Physics-Informed Data-Driven Surrogate Modeling for Full-Field 3D Microstructure and Micromechanical Field Evolution of Polycrystalline Materials

We have developed a machine learning-based crystal plasticity surrogate model (CP-SM) that can directly learn highly nonlinear material behavior during plastic deformation. CP-SM provides fast inference of spatially resolved three-dimensional (3D) microstructure and micromechanical fields and their evolution during plastic deformation, predicting the 22-dimensional material characteristics including a four-dimensional (4D) quaternion-based representation of crystal orientation, six-dimensional (6D) elastic and plastic strain tensors, and 6D stress at each location in the 3D structure. The predictions from CP-SM are orders of magnitude faster than and show good agreement with the deformation fields predicted by performing direct numerical simulations using spectral solvers. The fidelity of the CP-SM is further tested by assessing how well physics-based constraints are satisfied by the predicted 3D fields. We demonstrate our results on numerical simulations of uniaxially loaded copper.