Intelligent prediction of acoustic performance of landing gear using deep learning

Efficient prediction and evaluation of noise performance are crucial to the design and the optimization of landing gear noise. A systematic method is developed to predict and evaluate landing gear noise in the present study, termed as noise spectrum deep learning model (NSDL). In this algorithm, the encoder and decoder are designed to extract noise features and reconstruct noise data. Specifically, a loss function that takes the identification of both broadband noise and tone noise into account is utilized to guide the training direction of the model, aiming to improve the training efficiency and prediction results of the model. Afterward, the mapping relationship between landing gear experimental parameters and noise features is established by multi-layer perceptron. In this study, the detail of the algorithm is analyzed and discussed based on the results of wind tunnel noise experiment and numerical simulation. The results show that the proposed model can effectively and precisely predict landing gear noise under various conditions, including different flow speeds, angles of attack, number of wheels, and heights of the main strut. For the inflow velocity range of 34–75 m/s, the average error of the overall sound pressure level is restricted below 0.83% (0.6 dB). In case only the angle of attack is changed, the average error is reduced to be less than 0.36% (0.3 dB). The prediction results show that the landing gear noise is mainly broadband noise and tone noise mainly appears in the low frequency and intermediate frequency. With the increase in the inflow speed, the broadband noise increases gradually, and the frequency of tone noise gradually shifts to the high frequency band. Additionally, it is found that, for landing gear with four or six wheels, noise is very sensitive to angles of attack and wheel angles of attack. Consequently, the NSDL method shows significant potential in predicting the sound pressure level of landing gears and is expected to improve the efficiency of evaluation and optimization design for noise reduction of landing gear.