Predictions of Steady and Unsteady Flows Using Machine-Learned Surrogate Models
TimeFriday, 13 November 202011:50am - 12:15pm EDT
DescriptionThe research focuses on investigation of the ability of neural networks to learn correlation between desired modeling variables and flow parameters, thereby providing a surrogate model that can used in computational fluid dynamics (CFD) simulations. Investigation is performed for two different classes of problem: turbulence modeling and rotor modeling. For the turbulence modeling, a machine-learned model is applied for unsteady boundary layer flow, and the predictions are validated against direct numerical simulation (DNS) data and compared with 1-Eq. unsteady Reynolds Averaged Navier-Stokes (URANS) results. The machine-learned model performs much better than the 1-Eq. model due to its ability to incorporate the non-linear correlation between the turbulent stresses and rate-of-strain. The ongoing research is focusing on generating a larger high-fidelity solution database encompassing steady and unsteady boundary layer flows to train a generic machine-learned regression map, including investigation of approaches for data reductions. The development of the surrogate rotor model builds on the hypothesis that if a model can mimic the axial and tangential momentum deficit generated by a blade resolved model, then it should produce a qualitatively and quantitatively similar wake recovery. An initial validation of the hypothesis was performed, and the ongoing research is focusing on development of a regression map for the rotor axial and tangential momentum deficit. The final version of the paper will include results from the ongoing work.