Académique Documents
Professionnel Documents
Culture Documents
Abstract A new algorithm is presented for size opti- Previous works dealing with the optimal design of
mization of truss structures with any kind of smooth ob- truss structures considering elastic constraints were pub-
jectives and constraints, together with constraints on the lished by Schmidt and Miura (1976), Kirsch (1981), All-
collapse loading obtained by limit analysis, for several wood and Chang (1984), Rozvany and Zhou (1991),
loading conditions. The main difficulty of this problem Ohasaki(2001), Gil and Andreu (2001), among others.
is the fact that the collapse loading is a nonsmooth func- Literature concerning optimal design involving plastic
tion of the design variables. In this paper we avoid non- constraints is limited, however we can still mention the
smooth optimization techniques based on the fact that classical work by Hodge and Prager(1951) and also Lee
limit analysis constraints are linear by parts. Our ap- and Gordon (1981) and Kirsch(1981), Lellep and Puman
proach is based on a Feasible Directions Interior Point (2001), Lellep and Tungel (2002), Kaliszky and Lógó
Algorithm for nonlinear constrained optimization. Three (2002). The combination of simultaneous constraints of
illustrative examples are discussed. The numerical results elasticity and plasticity was discussed extensively over
show that the calculation effort when limit analysis con- thirty years ago, Rozvany and Cohn (1970). We also men-
straints are included is only slightly increased with re- tion the work of Muralhidar and Rao(1997) on topology
spect to classic constraints. optimization. The work by Rohan and Whiteman(2000)
is also interesting because it is applied to both trusses
Key words Limit Analysis Constraints, Truss Opti- and continua.
mization. For statically determinate structures, the first yield
condition is the ultimate capacity of the structure. How-
ever, for structures with multiple degrees of indetermi-
nacy, the collapse loading is normally much higher than
the loading that produces the first yield condition. The
ultimate capacity of a structure has become an impor-
1 tant issue in truss structural design. Limit analysis is an
Introduction alternative analytical procedure for obtaining the ulti-
mate loading of a structure for collapse. It determines
This work presents a new method for the automatic opti- the maximum safety factor, or factor of loading ampli-
mal design of truss structures, to achieve the minimum of fication, that can be supported by a structure of ideal
a given objective function. The design variables represent elastoplastic materials subjected to a specified external
the cross-sectional areas of bars or groups of bars. The loading, Drucker et al (1951),Prager and Hodge(1951),
truss may be subject to multiple load cases. The struc- Kirsch(1981), Feijóo and Zouain (1988, 1989), Zouain et
tural geometry is given and remains fixed (unchanged). al. (1993). In optimal design, constraints on the safety
factors with respect to the plastic collapse can then be
Received: date / Revised version: date introduced. As it is already known, the plastic design op-
J.Romero1 , P.C. Mappa2 , J. Herskovits2 and C.M.Mota timization problem can be formulated as a Linear Pro-
Soares3 gram, Kirsch(1981).
1 In this paper we consider the optimal design of struc-
Departamento de Engenharia Mecânica, CT - UFES,
tures with simultaneous constraints on smooth functions,
Vitória, ES, Brazil
2
Programma de Engenharia Mecânica, COPPE-UFRJ, like the elements stresses or nodal displacement, and on
caixa postal 68503, 21945-970, Rio de Janeiro, Brazil. the safety factor in relation to the plastic collapse. The
jose@optimize.ufrj.br main difficulty of this problem is the fact that the critical
3
IDMEC/IST - Instituto Superior Técnico, Lisbon, Portu- load is a nonsmooth function of the design variables.
gal We show that, in the present problem, the limit anal-
ysis constraints are linear by parts. In consequence, they
2
can be substituted by a set of linear constraints. The a boolean matrix relating the bars with the design vari-
resulting problem is solved with an Interior Point Al- ables.
gorithm for smooth nonlinear optimization. Some test Introducing now the design variables in the Limit
problems are solved in a very efficient way, Romero(1991). Analysis formulation, we have the collapse amplification
In the next section the Limit Analysis technique is factor as a function of x:
described. Then, the sensitivity of the critical load to
changes in the structure is discussed. In the section 4
max α
α, T
we describe the optimization problem as well as the lin- ᾱ(x) = (2)
ear program obtained in the particular case when only
such that: B T T − αP = 0
T − M Lx ≤ 0
limit analysis constraint are considered. The numerical
optimization procedure is presented in Section 5. Some
numerical results are presented in the following section
3
and finally we make our concluding remarks.
Sensitivity Study of Limit Analysis
design variables can also be included. This problem can paper we present a procedure to overcome this difficulty.
be expressed as a mathematical program as follows:
Let be I ≡ {I1 , I2 , ..., Ir } the set of all possible col-
lapse mechanism of the structure for different values of
minimize f (x) x. We denote by ᾱl (x) the function that represents the
x collapse amplification factor for the collapse mechanism
(5)
subject to: g(x) ≤ 0 Il . Then, in Problem (5) the limit analysis constraint can
αad − ᾱ(x) ≤ 0 be substituted by the set of constraints:
where f : <n → < is the objective and g : <n → <m
represent the inequality smooth constraints, including αad − ᾱl (x) ≤ 0, for l = 1, 2, .., r.
the side bounds.
These constraints are linear and their derivatives are
Since the feasible stresses for the limit analysis can be
given by (4). In practice, we do not include all this con-
taken less severe than the one for the elastic analysis, the
straints. At each iterate xk a limit analysis is carried out,
present approach gives lighter designs when compared
to obtain ᾱ(xk ) and the corresponding collapse mecha-
with the classical elastic methodology with displacement
nism I(xk ). If this is the first time that this mechanism is
and stress constraints.
obtained, then a new constraint αad − ᾱl (x) ≤ 0 is added
to the already existing set of constraints of the problem.
In the particular case when f (x) is the structural
The linearized constraints are stored in the function
weight, there is one loading condition and only the Limit
Analysis constraint is considered, Problem (5) is equiva-
lent to: h(x) ≡ [(αad − ᾱ1 (x)), (αad − ᾱ2 (x)), ..., (αad − ᾱl (x))],
We present now a numerical technique, see Romero(1991), (i) Compute (d0 , λ0 , φ0 ) by solving the linear system
to solve Problem (5) based on the Feasible Directions In-
terior Point Algorithm (FDIPA) described in Herskovits
(1998) and Herskovits(1995). FDIPA makes iterations in Bd0 + ∇g(xk )λ0 + ∇h(xk )φ0 = −∇f (xk ), (7)
the primal and dual variables to solve Karush-Kuhn-
Tucker first-order optimality conditions. At each itera- Λ∇g T (xk )d0 + G(xk )λ0 = 0. (8)
tion, a descent direction is defined by solving a linear
system. In a second stage, the linear system is perturbed Φ∇hT (xk )d0 + H(xk )φ0 = 0. (9)
so as to deflect the descent direction and obtain a fea-
sible descent direction. A line search is then performed If d0 = 0, stop.
to get a new interior point and to ensure global conver-
gence. There are first order, Newton, and quasi Newton (ii) Compute (d1 , λ1 , φ1 ) by solving the linear system
versions of FDIPA.
Bd1 + ∇g(xk )λ1 + ∇h(xk )φ1 = 0, (10)
The Limit Analysis constraints are nonsmooth and
FDIPA is an algorithm for smooth optimization. In this Λ∇g T (xk )d1 + G(xk )λ1 = −λ, (11)
4
360'' 360''
ure 4. The active constraints correspond to stresses and
5 1 3 2 1
displacements in case 1), to critical loadings and elas-
8 10 tic displacements in case 2) and to elastic stresses and
displacements and also critical loadings in case 3).
5 6 360''
7 9
6
3 4 2
4
p p
3000
2800
2600
objective
2400
500
Case 1
Case 2
Table 5 72 Bars Truss - Optimal Design - Cross-Sections
450 Case 3
areas in in2
400
350
250
1 1-4 0.6370 0.4438 0.6152
2 5- 12 0.3411 0.2864 0.3061
200
3 13-16 0.2152 0.2194 0.2131
150
4 17 18 0.2784 0.3173 0.3014
100 5 19-22 0.6886 0.7855 0.6280
6 23- 30 0.2459 0.2571 0.2544
50
0 2 4 6 8 10 12 14
7 31 - 34 0.1000 0.1000 0.1000
8 35 36 0.1000 0.1000 0.1000
Fig. 4 Iteration History for 25-Bars Truss 9 37- 40 0.7129 0.6249 0.6547
10 41 - 48 0.2349 0.2551 0.2526
11 49-52 0.1000 0.1000 0.1000
the bars are defined by sixteen design variables. The min- 12 53 54 0.100 0.1000 0.1000
imum value of the variables is of 0.1in2 . The yield stress 13 55 - 58 1.0352 0.9418 0.9470
is σγ = ± 25000 psi. The maximum value of nodal dis- 14 59- 66 0.2356 0.2557 0.2535
placements is 0.5in for all nodes in all directions. The 15 67 - 70 0.1000 0.1000 0.1000
optimal designs are shown in Table 5. The active con- 16 71-72 0.1000 0.1000 0.1000
straints correspond to stresses and displacements in case Weight (lb) 232.516 226.148 227.805
1), to critical loadings and elastic displacements in case ηad 3.5 - 3.2
2) and to elastic stresses and displacements and also crit- αad - 3.5 3.5
ical loadings in case 3).
550
Case 1
Case 2
Case 3
500
450
400
objctive
350
300
250
200
0 2 4 6 8 10 12 14 16 18
iterations
other. The choice of which method should be used in the Kaliszky, S.; Lógó, J. 2002:Plastic behaviour and sta-
project will depend on the conditions of operation of the bility constraints in the shakedown analysis and optimal
structure and the reliability requested in the project. The design of trusses. Structural and Multidisciplinary Opti-
algorithm used to solve the present problem behaves in mization,24, 118-124.
a very efficient way in the different examples explored.
Kirsch, U. 1981: Optimum Structural Design. McGraw-
Hill.
Acknowledgments
Lee,J.; Gordon, J. 1981: Transmission tower design anal-
ysis and design by linear programming. IEE. Transaction
This research was partially supported by CNPq (Brazil- on power Apparatus and systems, 100(4), 1989-2007.
ian Research Council), FAPERJ (Rio de Janeiro State
Research Council) and ICCTI(Portuguese Institute for Lellep, J.; Puman, E. 2001: Optimization of conical shells
International Scientific and Technological Cooperation). for Mises material. Structural and Multidisciplinary Op-
timization,22, 149-156.
References
Lellep, J.; Tungel, E. 2002:Optimization of plastic spher-
Allwood, R.; Chung, Y. 1984: Minimum weight design ical shells with a central hole.Structural and Multidisci-
of trusses by an optimality criteria methods. Int. J. for plinary Optimization,23, 233-240.
Num. meth. in Eng. 20, 697-713.
Lluı́s, G.; Andreu, A. 2001: Shape and cross-section op-
Cristiansen, E. 1981: Computation of limit loads. Int. timization of a truss structure.Computers an Structures,
J. for Num. meth. in Eng. 17,1547-1570. 79,681-689.
Drucker; D.C.; Greenberg, H.J.; Prager, W 1951: The Luenberger,G.D. 1984: Linear and nonlinear program-
safety factor of an elastic-plastic body in plane strain, J. ming. Second edition, London: Addison-Wesley Publish-
App. Mech. 18, 371-378. ing Company.
Feijóo, R.A.; Zouain, N. 1988: Formulation in rates and Ohsaki, M. 2001: Random Search method based on exact
increments for elastic-plastic analysis. Int. J. for Num. reanalysis for topology optimization of trusses with dis-
Meth. in Eng. 26,2031-2048. crete cross-section areas.Computers and Structures, 79,
673-679.
Feijóo, R.A.; Zouain, N. 1989: A contribution to struc-
tural plasticity by optimization techniques. Comp. plast. Prager, W. ; Hodge, P.G.1951: Theory of perfectly plas-
models soft. and appl. Ed. Owen, D.R.J. and Oñate. tic solids, John Wiley & Sons, New York.
Haftka,R.T.;Gurdal,Z.; Kamat, M.P. 1990: Elements of Rohan, E ; Whiteman, JR. 2000: Shape optimization of
Structural Optimization, 2nd Ed.,London: Kluwer Pub- elastic-plastic structures and continua. Computer Meth-
lishers. ods in Applied Mechanics and Engineering,187, Issues1-
2, 261-288.
Herskovits, J. 1995: A view on nonlinear optimization.
In Herskovits, J (ed.) Advances in Structural Optimiza- Romero,J. 1991: Projeto otimo de treliças restrições ao
tion; 71-116, London: Kluwer Academic Publishers. colapso plático. M.Sc. Thesis, COPPE, UFRJ, Brazil(in
Portuguese).
Herskovits, J. 1998: A Feasible Directions Interior Point
Technique for Nonlinear Optimization. JOTA - Jour- Rozvany, G.I.N.; Cohn, M. 1970: Lower bound optimal
nal of Optimization Theory and Applications; 99-1, 121- design of concret structures. J.Eng Mech. Div. ASCE,96,
146. 1013-1030.
Herskovits, J.; Santos, G. 1998: Feasible Arc Interior Rozvany, G.I.N.;Zouh, M. 1991: A note on truss design
Point Algorithms for Nonlinear Optimization. Fourth for stress and displacement constraints by the optimality
World Congress on Computational Mechanics; in CD- criteria method. Structural Optimization, 3, 45-50.
ROM , Argentina.
Shmith, L.; Miura, H. 1982: An Advanced structural anal-
Hodge, P.G.; Prager W. 1952: Limit design of reinforce- ysis/synthesis capability-access 1.AIAA Journal,29(2):1
ments of cut-outs in slabs. Bulletin of the American Math- 934-939.
ematical Society, 58, 68.
Sobieszcanski-Sobieski, J.; Berthelemy,J.F.,Riley K.M.
8