Optimization of cavitating flows simulation with data driven approach: from data assimilation to machine learning

Abstract

This paper investigates the application of data-driven approach to the optimization of cavitating flow simulations. An evaluation of the performance of commonly used RANS models (k-e, k-w and k-w SST) is presented by comparison with high fidelity data (DNS solution and X-ray experimental measurements). An ensemble based variational method is introduced and used to reconstruct the inlet velocity and calibrate the empirical parameters in the turbulence model and the cavitation model. Machine learning approach is discussed to construct a discrepancy function of the Reynolds stresses to address the RANS model-form uncertainty.