Synchronization and optimization of Large Eddy Simulation using an online Ensemble Kalman Filter

Abstract

An online Data Assimilation strategy based on the Ensemble Kalman Filter (EnKF) is used to study the features of parametric optimization and synchronization of the physical state when applied to scale-resolved numerical simulations. To this purpose, the method is combined with Large Eddy Simulation (LES) for the analysis of the turbulent flow in a plane channel, $R{e}_{\tau }\approx 550$. The algorithm sequentially combines the LES prediction with high-fidelity, sparse instantaneous data obtained from a Direct Numerical Simulation (DNS). It is shown that the procedure provides an augmented state that exhibits higher accuracy than the LES model and it synchronizes with the time evolution of the high-fidelity DNS data if the hyperparameters governing the EnKF are properly chosen. In addition, the data-driven algorithm is able to improve the accuracy of the subgrid-scale model included in the LES, the Smagorinsky model, via the optimization of a free coefficient. However, while the online EnKF strategy is able to reduce the global error of the LES prediction, a discrepancy with the reference DNS data is still observed because of structural flaws of the subgrid-scale model used.