03.4 - Applied Aerodynamics
A HYBRID APPROACH FOR RECONSTRUCTION OF TRANSONIC BUFFET AERODYNAMIC NOISE: INTEGRATING RANDOM FOREST AND COMPRESSIVE SENSING ALGORITHM
Q. Zhang, Northwestern Polytechnical University, China; W. Zhang, Northwestern Polytec, China
In response to the difficulty in obtaining high-precision aerodynamic noise data, this paper establishes a comprehensive standardization process for predicting the transonic buffet aerodynamic noise of RAE2822 airfoil. Firstly, a prior criterion is proposed based on flow correlation and the prediction accuracy of the Power Spectral Density (PSD) using the Random Forest (RF) algorithm. Subsequently, we determine whether the RF algorithm can be employed to directly obtain high-precision PSD results using this criterion. Successful PSD prediction is specifically determined when monitoring points simultaneously satisfy RMSE_corr<0.05 and RMSE_adj<0.05, the RMSE_corr represents the Root Mean Square Error (RMSE) of the auto-correlation and the cross-correlation coefficients, and the RMSE_adj indicates the RMSE of the PSD for the adjacent monitoring points. If not, we introduce an innovative approach by embedding the RF model into the Compressed Sensing algorithm reconstruction process (RF_CS). This method efficiently achieves high-precision Overall Sound Pressure Level (OASPL) and PSD reconstruction based on sparse sensor positions, demonstrating good robustness and generalization capabilities. Compared to the CS algorithm based on Proper Orthogonal Decomposition (POD_CS), this method achieves high-precision PSD (OASPL) reconstruction, with the RMSE has been reduced by a factor of 2 to 50 using 22 (9) sensor positions and 15 (12) basis functions, and the method does not exhibit phenomena such as high-frequency distortion or inflection point distortion.