Thermodynamically Consistent Machine Learning for Fracture Prediction in Heterogeneous Materials

Name: Thermodynamically Consistent Machine Learning for Fracture Prediction in Heterogeneous Materials
Uploaded: 2026-10-18T10:48:29.887947+00:00
Duration: 1019 s
Description: The tLaSDI-pGFINNs framework predicts fracture patterns and evolution in heterogeneous materials by embedding thermodynamic laws like energy conservation into a neural network. It uses parametric GENERIC Formalism-Informed Neural Networks (pGFINNs) to learn thermodynamic state variables in a lower-dimensional latent space, guided by a geometry-conditioned autoencoder for accurate reconstruction.

The tLaSDI-pGFINNs framework predicts fracture patterns and evolution in heterogeneous materials by embedding thermodynamic laws like energy conservation into a neural network. It uses parametric GENERIC Formalism-Informed Neural Networks (pGFINNs) to learn thermodynamic state variables in a lower-dimensional latent space, guided by a geometry-conditioned autoencoder for accurate reconstruction.

Open the full episode