Data-driven Bayesian inference of turbulence model closure coefficients incorporating epistemic uncertainty

Abstract We introduce a framework for statistical inference of the closure coefficients using machine learning methods. The objective this is to quantify epistemic uncertainty associated with model by experimental data via Bayesian statistics. tailored towards cases which limited amount available. It consists two components. First, treating all latent variables (non-observed variables) in as stochastic variables, sources probabilistic are quantified fully approach. defined consist parameters and other incorporating noise. Then, extracted from overall considering noise being zero. rigorously evaluated Markov-Chain Monte Carlo sampling assisted surrogate models. apply Spalart–Allmars one-equation turbulence model. Two test considered, including an industrially relevant full aircraft at transonic flow conditions, Airbus XRF1. Eventually, we demonstrate that uncertainties result into quantities interest prominent around, downstream, shock occurring over XRF1 wing. This data-driven approach could help enhance predictive capabilities computational fluid dynamics (CFD) terms reliable modeling extremes flight envelope if measured available, important context robust design virtual certification. plentiful information about also assist when it comes estimating influence on inferred coefficients. Finally, developed flexible can be applied different various