Professional Documents
Culture Documents
Reprinted from
INTERNATIONAL JOURNAL OF
SPACE STRUCTURES
Volume 24 · Number 1 · 2009
Key Words: Space structures, double layer grids, domes, shape optimization,
genetic algorithm.
7 Z
6 5 7
5 67
4 3 5
3 4 7 A
3 7
2 6
3
2 4 5 7
a 1 1 3 a H ri
1
2 3 7 zi B
1 4 5 r
6
Section a-a
gains the most effective state against the applied loads. The order of the polynomial is limited between 2
This situation optimizes the location of joints, their and circumferential divisions on the dome. Therefore,
load bearing area and the configuration of structure. the length of gene for the order of dome curve Ln is
In this research, optimization of single layer domes gained by the following equation:
with constant rise and span is studied. Since domes are
2 Ln ≥ nd − 1 (3)
formed by revolving a curve about a vertical axis, the
best equation for the curve and location of joints on where nd is the circumferential divisions.
this curve are considered as the variables of the Because the radius of dome rings is a continuous
optimization problem. For this purpose, first the variable, the gene length of radius for the ith ring Lr,i is
equation of the curve is determined and then by calculated by the following equation [10].
decoding the radius of each ring of the dome and the
log (ru , i − rl , i )/ e
use of curve equation, the height of joints on each ring Lr , i = INT 1 + ; i = 1,, 2, ..., nd (4)
is calculated. log 2
With regards to the geometry of domes, the where ru,i , rl,i and e are the upper bound, lower bound
equation of curve should have the following properties and required accuracy for the radius of the ith ring,
in the interval [A, B] (Fig. 2). respectively.
1. The curve should have its maximum at A. In this study, discrete variables are used for
∂z
i.e. = 0
optimizing the cross-sectional area of structural
∂r A members. These variables are selected from pipe
2. The curve should be descending in the interval sections with specified thickness and diameter.
∂2 z
[A, B]. i.e. 2 < 0
∂r 3. DECODING OF VARIABLES
where z and r are the height of joints on each ring and
After generating an initial population, the real value of
radius of each ring, respectively.
each design variable is evaluated by decoding, in order
According to above conditions, the curve is chosen
to evaluate the objective function and the magnitudes
to be a polynomial of order n:
of constraint violations.
z = a0 + a1r n (1) In topology optimization, the chromosome is
composed of two parts which includes variables of the
By applying the boundary conditions in Fig. 2 and presence or absence of joints and cross-sectional area
calculating constants (a0 and a1), the equation takes of each member group. Suppose the chromosome
the form: shown in Fig. 3 is an individual from the initial
2 n n population. It is produced for the topology
A ( 0, H ) and B ( S / 2, 0 ) ⇒ z = H − r + 1 (2)
S
optimization of the structure shown in Fig. 1. It is also
assumed that the members of this structure are divided
where H and S are the rise and the span of the dome, into 12 groups in which cross-sectional area of each
respectively. group is selected from 8 profiles.
Presence or Absence of
the Joints
IT1 IT2 IT3 IT4 …
1 1 0 1 1 1 1 0 0 0 1 0 1 0 1 0 1 0 0 0 1 0 1 1 1 0 0 1 0 1 1 1 0 0 0 1 0 0 0 0 1 0 1 0
Table 2. Categorizing symmetry positions of joints case, the purlins are designed, and their weights are
considered in the objective function.
Group Symmetry positions of joints
In geometry optimization, the chromosome is
1 37 composed of three parts which includes variables of
2 33, 34 dome curve order, radius of dome rings and cross-
3 31, 32, 35, 36 sectional area of each member group. Decoding the
4 19, 30 order of dome curve equation is accomplished by the
5 24, 25 following equation:
6 20, 21, 28, 29
Ln
n = ∑ Cn (i ) 2( Ln −i ) + 2
7 22, 23, 26,27
(5)
i =1
R j (rl , j − ru , j )
rj = rl , j + (6)
−1
Lr, j
2
Hs
where rj is the actual value of radius of the jth ring.
Hc
L
34
4. FORMULATION OF
23
OPTIMIZATION PROBLEM
In optimum shape design of the space structures, the
8 43 aim is to minimize the weight of the structure, under
42
some constraints on stresses, slenderness ratios and
7
displacements. The weight of the structure which
19 30 includes the weight of members, purlins and joints of
the space structures is taken as:
Figure 4. Top and bottom grids and support positions of
decoded structure.
Ne Nj
The first 8 bits in the chromosome relate to the minimize W = ∑ ρe Ai Li + ∑ Wj + Wp (7)
presence and absence of joints. Since the 3rd and 8th i =1 j =1
bits have a zero value, all the joints in 3rd and 8th
groups in Table 2 are omitted from the ground structure. where ρe, Ai, Li, Wj, Wp, Ne and Nj are the material
After decoding, the refined structure is shown in Fig. 4. density, cross-sectional area of ith element, length of
Now for decoding the cross-sectional area of members ith element, equivalent weight of jth joint, the weight
in any group, the remaining 32 bits of the chromosome of purlins, number of members and number of joints
are used. First, the value of each gene is transformed to after decoding of variables, respectively.
a decimal number [2]. Then, by referring to the table of In this study, Mero connector is used for the joints
profiles, properties of members in each group are of space structure. For calculating the equivalent
attained. After omitting the joints of top grid, a suitable weight of jth joint, first the diameter d of the bolt
cladding system is considered and the loads of any which connects the member to this joint is determined
removed joint are transformed to adjacent joints. In this by following formula [11]:
Table 3. Equivalent joint diameter in correspondence To solve a constrained optimization problem its
with the bolt diameter objective function should be modified in such a way
that the constrained problem should be converted to
Bolt diameter 12 16 20 24 27 an unconstrained one, with a modified objective
(mm) function. The modified objective function φ is defined
Equivalent joint 50 75 88 110 130 as [7]:
diameter (mm)
φ = W (1 + r C ) (13)
4Pmax
d= (8) Ne Nj
0.6π Fy C = ∑ ( g σ , i + g λ , i ) + ∑ g δ, (14)
j
i =1 j =1
where Pmax and Fy are maximum force applied to this
joint and yield stress of the bolt, respectively. Then by where C and r̄ are the penalty function and the
the use of design Table 3, appropriate diameter Dj for coefficient of penalty function, respectively.
this joint is selected, and its equivalent weight is The optimum shape design of space structures is a
calculated by the following equation [11]: minimization problem, and hence the fitness function
must be chosen such that the higher the weight of an
W j = 0.9 π ρ j D 3j (9) individual, the lower is its fitness and vice-versa. The
following relation is selected as the measure of
where ρj is the material density of jth joint. fitness [7]:
In this study, the AISC code provisions are
employed for the stress limits and local buckling Fi = φmax + φmin − φi (15)
criteria [12]. All its recommendations for stresses and
slenderness ratios in elements such as tension, where φmax , φmin and φi are the maximum and
compression, bending stresses and their combination minimum modified objective function value in a
are considered. The stress constraint gσ and the generation and the modified objective function value
slenderness ratio constraint gλ are taken as: of the ith individual, respectively.
where δk, δu are the displacement of joints in the kth In this research, the flat roof snow load is taken as
load combination and the allowable displacement, 150 kg/m2, and the roof slope factor is determined as
respectively. follows [14]:
Table 4. The comparison table for 25-bar space truss Table 5. Available pipe profiles
Optimal cross-sectional areas (cm2) No D(cm) t(cm) No D(cm) t(cm)
1 6.03 0.29 9 21.91 0.45
Element Erbatur et al. [21]
2 7.61 0.29 10 27.30 0.50
group GAOS-Level1 GAOS-level2 MGA 3 8.89 0.32 11 32.39 0.50
1 0.65 0.65 0.65 4 10.80 0.36 12 35.56 0.80
2 6.45 7.74 1.29 5 11.43 0.36 13 40.64 0.88
3 21.94 20.65 21.94 6 13.97 0.40 14 45.72 1.00
4 1.29 0.65 0.65 7 16.83 0.45 15 50.80 0.88
5 3.87 7.10 13.55 8 19.37 0.45 16 55.88 0.88
6 7.09 5.81 7.10
7 5.80 20.58 3.23
8 19.35 21.94 21.94 convergence history of this example using the GA and
Weight (kg) 233.86 224.12 220.58 the MGA is presented in Fig. 7.
7. NUMERICAL EXAMPLES
Z
1 2 After many experiments, the population size and the
number of generation in each stage were taken as 50
and 100, respectively. The probability of crossover and
mutation were taken as 0.9 and 0.005, respectively,
and to avoid the change of MGA operations into a
perfect random search method, for the presence of
3 4
6 joints a mutation rate 0.001 is employed. The modulus
5
of elasticity, material density and yield stress are taken
as 2.1 106 kg/cm2 , 7850 kg/m3 and 2400 kg/cm2,
respectively.
8
7
7.1. 16 16 Double Layer Grid
X
This is a square-on-square space structure with 605
9 joints, the topology of which is shown in Fig. 8. The
10
Y joint spacing in the top and bottom chord is 3 m, while
the depth of the double layer grid is 3.75 m. In this
example, the number of member groups after a
Figure 6. Configuration of 25-bar space truss. preliminary analysis was assumed as 3 each for column,
bottom, web and top grids, which resulted in 12 design
700 variables. Cross-sectional area of members is selected
600 from the pipe profiles available in Table 5. The
optimum topology design is compared with the
Minimum weight
500
GA optimized ground structure in Table 6. Comparing
MGA
400 results of optimization, the weight of the optimum
shape becomes 10.29% less than the ground structure.
300
The optimum shapes of structure are shown in Figs. 9 to
200 13.
100 The resulting configuration is interesting, and the
0 5 10 15 20 25 30 35 40 joints in some places are omitted. If the double layer
Number of generations
grid is replaced by a flat squire plate with semi fixed
Figure 7. Convergence history for 25-bar space truss. edges and distributed uniform loads, it can be seen that
the removed joints are related to the points of zero
Table 6. Properties of the optimum ground structure and optimum topology structure
Optimum ground structure Optimum topology structure
Number of joints 605 496
Number of elements 2108 1620
Weight (kg) 72938 66190
Generations 68 94
3.75 m
7m
15@3 m
Table 7. Properties of the optimum ground structure and optimum topology structure
Optimum ground structure Optimum topology structure
Number of joints 577 521
Number of elements 2400 1824
Weight (kg) 48038 36996
Generations 57 86
bending moment of the flat plate. Thus the optimal shown in Fig. 14.a. This optimum shape is similar to
places of the joints are analogous to the variation of the optimum topology design of a plate shown in Fig.
bending moment and internal stresses of similar flat 14.b. that was obtained in [22], but the optimum shape
plates. is not practical. In order to achieve a practical
If presence and absence of all of the joints are topology, existence of perimeter nodes in top and
considered as variable, some perimeter joints are bottom grids of space structure was not considered as
removed from the optimum shape of space structure variable.
7.2. Double Layer Space Pyramid Figure 19. Structure optimized with a second order
A double layer space pyramid with 577 joints, the polynomial.
topology of which is shown in Fig. 15 is selected. In
this example, the number of member groups after a the optimum shape is logic and the distribution of joints
preliminary analysis was assumed as 3 each for bottom, is analogous to the stress distribution of a continuum
web and top grids, which resulted in 9 design variables. pyramid.
Cross-sectional area of members is selected from the
pipe profiles available in Table 5. The optimum 7.3. Dome Geometry Optimization
topology design is compared with the optimized The dome shown in Fig. 18 has 5 meridional and 8
ground structure in Table 7. The weight of the optimum circumferential divisions. The rise and the span of the
shape is 23% better than the optimal weight of the dome are 8 m and 50 m, respectively. Members of the
ground structure. The optimum shapes of structure dome are collected in eighteen distinct groups with
are shown in Figs. 16 and 17. It can be observed that the manner of the members grouping being similar to
REFERENCES
[1] Parke, G. and Disney, P., Space Structures 5, Thomas
Telford, London, 2002.
[2]. Rajeev, S. and Krishnamoorthy, C.S., Discrete
optimization of structures using genetic algorithms,
Journal of Structural Engineering ASCE, 1992, 118(5),
1233–1238.
[3]. Goldberg, D.E., Genetic algorithms in search,
Figure 20. Structure optimized with a third order polynomial. optimization and machine learning. Massachusetts,
Addison-Wesley publishing Company, 1989.
[4]. Adeli, H. and Cheng, N.-T., Integrated Genetic
Fig. 2. Cross-sectional area of members is selected
Algorithm for Optimization of Space Structures, Journal
from the pipe profiles available in Table 8. After of Aerospace Engineering, ASCE, 1993, 6(4), 315–328.
ground structure optimization, it is concluded that [5] Adeli, H. and Cheng, N.-T., Augmented Lagrangian
second and third order polynomial weigh less than Genetic Algorithm for Structural Optimization, Journal
parabolas with other orders. For the optimum shapes of Aerospace Engineering, ASCE, 1994, 7(1), 104–118.
[6] Adeli, H. and Cheng, N.-T., Concurrent Genetic
that are shown in Figs. 19 and 20, top and bottom Algorithm for Optimization of Large Structures,
boundary of radius of circumferences and optimized Journal of Aerospace Engineering, ASCE, 1994, 7(3),
value of radius are given in Table 9. The weight 276–296.
percentage reduction of second and third order [7] Krishnamoorthy, C.S., Prasanna, V.P. and Sudarshan, R.,
Object-oriented framework for genetic algorithm with
polynomials are 19.8% and 32%, respectively. application to space truss optimization, Journal of
Computing in Civil Engineering, 2002, 16(1), 66–75.
8. CONCLUSION [8] Farsangi, E., Topological optimization of double layer
In this paper, an modified genetic algorithm (MGA) grids using genetic algorithms: Proceedings of 5th
was implemented for optimizing shape design of large International Conference on Space Structure, Telford,
space structures to confirm the ability of the GA in London, 2002, 459–468.
[9] Topping, B.H.V., Montero, G. and Montenegro, K.,
shape structural optimization. The following remarks Proceeding of the 8th International Conference on
can be made: Computational Structures Technology, Civil-Comp
This paper proposes a methodology for optimum press, Scotland, 2006.
topology design of large space structures, which takes [10] Kaveh, A. and Kalatjari, V., Topology optimization of
trusses using genetic algorithm, force method and graph
into account discrete variations in the member cross-
theory, International Journal for Numerical Methods in
sectional areas and the presence or absence of the Engineering, 2003, 58, 771–791.
joints. The proposed methodology provides a practical [11] Dianat, N., Space Structures Company, Private
and scientific basis to find the optimum shape of space communications, Tehran, Iran, 2005.
structures which can replace the trial and error [12] American Institute of Steel Construction, AISC Manual,
2005.
methods for removal of joints in these structures. The [13] Iranian Code of Practice for Seismic Resistant Design of
presented examples have demonstrated that by Building, BHRC Publication, Tehran, 2006.
employing the methodology efficient topology for [14] American National Standards. Minimum design loads for
space structures can be found. buildings and other structures, (ANSI A58.1), 1982.
[15] Ventsel, E. and Krauthammer, Th., Thin Plates and [19] Spears, W. M., Adapting crossover in a genetic
Shells: Theory, Analysis, and Application, Marcel algorithm, European Journal of Operational Research,
Dekker, New York, 2001. 1992, 4, 392–404.
[16] Kaveh, A. and Kalatjari, V., Genetic algorithm for [20] Coley, D., An introduction to genetic algorithms for
discrete-sizing optimal design of trusses using the force scientists and engineers, World Scientific Publishing,
method, International Journal for Numerical Methods in London, 1999.
Engineering, 2002, 55, 55–72. [21] Erbatur, F. Hasancebi, O. Tutuncu, I. and Kilic, H.,
[17] Salajegheh, E. and Gholizadeh, S. Optimum design of Optimal design of planar and space structures with
structures by an improved genetic algorithm using neural genetic algorithms, Journal of Computers and Structures,
networks, Advances in Engineering Software, 2005, 2000, 75, 209–224.
36(11–12), 757–767. [22] Liang, Q and Steven, G., A performance-based
[18] Salajegheh, E. and Heidari, A. Optimum design of optimization method for topology design of continuum
structures against earthquake by adaptive genetic structures with mean compliance constraint, Journal of
algorithm using wavelet networks, Structural and Computer methods in applied mechanics and
Multidisciplinary Optimization, 2004, 28(4), 277–285. engineering, 2001, 191, 1471–1489.