DYNAMIC STALL PREDICTION THROUGH COMBINING PHYSICAL MODELS AND MACHINE LEARNING

The dynamic stall problem has received much attention in the field of flight safety. However, highly accurate dynamic stall prediction remains a challenge due to the complexity of the flow. To make full use of the characteristics of different data sources to establish a reasonable dynamic stall aerodynamic time-domain prediction model, an embedded integrated neural network architecture is proposed, which can realize the fusion of typical multi-source data such as numerical simulation results, physical models and wind tunnel test data. The model effectively reduces the sample demand for unsteady wind tunnel test data in the dynamic stall problem, and significantly improves the accuracy and generalization capability in the dynamic stall prediction of wing and wide-body airliner standard models. For the large-scale nonlinear and unsteady dynamic stall aerodynamic performance prediction problem, the data fusion method embedded in a physical model shows stronger robustness and is more suitable for learning from small sample data than the traditional black-box model.