A Fault Diagnosis Method for Aerospace Bearings Based on CS-WGAN-PSO

To address the class imbalance issue in aero-engine bearing fault diagnosis, this paper proposes a novel method named cost-sensitive Wasserstein generative adversarial network with particle swarm optimization (CS-WGAN-PSO). The conditional Wasserstein generative adversarial network with gradient penalty (WGAN-GP) is optimized by the particle swarm optimization (PSO) algorithm to enhance the quality and diversity of synthetic samples. A dynamic focal loss function is introduced to improve the learning capability for minority classes. Finally, an attention-based convolutional neural network (CNN) is employed to achieve accurate fault classification. Experiments conducted on the HIT dataset demonstrate that the proposed CS-WGAN-PSO achieves superior diagnostic performance under various imbalance ratios. Notably, it maintains over 90% accuracy under an extreme 100:1 imbalance, significantly outperforming WGAN-GP and importance weighted autoencoder (IWAE), indicating strong potential for engineering applications.