Abstract To address the low fault diagnosis accuracy of spindle bearings in vehicle transmission systems under imbalanced data, a fault diagnosis method based on Double-Discriminator Generative Adversarial Network (DD-GAN) and Convolutional Kolmogorov-Arnold Network (Conv-KAN) is proposed. DD-GAN consists of two discriminators, a generator, and a loss aggregation mechanism. The discriminators have distinct architectures, with one incorporating an attention mechanism to enhance local feature discrimination. The loss aggregation mechanism balances the outcomes of both discriminators, preventing the generator’s optimization from being dominated by one. Conv-KAN, based on Kolmogorov-Arnold Network (KAN), captures complex, nonlinear relationships in high-dimensional data. By adapting KAN to simulate convolution-like operations, Conv-KAN enhances the extraction of spatial features from input data, improving fault diagnosis accuracy. The method’s workflow involves converting the original bearing vibration signals into time-frequency images, augmenting the imbalanced data using DD-GAN to normal levels, and then training the Conv-KAN model on the augmented dataset for fault diagnosis. Experiments show that DD-GAN generates high-quality bearing signal samples, with an FID value 69% lower than WGAN-GP. When the imbalance ratio is 1:100, the proposed method achieves an F1 score 26.09% higher than WGAN-GP with CNN. At a ratio of 1:200, the F1 score improves by 31.44%. These experimental results comprehensively validate the superiority of the fault diagnosis method based on DD-GAN and Conv-KAN, demonstrating its effectiveness in addressing the issue of low fault diagnosis accuracy in vehicle transmission system spindle bearing under imbalanced data conditions.
Hao et al. (Wed,) studied this question.