Physics-Informed Neural Networks for Multiscale Large Deformation Analysis of Metamaterials

Physics-informed neural networks (PINNs) have recently emerged as a promising alternative to traditional numerical methods for solving solid mechanics problems. In this work, we propose a novel PINN architecture designed for homogenisation problems of metamaterials under large deformation. The architecture incorporates periodic functions to ensure exactly imposed boundary conditions and employs an energy-based loss for efficient training. Three representative metamaterial structures—octet truss, gyroid, and spindoid—are selected as case studies. The results demonstrate that the proposed PINN achieves accuracy comparable to finite element analysis (FEA), while offering improved computational efficiency for high-volume-fraction structures. Beyond accuracy and speed, the meshfree nature and flexibility of PINNs provide clear advantages, highlighting their potential as a scalable tool for modelling complex materials.