Quantifying & analysing mixed aleatoric and structural uncertainty in complex turbulent flow simulations

Abstract Reynolds Averaged Navier Stokes models are the most popular approach for Computational Fluid Dynamics simulations of turbulent flows. Despite their popularity, these numerical models still need an appropriate quantification of margins and uncertainty for reliable engineering practice. In turbulent flow simulations, there are two kinds of uncertainties: aleatoric (those arising due to errors in initial conditions, material parameters, etc.) and structural or epistemic (those arising due to the turbulence model used). While these uncertainties have been explored in isolation, there are no studies that properly consider both types of uncertainty together as is the case in real life engineering applications. Considering these two sources of uncertainty in isolation limits our knowledge regarding the sources of uncertainty, their relative magnitudes, their interaction and the accuracy of uncertainty estimates. For the reasons given, a methodology is necessary to amalgamate uncertainties of different nature. In this paper is outlined a framework to carry out uncertainty quantification for such mixed uncertainty cases. The results are compared to those from aleatoric only and epistemic only studies to analyze the manner in which aleatoric and epistemic uncertainties interact and affect final results in complex turbulent flows and heat transfer. We compare and contrast the relative contributions of these sources to the overall uncertainty. The results obtained for our test cases exhibit that only considering aleatoric or epistemic uncertainty sources in isolation tends to severely under-predict the uncertainty in simulations. Utilizing the framework, we show that considering both sources together leads to a satisfactory prediction of the uncertainty in simulation results because of the underlying relations between these sources uncertainty in their propagation.