A flight test based deep learning method for transition heat flux prediction in hypersonic flow

Computational fluid dynamics predictions based on machine learning methods have become an important area of turbulence and transition research. However, the otherwise efficient and low-cost transition models based on Reynolds-averaged Navier–Stokes (RANS) methods have limited capability for dealing with hypersonic conditions, owing to the strong compressibility and multimodal features that are then present. This paper develops an augmented method for transition heat flux prediction. A deep neural network (DNN) is trained using flight test data from the China Aerodynamics Research and Development Center. The subject of the flight test is an inclined blunt cone on which temperature sensors are mounted. The training data consist of RANS solutions and flight test data, with the input being the mean strain/rotation rate tensor from RANS and the output the heat flux values from the flight test. The trained DNN model based on the RANS results can give heat flux values with similar accuracy to those from the flight test. For the blunt cone, the trained DNN model can accurately forecast the heat distribution caused by the Mack mode and the cross-flow transition under various inflow conditions, and the errors in the prediction results are all within 15%. Furthermore, the generalizability of the trained DNN model is also verified on an elliptic cone under different inflow conditions. This paper provides a new transition prediction approach with low computational cost and high accuracy. The proposed method solves the problem that the transition model fails in some working conditions and avoids re-modifying empirical criteria in the RANS model. It has both advantages of a transition model and flight tests and maintains the excellent potential for application.