Deep Learning Accelerated Design of Mechanically Efficient Architected Materials

Abstract

Lattice structures are known to have high performance-to-weight ratios because of their highly efficient material distribution in a given volume. However, their inherently large void fraction leads to low mechanical properties compared to the base material, high anisotropy, and brittleness. Most works to date have focused on modifying the spatial arrangement of beam elements to overcome these limitations, but only simple beam geometries are adopted due to the infinitely large design space associated with probing and varying beam shapes. Herein, we present an approach to enhance the elastic modulus, strength, and toughness of lattice structures with minimal tradeoffs by optimizing the shape of beam elements for a suite of lattice structures. A generative deep learning-based approach is employed, which leverages the fast inference of neural networks to accelerate the optimization process. Our optimized lattice structures possess superior stiffness (+59%), strength (+49%), toughness (+106%), and isotropy (+645%) compared to benchmark lattices consisting of cylindrical beams. We fabricate our lattice designs using additive manufacturing to validate the optimization approach; experimental and simulation results show good agreement. Remarkable improvement in mechanical properties is shown to be the effect of distributed stress fields and deformation modes subject to beam shape and lattice type.

Cubically Symmetric Mechanical Metamaterials Projected from 4th-Dimensional Geometries Reveal High Specific Properties in Shear

“…These methods are also often accompanied by various optimization frameworks employing metaheuristic search algorithms such as gradient-descent, 32−34 Bayesian methods, 35−37 and natureinspired algorithms. 13,38,39 With regard to the latter category of algorithms, swarm intelligence 40,41 and single or multiple objective genetic algorithm-based solutions 28,38,42,43 are among the most commonly used. Several examples of machine learning 28,44 and neural network-powered design tools 11,45−48 are also being applied in metamaterial design.…”

“…13,38,39 With regard to the latter category of algorithms, swarm intelligence 40,41 and single or multiple objective genetic algorithm-based solutions 28,38,42,43 are among the most commonly used. Several examples of machine learning 28,44 and neural network-powered design tools 11,45−48 are also being applied in metamaterial design. Recent publications suggest that new breakthroughs in machine learning have the potential to effectively solve complex design problems, which were previously not possible.…”

“…From among the papers that have considered shear loading, pentamode structures are interesting as they exhibit a vanishing shear modulus, i.e., these are solids that behave like fluids in that they are difficult to compress yet easy to deform. ,− As such, the “highly rubbery” behavior of pentamode structures has been found to be useful in a wide range of applications varying from 3D-printed shoes to seismic isolation devices . Other unusual metamaterial properties in shear include coupling effects between normal strain and shear, − which can usually be achieved through chiral architectures or structures that absorb , and dissipate energy in shear. , There are only a limited number of papers that focus on enhancing the shear properties of metamaterials, with some demonstrating high shear stiffness of lattices − and thin-wall-based structures. ,, A common practice is to incorporate computational design with finite element analyses ,,,− ,, to accelerate the development of new metamaterials. These methods are also often accompanied by various optimization frameworks employing metaheuristic search algorithms such as gradient-descent, − Bayesian methods, − and nature-inspired algorithms. ,, With regard to the latter category of algorithms, swarm intelligence , and single or multiple objective genetic algorithm-based solutions ,,, are among the most commonly used.…”

“…These methods are also often accompanied by various optimization frameworks employing metaheuristic search algorithms such as gradient-descent, − Bayesian methods, − and nature-inspired algorithms. ,, With regard to the latter category of algorithms, swarm intelligence , and single or multiple objective genetic algorithm-based solutions ,,, are among the most commonly used. Several examples of machine learning , and neural network-powered design tools ,− are also being applied in metamaterial design. Recent publications suggest that new breakthroughs in machine learning have the potential to effectively solve complex design problems, which were previously not possible. , These techniques are usually employed to perform parametric ,, or topology optimization , as well as multiscale optimization, − as a part of surrogate models , or inverse design frameworks. ,− …”

A Data-Driven Framework for Designing Microstructure of Multifunctional Composites with Deep-Learned Diffusion-Based Generative Models

“…Stress-strain curves are chosen as the model's result because they are difficult to predict, given their high dimensionality. This approach is conducted in many fields to reduce the computational resources for composite microstructures and lattice structures, protein structures, and origami/kirigami structures [40][41][42]. Beyond the prediction of stress-strain response, there have been numerous efforts to investigate the use of DL models for generating microstructures that have specific desired morphology [43][44][45][46][47][48][49][50][51], as well as models that can predict mechanical and thermal behavior based on the microstructure geometry [46,52].…”