Machine Learning Prediction of Dynamic Stall Loads using a Modal Approach

Modal decomposition and machine learning via a neural network are combined to effectively characterize and to predict dynamic stall behavior. Unsteady pressure measurements were taken over a wide range of flow conditions, and the pressure distribution for each test case was decomposed using Proper Orthogonal Decomposition into spatial modes and time-varying coefficients. The results under different conditions were used to develop a machine learning model that predicted time-varying coefficients, which were then used to reconstruct and predict pressure distributions. The predicted pressure distributions were subsequently utilized to estimate lift and moment coefficients. The pressure distributions, lift, and moment coefficients found using the machine learning model closely approximated the loads experienced during dynamic stall.