Equipment System-level Intelligent Fault Diagnosis Method Based on Data Fusion and Enhanced Graph Convolution

The existing component-level fault diagnosis methods fail to capture the fault coupling relationships between multiple components, leading to feature confusion and misjudgment in system-level fault diagnosis. This paper proposes a system-level fault diagnosis method for equipment based on data fusion and enhanced graph convolution. First, the information entropy values of each channel signal are calculated, and the multi-channel data-level information fusion is realized using an captive weight allocation strategy. Second, an independent convolutional neural network (CNN) is constructed for each component to obtain node attribute features. Then, a spatial topological relationship graph that reflects the real physical structure of the equipment system level is constructed as a priori knowledge, and an enhanced graph convolution (EGC) kernel is designed to model the fault coupling relationship between components. Finally, the system-level association features of each component are input into the diagnostic classifiers of each component to achieve system-level fault diagnosis. Experiments using a train bogie transmission system dataset for verification show that the proposed method achieves 100% diagnostic accuracy for all four components of the driving motor, gearbox, left axlebox, and right axlebox. It outperforms other fault diagnosis methods, verifying the superior performance of the proposed method to realize system-level fault diagnosis of equipment.