A0910
Title: Uncertainty quantification with a latent variable model
Authors: Mengyu Xu - University of Central Florida (United States) [presenting]
Abstract: To understand the behavior of a complex system, one is often interested in some key internal quantities that are not directly observable. The aim is to study the inference of the final output and the internal quantities of a complex system from controlled system inputs and multi-fidelity realizations of the latent internal quantities. A latent variable model is studied under a Bayesian framework. The model is two-step: (1) the inference of the latent internal variable given their noisy approximations and the system inputs; and (2) the study of the system outputs from the inferred internal variables. Linear and nonlinear approximations are employed in the second step. For the nonlinear approximation, Markov Chain Monte Carlo is employed. In addition, the inverse problem is studied, i.e., estimate the posterior of the internal quantities from their noisy measurements and the system inputs and output. This provides insight into the system's application of fault detection. The approach is verified against a numerical model, demonstrating its veracity.
