International Journal of

Advanced Structures and Geotechnical Engineering

ISSN 2319-5347, Vol. 02, No. 01, January 2013

Optimal Geometry of Pin-Jointed Plane frames using a

Hybrid Complex Method
BASHIR ALAM1, M. I. HAQUE2, QAISAR ALI1, KHAN SHAHZADA1, SYED MOHAMMAD ALI1,
AFZAL KHAN1, MUHAMMAD IBRAHIM1
1

Department of Civil Engineering, University of Engineering and Technology Peshawar, Pakistan

School Engineering and Applied Sciences, George Washington University, Washington DC, USA
Email: bashir@gwmail.gwu.edu, mhaque@gwu.edu, drqaisarali@nwfpuet.edu.pk, shah_civil2003@yahoo.com
bridge_doctor@yahoo.com, afzal5426@gmail.com, ibrahim_004@hotmail.com
2

ABSTRACT: This paper describes a computational procedure of obtaining minimum weight design of pinjointed plane frames using design variables that specify both the skeletal geometry and member sizes of the
structure. The entire design space is decomposed into two subspacesthe geometric design space, and the
member design space. The variables of the geometric design space contain the coordinates of the variable joints
of the structure, while the variables of the member design space contain the cross sectional dimensions of the
members. The optimal design vector of the geometric design space is found strictly by the mathematical
programming technique, and the complementary part of the design vector in the member design space is found
by the fully-stressed optimality criterion. To validate the applicability and proficiency of the proposed procedure
a numerical problem has been optimized and the results are compared with the previous results available in the
literature. The study demonstrates that the developed algorithm is capable of finding near-optimal design in a
some what robust and efficient manner.
Keywords: Shape Optimization, Pin-Jointed Frames, Box Complex Method, Fully Stressed Design
Introduction:
The goal of structural optimization is to select optimal
values of the design variables such that the predefined
objective function is minimized (or maximized) and
all the restrictions imposed by explicit and implicit
constraints are met. The objective function is a
function of the design variables, which provides a
basis for choice between alternative acceptable
designs. It can be the weight or stiffness of the
structure, cost of material, or any combination of
these or similar factors. The constraints could be
explicit such as the restriction on the width or height
of the structure, or implicit such as the restrictions on
stresses and displacements.
Shape or geometrical optimization introduces
additional design variables, which allow for boundary
movement. The coordinates of joints and the crosssectional sizes are treated as design variables and
optimized simultaneously. In general, the design
variables are assumed to be continuous and numerical
search algorithm is used to find optimum.
In order to overcome the difficulties in obtaining the
first order information in shape optimization and
shortcomings of local and global optimization search
methods, this paper presents a reliable and efficient
numerical procedure, referred to as the Hybrid
Complex Method. This procedure is developed by
combining a modified version of the Complex
Method of Box with the fully-stressed optimality
criterion which treats problems with continuous
design variables. Successive improvements in the
design are achieved quite effectively by decomposing
the design space into two spaces: the geometric design
space and the member design space.

The variables of the geometric design space are

obtained by a modified version of the Complex
Method of Box, while the variables of the member
design space are obtained by using the fully stressed
design criterion. The design routine allows multiple
load cases and design variables linkage. The Complex
Method of Box [1] has been applied successfully to
relatively small academic nature of structural
optimization problems in the past (Fu [3], Haque [5,
6], Lai [7], and Lipson [8-12]), and still accomplishes
the goals set for it in this study. The Complex Method
is a mathematical programming procedure for finding
an optimal solution of non-linear, constrained
optimization problems. This method derives its
acronym COMPLEX from two words, Constrained
and Simplex. The Complex Method was proposed
originally by M. J. Box in 1965, where he
demonstrated efficacy of the method in finding near
optimal solution to non-linear, constrained
optimization problems. It is a Zero-order method
optimization method; that is, it does not require either
the gradient of the objective function, or that of the
constraints.
To solve an optimization problem, a computational
methodology is developed consisting of three
logically separable phases: the optimization phase, the
structural analysis phase, and the design evaluation
phase. During the optimization phase, attempts are
made to improve by finding feasible points that are
successively closer to an optimum. In the structural
analysis phase, the structure, provisionally obtained in
the optimization phase, is analyzed and, finally, the
feasibility of the structure is checked in the design

IJASGE 020105 Copyright 2012 BASHA RESEARCH CENTRE. All rights reserved.

BASHIR ALAM, M. I. HAQUE, QAISAR ALI, KHAN SHAHZADA, SYED MOHAMMAD ALI,
AFZAL KHAN, MUHAMMAD IBRAHIM
evaluation phase. The developed procedure is applied
to the optimization of a pin-jointed plane frame used
by several previous investigators to study minimum
weight designs constrained by allowable stresses,
joint coordinates and member sizes. The formulation
worked quite well, and generally converged to better
solutions than those reported in the literature.
Proposed Modifications and Implementation of the
Complex Method
The modifications to the Complex Method as used in
this study are summarized as follows.
1) Separation of design space
In the Hybrid Complex Method proposed in this
study, the entire design space is decomposed into two
subspaces: the geometrical design space and the
member design space. The variables of geometrical
design space consist of the unknown joint coordinates
while variables of the member design space define the
cross sectional dimensions of the members. During
the Optimization process, the variables of the
geometrical design space are strictly obtained by the
Complex Method and the variables of the member
design space are found by using the fully-stressed
optimality criterion.
2) Feasibility of the initial design
An initial point (skeletal geometry) in an ndimensional geometric design space is chosen. In the
original procedure this point was required to be
feasible, but the present algorithm has been written in
such a way that if the initial chosen point is not
feasible it is made feasible by adjusting one or more
of the coordinates of the design vector.
3) Satisfying the implicit constraints
In the original procedure proposed by Box the points
in the initial complex which violated the implicit
constraints were moved halfway back towards the
centroid of the remaining, already accepted points.
The process of moving halfway in towards the
centroid is repeated until the point becomes feasible.
In the present method, an attempt is made to satisfy
all the implicit constraints for each randomly chosen
point during the sizing of the members in the member
design space, using the fully-stressed optimality
criterion. If it is impossible to satisfy all the implicit
constraints by this method, then, the original
procedure is used.
4) The improvement procedure
The improvement procedure has been modified in that
at every iteration the worst design is reflected through
the centroid of the remaining designs in the
geometrical design space to a new point. Then, when
this new point has been optimally sized, its objective
function is evaluated and compared with that of worst
design in the complex. If the new point is less, it is
accepted as a design improvement and termination
criteria are checked; if greater, instead of continuously
halving , it is halved only thrice and then centroid is
considered as a candidate for improvement. If

centroid is still greater than the worst, then a new

point is located at the midpoint of a line joining
centroid to the best point in the complex. If the
objective function is still greater than the worst, then
the worst point is replaced by the best design in the
complex.
5) Termination criteria
The procedure in 3 is repeated until a preset
termination criterion is reached. The first termination
criterion used in this study is based on the objective
function values of all k points in the complex. This
convergence criterion is met if the ratio of the
difference between the maximum objective function
value and the minimum objective function value to
maximum objective function value of the points in the
complex is less than or equal to the value of (a
user defined variable). The second criterion that is
checked for the convergence of the solution is a
measure of the design space spanned by the vertices
of complex. Finally, a constraint is placed on the
maximum number of iterations that may occur before
terminating the optimization. The optimization
process is terminated as soon as any of the
termination criteria is satisfied.
Sizing of Members
The design procedure used is an iterative technique,
and commences with the smallest cross sections of the
members whose skeletal geometry has been
established by the complex method. The program
takes each load condition and calculates the
deflections and forces. It determines the maximum
axial force for each member of the structure. After
completing the final loading condition, it summarizes
the condition of forces and deflections, by selecting
their highest value from the loading conditions for
each member. The cross-sectional design of the
members is undertaken in the member design space.
During this phase of calculations, the skeletal
geometry of the structure is regarded as fixed, while
its member cross-sections are determined based on
fully-stressed design. It then considers each group of
members and selects the maximum value for that
group. The program substitutes the modified crosssectional dimensions for the members, making the
necessary alterations to the stiffness matrix. The
analysis and design cycle is repeated until the
program is unable to modify any group of members.
This constitutes the final design for the structure
whose skeletal geometry has been established by the
complex method.
Example: Eighteen-Bar 2-D Pin-jointed Frame
The example is used to demonstrate the capability of
the Hybrid Complex Method to determine the best
shape and member sizes of the eighteen-bar 2-D Pinjoint frame, in order to minimize its weight without
exceeding the permissible stresses. This structure with
identical load condition has been analyzed in previous
works by Hansen and Vanderplaats [4], and Felix and
Vanderplaats [2].

International Journal of Advanced Structures and Geotechnical Engineering

ISSN 2319-5347, Vol. 02, No. 01, January 2013, pp 24-30

Optimal Geometry of Pin-Jointed Plane frames using a Hybrid Complex Method

The figure 1 shows initial geometry, loading

condition, member and node numbering for the
structure. The design data used in this example are
given in table 1. All members are solids and circular
in cross-sections, thus the dimension of each member
is the dimensional design variable. To satisfy the
practical construction requirements, the dimensional
design variables are further divided into four
independent groups as
Group A [ d1 d4 d8 d12 d16 ],
Group B
[ d2

d6 d10 d14 d18 ],

Group C [ d3
[ d5

d7 d11 d15 ] and Group D

d9 d13 d17 ].

The diameters of the members included in Group A,

Group B, Group C and Group D are denoted by d a, db,
dc and dd respectively in Figure 1. The coordinates in
the x and y directions of nodes 3, 5, 7, and 9 are
assigned as geometrical design variables. There are
four independent dimensional design variables and
eight independent geometric design variables. During
optimization process, the geometric variables are
modified by the Complex Method and member sizes
by fully-stressed optimality criteria. Eight times the
design optimization process has been repeated using
eight different seed numbers to generate eight initial
complexes. The final weights are presented in Table
2, which establishes the near global solution of the

problem. The optimum weight obtained in this study

is 3875.85 lb, which is about 0.80% lighter than the
design reported by Hansen and Vanderplaats [4], and
1.43% lighter than the optimum design presented by
Felix and Vanderplaats [2]. The final configuration is
shown in figure 2.
The computer run for the best design showed the
initial complex, randomly generated, consisted of
sixteen designs with total truss weights ranging from
4257.40 to 45925.33 lb. After 356 iterations the
stopping criteria was satisfied. The best design has a
total truss weight of 3875.82 lb. Even in the initial
complex, the best design is about 6% lighter than that
of Felix and Vanderplaats [2] and just 9% heavier
than that reported by Hansen and Vanderplaats [4].
Figure 3 depicts the progression of maximum and
minimum weights of the complex towards the
optimum solution. The worst design in the initial
complex is about 13 times heavier than the final
design. After 50 iterations, the worst design in the
complex becomes only 1.3 times heavier than the
final design, which shows how the optimization
problem with eight geometrical design variables
rapidly reaches the optimum.
The process of
optimization could have been terminated early by
setting higher convergence criteria and would still
have arrived at roughly the same solution. The
histories of geometric design variables and
convergence criteria shown are in figures 4 through 6.

Figure 1 Initial geometry of 18-bar 2-D truss

Figure 2 Final configuration of 18-bar 2-D

International Journal of Advanced Structures and Geotechnical Engineering

ISSN 2319-5347, Vol. 02, No. 01, January 2013, pp 24-30

BASHIR ALAM, M. I. HAQUE, QAISAR ALI, KHAN SHAHZADA, SYED MOHAMMAD ALI,
AFZAL KHAN, MUHAMMAD IBRAHIM
4

x 10

maximum weight in complex

minimum weight in complex

4.5

Structure weight in complex (lb)

(45925.33 lb)

3.5
3
2.5
2
1.5

(4257.40 lb)
13.3% heavier than the final design

1
final design (3875.82 lb)
0.5
0

50

100

150
200
250
300
Iteration number
Figure 3 History of max/min weight truss

350

400

300
node
node
node
node

Geometric design variables (in.)

250

3
5
7
9

y-coordinate
y-coordinate
y-coordinate
y-coordinate

200

150

100

50

50

100

150
200
250
Iteration number

300

350

Figure 4 History of geometric design variables (y-coordinates)

International Journal of Advanced Structures and Geotechnical Engineering
ISSN 2319-5347, Vol. 02, No. 01, January 2013, pp 24-30

400

Optimal Geometry of Pin-Jointed Plane frames using a Hybrid Complex Method

Table 1 Design data for the eighteen-bar 2-D truss

Modulus of
elasticity
10,000 ksi
0.1 lb/in3

Stress
constraints

20.00 ksi for all members

Side constraints

None for dimensional design

variable
125 in. in x-direction for
geometric design variables

Table 2 Optimum weights from eight runs using

different seeds
Run
Optimum Weight (lb)
1
3878.00
2
3898.80
3
3880.89
4
3888.41
5
3897.23
6
3914.93
7
3875.82
8
3898.88

1000
Geometric design variables (in.)

Weight density

node 9 x-coordinate
node 7 x-coordinate
node 5 x-coordinate
node 3 x-coordinate

1200

800

600

400

200

50

100

150
200
250
Iteration number

300

350

Figure 5 History of geometric design

Figure 6 History of convergence criteria variables (x-coordinates)

International Journal of Advanced Structures and Geotechnical Engineering

ISSN 2319-5347, Vol. 02, No. 01, January 2013, pp 24-30

400

BASHIR ALAM, M. I. HAQUE, QAISAR ALI, KHAN SHAHZADA, SYED MOHAMMAD ALI,
AFZAL KHAN, MUHAMMAD IBRAHIM
Conclusion:
A sequence of increasingly better designs for pinjointed plane frames with variable geometry can be
achieved quite effectively by decomposing the design
space into two subspaces the geometric design space
and the member design space. An important feature of
this resolution is that the design requirements can be
easily accommodated during the latter phase of sub
optimization in the member design space. The
complex method is able to search the design space of
the given model and is an effective tool in locating the
optimal skeletal geometry of the structure. The
improvement in function value is very rapid in the
initial iterations after the initial feasible complex is
established. It is simple to formulate the problem
generally in a manner suitable for Box Complex
method. It becomes very easy to apply the method
when there are more than a few constraints in the
problem. Multiple loading conditions can be
accommodated easily by this method without
significantly increasingly the computational effort.
The designer can examine the many different
optimum design states and can set some additional
goals or constraints. If for some reason it is not
possible to use optimum design, one of the slightly
heavier designs found could then be selected.
References:
[1] Box, M. J., 1965, A New Method of Constrained
Optimization and a Comparison with Other
Methods, Computer journal, 8 (1), pp. 42-52.
[2] Felix, J. and Vanderplaats, 1987, Configuration
Optimization of Trusses Subject to Strength,
Displacement and Frequency Constraints,
Journal of Mechanisms, Transmissions, and
Automation in Design, ASME, 109(2), pp. 233241.

[3] Fu, K. C. and Levey, G. E., 1978, Discrete

Frame Optimization by Complex-Simplex
Procedure, Computer & Structures, 9, pp. 207217.
[4] Hansen, S. R. and Vanderplaats, G. N., 1988,
An Approximation Method for Configuration
Optimization of Trusses, Proceeding of
AIAA/ASME/ASCE/ AHS 29th Structures,
Structural Dynamics and Material Conference,
pp. 1667-1676.
[5] Haque, M. I., 1985, Optimal Design of Plane
Frames by the Complex Method, Computers and
structures, 20, pp. 451-456.
[6] Haque, M. I., 1996, Optimal Frame Design with
Discrete Members Using the Complex Method,
Computers and structures, 59, pp. 847-858.
[7] Lai, Y. S. and Achenbach, J. D., 1973, Direct
Search Optimization Method, J. Structural
Division ASCE, 99(ST1), pp. 19-31.
[8] Lipson, S. L. and Haque, M. I., 1980, Optimal
Design of Arches Using the Complex Method, J.
Structural Division ASCE, 106, pp. 2509-2525.
[9] Lipson, S. L. and Gwin, L. B., 1977, The
Complex Method Applied to Optimal Truss
Configuration, Computers and Structures, 7, pp.
461-468.
[10] Lipson, S. L. and Gwin, L. B., 1977, Discrete
Sizing of Trusses for Optimal Geometry, J.
Structural Division ASCE, 103(ST5), pp. 10311046.
[11] Lipson, S. L. and Russell, A. D., 1971, Cost
Optimization of Structural Roof System,
J.
Structural Division ASCE, 97(ST8), pp. 20572071.
[12] Lipson, S. L. and Agrawal, K. M., 1974, Weight
Optimization of Plane Trusses,
J.
Structural Division ASCE, 100(ST5), pp. 865880.

International Journal of Advanced Structures and Geotechnical Engineering

ISSN 2319-5347, Vol. 02, No. 01, January 2013, pp 24-30