Probabilistic Global-Local Calibration of Crystal Plasticity Parameters for Additively Manufactured Metals Using Synthetic Data

Variability in additive manufacturing (AM) processes causes uncertainty in the microstructure and mechanical performance of AM metals. Uncertainty quantification (UQ) using experiments alone is costly, especially when mechanical allowables must be determined. Crystal plasticity (CP) simulations instead provide a physics-based approach to propagate microstructural uncertainty to mechanical behavior. However, establishing trust in CP models requires calibration and validation, ideally in a probabilistic setting to quantify parameter uncertainties. This work presents probabilistic calibration of a full-field CP model using synthetic global and local stress-strain data from a representative AM microstructure. A key question addressed in the work is how the type and amount of local data influences the calibrated parameter distributions. Challenges for overcoming the computational expense of Bayesian calibration are also addressed. The simulations are completed using a Python library, Materialize, developed at NASA Langley Research Center to streamline linkages between computational materials models and UQ in AM research.