Propeller noise performance prediction based on artificial intelligence method and rotor modal relationships

Efficient acquisition of high-confidence noise performance data is essential for optimizing propeller noise characteristics. A machine learning-based model, referred to as the Propeller Noise Spectrum Deep Learning (PNSDL) model, has been proposed for the prediction of propeller noise. This algorithm integrates the Noise Spectrum Deep Learning (NSDL) model to identify and predict the deep spectral noise features of the propeller, while a CNN-MHA(CM) model, based on rotor modal relationships, is employed to predict and correct the pure tone noise features of the propeller. Based on the flow field and noise mechanism of counter-rotating propellers calculated using the nonlinear harmonic method, this study conducts a quantitative analysis and discussion of the model's prediction performance based on wind tunnel noise test data. The results demonstrate that the PNSDL, by rotor modal relationships, substantially enhances the NSDL's ability to recognize and predict complex propeller noise features. For the prediction of single-rotor propeller spectral noise, 85% of the predicted conditions exhibit a relative error of less than 5%, while for counter-rotating propellers, 87.17% of the predicted conditions show a relative error of less than 5%. A primary cause of prediction errors in the PNSDL for certain propeller pure tone noises lies in the limited explanatory power of the rotor modal relationships for complex propeller noise features. Continued exploration of more intricate noise mechanisms and their integration into the model is expected to further enhance the model's prediction accuracy.