See discussions, stats, and author profiles for this publication at:

https://www.researchgate.net/publication/221185084

Sequential and Distributed

Evolutionary Computations in
Structural Optimization

Conference Paper in Lecture Notes in Computer Science · January 2004

DOI: 10.1007/978-3-540-24844-6_167 · Source: DBLP

CITATIONS READS

16 12

5 authors, including:

Tadeusz Burczynski Arkadiusz Poteralski

Silesian University of Tec… Silesian University of Tec…
54 PUBLICATIONS 297 34 PUBLICATIONS 114
CITATIONS CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these

related projects:

Numerical analysis and optimization of materials with complex

properties and microstructure View project

Intelligent design of 2D nanostructures based on molybdenum

View project

All content following this page was uploaded by Arkadiusz Poteralski on 27 December 2016.

The user has requested enhancement of the downloaded file.

 Sequential and Distributed Evolutionary
Computations in Structural Optimization

Tadeusz Burczyński1,2 , Wac
law Kuś1 , Adam D
lugosz1 , Arkadiusz Poteralski1 ,

and Miros
law Szczepanik1
1
Department for Strength of Materials and Computational Mechanics,
Silesian University of Technology, Konarskiego 18a, 44-100 Glwice, Poland
[email protected],
2
Institute of Computer Modelling, Cracow University of Technology
Cracow, Poland

Abstract. The aim of the paper is to present the application of the

sequential and distributed evolutionary algorithms to selected structural
optimization problems. The coupling of evolutionary algorithms with
the finite element method and the boundary element method creates a
computational intelligence technique that is very suitable in computer
aided optimal design. Several numerical examples for shape, topology
and material optimization are presented.

1 Introduction
Evolutionary methods have found various applications in mechanics, especially in
structural optimization [2], [5]. The main feature of such applications is the fact
that design process of artificial systems like structural or mechanical components
is simulated by biological processes based on heredity principles (genetics) and
the natural selection (the theory of evolution). The paper is devoted to structural
optimization using sequential and distributed evolutionary algorithms. Solutions
of optimization problems are very time consuming when sequential evolutionary
algorithms are applied. The long time of computations is due to the fitness
function evaluation which requires solution of direct (boundary-value or initial
boundary-value) problems. The fitness function is computed with the use of the
boundary element method (BEM) or the finite method (FEM) [8]. In order to
speed up evolutionary optimization the distributed evolutionary algorithms can
be proposed instead of the sequential evolutionary algorithms [4]. This paper
is extension of previous papers devoted to optimization using sequential and
distributed evolutionary algorithms in thermoelastic problems [3],[7], shape and
topology optimization of 2D and 3D structures [6],[10].

2 Formulation of the Evolutionary Design

A body, which occupies a domain Ω bounded by a boundary Γ , is considered.
Boundary conditions in the form of the displacement and traction fields are pre-
scribed. In the case of dynamical problems initial conditions are also prescribed.

L. Rutkowski et al. (Eds.): ICAISC 2004, LNAI 3070, pp. 1069–1074, 2004.
c Springer-Verlag Berlin Heidelberg 2004


1070 T. Burczyński et al.

One should find optimal shape or topology of the structure by minimizing an

objective functional:
minJo (Ch) (1)
Ch

with imposed constraints:

Jα (Ch) = 0, α = 1, 2, .., A; Jβ ≤ 0, β = 1, 2, .., B (2)

Ch is a vector of design parameters which is represented by a chromosome with

floating point representation

Ch = [g1 , g2 , .., gi , .., gN ] (3)

where restrictions on genes are imposed in the form

giL ≤ gi ≤ giR , i = 1, 2, .., N (4)

Genes are responsible for shape, topology and material parameters of the
structures. General form of the objective functional Jo and performance func-
tional Jα and Jβ can be expressed in structural optimization as follows



J = Ψ (σ,ε,u, T ) dΩ + ϕ(u, p, T, q)dΓ (5)
Ω Γ

where Ψ is an arbitrary function of stress σ, strain ε, displacement u and tem-

perature T fields in the domain Ω, respectively, ϕ is an arbitrary function of
displacement u, traction p, temperature T and heat flux q fields on the bound-
ary Γ , respectively.
Using the penalty function method the optimization problem (1) and (2) is
transformed into non-constrained problem and the fitness function consists of
functionals Jo , Jα and Jβ . In order to evaluate the fitness function one should
solve the boundary-value problem using FEM or BEM.

3 Shape, Topology, and Material Parametrization by

Genes

The geometry of the structure is specified by NURBS - Non Uniform Rational

B-Spline. Co-ordinates of control points of the NURBS play the role of genes.
The distribution of material properties as Young’s modulus E (x) , x ∈ Ω in
the structure is describing by a surface W (x) , x ∈ H 2 (for 2-D) or a hyper
surface W (x) , x ∈ H 3 (for 3-D). W (x) isstretched under H d ⊂ E d , (d = 2, 3)
and the domain Ω is included in H d , i.e. Ω ⊆ H d . The shape of the surface
(hyper surface)W (x) is controlled by genes gi , i = 1, 2, .., N, which create the
chromosome (3). Genes take values of the function W (x) in interpolation nodes
xj , i.e. gj = W (xj ) , j = 1, 2, ..., N . If the structure is discretized by FEM
 Sequential and Distributed Evolutionary Computations 1071

the assignation of Young’s moduli to each finite element is performed by the

mapping [6]:
Ee = W (xe ) , xe ∈ Ωe , e = 1, 2, ..., R (6)
It means that each finite element contains different material. When the value
of Young’s modulus E for the e-th finite element is included in the interval
0 ≤ Ee < Emin , the finite element is eliminated. Otherwise the finite element
remains having the value of the Young’s modulus from this material.

4 Distributed Evolutionary Algorithms

Distributed evolutionary algorithms [1],[4],[11] are based on the theory of co-
evolutionary algorithms. The process of evolution is faster, if isolated subpopu-
lations of small number of interchangeable individuals evolve. In this algorithm
a population of individuals is divided into several subpopulations. Each of the
subpopulations evolves separately, and from time to time only a migration phase
occurs, during which a part of individuals is interchanged between the subpopu-
lations. The distributed evolutionary algorithm works as a few isolated sequential
evolutionary algorithms [9] communicating between each other during migration
phases. The evolutionary optimization is performed in a few steps (Figure 1).

Fig. 1. The distributed evolutionary algorithm (one subpopulation)

At the beginning the starting population of chromosomes is generated ran-

domly. The floating point representation is used. The population is divided into
M subpopulations. Then the fitness function values for every chromosome are
computed. The evolutionary algorithm operators as the crossover (simple, arith-
metical and heuristic) and the mutation (uniform, boundary, non-uniform and
Gaussian) are applied next. When the migration phase occurs, some chromo-
somes from subpopulations migrate to other subpopulations. The topology of
1072 T. Burczyński et al.

the migration decides between which subpopulations the migration occurs. The
elementary parameters defining the migration are: the migration frequency, the
number of migrating individuals, the cloning or the eliminating an emigrating
individual from a subpopulation, the selection of emigrating individuals, the way
of introducing immigrating individuals to a subpopulation, the topology of the
connections among subpopulations. The migration frequency defines the number
of generations after which an individual emigrates. When the migration is too
frequent, other subpopulations may be dominated by the solutions achieved in
the current subpopulation. If it is too rare, in an extreme situation subpopulati-
ons will be isolated and they will evolve independently. The number of migrating
individuals defines which part of a population will emigrate. If this number is too
big, the algorithm starts to behave like a sequentional algorithm with a single
population. The ranking selection creates the offspring subpopulation based on
the parent subpopulation modified by evolutionary operators. When selection is
performed immigrated chromosomes from other subpopulations are also consi-
dered. The next iteration is performed if the stop condition is not fulfilled. The
end computing condition can be expressed as the maximum number of iterations
or the best chromosome fitness function value. The speedup, which is a measure
of increasing of distributed computations, is defined as: ko = ttn1 where tn me-
ans the time of performing the computation when n processing units (usually
processors) are used, and t1 - the computation time on a single unit.

5 Numerical Examples of Evolutionary Design

5.1 Shape and Topology Evolutionary Design of Thermomechanical
Structures
A square plate with a circular void is considered (Figure 2a). For the sake of
the symmetry only a quarter of the structure is taken into consideration. The
considered quarter of the structure contains a gap bounded by an unknown in-
ternal boundary shown in the Figure 2b. The values of the boundary conditions
are: T10 = 3000 C, T20 = 200 C, q0 = 0, p0 = 100kN/m, α1 = 1000W/m2 K,
α2 = 20W/m2 K, u0 = 0. The model consists of 90 boundary elements. The
optimization problem consists in searching an optimal: a) shape of the inter-
nal boundary, b) width of the gap, c) distribution of the temperature T 0 ∗ on
the internal boundary for minimization of the radial displacements given by the
2n
functional (5) with Ψ = 0 and ϕ = (u/uo ) , where u is a field of boundary
displacements where tractions p0 are prescribed, u0 is a reference displacement,
n is natural number. Shape of the internal boundary was modeled using NURBS
curve which consists of 7 control points, whereas width of the gap and tempera-
ture T 0 ∗ using NURBS curve consist of 6 control points (Figure 2c). For the sake
of the symmetry along line AB (Figure 2c) the total number of design parameters
was equal to 13. The range of the variability of each control point for the width
of the gap is between 0.2 and 0.8, whereas for the temperature is between 50 C
and 800 C. The number of subpopulations was 2 and the number of chromosomes
in each subpopulation was 10. Figure 2d shows results of the optimization.
 Sequential and Distributed Evolutionary Computations 1073

a) b)

c) d)

Fig. 2. Square plate with circular void: a) geometry, b) boundary conditions, c) NURBS
curves, d) best result

a) b)

c) d)

Fig. 3. Evolutionary design of a bicycle frame: a) plate and boundary conditions, best
result, b) proposed frame; Evolutionary design of 3D structure: c) geometry, d) best
result

5.2 Shape, Topology, and Material Evolutionary Optimization

Two numerical examples are considered for this kind of evolutionary design.
The first example (Figure 3a,b) refers to evolutionary design of 2-D structures:
(a bicycle frame - plain stress). The structures is discretized by triangular finite
elements and subjected to the volume constraints. The second example (Figure
3c,d) is the 3-D problem where the body is discretized by hexahedral finite
elements and subjected to the stress constraint. Using proposed method, material
properties of finite elements are changing evolutionally and some of them are
eliminated. As a result the optimal shape, topology and material or thickness of
the structures are obtained.
1074 T. Burczyński et al.

6 Conclusions
An effective tool of evolutionary design of structures has been presented. Using
this approach the evolutionary process of creating the shape, topology and ma-
terial is performed simultaneously. The important feature of this approach is its
great flexibility for 2-D and 3-D problems and the strong probability of finding
the global optimal solutions.

Acknowledgements. This research was carried out in the framework of the

KBN grant no. 4T11F00822.

References
1. Aleander J.T., An Indexed Bibliography of Distributed Genetic Algorithms, Uni-
versity of Vaasa, Report 94-1-PARA, Vaasa, Finland, (2000).
2. Burczyński T., Beluch W., D
lugosz A., Kuś W., Nowakowski M. and Orantek P.:
Evolutionary computation in optimization and identification. Computer Assisted
Mechanics and Engineering Sciences, Vol.9, No. 1 2002, pp. 3-20.
3. Burczyński, T., D
lugosz, A., Evolutionary optimization in thermoelastic problems
using the boundary element method. Computational Mechanics, Vol. 28, No. 3-4,
Springer 2002.
4. Burczyński, T., Kuś, W., Distributed evolutionary algorithms in optimization of
nonlinear solids. In: Evolutionary Methods in Mechanics (eds. T. Burczyński and
A. Osyczka). Kluwer, 2004.
5. Burczyński T., Osyczka A. (eds): Evolutionary Methods in Mechanics, Kluwer,
Dordrecht 2004.
6. Burczyński T., Poteralski A., Szczepanik M., Genetic generation of 2D and 3D
structures. Computational Fluid and Solid Mechanics (ed. K.J. Bathe), Elsevier,
Amsterdam 2003, pp. 2221-2225.
7. D
lugosz, A., Evolutionary computation in thermoelastic problems. In: Evolutio-
nary Methods in Mechanics (eds. Burczynski, T. and Osyczka, A.), Kluwer, 2003.
8. Kleiber, M., (ed.), Handbook of Computational Solid Mechanics, Springer, Berlin,
1998.
9. Michalewicz Z.: Genetic Algorithms + Data Structures = Evolutionary Algorithms,
Springer-Verlag, Berlin 1996.
10. Poteralski A., Burczyński T., Szczepanik M., Evolutionary optimisation of 3-D
structures. Proc. CMM 2003 15th International Conference on Computer Methods
in Mechanics, Wis
la, 2003.
11. Tanese R.: Distributed Genetic Algorithms. Proc. 3rd ICGA, pp.434-439, Ed. J.D.
Schaffer. San Mateo, USA, (1989).

View publication stats