Helical model based on artificial neural network for large eddy simulation of compressible wall-bounded turbulent flows

Abstract

Similar to the kinetic energy cascade, a helicity cascade is also a basic and key process in the generation and evolution of the turbulent flows. Furthermore, the helicity flux (HF) plays a crucial role between two scales in the helicity cascade. In this study, we will supply a new helical model constrained by the helicity flux for the large eddy simulation of the compressible turbulent flows. Then, in order to obtain a more precise HF, the local coefficient of the modeled HF is determined by the artificial neural network (ANN) method. The new model combines merits of the high robustness and the correlation with the real turbulence. In the test case of the compressible turbulent channel flow, the new model can supply a more accurate mean velocity profile, turbulence intensities, Reynolds stress, etc. Then, for the test in the compressible flat-plate boundary layer, the new model can also precisely predict the onset and peak of the transition process, the skin-friction coefficient, the mean velocity in the turbulent region, etc. Moreover, the ANN here is a semi-implicit method, and the new model would be easier to be generalized to simulate other types of the compressible wall-bounded turbulent flows.