Ensemble Kalman method for learning turbulence models from indirect observation data

In this work, we propose using an ensemble Kalman method to learn a nonlinear eddy viscosity model, represented as tensor basis neural network, from velocity data. Data-driven turbulence models have emerged promising alternative traditional for providing closure mapping the mean velocities Reynolds stresses. Most data-driven in category need full-field stress data training, which not only places stringent demand on generation but also makes trained model ill-conditioned and lacks robustness. This difficulty can be alleviated by incorporating Reynolds-averaged Navier–Stokes (RANS) solver training process. However, would necessitate developing adjoint solvers of RANS requires extra effort code development maintenance. Given difficulty, present with adaptive step size train neural-network-based indirect observation To our knowledge, is first such attempt modelling. The verified flow square duct, where it correctly learns underlying Then generalizability learned evaluated family separated flows over periodic hills. It demonstrated that one predict similar configurations varying slopes.