CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Bayesian polynomial chaos in multi-fidelity modelling Authors:  Pranay Seshadri - Imperial College London (United Kingdom) [presenting]
Andrew Duncan - Imperial College London (United Kingdom)
Abstract: Bayesian polynomial chaos, a Gaussian process analogue to polynomial chaos, will be introduced. Polynomial chaos represents a set of methodologies for delivering efficient aleatory uncertainty estimates for computer models. It has garnered significant industrial uptake within engineering, having seen applications in aerospace, mechanical, civil, geothermal and wind sectors. We argue why our Bayesian re-formulation of polynomial chaos is necessary and proceed to define it mathematically. The thrust will be on how Bayesian polynomial chaos is tailored for multi-fidelity uncertainty quantification, where one has to negotiate data from multiple models of varying fidelity and associated experimental data. An application of the proposed methodology on a gas turbine is presented.