Estimating Model-Form Uncertainty in RANS Turbulence Closures for NICFD Applications

Abstract

The investigation of applications entailing non-ideal compressible fluid flows largely relies on Computational Fluid Dynamics (CFD) simulations. Popular CFD models rely on the Reynolds-Averaged Navier-Stokes equations (RANS). In RANS computations, turbulence models must be employed to reconstruct the Reynolds stress term arising from the time-averaged decomposition of the Navier-Stokes equations. Though literature is teeming with works supporting the development of turbulence closures for flows of fluids of common interest, little, if not just exploratory works, can be found regarding non-ideal compressible fluid flows. Moreover, a scarce amount of experimental data prevents the empirical development of turbulence models of general validity for non-ideal flows. Recently, formal estimation techniques have been developed to provide a characterization of the prediction uncertainty related to the model-form error inherent to the structure of turbulence closures. Here, we present the application of the Eigenspace Perturbation Method (EPM) from Emory et al. (2011) to a use case fundamental to non-ideal compressible fluid flows i.e., the backward facing step. The use case consists of an infinite width channel, and it is simulated both in the subsonic and supersonic regimes. Results show how predictions are affected by the model-form uncertainty inherent turbulence closures.