Machine learning-enabled inverse design of bimaterial thermoelastic lattice metamaterials

In this work, we develop a general framework combining machine learning and high-throughput simulations for the inverse design of bimaterial thermoelastic lattice metamaterials with unconventional and tailorable effective properties. As a representative case study, the inverse design of a hybrid-honeycomb structure exhibiting negative Poisson’s ratios and negative thermal expansion coefficients is considered. A large dataset is generated through computational homogenization of structures with varying geometrical parameters and base material properties. A forward ML model is first trained to predict the effective thermoelastic properties for a given structure design. Subsequently, inverse ML models are developed to suggest design variables for prescribed target properties. To address different practical scenarios, six inverse models with different combinations of target properties and design variables are proposed. The trained forward model is integrated into the loss functions of inverse models and is further employed to generate additional datasets for cases with fixed materials. These ML models with demonstrated good performance can be applied to efficiently solve inverse design tasks involved in the practical application of the thermoelastic structure as structural or functional components in aerospace systems and temperature-adaptive structures. The methodology can be extended to the inverse design of other thermoelastic metamaterials. • Machine learning-enabled inverse design of thermoelastic lattice metamaterials is proposed. • High-throughput computational homogenization is integrated with machine learning methods. • Forward and inverse machine learning models with good performance are attained. • Multiple inverse design scenarios are demonstrated through representative examples.