Verification, validation, uncertainty quantification and aggregation for engineering computational models in industrial applications

Abstract

Computational models and physical measurements are vital assets in understanding and optimizing the design and operation of engineered systems. However, the credibility of both models and experiments is challenged by multiple sources of uncertainty and lack of resources. The field of verification, validation, and uncertainty quantification (VVUQ) pursues systematic procedures to build credibility in physics-based computational models in the eyes of decision-makers, while maximizing the value of limited measurements. This research focuses on bridging the gap between the wealth of advancements in the VVUQ literature and the practical challenges in their application within industrial settings. Contributions of this dissertation include (i) development of an end-to-end Bayesian VVUQ framework for uncertainty quantification and aggregation, (ii) estimation of discretization error in numerical solutions with adaptively refined discretization, (iii) a novel dimension-reduced surrogate model methodology, (iv) exploration of model discrepancy in the dimension-reduced space, (v) and a multi-metric validation assessment approach for multivariate (correlated) model outputs. These contributions are illustrated through application to a realistic gas turbine engine heat transfer model.