Beruflich Dokumente
Kultur Dokumente
net/publication/261503088
CITATIONS READS
0 27
2 authors, including:
Yin Li
Concordia University Montreal
2 PUBLICATIONS 0 CITATIONS
SEE PROFILE
All content following this page was uploaded by Yin Li on 15 March 2016.
$EVWUDFW²7KLV SDSHU SUHVHQWV D K\EULG JHQHWLF DOJRULWKP be needed since they do not guarantee a convergence toward
*$ IRU WKH RSWLPL]DWLRQ RI VWUXFWXUDO VXSSRUW ORFDWLRQV 7KH global optima.
REMHFWLYHIXQFWLRQFRQVLGHUHGLVHLWKHUWKHPD[LPXPGHIOHFWLRQRU
PD[LPXP EHQGLQJ PRPHQW RI WKH VWUXFWXUH 7KH REMHFWLYH Two types of non-gradient methods, namely the sub-
IXQFWLRQLVPLQLPL]HGVXEMHFWHGWRWKHGHVLJQFRQVWUDLQWVWKDWDUH problem approximation method and the genetic algorithm (GA)
LPSRVHG DV LQHTXDOLW\ FRQVWUDLQWV RQ WKH GHVLJQ YDULDEOHV 7KH method, have been established as an attempt to overcome the
SURSRVHG DOJRULWKP LQWHJUDWHV WKH FRQFHSWV RI WKH *$ PHWKRG above-mentioned problems found in the gradient-based
DQG WKH ILQLWH HOHPHQW PHWKRG $ UHDOFRGHG PHWKRG LV XVHG WR methods. The sub-problem approximation method requires
UHDOLVWLFDOO\ UHSUHVHQW WKH YDOXHV RI WKH GHVLJQ YDULDEOHV 7KUHH only the values of the dependent variables, and not their
*$ RSHUDWRUV FRQVLVWLQJ RI WKH QRUPDOL]HG JHRPHWULF UDQNLQJ derivatives. In other words, the calculation of the sensitivities
VHOHFWLRQ SURFHGXUH WKH DULWKPHWLF FURVVRYHU DQG WKH QRQ is not required to optimize the support locations. Thus, this
XQLIRUPPXWDWLRQDUHSURSRVHG)LQLWHHOHPHQWPHWKRGLVXVHGWR method is more convenient and more efficient than the
FRPSXWHWKHYDOXHVRILPSOLFLWREMHFWLYHIXQFWLRQV7KHUHVXOWVRI gradient-based methods. However, this method may not be as
WZR QXPHULFDO H[DPSOHV RQH E\ PLQLPL]LQJ WKH PD[LPXP accurate as the gradient-based methods. The weakness of the
GHIOHFWLRQ DQG WKH RWKHU E\ PLQLPL]LQJ WKH PD[LPXP EHQGLQJ sub-problem approximation method is discussed in Section 5.
PRPHQW RI WKH VWUXFWXUH LQGLFDWH WKDW WKH SURSRVHG PHWKRG LV
DFFXUDWH DQG FRPSXWDWLRQDOO\ HIILFLHQW LQ RSWLPL]LQJ VXSSRUWLQJ A genetic algorithm is a search/optimization technique
ORFDWLRQV based on the survival of the fittest theory and natural selection
proposed by Darwin. The genetic algorithm has the
.H\ZRUGV²*HQHWLF DOJRULWKP 6WUXFWXUDO VXSSRUW ORFDWLRQ advantages that it is able to deal with discrete optimum design
)LQLWH HOHPHQW PHWKRG 0D[LPXP EHQGLQJ PRPHQW 0D[LPXP problems, does not need to know the derivatives of objective
GHIOHFWLRQ functions and has the capability of identifying global optimal
values of the objective function. Due to these advantages, the
I. INTRODUCTION GA method has found numerous applications in the optimized
design of structures. A paper by Goldberg and Samtani [8]
The optimization of structural support locations has been examined the application of the GA method to the
the subject of extensive studies during the past few decades. optimization design of structures. More recent development of
This topic is of great importance to improve structural the GA method can be found in papers by Rajeev and
performance such as in reducing the maximum deflection or Krishnamoorthy [9], Pezeshk et al. [10], Saka and Kameshki
bending moment [1-3], increasing the fundamental natural [11], Yun and Kim [12], among others. Although much work
frequency [4], and increasing the buckling loads of structures has been performed in this area, the application of the GA to
[5]. Various methods have been developed for determining the the optimization of structural support locations is quite limited.
optimal locations of supports in a structure by minimizing/ An attempt was made to apply GA to find optimal support
maximizing an objective function subjected to certain location of beams [4]. In their paper, the main aim is to
constraints. These methods may be broadly divided into two determine the optimal support location in a beam, with the
categories as: (1) gradient-based methods; and (2) non- objective of maximizing the fundamental natural frequency.
gradient methods. The problem was solved by a simple genetic algorithm with
.Gradient-based methods are based on a heuristic search reproduction, crossover, and mutation operators. The results
strategy [6]. The first order optimization method and obtained from GA are reasonably accurate. However, there is
evolutionary shift method are two representative examples of room for further improvement in the efficiency, which is a
such methods. Extensive reviews of these two methods are very important issue when this method is applied to a complex
found in [2, 7]. However, these methods involve the design structure.
sensitivity analysis, and therefore require a relatively long So far, to the knowledge of the authors, the GA has not
computation time. Manual intervention to the programs may been applied to the optimization of structural support locations,
978-1-4673-5851-4/13/$31.00 2013
c IEEE 1
with the objective of minimizing the maximum deflection or III. SOLUTION OF OPTIMAL SUPPORT LOCATION PROBLEM
bending moment. In addition, there is no comprehensive study For illustrative purposes, two solution methods for the
carried out to assess the relative performance of the GA and the
optimal support location problem mentioned previously are
above-mentioned two alternative methods (the first order
optimization method and sub-problem approximation method) briefly reviewed in this section. One is the sub-problem
for solving the optimal support location problem. The approximation method and the other is the first order
parameters of the GA may affect the optimal results. However, optimization method. These two methods have been already
there is no guidance regarding how to select appropriate values integrated into a multi-physics commercial finite element code
for the parameters. The objective of this paper is to apply GA as optimization tools [7]. More detailed description of the two
method to solve the optimal support location problem, with the methods can be found in [7].
objective of minimizing the maximum deflection or bending The sub-problem approximation method can be used to
moment. Its accuracy, efficiency and advantages over above-
provide a rapid solution to the optimal support location
mentioned two alternative methods namely the first order
problem, without using the derivatives of objective functions.
optimization method and sub-problem approximation method,
will be discussed in detail. This method has three major steps. Firstly, the objective
function expressed by Eq. (1) or Eq.(3) is approximated using
least squares fitting of a number of design sets generated
II. PROBLEM STATEMENT OF OPTIMAL SUPPORT LOCATIONS randomly. Secondly, the constrained optimal support location
The optimal support location problem is to find a design problem described above is converted into an unconstrained
variable vector representing support locations in order to form by using penalty functions. Finally, a sequential
minimize a structural performance function. This function may unconstrained minimization technique is used to solve the
be the maximum defection or maximum bending moment of unconstrained optimal support location problem. The
the structure subjected to the design constraints that are algorithm proceeds iteratively until convergence is achieved,
imposed as inequality constraints on the design variables. or when the maximum number of iterations is reached in the
In this paper, we have considered two optimal support presented study.
location problems. One is to minimize the maximum defection
Different from the sub-problem approximation method,
of structures and the other is to minimize the maximum
the first order optimization method needs to know the
bending moment of structures. Of course, other optimal
support location problem such as maximizing the fundamental derivatives of the objective functions. Instead of using least
natural frequency of structures can also be considered. The square fitting, the constrained optimal support location problem
two optimal support location problems may be described above is directly transformed into an unconstrained
correspondingly stated as follows: form. Various steepest descent and conjugate direction
searches are performed during each iteration until convergence
(1) For the optimal support location problem of is reached. Each iteration is composed of sub-iterations that
minimizing the maximum defection of structures include search direction and derivatives computations.
Minimize max ( į L ( ; ) ), L 1,..., N
IV. THE PROPOSED SOLUTION METHOD FOR OPTIMAL
SUPPORT LOCATION PROBLEM
subjected to constraints / M d ; M d 8 M , M 1,..., Q
When the objective functions in Eqs (1) and (3) are
where GL (; ) is the absolute deflection of the L th node of explicitly expressed in terms of the design variables, the
optimal support location problem can be solved easily using a
the structural system; ; is the design variable vector conventional GA. Unfortunately, the objective functions in Eqs.
representing the support locations; ; M is the nodal (1) and (3) are usually an implicit function of the design
variables. In other words, the explicit expressions for the
coordinates of the finite element model specifying the M th
objective functions in Eqs. (1) and (3) are not available. The
location of support; / M and 8 M are the lower and upper solution of the optimal support location problem is then not
straightforward. To overcome the problem of using a
bounds of the location of support, respectively. N and Q are
conventional GA, a hybrid GA is developed. In the proposed
the total number of nodes and the total numbers of supports in
the finite element model, respectively. method, the finite element method (FEM) is used to compute
(2) For the optimal support location problem of the value of implicit objective functions and the optimal
minimizing the maximum bending moment of structures support location problem can then conveniently be solved by
using the conventional GA. Some important elements of the
Minimize max ( 0 L ( ; ) ), L 1,..., N proposed method are briefly discussed in the following sections.