| Abstract Scope |
Integrating equilibration-preserving deep learning (DL) techniques that satisfy equilibrium in the strong form with DOFKILL/DOFALIVE-based multiphysics frameworks offers a promising approach for modeling complex material behaviors related to structural, thermal, and diffusion physics under damage. Equilibration-preserving DL, such as Machine Learning enabled Stiffness (MLeS), ensures that learned approximations maintain physical constraints, making it suitable for multiphysics analogs of structural damage analysis. A combination (MLeS/traditional approach) with DOFKILL/DOFALIVE methods—where degrees of freedom (DOF) are selectively activated or deactivated based on combined metrics of damage and multiphysics—enables efficient partitioning of damage into brittle and ductile forms. This synergy allows the model to dynamically adapt its complexity, activating DOFs only where needed, reducing computational load without sacrificing accuracy. An example in Hydrogen Embrittlement (HE) will demonstrate damage partitioning and how coupled physics influences partitioning, order, and extent of damage. In addition, the accuracy, speedup, and DOF dropout/pickup aspects will be further discussed. |