Uncertainty Quantification of Multivariate Correlated Output in the Simulation of Airfoil Flow

Abstract

Due to the limitation of human cognition, there are uncertainties in the turbulence model closure coefficients that cannot be ignored. The key to assess model objectively is quantifying the uncertainty of model coefficients scientifically. In the CFD simulation of flow around the airfoil, wall pressure distribution is a multivariate correlated quantity. This paper developed a surrogate model method based on artificial neural network to characterize the mapping relationship between multivariate input and multivariate output. In order to reduce the output dimension and modelling difficulty, this paper use Proper Orthogonal Decomposition (POD) method. An RAE2822 benchmark case under transonic conditions is used to study the uncertainty of Spalart-Allmaras (SA) turbulence model coefficients and verify the method developed by this paper.