A mesh-based neural energy method for the simulation of heterogeneous composites
This work introduces a mesh-based neural energy method (M-NEM) to simulate complex materials like heterogeneous composites. It improves upon conventional methods by using shape functions for mesh discretization, which suppresses oscillations and replaces costly automatic differentiation with algebraic derivative calculations. The M-NEM achieves significantly higher accuracy and computational efficiency than standard methods.