Intelligent Computation and Analysis of Mechanical Behaviour in Piezoelectric Metamaterials Based on Physics-Informed Neural Networks

Piezoelectric metamaterials, serving as critical functional media in high-end equipment, face significant design challenges due to the mesh bottlenecks of traditional finite element methods and the interpretability shortcomings of purely data-driven models. Physical Information Neural Networks (PINNs) establish a robust scientific machine learning paradigm by embedding physical equations, offering an innovative solution to these predicaments. This paper systematically reviews recent advancements of PINNs in piezoelectric metamaterial analysis and design: drawing upon multiscale modelling theory, it elucidates PINNs' mesh-free advantages in handling high-dimensional parameters and their exceptional capability in solving small-sample inverse problems; subsequently, it explores their application paradigms in constructing high-fidelity forward surrogate models and accelerating efficient topology optimisation. Finally, this paper summarises key computational challenges in multi-physics coupling scenarios and outlines potential pathways towards achieving high-fidelity intelligent design, aiming to bridge the existing gap between theoretical modelling and engineering practice in piezoelectric metamaterials.