To address the challenge of efficiently evaluating vehicle aeroacoustic
characteristics during the preliminary modeling stages, this article proposes an
interior wind noise prediction method based on styling features and the
squeeze-and-excitation module and Xception network. Drawing upon wind tunnel
test data from sport utility vehicles of various dimensions operating under a
120 km/h condition, key styling parameters were identified based on vehicle
geometry. Furthermore, to overcome the limitations associated with small sample
sizes, a Wasserstein generative adversarial network is introduced to augment the
original wind noise dataset. Subsequently, a complex nonlinear correlation is
established between the acquired geometric parameters and the interior wind
noise level. Validation results demonstrate that the proposed model achieved
average MAPE values of 3.18%, 3.75%, and 3.26%, along with average RMSE values
of 1.12 dB, 1.20 dB, and 1.19 dB, respectively, across three independent
validation vehicles, indicating its promising efficiency and accuracy for
vehicle wind noise evaluation. This research provides a streamlined and robust
methodology to expedite the evaluation of interior wind noise, offering a
practical technical reference for aerodynamic noise control during the
preliminary styling stages.
