Data Refinement for Confidence Management in Model-Based Predictions

摘要

The activities under this grant were in the general area of Verification and Validation using the Polynomial Chaos formalism developed by the PI over the course of the past 15 years. In particular, three significant questions were formulated and addressed in the course of this research: (1) How to develop polynomial chaos representations of physical parameters from experimental measurements of these parameters; (2) How to propagate the uncertainty in these parameters (as reflected in their Polynomial Chaos representations) into the dynamical behavior of the physical system; (3) How important is data refinement: what is the significance, on the predictive value of a computational model, of collecting additional experimental measurements as opposed to performing further numerical refinement (using, for example, mesh refinement). The remainder of this report will review the highlights of each of the above three topics.