COMPUTER PROGRAM DEVELOPMENT OF TRUSS STRUCTURAL

CONFIGURATION OPTIMIZATION

Wahyu Kuntjoro, Jamaluddin mahmud

Faculty of Mechanical Engineering, Universiti Teknologi MARA, 40450 Shah Alam,

Selangor, Malaysia

Abstract
An optimum structural design is a design, which minimize (or maximize) a
certain objective function, and at the same time meets its design
requirements. In most of cases, the objective functions are weight of
structures. The design requirements are strength and stiffness among
others. Normally, the optimization is performed on the size of the
structural members. Performing structural optimization by involving its
configuration, that is the location of structural joints, as the design
variables will offer more flexibility in the final design and also open up the
design space. Design requirements are formulated as the design
constraints while the weight of structure is formulated as the objective
function, which includes the joint location as the design variables. Linear
Extended Interior Penalty Function is employed to optimize the structure,
with Finite Element Analysis is used to obtain the truss structural
responses. This paper describes the development of computer program of
truss optimization configuration. Some results are shown.

Keywords: Structural Configuration Optimization, Truss Optimization,

Mathematical Programming.

1. Introduction
An optimum design is a design, which minimize (or maximize) a certain objective
function, and still meets its design requirements. In structural domain, in particular the
lightweight structure, the weight is often to be minimized. On the other hand, the
structure shall meet its design requirements with regard to strength, stiffness, and
others. It is obvious that design optimization will be conducted iteratively until a
minimum weight structure, which still meet the design requirement, is obtained. The
advanced of computer makes possible the design optimization process to be conducted
in systematic and consistent manners. Optimization techniques which are based on
optimality criteria approach[2], mathematical programming[1,2,3,7], and genetic
algorithm are widely employed now.

This structural optimization which is reported in this paper is based on the mathematical
programming. In mathematical programming, the weight to be minimized is formulated
as the objective function. The objective function is a function of a set of design variables.
Design variables are structural parameters, which values to be varied during the
optimization (iteration) process. Design variables normally are thickness (for plates),
cross sectional area (for bars), or second moment of area (beams). The design
requirements are formulated as the design constraints. For example, design constraints
will require that working stress is less than strength, and displacement magnitude is
limited. Again design constraints are functions of design variables. Figure 1 shows the
flow chart of design optimization using mathematical programming. The optimization is
stopped if the design is converging. Convergence criteria includes the convergence in
design variables, or/and convergence in objective function, and also requires that design
constraints are satisfied.

Define:
1. Design variables
2.Objective function
3. Design constraints

Initial design

Analysis

Calculate constraints

Y
Converging? Stop

redesign using
optimization methods

Figure 1 Flow chart of the optimization procedure.

As mentioned before, normally the optimization is performed on the size of the structural
members. Performing structural optimization by involving its configuration, that is the
location of structural joints, as the design variables will offer more flexibility in the final
design and also open up the design space. In this case, the set of design variables will
consists of the locations of structural joints, in addition to the sizes of structural
members.

Analysis stage can be performed using relevant structural analysis approach, depends to
the problem at hands. However, for complex structures with large numbers of
components, the finite element method is normally used in the analysis stage. In its
analysis, finite element method is often based on entities such as cross sectional area for
truss, or second moment of area for beam. This is in line with the entities used as the
design variables in the optimization process. Addition of joint locations are not really a
problem, as joint locations (nodal points) indeed are parameters in finite element
method, which are employed in the setting-up of structural stiffness matrix.

A research in the development computer program for truss structural configuration

optimization is reported in this paper. Strength and displacement are treated as the
design requirement. FORTRAN language is used in the development of the software.
2. Optimization Theory

2.1. Optimization Statement

In mathematical programming, the problem of structural design is formulated into a
general mathematical expression. Structural optimization problem with non-linear
constraints can be expressed as, finding (x), hence the objective function

f(x) is minimum (1)

and satisfy the design constraints,

gj(x) ≤ 0 j=1,m (2)

where (x) = (x1, x2, x3, ..., xn) are design variables, and gj is inequality constraint.
In searching for optimum, the design is change iteratively. Iteration procedure is
performed according the following equation,

(xq+1) = (xq) + α Sq (3)

where S is the search direction in the design space and α is the step length (at direction
of S). Search procedure can be regarded as searching the optimum point in a design
space. For design problem of two variables, optimization procedure is illustrated in
Figure 2.

X2 g1

x1 g4
X1

X2 S1

S2
C1 < C2 < ... < C5
X4 S3
X3

optimum point
g2

f = C1
f = C4 f = C5
f = C2
X1
g3
f = C3

Figure 2. Searching for optimum in a 2-D design space

2.2. Extended Interior Penalty Function Method

Linear Extended Interior Penalty Function (LEIPF) [1] is used for mathematical
programming optimization. This method is a further development of the Interior Penalty
Function method, which is under the category of Sequential Unconstrained Minimization
methods. In LEIPF, based on equation (1) and (2), a pseudo objective function is set-up
as follow:

Φ( x, r ) = f ( x ) + r P(x ) (4)

P(x) is penalty function which is defined as follow,

 m
P( x ) = ∑ ~
g j( x)
j=1

~ 1
g j ( x) = − g j ( x) ≤ ε (5)
g j ( x)
~ 2ε − g j ( x )
g j ( x) = − g j ( x) > ε
ε2

Parameter ε is a small negative number, which points out the transition from interior
penalty mode to extended penalty.

ε = - C ra 1/3 ≤ a ≤ 1/5 (6)

Equation (4) is minimized sequentially. Each iteration sequence is an unconstrained

minimization. The value of r is updated at every sequence. Figure-3 shows the flowchart
of LEIPF algorithm. The search direction used is according to Powell Method [1,2,3].

START

Define: Xo, r, c, a, gama

Calculate
Constraints

Minimize (X,r,eps.) as
unconstrained
optimization

Y
converging Stop

r = (gama)(r)

Figure 3. LEIPF Algorithm

2.3. Design Approximation

Optimization process frequently needs to involve a huge number of constraints,
thousands of constraint equations. There is possibility that many constraints are not
critical or has no influence to optimum design. The same thing applies to design
variables. Some of the variables might not have effects to the optimum. On the other
hand, structural responses are often the implicit function of design variables. In this
case, it is very inefficient to perform a complete analysis using the finite element method
for a small change in design (due to small change in one design variable). Approximation
concepts try to handle this problem by reducing the number of finite element analysis in
an optimization process.

Approximation concept to use is based on first order (linear) Taylor series [2].

rj(xk + ∆xk) ≅ rj(xk) + ∇rjT(xk). ∆xk (7)

The above approximation is linear against a change in design variable, ∆xk. Hence, it is
not always necessary to use finite element analysis, but it is sufficient to use
approximation technique. From the above equation, it is clear, that information of
response derivative is required.

2.4. Analysis and Response Derivative

Structural analysis is based on displacement method [5,6]. The structure is assumed as
truss structure. A finite element analysis for truss structure has been written. Linear
approximation is performed in finding the structural responses, e.g., stress and
displacement. For that purpose, it is necessary to obtain the response derivative with
respect to the design variables, that is the design sensitivity. The calculation of stress
derivative ∂σ/∂x, and displacement derivative ∂δ/∂x, is obtained through the finite
difference approach.

3. Programming
Optimization programming was performed using Fortran language. Unconstrained
optimization was programmed by making use various routines available at [4]. LEIPF
algorithm is based on the one in literature [1]. Finite element method for truss analysis
is based on [5,6]. The program is a further development of the previously developed
prototype which deals with member size optimization problem [7].

The optimization algorithm as a whole, which shows the cycle of optimization and finite
element analysis is given in figure-4.

FINITE ELEMENT ANALYSIS

Structural Response
Initial Design Sensitivity Analysis
Analysis

Problem Linearization

Better Design Optimization Algorithm

Figure-4 The cycle of optimization process and finite element analysis

4. Formulation of Structural Configuration Optimization
A configuration optimization will be illustrated through a design optimization of a simple
three-bar truss structure. A three-bar truss structure with its associated structural
parameters and loading are shown in Figure 5.

B C
1 2 3

1 2 3
A
P
4

Figure 5 A three-bar truss structure.

The structure is simply supported at three nodes, 1, 2, 3, and loaded at node 4 with a
concentrated force P. The location of joints (nodals 1, 2, 3, and, 4) can be described
based on the values of A, B, and C. The cross sectional area of members 1, 2, and 3, are
A1, A2, and A3 respectively. The modulus elasticity of each bar is E (assuming all bars
are made of the same material), the weight density is RHO, and the maximum allowable
stress (strength) is S.

In structural configuration optimization, the design variables are joint locations, A, B,

and C, in addition to the member cross sectional areas A1, A2, and A3. The objective
function is the total weight W of the structure, which is a function of design variables and
includes the mass density in the objective equation. The design constraints involve
strength requirement for each bar, that is the stress at each bar must be not higher than
the strength, and displacement at node 4. To make the final design to be meaningful,
side constraints are required. In this case, the dimensions A, B, and C, are bounded in
such a way to make the final dimensions of the optimum truss to be realistic. For
example, no negative dimensions are allowed. Similarly, member cross sectional area
should be bounded to yield realistic final sizes of the members.

500mm 500mm
1 2 3

1 2 3
A

10,000 N

Figure 6 A three-bar truss structure.

5. Case Study
5.1. Case-A, Three-Bar Structure, Configuration Optimization.
A three-bar truss structure with its associated structural parameters and loading are
shown in Figure 6 [8]. The structure is simply supported at nodes, 1, 2, 3, and loaded at
node 4. The location of joints (nodals 1, 2, 3, and, 4) are described based on the values
of A and B (B is fixed at 500 mm). The cross sectional area of members 1, 2, and 3, are
A1, A2, and A3 respectively. The modulus elasticity of each bar, E is 71E3 MPa
(assuming all bars are made of the same material). The weight density, RHO is 2800
Kg/m3. The maximum allowable stress (strength), S is 100 MPa. It is required to obtain
the optimum design of the structure.

From the problem definition, it is obvious that the design variables 1 to 4 are A1, A2, A3
and A respectively. The structure is symmetrical in vertical axis. The following geometric
properties are used:
Minimum member cross-sectional area = 50 mm2 (initial value = 100 mm2)
Range of dimension A = 300 to 700 mm (initial value = 500 mm)

The optimum parameters are listed below:

Weight = 0.239 Kg
A1 = 50.5 mm2 A2 = 87.0 mm2 A3 = 50.5 mm2
A = 303.1 mm
Stress-1 = 26.6 Mpa Stress-2 = 98.9 Mpa Stress-3 = 26.6 MPa
Displ-4 = 0.42 mm

Weight History

0.9
0.8
0.7
0.6
Weight, Kg

0.5
Weight (Kg)
0.4
0.3
0.2
0.1
0
1 3 5 7 9 11 13 15 17
Iteration

Figure 7(a)

Design Variables History

600
Area (mm2), Height (mm)

500

400 A1 (mm2)
A 2 (mm2)
300
A3 (mm2)
200 H (mm)
100

0
1 3 5 7 9 11 13 15 17
Iteration

Figure 7(b)
 Displacement History

0.45
0.4
0.35
0.3

disp, mm
0.25
d (mm)
0.2
0.15
0.1
0.05
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Iteration

Figure 7(c)

Stress History

120

100
Stress (MPa)

80 S1 (Mpa)
60 S2 (Mpa)
S3 (Mpa)
40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Iteration

Figure 7(d)

Figure 7 History of optimization, Case A.

The history of optimization is shown in Figure 7. As can be seen, member 2 is the critical
member from stress point of view, as the stress (stress-2) is close to maximum limit
which is 100 MPa. The dimension A tends to go to lower limit to reduce the volume of
the members. Displacement is not critical as the magnitude of displacement, 0.42 mm,
is far below the allowed displacement value which is 5 mm. Looking at the external
forces direction, it is quite logical that the structure tries to carry the load through its
central member, as these will make the forces flow directly to the supports. For this
reason, the optimization tries to minimize the cross-sectional area of members 1 and 3.

5.2. Case-B, Nine-Bar Structure, Configuration Optimization.

Figure 8 shows a nine bar structure to be optimized [8]. The configuration is optimized
with variation of vertical location of joints 4 and 6. Members are linked to obtain the
following design variables: area member 1 and 2 as var-1, area member 3 and 4 as var-
2, area member 5 and 9 as var-3, area member 6 and 8 as var-4, and area member 7
as var-5. The vertical location of joints 4 and 6 (designated as Y) is var-6. Maximum
allowable stress is 100 MPA, and the allowable displacement is 5 mm. Initial area is 100
mm2, and the initial value of Y is 500 mm. Minimum member area is 50 mm2, while the
value of Y can vary between 300 mm to 700 mm.

The optimum condition is given as follows:

 X1=50.5mm2 X2=108.6mm2 X3= 101.4mm2 X4=118.1mm2 X5=79.7mm2
X6=303.380mm
Weight = 1.12 Kg
stress1=0 stress 2=0 stress 3=-98.9 MPa
stress 4=-98.9 MPa stress 5=-98.6 MPa stress 6=99.0 MPa
stress 7=98.7 MPa stress 8=99.0 MPa stress 9=-98.6 MPa
disp2=-2.4 mm

At optimum, the height of nodes 4 and six reduce to its lower limit. The stress at
member-1 and member-2 is zero, and the area goes down to its minimum alowable.
Other members are stressed to the allowable stress value. The weight history is given in
Figure 9.

500 500
4 3 5 4 6

6 7 8 9
Y 5

1 1 2 2 3
20,000
Figure 8 Nine bar structure.

Weight History

3
2.5
2
1.5 Weight (Kg)
1
0.5
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Iteration

Figure 9 Weight history, Case B.

6. Conclusion
An optimum structural design is a design, which minimize (or maximize) a certain
objective function, and still meets its design requirements. For many structures, more
especially in aircraft structures, the objective function normally is weight of structure,
while the design requirements are strength and stiffness among others. Normally, the
optimization is performed on the size of the structural members. Performing structural
optimization by involving its configuration, that is the location of structural joints, as the
design variables will offer more flexibility in the final design and also open up the design
space. A computer program in FORTRAN of structural configuration optimization has
been developed and described in this paper. Truss structure has been used as the case
study.

In structural configuration optimization (truss structures), the design variables are joint
locations (or nodal coordinates), in addition to the member cross sectional areas. The
objective function is the total weight W of the structure, which is the function of design
variables. The design constraints are based on the design requirements, which in the
case looked at in this paper involve strength and displacement requirement. Side
constraints are introduced to the joint locations and member areas to yield realistic
optimum structure.

The optimization results show the flexibility offered by configuration optimization by

opening up design space, which now includes joint locations. It was also found that side
constraints will have a profound effect to the outcome of the optimization.

7. References
1. Vanderplaats, GN. Numerical Optimization Techniques for Engineering Design: with
Applications. McGraw-Hill Book Company, 1st ed., USA, 1984.
2. Morris, AJ. Foundations of Structural Optimization: A Unified Approach.
John Wiley & Sons, 1st ed., UK, 1982.
3. Arora, JS. Introduction to Optimum Design.
McGraw-Hill Book Company, 1st ed., USA, 1989.
4. Press, WH; Flannery, BP; Teukolsky, SA; Vetterling, WT. Numerical Recipes.
Cambridge University Press, USA, 1989.
5. Cook, RD. Concepts and Applications of Finite Element Analysis. 2nd ed.
John Wiley and Sons, USA, 1981.
6. Weaver, W. Jr; Johnston, PR. Finite Element for Structural Analysis.
Prentice-Hall Inc., USA, 1984.
7. Kuntjoro, W; Husain, AK; Berhan, MN. Angle-Profile Truss Structure Optimization
using Measurable Design Variables with Stress and Displacement Constraints,
Experimental and Theoretical Mechanics 2001, Bandung, Indonesia, June-2001
8. Kuntjoro, W; Mahmud, J. Optimization of Structural Configuration using Mathematical
Programming, STG Report, Bureau of Research and Consultancy, Universiti Teknologi
MARA, Malaysia, 2004