EUROPEAN CONGRESS AND EXHIBITION ON ADVANCED MATERIALS AND PROCESSES - EUROMAT 2021

This talk will first distinguish the most relevant issues concerning uncertainty in developing models for structure-property relations of materials to support development and deployment in applications. Both model form uncertainty and uncertainty of model parameters will be considered, with a focus on nonequilibrium dislocation-mediated properties/responses of metallic systems. We will distinguish between traditional “big data” applications prominent in computer science and “small data” problems in high value materials of interest for which development and experimental pathways are costly and therefore limited. Use of fast acting reduced order models that incorporate high order spatial statistics to project structure property relations will be demonstrated to characterize extreme value (rare event) high cycle fatigue behavior of alpha-beta Ti systems with various textures to facilitate comparison of microstructures. We will use machine learning strategies with Bayesian updating to quantify uncertainty associated with exercising each of several crystal plasticity frameworks relative to experimental datasets based on spherical indentation for these materials, and go one step further to fuse disparate models so as to incorporate their most salient model elements and reduce overall model form uncertainty while fitting parameters. Finally, In many cases of interest, parameters of mesoscopic reduced order models must be informed from a combination of lower scale simulations (e.g., atomistics) along with experiments to respect the uncertainty associated with interatomic potential model form; we apply such a strategy to estimate parameters of a crystal plasticity flow rule for bcc Fe, introducing an inter-scale model discrepancy formulation to account for gaps between atomic scale and intermediate scale in dislocation generation that contribute to uncertainty in the mesoscale flow rule.