Machine learning powered inverse design for strain fields of hierarchical architectures

Abstract

Hierarchical architectures are complex structures composed of multiple materials arranged at a microstructural level to achieve specific macroscopic properties. Despite the advantages offered by hierarchical architectures which are offering broad design freedom, this extensive design space also poses significant challenges for inverse designing hierarchical architectures. This paper addresses the inverse design of strain fields for hierarchical architectures by integrating efficient forward prediction with precise inverse optimization. Forward prediction models are developed to accurately predict the physical properties and performance metrics of these materials, while inverse optimization algorithms determine the optimal material distribution to achieve desired outcomes. We propose a machine learning approach that utilizes a recurrent neural network (RNN)-based forward prediction model trained on finite element analysis data, achieving over 99% accuracy. An evolutionary algorithm-based inverse optimization model is then used to identify the optimal material configuration to reach the desired strain fields. The results, validated through simulation and experimental testing, demonstrate the potential of machine learning to accelerate the design and optimization of strain fields in hierarchical architectures, paving the way for advanced material applications in the fields of aerospace engineering, biomedical devices, robotics, structural engineering, and energy storage systems.