Abstract
Turbulence is one of the most important phenomena in fluid dynamics. In general, turbulent phenomena can be resolved more clearly with Large Eddy Simulation (LES) compared with Unsteady Reynolds Averaged Navier-Stokes (URANS), and the numerical solution shows good agreements with that based on Direct Numerical Simulation (DNS). However, more time and computational power are needed on LES than those on URANS. If possible, the ideal simulation method is that the method is able to resolve the turbulent phenomena same quality as the results based on DNS and LES with less time and less computational power same as that on URANS. This paper shows an adaptive simulation method based on URANS and Ensemble Kalman Filter (Enkf) to reproduce the flow fields based on LES. In this study, a two-dimensional turbine cascade flow has been solved with URANS and LES. The adaptive simulation method has been also applied to the cascade flow. As the results, in the flow field of URANS with the assimilated turbulence model’s parameters, the separation phenomenon and the boundary layer thickness was close to that of the time averaged LES.
Adaptive Simulation-Based Training of Artificial-Intelligence Decision Makers Using Bayesian Optimization
