A generative-transfer learning framework for intelligent fault diagnosis of bearings with incomplete vibration signals under strong operational shift

Abstract Under prolonged, complex operating conditions, bearings’ incipient fault signals are weak and heavily masked by noise, threatening safe operation and maintenance. To address the three major challenges in real vibration data—numerous outliers, large variations in operating conditions, and scarce labeled samples—this study proposes a fault-diagnosis framework that combines deep generative models with transfer learning. During data preprocessing, outliers are detected using the interquartile range method. A conditional variational autoencoder (CVAE) was created by combining a bidirectional long short-term memory network and a transposed convolutional network to replace outliers and enhance data. Subsequently, multidimensional fault features were extracted from three domains: the time domain, frequency domain, and time–frequency domain. Building upon this foundation, transfer learning methods were introduced. By employing feature alignment based on physical semantic mapping and Z -score distribution alignment techniques, the approach effectively mitigates semantic feature discrepancies and distribution shifts between the source and target domains. Experiments demonstrate that the gradient-boosted classifier achieves a diagnostic accuracy of 79.6% in the test set. Furthermore, the target domain confidence score is higher than that of mainstream models such as domain-adversarial neural network and multi‐kernel maximum mean discrepancy. In addition, ablation experiments have verified the superiority and rationality of the CVAE model architecture and the proposed transfer method. Finally, an analytical framework covering pre-, during, and post-event phases was established, enhancing the transparency and credibility of the intelligent diagnostic model. The research findings provide a reliable theoretical basis and engineering solutions for the intelligent operation and maintenance of high-speed train bearings.