Stall Ang e of Attack P rediction of Ridge Ice on Airfoil Using Deep Neural Networks

The study of ridge ice stall is of great significance to grasp the aerodynamic hazard mechanism and guide the anti-icing system design. Aiming at the forward quarter round ridge ice, this paper applied deep neural network to classify the data, so as to predict the stall angle of attack. The flow field of NACA23012 airfoil under different ridge shapes, Reynolds numbers and angles of attack are obtained through numerical calculation. Flow results are divided into dangerous and safe cases according to whether the separated flow behind the ridge ice is bonded to the airfoil surface again. The flight safety prediction accuracy of the classification model is higher than 90%. The results based on the prediction model show that the stall angles of attack are greatly reduced in ridge ice states; the change of ridge radius plays a decisive role in aerodynamic performance; the anti-icing system shall focus on avoiding the ice ridge radius greater than 0.005 times the chord length and try to avoid the ridge ice growing at the chord position of 0.12-0.16 times the chord length.