Prediction of vehicle front wall sound insulation performance using Wasserstein generative adversarial network and Res-InceptionNet

The acoustic package of the front wall system plays a critical role in automotive design, directly affecting in-vehicle noise quality and passenger comfort. However, accurately modeling its acoustic performance remains a significant challenge due to the inherent nonlinearity and complexity of sound absorption and insulation mechanisms. These challenges hinder precise performance prediction and optimization in engineering applications. While data-driven prediction methods have shown strong potential in addressing complex nonlinear problems, their effectiveness is often constrained by the limited availability of comprehensive sample data. To overcome this limitation, this study proposes a prediction method that integrates small-sample augmentation with a Res-InceptionNet model. The original small dataset is expanded using a Wasserstein generative adversarial network, and the enhanced dataset is then used to develop a prediction model leveraging the learning capabilities of Res-InceptionNet. This approach is applied to predict and analyze the transmission loss of the front wall system and its components. The proposed method achieves a prediction accuracy of 96%, with a root mean square error of 0.824, demonstrating its effectiveness and reliability in assessing the acoustic performance of the front wall system. Beyond its application to automotive acoustic package design and optimization, this research provides a methodological framework that can be extended to other vehicle noise prediction and analysis challenges.