Constitutive model characterization and discovery using physics-informed deep learning

Constitutive models are fundamental blocks of modeling physical processes, where they connect conservation laws with the kinematics system. They often expressed in form linear or nonlinear systems ordinary differential equations (ODEs). Within regimes, however, it is challenging to characterize these constitutive models. For solids and geomaterials, relations that relate macroscopic stress strain quantities described using highly nonlinear, constrained ODEs their mechanical response at different stages both reversible irreversible deformation process. A recent trend leverages complex neural network architectures construct model-free material models, such networks inefficient demand significant training data. Therefore, we believe theory-based parametric elastoplasticity still most efficient predictive. To alleviate task characterization discovery here, present a physics-informed (PINN) formulation for stress–strain modeling. The main obstacle address have inequality constraints theory embedded PINN loss functions. These crucial find correct yield surface plastic flow. We also show calibration new datasets can be performed very efficiently enhanced performance achieved even case discovery. This framework requires single dataset characterization. Although only focus on similar analogies used any