6th World Congresses of Structural and Multidisciplinary Optimization

Rio de Janeiro, 30 May - 03 June 2005, Brazil

TOPOLOGY OPTIMIZATION DESIGN OF FUNCTIONALLY GRADED STRUCTURES

Emilio Carlos Nelli Silva1 and Glaucio H. Paulino2
1

Department of Mechatronics and Mechanical Systems. Escola Politcnica da Universidade de So Paulo, So Paulo SP Brazil 1 ecnsilva@usp.br 2 Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, Urbana IL USA 2 paulino@uiuc.edu

1. Abstract Functionally Graded Materials (FGMs) possess continuously graded material properties and are characterized by spatially varying microstructures. Such materials are studied in conjunction with the concept of topology optimization design which determines holes and connectivities of the structure by adding and removing material in the extended fixed design domain. The objective is to design FGM structures by using the concept of continuum topology optimization which considers a continuum distribution of the design variable inside the finite element domain. The traditional formulation for stiffness design problem is considered where the objective is to find the material distribution that minimizes the mean compliance. Two distinct, but related, applications are considered. The first application considers the objective of designing a structure in an FGM domain where the properties change in a certain direction according to a specified law, which leads to a structure with asymmetric stiffness properties. A new material model, called FGM-SIMP (Functionally Graded Material Solid Isotropic Material with Penalization), is defined based on the traditional SIMP model. The second application consists of finding the optimal material gradation law inside the design domain. In this case we expect to obtain intermediate material properties inside the design domain. A material model obeying the Hashin-Strikman bounds is applied. Because current FGM manufacturing techniques emphasize layered systems, a layered material constraint is adopted. Different from the traditional topology optimization problem which focuses on a 0-1 design, we seek intermediate properties at the end of the optimization process. The optimality criteria method is applied to solve the optimization problem. The algorithm was implemented using C language. A few numerical examples are given to illustrate the proposed methods. It is shown that the resulting topology can change significantly as a function of the material gradient. 2. Keywords: Functionally graded materials (FGMs), Topology optimization, Compliance minimization. 3. Introduction Functionally Graded Materials (FGMs) are materials that possess continuously graded properties with gradual change in microstructure [1,2]. These materials take advantage of desirable features of its constituent phases. For instance, in a thermal protection system, FGMs take advantage of heat and corrosion resistance typical of ceramics, and mechanical strength and toughness typical of metals. Topology optimization is a powerful structural optimization method that combines a numerical solution method, usually the Finite Element Method (FEM), with an optimization algorithm to find an optimal material distribution inside a given domain [3]. It is well-known that the optimum topology optimization result consists in a structure with intermediate (or composite) material. In this sense the concept of topology optimization is strongly related to the concept of FGM materials, which essentially considers a continuous transition of material properties. In traditional topology optimization formulations, the design variable is defined in a piecewise fashion in the discretized domain, which means that continuity of the material distribution is not realized between finite elements. However, considering the topology optimization results as an FGM-type, a more natural way of representing the material distribution emerges by using the concept of the graded finite elements, which leads to a continuous representation of material properties [4] that are interpolated inside the finite element using the FE shape functions. In fact, recent works [5,6] have suggested considering the continuum distribution of the design variable inside of the finite element in the topology optimization formulation, and this new formulation alleviates the checkerboard problem [7], which is an old problem in topology optimization. By means of the continuum model approach, the design of FGM structures can be fully achieved by applying topology optimization because a continuous change of material properties is considered inside the design domain. The design of FGM structures using topology optimization has been considered in previous works [8, 9] focusing mainly on thermal and thermomechanical applications, including transient analysis, by defining the design variable in a piecewise fashion in the discretized domain. On the other hand, the objective of the present work is to design FGM structures by using the concept of continuum topology optimization [6]. In this method, a continuum distribution of the design variable inside of the finite element domain is considered allowing us to represent a fully continuous FGM material during the design problem. As objective function, the traditional formulation for stiffness design problem is considered initially where the objective is to find the material distribution that minimizes the mean compliance. Two distinct, but related, applications are considered here: Design of a new material/structural configuration by considering the distribution of the FGM itself in the domain, which leads to a structure with asymmetric stiffness properties. Design of an FGM optimal layering configuration. Because current FGM manufacturing techniques emphasize layered systems, a layered material constraint is adopted.

The remainder of this paper is organized as follows. In section 4, the theoretical formulation is described by discussing the continuous topology optimization concept and the formulation of the design problem. In section 5, the design problem formulation is described. In section 6, implementation and computational procedures are briefly discussed. In section 7, some numerical results are presented to illustrate some of the features of the proposed method. Finally, in section 8, conclusions are inferred and a potential direction for future work is provided. 4. Theoretical Formulation

4.1. Basic topology optimization formulation

Topology optimization is based on two main concepts [10]: the extended design domain and the relaxation of the design domain. The extended design domain is a large fixed domain that must contain the whole structure to be determined by the optimization procedure. The objective of topology optimization is to determine the holes and connectivities of the structure by adding and removing material in the domain. Thus, the topology optimization problem is defined as a problem of finding the optimal distribution of material in the extended domain. Since the extended domain is fixed, the Finite Element model domain is not changed during the optimization process, which simplifies the calculation of derivatives of any function defined over the extended domain. The relaxation of the design problem is associated to the change of material from solid (one) to void (zero). The discrete problem, where the amount of material in each element can assume only values equal to either one or zero, is an ill-posed problem, that is, it does not present a solution. Thus, the problem must be relaxed by allowing the material to assume intermediate property values during the optimization procedure which can be achieved by defining a material model. Essentially, the material model approximates the material distribution by defining a function of a continuous parameter (design variable) that determines a mixture of two materials throughout the domain. By allowing the appearance of intermediate (or composite) materials rather than only void or full material - in the final solution, enough relaxation is provided for the design problem. Since the beginning of topology optimization implementation, the design variables that determine the mixture law were approximated in a piecewise constant fashion in the FE implementation, which means that the continuity of the material distribution is not realized between elements. Recent works [6, 11] have suggested considering the continuum distribution of the design variable inside the finite element by means of its interpolation using the FE shape functions. Thus, the design variables would be defined for each element node instead of each finite element as usual. Although this formulation cannot solve the mesh-dependency problem, it reduces the checkerboard problem. In this work, a topology optimization material model based on the so-called density method or artificial power law approach will be employed together with a filtering technique (to control the mesh dependency) [3]. They will be discussed in the next section.

4.2. Material models

This work considers two types of FGM structural design. The first problem is related to the design in an FGM domain, that is, a domain where the properties change in a certain direction according to a specified model. The second problem consists in finding the optimal property variation of the FGM in the domain to achieve the mean compliance design. Thus, different material models are considered, as described below. 4.2.1 FGM-SIMP material model The traditional SIMP (Solid Isotropic Material with Penalization) material model [3] states that in each point of the domain, the material property is given by:

E H = p E0

(1)

where EH and E0 are the Young modulus of the homogenized material and basic material that will be distributed in the domain, respectively, is a pseudo-density describing the amount of material in each point of the domain which can assume values between 0 and 1, and p is a penalization factor to recover the discrete nature of the design. For equal to 0 the material is equal to void, and for equal to 1 the material is equal to solid material. Now, consider the objective of designing a structure in an FGM domain, that is, a domain where the properties change in a certain direction according to a specified law. In this case, the property E0 considered above is not constant along the domain, but it depends on the position x. Thus, the previously defined SIMP material model can be written in the following form:

E H = p E0ex + y

(2)

where and are coefficients that define the change of material property in the domain, and x and y are the Cartesian coordinates. This is a common situation when dealing with FGM materials, which it is expected to result in non-symmetric designs. This material model is called FGM-SIMP material model. 4.2.2 Material models for layered FGM structures Consider two-phase material systems. Thus, the FGMs represent the transition between two base materials, and the objective is to find the optimal FGM property variation such that the material model allows local distribution of the two materials in the domain. In this case, a material model based on a density method approach was defined for the Lame constants, and it states that in each point of the domain:

 = B e and = ln ( A /B ); = B e and = ln ( A / B )

(3)

where A, A, and B, B are the Lame constants for materials A and B, respectively. Thus, the elasticity (Young) modulus and Poissons ratio can be obtained through the well-known equations:

E=

(3 + 2 ) ; = + 2( + )

(4)

In the material model defined by Eq.(4), no penalization is applied because we are interested in solutions with intermediate materials, which are closely related to the continuous nature of FGMs. In this case, even though we are using a material model based on a density method approach, we do not expect the traditional mesh-dependency problem. This material model satisfies the HashinShtrikman bounds [12], as illustrated by Figure 1.

Materials Models
10 9 8 7 6 5 4 3 2 1 0

Young Modulus

0.1

0.2

0.3

0.4

0.5 0.6 Pseudo-Density

0.7

0.8

0.9

Hashin-Shtrikman upper bound Hashin-Shtrikman lower bound first model (equation 4)

Figure 1 Material model given by Eq.(4) and Hashin-Shtrikman bounds.

By considering a discretized domain into finite elements, the concept of the continuum distribution of design variables based on the CAMD method (discussed above) is adopted. Thus, Eqs.(2) and (4) are considered for each element node, and the material property (Young modulus) inside each finite element is given by:

( x) =

I =1

I N I ( x)
(5)

where I is the nodal design variable, NI is the finite element shape function and nd is the number of nodes in each element. This formulation allows us to have a continuous distribution of material along the design domain instead of the traditional piecewise material distribution used by previous formulations of topology optimization.

5. Design problem formulation As objective function, the traditional formulation for stiffness design problem is considered where the objective is to find the material distribution that minimizes the mean compliance [3]:

Cmean =

t .u d

(6)

where t and u denote the traction and displacements, respectively, on the boundary of the domain. Then, the general continuous form of the topology optimization problem for stiffness design can be defined as: min Cmean (material distribution (x) for ) Such that Equilibrium equation

d des

0 1
By means of the FEM formulation for the discretized domain, the discrete optimization problem is restated as follows: min Cmean=UTF I (for each node) Such that KU=F

I =1

I Vdes

0 I 1
filtering technique where the filtering technique allows control of mesh-dependency (using the material model of Eq. (2)) or checkerboard problems. 6. Implementation and Computational Procedures A flow chart of the optimization algorithm is shown in Figure 2. The algorithm was implemented using C language and the visualization of results was performed using a simple MATLAB code. The design variables are the pseudo-density I, which can assume different values in each finite element node. Four node bilinear elements, considering a plane stress formulation, are used in the finite element formulation [13].

Initialization and Data Input

Calculation of Mean Compliance and Constraint

Converged?
N

Plotting of the Results

Sensitivity Analysis

Optimization (Optimality Criteria)

Update of Design Variables

Figure 2 - Flow chart of the topology optimization procedure.

Because the traditional stiffness design problem is considered, the optimality criteria method was applied to solve the optimization problem. This method is well-described in reference [3]. The filtering technique applied is presented in detail in reference [14] and is not discussed here. We also apply the so-called continuation method [15] where the penalization coefficient starts with a value equal to 1 and, after some convergence of objective function is obtained, the value is changed to 3. This method guarantees that the final result will be close to a global optimum. By starting with the penalization coefficient equal to 3 the algorithm may get trapped in a local optimum, which is not of interest. The use of continuation method has been a usual approach in topology optimization, thus, it was applied in this work. The optimization method requires the sensitivity of objective function in relation to the design variables. The sensitivity calculation for the mean compliance design problem considering the concept of the continuous approximation of material distribution was developed in [6] and it is equivalent to:
mi H Cmean T E = N I (x) dVe e I e =1

(7)

where mI is the number of elements associated with I-th node. The derivative material model formulation described in Eqs.(2) and (4). 7. Numerical Examples

can be easily calculated by considering the

The first example is related to the structure design in an FGM domain, that is, a domain where the properties change in a certain direction according to a specified law. This design, called graded structure design, is obtained by distributing the FGM material itself in the design domain. The idea is to obtain a structure with asymmetric stiffness properties. The distribution of only one basic material is considered with FGM law following Eq. (2). The normalized Young`s modulus E0 is equal to 10. The design domain considered with corresponding applied distributed load is shown in Figure 3(a). A mesh of 90 X 45 elements is employed. A volume material constraint of 30% is considered in all examples. The designs are obtained considering the material variation along either the x direction ( = 0) or the y direction ( = 0). A filter with radius equal to 1.5 was applied to all the results. For comparison purposes the topology optimization result obtained considering no material variation in either x or y direction ( = = 0) is shown in Figure 3(b).

1 1

45 90

(a) (b) Figure 3 (a) Initial domain considered (90 x 45 mesh) and corresponding applied loads and boundary conditions; (b) topology obtained considering the standard homogeneous SIMP material model, 30% volume constraint, and filtering (radius equal to 1.5). Figures 4(a), (b), (c), and (d) show the results obtained considering material variation in x direction for values of equal to 0.02 and 0.06, and equal to 0.02 and -0.06, respectively. By comparing with the homogeneous result, Figure 3(b), where no material variation is considered, one notices that, depending on the values of and , the method enlarges the thickness of the structure close to the regions where the material has a lower stiffness to guarantee the minimization of the structural mean compliance. Notice that, in Figures 4(a), (b), and (c), there is a relatively small change in the structural topology, however in case (d), there is a significant change.

(a)

(b)

(c)

(d)

Figure 4 (a) Topology obtained using the FGM-SIMP material model in the x direction with =0.02, and filter (radius equal to 1.5); (b) same, with =0.06; (c) same, with =0.02; (d) same, with =-0.06. The second example consists of finding the optimal property variation of a layered FGM in the domain to achieve the mean compliance design. In this last example, three different types of layered structures are considered such as horizontal, vertical and 45 inclined layers. The design domain considered, with corresponding applied distributed load and boundary conditions, is shown in Figure 5. A mesh of 100 X 20 elements is considered. The normalized Young`s modulus of the two basic materials are equal to 1 and 10, respectively; and the corresponding Poisson`s ratios are equal to 0.3 (constant). The material model described in Eq. (4) is considered. A volume material constraint equal to 20% is considered for the stiffer material. A symmetry constraint in the material distribution was considered for the vertical layers since the expected result is symmetric. No penalization is applied since we are interested in solutions with intermediate materials, which are closely related to the continuous nature of FGMs.

1 1 1 100 2
Figure 5 Design domain (100 x 20 mesh), with applied loads and boundary conditions, for the three-point bending beam. Figure 6(a) illustrates the results obtained considering the design of a horizontally layered FGM structure. Notice that the method concentrates the stiffer material in the upper and lower layers of the domain where the stresses are higher due to the bending effect. Figure 6(b) illustrates results obtained considering the design of a vertically layered FGM structure. A filter [14] with radius equal to 1.5 is considered. Figure 6(c) illustrates results obtained considering the design of a 45 inclined layer FGM structure. It can be noticed that the amount of stiffer material is increased close to the applied force, and more at the left support region than on the right support.

20

(a)

(b)

(c) Figure 6 (a) Design of horizontally layered FGM structure and corresponding plot of pseudo-density as a function of layers; (b) same for a vertically layered FGM structure with a filter of radius equal to 1.5; (c) same, for a 45 inclined layer FGM structure. 8. Conclusions This work shows that continuum topology optimization can be successfully applied to design FGM structures. The material distribution is based on graded finite elements [4] that allow change of material properties inside the design domain in a continuous manner (which is close related to the FGM concept). A new design is obtained where the distribution the FGM material itself is considered in the design domain leading to a structure with asymmetric stiffness properties. Thus, novel types of structures can be obtained by exploring the FGM idea. The length-scale of material gradation (e.g. 1/) competes with the geometric length-scale of the design problem and leads to novel structural topologies. As future work, the design of composite unit cells made of FGMs will be considered using the continuum topology optimization concept. 9. Acknowledgments We acknowledge the USA NSF through the project CMS#0303492 Inter-Americas Collaboration in Materials Research and Education (PI, Prof. W. Soboyejo, Princeton University and FAPESP (Fundao de Amparo Pesquisa do Estado de So Paulo). We also thank Mr. Fernando Viegas Stump for his help with the numerical results. 10. References 1. 2. 3. 4. 5. 6. 7. 8. 9. Hirai T. Functionally Graded Materials. Materials Science and Technology: Processing of Ceramics, Part 2, R.J.Brook, ed., VCH Verlagsgesell-schaft mbH, Weinheim, Germany, 1996, 17B: 292-341. Suresh S. & Mortensen A. Fundamentals of Functionally Graded Materials. IOM Communications, London, 1998. Bendse M. P. & Sigmund O., Topology Optimization: Theory, Methods and Applications. Springer-Verlag, Berlin, 2003. Kim J. H. & Paulino G. H. Isoparametric graded finite elements for nonhomogeneous isotropic and orthotropic materials. ASME Journal of Applied Mechanics, 2002, 69(4):502-514. Jog C. S. & Haber R. B. Stability of finite element models for distributed-parameter optimization and topology design. Computer Methods in Applied Mechanics and Engineering, 1996, 130: 203-226. Matsui K. and Terada K. Continuous approximation of material distribution for topology optimization, International Journal for Numerical Methods in Engineering, 2004, 59:1925-1944. Diaz A. R. & Sigmund O. Checkerboard patterns in layout optimization. Structural Optimization, 1995, 10(1): 4045. Turteltaub S. Functionally Graded Materials for Prescribed Field Evolution. Computer Methods in Applied Mechanics and Engineering, 2002, 191: 2283-2296. Turteltaub S. Optimal material properties for transient problems. Structural and Multidisciplinary Optimization, 2001, 22:157166.

10. Bendse M. P. & Kikuchi N. Generating optimal topologies in structural design using a homogenization method. Computer Methods in Applied Mechanics and Engineering, 1988, 71(2): 197-224. 11. Rahmatalla S. F. and Swan C. C. A Q4/Q4 continuum topology optimization implementation. Structural and Multidisciplinary Optimization, 2004, 27: 130-135. 12. Hashin Z. and Shtrikman S. A Variational Approach to the Theory of the Elastic Behaviour of Multiphase Materials, Journal of the Mechanics and Physics of Solids, 1963, 11(2): 127-140. 13. Bathe K. J. Finite Element Procedures. Prentice-Hall, Englewood Cliffs, 1996. 14. Sigmund O. A 99 line topology optimization code written in Matlab. Structural and Multidisciplinary Optimization, 2001, 21(2): 120-127. 15. Stolpe M. and Svanberg K. On the trajectories of penalization methods for topology optimization. Structural and Multidisciplinary Optimization, 2001, 21: 128-139.