Abstract
We demonstrate that the consideration of material uncertainty can dramatically impact the optimal topological microstructural configuration of mechanical metamaterials. The robust optimization problem is formulated in such a way that it facilitates the emergence of extreme mechanical properties of metamaterials. The algorithm is based on the bidirectional evolutionary topology optimization and energybased homogenization approach. To simulate additive manufacturing uncertainty, combinations of spatial variation of the elastic modulus and/or, parametric variation of the Poisson’s ratio at the unit cell level are considered. Computationally parallel Monte Carlo simulations are performed to quantify the effect of input material uncertainty to the mechanical properties of interest. Results are shown for four configurations of extreme mechanical properties: (1) maximum bulk modulus (2) maximum shear modulus (3) minimum negative Poisson’s ratio (auxetic metamaterial) and (4) maximum equivalent elastic modulus. The study illustrates the importance of considering uncertainty for topology optimization of metamaterials with extreme mechanical performance. The results reveal that robust design leads to improvement in terms of (1) optimal mean performance (2) least sensitive design, and (3) elastic properties of the metamaterials compared to the corresponding deterministic design. Many interesting topological patterns have been obtained for guiding the extreme material robust design.
Similar content being viewed by others
Introduction
Metastructures are metamaterial induced concepts implanted in structural design. The exploration of metamaterials was heavily steered by the pathbreaking investigation of electromagnetic metamaterials, which exhibited properties like, negative permittivity and permeability^{1}. This paved the way for mechanical metamaterials^{2}. These metamaterials are engineered to derive their fundamental mechanical properties from the geometry of their structural building blocks (periodicity or translational symmetry), instead of the constituting materials^{3,4}. Mechanical metamaterials have gained wide popularity in engineering applications due to their superior properties compared to conventional materials found in nature. It has been found that by varying one or more material constants such as, bulk modulus, shear modulus, Poisson’s ratio and elastic modulus, extraordinary materials such as, ultra light weight and high strength aerospace shells, functionally graded electromagnetic sensors and energy absorbing dampers can be secured^{5,6}.
For exploring such exciting possibilities of discovering new materials with extreme properties, determining their optimal configuration at the very conceptual microstructural design stage is indispensable^{7}. A recent works with multimaterial lattices^{8}, prestressed materials^{9} and piezoembedded damped lattice metamaterials^{10,11} show some ways of achieving extreme homogeneous properties. In this scenario, topology optimization (TO) is wellknown to be an effective computational analysis tool which renders greater design freedom and yields the best material layout given a prescribed design domain and boundary condition^{12}. Depending on the algorithm’s performance and versatility, various successful TO methods have emerged to the present^{13}, few popular ones being the solid isotropic material with penalization (SIMP), evolutionary structural optimization (ESO), level set method (LSM), and moving morphable components (MMC). In fact, the popularity of TO has even led to its integration with relatively newer numerical discretization schemes (compared to finite element method) such as, isogeometric analysis^{14}. A simple illustration of multiscale design and topology optimization of the microstructure of metamaterial is presented in Fig. 1a and b, respectively.
The success of TO is clearly evident from its extensive applications for material microstructural design in recent years. In this regard, few noteworthy literatures have been discussed as follows. Gibiansky and Sigmund^{15} investigated multiphased elastic composites and obtained topological design corresponding to extreme bulk and shear modulus. For obtaining maximum bulk and shear modulus, the optimal topology of cellular microstructures was determined using ESO^{16}. A compact code in MATLAB based on the SIMP and energybased homogenization method (EBHM) was developed for extreme material multiscale structural design^{17}. Clausen et al.^{18} designed and fabricated 3D auxetic material microstructures undergoing large deformations. Long et al.^{19} performed topology optimization to maximize the effective Young’s modulus, so as to obtain the optimal distribution of 3D material microstructures whose constituent phases consist of nonidentical Poisson’s ratios. A novel TO method was presented based on the independent pointwise density interpolation to obtain a bimaterial chiral metamaterial^{20}. Chen and Huang^{21} designed 3D chiral metamaterials based on the couplestress homogenization and maximized the chiralty (the coupling effects between tension/compression and torsion) of the cellular structure. Xu et al.^{14} achieved designs of ultralightweight architected materials along with extremal properties such as, maximum bulk and shear modulus by using isogeometric TO. Petalshaped auxetic metamaterials were designed where a backpropagation neural network was coupled with isogeometric TO to evaluate their macroscopic equivalent properties^{22}. Zheng et al.^{23} proposed two numerical schemes within the evolutionary optimization framework to obtain 2D auxetic metamaterials with favourable characteristics, and experimentally validated their special properties. Ye et al.^{24} proposed a method for the simultaneous adaptive design of gradually stiffer mechanical metamaterials along with auxetic property.
Although the above published articles related to microstructural TO explore multiple potent research prospects to achieve extreme material properties, they are restricted to the deterministic optimization framework. In this regard, the literature is found to be scarce in studies addressing uncertainties in manufacturing processes. The following few works highlight the importance of considering manufacturing variability and their effect on the structural design. Most nano, micro and macro structures fabricated using ebeam lithography, etching and milling processes, respectively are vulnerable to manufacturing uncertainties. Even the structures meticulously designed and optimized may experience inferior performance or in an extreme scenario, lose their functionality due to the wear of machining tools, under or overetching, or malcalibrated ebeam equipment. To address few of the above aspects, robust topology optimization (RTO) for structural design was presented accounting for manufacturing errors in^{25,26,27}. Specifically, the effect of overetching (erosion) and underetching (dilation) was modelled for structures produced by milling or etching, which may cause the structural parts to become thinner or thicker than intended. For the works^{25,26}, the problem was posed as a worst case design formulation and the error effects were simulated by a Heaviside projection threshold. While they considered only uniform manufacturing errors (i.e. constant in magnitude over the entire design domain), the method was extended to account for nonuniform manufacturing errors (i.e. with spatially varying magnitude)^{27}. For doing so, a probabilistic approach was followed where the projection threshold was represented by a (nonGaussian) random field. Jansen et al.^{28} developed an RTO framework which incorporated the spatially varying geometric imperfections due to misalignment and misplacement of material often encountered in slender (civil engineering) structures. These imperfections should be accounted for in accordance to the provision recommended in the Eurocode for design of steel structures and by the Joint Committee on Structural Safety. Note that the misplacement errors cause a perturbation in the spatial location of the material, whereas the errors due to etching, add or remove material at the structural surface. An ESO based RTO framework was developed for multimaterial structures considering interval loading uncertainty in^{29}. Efficiency was enhanced by (1) orthogonal decomposition of the loading uncertainty which decoupled the stochastic problem from the optimization loop and (2) sensitivity analysis. Material uncertainty modelled as imprecise probability was integrated with the multiscale concurrent ESO based TO framework in^{30}. The RTO methodology adopted the type I hybrid random interval model. An ESO based RTO framework was developed for multiscale 2D and 3D structures considering multiple random loading conditions in^{31}. Decoupled sensitivity analysis was observed to improve the computational efficiency. An ESO based RTO methodology was proposed for actuatorcoupled structures considering hybrid uncertainties in^{32}. The hybrid random interval model simulated the stochastic mechanical and piezoelectric parameters and perturbation analysis was employed to evaluate the statistical quantities.
While the above works simulated manufacturing uncertainties numerically and investigated their effect on the structural topology, the following studies have carried out experimental analysis to probe manufacturing uncertainties. It has been shown that the emission rate of a 3D printer based on fused deposition modeling principle increased significantly with the increase in the extrusion temperature^{33}. Melenka et al.^{34} found that the per cent infill has a significant effect on the longitudinal elastic modulus and ultimate strength of the test specimens, whereas print orientation and layer thickness fail to achieve significance. Dimensional analysis of test specimens also showed that the test specimen varied significantly from the nominal print dimensions. The manufacturing tolerances of metamaterial samples produced by a selective laser sintering process were obtained to simulate the manufacturing uncertainty in massproduced industrial applications^{35}. It was found that even small levels of variability, given by less than 1 % for the mass and less than 3 % for the elastic modulus, has a significant effect on the overall vibration attenuation performance.
While the first of the above three paragraphs discusses works on deterministic TO for extreme metamaterial design, the second paragraph illustrates the numerical simulation of manufacturing uncertainty and robust topological design. Finally, the above paragraph highlights a few instances of experimental exploration of manufacturing uncertainties. Motivated by the need to consider manufacturing uncertainties for metamaterial design, this work attempts to bridge the above points. Thus, we integrate these aspects to present an RTO formulation specifically for extreme metamaterial design by minimizing the effect of potential material uncertainties. To the best of the authors’ knowledge, this is one of the first applications of robust microstructural design for obtaining extreme mechanical properties of metamaterials. Few instances of readytoprint robust microstructural topologies of extreme metamaterials achieved as a part of this study have been presented in Fig. 1c for illustrative purpose and the interest of readers.
Methodology
The work integrates the material uncertainty model (simulated by inhouse developed codes) with deterministic topology optimization^{36} to develop a parallel Monte Carlo based robust topology optimization framework. A flowdiagram of the proposed RTO framework is presented in Fig. 2. The main intent is to illustrate the effect of material uncertainties on the microstructural topologies and derive robust configurations for extreme material design. As shown in Fig. 2, the entire methodology comprises of three steps: (1) the userdefined input module where the system and the optimization parameters are initialized, (2) the analysis module where the expensive computations take place involving uncertainty quantification nested within the optimization loop. To reduce the computational cost, a parallel version of Monte Carlo simulation is employed to simulate the randomness within each optimization iteration. The details of this step regarding the optimization algorithm, finite element model, energybased homogenization and sensitivity computations are presented subsequently. (3) The postprocessing module is where the resulting robust microstructural topologies with extreme properties are exported to a CAD software for 3D printing.
Homogenization
Considering linear elastic behaviour, the equivalent constitutive behaviour of periodic structures can be determined by homogenization technique. This work employs the energybased homogenization approach to predict the macroscopic equivalent elastic properties of microstructures^{36}. The approach is based upon energy conservation with respect to stress and strain. The homogenized stiffness tensor \(E^{H}_{ijkl}\) can be expressed in terms of mutual energies as
where Y represents the base cell, \(E_{pqrs}\) is the elasticity tensor in index notation, \(\varepsilon _{pq}^{A(ij)}\) and \(\varepsilon _{rs}^{A(kl)}\) denote the superimposed strain fields. When the base cell is discretized into N finite elements, Eq. (1) can be approximated as
where \(\varvec{u}_{e}^{A(kl)}\) are the element displacement solutions corresponding to the unit test strain fields \(\varepsilon ^{0(kl)}\) and \(\mathbf {k}_{e}\) denotes the element stiffness matrix. It is to be noted that for 2D problems, \(11 \equiv 1\), \(22 \equiv 2\), and \(12 \equiv 3\). Thus, Eq. (2) can be concisely represented as
where \(Q_{ij}\) is given by
In Eq. (3), \(q_{e}^{(ij)}\) are the element mutual energies and can be expressed as
Periodic boundary conditions
The displacements on a pair of opposite boundaries of a 2D base cell
where notations \(k\) and \(k+\) represent the pair of two opposite parallel boundary surfaces perpendicular to the \(k{{\rm th}}\) direction. One can remove the unknown periodic fluctuation term \(u_{i}^{*}\) by using the difference between the displacements as,
For a given unit cell, the quantity \((y_{j}^{k+}  y_{j}^{k})\) is constant. Thus, the right side of Eq. (7) is a constant for a specified \(\varepsilon _{ij}^{0}\). This boundary conditions can be incorporated within the finite element model by restraining the corresponding pairs of nodal displacements. Note that this boundary configuration satisfies the periodicity and continuity conditions in terms of displacement and stress employing displacementbased finite element analysis^{37}.
Optimization approach
The unit cell is discretized into N finite elements, resulting into equal number of density design variables \(\rho \in \mathbb {R}^{N}\). The element level elastic modulus can be defined by the modified SIMP method^{38} as
where \(E_{\text {min}}\) denotes the elastic modulus of the Ersatz material, an approximation for void material to prevent singularity of the stiffness matrix, \(0<\rho _{e}<1\), limits corresponding to Ersatz and solid materials, p is a penalization factor for driving the density towards the black and white solution and \(E_0\) denotes the elastic modulus of solid material.
The deterministic optimization problem can be expressed as
where the objective function \(c(E_{ijkl}^{H})\) is a function of the homogenized stiffness tensors, \(\mathbf {K}\) is the global stiffness matrix, \(\varvec{u}^{A(kl)}\) and \(\varvec{f}^{kl}\) are the global displacement and external force vectors of the test case (kl), respectively, d, \(v_e\) and v denote the spatial dimension, element volume and upper limit of the volume fraction, respectively.
The motive of this work is to determine the extreme properties of material microstructures, such as maximum bulk modulus, maximum shear modulus, minimum Poisson’s ratio and maximum equivalent elastic modulus. The corresponding objective functions can be defined as

Design of maximum bulk modulus:
$$\begin{aligned} c = (E_{1111} + E_{1122} + E_{2211} + E_{2222}) \end{aligned}$$(10) 
Design of maximum shear modulus:
$$\begin{aligned} c = E_{1212} \end{aligned}$$(11) 
Design of minimum Poisson’s ratio:
$$\begin{aligned} c = E_{1122} + E_{2211} \delta E_{1212} \end{aligned}$$(12) 
Design of maximum equivalent elastic modulus:
$$\begin{aligned} c = \frac{1}{2}(E_{1111} + E_{2222}) \end{aligned}$$(13)
Note that the topological design of minimum and negative Poisson’s ratio \((E_{1122}/E_{1111})\) has always been a difficult issue. It is observed from the literature that auxetic properties can be obtained by imposing additional constraints on isotropy or bulk modulus^{36}. An equivalent expression in the form of Eq. (12) is used for this purpose and adopted from^{23}. Selection of the term \(\delta\) in Eq. (12) is discussed later in the numerical illustration section.
The solution scheme for solving the equilibrium equation in Eq. (9) adopted here can be found in Xia and Breitkopf^{36}. After obtaining the finite element displacement solution, the optimization problem in Eq. (9) is solved by using optimality criteria method. The heuristic updating criterion^{13} is expressed as
where m is the step size, \(\eta\) is the damping coefficient and \(B_e\) can be determined by using the following optimality condition
where \(\lambda\) is the Lagrange multiplier. It imposes the satisfaction of the material volume fraction constraint. The numerator \(\frac{\partial c}{\partial \rho _{e}}\) which is the sensitivity of the objective function can be determined by using the adjoint method as
where \(\mathbf {k}_{0}\) represent the element stiffness matrix with unit elastic modulus. For a uniform mesh, the element volume \(v_e = 1\) and hence, \(\frac{\partial V}{\partial \rho _{e}} = 1\). To ensure convergence of the optimization problem Eq. (9), filtering techniques are used which are illustrated next.
Filtering
To eliminate the formation of a checkerboard pattern and resolve the meshdependent issue, a filter is applied either to the sensitivities or the densities^{38,39}. The sensitivity based filtering modifies the sensitivities \(\frac{\partial c}{\partial \rho _{e}}\) (Eq. (16) as,
where \(\gamma = 10^{3}\) is a small positive number incorporated to avoid the denominator becoming zero. \(H_{ei}\) is a weight term defined in Eq. (18). \(N_e\) represents the set of elements i whose centretocentre distance to element e, \(\Delta (e,i)\), is less than the filter radius \(r_{\text {min}}\). The size of the filter radius can be used to control the minimum size of the emerging features in the design domain.
The density based filtering converts the original densities to the filtered densities \(\rho _e\) as,
The sensitivities with respect to \(\rho _j\) can be obtained as,
Note that \(\frac{\partial c}{\partial \tilde{\rho }_{e}}\) can be evaluated by Eq. (16) upon replacing \(\rho _e\) with \(\tilde{\rho }_e\).
Robust topology optimization
The objective function considered for the robust topology optimization is
where \(\mu _c\) and \(\sigma _c\) are the mean and standard deviation of the deterministic objective function c, respectively.
Sensitivity calculation
Differentiating \(c_{\text {RTO}}\), we obtain
Now considering the first term of Eq. (22) and expanding it, we obtain
where we have represented the expectation based on \(N_{\text {simu}}\) number of samples. From Eq. (23), we can see that the first term of Eq. (22) can be computed based on the expectation of the topological derivatives in the deterministic case. For the second term in Eq. (22), we have
where \(\mathop {\mathrm{var}}\left( \bullet \right)\) indicates the variance operator. Considering the first term in Eq. (24) and expanding it similar to Eq. (23), we have
Finally, considering the second term in Eq. (24) and using Eq. (23), we have
Substituting Eqs. (26) and (25) into Eq. (24), we obtain
Finally, substituting Eqs. (27) and (23) into Eq. (22), we obtain
From Eq. (28), it is obvious that the topological derivatives for RTO can be computed based on conventional/deterministic topological derivatives.
Numerical illustration: results and discussion
Implementation details
In this work, the design domain is considered to be square and discretized into square plane stress elements. The volume fraction is taken to be 0.5. The unit cell is discretized into \(3 \times 3\) elements. The influence of an initial design can affect the final optimal topology significantly. For this work, a square shaped void region at the center of the unit cell is adopted as the initial design^{40} as shown in Fig. 2. The small void is generally inserted in the initial design to stimulate the evolutionary optimization process.
It has been reported in multiple studies that topology optimization for the design of extreme material properties is prone to multiple local minima. The final optimal solution is found to be extremely sensitive to the initial topology, shape of the unit cell, filter radius \((r_{\text {min}})\), penalization factor (p), filtering approach (sensitivity filtering or density filtering) and others. Therefore, for selecting the appropriate parameter values, few rules of thumb have been followed. The optimization problem Eq. (9) is nonconvex for \(p\ge 1\). Although the high values of penalization factor will result in more distinct topologies, the algorithm is likely to get trapped in a local minima^{17}. A smaller filter radius value generally results in a better solution (in terms of a detailed microstructure) as higher frequency details are visible through lowpass filter. Compared to sensitivity filtering, density filtering is observed to yield less sensitive designs and hence preferable for material microstructure design.
Parametric study has been carried out by considering various combinations of stochastic material parameter models such as, spatial variation via random field and random parametric models. The stochastic models assumed are in line with the literature (discussed in the introduction section) where spatially varying and parametric models simulate nonhomogenous and uniform manufacturing errors, respectively. For stochastic simulations, the elastic modulus and Poisson’s ratio are considered to be random and follow lognormal distribution with 5\(\%\) variability. It is practical to assume that all realizations corresponding to these random parameters will be positive and therefore, they are assumed to be lognormally distributed. In doing so, 1000 MCS is performed within every optimization iteration to simulate the material variability. This has been carried out in a costeffective manner using parallel computing (employing all 8 cores of an Intel Xeon processor). Four case studies have been undertaken to obtain extreme properties of material microstructures, such as maximum bulk modulus, maximum shear modulus, minimum negative Poisson’s ratio (auxetic metamaterials) and maximum equivalent elastic modulus.
The initial design is the same for all the following cases studied (except type III). The nominal values of the elastic modulus and Poisson’s ratio at the element level are adopted as 1 and 0.3, respectively. For the stochastic cases, their mean values are same as that of the nominal values. The length scale used for discretizing the random field model is taken as 0.2. The results are represented by reporting the optimal unit cell, \(3 \times 4\) periodic arrangement, equivalent elastic matrices and robust optimal solutions for each of the extreme material configurations.
Maximum bulk modulus (Type I)
Table 1 illustrates the optimal solutions for the case of type I. To obtain these results, density based filtering and the following parameter values are considered: \(p=5\) and, \(r_{\text {min}}=2\). It can be observed from Table 1 that uncertainty affects both the topological configuration and the equivalent elastic matrix. The stiffness terms \(Q_{11}\) and \(Q_{22}\) are observed to increase for all types of uncertainty models compared to the deterministic case. Specifically, the greatest rise is observed for the random field cases. The term \(Q_{33}\) decreases due to uncertainty. The offdiagonal term \(Q_{12}\) reduces for all uncertainty models and the deterministic model achieves the maximum value. Interestingly, for the hybrid input uncertainty modelling case, the resulting structure shows nonzero (negative) \(Q_{13}\) and \(Q_{23}\) terms. This indicates coupling of axial and shear deformations/stress states. Although there is a slight difference between the terms \(Q_{11}\) and \(Q_{22}\) for the deterministic and random field models, the difference is significant for the hybrid uncertainty model, exhibiting strong anisotropic behaviour. In terms of mean value of the objective function, the random field cases lead to higher value of bulk modulus compared to the deterministic and parametric uncertainty cases. Also, the random field cases result in a more robust topological design indicated by lower values of the standard deviation of objective function.
Maximum shear modulus (Type II)
Table 2 illustrates the optimal solutions for the case of type II. To obtain these results, density based filtering and the following parameter values are considered: \(p=5\) and, \(r_{\text {min}}=2\).
It can be observed from Table 2 that uncertainty affects both the topological configuration and the equivalent elastic matrix. The stiffness terms (diagonal entries) \(Q_{11}\), \(Q_{22}\) and \(Q_{33}\) and the offdiagonal term \(Q_{12}\) are observed to increase for the random field and hybrid input uncertainty models compared to the deterministic case. These terms achieve lower values for the parametric uncertainty model compared to the deterministic case. Interestingly, for the random field and hybrid input uncertainty models, the resulting structure show (1) nonzero values of \(Q_{13}\) and \(Q_{23}\) terms and (2) \(Q_{11} \ne Q_{22}\). (1) indicates coupling of axial and shear deformations/stress states and (2) illustrates anisotropic behaviour. In terms of mean value of the objective function, the random field and hybrid input uncertainty models lead to higher value of shear modulus compared to the deterministic and parametric uncertainty cases. The parametric stochastic model results in the minimum value of shear modulus. Also, the random field and hybrid input uncertainty models result in a more robust topological design indicated by lower values of the standard deviation of objective function.
Minimum Poisson’s ratio (Type III)
Density based filtering and the following parameter values are considered: \(p=5\) and, \(r_{\text {min}}=5\), for the type III case. The unit cell is discretized into 200 elements along both the horizontal and vertical directions. As the term \(\delta\) (refer Eq. (12)) causes significant sensitivity to the topological design of auxetic metamaterials^{23}, a parametric study is performed in Table 3 to investigate the effect of \(\delta\) on the resulting topologies with minimum Poisson’s ratio.
It can be observed from Table 3 that out of the adopted \(\delta\) values, only \(\delta\) = 0.5 and 1 lead to negative mean Poisson’s ratio corresponding to the deterministic and each of the stochastic models. \(\delta\) = 0.5 has been selected as lower Poisson’s ratios are obtained compared to the case \(\delta\) = 1. The corresponding row has been highlighted to indicate the best and the most consistent performance in Table 3 for clarity. Table 4 illustrates the optimal solutions for the case of type III, corresponding to \(\delta\) = 0.5.
It can be observed from Table 4 that uncertainty affects both the topological configuration and the equivalent elastic matrix. Although the overall resulting topologies from the deterministic and parametric uncertainty models are very similar with fine differences at the boundary. The stiffness terms \(Q_{11}\) and \(Q_{22}\) are observed to increase for the random field and hybrid uncertainty models and decrease for parametric uncertainty compared to the deterministic case. The stiffness term \(Q_{33}\) is observed to increase for all stochastic models compared to the deterministic case. The offdiagonal term \(Q_{12}\) is negative for all the cases reported and observed to increase for the parametric stochastic model and decrease for the random field and hybrid uncertainty cases than that of the deterministic one. Interestingly, for the random field and hybrid input uncertainty models: (1) the terms \(Q_{13}\) and \(Q_{23}\) are nonzero (negative), which illustrates coupling of axial and shear deformations/stress states and (2) \(Q_{11} \ne Q_{22}\), which indicates anisotropic behaviour. The random field and hybrid input uncertainty models lead to lower values of the mean objective function compared to the deterministic and parametric uncertainty cases. Specifically, the parametric and hybrid stochastic models result in the maximum and minimum value of Poisson’s ratio, respectively. The deterministic, random field and hybrid input uncertainty models result in the same level of robustness of the topological design indicated by the standard deviation of the objective function. The parametric uncertainty model achieves the most robust design by a small margin. The results in Table 4 illustrate that negative Poisson’s ratio (given by \(Q_{12}/Q_{11}\)) has been achieved for all the case studies undertaken. This is worth mentioning as the design of auxetic metamaterials by topology optimization is a challenging task on its own and even more difficult with uncertainties.
Maximum equivalent elastic modulus (Type IV)
Table 5 illustrates the optimal solutions for the case of type IV. To obtain these results, density based filtering and the following parameter values are considered: \(p=5\) and, \(r_{\text {min}}=2\).
It can be observed from Table 5 that uncertainty affects both the topological configuration and the equivalent elastic matrix. The resulting topologies obtained from the random field and hybrid random input models are the same. The stiffness terms (diagonal entries) \(Q_{11}\), \(Q_{22}\) and \(Q_{33}\) are observed to increase for the random field compared to the deterministic case. The stiffness terms \(Q_{11}\) and \(Q_{33}\) are observed to increase for the hybrid random input model compared to the deterministic case. These terms achieve lower values for the parametric uncertainty model compared to the deterministic case. The offdiagonal term \(Q_{12}\) is observed to be higher for all the stochastic models than that of the deterministic one. Interestingly, for the hybrid input uncertainty models, the resulting structure shows anisotropic property indicated by \(Q_{11} \ne Q_{22}\) and illustrates coupling of axial and shear deformations/stress states indicated by nonzero \(Q_{13}\) term. In terms of mean value of the objective function, the random field and hybrid input uncertainty models lead to higher value of equivalent elastic modulus compared to the deterministic and parametric uncertainty cases. The parametric stochastic model results in the minimum value of equivalent elastic modulus. Also, the random field and hybrid input uncertainty models result in a more robust topological design indicated by lower values of the standard deviation of objective function.
Conclusions
This section summarizes the results obtained from the numerical investigation. Few critical points have been enumerated below:

It has been observed that the material uncertainties affect the resulting microstructural topologies and mechanical performance significantly.

Out of the input stochastic material models, the random field model followed by the hybrid one lead to desirable performance in terms of (1) optimal mean performance, (2) most robust performance and (3) elastic property, indicated by the mean value of objective function, standard deviation of objective function and stiffness terms of the equivalent elastic matrix, respectively, for types I, II and IV.

For type III, the hybrid uncertainty model followed by the random field leads to desirable output performance (based upon combination of the above points (1)–(3)). This point along with the above are indicative of the degree of sensitivity of input stochastic models on the optimal microstructural topology. Thus, these observations have a practical relevance as one needs to be more cautious in fabricating the metastructures which follow a more sensitive input stochastic model.

The robust design leads to improved material properties compared to the deterministic design.

Multiple interesting topological patterns have been found for guiding the robust metamaterial design. Importantly, all of the resulting topologies have tessellating property, which make them feasible for 3D printing. Thus, there can be a possible extension of the present computational work as the resulting robust topological configuration of materials with extreme properties are practically realizable.

In addition to the few resulting microstructural configurations showing isotropic properties, interestingly, the others yielded configurations which showed anisotropic property. Some cases led to positive and negative coupling of axial and shear deformations/stress states. These phenomenon were primarily observed for the random field and hybrid uncertainty cases. This observation may be attributed to the fact that the spatial variability (inhomogeneity) in the microstructure induces anisotropy and coupling phenomena in the material. Although this requires further investigation and achieving this was unintentional but will prove to be useful for designing material microstructures with dual/simultaneous properties.

Considering the fact that the design of auxetic metamaterials by topology optimization is a challenging task on its own and even becomes more difficult to obtain robust designs with microstructural uncertainties, it is worth noting that negative Poisson’s ratio was achieved for all the undertaken cases of type III using the proposed framework by adopting appropriate parameters (Table 4). However, one should be careful in choosing the parameters considering their high sensitivity on the resulting topologies (for example, \(\delta\) of Eq. (12)) as evident from Table 3. It is also recommended to thoroughly scrutinize the existing equivalent expressions for minimizing the Poisson’s ratio and is highly sensitive to the problem type or application in hand.

The effect of uncertainties on the material microstructures is so high that even considering the accuracy and high precision of additive manufacturing techniques, accounting nominal amount of uncertainty for material design is recommended, at least in the conceptual design stage before the manufacturing. In fact, there is nothing at stake, as it is shown in this study that robustness improves the design in terms of mechanical performance.
Data availability
All relevant data are available from the corresponding author upon reasonable request, and/or are included within the main part and Supplementary Information.
References
Solymar, L. & Shamonina, E. Waves in Metamaterials (Oxford University Press Inc, USA, 2009).
Christensen, J., Kadic, M., Kraft, O. & Wegener, M. Vibrant times for mechanical metamaterials. Mrs Commun. 5, 453–462 (2015).
Hussein, M., Leamy, M. & Ruzzene, M. Dynamics of phononic materials and structures: Historical origins, recent progress, and future outlook. Appl. Mech. Rev. 66, 040802 (2014).
Surjadi, J. U. et al. Mechanical metamaterials and their engineering applications. Adv. Eng. Mater. 21, 1800864 (2019).
Lu, Y., Yang, Y., Guest, J. K. & Srivastava, A. 3d phononic crystals with ultrawide band gaps. Sci. Rep. 7, 43407 (2017).
Yu, X., Zhou, J., Liang, H., Jiang, Z. & Wu, L. Mechanical metamaterials associated with stiffness, rigidity and compressibility: A brief review. Prog. Mater. Sci. 94, 114–173 (2018).
Choi, M.J., Oh, M.H., Koo, B. & Cho, S. Optimal design of lattice structures for controllable extremal band gaps. Sci. Rep. 9, 9976 (2019).
Mukhopadhyay, T., Naskar, S. & Adhikari, S. Anisotropy tailoring in geometrically isotropic multimaterial lattices. Extreme Mech. Lett. 40, 100934 (2020).
Karlicic, D., Cajic, M., Chatterjee, T. & Adhikari, S. Wave propagation in mass embedded and prestressed hexagonal lattices. Compos. Struct. 256, 113087 (2021).
Singh, A., Mukhopadhyay, T., Adhikari, S. & Bhattacharya, B. Equivalent inplane voltagedependent elastic moduli of piezoelectric 2d lattices. Int. J. Solids Struct. 208–209, 31–48 (2021).
Adhikari, S., Mukhopadhyay, T. & Liu, X. Broadband dynamic elastic moduli of honeycomb lattice materials: A generalized analytical approach. Mech. Mater. 157, 103796 (2021).
Osanov, M. & Guest, J. K. Topology optimization for architected materials design. Ann. Rev. Mater. Res. 46, 211–233 (2016).
Bendsoe, M. P. & Sigmund, O. Topology Optimization: Theory, Methods and Applications (Springer, Berlin, 2004).
Xu, J., Gao, L., Xiao, M., Gao, J. & Li, H. Isogeometric topology optimization for rational design of ultralightweight architected materials. Int. J. Mech. Sci. 166, 105103 (2020).
Gibiansky, L. V. & Sigmund, O. Multiphase composites with extremal bulk modulus. J. Mech. Phys. Solids 48, 461–498 (2000).
Huang, X., Radman, A. & Xie, Y. Topological design of microstructures of cellular materials for maximum bulk or shear modulus. Comput. Mater. Sci. 50, 1861–1870 (2011).
Xia, L. Multiscale Structural Topology Optimization, 1st edn. (ISTE Press, Elsevier, 2016). https://doi.org/10.1016/C20150012540.
Clausen, A., Wang, F., Jensen, J. S., Sigmund, O. & Lewis, J. A. Topology optimized architectures with programmable poisson’s ratio over large deformations. Adv. Mater. 27, 5523–5527 (2015).
Long, K., Du, X., Xu, S. & Xie, Y. M. Maximizing the effective young’s modulus of a composite material by exploiting the poisson effect. Compos. Struct. 153, 593–600 (2016).
Zhang, H., Luo, Y. & Kang, Z. Bimaterial microstructural design of chiral auxetic metamaterials using topology optimization. Compos. Struct. 195, 232–248 (2018).
Chen, W. & Huang, X. Topological design of 3d chiral metamaterials based on couplestress homogenization. J. Mech. Phys. Solids 131, 372–386 (2019).
Wang, Y., Liao, Z., Shi, S., Wang, Z. & Poh, L. H. Datadriven structural design optimization for petalshaped auxetics using isogeometric analysis. CMES Comput. Model. Eng. Sci. 122, 433–458 (2020).
Zheng, Y., Wang, Y., Lu, X., Liao, Z. & Qu, J. Evolutionary topology optimization for mechanical metamaterials with auxetic property. Int. J. Mech. Sci. 179, 105638 (2020).
Ye, M., Gao, L. & Li, H. A design framework for gradually stiffer mechanical metamaterial induced by negative poisson’s ratio property. Mater. Des. 192, 108751 (2020).
Sigmund, O. Manufacturing tolerant topology optimization. Acta Mechanica Sinica 25, 227–239 (2009).
Wang, F., Lazarov, B. S. & Sigmund, O. On projection methods, convergence and robust formulations in topology optimization. Struct. Multidiscip. Optim. 43, 767–784 (2011).
Schevenels, M., Lazarov, B. & Sigmund, O. Robust topology optimization accounting for spatially varying manufacturing errors. Comput. Methods Appl. Mech. Eng. 200, 3613–3627 (2011).
Jansen, M. et al. Robust topology optimization accounting for misplacement of material. Struct. Multidiscip. Optim. 47, 317–333 (2013).
Zheng, Y., Da, D., Li, H., Xiao, M. & Gao, L. Robust topology optimization for multimaterial structures under interval uncertainty. Appl. Math. Model. 78, 627–647 (2020).
Wu, Y., Li, E., He, Z., Lin, X. & Jiang, H. Robust concurrent topology optimization of structure and its composite material considering uncertainty with imprecise probability. Comput. Methods Appl. Mech. Eng. 364, 112927 (2020).
Cai, J., Wang, C. & Fu, Z. Robust concurrent topology optimization of multiscale structure under single or multiple uncertain load cases. Int. J. Numer. Methods Eng. 121, 1456–1483 (2020).
He, Z. C. et al. Robust topological design of actuatorcoupled structures with hybrid uncertainties. Acta Mech. 231, 1621–1638 (2020).
Stabile, L., Scungio, M., Buonanno, G., Arpino, F. & Ficco, G. Airborne particle emission of a commercial 3d printer: the effect of filament material and printing temperature. Indoor Air 27, 398–408 (2017).
Melenka, G. W., Schofield, J. S., Dawson, M. R. & Carey, J. P. Evaluation of dimensional accuracy and material properties of the makerbot 3d desktop printer. Rapid Prototyp. J. 21, 618–627 (2015).
Fabro, A. T., Meng, H. & Chronopoulos, D. Uncertainties in the attenuation performance of a multifrequency metastructure from additive manufacturing. Mech. Syst. Signal Process. 138, 106557 (2020).
Xia, L. & Breitkopf, P. Design of materials using topology optimization and energybased homogenization approach in matlab. Struct. Multidiscip. Optim. 52, 1229–1241 (2015).
Xia, Z., Zhou, C., Yong, Q. & Wang, X. On selection of repeated unit cell model and application of unified periodic boundary conditions in micromechanical analysis of composites. Int. J. Solids Struct. 43, 266–278 (2006).
Sigmund, O. Morphologybased black and white filters for topology optimization. Struct. Multidiscip. Optim. 33, 401–424 (2007).
Andreassen, E., Clausen, A., Schevenels, M., Lazarov, B. S. & Sigmund, O. Efficient topology optimization in matlab using 88 lines of code. Struct. Multidiscip. Optim. 43, 1–16 (2011).
Amstutz, S., Giusti, S. M., Novotny, A. A. & de Souza Neto, E. A. Topological derivative for multiscale linear elasticity models applied to the synthesis of microstructures. Int. J. Numer. Methods Eng. 84, 733–756 (2010).
Acknowledgements
Dr. T. Chatterjee and Prof. Michael I. Friswell gratefully acknowledge the support of the Engineering and Physical Sciences Research Council through the award of a Programme Grant “Digital Twins for Improved Dynamic Design”, grant number EP/R006768.
Author information
Authors and Affiliations
Contributions
TC conducted the experiments, generated the results and prepared the manuscript draft. All the authors reviewed the manuscript, critically examined the results and checked the data and mathematical equations.
Corresponding author
Ethics declarations
Competing interest
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Chatterjee, T., Chakraborty, S., Goswami, S. et al. Robust topological designs for extreme metamaterial microstructures. Sci Rep 11, 15221 (2021). https://doi.org/10.1038/s4159802194520x
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s4159802194520x
This article is cited by

Machine Learning in Biomaterials, Biomechanics/Mechanobiology, and Biofabrication: State of the Art and Perspective
Archives of Computational Methods in Engineering (2024)

Topology optimization under microscale uncertainty using stochastic gradients
Structural and Multidisciplinary Optimization (2023)

Ideal nodal rings of onedimensional photonic crystals in the visible region
Light: Science & Applications (2022)

Deep transfer operator learning for partial differential equations under conditional shift
Nature Machine Intelligence (2022)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.