Dynamic properties and deep learning-based optimal design of hybrid shape-memory-alloy sandwich beams

Abstract

This study explores the design of sandwich composite materials integrating a shape-memory-alloy core. The energy-dissipating property of shape memory alloys (SMAs), associated with their underlying phase transformation, can help enhance the overall damping capacity while maintaining required stiffness performance. However, the design of the core poses significant challenges due to the highly nonlinear behavior of SMAs. In this work, an optimization framework for the design of nonlinear sandwich beams with SMA cellular cores is presented. The approach relies on a combination of a deep learning (DL)-based surrogate model with an evolutionary optimizer. While trained on a limited number of configurations, the surrogate model can accurately predict the nonlinear response of the SMA sandwich beam (SSB) integrating an arbitrary core design, accelerating the nonlinear numerical simulations by six orders of magnitude. Then, by incorporating the surrogate model into an evolutionary algorithm, the multiobjective optimization problem can be solved under geometric constraints. In particular, optimal core designs are sought to maximize damping while simultaneously satisfying stiffness and SMA volume requirements. The obtained optimal configurations are then validated via full field finite element analysis, showing that the proposed framework can effectively deliver SSB designs with tailored performance.