Uncertainty Quantification for Crystal Plasticity Modeling Using a Bayesian Inferential Framework

摘要

This work is to support the development of Integrated Computational Materials Engineering (ICME) and the successful integration of material information in computational models spanning a range of length and time scales. In order to successfully integrate models and produce reliable material predictions, various sources of uncertainty must be taken into consideration and the fidelity of model predictions must be quantified. Statistical tools for UQ are applied to CP modeling to create a framework to quantify the accuracy of model predictions given the calibration data available by taking into account various sources of uncertainty. Furthermore, the use of reduced-order models are explored for design work since the models traditionally used for design purposes are very expensive and difficult to optimize and evaluate for UQ. Adapting UQ tools for CP modeling provides not only a quantification on the level of model fidelity but also allows uncertainty to be propagated between models through the creation of model predictive distributions. A framework for UQ is developed with the reduced-order VPSC model using Bayesian Inference and a Metropolis-Hastings MCMC algorithm.