Professional Documents
Culture Documents
Thai Et Al. - 2017 - Optimum Design of Cable Nets by Using Genetic Algorithm
Thai Et Al. - 2017 - Optimum Design of Cable Nets by Using Genetic Algorithm
DOI 10.1007/s13296-017-9025-6
ISSN 1598-2351 (Print)
ISSN 2093-6311 (Online)
www.springer.com/journal/13296
Abstract
This paper presents a generalized procedure to optimize cable nets by using the Genetic Algorithm (GA). The finite element
program employing 2-, 3- and 4-node isoparametric curved cable elements is utilized to deal with the nonlinear behavior of
cable nets under static loads. The allowable stress and the maximum displacement are considered as optimization constraints
while the minimum volume is selected as an objective function. To validate the accuracy and efficiency of the proposed
procedure, four optimization examples originated from nonlinear analysis problems of cable nets are introduced.
Keywords: cable nets, nonlinear analysis, finite element method, optimization, genetic algorithm
programming, where a few loading conditions as well as of design variables and practical constraints. Recently,
initial prestress values were the design parameters. Wang Yamamoto et al. (2011), Koohestani (2012) and Lee et al.
(1986) studied the effects of self-weight, concentrated (2014) successfully utilized GA when dealing with
loads and the distribution of loads on the optimal shape of tensegrity structures.
isolated cable. Another prominent optimization research The goal of this paper is to present a generalized
for cable structures was carried out by Braga and D’Asdia procedure to seek the optimal volume of cable nets based
(1981). In their work, a method for the optimization of on GA by using the geometrically nonlinear finite element
plane cable with distributed loads was proposed. method. In the first content of this paper, the formulation
From the previous optimization researches of cables, it of a spatially distributed cable member under the pretension
can be seen that the analytical method is the dominant and the external forces is developed by using the 2-, 3-
approach to handle the highly nonlinear effect of cables. and 4-node isoparametric curved cable elements. In
This approach could only be put into practice efficiently addition, the linear and nonlinear tangent stiffness
for the quite simple model of cable structures like isolated matrices and the force vector are evaluated based on the
cables or plane cable systems. For a more complicated total Lagrangian formulation. The application of GA and
and practical system such as cable net which composes a optimal formulations are discussed in the next part. In
number of isolated cable elements, it is unreasonable to numerical examples, the four types of cable nets are
apply the previous analytical methods to the optimization introduced and the accuracy of isoparametric curved
problem since it would be expected to require a lot of cable elements is demonstrated by comparing the obtained
computational effort to handle a complex problem. To results with the previous published ones. For the optimization
overcome difficulties from the analytical approaches for a problem, allowable stress and limited displacement are
complicated optimization problem, it is necessary to use considered simultaneously as the design constraints. The
the finite element method as an efficient approach. optimal results are also investigated to evaluate the
There have been a number of robust algorithms to solve effects of constraint values as well as the numbers of
optimization problems. Among them, the Genetic Algorithm design variables on optimal results. Some remarkable
(GA) is one of the most famous techniques widely conclusions are presented in the final part of paper.
employed. The essence of GA is generally based on the
natural evolution (Yang, 2014). Up to now, a huge number 2. Cable Formulation
of researches in different fields have proven that GA is an
effective technique to handle searching as well as optimization 2.1. Incremental equation of elastic continuum
problems, especially when the number of possibilities is The configuration of a curved cable is presented from
enormous. For example, Lute (2011) obtained the relative the theory of elastic continuum since the cable undergoes
cost of the cable-stayed bridge including a great number large deformation under self-weight or external load. The
{ }
T
U = u11 , u12 , u31 ,L, u1n , u2n , u3n (8)
where 0e and 0η are the linear and nonlinear components,
respectively, of the incremental Green strain increment,
The linear and nonlinear strains in Eq. (4) can be
and the infinitesimal lengths 1ds and 2ds are expressed as
expressed in matrix manner as follows
follows:
0e = BL .U (9a)
( ) ( )( )
1 2
ds = d 0 xi + d 01ui d 0 xi + d 01ui (2a)
0η = B NL .U (9b)
( ds ) = ( d x + d u )( d )
2 2 0 2 0
i 0 i xi + d02ui (2b)
where B L = B0 + Bu and B NL are the linear and
2 1 nonlinear strain-displacement matrices, respectively. The
where ui = − 0 ui 0 ui
denotes the incremental displacement.
small strain-displacement relationship matrix B0 can be
The incremental equation for an elastic continuum can be
written as
expressed by the principle of virtual work as
B0 = ⎡⎣B10 B02 ... B0n ⎤⎦
∫ V 0 Sδ ( 0 ε ) d
0
0
V + ∫0 1
V 0
Sδ ( 0η ) d 0V =δ ( R) − δ ( R)
2
0
1
0
(10)
(3) where
0
where d V is the differential volume; 0S is the component
⎡ dx dN dx2 dN i dx3 dN i ⎤
of the Second Piola-Kirchhoff stress increment tensor and Bi0 = ⎢ 0 1 0 i 0 0 0 ⎥ (11)
⎣ ds ds ds ds ds 0ds ⎦
d 0 xi dui dui d 01ui 1 dui dui
0e = + 0η = (4) in which
0 0 0 0
ds ds ds ds 2 0 ds 0 ds
dNi d ξ dNi 1 dNi
= =
In the right-hand side of the Eq. (3), δ ( )
2
0R is the 0
ds 0
ds d ξ J dξ
(12a)
δ ( R ) = ∫0 Sδ ( e ) d V
1
V
1
0 0
0
where
1
1 T
⎡ du dN du dN du dN ⎤
K NL = ∫ 0 SAB NL B NL Jdξ (17)
B iu = ⎢0 1 0 i 0 2 0 i 0 3 0 i ⎥ (15) −1
⎣ ds ds ds ds ds ds ⎦
1
1 1 T
The nonlinear strain-displacement matrix is defined as 0F = ∫ 0 SAB L Jdξ (18)
−1
⎡ dN1 dN n ⎤
⎢ dξ 0 0 0 0 ⎥ where E and A are the Young’s modulus and the cross-
⎢ dξ ⎥ sectional area of cable element, respectively. In this study,
1⎢ dN1 dN n ⎥ the well-known Newton-Raphson method is employed to
B NL = ⎢ 0 0 L 0 0 ⎥
J⎢ dξ dξ ⎥
deal with the geometric nonlinear equations.
⎢ dN1 dN n ⎥
⎢ 0 0 0 0 ⎥ 3. Design Optimization
⎣ dξ dξ ⎦
(16) Genetic Algorithm (GA) is a famous computational
searching technique inspired by the Darwin’s theory of
Substituting above expressions into Eq. (3), the
evolution for the process of natural selection. Until now,
equilibrium equation of cable is obtained as follows
there have been a plenty of research by GA as a stochastic
algorithm to seek good or optimal solutions in various
( K L + K NL ) U = 20 R − 01F (16)
fields of science and engineering (Yang, 2014). A great
number of papers and text books presented detail descriptions
in which K L and K NL denote the linear and nonlinear
about this technique. Therefore, this section discusses
incremental stiffness matrices, respectively; 01F is the
briefly features of GA and focuses on the implementation
vector of nodal point force at the configuration C1; 20 R
of method to optimization problems of cable nets.
is the vector of externally applied nodal point loads at
configuration C2. The Gauss integration scheme is
3.1. Genetic algorithm (GA)
employed to perform the numerical evaluation of these
The general procedure of GA can be succinctly summarized
terms as
as revolutions of a population through generations to find
1 the most fitted individual with respect to the given conditions.
T
KL = ∫ EAB L B L Jdξ (16) An individual or so-called chromosome is the representation
−1 of a possible solution expressed by an encoded binary
important features of GA. The simple illustration of between the constraint values of allowable stress, limited
structure for a chromosome and its sequence of variable displacement and the optimal volumes are discussed in
is shown in Fig. 4. To determine the required binary examples of plane and spatial cable nets. Besides that, the
strings to represent for the encoded variable, it is effects of number of design variables on the optimal
necessary to define the number of its possibilities, which results and the optimization convergence rate are also
are expressed as the side constrains or the range of design investigated in examples of the hyperbolic paraboloid
value. The side constrains are expressed by the maximum cable and saddle nets. For all numerical examples, 2-, 3-
and minimum values of design space as and 4-node elements are employed to model the cable
nets and the total number of elements are similar in all
Amin ≤ Ai ≤ Amax (21a) cases
Table 2. Constraint parameters for plane cable net Table 3. Optimal cross-sectional areas A (mm2) and
pretension forces F (kN) with different values of
Constant Condition allowable stress and limited displacement
Allowable stress σ a =200, 410, 600 MPa
σa δ limit
Limited displacement δ limit =0.2, 0.45, 0.7 m A A+F A+2F
(MPa) (m)
Cross-sectional area 1 mm2 ≤ A ≤ 500 mm2, ∆A =1 mm2
200 0.45 A = 303 A= 299 A = 299
Pretensioned force 1 kN ≤ F ≤ 100 kN, ∆F =1 kN
F= 4.88 Fi = 5.26
Fh = 4.13
properly. Figures 6 and 7 show the effects of stress and 410 0.20 A = 299 A = 149 A = 148
displacement constraints, respectively, on the optimal F = 49.99 Fi = 55.24
results. It is seen that the results in such case having only Fh = 22.89
cross-sectional area variable and case having it together 410 0.45 A = 145 A = 145 A = 145
with the pretension force variable become significantly F = 24.676 Fi = 34.63
different when the magnitude of displacement is relatively Fh = 13.65
small and the value of allowable stress is relatively large. 410 0.70 A = 145 A = 142 A = 142
The efficiency of solutions is evaluated by employing two
F = 10.35 Fi = 12.51
parameters: Δ S = σ max σ a and Δ D = δ max δ limit
Fh = 6.05
which indicate how the constraints are involved in the
optimization procedure. It is seen in Table 4 that the 600 0.45 A = 145 A= 99 A= 99
allowable stress is fully utilized in case that the pretension F = 36.87 Fi = 50.59
force is considered as a design variable. While the Fh = 9.36
displacement ratio Δ D becomes critical when the cross-
sectional area is the unique design variable.
cable segments arranged in a 4 m×4 m quadrilateral form
4.2. Spatial net as shown in Fig. 8. The cable net is symmetric in both x-
The spatial cable net considered has 38 pretension and y-directions and the z-coordinates of nodes for a
1190 Son Thai et al. / International Journal of Steel Structures, 17(3), 1183-1198, 2017
Figure 6. Optimal cross-sectional areas for plane cable net Figure 7. Optimal cross-sectional areas for plane cable net
in case that δ limit =0.45 m and σ a =200, 410, and 600 in case that σa =410 MPa and δ limit =0.2, 0.45, and 0.7 m.
MPa.
Table 7. Optimal cross-sectional areas (mm2), pretension forces (kN) and volumes (10−2 m3) for spatial cable net with
different values of allowable stress and limited displacement
σ a (MPa) δlimit (m) 2A 5A 2A+2F 5A+5F
300 0.045 A1 = 92 A1 = 100 A1 = 60 A1 = 10
A2 = 436 A2 = 80 A2 = 340 A2 = 10
A3 = 90 F1 = 0 A3 = 60
A4 = 360 F2 = 90 A4 = 330
A5 = 440 A5 = 370
F1 = 0
F2 = 0
F3 = 0
F4 = 100
F5 = 130
V = 3.9112 V = 3.5381 V = 2.9568 V = 2.6590
362 0.020 A1 = 85 A1 = 190 A1 = 140 A1 = 10
A2 = 1195 A2 = 80 A2 = 400 A2 = 40
A3 = 90 F1 = 20 A3 = 120
A4 = 720 F2 = 120 A4 = 300
A5 = 1170 A5 = 370
F1 = 0
F2 = 10
F3 = 50
F4 = 100
F5 = 140
V = 9.3781 V = 7.3392 V = 4.0384 V = 2.7091
362 0.045 A1 = 90 A1 = 80 A1 = 60 A1 = 10
A2 = 380 A2 = 110 A2 = 280 A2 = 10
A3 = 80 F1 = 0 A3 = 50
A4 = 290 F2 = 90 A4 = 250
A5 = 360 A5 = 300
F1 = 0
F2 = 0
F3 = 10
F4 = 90
F5 = 120
V = 3.4681 V = 3.0209 V = 2.5204 V = 2.0852
362 0.070 A1 = 80 A1 = 80 A1 = 20 A1 = 10
A2 = 360 A2 = 70 A2 = 250 A2 = 10
A3 = 70 F1 = 0 A3 = 10
A4 = 300 F2 = 90 A4 = 230
A5 = 360 A5 = 270
F1 = 0
F2 = 0
F3 = 0
F4 = 80
F5 = 70
V = 3.2161 V = 2.9243 V = 2.3429 V = 1.8503
420 0.045 A1 = 120 A1 = 70 A1 = 40 A1 = 10
A2 = 340 A2 = 110 A2 = 270 A2 = 10
A3 = 70 F1 = 10 A3 = 60
A4 = 250 F2 = 90 A4 = 220
A5 = 350 A5 = 250
F1 = 0
F2 = 0
F3 = 0
F4 = 90
F5 = 100
V = 3.4456 V = 2.7540 V = 2.2865 V = 1.8354
1194 Son Thai et al. / International Journal of Steel Structures, 17(3), 1183-1198, 2017
47 -1600 4.63 0.00 -22.52 4.64 0.00 -22.17 4.64 0.00 -22.32 4.64 0.00 -22.52 4.63 0.00 -22.52 4.63 0.00 -22.52 4.63 0.00 -22.52
48 -1667 4.54 0.00 -45.87 4.54 0.00 -45.51 4.54 0.00 -45.66 4.55 0.00 -45.89 4.54 0.00 -45.86 4.54 0.00 -45.86 4.54 0.00 -45.86
52 -600 -0.92 0.00 5.86 -0.96 0.00 6.27 -0.94 0.00 6.10 -0.92 0.00 5.86 -0.92 0.00 5.87 -0.92 0.00 5.87 -0.92 0.00 5.87
72 -1848 3.85 -0.78 -30.10 3.83 -0.76 -29.82 3.84 -0.77 -29.94 3.85 -0.78 -30.12 3.85 -0.78 -30.10 3.85 -0.78 -30.10 3.85 -0.78 -30.1
81 -2867 4.10 2.80 11.15 4.10 2.80 11.24 4.10 2.80 11.21 4.11 2.80 11.16 4.10 2.80 11.16 4.10 2.80 11.16 4.10 2.80 11.16
85 -1032 -5.40 1.87 32.15 -5.44 1.88 32.38 -5.42 1.88 32.28 -5.40 1.87 32.17 -5.40 1.87 32.16 -5.40 1.87 32.16 -5.40 1.87 32.16
Optimum Design of Cable Nets by Using Genetic Algorithm 1197
Table 12. Constraint parameters for saddle net (2) When the pretension forces are considered as design
Constant Condition variables in optimization procedure, the stress constraints
are fully utilized.
Allowable stress σ a =258.5 MPa
(3) The appropriate pretension forces applied to cable
Limited displacement δ limit =0.00972 m
nets can help the structure reduce the displacement and
Cross-sectional area 100 mm2 ≤ A ≤ 400 mm2, ∆A =1 mm2
then the optimal volumes can be obtained as well.
Table 13. Optimal cross-sectional areas (mm2) and
(4) The pretension forces play a significant role on
volumes (10−1 m3) for saddle net with different numbers obtaining the optimal volume when the displacement
of cross-sectional area design variables constraint is relatively small or the allowable stress is
relatively large.
A 2A 16A (5) The results obtained from the proposed optimization
A= 305 A1 = 311 A1 = 301 procedure can be used as a reliable information for the
A2 = 258 A2 = 368 preliminary design of cable nets.
A3 = 270
A4 = 262 Acknowledgments
A5 = 229
A6 = 197 This research was supported by a grant (17AUDP-
A7 = 263 B100343-03) from Architecture & Urban Development
A8 = 229 Research Program funded by Ministry of Land, Infrastructure
A9 = 332 and Transport of Korean government.
A10 = 327
A11 = 370 References
A12 = 259
A13 = 296 Ali, H. M. and Abdel-Ghaffar, A. M. (1995). “Modeling the
A14 = 264 nonlinear seismic behavior of cable-stayed bridges with
A15 = 261 passive control bearings.” Computers and Structures, 54,
A16 = 222 pp. 461-492.
Braga, F. and D’Asdia, P. (1981). “Analysis and optimization
V = 0.2176 V = 0.2032 V = 0.1989
of plane cable trusses.” Journal of Structural Mechanics,
9, pp. 253-269.
variables. It is evident that the case with 16 cross- Chen, Z. H., Wu, Y. J., Yin, Y., and Shan, C. (2010).
sectional areas needs the smallest generation to obtain the “Formulation and application of multi-node sliding cable
optimal volume. element for the analysis of suspen-dome structures.”
Finite Elements in Analysis and Design, 46, pp. 743-750.
Coyette, J. P. and Guisset, P. (1988). “Cable network
5. Conclusions analysis by a nonlinear programming technique.”
Engineering Structures, 10, pp. 41-46.
Based on the combination of the geometric nonlinear Desai, Y. M., Popplewell, N., Shah, A. H., and Buragohain,
finite element analysis and GA, a generalized approach to D. N. (1988). “Geometric nonlinear static analysis of
optimize the volume of cable nets is proposed in this cable supported structures.” Computers and Structures,
study. The suggested procedure has ability to find the 29, pp. 1001-1009.
optimal volumes of cable nets. Cross-sectional areas and Gambhir, M. L. and Batchelor, B. (1979). “Finite element
pretension forces are taken into consideration as design study of the free vibration of 3-D cable networks.”
variables, while stress and displacement conditions are International Journal of Solids and Structures, 15, pp.
considered as optimal constraints. The accuracy of employing 127-136.
isoparametric cable elements is demonstrated through Huang, N. C. (1975). “Minimum weight design of elastic
four numerical examples. Optimization problems are also cables.” Journal of Optimization Theory and Applications,
introduced to show the efficiency of the optimization 15, pp. 37-49.
Jayaraman, H. B. and Knudson, W. C. (1981). “A curved
procedure. Consequently, the following conclusions are
element for the analysis of cable structures.” Computers
drawn as follows: and Structures, 14, pp. 325-233.
(1) For cable nets under the concentrated point loads, Kim, N. I., Thai, S., and Lee, J. (2016). “Nonlinear elasto-
the results given by the minimum number of 2-, 3- and 4- plastic analysis of slack and taut cable structures.”
node elements are in great agreement 13with those from Engineering with Computers, 32, pp. 615-627.
other references and do not show any noticeable Koohestani, K. (2012). “Form-finding of tensegrity structures
difference between each element. However, it should be via genetic algorithm.” International Journal of Solids
noted here that all cables in net structures are assumed to and Structures, 49, pp. 739-747.
be the taut ones. Kwan, A. S. K. (1998). “A new approach to geometric
1198 Son Thai et al. / International Journal of Steel Structures, 17(3), 1183-1198, 2017
nonlinearity of cable structures.” Computers and Structures, Swaddiwudhipong, S., Wang, C. M., Liew, K. M., and Lee,
67, pp. 243-252. S. L. (1989). “Optimal pretensioned forces for cable
Lee, S., Woo, B. H., and Lee, J. (2014). “Self-stress design networks.” Computers and Structures, 33, pp. 1349-1354.
of tensegrity grid structures using genetic algorithm.” Thai, H. T. and Kim, S. E. (2011). “Nonlinear static and
International Journal of Mechanical Sciences, 79, pp. 38- dynamic analysis of cable structures.” Finite Elements in
46. Analysis and Design, 47, pp. 237-246.
Lewis, W. J., Jones, M. S., and Rushton, K. R. (1984). Thai, S., Kim, N. I., and Lee, J. (2017a). “Isogeometric cable
“Dynamic relaxation analysis of the nonlinear static elements based on B-spline curves.” Meccanica, 52, pp.
response of pretensioned cable roofs.” Computers and 1219-1237.
Structures, 18, pp. 989-997. Thai, S., Kim, N. I., and Lee, J. (2017b). “Free vibration
Lute, V., Upadhyay, A., and Singh, K. K. (2011). “Genetic analysis of cable structures using isogeometric approach.”
algorithms-based optimization of cable stayed bridges.” International Journal of Computational Methods, 14, pp.
Journal of Software Engineering and Applications, 4, pp. 1750033.
571-578. Thevendran, V. and Wang, C. M. (1985). “Minimum weight
Nelder, J. A. and Mead, R. (1965). “A simplex method for design of cables with supports at different levels.”
function minimization.” The Computer Journal, 7, pp. International Journal of Mechanical Sciences, 27, pp.
308-313. 519-529.
Ozdemir, H. (1979). “A finite element approach for cable Tibert, G. (1999). Numerical analyses of cable roof
problems.” International Journal of Solids and structures, Royal Institute of Technology, Department of
Structures, 15, pp. 427-437. Structural Engineering.
Salehi Ahmad Abad, M., Shooshtari, A., Esmaeili, V., and Wang, C. M., Pulmano, V. A., and Lee, S. L. (1986). “Cable
Naghavi Riabi A. (2013). “Nonlinear analysis of cable optimization under self-weight and concentrated loads.”
structures under general loadings.” Finite Elements in Journal of Structural Mechanics, 14, pp. 191-207.
Analysis and Design, 73, pp. 11-19. Yamamoto, M., Gan, B. S., Fujita, K., and Kurokawa, J.
Selleri, F. and Spadaccini, O. (1977). “Optimal design of (2011). “A genetic algorithm based form-finding for
prestressed plane cable structures.” Journal of Structural tensegrity structure.” Procedia Engineering, 14, pp.
Mechanics, 5, pp. 179-205. 2949-2956.