Two-scale topology optimization of macrostructure and porous microstructure composed of multiphase materials with distinct Poisson’s ratios

Negative Poisson’s ratio NPR material attracts a lot of attentions for its unique mechanical properties. However, achieving NPR is at the expense of reducing Young’s modulus. It has been observed that the composite stiffness can be enhanced when blending positive Poisson’s ratio PPR material into NPR material. Based on the respective interpolation of Young’s modulus and Poisson’s ratio, two concurrent topology optimization problems with different types of constraints, called Problem A and B, are respectively discussed to explore the Poisson’s ratio effect in porous microstructure. In Problem A, the volume constraints are respectively imposed on macro and micro structures; in Problem B, besides setting an upper bound on the total available base materials, the micro thermal insulation capability is considered as well. Besides considering the influence of micro thermal insulation capability on the optimized results in Problem B, the similar and dissimilar influences of Poisson’s ratios, volume fractions in Problem A and B are also investigated through several 2D and 3D numerical examples. It is observed that the concurrent structural stiffness resulting from the mixture of PPR and NPR base materials can exceed the concurrent structural stiffness composed of any individual base material.


INTRODUCTION
With a near 30-year development, the structural topology optimization has emerged a variety of approaches including homogenization method Bendsøe and Kikuchi 1988 , solid isotropic material with penalization SIMP method Bendsøe 1989, Mlejnek 1992, Zhou and Rozvany 1991 , evolutionary structural optimization ESO method Xie and Steven 1993 and the updated version, bi-directional evolutionary structural optimization BESO method Querin et al. 2000 , level set method Wang et al. 2003 and phase field method Bourdin and Chambolle 2003 .Quite recently, an explicit topology optimization approach based on so-called moving morphable components MMC solution framework was presented by Guo, Zhang, and Zhong 2014 .The concept of concurrent design can be traced to Lakes 1993 .The pioneering work of Rodrigues, Guedes, and Bendsoe 2002 was to introduce the topology optimization technique into concurrent design.However, any variable microstructure causes a high computational cost and manufacture difficulties.Later a more popular concurrent design method was developed by Liu, Yan, and Cheng 2008 .In this method, uniform microstructure can be achieved which overcomes the issues mentioned above.During the optimization process, effective material properties are computed through the numerical homogenization technique and integrated into the analysis of macrostructure.Subsequently, a series of researches were carried out under this framework.Yan, Cheng, and Liu 2008 suggested that the structure composed of porous material has lower stiffness than the fully solid material structure.Moreover, Niu, Yan, and Cheng 2009 also observed that the structure without porosity on micro-scale has higher fundamental frequency than the structure with porosity on micro-scale.A similar viewpoint is expressed in literature Vicente et al. 2016 .Taken minimized frequency response as the objective, under the BESO framework, Vicente et al. 2016 revealed that the fully solid material structure has a better result compared to the structure consisted of porous microstructure.The same conclusion can be obtained through porous microstructure converging to isotropic solid material under coupled volume constrains Sivapuram, Dunning, and Kim 2016 .The above researches manifest that the structure made of porous material should consider the multifunctional demands.Basing on this idea, Yan, Cheng, and Liu 2008, Deng, Yan, and Cheng 2013, Yan et al. 2014 and Long, Wang, and Gu 2018 discussed the concurrent optimization of thermoelasticity and thermal conduction.Besides the discussion about the multifunctional applications, the concurrent approaches in considering multiphase materials are concerned as well in recent years.Especially Xu et al. Xu, Jiang, and Xie 2015, Xu and Xie 2015, Xu et al. 2016 built the concurrent optimization models in regard to multiphase materials respectively under harmonic, random and mechanical-thermal coupled loads.Da et al. 2017 focused on the super multiphase materials problem in concurrent optimization.Besides the homogenization theory, other approaches are also employed in concurrent optimization.Basing on super element technique, Zhang and Sun 2006 achieved the scale-related cellular materials and layered structures.Yan, Hu, and Duan 2015 adopted extended multiscale finite element method to study the size-effect in concurrent structure which is composed of lattice materials.Xia and Breitkopf 2014 introduced FE 2 model into concurrent optimization to solve the computational cost problem caused by multiscale nonlinearity.
Recent researches indicate that exceptional Poisson's ratios, e.g., Poisson's ratio towards to the thermodynamic limit 0.5, negative Poisson's ratio NPR , are helpful to improve the composite stiffness, called as Poisson's ratio effect Liu, Zhang, andGao 2006, Lim 2010 .Moreover, when Young's moduli of the positive Poisson's ratio PPR and NPR base materials are close to each other, the maximum enhancement in stiffness is observed to happen with the PPR and NPR respectively approaching to the thermodynamic limits 0.5 and -1 Kocer, McKenzie, and Bilek 2009 .Successive investigations validate that Poisson's ratio effect is influenced by Poisson's ratio, Young's modulus, dosages, shapes and distributions of base materials Zuo andXie 2014, Shufrin, Pasternak, andDyskin 2015 .Young's modulus and Poisson's ratio are respectively interpolated for employing topology optimization technique to exploit Poisson's ratio effect in sandwich structured composites Strek et al. 2014 .Besides maximizing the effective Young's modulus Long et al. 2016, Long et al. Long, Han, and Gu 2017, Long et al. 2018 extended the above model to reveal the Poisson's ratio effect in concurrent structure under volume fraction constraint and mass constraint.
In this paper, basing on the respective interpolation of Young's modulus and Poisson's ratio, Poisson's ratio effect in porous microstructure is revealed.Taken structural compliance as the objective, two optimization problems with different types of constraints, named as Problem A and B, are respectively discussed.In Problem A, the independent volume constraints are respectively imposed on macro and micro structures; in Problem B, besides setting an upper bound on the total available base materials, the thermal insulation capability in microstructure is also considered.The effective elasticity and thermal conductivity matrices obtained from the homogenization technique are respectively used in macro structural analysis and micro thermal insulation constraint.The remainder of the paper is organized as follows.Two concurrent optimization problems considering Poisson's ratio effect are established and described in Section 2. Section 3 provides the sensitivity analyses.The numerical implementation is given in Section 4. Section 5 presents several 2D and 3D numerical examples to illustrate the proposed models are effective to reveal the Poisson's ratio effect in concurrent optimization.Concluding remarks are given in Section 6.

CONCURRENT TOPOLOGY OPTIMIZATION PROBLEM WITH PPR AND NPR MATERIALS
Considering a two-scale system, as shown in Fig. 1, the macrostructure , i 1, 2, …, M, M is the total number of elements on macro-scale is taken as the macro design variable, with representation of the relative density of the th i macro element.d is a small predeter- mined value to avoid numerical singularity in optimization whose value is 0.001 in this paper.jective, two kinds of concurrent optimization problems are discussed to reveal the Poisson's ratio effect.In Problem A, the independent volume fraction constraints are imposed on macrostructure and microstructure; in Problem B, besides the coupled volume relationship between the macrostructure and microstructure, the thermal insulation constraint is conducted on micro-scale.
Figure 1: A two scale system: a macrostructure; b periodic porous material; c porous microstructure.

Problem A
In Problem A, the independent volume constraints are imposed on macrostructure and microstructure, the corresponding mathematical formula can be expressed as : f are the prescribed volume fractions of base materials and PPR base material on micro-scale.As known in Eq. 1 , the volume fractions of the macrostructure and microstructure are respectively defined and there is no coupling relationship between the macro and micro design variables.That is to say the porous requirement on micro-scale is mandatory.

Problem B
In Problem B, the total dosages of the base materials and micro thermal insulation capability are simultaneously defined.The concurrent optimization model can be correspondingly expressed as Find : { , , }( 1,2,..., ; 1,2,..., ) Minimize: Constraint I : In Eq. 2 , Constraint II and III respectively set the dosages of the base materials and PPR base material in macro and micro design domains.That is to say the macro and micro design variables have strong coupling relationship in volume fraction constraints.
Constraint IV describes the thermal insulation capability related to the microstructure.As to orthotropic material, the thermal insulation capability can be appraised according to the average of the diagonal elements in the effective thermal conductivity matrix de Kruijf et al. 2007 .
H ss k is the th s diagonal element in the effective thermal conductivity matrix.lim k is the upper bound of the thermal conductivity.Actually, the thermal insulation con- straint can also be understood as a porous requirement like volume fraction constraint, which not only restricts the dosages of the base materials on micro-scale but also plays a role of connecting the base materials.

Material interpolation scheme for structural analysis
Using SIMP scheme, the Young's modulus and Poisson's ratio in micro element j can be written as ) ) (1 ) in which E and v respectively denote the Young's modulus and Poisson's ratio.Superscript numbers 1 and 2 characterize the PPR and NPR base materials, respectively.The penalization factors have the values of 4 a= and 1 b = for better convergence and clearer topologies Long et al. 2016 .For succinct programming, the Young's modulus and Poisson's ratio are substituted by the Lamé's parameters.As to 2D plane stress problem, the Lamé's parameters can be expressed as Two-scale topology optimization of macrostructure and porous microstructure composed of multiphase materials with distinct Poisson's ratios Latin American Journal of Solids and Structures, 2018, 15 11 , e133 5/21 2 / (1 ) In 3D problem, the first Lamé's parameter needs to be modified as /(1 )/(1 2 ) , while the second Lame's parameter j m is kept the same with Eq. 4b .
With the help of Eq. 4 , the elasticity matrix in microstructure can be split into where l D and m D are the constant matrices with the expressions of stress problem, and Effective elasticity matrix D H can be calculated by the numerical homogenization theory with the format of Lamé's parameters Andreassen and Andreasen 2014, Hassani and Hinton 1998 where |V| is the volume of unit cell; I is the identity matrix; b is the micro strain-displacement matrix; the micro displacement field u can be acquired through the FE analysis with applying the periodic boundary conditions within the unit cell ( ) The right hand side in Eq. 7 defines the external forces corresponding to the uniform strain fields, e.g., two normal unit strains in x and y directions and one shear unit strain for 2D cases, and three normal unit strains in x , y and z directions and three shear unit strains for 3D cases.
The macro elasticity matrix can be determined by the following interpolation where the penalization factor p retains the typical value of 3.
From the above analysis, effective elasticity matrix D H plays a connection role between macrostructure and microstructure.
On macro-scale, FE analysis is again performed to obtain the displacement vector.The macro global stiffness matrix K can be assembled by the elemental stiffness matrix i K where B denotes the macro strain-displacement matrix.When considering the Problem B, the effective thermal conductivity of the porous microstructure is needed.Similarly, the elemental thermal conductivity can be interpolated with the SIMP scheme where (1)   k and (2 )   k are respectively used to represent the thermal conductivity of PPR and NPR materials.It is supposed that (1)   (2 ) . g is the penalization factor with the value of 4 Jia et al. 2016 .Furthermore, the micro thermal conductivity matrix can be obtained as where 0 k is the basic thermal conductivity matrix when thermal conductivity is 1.
Similarly, the effective thermal conductivity matrix H k can also be evaluated through the numerical homogenization method where I s represents the identity matrix in the process of thermal conductivity homogenization; χ denotes the induced temperature gradient field which can be computed from the uniform gradient temperature fields.More details in implementation of homogenized elasticity and thermal conductivity matrices can be referred to literature Andreassen and Andreasen 2014 .

SENSITIVITY ANALYSES FOR MACROSTRUCTURES AND MICROSTRUCTURES
Basing on the adjoint variable method Haug, Choi, and Komkov 1986 , the derivatives of the objective to the design variables can be calculated as According to the mapping method Liu et al. 2002 , the expression of k 1, 2 in Eq. 14 can be written as With the aid of Eq. 5 , the above equation can be further given as Two-scale topology optimization of macrostructure and porous microstructure composed of multiphase materials with distinct Poisson's ratios Latin American Journal of Solids and Structures, 2018, 15 11 , e133 7/21 where j j E l ¶ ¶ has distinct expressions in 2D plane stress and 3D problems, as 2 / ( 1) / (1 )/ (1 2 ) Furthermore, the expression of ¶ ¶ in Eq. 17 can be calculated as ) ) Similar as ¶ ¶ also possesses distinct expressions in 2D plane stress and 3D problems, as (1 ) / (1 ) (1 2 ) / ( 1) / (1 2 ) For In Eq. 17b , j j E m ¶ ¶ and j j v m ¶ ¶ can be respectively derived from Eq. 4b 1/(2( 1)) Eqs. 14 -16 clearly indicate that the sensitivity expressions for micro design variables are split into two parts with the help of Lamé's parameters, which simplifies the deduction process and programming.
In Problem A, the derivatives of volume fractions with the design variables on macro and micro scales can be respectively given as When considering Problem B, the derivatives of volume fractions with respect to the design variables can be followed as Two-scale topology optimization of macrostructure and porous microstructure composed of multiphase materials with distinct Poisson's ratios Latin American Journal of Solids and Structures, 2018, 15 11 , e133 8/21 In this article, the heuristic sensitivity filtering technique Sigmund 1997 is adopted to eliminate checkerboard patterns and mesh dependence.
For design variables i P and 1 j r respectively representing the material or void on macro and micro scales, the classical mesh-independency filter with weighting the element density is employed.Taken i P as an example In Problem A, the optimality criteria OC method is employed in macro-scale optimization, and the method of moving asymptotes MMA Svanberg 1987 is utilized as the optimizer in micro-scale.In Problem B, only MMA is adopted as an optimization solver.For ensuring close Young's moduli to each other, the base materials are assumed to be polymers and designed in the form of porous microstructures.In all the numerical examples, the Young's modulus of the base material 1 is defined as (1) 2.1 E = . The Poisson's ratio of the base material 1 is supposed to be (1) 0.4 in the first four examples and (1) 0.35 v = in the last example.For easy identification, PPR and NPR base materials in microstructures are respectively plotted in blue and red.Although the strategy with overcoming intermediate densities is adopted, a handful of intermediate densities still exist in the final topologies.For more meaningful comparisons, all the optimized results are executed with the 0-1 post processing Sigmund and Maute 2013 .

Problem A
In Problem A, the Young's modulus of the base material 2 is set as (2) 1.8 E = .Example I The Poisson's ratio effect in concurrent design is illustrated in this example.As shown in Fig. 4, the dimensions of the macro design domain are: length L 50, height H 100. The left edge is fully constrained and subjected to a vertical downward load, F 10, at the center of the right edge. ( )2 v is allowed to vary from -0.4 to -0.9.The target macro and micro volume fractions are predefined as Fig. 5 plots the evolution histories of the macro and micro densities with ( ) 2 v -0.4.This process clearly illustrates that the macrostructure and microstructure rapidly converge to stable topologies within fewer iteration steps.Fig. 6 provides the resulting compliance under various ( ) 2 v .Several quintessential topologies with ( )  2 v -0.4,-0.6, -0.7 and -0.9 are inserted above the resulting compliance curve.For the sake of comparison, the optimized results with exclusive NPR base material are also presented in Fig. 6.At the mean time, the optimized macrostructure and microstructure from exclusive PPR base material are shown in Fig. 7.The corresponding compliance is C 3946.37.From Figs. 6 and 7, it is known that the macrostructure composed of PPR and NPR base materials can provide a lower system compliance than that from the exclusive PPR or NPR base material.Moreover, the resulting compliance presents a tendency to decline with the decrement of ( ) 2 v from -0.4 to -0.9 and the minimum compliance occurs when the difference of ( ) v 1 and ( ) v 2 reaching the maximum ( ) v 1 0.4 and ( ) v 2 -0.9 in this example .f varies from 20% to 40% when ( ) 2 v -0.7.For clarity, the objective values corresponding to the above mentioned volume fractions are highlighted in Table 1.Example III A 3D cantilever beam is optimized in this example to demonstrate the adaptability of the proposed model in 3D concurrent design.Fig. 10 shows the admissible design domain in macrostructure with sizes of length L 48, height H 30, and width B 8. The left hand side surface is fully constrained and an evenly distributed load F 5 is applied at the middle of the right surface.The Poisson's ratio of the base material 2 is set as ( ) 2 v -0.7.The volume fraction constraints at two scales are set as m ac f 30%, mic,b f 80%, and mic,b1 f 40%.Table 2 lists the resulting two-scale topologies and effective elasticity matrix from hybrid of PPR and NPR base materials.At the mean time, the results of traditional structures made of exclusive PPR and NPR base materials with identical macro volume constraint m ac f 30% are provided as well in Table 2.According to the previous research Niu, Yan, and Cheng 2009, Vicente et al. 2016, Yan, Cheng, and Liu 2008 , the macrostructure composed of porous microstructure cannot provide higher stiffness than the traditional structure without porosity on microscale.However, from Table 2, a lower resulting compliance is observed in the concurrent design from porous microstructure.It is should be attributed to the Poisson's ratio effect.2.5683 -0.0935 -0.1389 0.0000 0.0001 0.0000 -0.0935 2.0672 0.4953 -0.0000 0.0002 0.0000 -0.1389 0.4953 1.1611 -0.0000 0.0000 0.0000 0.0000 -0.0000 -0.0000 1.1608 -0.0000 0.0000 0.0001 0.0002 0.0000 -0.0000 0.6976 -0.0000 0.0000 0.0000 0.0000 0.0000 -0.0000 0.9563 v 2 -0.6 are also given in Fig. 12.All the detailed data are summarized in Table 3. From Fig. 12, it is can be easily found that the macro and micro structures respectively exhibit the disparate topological configurations with increasing of b1 f .In order to manifest the stiffness enhancement is not caused by the variations of the macro and micro topological configurations, the optimized results with ( ) v 2 0.4 are also pre- sented in Fig. 12.In Fig. 12, it is noted that the resulting compliance from exclusive base material 2 is lower than the value of exclusive base material 1.Moreover, when ( ) v 2 0.4, the minimum resulting compliance from multiphase materials is found to be between the values of exclusive base material 1 and 2, i.e., the concurrent stiffness is not be enhanced by blending two PPR base materials.However, when ( )  v 2 -0.3, -0.4 and -0.6, the resulting stiffness can be enhanced at certain ranges of   f , e.g., b1 f 30% and 40%.For better illustration, taken b1 f 40% as an example, the objective value is revaluated after exchanging the material parameters for each other, i.e., respectively take ( ) v 2 0.4 and ( ) v 2 -0.6 into the topologies of hybrid of PPR and NPR base materials and hybrid of two PPR base materials.The detailed data are listed in Table 4.The comparisons show that both the original optimized results have lower compliance values which illustrates that with the variation of ( ) v 2 , the diverse distributions of base materials seem reasonable.This is mainly due to the stiffness of the base material 2 changes with the variation of ( ) v 2 .Taken Example V This example is the extension of example IV, with an attempt to reveal the influence of thermal insulation constraint on the final results.The design domain, boundary condition and external load are kept the same with the example IV.The Poisson's ratio of the base material 2 is set as ( )  v 2 -0.5, and the total dosages of the base material 1 is fixed at b1 f 20%.The thermal insulation constraint varies from 2 to 3.5.Respective interpolation scheme of Young's modulus and Poisson's ratio can realize the optimization with distinct Poisson's ratios.Taken the minimized structural compliance as the objective, two optimization problems with different types of constraints are proposed to reveal the Poisson's ratio effect stiffness enhancement in concurrent structure.In Problem A, the independent volume fraction constraints are applied on macrostructure and microstructure.In Problem B, besides building the coupled volume relationships between macrostructure and microstructure, the thermal insulation constraint is also considered on micro-scale.Several typical numerical examples clearly demonstrate that the hybrid of PPR and NPR base materials on micro-scale can improve the stiffness of concurrent structure.More conclusions can be obtained as follows

Under various lim
(1) In both Problem A and B, the stiffness of the concurrent structure can be constantly enhanced with the difference between the Poisson's ratios of the base materials increasing.
(2) The resulting compliance varies with the volume fractions of base materials.Moreover, the volume fraction ranges whose objective values can simultaneously exceed the results of exclusive PPR and NPR base materials are gradually enlarged with decreasing of NPR.In Problem A, the optimal proportion of base materials tends to reduce the dosages of PPR base material in microstructure with the decrement of NPR.However, the optimal volume fraction in Problem B does not change with the variation of NPR.Moreover, in Problem B, the macrostructure strongly interacts with microstructure, e.g., with continuously increasing the volume fraction of base material 1, the volume fractions of macrostructure and microstructure respectively increase and decrease monotonically.
(3) When relaxing the thermal insulation requirement, the resulting compliance decreases monotonically.Similar as the volume faction constraint in Problem B, the thermal insulation constraint significantly influences the dosages of base materials between macro and micro scales, e.g., with gradually relaxing the thermal insulation constraint, the volume fractions of macrostructure and microstructure respectively decrease and increase monotonically.
Fig. 1 a is comprised of periodic materials Fig. 1 b whose unit cell Fig. 1 c is constituted by isotropic PPR and NPR base materials.The microstructure is assumed to be orthotropic.In Fig. 1 c , the PPR and NPR base materials are respectively represented by blue and red.
material.Adopting the structural compliance as the ob- Two-scale topology optimization of macrostructure and porous microstructure composed of multiphase materials with distinct Poisson's ratios Latin American Journal of Solids and Structures, 2018, 15 11 , e133 9/21 Key steps for the concurrent design of macrostructure and microstructure with PPR and NPR base materials are given in Fig. 2.

Figure 2 :
Figure 2: Flowchart of concurrent design with PPR and NPR base materials

Figure 3 :
Figure 3: Density distribution in micro design domain: a initial design I; b initial design II; c initial design III.

Figure 4 :
Figure 4: A 2D cantilever beam for example I with length L 50, height H 100.

Figure 5 :
Figure 5: Evolution histories of macro and micro densities.

Figure 6 :
Figure 6: The influence of NPR on the resulting compliance

Figure 7 :
Figure 7: The optimized results composed of exclusive PPR base material: a macrostructure; b microstructure.

Figure 8 :
Figure 8: A 2D long beam for example III with length L 300, height H 50.

Fig. 9
Fig. 9 provides the variation trend of resulting compliance under various mic,b1 f

Jiao 21 Figure 9 :
Figure 9: The influence of the volume fraction of base material 1 on the resulting compliance.
Figure 10: A 3D cantilever for example III with length L 48, height H 30 and width B 8.
Figure 11: A 2D cantilever beam for example IV with length L 160, height H 100.

b1f.f
The above analysis proves that the stiffness enhancement is caused by the hybrid of PPR and NPR base materials.Similar with the Problem A, in Problem B the resulting compliance also decreases monotonically with the decrement of ( ) v 2 from -0.3 to -0.6 .Moreover, with the reduction of ( ) v 2 , the ranges of b1 f for enhanced stiffness are keeping expanding, i.e., b1 f is equal to 10% when ( ) varies from 10% to 20% when ( ) v 2 -0.6.For clarity, the objective values corresponding to the above mentioned volume fractions are highlighted in Table3.However, unlike in Problem A, the optimal volume fraction does not change with the variation of NPR in Problem B. That is to say the resulting compliance increases monotonically with the increment of b1 f .

Figure 12 :
Figure 12: The influences of NPR and volume fraction of base material 1 on the resulting compliance.

Figure 13 :
Figure 13: The influences of b1 f on mac f

k
, Fig. 14 a illustrates the resulting compliance from the hybrid of PPR and NPR base ma- terials with optimized macro and micro topologies inserted.From Fig. 14 a , it is known that the resulting compliance decreases with the increment of lim k .When lim 2 k = , the resulting compliance lies between the values from exclusive base material 1 and 2. However, the resulting stiffness is enhanced when lim k equals 2.5, 3.0 and 3.5.Furthermore, it is observed in Fig. 14 a that the macro and micro structures respectively present distinct topologies under various lim k .For illustrating the stiffness enhancement is not caused by the variations of topological configurations, the optimized results from two PPR base materials, i.e., 35, are given in Fig. 14 b .In Fig. 14 b , it is obvious that although the macro and micro structures have the similar topological configurations as shown in Fig. 14 a , the resulting compliance from two PPR base materials always lies between the values of exclusive base material 1 and 2. That is to say the stiffness enhancement is induced by Poisson's ratio effect.The above analysis illustrates that besides NPR and volume fractions of the base materials, the thermal insulation constraint can also influence the Poisson's ratio effect in concurrent structure.

Figure 14 :,
Figure 14: The influence of the thermal insulation constraint on the resulting compliance: a hybrid of PPR and NPR base materials b hybrid of two PPR base materials.

R
is the set of elements e for which the center-to-center distance ( )

Table 2 :
The comparisons of concurrent structure and traditional structure.

Table 3 :
The resulting compliance under various b1 f when