Tuning Mechanical Properties in Polycrystalline Solids Using a Deep Generative Framework

Abstract

Crystalline solids, which often possess distinguished mechanical properties among distinct orientations, may be potentially further recombined into various composite materials as an effective design strategy. [1][2][3][4][5][6] Among different types of composite design methods, better mechanical properties may be achieved by making use of the anisotropy and grain boundary effects seen in nanomaterials. [7][8][9] Considering any given space of a material, there is potential for microstructural design to improve its performance. However, in most cases, the cost of simulation or experiment will rapidly increase along with the daunting complexity of the sheer number of possible combinations within a given design space, and therefore the traditional process of composite design with experiment or simulation on crystalline solids is not capable of discovering new composite materials in a productive way. Here, the use of artificial intelligence (AI) methods [10,11] can be a powerful way to accelerate the discovery process [12][13][14][15][16][17][18][19][20][21] and, as we will show in this article, can also provide important mechanistic insights into the structure-property relationships. As such it complements other multiscale models in our quest to better understand how materials perform under extreme conditions.We adopt a framework that combines a deep learning (DL) model with a genetic algorithm (GA) [12,15,21] to rapidly search and generate composite materials on demand-simulating "nature's evolution paradigm" [19,22] -to achieve different mechanical properties, a process that enables us to realize "inverse design" through a continuous assessment of variations in design suggestions. Figure 1 shows a summary view of the method reported in this study. The deep learning model in this framework serves as a surrogate model for molecular dynamics (MD) simulation to help us obtain certain properties much faster, [15] offering a novel way to discover key mechanical feature of a material. As a result, the GA can perform a computationally efficient optimization of polycrystalline composites by using the DL-derived simulation results as feedback. Overall, the goal of this framework is to automatically determine optimized polycrystalline solid design

Untapped Potential of Deep Eutectic Solvents for the Synthesis of Bioinspired Inorganic–Organic Materials

“…Generative neural network models can also be used to generate de novo protein designs (Figure 9) with specific mechanical behaviors by coupling it with coarse grained analysis and has been demonstrated to solve the inverse problem of correlating and tuning polycrystalline materials to mechanical properties in which this framework can be used to guide de novo biomaterial design. 360 Similar approaches to what has been accomplished previously can be taken by studying DESs. A model could be developed that is capable of both understanding DESs and even discovering new DESs.…”

Optimization of vascular structure of self-healing concrete using deep neural network (DNN)

“…Besides, the optimization may be stuck in sub-optimal when using some gradient-based topology optimization methods. Recently, optimization with machine learning (ML) methods has been proposed in material design for computational efficiencies [26] as well as capacity of overcoming local minima [27]. Gu and her colleagues designed composite materials with better quality using generative deep neural network [28] and deep reinforcement learning [29].…”

Inverse design of truss lattice materials with superior buckling resistance

“…From the computational side, deep-learning models 20 , specifically, represent a promising alternative to physics-based models for materials design, providing much faster (several orders of magnitude) yet accurate structure-property relationship predictions, thus allowing efficient design space exploration [21][22][23][24][25][26][27][28] . From composites [29][30][31][32][33] , through complex symmetric architectured materials 34 , stretchable kirigami-inspired-cut materials 35,36 , spinodoid metamaterials 37 , up to polycrystalline solids 38 , several studies have attempted to solve the inverse design problem exploiting the powerful computational and predictive capabilities provided by deep-learning techniques, mainly deep neural networks, used either as generative 31,36,39,40 or surrogate forward models coupled with other optimization methods 32,41 (e.g., evolutionary algorithms). Yet, only a few studies have provided solutions for the inverse design of truss lattice materials, mainly focusing on (i) pre-existing architectures conveniently modified to obtain lattices with desired properties, such as tunable stiffness anisotropy 42 and stronger micro-lattices with arranged defects 43 ; (ii) single complex 3D novel unit cells with load carrying applications, such as stronger lattice cores for sandwich structures 44,45 ; (iii) basic architectures (such as a square lattice), on which reinforcements are non-uniformly added in order to match the desired mechanical response from a database 46 ; (iv) targeted linear properties (in solid mechanics terms), such as lattice' stiffness 39,41,42,47 and Poisson's ratio 48 .…”

“…It is worth underlying that the idea of combining an ML model with a genetic algorithm to solve inverse problems is not new in general 24,32,41,49 . In solid mechanics, specifically, strong yet tough polycrystalline materials subject to complex fracture mechanisms 38,50 (e.g., crack branching) as well as stiff and strong digital composites 32 have been inverse-designed by exploiting deep learning models as efficient surrogate solvers combined with genetic algorithms. We hence take advantage of the demonstrated ability of this combined model to discover highperforming designs in complex mechanics applications and simplicity, applying it in a bottom-up framework to inversedesign truss lattices undergoing high nonlinear phenomena (see next).…”