A Data-Driven Algebraic Intermittency Model for Separation-Induced Transitional Flows

Laminar-to-turbulent transition can occur through several distinct mechanisms. These varied transition pathways pose significant challenges for existing transition models, which may struggle to accurately predict key features such as transition onset, length, separation, and reattachment. Data-driven techniques have recently been explored to improve the performance of existing RANS models in predicting transitional flows through data assimilation. However, their application to separation-induced transitional flows for isolated airfoils is less explored.

In this work, we try to bridge this gap by presenting a training approach to improve the ability of a baseline RANS turbulence model to predict separation-induced transitional flows over low Reynolds number airfoils. We employ the learning and inference assisted by feature space engineering framework to train the underlying model parameters to predict the intermittency field that augments the baseline turbulence model. The training is performed using steady state airfoil LES data at different angles of attack. Then, the trained model is applied to new steady state unseen cases with different airfoil sections and Reynolds numbers. Eventually, the trained model is used to simulate the flow over pitching airfoils at low Reynolds number to characterize their dynamic stall onset. The performance of the trained model is evaluated in comparison to LES data and other existing transition models.