Non-linear mechanical field reconstruction coupling recurrent neural networks with physics-informed graph neural networks

Name: Non-linear mechanical field reconstruction coupling recurrent neural networks with physics-informed graph neural networks
Uploaded: 2026-10-15T07:29:25.852121+00:00
Duration: 421 s
Description: The proposed framework couples a Long Short-Term Memory (LSTM) network with a Physics-Informed Graph Neural Network (P-DivGNN) to reconstruct local stress fields in heterogeneous materials under complex, history-dependent loading. The LSTM captures the temporal response of the material, while the GNN uses this history to spatially map out where stresses occur at each time step.

The proposed framework couples a Long Short-Term Memory (LSTM) network with a Physics-Informed Graph Neural Network (P-DivGNN) to reconstruct local stress fields in heterogeneous materials under complex, history-dependent loading. The LSTM captures the temporal response of the material, while the GNN uses this history to spatially map out where stresses occur at each time step.

Open the full episode