Friction measurement of aircraft wing based on optimized FlowNet2.0

Abstract

Few studies have applied the deep optical flows model to global friction measurements of aircraft wing. This study used an optimized FlowNet2.0 model to measure friction of wing based on fluorescent oil film, which achieved the first integration of deep learning and friction measurement. Two input images of the traditional FlowNet2.0 model were extended to multiple images so that more flow features and details could be with them. It is the specific part of optimization that will also further improve the measurement accuracy of FlowNet2.0. Simulation experimental results show that the optimized FlowNet2.0 model reduces the Mean Absolute Percentage Error (MAPE) error by 8.51% and increases Root Mean Square Error (RMSE) by only 0.0138 when compared to the hybrid optical flow method, which indicate that the optimized FlowNet2.0 model has great potential for friction measurement. Measurements in continuous transonic wind tunnel tests demonstrate that FlowNet2.0 can calculate clearer and more accurate flow velocity than hybrid optical flow, and the solved friction magnitude distribution is consistent with the actual flow, which will be of great practical application in wind tunnel engineering.