Quantifying Multiple Types of Uncertainty in Physics-Based Simulation Using Bayesian Model Averaging

Given experimental data measured from an engineering system, response predictions by a stochastic simulation model involve both parametric uncertainty and random errors. Also, model form uncertainty arises when two or more simulation models predict the responses of an engineering system because it is beyond capability to identify the best approximating model among the considered model set. In this research, a methodology is developed to quantify model probability using measured deviations between experimental data and model predictions of the data under a Bayesian statistical framework. Model averaging is used to combine the predictions of a system response by a model set into a single prediction. Then, a nonlinear spring-mass system is used to demonstrate the process for implementing model averaging. Finally, the methodology is applied to the engineering benefits of a laser peening process, and a confidence band for a residual stress field is established to indicate the reliability of the composite prediction of the stress field. [PUBLICATION ABSTRACT]