Deep learning-based wind noise prediction study for automotive clay model

Analyzing and mitigating wind noise in automobiles is a significant issue within the realm of noise, vibration, and harshness. Due to the intricate nature of aeroacoustic generation mechanisms, current conventional methods for wind noise prediction exhibit limitations. Hence, deep learning methods are introduced to investigate wind noise in the side window area of an automotive clay model. During aeroacoustic wind tunnel experiments, side window vibration data and noise data from the driver were collected at experimental wind speeds of 100 km h-1, 120 km h-1, and 140 km h-1, respectively. These data samples were obtained to be used for training and validation of the wind noise model. Convolutional neural network (CNN), residual neural network (ResNet) and long short-term memory neural network (LSTM) algorithms were separately employed to reveal the complex nonlinear relationship between wind noise and its influencing factors, leading to the establishment of a wind noise prediction model. Simultaneously, these deep learning methods were compared with backpropagation neural network (BPNN), extreme learning machine (ELM), and support vector regression (SVR) methods. Conclusion indicates that the LSTM wind noise prediction model not merely exhibits higher accuracy, but furthermore demonstrates superior generalization capabilities, thereby substantiating the superiority of this method.