| description abstract | The fault diagnosis of slewing bearings is crucial for modern industry. However, operational constraints and high signal acquisition costs limit the number of available diagnostic samples, leading to decreased diagnostic accuracy. This study proposes a novel fault diagnosis method for slewing bearings based on audible sound signals, termed Time Generative Adversarial Network (Time GAN)–Tabular Prior-Data Fit Network (TabPFN). It is a hybrid approach that integrates the capabilities of Time GANs and the TabPFN. The method leverages the feature enhancement capabilities of Time GAN and the probabilistic modeling strengths of TabPFN to improve fault diagnosis accuracy. This method utilizes low-cost, easily obtainable audible sound signals as input. By employing Time GAN, the original data features are enhanced, generating new training samples. Subsequently, the TabPFN framework constructs a substantial amount of synthetic data with causal relationships, facilitating Bayesian inference. Experimental results demonstrate that the proposed method effectively identifies various fault types with small-sample sizes, achieving an accuracy of 96.5%, approximately 10% higher than existing algorithms. Furthermore, this method exhibits high diagnostic accuracy and strong generalization capabilities, making it a robust solution for slewing-bearing fault diagnosis. | |
