Construction of a flow separation detection method by machine learning

Abstract

Flow separation has been attracting great interest because it affects lift reduction and drag increase. In particular, development of control methods designed to reduce flow separation requires predicting the occurrence of separation. However, predicting the occurrence of flow separation is a challenging problem due to the strong nonlinearity and high degrees of freedom of flow phenomena. Additionally, the importance of physical quantities for accurate prediction remains unclear. To address these challenges, machine learning methods are considered effective as they are data-driven and superior in capturing nonlinearities embedded in data. In this study, we use machine learning methods to predict the occurrence of flow separation. We deal with flow separation phenomena in a channel flow with a bump. As a specific problem setting, we demonstrate the prediction of the time evolution of the wall shear stress field behind the bump using a machine learning model. Then, reattachment points are determined using the sign of the predicted wall shear stress field to identify the occurrence of flow separation. The results show that the proposed method can well predict the occurrence of flow separation. In addition, we investigate the dependence of prediction accuracy on the input components and the prediction time range.