Multi-Physics-informed deep learning approach for computational structural mechanics

Finite element method (FEM) has been popular for decades for structure analysis by solving partial differential equations (PDEs). Recently, physics-informed deep learning has emerged as a promising approach for solving PDEs and shows great advantages over FEM in the computational efficacy. However, it is still a challenge for the structural mechanics computation since it involves solving higher-order PDEs as the governing equations are fourth-order nonlinear PDEs. Therefore, we develop a novel multi-physics-informed neural network (m-PINN) framework via combining multiple neural networks (NNs) where each NN representing physics information such as geometrical compatibility, constitutive relation, and static equilibrium. The verification result demonstrates that m-PINN has the same accuracy as FEM for the computation of beam and shell structures. The proposed method has the potential to be a new paradigm in the structural mechanics computation as an alternative of FEM, and could serve as an intelligent computational module in digital twin system.