Intelligent fault diagnosis of high-speed train axle box bearings under parameter fluctuation working conditions using continuous cohomology-based meta-transfer learning framework

Due to the challenges associated with acquiring fault data from high-speed train axle box bearings operating under parameter fluctuation working conditions, the diagnostic accuracy of data-driven fault diagnosis methods often suffers from insufficient fault information. To address this issue, we propose an intelligent fault diagnosis strategy utilizing a novel continuous cohomology-based meta-transfer learning (CCMTL) framework. This approach aims to enhance fault identification accuracy by developing comprehensive feature representations despite limited data availability. Initially, acoustic emission data are transformed into a barcode chart using the continuous cohomology mechanism, from which fault topology features are derived. Subsequently, these features are input into a specially designed meta-transfer model for multi-task learning, thus forming a pre-trained model. This pre-training model undergoes further refinement both externally and internally to optimize its ability to adapt quickly to few-shot tasks. According to carry out the abovementioned steps, the proposed method can effectively overcome the challenges (e.g., data scarcity, difficulty in fault feature extraction, low robustness, and generalization ability) in the fault diagnosis of high-speed train axle box bearings under parameter fluctuation working conditions. Compared with existing fault diagnosis methods, the proposed CCMTL algorithm has three obvious advantages: (1) improve the fault diagnosis robustness of high-speed train axle box bearings, (2) reduce the cost of data annotation, and (3) adapt to a wide range of working conditions and parameter fluctuations. Experimental findings validate the effectiveness and feasibility of the proposed approach in scenarios requiring few-shot fault diagnosis. Specifically, when compared with current advanced diagnostic methods, the CCMTL method demonstrates superior diagnostic accuracy for fault diagnosis in high-speed train axle box bearings under parameter fluctuation working conditions. In conclusion, the practicality and generalization ability of the proposed CCMTL-based fault diagnosis approach are thoroughly demonstrated through comprehensive ablation studies. On the other hand, the designed CCMTL framework is not limited to specific machinery or operational conditions. With proper training and adjustments, the model can be applied to other industrial equipment with similar fault patterns and data characteristics. In summary, although this article emphasizes a specific application scenario (e.g., parameter fluctuation working conditions), the proposed method possesses broad applicability and can be extended to other similar industrial fault diagnosis tasks.