Field Inversion and Machine Learning for turbulence modelling applied to three-dimensional separated flows
Joel Ho, Alastair West
Abstract
View Video Presentation: https://doi.org/10.2514/6.2021-2903.vid The Field Inversion and Machine Learning (FIML) method was applied to augment the k-ω SST turbulence model to improve the modelling of separated flows. Various flow cases including the 3D FAITH hill were inverted for a correction factor in the production term of the specific dissipation rate equation. The adjoint method was employed in the inversion. Neural network and random forest machine learning models were trained on the inverted results and tested on three unseen cases including a 3D non-axisymmetric bump. The generality of the models were tested and quantified with local outlier factor as an extrapolation metric. The results between applying the model at every iteration or as a one-time-correction on the converged uncorrected flowfield are discussed. Application of the FIML method improved the prediction of mean velocity and turbulent kinetic energy in all of the tested cases.