International Journal of Solids and Structures 42 (2005) 5848–5871

www.elsevier.com/locate/ijsolstr

Benchmark case studies in structural design optimization

using the force method
R. Sedaghati *

Department of Mechanical and Industrial Engineering, Concordia University, 1455 de Maisonneuve Blvd. West,
Montreal, Que., Canada H3G 1M8

Received 28 October 2004

Available online 14 April 2005

Abstract

A structural optimization algorithm is developed for truss and beam structures under stress–displacement or fre-
quency constraints. The algorithm combines the mathematical programming based on the Sequential Quadratic Pro-
gramming (SQP) technique and the finite element technique based on the Integrated Force Method. A new
approach based on the single value decomposition technique has been developed to derive the compatibility matrix
required in the force method. Benchmark case studies illustrate the procedure and allow the results obtained to be com-
pared with those reported in the literature. It is shown that the computational effort required by the force method is
significantly lower than that of the displacement method and in some cases such as structural optimization problems
with multiple frequency constraints, the analysis procedure (force or displacement method) significantly affects the final
optimum design and the structural optimization based on the force method may result in a lighter design.

2005 Elsevier Ltd. All rights reserved.

Keywords: Structural design; Optimization; Finite element force method

1. Introduction

Equilibrium equations combined with the compatibility conditions are generally fundamental to analysis
methods in structural mechanics. The equilibrium equations are basically balance of elemental forces. For
statically determinate structures, it is well known that the equilibrium equations alone, expressed in terms
of forces, are adequate to calculate unknown elemental forces. However, equilibrium equations are not suf-
ficient to solve general structural analysis problems, as they have to be augmented by the compatibility

*
Tel.: +1 514 848 7971; fax: +1 514 848 8635.
E-mail address: [email protected]

0020-7683/$ - see front matter
2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijsolstr.2005.03.030
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5849

conditions. In other words, equilibrium equations are indeterminate in nature, and determinacy for a con-
tinuum is achieved by adding the compatibility conditions. If the equilibrium equations are written in terms
of nodal displacements, the compatibility conditions can be indirectly satisfied and the number of equations
and the displacement unknowns are identical. This is basically the essence of displacement method.
A structure in the force method of analysis can be designated as structure (n, m), where (n, m) are the
force and displacement degrees of freedoms (fof, dof), respectively. The n component force vector F must
satisfy the m equilibrium equations along with r = (n
m) compatibility conditions. If n = m, the structure
is determinate and its analysis is trivial, i.e. the equilibrium equations are sufficient to find the element
forces. The emphasis here is on the analysis of indeterminate structure for which n > m. There are at present
two main force formulations, the Classical Force Method (CFM) and the Integrated Force Method (IFM).
Both the CFM and IFM use the same equilibrium equations. The equilibrium matrix is a (m · n) banded
rectangular matrix, which is independent of the material properties and design parameters of the indeter-
minate structure (n, m). The equilibrium matrix can be easily assembled from elemental equilibrium matri-
ces using the finite element analysis. In CFM first equilibrium equation is satisfied and then using
compatibility conditions, the r redundant forces will be obtained. In the Classical Force Method, the com-
patibility conditions are generated by splitting the structure (n, m) into a determinate basis structure (m, m)
and r redundant members. The compatibility conditions are written in the redundant members by establish-
ing the continuity of deformations between the r redundant members and the basis structure (m, m) for the
external loads, thus the redundant members are the primal variables of the compatibility conditions in the
CFM. This procedure was originally developed by Navier (Timoshenko, 1953) for the analysis of indeter-
minate trusses.
Patnaik (1986) and Patnaik et al. (1991) developed the IFM method. In IFM, the compatibility matrix is
obtained by extending St. VenantÕs principle of elasticity strain formulation to discrete structural mechanics
(Patnaik and Joseph, 1986). Both equilibrium equations and compatibility conditions are satisfied simulta-
neously. The compatibility conditions are generated without any recourse to redundant members and the
basis determinate structure.
Structural analysis and optimization algorithms developed in recent years have generally been based on
the displacement method(Venkayya, 1978; Canfield et al., 1988; Mohr, 1992; Mohr, 1994; Flurry and Sch-
mit, 1980; Haftka and Gurdal, 1992). Commercial finite element programs are based on the displacement
method and very few investigations have been reported in structural optimization using the finite element
force method. In the present study, the linear analysis based on the Integrated Force Method has been used
to analyze and optimize the truss and beam structures under stress, displacement and frequency constraints.
It is intended to investigate the efficiency of the force method in the structural optimization of the truss and
beam structures, under displacement, stress and frequency constraints, by solving the equilibrium and com-
patibility equations simultaneously. A direct method has been used to generate the compatibility matrix for
indeterminate structures, which, is based on the displacement–deformation relationship and the Singular
Value Decomposition (SVD) technique without the need to select the consistent redundant members.
The equilibrium matrix is also generated automatically through the finite element analysis.
In most recent works, reported in literatures, the optimization algorithms were mainly based on the opti-
mality criterion technique because of its computational efficiency. For example Saka (1984), Khan (1984),
Khot (1984), Venkayya and Tischler (1983), Grandhi and Venkayya (1988), Konzelman (1986), Khan and
Willmert (1981), McGee and Phan (1991), Khot and Kamat (1985), Saka and Ulker (1992), and Sedaghati
and Tabarrok (2000) employed optimality criteria methods in minimization of the weight of truss and beam
structures under stress, displacement, frequency or stability constraints. Modern optimality criterion algo-
rithms would involve the case of satisfying the multiple constraints (scaling) and Karush–Kuhn–Tuker
(KKT) condition (resizing) alternatively. However when the cross-sectional area and principal moment
of inertia are nonlinearly related (frame structures), the scaling procedures, normally used in the optimality
criterion methods, are approximate in nature and the scaling itself needs an iteration procedure. In this
5850 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

study the powerful nonlinear mathematical programming method, based on the Sequential Quadratic
Programming (SQP) technique, has been utilized as the optimization algorithm to find the true optimum
solution and the results have been compared with those obtained using optimality criterion technique.
The application and efficiency of the proposed methodology is illustrated by minimizing the weight of
different benchmark truss and beam structures under displacement, stress and frequency constraints. It
is shown that by using either the force or displacement method, as an analyzer does not affect the final opti-
mum solutions of the problems with stress and displacement constraints. However, the force method is
more computationally efficient than the displacement method. Moreover it has been demonstrated that
using force or displacement method as an analyzer may affect the final optimum solution in the problems
under multiple frequency constraints and the force method may cause lighter design. Finally it is found that
using Sequential Quadratic Programming method as the optimiser may result lighter design in comparison
to the optimality criterion technique commonly used in the literature.
In the following sections, a short description of the structural analysis using the force method is pre-
sented and subsequently the optimization algorithm is explained. Finally, the application of the algorithm
is illustrated by structural optimizing of different benchmark case studies.

2. Structural analysis using the force method

A discrete finite element structure with m and n displacement and force degrees of freedom, respectively
has m equilibrium equations and r = n
m compatibility conditions. In static problems the equilibrium
equations in the displacement formulation can be written as
KU ¼ P ð1Þ
where K is the system stiffness matrix of the structure (obtained by assembling the stiffness matrices of the
individual elements), U is the nodal displacement vector and P is the external applied load vector. The com-
patibility conditions have been satisfied implicitly during the generation of Eq. (1). The equivalent form of
Eq. (1) in the integrated force formulation can be written as (Patnaik, 1986; Patnaik et al., 1991)
SF ¼ P ð2Þ
where F is the element force vector. The matrix S and vector P* can be obtained through the combination
of the m equilibrium equations as
QF ¼ P ð3Þ
and r compatibility equations as
CD ¼ 0 ð4Þ
where the element deformation vector D can be related to the element force vector F in accordance to
D ¼ GF ð5Þ
thus
2 3 2 3
Q P
6 7 6 7
S ¼ 4    5; P ¼ 4    5 ð6Þ
C G 0
where Q, C and G are the (m · n) equilibrium matrix, (r · n) compatibility matrix and the (n · n) flexibility
matrix, respectively. One should note that the matrices Q, C and G are banded and they have full-row ranks
of m, r and n, respectively. The matrices Q and C also depend on the geometry of the structure and
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5851

therefore, are independent of the material properties. For a finite element idealization, the generation of the
equilibrium matrix Q and the flexibility matrix G is straightforward and can be obtained automatically.
However the automatic generation of the compatibility matrix C is a laborious task in the standard force
method. Moreover, the generation of C in the Integrated Force Method is based on the elimination of the m
displacement degrees of freedom from the n elemental deformations.
The displacement–deformation relationship for the discrete structures can be obtained by equating the
internal strain energy and the external work and can be explained as
D ¼ QT U ð7Þ
Eq. (7) relates the n deformations to the mnodal displacement degrees of freedom and hence, the r = n
m
compatibility equations can be arrived through the elimination of the m nodal displacements from the n
deformations. To obtain the compatibility matrix, one may express the nodal displacements in terms of
the member deformations by using Eq. (7) as

1 pinv
U ¼ ðQQT Þ QD ¼ ðQT Þ D ð8Þ
T pinv T
where the matrix (Q ) denotes the Moore–Penrose pseudo-inverse of Q . Considering Eqs. (7) and (8),
we may have
BD ¼ 0 ð9Þ
where
B ¼ ½I n
QT ðQT Þpinv  ð10Þ
where In is the (n · n) identity matrix. Eq. (9) is similar to the compatibility equations given by Eq. (4).
However, the matrix B is a (n · n) matrix with rank of r inferring that the rows of matrix B are dependent
on each other. In order to extract the (r · n) compatibility matrix C form the matrix B, i.e. to reduce the
matrix B to matrix C, the Singular Value Decomposition (SVD) method has been employed (Golub and
Van Loan, 1996). By applying the SVD method to B, we obtain
B ¼ RRT T ð11Þ
where R and T are the (n · n) orthogonal matrices and
" #
K 0
R¼ ð12Þ
0 0 nn
with K ¼ diagf n1 n2  nr g, where n1 P n2 P    P nr > 0 are positive numbers. It follows that


C
B¼R ð13Þ
0
Therefore the (r · n) compatibility matrix C can be represented by
T
C ¼ K½ T1 T2    Ti  Tr  ð14Þ
where the vector Ti denotes the ith column of the matrix T.
Although Eq. (7) is quite adequate to determine the element deformations using the nodal displacements,
however it is not sufficient to obtain the nodal displacements using the element deformations or forces since
the redundant structures are represented by the rectangular equilibrium matrix Q with no inverse. This im-
plies that the compatibility equations should be merged with the equilibrium equations. For this reason,
using S instead of Q in Eq. (7) and solving for the nodal displacements U, we obtain
U ¼ JD or U ¼ JGF ð15Þ
5852 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

where
J ¼ m rows of S
T ð16Þ
In frequency analysis problems, the equations of the motion in the displacement formulation can be written
as
€ þ KU ¼ 0
MU ð17Þ
€ is acceleration vector. Considering Eq. (15) and noting that
where M is the stiffness matrix of system and U
KU in the displacement method is equivalent to SF in the force method, Eq. (17) may be written as
€ þ SF ¼ 0
M F ð18Þ
where
2 3
f ¼ MJG
M
6 7
M ¼ 4  5 ð19Þ
0
Eq. (18) represents the frequency equation in the framework of the force formulation. In free vibration
analysis, it is assumed that element forces are harmonics in time, F ¼ F sinðxtÞ, where x and F are circular
frequency and force mode shape, respectively. Considering this, Eq. (18) can be written as
SF
x2 M  F ¼ 0 ð20Þ
The natural frequencies can be obtained easily from Eq. (20) by using an eigenvalue extraction algo-
rithm. To overcome some computational difficulties during the analysis, the (n · n) system of equations ex-
pressed by Eq. (20) can be reduced to a (m · m) system by taking advantage of the null matrices. To obtain
this, the matrices in Eq. (20) are partitioned into the redundant and basis determinate structure as
2 3
" #" # f mm M
M f rr " #
S mm S mr Fm 6 7 Fm

x2 6
4  5
7 ¼0 ð21Þ
S rm S rr Fr Fr
0
or
f mm Fm þ M
S mm Fm þ S mr Fr
x2 ð M f rr Fr Þ ¼ 0
ð22Þ
S rm Fm þ S rr Fr ¼ 0
Eliminating of Fr from the (n · n) system of equations in Eq. (22) results in the reduced (m · m) subsys-
tem as
2 f f
1
ðS mm
S mr S
1
rr S rm ÞFm
x ð M mm
M rr S rr S rm ÞFm ¼ 0 ð23Þ
and
Fr ¼ S
1
rr S rm Fm ð24Þ
The solution of the reduced eigenvalue problem expressed by Eq. (23) gives all the eigenvalues, whereas
both Eqs. (23) and (24) are used to calculate the force eigenvectors. Once the force mode shapes are deter-
mined, the displacement mode shapes can be generated by using Eq. (15).
It is noted that the selection of the redundant members is not unique and there are multiple combina-
tions of redundant and basis determinate structure for an indeterminate structure. However the redundant
members should be selected so that the remaining determinate structure is not a mechanism. This consistent
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5853

set of redundant members will ensure the existence of the inverse of Srr. If the equilibrium equations in
Eq. (3) is rewritten in the following form:
Qr Fr þ Qm Fm ¼ P ð25Þ
the selection of the consistent set of redundant members and basis determinate structure is such that the
rank of the matrix Qm is equal to m. The violation of this condition makes the matrix Qm singular. Here,
a robust technique based on the Gauss elimination technique and work of Robinson (1965) has been devel-
oped and applied to automatically identify the consistent set of redundant members and basis determinate
structure. The proposed technique is outlined as follows:
I. Augment the equilibrium matrix Q with the external load P as ½ Q P ; II. Select one of the nonzero
elements in the first row of the augmented matrix and divide all the elements in this row by this number; III.
Multiply the first row by the coefficient of the corresponding element in the second row (if it is not zero) and
subtracted from the second row; IV. Continue this procedure for each of the remaining rows; V. The col-
umn corresponding to that element has now a one in the first row and zeros in all other rows; VI. Repeat
the same process from steps II through IV, in turn for the remaining rows until either all of the rows are
exhausted or all of the remaining rows have all zeros as elements; VII. All the m unit columns are indepen-
dent and they correspond to the basis determinate structure. The remaining columns correspond to the con-
sistent redundant members.
As mentioned before, consistent redundant members selected are not unique since the redundancy is
dependent on the order in which the equations are generated and by the selection of the nonzero element
in each row when applying the Gaussian elimination procedure.

4. Optimization algorithm

The optimization problem for truss and beam type of structures can be defined mathematically as min-
imizing the structural mass represented as
( )
X
Ne
Min MðAÞ ¼ qi Li Ai ð26Þ
i¼1

subject to the (Nr + NU) stress and the displacement constraints (behaviour constraints)
gi ðAÞ ¼j ri =

ri j
1 6 0 i ¼ 1; . . . ; N r
ð27Þ
gj ðAÞ ¼j U j =U j j
1 6 0 j ¼ 1; . . . ; N U
or subject to Nx natural frequency constraints (behavior constraints)
~ 2k
x2k 6 0 k ¼ 1; 2; 3; . . . ; N x
gk ðAÞ ¼ x ð28Þ
and the NA side constraints on the design variables
Ai
Ai 6 0 i ¼ 1; 2; 3; . . . ; N A ð29Þ
where qi, Ai and Li are the density, the cross-sectional area and the length of the ith element, respectively. M
is the total mass of the structure and Ai is the lower limit on the ith design variable. Furthermore, ri, r
i are
the stresses in the ith element and its allowable limit value, respectively, and Uj, U j are the constrained dis-
placement on the jth degrees of freedom and its allowable limit value. xk and x ~ k are the kth natural fre-
quency and its specified value, respectively.
It should be pointed out that the analysis and optimization are two separate modules. The equilibrium
and compatibility equations in the force method, Eq. (2), and the frequency equation, Eq. (21), are satisfied
5854 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

directly during the finite element analysis (analysis module) and then the results are passed to the optimi-
zation module. Thus it is not necessary to take into account Eq. (2) or Eq. (21) as the equality constraints
inside the optimization algorithm.
In this study, the Sequential Quadratic Programming (SQP) method has been applied to solve the opti-
mization problem discussed above. The implementation of the SQP method was performed in MATLAB
(Coleman et al., 1999). Based on the work done by Powell (1978), the method allows one to closely mimic
the NewtonÕs method for the constraint optimization just as it is done for the unconstraint optimization.
SQP is indirectly based on the solution of the KKT conditions. Given the problem description in Eqs.
(26)–(29), the principal idea is the formulation of a QP sub-problem based on a quadratic approximation
of the Lagrangian function as follows:
XNt
LagðA; kÞ ¼ MðAÞ þ ki  gi ðAÞ ð30Þ
i¼1

where Nt is the total number of constraints. At each major iteration of the SQP method a QP sub-problem
is solved. The solution to the QP sub-problem produces an estimate of the Lagrange multipliers, ki, and a
search direction vector dm in each iteration m, which is used to form a new iteration as
Amþ1 ¼ Am þ am dm ð31Þ
m
The step length parameter a is determined through a one-dimensional minimization in order to produce
a sufficient decrease in the merit function. At the end of the one-dimensional minimization, the Hessian of
the Lagrangian, required for the solution of the next positive definite quadratic programming problem, is
updated using the Broyden–Fletchet–Goldfarb–Shanno (BFGS) formula as
qðmÞ qT ðmÞ H T ðmÞ dðmÞ dT ðmÞ H ðmÞ
H ðmþ1Þ ¼ H ðmÞ þ
ð32Þ
qT ðmÞ dm dT ðmÞ H ðmÞ dðmÞ
where
dm ¼ Amþ1
Am ð33Þ

qm ¼ rLagðAmþ1 ; kmþ1 Þ
rLagðAm ; km Þ ð34Þ

m+1
and H is the approximation of the Hessian of ÔLagÕ at A .
Powell (1978) recommends keeping the Hessian positive definite even though it may be positive indefinite
at the solution point. To guarantee that the updated Hessian matrix Hm+1 remains positive definite, he sug-
gests replacing qm by hqm + (1
h)Hmdm where qm is given by Eq. (34), and h is determined by
8
<1 if dTv qv P 0.2dTv Hdv
h¼ 0.8dTv Hdv ð35Þ
: Tv v Tv v if dTv qv < 0.2dTv Hdv
0.8d Hd
d q
It should be noted that the stress–displacement and frequency gradient functions, in Eqs. (27) and (28),
are not both smooth and convex functions, thus the local optimum result may be achieved using the gra-
dient-based algorithms such as the above SQP algorithm. In this study, several randomly generated initial
points have been used for the SQP algorithm in order to make sure that the optimal solution is either a
global solution or very close to the global solution.

5. Benchmark case studies

A brief description of the test cases are outlined in Table 1.

 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5855

Table 1
Benchmark Case Studies on structural optimization under stress, displacement and frequency constraints

(continued on next page)

5856 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

Table 1 (continued)
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5857

5.1. Stress and displacement constraints

Five examples on the analysis of the planar and space truss and frame structures shown in Figs. 1–5 (see
Table 1) illustrate the proposed procedure and allow the results to be compared with those reported in lit-
eratures. All information regarding the material and geometrical characteristics, applied external loads, side
constraints (minimum and maximum cross-sectional areas) and behaviour constraints (stress–displacement
constraints) are provided in the relative figures. It is intended to show that in the structural optimization
problems, with the stress and displacement constrains, the analysis procedure (either the force method
or the displacement one) does not affect the final optimum design. Furthermore, it is to establish the fact
that the design optimization procedure based on the force method is more efficient than the displacement
method.

5.1.1. The 10-bar planar truss structure

The 10-bar planar truss is shown in Fig. 1 (see Table 1). This structural design problem is a classical
example in the literature and has been extensively investigated using linear analysis based on the displace-
ment method. The number of displacement degrees of freedom is m = 8 and the number of force degrees of
freedom n = 10. Thus, the number of redundancy is r = 2.
A minimum mass of 2299.65 kg (5069.85 lbm) was obtained using both the displacement method (DM)
and the force method (FM). The initial cross-sectional area for all the elements is 9677.4 mm2 (15 in.2),
which is an infeasible guess. The horizontal negative displacement constraint at node 1 and the tension
stress constraint in element 5 are active at optimum point. The final results are tabulated in Table 2.
The iteration history is shown in Fig. 9. The CPU time required by the FM was significantly lower than
the DM, thus confirming the efficiency of the FM. The cross-sectional areas in elements 2, 5 and 10 reached
to their minimums in both analyses. It is interesting to note that both the stress constraint and the cross-
sectional area are active in element 5. The results obtained match exactly the results reported by Flurry and
Schmit (1980) and Haftka and Gurdal (1992), who solved the problem using the dual method and approx-
imation concepts in linear analysis based on the displacement method.

Table 2
The final design solution for the cross-sectional areas (in.2) for the 10-bar planar truss
Member DM FM
1 30.5218 30.5218
2 0.1000 0.1000
3 23.1999 23.1999
4 15.2229 15.2229
5 0.1000 0.1000
6 0.5514 0.5514
7 7.4572 7.4572
8 21.0364 21.0364
9 21.5284 21.5284
10 0.1000 0.1000
Mass (lbm) 5060.85 5060.85
No. of iterations 237 237
No. of A.C.a 5 5
CPU time (s) 11.24 4.34
a
A.C.: Active Contraints.
5858 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

Fig. 9. Iteration history for the 10-bar planar truss using the force and displacement methods.

9000
Initial areas =5 in2 ; Infeasible guess
8000 Initial areas =15 in2 ; Infeasible guess
Initial areas =20 in2 ; Feasible guess

7000

6000
Mass (lbm)

5000

4000

3000

2000

1000
0 50 100 150 200 250 300 350 400
Iteration

Fig. 10. Iteration history for different initial areas for the10-bar planar truss.

The problem was also solved using different initial cross-sectional areas for the elements. The results ob-
tained were exactly the same as in the previous case. The iteration history for three different initial cross
sectional areas (using the force method) is shown in Fig. 10.

5.1.2. The 25-bar space truss structure

The 25-bar space truss structure, shown in Fig. 2 (see Table 1), has previously been investigated by Saka
using the linear analysis based on the displacement method and with the optimality criterion approach
(Saka, 1984). The structure has identical symmetries about the X–Z and Y–Z planes, so that the design var-
iable linking is used to impose symmetry on the structure and hence, only eight design variables are iden-
tified here. The structure is subjected to single load case as shown in Table 3.
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5859

Table 3
Nodal load components (N) for the 25-bar space truss structure
Node Coordinate directions
X Y Z
1 80000 120000 30000
2 60000 100000 30000
3 30000 0 0
6 30000 0 0

The allowable compressive stress, rac, is determined in accordance to the AISC codes (AISC, 1978) in
which
(
p2 E=S 2R for S R > C c
rac ¼ 2 2
ð36Þ
rat ð1
0.5S R =C c Þ for S R < C c
where the slender ratio ofpeach member is SR = L/RG (L is the length and RG is the radius of gyration for
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
each member) and C c ¼ 2p2 E=rat (rat is the allowable tension stress). Thus, the value of the allowable
compressive stress varies during the optimization process. All the members have the pipe-type cross-sec-
tions with SR = aAb, where A is the cross-sectional area in square centimeter and the constants a and b
are selected as 0.4993 and 0.6777, respectively. The number of degrees of freedom for the displacement
is m = 18, and that of the force is n = 25. Therefore, the number of the redundancy is found to be r = 7.
Without linking the design variables, the number of the design variables is 25 and, the number of the con-
straints is 54. On the other hand, by linking the design-variables into eight groups, the number of the design
variables reduces to 8, and the number of the constraints would change to 20.
A minimum value of the mass of 649.7 kg (1432.3 lbm) was obtained using both the displacement meth-
od (DM) and the force method (FM). The final results for both displacement and force methods are pre-
sented in Table 4 with their iteration histories illustrated in Fig. 11. The initial cross-sectional area for all
the elements is chosen as 2000 mm2 (3.1 in.2). The CPU time required for the FM is significantly lower than
that of the DM indicating its superior efficiency. It ahs been realized that the compressive stress constraint
in the elements 1, 2, 6, 10, 13, 16, 18 and 24 (one member from each group was selected) are found to be
active.

Table 4
Final design solutions for the cross-sectional areas (mm2) for the 25-bar space truss structure
Design variables Member DM FM
1 1 232.7 232.7
2 2–5 1150.6 1150.6
3 6–9 895.1 895.1
4 10,11 230.4 230.4
5 12,13 223.3 223.3
6 14–17 1018.4 1018.4
7 18–21 950.2 950.2
8 22–25 1443.5 1443.5
Mass (kg) 649.7 649.7
No. of iterations 316 424
No. of A.C.a 8 8
CPU time (s) 50.64 16.31
a
A.C.: Active Constraints.
5860 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

Fig. 11. Iteration histories for the 25-bar space truss structure using the displacement and force methods.

1400
Initial areas=2000 mm2 ; Feasible guess
1200 Initial areas=500 mm2 ; Infeasible guess

1000
Mass (kg)

800

600

400

200
0 200 400 600 800 1000
Iteration

Fig. 12. Iteration histories for different initial areas for the 25-bar space truss.

To confirm that the global optimum has been caught, the problem was again solved, using different ini-
tial values of the cross-sectional areas for all the elements, and results were found to be exactly the same as
for the previous case. The iteration histories for two different initial values of the cross-sectional area, using
the force method, is shown in Fig. 12.
A minimum value of the weight of 921 kg (2030.5 lbm) has been reported obtained by Saka, using the
displacement method combined with the optimality criterion based only on satisfying the displacement con-
straints (Saka, 1984). The stress constraints were satisfied through the stress-ratio technique in the linear
analysis. Flurry and Schmit (1980) have also solved this problem by using the dual method and the approx-
imation concept and analysis based on the displacement method. The structure was also analyzed using the
data provided by Fleury and Schmit and identical results were obtained.
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5861

5.1.3. The 72-bar space truss structure

The 72-bar space truss structure, shown in Fig. 3 (see Table 1), is a relatively large size problem. The
structure is subjected to two different loading inputs. In the first case, only the node 1 is subjected to a pull
load of 22.25 kN (5000 lbf) in the X- and Y-directions and a push load of
22.25 kN (
5000 lbf) in the Z-
direction. However, for the second case, all the nodes of 1–4 are subjected to a push load of
22.25 kN
(
5000 lbf) in the Z-direction. The number of displacement degrees of freedom is m = 48, and the number
of the force degrees of freedom is n = 72, and hence, the number of the redundancy will be r = 24. Without
linking together the design variables, the number of the design variables is 72 and the number of the
constraints is found to be 152. However, by linking the design variables into 16 groups, the number of
the design variables becomes 16, and the associated number of the constraints reduces to 40.
The minimum value of the mass, for both the DM and the FM, was obtained as 172.20 kg (379.615 lbm).
The final results are presented in Table 5 and the iteration histories are illustrated in Fig. 13. The required
computational time for the FM is significantly lower than the DM, illustrating the efficiency of the FM over
the DM for the analysis of large size problems. A closer examination of the results reveals that the nodal
displacement constraints at node 1 in the X- and Y-directions for the first load case, and the stress con-
straints in the elements 1–4 for the second load case are active. The cross-sectional areas in groups 7, 8,
11, 12, 15 and 16 reached their minima in both analyses. The optimum results matches exactly with the
solution reported by Flurry and Schmit (1980), who solved the problem using the dual method and the
approximation concepts based on the displacement method. Furthermore, the problem was once again
solved when the design variables were not linked together. The results using the FM and the DM were
found to be almost identical and the computational time for the FM was approximately one half of that
required by the DM. The optimum value of the mass was reduced to 131 kg (288.8 lbm), demonstrating
that when the symmetry is not imposed on the structure a significantly lower value of the mass can be
obtained for the final design. In this case the number of active constraints was 47. The displacement
constraints at node 1 in both X- and Y-directions for the first load case, and the stress constraints in the

Table 5
Final design solutions for the cross-sectional areas (mm2) for the 72-bar space truss
Design variable Members DM FM
1 1–4 100.97 100.97
2 5–12 352.00 352.00
3 13–16 264.77 264.77
4 17,18 367.55 367.55
5 19–22 337.87 337.87
6 23–30 333.61 333.61
7 31–34 64.52 64.516
8 35,36 64.52 64.516
9 37–40 818.32 818.32
10 41–48 330.13 330.13
11 49–52 64.52 64.52
12 53,54 64.52 64.52
13 55–58 1216.90 1216.90
14 59–66 330.52 330.52
15 67–70 64.52 64.52
16 71,72 64.52 64.52
Mass (kg) 172.20 172.20
No. of iterations 556 557
No. of A.C. 9 9
CPU time (s) 274.23 107.10
5862 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

Fig. 13. Iteration histories for the 72-bar space truss structure for both the force and displacement methods with variable linking.

members 1, 2, 4 and 19 for the second load case were found to be active. The cross-sectional areas in the
elements 5, 8, 9, 12 to16, 18, 24, 25, 28, 29, 31 to 36, 38, 40, 41, 48 to 54 and 56 were reached to their minima
and the iteration histories for this case is illustrated in Fig. 14.

5.1.4. The 10-member frame (two-story and two-bay)

The 10-member frame structure consists of three stories and two bays and illustrated in Fig. 4 (see Table
1). The initial cross-sectional area was set as 16129 mm2 (25 in.2) that being the same for all the elements.
The following empirical relationships were used for the area A, the section modulus S, and the moment of
inertia I (Khan, 1984).

Fig. 14. Iteration histories for the 72-bar space truss structure with no variables linking.
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5863
"
S ¼ 1.6634A1.511
0 6 A 6 15
I ¼ 4.592A2
"
S ¼ ð281.077A2 þ 84100Þ0.5
290 ð37Þ
15 < A 6 44
I ¼ 4.638A2


S ¼ 13.761A
103.906
44 < A 6 100
I ¼ 256.229A
2300
where A is the area measured in square inches. The above relationship is stated for a steel section in accor-
dance to the AISC code (AISC, 1978).
A minimum weight of 3307.23 kg (7291.19 lbm) is obtained. Final design solutions for both displacement
and force method are tabulated in Table 6. The horizontal displacement constraint at node 4 and the stress
constraints on element 6 are identified as active. The horizontal displacement at node 3 is close to being
active and the cross-sectional area of the elements 3, 4 and 10 reached their minimum size. The results were
compared with those reported in the literature. As an example, Khan (1984) used a displacement based lin-
ear analysis with the optimality criterion technique and has obtained a minimum value of the weight
3391.87 kg (7477.79 lbm) with the horizontal displacement of the nodes 3 and 4 being active constraints
(no active stress constraint). This problem has also been solved using the CONMIN code and a minimum
value of the weight 3969.97 kg (8752.29 lbm) is reported (Vanderplaats, 1973).
The computational time for the force method was significantly lower than that required by the displace-
ment method, again pointing out the efficiency of the force method when applying the force method to the
frame-type structures.

5.1.5. The 25-member frame (three-story and three-bay)

The 25-member frame structure, shown in Fig. 5 (see Table 1), corresponds to a three stories and three
bays structure. The numerical values of the material properties and the stress limit and the relationship
between the cross-sectional area, section modulus and the moment of inertia are all the same as those men-
tioned in example 5.1.4. The minimum value of the weight is obtained to be 9508.32 kg (20962.26 lbm). The
final results are tabulated in Table 7. The horizontal displacement constraints at the nodes 2 and 10, as well
as the stress constraints on the elements 1, 2, 3, 5, 9, 12, 14 and 17, are active both in the linear and non-
linear analysis. The cross-sectional areas for the frame elements 6, 11, 13, 15, 18, 20, 24 and 25 reached their

Table 6
Final design solutions for the cross-sectional areas (mm2) for the 10-member frame structure
Members DM FM
1 28387 28387
2 23682 23682
3 3226 3226
4 3226 3226
5 46255 46255
6 10241 10241
7 7236 7236
8 16433 16433
9 16243 16243
10 3226 3226
Mass (kg) 3307.23 3307.23
No. of iterations 620 608
No. of A.C. 5 5
CPU time (s) 63.52 20.60
5864 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

Table 7
Final design results for the cross-sectional areas (mm2) for the 25-member frame structure
Members Linear analysis
DM FM
1 10007 10007
2 7146 7147
3 4233 4233
4 16210 16210
5 20247 20246
6 3226 3226
7 51757 51749
8 14878 14885
9 31429 31429
10 53870 53862
11 64516 64516
12 20581 20581
13 3226 3226
14 19581 19581
15 3226 3226
16 4826 4832
17 13153 13154
18 3226 3226
19 16165 16168
20 3226 3226
21 18751 18754
22 13546 13556
23 46754 46739
24 3226 3226
25 3226 3226
Mass (kg) 9508.32 9508.32
No. of iterations 1849 1665
No. of A.C. 18 18
CPU time (s) 479.93 299.50

minimum size. This problem was also solved by Khan (1984) using the displacement based linear analysis
and the optimality criterion technique, having obtained a minimum weight of 10049.77 kg (22155.95 lbm)
with just the horizontal displacement at nodes 2 and 11 being active (no active stress constraint). Once
again results obtained from the finite element force method, performed in this study, indicate a superior
advantage over the ones obtained from the displacement method, as illustrated in Table 7.

5.2. Frequency constraints

In this section the efficiency of the force method in design optimization under frequency constraints have
been examined. Three examples shown in Figs. 6–8 (see Table 1) illustrate the proposed procedure and al-
low the results to be compared with those reported in literatures. Similar to the problems investigated in
Section 5.1, all information regarding the material and geometrical characteristics, side constraints and
behaviour constraints (natural frequencies) are provided in the relative figures.

5.2.1. 10-Bar planar truss structure with nonstructural lumped masses

The 10-bar planar truss with lumped massed at each of its free nodes is shown in Fig. 6 (see Table 1).
Recall that the number of displacement degrees of freedom is m = 8, the number of force degrees of
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5865

freedom is n = 10 and the number of redundancy is r = 2. At the initial design, all the cross-sectional areas
are 129.03 cm2 (20 in.2) and the initial mass is 3810.39 kg (8392.94 lbm). This problem was investigated by
Venkayya and Tischler (1983), as well as by Grandhi and Venkayya (1988) using the optimality criterion
and displacement method. First, the structure was designed with a fundamental frequency of 14 Hz alone,
using both the displacement and force methods. A minimum weight of 2637.85 kg (5810.24 lbm) was ob-
tained. The number of iterations required using the FM was lower than that required by the DM. The final
results for the cross-sectional areas and fundamental frequency are tabulated in Tables 8 and 9, respec-
tively. Venkayya and Tischler (1983) have reported a minimum mass of 3026.17 kg (6665.577 lbm), where
the optimum design was taken as input to compute the specified natural frequency. A fundamental natural
frequency of 14.47 Hz was obtained in the present analysis. A simulation carried out using the solution re-
ported in Venkayya and Tischler (1983) as the initial design resulted in a final design, which converged to a
lighter solution 2637.85 kg (5810.24 lb obtained before. The current design was based on a consistent mass

Table 8
Final design for the cross-sectional areas (cm2) for various frequency constraints (Hz) for the 10-bar planar truss structure
Element no. DM FM
x1 = 14 x2 = 25 x1 = 7 x1 = 14 x2 = 25 x1 = 7
x2 P 15 x2 P 15
x3 P 20 x3 P 20
1 219.909 48.166 38.619 219.903 48.123 38.245
2 47.916 35.852 18.239 47.916 35.832 9.916
3 219.909 48.194 38.252 219.903 48.200 38.619
4 47.916 35.852 9.910 47.916 35.884 18.232
5 0.645 14.800 4.419 0.645 14.826 4.419
6 0.645 7.632 4.200 0.645 7.632 4.194
7 123.626 41.135 24.110 123.626 41.103 20.097
8 123.626 41.142 20.084 123.626 41.181 24.097
9 54.677 13.200 11.452 54.677 13.200 13.890
10 54.677 13.194 13.897 54.677 13.187 11.4516
Mass (kg) 2637.85 871.92 537.01 2637.85 871.92 537.01
No. of iteration 256 1035 637 237 973 705
No. of A.C. 3 1 2 3 1 2
CPU time (s) 10.54 40.81 25.96 7.51 28.70 21.62

Table 9
Final design of natural frequencies (Hz) in different frequency constraints for the 10-bar planar truss structure
Freq. no. Initial design DM FM
x1 = 14 x2 = 25 x1 = 7 x1 = 14 x2 = 25 x1 = 7
x2 P 15 x2 P 15
x3 P 20 x3 P 20
1 11.23 14.00 8.01 7.00 14.00 8.01 7.00
2 33.05 18.01 25.00 17.62 18.01 25.00 17.62
3 36.85 29.40 25.00 20.00 29.40 25.00 20.00
4 68.26 34.55 26.68 20.00 34.55 26.68 20.00
5 75.86 49.36 32.83 28.20 49.36 32.83 28.21
6 85.18 53.11 40.92 31.07 53.11 40.94 31.07
7 85.74 85.10 62.52 47.68 85.10 62.52 47.68
8 103.10 90.41 64.79 52.35 90.41 64.78 52.35
5866 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

matrix. A simulation using a lumped mass matrix resulted in a final design with a minimum mass of 2895.53
kg (6377.82 lbm) suggesting that a lumped mass matrix may have been used by Venkayya and Tischler. To
confirm this finding, the optimum result reported in Venkayya and Tischler (1983) was used as input to
compute the fundamental natural frequency based on the lumped mass matrix. A fundamental natural fre-
quency of 13.96 Hz was obtained.
To demonstrate the application of the algorithm for designing a structure with other specified natural
frequencies, the structure was designed for a second natural frequency of 25 Hz. A minimum mass of
871.92 kg (1920.52 lbm) was obtained. Grandhi and Venkayya (1988) reported a minimum mass of
1018.69 kg (2243.8 lbm). Here, the optimum design was used as input to compute the second natural
frequency, resulting in a solution of 25.37 Hz for the second natural frequency. The final results for the
cross-sectional areas and the fundamental frequency are given in the Tables 8 and 9.
Finally, the structure was designed under multiple natural frequency constraints given by x1 = 7 Hz,
x2 P 15 Hz and x3 P 20 Hz. A minimum mass of 537.01 kg (1182.85 lbm) was obtained. Upon closer
inspection, the results reveal that the optimum cross-sectional areas for elements 9 and 10 obtained using
the FM is different from that using the DM. It is noted that the optimum masses for both the FM and DM
are exactly the same and so are the final natural frequencies. It is inferred that, as the optimizer is very sen-
sitive to the output results from the FM and the DM, a small difference causes the optimizer to select a
different path. It is interesting to note that, for this case, the number of iterations required by the DM is
now lower than that of FM. However, the computational time using the FM is still lower than that of
DM. Grandhi and Venkayya (1988) have reported a minimum weight of 594.01 kg (1308.4 lbm). The final
cross-sectional areas and natural frequencies are given in Tables 8 and 9. The variation of the optimum
mass with the first and second natural frequency limits is shown in Fig. 15. It is observed that, when increas-
ing the fundamental natural frequency limit, the optimum mass increases drastically from 157.95 kg (347.91
lbm) for a fundamental natural frequency limit of 4 Hz to 11448.38 kg (25216.7 lbm) for a fundamental
natural frequency limit of 22 Hz. However, a less dramatic change is observed for the second natural fre-
quency limit. The optimum mass increases from 21.26 kg (46.83 lbm) for a second natural frequency limit of
4 Hz to 640.20 kg (1410.14 lbm) for second natural frequency limit of 22 Hz.

13620
Fundamental Frequency
Second Frequency
11350
Optimum Mass (kg)

9080

6810

4540

2270

0
2 4 6 8 10 12 14 16 18 20 22 24
Frequency (Hz)

Fig. 15. History of the variation of the optimum mass with respect to the fundamental and second frequencies for the 10-bar planar
truss.
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5867

5.2.2. 72-Bar space truss structure with nonstructural lumped masses

The 72-bar space truss structure with lumped masses at its four top nodes (nodes 1–4) is shown in Fig. 7
(see Table 1). Recall that the number of displacement degrees of freedom is m = 48 and the number of force
degrees of freedom is n = 72. Therefore, the number of redundancies is r = 24. Here the design variables
(cross-sectional areas) have been linked into 16 groups so the number of the design variables becomes
16, and the associated number of the constraints reduces to 40.
This problem was investigated by Konzelman and Tabarrok and Konzelman (1986) using the dual
method based on approximation concepts for optimization and the finite element method based on dis-
placement method for analysis. Here, the structure was designed for a fundamental frequency of 4 Hz,
using both the displacement and the force methods. A minimum mass of 287.09 kg (632.361 lbm) was ob-
tained. For this example, the displacement method was found to be computationally more efficient than the
force method approach. The reason is that SVD technique used to develop the compatibility is not a cheap
technique and when the number of redundancy increases (as this example), generating the compatibility
matrix using SVD becomes computationally expensive. Thus, in optimal design problems with frequency
constraints, the analysis of not highly redundant structures using the force method is not necessarily more
efficient than the DM. The final results for the cross-sectional areas and frequencies are given in Tables 10
and 11 and they are in excellent agreement with those reported by Konzelman (1986) who reported a min-
imum mass of 287.09 kg (632.36 lbm). Due to the symmetry imposed by the structural geometry and the
linking scheme, the eigenvalue corresponding to the fundamental mode of vibration is a repeated eigenvalue
of multiplicity two. This means that in the initial and optimum design, the first and second modes of vibra-
tion have the same natural frequencies. Thus, a small change or deviation in the geometry of the structure
can switch the mode of vibration from first to the second mode.

Table 10
Final design for the cross-sectional areas (cm2) for the various frequency constraints (Hz) for the 72-bar space truss structure
Element no. DM FM
x1 = 4 x1 = 4 x1 = 4 x1 = 4
x3 P 6 x3 P 6
1–4 4.717 3.499 4.717 3.499
5–12 5.514 7.932 5.514 7.932
13–16 0.645 0.645 0.645 0.645
17–18 0.645 0.645 0.645 0.645
19–22 11.750 8.056 11.750 8.056
23–30 5.573 8.011 5.573 8.011
31–34 0.645 0.645 0.645 0.645
35–36 0.645 0.645 0.645 0.645
37–40 18.950 12.812 18.950 12.812
41–48 5.607 8.061 5.607 8.061
49–52 0.645 0.645 0.645 0.645
53–54 0.645 0.645 0.645 0.645
55–58 25.935 17.279 25.934 17.279
59–66 5.628 8.088 5.628 8.088
67–70 0.645 0.645 0.645 0.645
71–72 0.645 0.645 0.645 0.645
Mass (kg) 287.092 327.605 287.092 327.605
No. of iteration 544 379 510 379
No. of A.C. 9 10 9 10
CPU time (s) 283.78 200.37 302.96 227.62
5868 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

Table 11
Final design results for the natural frequencies (Hz) with different sets of frequency constraints for the 72-bar space truss structure
Freq. no. Initial design DM FM
x1 = 4 x1 = 4 x1 = 4 x1 = 4
x3 P 6 x3 P 6
1 3.113 4.000 4.000 4.000 4.000
2 3.113 4.000 4.000 4.000 4.000
3 5.374 5.001 6.000 5.001 6.000
4 9.425 6.505 6.247 6.505 6.247
5 13.189 8.595 9.074 8.595 9.074

It is noted that when the eigenvalues are repeated, the structures becomes extremely sensitive to any
change in design variables. Usually the natural frequencies of the first few modes are important and to sep-
arate these eigenvalues in the optimum design, the frequency constraints of the first few modes have to be
separated. In this example, due to the intrinsic nature of symmetry in the structure, any attempt to separate
the fundamental and second natural frequencies in the optimum design failed. To quantify the performance
of the algorithm for multiple frequency constraint problems, the structure was designed using the displace-
ment methods and force method for x1 = 4 Hz and x3 P 6 Hz. A minimum mass of 327.605 kg (721.597
lbm) was obtained. The results are given in Tables 10 and 11.

5.2.3. Member frame (two-story and one-bay) with nonstructural distributed mass
The 6-member frame is illustrated in Fig. 8 (see Table 1). This problem has been studied by Khan and
Willmert (1981) and McGee and Phan (1991) using the optimality criterion method combined with the
finite element method based on the displacement formulation.
The moment of inertia, I is empirically related to the area, A, by the following expressions (Khan and
Willmert, 1981; McGee and Phan, 1991):
I ¼ 4.6248A2 0 6 A 6 44
ð38Þ
I ¼ 256A
2300 44 < A 6 88.2813
where A is the area measured in square inches. At the initial design, all the cross-sectional areas are equal
to193.55 cm2 (30 in.2) with an initial mass of 5034 kg (11088 lbm). First, the structure was optimized using
both the displacement and force method giving a fundamental natural frequency of 78.5 rad/s. A minimum
mass of 4272.32 kg (9410.39 lbm) was obtained. It is interesting to note that the final design variables
(cross-sectional areas) are different in the displacement method and forced method solutions, suggesting
that the optimum solution is not unique. However, the final natural frequencies are the same. It is noted
that although the optimum results obtained using the force method and displacement method are different;
they resulted in the same optimum mass and same final natural frequency. Therefore, both are optimum
solutions. The results are given in Tables 12 and 13. Khan and Willmert (1981) and McGee and Phan
(1991) report a minimum weight of 4341 kg (9561 lbm) and 4456 kg (9815 lbm), respectively. To verify that
the optimality criterion employed by Khan and Willmert (1981) and McGee and Phan (1991) may have
produced a local minimum, another simulation was performed, starting with the solution reported in Khan
and Willmert (1981) and McGee and Phan (1991). This solution process resulted in a design change and
converged to a lighter solution of 4272.32 kg (9410.39 lbm). Thus, it is confirmed that the solution in Khan
and Willmert (1981) and McGee and Phan, 1991 does not represent a local minimum.
The structure was again designed using multiple natural frequency constraints of x1 = 78.5 rad/s and
x2 P 180 rad/s. Surprisingly, the optimum mass of 4365.56 kg (9615.78 lbm) using force method and
4418.46 kg (9732.3 lbm) using displacement method was obtained. As explained before, this specific prob-
lem is path dependent and the slightly difference in output results from analyzers (force method and
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5869

Table 12
Final design for the cross-sectional areas (cm2) for different sets of frequency constraints (rad/s) for the 6-member frame
Element no. DM FM
x1 = 78.5 x1 = 78.5, x2 P 180 x1 = 78.5 x1 = 78.5, x2 P 180
1 215.867 120.556 51.088 206.289
2 51.088 141.802 215.551 62.746
3 51.088 283.870 365.982 138.767
4 367.203 227.761 51.088 297.876
5 51.088 51.088 51.088 51.088
6 253.087 228.549 253.799 256.361
Mass (kg) 4272.35 4418.46 4272.32 4365.56
No. of Iteration 320 726 258 246
No. of A.C. 4 2 4 3
CPU time (s) 15.31 34.29 11.76 11.21

Table 13
Final design results for the natural frequencies (rad/s) for different sets of frequency constraints for the 6-member frame
Freq. no. Initial design DM FM
x1 = 78.5 x1 = 78.5, x2 P 180 x1 = 4 x1 = 78.5, x2 P 180
1 69.044 78.500 78.500 78.500 78.500
2 286.840 146.670 220.806 146.668 180.000
3 380.324 268.399 436.420 268.350 371.289
4 476.168 350.723 486.975 350.667 418.804
5 499.720 465.900 540.125 465.780 485.897

displacement method) may have caused a different optimum solution. Investigating the final natural fre-
quencies obtained from the displacement method and the force method, it was revealed that in the force
method, the inequality constraint is active in the optimum solution, and this is not observed in the solution
from the displacement method. This is the reason for a lighter mass obtained using the force method. For
this case, the force method performed better computationally than the displacement method. It can be in-
ferred that for frequency constraint problems, the computational time totally depends on the iteration
number.

6. Conclusion

A structural optimization algorithm, using integrated force formulation technique, has been developed
to minimize the mass of truss and frame type structures under the stress, displacement and frequency con-
straints. The required compatibility matrix in the formulation has been derived directly by utilizing a dis-
placement–deformation relationship and the Single Value Decomposition Technique. Moreover, the
Sequential Quadratic Programming method has been adopted to optimize the truss and frame structures.
The main objective of this study is to investigate the relative performance of the force and displacement
methods in the design and optimization of different planar and space structures with the stress–displace-
ment and frequency constraints. It is found that the optimization technique that is based on the force meth-
od is computationally far more efficient than the displacement method for structural design optimization
under stress–displacement constraints. It is also concluded that, in some problems with multiple frequency
constraints, the optimization based on the force method may result in a lighter design. It is noted that this is
not a general case, and they are specific results for the examples presented.
5870 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871

Last but not least, from the results obtained from eight different benchmark examples, it is demonstrated
that the Sequential Quadratic Programming method could result into a lighter optimal design of space
structures when compared to the conventionally used optimality criterion techniques. The proposed meth-
odology has proved to be extremely efficient in the analysis and optimization of the truss and frame type
space structures.

Acknowledgment

Support by the Natural Sciences and Engineering Research Council (NSERC) of Canada is gratefully
acknowledged.

References

AISC, 1978. Specification for the Design, Fabrications and Erection of Structural Steel for Buildings. American Institute of Steel
Constructions, Chicago IL, USA.
Canfield, R.A., Grandhi, R.V., Venkayya, V.B., 1988. Optimum design of structures with multiple constraints. AIAA Journal 26, 78–
85.
Coleman, T., Branch, M.A., Grace, A., 1999. Optimization toolbox for use with Matlab. UserÕs Guides, Version 2. The MathWorks
Inc.
Flurry, C., Schmit Jr., L.A., 1980. Dual methods and approximation concepts in structural synthesis. NASA CR-3226.
Golub, G.H., Van Loan, C.F., 1996. Matrix Computations, third ed. The Johns Hopkins University Press, Baltimore and London.
Grandhi, R.V., Venkayya, V.B., 1988. Structural optimization with frequency constraints. AIAA Journal 26, 858–866.
Haftka, R.T., Gurdal, Z., 1992. Elements of Structural Optimization, third ed. Kluwer Academic Publishers.
Khan, M.R., Willmert, K.D., 1981. An efficient optimality criterion method for natural frequency constrained structures. Computers
& Structures 14, 501–507.
Khan, M.R., 1984. Optimality criterion techniques applied to frames having general cross-sectional relationships. AIAA Journal 22
(5), 669–676.
Khot, N.S., Kamat, M.P., 1985. Minimum weight design of truss structures with geometric nonlinear behavior. AIAA Journal 23, 139–
144.
Khot, N.S., 1984. Optimization of structures with multiple frequency constraints. Computers & Structures 20 (5), 869–876.
Konzelman, C.J., 1986. Dual Methods and Approximation Concepts for Structural Optimization, M.A.Sc thesis, Department of
Mechanical engineering, University of Toronto.
McGee, O.C., Phan, K.F., 1991. A robust optimality criteria procedure for cross-sectional optimization of frame structures with
multiple frequency limits. Computers & Structures 38, 485–500.
Mohr, G.A., 1992. Finite Elements for Solids, Fluids, and Optimization. Oxford University Press.
Mohr, G.A., 1994. Finite element optimization of structures-I. Computers & Structures 53, 1217–1220.
Patnaik, S.N., Joseph, K.T., 1986. Generation of the compatibility matrix in the integrated force method. Computer Methods in
Applied Mechanics and Engineering 55, 239–257.
Patnaik, S.N., Berke, L., Gallagher, R.H., 1991. Integrated force method versus displacement method for finite element analysis.
Computers & Structures 38 (4), 377–407.
Patnaik, S.N., 1986. The integrated force method verses the standard force method. Computers & Structures 22 (2), 151–163.
Powell, M.J.D., 1978. A fast algorithm for nonlinearly constrained optimization calculations. In: Watson, G.A. (Ed.), Lecture Notes in
Mathematics and Numerical Analysis. Springer-Verlag, p. 630.
Robinson, J., 1965. Automatic selection of redundancies in the matrix force method: the rank technique. Canadian Aero Space Journal
11, 9–12.
Saka, M.P., Ulker, M., 1992. Optimum design of geometrically nonlinear space trusses. Computers & Structures 42, 289–299.
Saka, M.P., 1984. Optimum design of space trusses with buckling constraints. Proceeding of the Third International Conference on
Space Structures. University of Surrey, Guildford, UK.
Sedaghati, R., Tabarrok, B., 2000. Optimum design of truss structures undergoing large deflections subject to a system stability
constraint. International Journal for Numerical Methods in Engineering 48 (3), 421–434.
Timoshenko, S., 1953. History of Strength of Material. McGraw-Hill, New York.
 R. Sedaghati / International Journal of Solids and Structures 42 (2005) 5848–5871 5871

Vanderplaats, G.N., 1973. CONMIN- A FORTRAN program for constrained function minimization. UserÕs Manual, NASA TMX-
G2.
Venkayya, V.B., Tischler, V.A., 1983. Optimization of Structures with Frequency Constraints. Computer Methods for Nonlinear
Solids and Structural Mechanics ASME AMD-54, 239–259.
Venkayya, V.B., 1978. Structural optimization: a review and some recommendations. International Journal for Numerical Methods in
Engineering 13, 203–228.