A data-driven memory model for solving turbulent flows with the pseudo-direct numerical simulation method

Abstract

It is well known that the inherent three-dimensional and unsteady nature of turbulent flows is a stumbling block for all approaches aimed at resolving their spatial and temporal variability. The pseudo-direct numerical simulation (P-DNS) method for turbulent flows, proposed by the authors in a previous publication, focused on resolving the spatial variability, leaving the task of solving the temporal evolution to a highly simplified, parameter dependent model, to be adjusted in a case by case basis. Although some auspicious results were obtained, the applicability of P-DNS for problems of industrial interest required a more sophisticated method to deal with the temporal variability. In this sense, the present work proposes a new, parameter free, data-driven memory model for P-DNS. The model is based on the study of off-line DNS solutions of turbulent flows transitioning between statistically steady states in simple domains. The new P-DNS model is tested and successfully compared against existing methods in selected three-dimensional turbulent flows.