Finite element modeling is crucial for structural analysis but computationally intensive, particularly under dynamic loading.
A Multiple Input Operator Network (MIONet) is introduced to predict structural responses accurately under moving loads by encoding temporal dynamics with a second trunk network.
MIONet can predict responses continuously over space and time, eliminating the need for stepwise modeling and achieving FEM level accuracy within seconds.
The model uses physics-informed loss for efficiency and physical consistency, and a Schur complement formulation to reduce training domain and computational costs while maintaining accuracy.