discrete points spaced at h = L/n, in which n is an odd number to which the beam-column is
­u ½
divided. This set of simultaneous equations may be written in matrix form: [ A] ® ¾ = 0 . In this equation
¯β ¿
the matrix [A] is a set of the coefficients Aij representing combinations of the cross-section properties (By,
CT(z), Cw(z), yo(z), and K( z) ), the load parameters ( P and Mo) and the length of the member (L). In
order to compute the stiffness of a cross-section it is necessary to know how much of the section is plastic
and how much of the section is elastic, and where the corresponding regions are located on the cross-
section. The non-dimensionalized M/My, φ/φy, P/Py, relationships about the strong-axis for an I-shape
section have been determined by Jimenez [4]. Figure 4 shows these relationships for the following cases
of yielding:

Case 1 Case 2 Case 3 Case 4

Figure 4: Yielded patterns for wide-flange cross-section.

In outline form, the steps that are used in computing the critical moment Mocr for steel web-tapered beam-
columns are as follows:

Gabriel A. Jiménez

Given: (L/r), Assume: In-plane analysis:

Compute moment, curvature
(P/Py)A, ρ (Mo/Mp)B
(extent of yielding) By,
CT(i),Cw(i), yo(i), K( i )
Compute determinant
of the coefficient Formulate the finite-difference-
matrix [A] equations (out-of-plane equations: (2b)
and (2c))

This process is repeated for different load levels until a zero value for the determinant is found.

To create the finite element model using ANSYS, a commercially available finite element program,
several steps had to be performed including element selection, laying out the mesh and determining
boundary conditions. The finite element mesh is comprised of BEAM188 elements. BEAM188 elements
are suitable for analyzing slender to moderately stubby/thick structures. This element is based on
Timoshenko beam theory. Shear deformation effects are included. The BEAM188 is a quadratic beam
element in 3-D. This element is well-suited for linear, large rotation, and/or large strain nonlinear
applications. Furthermore, the provided stress stiffness terms enable the elements to analyze flexural,
lateral, and torsional stability. The cross-section associated with the element may be linearly tapered.
Elasticity and plasticity models are supported.


Solutions for the elastic and inelastic lateral-torsional buckling of steel web tapered beam-columns
were computed using both the finite difference method and a commercially available finite element
program. The beam-column elements are subjected to an axial force and to bending moments applied at
both ends of the member. Figure 5 compares the Finite Difference (FD) solution with the ANSYS
solution for a typical tapered beam-column subjected to the forces shown. In this case γ represents the
taper ratio, L/rx represents the slenderness parameter about the x axis, rx is the radius of gyration about the
x axis, and Mp is the plastic moment.
Lateral Torsional Buckling Strength ofTapered Beam- (P/Py)A = 0.2, γ = 1

0.8 M

(M/Mp) A

0.5 Finite
Difference M
0.4 Ansys
0.3 p



0 20 40 60 80 100 120

Figure 5: Comparison Between FD and ANSYS models.

Gabriel A. Jiménez

It appears that the ANSYS solution produces more conservative results for the slenderness ratios between
40 to 90. This behavior is due to the gradual yielding in the ANSYS model versus the four defined
yielded patterns from Figure 4 utilized in the FD method. Figure 6 corresponds to the overall buckled
shape of the tapered beam-column as depicted by ANSYS. Yielding of both flanges at the smaller end is
evident. A close-up look of the smaller end is depicted in Figure 7.

Figure 6: Overall view of buckled tapered beam-column.

Figure 7: Close up view of yielded flanges/web at the smaller end of a tapered beam-column.

Comparisons were made between the 1999AISC-LRFD code and this study. Figure 8 illustrates a
typical case of a tapered beam-column subjected to compressive axial load and end moments for lateral-
torsional buckling (LTB). The beam-column problem is treated in the 1999 Specification in the form of
an interaction equation. It can be seen that for values of M/Mp greater than about 0.4 the predictions of
the interaction equation are unsafe. The unconservative results predicted by the use of the beam equation
in the AISC Specification are typical for different tapering ratios with unsafe discrepancies up to 25 %
between the advanced analysis and the Specification.

Gabriel A. Jiménez

Figure 8: LTB strength of beam-column.

This paper presents studies for the out-of-plane behavior of tapered beam-columns using ANSYS and
the Finite Difference method. It was shown that the ANSYS solution produces slightly more conservative
results due to the progressive yielding of the flanges and web. It was found that for medium to short
beams and beam-columns, the 1999 Specification [1] predicts strengths on the non-conservative side,
with maximum discrepancies of about 25% between advanced analysis approaches and those given the
specifications. Jimenez and Galambos [5] suggested an alternate set of equations to better predict the
strength capacity of tapered beams and beam-columns.
Furthermore, additional studies are needed to evaluate the use of the 2005 AISC provisions [2] with
previous specifications for tapered members.


[1] American Institute of Steel Construction, Load and resistance factor design specification for
structural steel buildings, AISC, 1999
[2] American Institute of Steel Construction, Load and resistance factor design specification for
structural steel buildings, AISC, 2005
[3] Lee G. Morrell M. and Ketter R., “Design of tapered members”, Welding Research Council
Bulletin, 173, 1-32, 1972.
[4] Jimenez G., “Inelastic stability of tapered structural members”, Doctoral Dissertation, Minneapolis,
Minnesota, 1998.
[5] Jimenez G., and Galambos T.V., “Inelastic lateral-torsional buckling of tapered beam-columns”,
Proceedings, Annual Technical Session, Structural Stability Research Council, 1998.
[6] White D., and Kim, Y. D., “A prototype application of the AISC (2005) Stability and Design
Provisions to metal building structural systems,” Metal Building Manufacturers Association, 2006.

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010


Peter Osterrieder*, Stefan Richter*, Matthias Friedrich**

* Faculty of Architecture, Civil Engineering and Urban Planning,

Brandenburg Technical University BTU Cottbus, D-03044 Cottbus, FRG,
e-mails: lsud@statik.tu-cottbus.de
** Friedrich+Lochner GmbH, D-01067 Dresden, FRG

Keywords: Lateral torsional buckling, local hinges, first-yield design, warping torsion of open thin-
walled members.

Abstract. Continuous steel beams used for girders, purlins, crane girders, etc. are subjected to lateral
torsion buckling. In many cases construction joints of those beams are carried out as moment hinges
rather than as rigid connections. The study is concerned with the effect of these hinges on the lateral
stability behaviour and their effect on the design. Based on a numerical approach critical lateral
buckling loads are calculated for different hinge locations and moment distributions. Finally load
capacities obtained from equivalent slenderness approach are compared to those from first-yield
criterion with internal forces from theory 2nd order analysis.

Figure 1 shows common type of construction joints for continuous I-beams which will be assembled
on the construction site. These joints are able to transfer shear forces but only small fractions of the
bending, torsion and warping moments associated with 3-dimensional loading and nonuniform torsion.
Therefore in the structural analysis it is assumed for the hinge that lateral displacements v and w are
compatible and
My Mz Mx MZ 0 (1)

Figure 1: Examples of common construction joints for continuous beams

P. Osterrieder et al.

While moment hinges may be easily considered in any commercial computer program for in-plane
stress analysis the allowance of local static boundary conditions in the eigenvalue calculation required for
lateral buckling design is rather unusual. To account for local boundary conditions within a displacement
based finite element formulation basically two alternatives are available. Applying as usual static
condensation to local element stiffness matrix leads to coupled elastic and geometric element matrices
which are not suitable for numerical eigenvalue solution applying vector iteration. Introducing instead
double nodes and incorporating coupling conditions during assembly of the system stiffness results in
uncoupled elastic and geometric system stiffness matrices. The latter approach has been applied to a FE-
program for nonlinear analysis of 3-D beams with open thin-walled cross sections [1]. Bifurcation loads
from this numerical approach are applied to check for lateral buckling according to Eurocode 3 [2,
paragraph]. Design loads from this approach are compared to ultimate loads obtained by an elastic
theory 2nd order analysis considering geometric out-of-plane imperfections according to [2, paragraph


As described above the eigenvalue problem of a continuous beam with a local hinge has been solved
within a geometric nonlinear formulation by introduction of double nodes with subsequent coupling of
dofs during the assembling process of elastic and geometric system stiffness matrices. Two procedures
are available for the numerical solution of the general eigenvalue problem det(A-ȜB) = 0. First a rather
simple algorithm based on a modified inverse iteration with random generated starting vectors and
automatic shifting is started. If no convergence is reached subspace iteration with QZ-Algorithm is
initiated. To verify the procedure bifurcation loads for beams with moment hinge at midspan (fig. 2) have
been compared with closed-form and numerical solutions.

L = 10 m, IPE360 section
q applied at centroid

q cr 58.9 kN/ m ș'x (0) = 0, ș'x (L) = 0

q cr 32.4 kN/ m ș'x (0) z 0, ș'x (L) z 0
B.C. Tx Ty Tz 0 at x 0, x L
Figure 2: Lateral torsional buckling of beam with hinge

Critical loads for a cantilever beams which is equivalent to the structural system in figure 2 with a
span of 5m are calculated from analytical equations in [3, eqns.(9.14), (9.17)] as qcr = 57,2 kN ( ș'x 0 at
both ends) and qcr = 27.9 kN ( ș'x z 0 at both ends) with ș'x as the warping displacement. Numerical
results for the cantilever beam obtained from [1] are identical to those given for the entire beam with the
hinge in figure 2.
Specific attention is required, when the joint design in addition cannot transfer shear forces in y-
direction at all or only with considerable web deflections. In this case the left and the right side of the
beam is essentially uncoupled and so lateral buckling behaviour of the two substructures.

Figure 3: Lateral torsional buckling mode of mono symmetric beam with moment hinge

P. Osterrieder et al.

If the cross section is not doubly symmetric as shown in figure 3, displacements v und w in the
direction of the major y- and z-axes are related to the shear centre of the cross section.


3.1 Structural idealization

To check any continuous beam as shown in figure 4 for lateral torsional buckling applying equivalent
slenderness procedure the associated lowest eigenvalue qcr for the entire structural system is required.

Figure 4: Continuous beam with moment hinge in midspan

Most common qcr is calculated alternatively in engineering practice for a virtually cut out single span
beam only with appropriate geometric and static boundary conditions at the respective supports. The
lateral torsional stability is governed by weakest – the most slender – beam. The interaction between the
single spans depends not only on the major axis bending moments but especially on the boundary
conditions for out-of-plane bending (v, șz) and torsion (șx, ș'x ). While in most real structures lateral
deformations v and torsional rotations șx at the supports will be restrained, boundary conditions for șz and
ș'x depend strongly on the out-of-plane bending and torsional stiffness of the adjacent spans. To illustrate
the sensitivity of the buckling load qcr with respect to the b.c. critical loads are calculated for the three
single spans in fig. 4 with and without warping restraint at the ends and related to the critical value qcr =
24,94 kN/m for the entire system. Results in table 1 are for L1 = L2 = 8m, L = 10m, Į = 0.3 and IPE330
section. To avoid kinematics in the eigenvalue analyses for span II the rotation at the right end about the
major y-axis is completely restrained.
For critical loads in row 2 it is assumed that the rotation șz about the weak z-axis is completely
restrained for all beams at the intermediate supports. Results in row 3 are calculated for unrestrained
rotations șz in a for the left beam, in b for the right beam and the rotation spring stiffness

EI z ª kNm º
kT z 3 620, 6 « » (2)
L1 ¬ rad ¼

at both supports of the beam with hinge. From the results in table 1 it is obvious, that buckling of the
middle span governs the stability problem and further that the middle span is elastically restrained against
out-of-plane bending and warping by both outer spans. More general it can be concluded that for most
practical problems of continuous beams the span containing the hinge will be relevant for stability

P. Osterrieder et al.

design. Thus the critical load may be approximated considering only the span with the moment hinge.
Fully warping restraint at both supports leads to the upper bound and free warping to the lower bound for
the critical eigenvalue.

Table 1: Critical loads qcr

span I span II span III
1 T'x,a 0 T'x,a z0
T'x,a 0 T'x,a z 0 T'x,b 0 T'x,b z 0
T'x,b 0 T'x,b z0
2 1.58 1.26 1.22 0.95 1.31 1.11
3 1.55 1.26 1.20 0.95 1.19 1.05

From the smallest critical load the critical moment and the dimensionless lateral torsional buckling

Wy f y
OLT (3)
M cr

has to be evaluated for the subsequent design check.

3.2 Critical Loads for single span beams with a moment hinge

Figure 5: Bending moment distribution for single span beam with hinge

For a single span beam with a moment hinge as specified in eqn. (1) at distance ĮL from the left end
(fig. 5) the moment Mb at the right end depends further on the moment Ma and the uniformly distributed
load q

§ qL2 Ma ·
Mb E¨  ¸ (4)
© 2 D ¹
As long as the hinge is close to the centre of the span the left and the right segments will interact in
lateral torsional buckling depending on the bending moment distribution. When the hinge gets closer to
the right or left bearing, the shorter beam segment will support the longer segment in the out-of-plane
behaviour. For a hinge very close to one of the bearings the out-of-plane boundary conditions at the hinge
for the remaining longer segment may be approximated as

v 0 T z z 0 T x z 0 T x' z 0 (5)

Figure 6 shows a comparison of critical loads and associated eigenmodes for a beam with uniformly
distributed load q, a hinge at Į = 0.3 and equivalent bending moment distributions about major y-axis in

P. Osterrieder et al.

both systems. The total buckling mode is governed by the weaker beam segment and differs from the
partial mode only slightly with an increase in the critical load for the partial system of 11%.

Bending moment distribution

qcr = 6.87 kN/m

Buckling mode of beam with hinge

Out-of-plane B.C.
qcr = 7.61 kN/m = 1.11·6.87

Buckling mode of right beam segment

Figure 6: Buckling modes and buckling loads for total and partial structure

Results in table 2 for a beam with a total span of 10 m, IPE300 section, varying hinge location Į, out-
of-plane b.c. as shown in fig. 6 and bending moment Ma = 0 confirm this observation more generally.

Table 2: Comparison of critical loads

ǹ qcr [kN/m] total beam qcr [kN/m] right segment
0.5 8.30 13.09
0.4 7.42 9.67
0.3 6.87 7.61
0.2 6.10 6.25
0.1 5.29 5.30

In figures 7 to 9 the dimensionless critical buckling loads

qcr L3
qcr (6)
EI z GI t

for beams with hinge b.c. given in eqn. (1) and șz = 0 at both supports are plotted over the stiffness

k (7)
L GI t

In [3] and similar in [4] it has been shown, that critical loads presented in this dimensionless form are
applicable to almost any beam with hot-rolled doubly symmetric I-section. From figs. 7b to 9b it follows
that for beams restrained against warping at the ends the relation between the dimensionless stiffness and
the dimensionless critical load is almost linear. For beams with free warping b.c. at both ends the relation
is found to be highly nonlinear with almost asymptotic behaviour for increasing stiffness.

P. Osterrieder et al.

Figure 7: Critical loads for beam with hinge at Į = 0.4

Figure 8: Critical loads for beam with hinge at Į = 0.5

Figure 9: Critical loads for beam with hinge at Į = 0.6

P. Osterrieder et al.


4.1 Lateral torsional buckling resistance according to Eurocode 3

The resistance according to EC3 [2,] for hot rolled sections is

§ ·M ­ d 1,0
M pl ,y 1 °
EC 3
M R,d F LT
¨ )  ) 2  EO 2
¸ pl ,y
¸ JM1
®d 1 ) LT 2 º (8)
0,5 ¬ª1  D LT OLT  OLT ,0  EOLT ¼
© LT LT LT ¹ ° O2
¯ LT

The following investigation was carried out for beams with variable span L, moment hinges and
IPE300 section (section class 1 [2, table 5.2], buckling curve b [2, table 6.4] and imperfection coefficient
ĮLT = 0,34 [2, table 6.4]). Further it is conservatively assumed that ȕ = 0,75, Ȗ M1 = 1,1, ȜLT,0 = 0, 4 and
f = 1 . For lateral buckling capacities with Į = 0.5 (fig. 10a) Ma = Ma = qL2/8 and for Į = 0.4 (fig. 10b)
and Ma = qL2/12 the moment Mb = 0.175qL2. Again boundary conditions at both ends are such that
rotations about z-axis are restrained and warping unconstrained.

Figure 10: Dimensionless lateral torsional buckling resistance

Figures 10a and 10b show dimensionless load capacities depending on the dimensionless buckling
slenderness O LT . Curves qu,EC3 in fig. 10 are obtained by dividing the ultimate load derived from eqn.
(8) by the load qpl associated with fully plastic moment. They agree with the Ȥ LT -distribution. It is to
notice that eqn. (8) does not take into account the effect of shear forces. Therefore the cross section
resistance has to be checked additionally and will restrict the capacity - specifically for small buckling
slendernesses - with respect to provisions in [2, 6.2.8].
4.2 Lateral torsional buckling resistance based on theory 2nd order analysis

Figure 11: Lateral torsional buckling mode of tapered beam with moment hinge

P. Osterrieder et al.

For stability design of more general structures like continuous beams with discrete or continuous
elastic support, arbitrary boundary conditions, intentionally out-of-plane loading and variable cross
section (see fig. 11) a theory 2nd order three-dimensional stress analysis with geometric out-of-plane
imperfections leads to a generally applicable approach.
For class 1 cross sections two criterions [5], [6] are available for definition of the ultimate load
- theory 2nd order first yield criterion
- theory 2nd order first hinge criterion
For first yield design the von-Mises stress ıv anywhere along the beam axis within the cross section,
obtained from a theory second order analysis of the geometrically imperfect structure with linear elastic
material behaviour must satisfy
max ıVII ( ı xII,d )2  3( IJ dII )2 d (9)
ȖM 0
In (9) VIIx,d is the theory 2nd order axial stress due to the combined action an axial force N, bending
moments M y and Mz about principal axis and the warping moment MȦ from nonuniform torsion. The
theory 2nd order shear stress WdII is calculated from shear forces Vy and Vz and the St. Venant’s torsion
moment MTP. Graphs qu,el,v in figs. 10 are obtained by dividing the ultimate load derived from eqn. (9)
by the qpl. From comparison with the qu,el,x curve, which neglects shear stresses in eqn. (9) it is obvious,
that in the stocky slenderness area the capacity is essentially limited by the shear stresses. In the moderate
slender area the first yield criterion leads, depending on the moment distribution, to somewhat higher
capacities than the equivalent slenderness approach of EC3. For very slender structures shear stresses do
not count and capacities are very similar.

A procedure has been developed for lateral torsional buckling design of continuous beams with
moment hinges. It has been alternatively applied for equivalent slenderness procedure in EC3 and for first
yield criterion on the basis of a geometrically nonlinear theory 2nd order stress analysis including
geometric imperfections.

[1] BTII-Programm, Biegetorsionstheorie II.Ordnung, Friedrich + Lochner GmbH - Software für Statik
und Tragwerksplanung, Version 03/09, Stuttgart/Dresden.
[2] DIN EN 1993-1-1, Eurocode 3: Bemessung und Konstruktion von Stahlbauten, Teil 1-1:
Allgemeine Bemessungsregeln und Regeln für den Hochbau, 2005.
[3] Trahair, N.S., Flexural-Torsional Buckling of Structures, CRC Press Boca Ration, 1993, ISBN
[4] Lindner, J., Stabilisierung von Trägern durch Trapezbleche, Stahlbau 1/1987, Ernst & Sohn
[5] Osterrieder, P., Voigt, M., Saal, H., Vergleichende Betrachtungen zum Biegedrillknicknachweis
nach DIN 18800 Teil 2 (Ausgabe März 1988), Stahlbau 58 (1989), Heft 11, 341 – 347
[6] Osterrieder, P., Kretzschmar, J., First-hinge analysis for lateral buckling design of open thin-walled
steel members, Journal of Constructional Steel Research 62 (2006), pp. 35-43

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010



Benjamin Braun* and Ulrike Kuhlmann*

* Institute for Structural Design, University of Stuttgart, Germany

e-mail: sekretariat@ke.uni-stuttgart.de

Keywords: Stability, plate buckling, interaction, steel structures

Abstract. This paper focuses on the design of steel plated girders under combinations of transverse
loading, bending moment and shear force. In the member states of the European Union the design of
slender steel plates is covered by EN 1993-1-5:2006. Although conclusions from literature show a rather
significant interaction between transverse loading and shear force (F-V), no consideration of this type of
interaction is made in Section 7.2 of EN 1993-1-5:2006. In order to close that gap experimental and
numerical studies were undertaken to analyse the stability behaviour and to identify the influence of key
parameters. Based on that an F-V proposal is developed which is completed by the consideration of a
bending moment so that finally a fully usable F-M-V interaction equation is proposed.

Steel plated structures occur as part of slender structural systems due to their advantageous strength-
to-weight ratio which allows especially aesthetical solutions. Transverse stiffeners are usually provided at
locations where forces are applied locally. However, this is not possible if the position of the load
introduction is transient e.g. in case of bridge girders being incrementally launched or for deep crane
runway beams. In both cases high transverse forces have to be introduced into the slender steel webs of
the girder, often with high bending moment and shear force at the same time, see figure 1.
Although conclusions from literature show a rather significant interaction between transverse loading
and shear force (F-V), no consideration of this type of interaction is made in Section 7.2 of
EN 1993-1-5 [1]. In order to close that gap experimental and numerical studies were undertaken to
analyse the stability behaviour and to identify the influence of key parameters. Based on that an F-V
proposal is developed which is completed by the consideration of a bending moment so that finally a fully
usable F-M-V interaction equation is proposed.
Before current proposals are evaluated and improvements are proposed, considerations on the general
formulation of an interaction equation and on the choice of the verification point are presented in the
following sections.

a) combined loading b) longitudinal stresses c) shear stresses d) transverse stresses

Figure 1: Load combinations of a transversely loaded panel.

Ulrike Kuhlmann and Benjamin Braun


In addition to a development of a design method for a single load case, the formulation of an
interaction equation puts additional difficulties which are:

ƒ Limited number of available data. In general, there is only a limited number of available data
points particularly from experiments because the ratio of loading is added as an important
parameter so that in order to be able to identify the parameter variation within a given load ratio a
similarly large number of specimens should be tested than for a basic load scenario.

ƒ Reference value for basic loading. The reference strength for basic loading coming from
resistance models usually has a variation itself. The interaction equation can therefore only be as
good as the resistance model for the reference strength. It is desirable to know the experimental
ultimate load from basic loading for each interaction test series, otherwise an assumption based on
a resistance model has to be made.

In the following, the interaction between transverse loading and shear force is exemplarily used to
illustrate the aforementioned difficulties and to explain the decisions which were taken in the formulation
of the proposals later on. Figure 2a) shows the F-V interaction with reference strengths based on basic
loadings from experimental and numerical studies in which nothing else than the load parameter was
varied in comparison to the interaction case. In contrast to this figure 2b) shows the same interaction data
but with reference strengths according to EN resistances. It can be shown that the data increasingly
scatters for the EN reference strengths due to the variation of the design model itself. In order to draw a
concise conclusion on the interaction behaviour, it would be necessary to eliminate the effect of the
reference strength's design model. In the F-V study, this was done while studying the effect of parameters
where experimental and numerical reference strengths are also referenced. Thus we can state that a
procedure similar to figure 2a) is better suited to analyse the interaction behaviour.

a) experimental and numerical b) EN reference strengths

reference strengths
Figure 2: Evaluation of interaction data with FR and VR according to experimental and numerical
reference strengths.

Besides that the quality of the reference strength influences the interaction data, so it is not the best
way to evaluate or define an interaction equation based on design models for reference strengths. Imagine
that only the experimental data would be available in figure 2b). In that case only two data points lie in
the quadrant which is relevant for interaction and the interaction equation would be less strict than in the
case when the numerical data are additionally considered. The parameters of the experiments are covered

Ulrike Kuhlmann and Benjamin Braun

rather by the means and upper fractiles of the resistance models which would lead to an underestimation
of interaction. It can be shown that if parameters are chosen such that they cover the lower tail of the
resistance models variation, as done in the numerical simulation, interaction becomes more severe. Of
course, if the definition of interaction equations is coupled to application ranges which restrict the use e.g.
to the parameters of the experiments, interaction may be defined more loose. However, in view of general
applicability and safety, the reference strengths should be based on corresponding experimental and
numerical basic loadings whenever possible.


A plate is usually not subjected to constant stresses but rather to stress gradients. Focusing on the
resistance to transverse loading, it is presumed that the worst case is when the patch loading is placed at
the centerline of the plate. At this location also the bending stress induced by the transverse loading
becomes extremal. For the interaction between transverse loading and shear force, however, there are
basically two choices which reference load can be assigned to each axis of an interaction diagram:
ƒ The applied patch load F is related to the pure patch loading resistance FR and the maximum
internal shear force Vint,max is related to the pure shear resistance VR.
ƒ The applied patch load F is related to the pure patch loading resistance FR and the applied shear
force is corrected by 0.5-times of the applied patch load.
Although the maximum value of the internal shear force can be easily attained from the distribution of
internal forces, its use turned out to be disadvantageous in an interaction diagram because the verification
of the pure patch loading resistance already includes the shear force (=0.5·F) which is induced by the
patch load.
The second approach subdivides the combined loading into the two basic load cases "transverse
loading" and " shear force" which can be composed to create one load combination. Thus, the influence of
shear stresses which are caused by the transverse loading can be better accounted for. Then the
verification point coincides with the one for bending moment. For these reasons the verification point at
the centerline of the transverse loading, i.e. at maximum bending moment and average shear, is taken.


The reference strengths are important parameters and a lot of progress has been made for patch load
resistance models as summarised in [2]. Therefore an evaluation of the interaction equations is performed
not only for EN resistances but also for patch load resistance models which have been recently developed.
By doing this it is assured that the newly developed F-M-V interaction equation is applicable to future
developments. The following advanced patch load resistance models which have been mainly developed
in the frame of the COMBRI research project [3] are compared besides current EN resistances:
ƒ The proposal for girders without longitudinal stiffeners according to Gozzi [4] which follows the
general procedure of current EN 1993-1-5 but which has been further developed with regard to
the yield load and the reduction function.
ƒ The proposal for girders with longitudinal stiffeners according to Davaine [5] which can be used
with current EN 1993-1-5. It has been developed by adding the critical load of the directly loaded
subpanel and by modifiying the reduction curve.
ƒ Another proposal for girders with longitudinal stiffeners according to Clarin [6] which is
harmonised with the improved resistance model for girders without longitudinal stiffeners
according to Gozzi. It also uses the critical load of the directly loaded subpanel but here the
reduction curve for unstiffened cases is also used for girders with longitudinal stiffeners.

Ulrike Kuhlmann and Benjamin Braun

Regardless of girders without and with longitudinal stiffeners the bending moment resistance is
determined according to Section 4 and the shear resistance according to Section 5, both EN 1993-1-5 [1].
In order to evaluate the quality of the proposed interaction equations with regard to the different patch
load resistance models statistical analyses are performed. In the statistical evaluation the test result Re
which can be of experimental or numerical origin is consistently compared to the calculated resistance Rt
of the chosen engineering model under the same load ratio. From a constant load ratio two scalar load
amplification factors can be determined. The quotient of the two scalar load amplification factors
represents a key figure and on that basis a vectorial comparison is carried out for each pair of tested and
calculated resistances.


Based on own experimental and numerical investigations in the frame of the COMBRI research
project [3] a Finite Element model has been established with ANSYS® software [7] and verified with a
good agreement between experimental and numerical results. With this model the complex load paths in
the steel plates were followed and the interaction behaviour for varying F-V load ratio could be described.
The investigations showed that the interaction between transverse loading and shear force is significant.
However, current design standards such as EN 1993-1-5 [1] cover only the interaction between transverse
loading and bending moment. The evaluation of proposed interaction equations from literature led to the
conclusion that the proposals made on the basis of cold-formed trapezoidal beams and hot rolled sections
are not applicable to slender steel plates. On the other hand the interaction equation proposed by Roberts
and Shahabian [8] was approved for short loading lengths ss/hw < 0.25. For longer loading lengths,
however, their interaction equation does not lead to safe results. The lack of a F-V interaction equation in
EN 1993-1-5 and the results from the experimental and numerical studies indicated that for the interaction
between transverse loading and shear force the formulation of an appropriate interaction equation is
required. Following the principles which were set up in sections 2 and 3, interaction equation (1) has been
F §¨ V  0.5 ˜ F ·¸
 d 1. 0 (1)
FR ¨© Vb ,R ¸
FR is the transverse loading resistance of the cross section according to Section 6, EN 1993-1-
5, or according to the advanced resistance models by Gozzi [4], Davaine [5] and Clarin [6];
Vb,R is the shear resistance of the cross section according to Section 5, EN 1993-1-5.

The evaluation of the different resistance models showed that the proposed interaction equation is
safe sided not only for girders without but also with longitudinal stiffeners. The statistical evaluation is
given in table 1. Detailed results of the study can be found in [9].

Table 1: Statistical evaluation of the F-V interaction equation.

Model Mean value Standard deviation Lower 5%-fractile
Gozzi (2007) 1.309 0.184 1.007
Davaine (2005) 1.574 0.139 1.346
Clarin (2007) 1.556 0.126 1.348


Plates under transverse loading are unavoidably subjected to bending moment so that this interaction
has been already addressed in a number of research works which cannot be fully listed here. A

Ulrike Kuhlmann and Benjamin Braun

comprehensive summary of interaction equations can be found e.g. in [10]. The statistical evaluation of
the different interaction equations based on EN reference strengths is given in table 2. The Roberts
proposal is the most conservative one, whereas the Bergfelt proposal is the most favourable one. However
it can be shown that all proposals perform similar and that the trilinear EN approach is simple though
appropriate. And although the EN interaction equation was determined on the basis of EN reference
strengths, it can be shown that it could be further used for welded sections even if the advanced resistance
model of Gozzi is used. However, the objective to propose a single F-M-V interaction equation led to the
development of a F-M interaction equation which can be consistently merged with the F-V proposal, see
equation (1).

Table 2: Statistical evaluation of F-M interaction equations in chronological order.

Model Mean value Standard deviation Lower 5%-fractile
Bergfelt (1971) 1.541 0.264 1.107
Roberts (1981) 1.584 0.237 1.193
Elgaaly (1983) 1.552 0.254 1.134
Ungermann (1990) 1.567 0.239 1.174
Johansson & Lagerqvist (1994) 1.544 0.258 1.120
EN 1993-1-5 (2006) 1.548 0.255 1.128

Following the principles which were set up in sections 2 and 3, an interaction equation based on the
general format according to equation (2) has been developed.
F § M ·
¨ ¸ d 1.0 (2)
FR ¨© M R ¸¹

Equation (2) is fitted as lower bound curve to a small FE study of Davaine and the F-M database
which has been evaluated with the advanced resistance models of Gozzi and Clarin. The comparison is
shown in figures 4 and 5. The difference between both figures is the reference strength which has been
chosen for the bending moment resistance. Data points inside the interaction curve can be disregarded
since they are close to basic loading cases so that their deviation is considered as inherent to the resistance
models for the reference strengths. In figure 4 the reference strengths for the bending moment resistance
MR is based on the relevant cross-section class, i.e. Mpl or Mel. The parameter c is determined and
rounded off to a single decimal place so that c = 5.0. It can be shown that for high levels of bending
moment the interaction curve hardly catches the distribution of data points. For that reason, in a second
step the plastic moment resistance irrespective of the cross-section class was chosen as reference strength,
as it is similarly used in the M-V interaction of Section 7.1, EN 1993-1-5. The parameter c is determined
and rounded off to a single decimal place so that c = 3.6. The results are shown in figure 5. It can be
shown that the data is slightly more homogenouos though hardly perceptible. The statistical evaluation of
both proposals which is given in tables 4 and 5 supports this. In table 3 the results of current EN rules are
given and a comparison shows that in both cases an improvement exists which can be identified by
comparing especially the standard deviation. However, in terms of statistical quality both proposals are
almost identical.
This consistency and the data scatter which is perceived to be slightly more homogenuous leads to the
adoption of the plastic bending moment resistance as reference value. Thus, the consistent definition of
the F-V and F-M interaction equations as continuous function enables the merging of both criteria. The
full F-M-V interaction equation becomes equation (3). In addition the resistance criteria of the basic
loadings according to section 4.6, 5.5 and 6.6, EN 1993-1-5, should be met. The resulting interaction
surface is illustrated in figure 6.

Ulrike Kuhlmann and Benjamin Braun

Figure 4: F-M interaction proposal, MR based on relevant cross-section class

Figure 5: F-M interaction proposal, MR = Mpl,R irrespective of the cross-section class

(here: Mel,R/Mpl,R = 0.88)

Table 3: Statistical evaluation of the F-M interaction equation according to EN 1993-1-5.

Girder type Mean value Standard deviation Lower 5%-fractile
with longitudinal stiffeners 1.548 0.255 1.128
without longitudinal stiffeners 1.598 0.315 1.080

Table 4: Statistical evaluation of the F-M interaction equation with MR = MR,EN and c = 5.0.
Model Mean value Standard deviation Lower 5%-fractile
Gozzi (2007) 1.458 0.229 1.081
Davaine (2005) 1.404 0.204 1.067
Clarin (2007) 1.462 0.243 1.063

Ulrike Kuhlmann and Benjamin Braun

Table 5: Statistical evaluation of the F-M interaction equation with MR = Mpl,R and c = 3.6.
Model Mean value Standard deviation Lower 5%-fractile
Gozzi (2007) 1.466 0.229 1.089
Davaine (2005) 1.414 0.204 1.078
Clarin (2007) 1.466 0.242 1.068

3.6 1.6
F §¨ M ·
§ V  0.5 ˜ F ·
 ¨ ¸ d 1.0 (3)
FR ¨© M pl ,R ¸ ¨ V ¸
¹ © b ,R ¹
FR is the transverse loading resistance of the cross section according to Section 6, EN 1993-1-
5, or according to the advanced resistance models by Gozzi [4], Davaine [5] and Clarin[6];
Mpl,R is the plastic resistance of the cross section consisting of the effective area of the flanges
and the fully effective web irrespective of its section class;
Vb,R is the shear resistance of the cross section according to Section 5, EN 1993-1-5.

Figure 6: F-M-V interaction surface

Based on the research work of the COMBRI research project [3] and beyond [11], evident and
necessary improvements regarding the interaction criteria of steel plates are reported in this paper. At the
beginning thorough considerations on the formulation of interaction criteria and verification points were
made in general. In detail a new interaction equation for the effective width method in case of transverse
loading, bending moment and shear force has been proposed which is summed up below.
First a comparison of the experimental and numerical F-V results with known tests from literature
showed that the interaction between transverse loading and shear force is not negligible. A comparison
with proposals from literature showed that only few approaches exist which do not appropriately describe
the interaction behaviour e.g. with regard to the influence of the long loading lengths. Especially for
longer loading lengths the formulation of a new interaction equation was required, see equation (1). By
choosing the verification point at the centerline of the transverse loading the part of the shear force which

Ulrike Kuhlmann and Benjamin Braun

is induced by the transverse loading and which is already included in the resistance model can be
accounted for. As a result not only the smallest data scatter is found but also a conclusive subdivision of
the interaction case into the basic loadings "transverse loading" and "shear force" is possible which makes
an interaction verification for a transverse loading without additional shear force obsolete.
Plates under transverse loading are unavoidably subjected to bending moment so that this interaction
has already been addressed in a number of research works which have been thoroughly evaluated in this
work. The performance of all proposals is similar and it could be shown that the trilinear EN approach is
simple though appropriate. However, the objective to propose a single F-M-V interaction equation led to
the development of a F-M interaction equation which can be consistently merged with the F-V proposal,
see equation (3). The verification point is naturally chosen at the centerline of the transverse loading
which is the location where also the maximum bending moment occurs.
The new formulation is based on the experimental and numerical data set from own work and from
literature and a statistical evaluation proves the applicability of the equation not only to current EN
resistance models but also to the improved resistance models developed by Gozzi, Davaine and Clarin,
for unstiffened and for longitudinally stiffened girders.

[1] EN 1993-1-5, Eurocode 3: Design of steel structures - Part 1-5: Plated structural elements, 2006.
[2] Kuhlmann, U., Johansson, B., Raoul, J., Braun, B., Gozzi, J., Clarin, M., Davaine, L. and Martin,
P.-O., “A Survey on Patch Loading Models for Bridge Launching”, Proc. of the IABSE Symposium,
Weimar, Germany, 2007.
[3] COMBRI, Competitive Steel and Composite Bridges by Improved Steel Plated Structures, Final
Report, RFCS contract no. RFS-CR-03018, 2007.
[4] Gozzi, J., Patch Loading Resistance of Plated Girders - Ultimate and serviceability limit state,
Doctoral thesis 2007:30, Division of Steel Structures, Luleå University of Technology, Sweden,
[5] Davaine, L., Formulation de la résistance au lancement d’une âme métallique de pont raidie
longitudinalement - Résistance dite de "Patch Loading", Doctoral Thesis, L’Institut National des
Sciences Appliquées de Rennes, France, 2005.
[6] Clarin, M., Plate Buckling Resistance - Patch Loading of Longitudinally Stiffened Webs and Local
Buckling, Doctoral thesis 2007:31, Division of Steel Structures, Luleå University of Technology,
Sweden, 2007.
[7] ANSYS® v10.0, ANSYS Inc., Canonsburg, Pennsylvania, USA.
[8] Roberts, T.M. and Shahabian, F., “Design procedures for combined shear and patch loading”, Proc.
of the Institution of Civil Engineers, Structures & Buildings, 140(August), 219-225, 2005.
[9] Kuhlmann, U., Braun, B. and Johansson, B., “The interaction behaviour of slender plates subjected
to shear force and patch loading”, Proc. of the 5th International Conference on Thin-walled
Structures, Brisbane, Australien, 2008.
[10] Lagerqvist, O., Patch Loading - Resistance of steel girders subjected to concentrated forces,
Doctoral thesis 1994:159 D, Division of Steel Structures, Luleå University of Technology, Sweden,
[11] Braun, B., Design of steel plates under combined loading, Doctoral thesis, Institute for Structural
Design, University of Stuttgart (in preparation).

E. Batista, P. Vellasco, L. de Lima (Eds.)
Rio de Janeiro, Brazil, September 8 - 10, 2010


Rolando Chacón*, Enrique Mirambell* and Esther Real*

*Construction Engineering Department. Universitat Politècnica de Catalunya, Barcelona, Spain

e-mails: rolando.chacon@upc.edu, enrique.mirambell@upc, esther.real@upc.edu

Keywords: Hybrid girders, patch loading, EN1993-1-5.

Abstract. The structural case of hybrid girders subjected to patch loading is treated identically than
the one for homogenous specimens in EN1993-1-5. The EN1993-1-5 formulation is based upon a
theoretical model which consists of a plastic resistance Fy partly reduced by a resistance function χF,
this latter accounting for instability. The EN1993-1-5 formulation predicts that the ultimate load
capacity of steel plate girders subjected to patch loading increases, among others, with the hybrid
grade fyf/fyw. In this work, an appraisal of the theoretical and numerical responses of hybrid and
homogenous specimens subjected to patch loading is presented. Some peculiarities concerning the
treatment of hybrid girders are pinpointed though. It is numerically demonstrated that the moment
capacity of the flanges (and thus, fyf/fyw) does not play any role in the resistance of girders predicted in
EN1993-1-5. Accordingly, a design proposal which enhances the current formulation is provided.

A girder is deemed as being hybrid when it is fabricated with different steel strengths for the flange
and web panels. In hybrid design, the nominal yield strength of one or both flanges is larger than the
nominal yield strength of the web. This type of girder is popular as the girder yields a greater flexural
capacity at lower cost and weight compared to a homogeneous girder [1]. On the other hand, patch
loading phenomena has been widely analyzed since the early sixties. Experimental and theoretical
analyses have pinpointed the typical failure mechanisms of girders subjected to patch loading and
consequently ultimate load predictions are nowadays available [2]-[4]. Broadly speaking, two
magnitudes have been given to describe the resistance of members subjected to this sort of loads. The
former defines a plastic resistance Fy of the member whereas the latter, an elastic critical load Fcr. The
former has been generally obtained by limit analysis whereas the latter, by theoretical formulae properly
calibrated with numerical simulations. The factual situations to which these members are subjected lie
inside a blurred transition between yielding and instability. It is well known that the root square of the
ratio between the plastic resistance Fy and the elastic critical load Fcr is commonly referred to as the
slenderness parameter. Admittedly, there exists a direct relation between this slenderness and the actual
failure mode. This relation has been labeled in the European guidelines for the design of plated
structural elements EN1993-1-5 [5] as the resistance function. The patch loading phenomena has been
harmonized to this procedure.
Despite the vast amount of research devoted separately to both hybrid girders and the patch loading
field, the research work that matches both subjects is rather scant [6]-[8]. In this paper, the resistance of
hybrid steel plate girders subjected to patch loading is dealt with simultaneously. Theoretically, it is
demonstrated that the EN1993-1-5 formulation predicts the resistance of plate girders subjected to patch
loading as a monotonic increasing function with, among other parameters, the flange yield strength fyf
(and consequently, with the hybrid grade fyf/fyw). Numerically, a vast study aimed at comparing
numerical and theoretical results is presented. In this study, it is found that the results obtained with the

Rolando Chacón et al.

EN1993-1-5 provisions do not reproduce satisfactorily the trends obtained numerically. Alternatively, a
design proposal in accordance with the procedure implemented in EN19931-5 and aimed at correcting
the aforementioned anomaly is proposed.


2.1 Hybrid girders

The European design rules EN1993-1-5 overtly consider the hybrid girder usage. These rules
recommend a maximum value of Φh=fyf/fyw<2,0. The treatment of hybrid girders is identical to one of
homogeneous prototypes except for the following remarks. First, as the resistance to direct stresses of
plate girders is calculated by using the effective area of the cross-section, in the particular case of hybrid
design, fyf must be used in determining the effective area of the web for plate buckling purposes.
Second, for the particular case of hybrid plate girders it is indicated that the potential yielding of the
web must be taken into account in direct stresses verifications.
2.2 Patch loading
Regarding the verification to patch loading, the general approach currently included in EN1993-1-5
is based upon a plastic resistance Fy which is partially reduced by means of the resistance function χF.
The plastic resistance is based upon the mechanical model suggested in [2]. In eq. (1), ly is the yield-
prone effectively loaded length and is calculated from geometrical and mechanical properties of the
girders by using equations (2), being “ly” not greater than the distance between adjacent transverse
stiffeners “a”. χF takes instability into account by means of equations (3) and (4) and can be