Machine learning-enabled self-consistent parametrically-upscaled crystal plasticity model for Ni-based superalloys

This paper introduces a concurrent multiscale modeling framework for developing parametrically-upscaled crystal plasticity models (PUCPM) for crystalline metals that are characterized by multiple phases in their intragranular microstructure. Specifically, Ni-based superalloys with distributions of gamma ' precipitates in the gamma matrix in their sub-grain microstructure are modeled in this study. The interaction of the gamma - gamma ' phases with nano-scale dislocation mechanisms determine the elasto-plastic behavior of the material across multiple scales. The PUCPM is designed to explicitly account for the morphological and configurational statistics of these gamma - gamma ' intragranular microstructures in its crystal plasticity constitutive coefficients. This approach provides a thermodynamically-consistent foundation to enable microstructure-aware material simulation at the higher scale. Establishing this multiscale characteristic requires an automated toolchain of computational methods to generate and embed heterogeneous, statistically-equivalent representative volume elements (SERVEs) into the concurrent multiscale simulation domains for self-consistent homogenization. The self-consistency condition is enforced through an optimization strategy invoking a series of coupled nonlinear finite element solutions. Supervised and unsupervised machine learning methods are integrated with the physics-based modeling at all stages of model development to overcome computational limitations and to provide the final, connecting map between PUCPM constitutive coefficients and gamma - gamma ' microstructural descriptors. The resulting PUCPM benefits from orders of magnitudes of speedup compared to the equivalent explicit representation of the lower scale microstructure. This advantage enables unique model capabilities for the multiscale analysis of deformation and failure in materials and location-specific design.(c) 2022 Elsevier B.V. All rights reserved.