Advanced
Warping thermal deformation constraint for optimization of a blade stiffened composite panel using GA
Warping thermal deformation constraint for optimization of a blade stiffened composite panel using GA
International Journal of Aeronautical and Space Sciences. 2013. Dec, 14(4): 334-340
Copyright © 2013, The Korean Society for Aeronautical & Space Sciences
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/bync/3.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
  • Received : September 11, 2013
  • Accepted : November 12, 2013
  • Published : December 30, 2013
Download
PDF
e-PUB
PubReader
PPT
Export by style
Share
Article
Author
Metrics
Cited by
TagCloud
About the Authors
Akira Todoroki
atodorok@ginza.mes.titech.ac.jp
Takumi Ozawa

Abstract
This paper deals with the optimization of blade stiffened composite panels. The main objective of the research is to make response surfaces for the constraints. The response surface for warping thermal deformation was previously made for a fixed dimension composite structure. In this study, the dimensions of the blade stiffener were treated as design variables. This meant that a new response surface technique was required for the constraints. For the response surfaces, the lamination parameters, linear thermal expansions and dimensions of the structures were used as variables. A genetic algorithm wasadopted as an optimizer, and an optimal result, which satisfied two constraints, was obtained. As a result, a new response surface was obtained, for predicting warping thermal deformation.
Keywords
1. Introduction
For laminated Carbon Fiber Reinforced Polymer (CFRP) composites, optimization of the stacking sequences, and of the dimensions of the structures is essential, because of their strong anisotropic properties. Miki [1] and Fukunaga [2] both proposed a graphical optimization method using the lamination parameters. For practical laminated CFRP structures, however, the available fiber angles are usually limited to a small set of fiber angles, because of a lack of experimental data. In addition, multiple constraints on the fiber angles exist from specific empirical rules, such as the four-contiguous-ply rule, to prevent large matrix cracking. These facts make the optimization of the stacking sequences a combinatorial optimization problem, with combinatorial constraints.
For optimization of the stacking sequence of laminated CFRP composites, Genetic Algorithms (GAs) have been adopted in many research programs [3 - 15] . Narita proposed a layer-wise optimization method for stacking sequence design [16] .
In previous research, we proposed a fractal branch and bound method (FBB), for optimizing the stacking sequence of laminated carbon fiber composites [17 , 18] . This method employs a quadratic polynomial for the response surface using the lamination parameters, such as the buckling load, to approximate the objective functions. The method involves low computational costs, and a practical optimal result can be obtained in milliseconds, by means of a deterministic process. This method has been successfully applied to the problem of establishing the maximization of the buckling load of a laminate [17 , 18] , and for determining the maximization of the flutter speed limit [19] with constraints.
For a stiffened panel made from a laminated CFRP structure, the dimensions of the stiffener and panel have to be simultaneously optimized, in addition to optimizing the stacking sequences of the panel and stiffeners. We havepublished papers that deal with the modified efficient global optimization method using a multi-objective GA [20] : the improvement in the objective function, and the possibility of satisfying the constraints are treated as two objective functions.
To optimize, while preventing any fracture of a laminated CFRP composite structure, it is difficult to obtain an approximate surrogate model to predict the fracture load, when checking the fracture constraints. When a quadraticpolynomial was used to predict the fracture of the CFRP laminate, the surrogate model provided poor estimates for this condition. In a previous paper, this problem was solved using the Kriging response surface [21] . This means that the FBB method cannot be applied to the fracture constraint, because it requires a response surface approximation of a quadratic polynomial. We proposed a new response surface technique to approximate the warping thermal deformation caused by the curing process [22] , although the method was limited to the fixed dimensions of a blade stiffened composite structure.
In the present study, therefore, a GA is adopted to simultaneously optimize the stacking sequences and dimensions of a target composite structure, without using the FBB method. To optimize the number of plies for the GA, intron genes that represent the empty position of a chromosome have been proposed in [8] . To simplify the implementation, the intron genes were not used here. Therefore, a new simple crossover operation between chromosomes of different lengths is proposed. The main objective of this paper is to investigate the effectiveness of the response surface approximation of the warping thermal deformation for the blade stiffened composite structures, against the various dimensions of the stiffener. For the optimization process in the present study, the buckling load and warping of the thermal deformation of the target stiffened laminated composite structure are treated as given constraints. Here, the objective function of the GA is reduction in the weight of the laminated composite structure with the two constraints.
2. Optimization problem
In the present study, a blade-stiffened laminated composite panel is dealt with, as shown in Fig. 1 . The blade stiffened composite structure consists of four unit structures. The design variables and the material properties are shown in Tables 1 and 2 , respectively
The objective function of the optimization problem is the weight of the unit structure of the blade stiffened composite panel. Because the composite plates are thin, the weight of the unit composite structure can be calculatedapproximately, as follows:
Lager Image
where, tp is the ply thickness (0.132 mm), ρ is the specific gravity of the CFRP ( ρ =1.3), nρ is half the number of plies
Lager Image
Blade stiffened composite structure
Range of design variable
Lager Image
Range of design variable
Material properties
Lager Image
Material properties
in the panel, and ns is half the number of plies in the blade stiffener. The objective is to minimize the weight with the two constraints: the maximum warping caused by the thermal deformation, and the buckling stress.
For the analysis of thermal deformation and the buckling load, the commercially available FEM code ANSYS ver.12 was used. Linear laminated structural shell elements were adopted for the analyses. The total number of elements was4,160. Fig. 2 shows the boundary condition for the buckling analysis. The reference compression stress of 1.75 MPa was applied along the panel edge, as shown in Fig. 2 . The composite panel had a simply supported boundary. As a buckling constraint, a buckling stress smaller than 1.75 MPa must be avoided.
For the warping analysis caused by thermal deformation, the displacement and rotation of the bottom center node of the plate was fixed. The applied temperature change was ΔT=−228℃, which came from the curing process. The maximum affordable thermal warping deflection was set to 2
Lager Image
Boundary condition and loading condition of the buckling analysis
Lager Image
Thermal deformation types.
mm. The change in the stacking sequence of the stiffener and panel causes a mismatch of the linear thermal expansion coefficients. A thermal expansion mismatch may cause a large thermal deformation of the target structure; from afabrication point of view, this should be prevented.
Thermal deformation was checked using quadratic response surfaces. The process for using the response surfaces to check the thermal deformation constraint is described in detail in [22] . Although all dimensions of the composite structure are fixed in [22] , the dimensions are variables in this study, which is one of the original points of the study. Thermal deformation is classified into two types, as shown in Fig. 3 . Quadratic polynomial responses surfaces were made for the two types of deformation modes.
We considered the case where the x - and y -coordinates were the longitudinal and transverse directions of the laminated plate. When αx P < αx S and αy P > αy S are satisfied, the deformation mode is Type A (like a saddle). When αx P > αx S and αy P < αy S are satisfied, the deformation mode is Type B (like a saddle). When αx P αx S and αy P αy S are satisfied, the deformation mode is a simple extension type, because both the plate and stiffener deform similarly. Although the classification method is the same as that shown in [22] , the method was investigated in this study.
It can be readily concluded that the small differences in the linear thermal expansion coefficients of the plate and stiffener are easily warped. Thus, the simple extension mode was neglected in this study.
After the classification of the thermal deformation mode, two types of response surfaces were made, using the lamination parameters, linear thermal expansion coefficients of the plate and stiffener, and the dimensions of the design variables in Table 1 . This means that V 1 *P , V 2 *P , W 1 *P , W 2 *P (lamination parameters of the plate), V 1 *S , V 2 *S , W 1 *S , W 2 *S (lamination parameters of the stiffener), αx P , αy P , αxy P , αx S , αy S and αxy S (the linear thermal expansion coefficients), and b 2 , h , np , ns (dimensions of the composite structures) were variables of the response surface for the thermal deformation. In this method, two response surfaces were made, to predict the thermal deformation of the laminated CFRP. The D-optimal laminates [5] were selected from the set of candidate Types A and B. Four hundred laminates were selected, to make the two response surfaces.
To create a response surface of the buckling stress for plates with various numbers of plies, it was not sufficient to use a simple quadratic polynomial response surface, because the buckling stress was significantly affected bythe plate thickness. Let us consider a simply supported rectangular plate, subjected to a single axial compression stress. The buckling stress is proportional to the square of the plate thickness. This means that a response surface canbe made, when the buckling stress is divided by the square of the thinner thickness of the blade stiffened composite structure. In this study, the buckling stress factor f , which is the coefficient of the reference stress divided by the square of the thinner plate, is used, instead of directly using the buckling stress.
The buckling stress factor f was approximated, using the Kriging response surface. The description of the Kriging response surface is detailed in [20] . To make the response surface of the buckling stress factor, Latin Hypercube Sampling (LHS) was used. The total number of calculations was 251 for the buckling response surface.
3. Genetic algorithm coding
The GA used was a general simple GA, but the crossover was a modified new one. The violation of the constraints on thermal deformation and the buckling stress ratio were implemented in the objective function of Equation (1) aspenalties. When the composite structure violated the two constraints calculated from the response surfaces, penalty values were added into the objective function. In this study, the penalty factors were ηd (thermal deformation)=1.0 and ηb (buckling stress)=1.0 This indicates that the true objective function is described as follows:
Lager Image
For a real number of design variable, the real number was used as a chromosome. when the random number function ( rand ( )) was used, the real number valuables were initialized, as follows:
Lager Image
where, ximax is the maximum value of the design variable xi , ximin is the minimum value of the design valuable xi , rand_max is the maximum value of the integer random number function, and rand ( ) is the random number function. When Equation (3) is implemented in the actual source code, the programmer should be careful of the effect of rounding off the integer numbers. Crossover between the two real number chromosomes is the simple mean value. Mutation is the movement of the real number between ximax and ximin with equal probability.
For a stacking sequence, an array of the same length as the number of plies was used. For example, (0,2,1,1) meant [0/90/45/-45]s. The recessive gene-like repair method is described in [4] . When the number of genes was decreased in the mutation process for the number of plies, the inner plies were deleted until the number of plies is satisfied , to prevent a large change in the bending lamination parameter. When the number of genes was increased in the mutation process, the innermost ply was added. The mutation of the number of plies was limited to one ply in this study.
When two chromosomes with a different numbers of plies were selected for the crossover operation, a new normalized coordinate method was adopted, to implement the crossover between the two different-length chromosomes.The normalized coordinate was defined for the stacking sequence. We assume that all the stacking sequences can be normalized by the number of plies. For example, a laminate [0/90/0/90]s has 0°-ply from 0.0 up to 0.25, and 90°-ply from 0.25 up to 0. 5. In the present study, the crossover operation was performed using this normalized coordinate. Let us consider the case in which a four-ply laminate ([0/90/0/90]s)and six-ply laminate ([45/-45/45/-45/90/0]s) were selected as parents, and the child was a five-ply laminate. The five-ply laminate had five gene loci: 0-0.2, 0.2-0.4, 0.4-0.6, 0.6-0.8 and 0.8-1.0. The coordinate of the gene locus was the midpoint of the segment. For example, 0.1 was the gene locus of the first segment. For the fiber angle of the gene locus, a gene from the parents at the same gene locus was selected with a 50 % probability. If the four-ply laminate was selected as the first ply, 0°-ply was set to the first gene locus of the child. The rest of the gene loci were decided similarly.
A two-stage optimization process was adopted. In the first stage, the dimensions of the composite structures were decided. In the second stage, a stacking sequence optimization was performed, to obtain the best performanceof the fixed dimensions, taking into account the constraints. For both processes, the GA was adopted. In the second optimization process, however, to obtain the best stacking sequences, nearby laminates in the lamination parameterdesign space were searched, using the same method as in [23] . This means the GA used here was not a pure simple GA, but included searching of the adjacent laminates. The optimizations were performed five times, to prevent errors based on probability.
4. Results and discussion
Figures 4 (a) and (b) show the results of the response surface of the warping thermal deformation. The abscissa gives the FEM results, and the ordinate is the estimated results, using the response surface. The adjusted coefficient of the determination is 0.97 for Mode A, and 0.98 for Mode B. These high values for the adjusted coefficient of the determination indicate that both response surfaces provided extremely good approximations, similar to those shown in
Lager Image
Results of response surfaces of the two thermal deformation types.
Optimal blade stiffened composite structure
Lager Image
Optimal blade stiffened composite structure
[22] . Even when the dimensions were variables, the response surfaces using quadratic polynomials provided very good approximations.
The cross validation of the Kriging response surface is shown in Fig. 5 . The cross validation is different from that in Fig. 4 . As the Kriging response surface gives exactly the same value as that given in the training data, this study used cross validation for the Kriging response surface. In the cross validation, one item of data was deleted, and
Lager Image
Buckling stress cross-validations
a Kriging response surface was made. Using the Kriging response surface, the deleted item of data was estimated. The estimation, therefore, was a new case of the estimations, which is why the Buckling stress Kriging response surfacelooks worse than the warping thermal deflection response surface. The Kriging response surface, therefore, gives an affordable response approximation.
The optimal results obtained here are shown in Table 3 . The buckling mode of the optimal composite structure is shown in Fig. 6 . The warping thermal deformation mode of the optimal result is shown in Fig. 7 . As mentioned in the optimization process, the optimization method includes adjacent searching, to confirm the optimality in the lamination parameter design space. The result obtained is the approximately true optimal result. The results are confirmed using FEM analyses.
5. Conclusions
The present study investigated the warping thermal deformation constraint for various dimensions, using blade stiffened composite structures. Lamination parameters, linear thermal coefficients and dimensions of the composite structures were used as variables of quadratic polynomials. As a result, the response surfaces gave very good approximations, and the warping thermal deformation was evaluated using the response surfaces. The GA was adopted, to obtain the optimal composite structures with the buckling stress, and the warping thermal deformation constraints. The optimization successfully obtained the optimal result.
Lager Image
Buckling mode of yhe optimal composite structures.
Lager Image
Thermal deformation distribution of the opt icmomal pcoosmitpeo sstrituec ture structure.
References
Miki M. 1985 “Design of Laminated Fibrous Composite Plates with Required Flexural Stiffness” ASTM STP 846 387 - 400
Fukunaga H. , Vanderplaats G. 1992 “Stiffness Design Method of Symmetric Laminates Using Lamination Parameters” AIAA Journal 30 (11) 2791 - 2795    DOI : 10.2514/3.11304
Le Riche R. , Haftka R. T. 1993 “Optimization of Laminate Stacking Sequence for Buckling Load Maximization by Genetic Algorithm” AIAA Journal 31 (5) 951 - 956    DOI : 10.2514/3.11710
Todoroki A. , Haftka R. T. 1998 “Stacking sequence optimization by a genetic algorithm with a new recessive gene like repair strategy” Composites Part B 29 (3) 277 - 285    DOI : 10.1016/S1359-8368(97)00030-9
Todoroki A. , Ishikawa T. 2000 “Design of Experiments for Stacking Sequence Optimizations with Genetic Algorithm using Response Surface Approximation” Composite Structures 6 (3-4) 349 - 357
Liu B. , Haftka R. T. , Akgün M. , Todoroki A. 2000 “Permutation Genetic Algorithm For Stacking Sequence Design Of Composite Laminates” Computer Methods in Applied Mechanics and Engineering 186 (2-4) 357 - 372    DOI : 10.1016/S0045-7825(99)90391-2
Park J. H. , Hwang J. H. , Lee C. S. , Hwang W. 2001 “Stacking sequence design of composite laminates for maximum strength using genetic algorithms” Composite Structures 52 (2) 217 - 231    DOI : 10.1016/S0263-8223(00)00170-7
Todoroki A. , Sasai M. 2003 “Optimizations of stacking sequence and number of piles for laminated cylinders using GA with intron genes” Advanced Composite Materials 12 (4) 331 - 344    DOI : 10.1163/156855103322986592
Cho M. H. , Rhee S. Y. 2003 “Layup Optimization Considering Free-Edge Strength and Bounded Uncertainty of Material Properties” AIAA Journal 41 (11) 2274 - 2282    DOI : 10.2514/2.6821
Lin C. , Lee Y. 2004 “Stacking sequence optimization of laminated composite structures using genetic algorithm with local improvement” Composite Structures 63 (3-4) 339 - 345    DOI : 10.1016/S0263-8223(03)00182-X
Kang J. H. , Kim C. G. 2005 “Minimum-weight design of compressively loaded composite plates and stiffened panels for postbuckling strength by Genetic Algorithm” Composite Structures 69 (2) 239 - 246    DOI : 10.1016/j.compstruct.2004.07.001
Abouhamze M. , Shakeri M. 2007 “Multi-objective stacking sequence optimization of laminated cylindrical panels using a genetic algorithm and neural networks” Composite Structures 81 (2) 253 - 263    DOI : 10.1016/j.compstruct.2006.08.015
Almeida F. S. , Awruch A. M. 2009 “Design optimization of composite laminated structures using genetic algorithms and finite element analysis” Composite structures 88 (3) 443 - 454    DOI : 10.1016/j.compstruct.2008.05.004
Marín L. , Trias D. , Badalló P. , Rus G. , Mayugo J. A. 2012 “Optimization of composite stiffened panels under mechanical and hygrothermal loads using neural networks and genetic algorithms” Composite structures 94 (11) 3321 - 3326    DOI : 10.1016/j.compstruct.2012.04.024
Lee D. S. , Morillo C. , Bugeda G. , Oller S. , Onate E. 2012 “Multilayered composite structure design optimization using distributed/parallel multi-objective evolutionary algorithms” Composite structures 94 (3) 1087 - 1096    DOI : 10.1016/j.compstruct.2011.10.009
Narita Y. 2003 “Layerwise optimization for the maximum fundamental frequency of laminated composite plates” Journal of Sound and Vibration 263 (5) 1005 - 1016    DOI : 10.1016/S0022-460X(03)00270-0
Terada Y. , Todoroki A. , Shimamura Y. 2001 “Stacking Sequence Optimizations Using Fractal Branch and Bound Method for Laminated Composites” JSME International Journal., Series A 44 (4) 490 - 498    DOI : 10.1299/jsmea.44.490
Todoroki A. , Terada Y. 2004 “Improved Fractal Branch and Bound Method for Stacking Sequence Optimizations of Laminated Composite Stiffener” AIAA Journal 42 (1) 141 - 148    DOI : 10.2514/1.9038
Hirano Y. , Todoroki A. 2004 “Stacking sequence optimizations for composite laminates using fractal branch and bound method (Application for Supersonic Panel Flutter Problem with Buckling Load Condition)” Advanced Composite Materials 13 (2) 89 - 107    DOI : 10.1163/1568551041718044
Todoroki A. , Sekishiro M. 2008 “Modified Efficient Global Optimization for a Hat-stiffened Composite Panel with Buckling Constraint” AIAA Journal 46 (9) 2257 - 2264    DOI : 10.2514/1.34548
Todoroki A. , Shinoda T. , Mizutani Y. , Matsuzaki R. 2011 “New surrogate model to predict fracture of laminated CFRP for structural optimization” Journal of Computational Science and Technology,JSME 5 (1) 26 - 37    DOI : 10.1299/jcst.5.26
Todoroki A. , Ozawa T. , Mizutani Y. , Suzuki Y. 2013 “Thermal deformation constraint using response surfaces for optimization of stacking sequences of composite laminates” Advanced Composite Materials 22 (4) 265 - 279    DOI : 10.1080/09243046.2013.805502
Todoroki A. , Sasai M. 2002 “Stacking Sequence Optimizations Using GA with Zoomed Response Surface on Lamination Parameters” Advanced Composite Materials 11 (3) 299 - 318    DOI : 10.1163/156855102762506335