Recurrent Neural Network Model for Laminar-Turbulent Transition

A physics-based model has been proposed for end-to-end prediction of laminar-turbulent transition in boundary layer flows. Traditional methods lack generalizability to multiple instability mechanisms and different flow configurations, especially where transition is dependent on a large set of parameters. Neural network methods allow for higher dimensional input features, however previously proposed neural network models follow an involved methodology of predicting instability growth rates over a broad range of frequencies, which are then integrated to obtain N-factor curves for each frequency, and, then transition location is determined via empirical correlation of the envelope N-factor. We propose a recurrent neural network (RNN) with a simplified workflow, by directly predicting the N-factor envelope and, hence, the transition location. The model processes the flow information in a physically consistent manner, providing a direct link with the physics of the underlying transition mechanism. Furthermore, the simplified workflow requires minimal user expertise, allowing non-expert users to apply the RNN model to multiple instability mechanisms. The proposed model has been analyzed for an extensive dataset of two-dimensional boundary-layer flows over a diverse set of airfoils.