Physics-Informed Deep Energy Approach with Phase-Field Modeling for Coupled Thermo-Mechanical and Hydro-Mechanical Fracture
Please login to view abstract download link
Fracture modeling under coupled thermal, mechanical, and hydraulic loading is challenging due to nonlinear crack evolution and strong multiphysics interactions. Conventional numerical methods often rely on crack tracking, remeshing, or local mesh refinement, increasing computational cost and complexity. Phase-field fracture models provide a robust alternative by representing cracks through a diffusive formulation without prescribing crack paths. This work presents a physics-informed machine learning framework that integrates phase-field fracture modeling with the Deep Energy Method. Governing variational principles are embedded directly into the training objective of physics-informed neural networks, enabling mesh-free solutions without labeled data while preserving thermodynamic and mechanical consistency. The framework is applied to coupled thermo-mechanical and hydro-mechanical fracture problems, and benchmark examples demonstrate its ability to capture fracture initiation and evolution under multiphysics loading.
