CROSS-SCALE, CROSS-DOMAIN MODEL VALIDATION BASED ON GENERALIZED HIDDEN MARKOV MODEL AND GENERALIZED INTERVAL BAYES' RULE

摘要

Reliable simulation protocols supporting integrated computational materials engineering (ICME) requires uncertainty to be quantified. In general, two types of uncertainties are recognized. Aleatory uncertainty is inherent randomness, whereas epistemic uncertainty is due to lack of knowledge. Aleatory and epistemic uncertainties need to be differentiated in validating multiscale models, where measurement data for unconventionally very small or large systems are scarce, or vary greatly in forms and quality (i.e., sources of epistemic uncertainty). In this paper, a recently proposed generalized hidden Markov model (GHMM) is used for cross-scale and cross-domain information fusion under the two types of uncertainties. The dependency relationships among the observable and hidden state variables at multiple scales and physical domains are captured by generalized interval probability. The update of imprecise credence and model validation are based on a generalized interval Bayes' rule (GIBR).