Speed Adaptive Graph Convolutional Network for Wheelset-Bearing System Fault Diagnosis Under Time-Varying Rotation Speed Conditions

PurposeThe wheelset-bearing system is one of the key parts of the high-speed train. Fault diagnosis is the guarantee of its safe operation. But its vibration signal contains a lot of noise and is not stable. Existing intelligent algorithms are difficult to extract fault features of it.MethodsTo solve this problem, a speed-adaptive graph convolutional network (SAGCN) for fault diagnosis of the wheelset-bearing system is proposed. The model can fuse three types of information. First, the deep convolutional layer is used to extract the features of vibration signals and speed signals to improve speed adaptability. Then, feature compression and connection are carried out by global average pooling (GAP). The graph construction layer (GCL) is proposed to extract the spatial topology. Finally, the obtained three types of fault information are input into GCN for data fusion and fault classification to improve the noise robustness of the model.ResultsThe time-varying rotation speed samples are obtained from the test rig. Compared with the five methods, the proposed model can achieve the highest diagnostic accuracy, with an average accuracy of 98.83%. Two sample sets with different SNRs are constructed for experiments, and the average accuracy of the proposed model can still reach 97.54% and 96.21%.ConclusionThe experimental results show that the proposed SAGCN has good speed adaptability and noise robustness. So this study can be beneficial to the fault diagnosis of wheelset-bearing systems under time-varying speed conditions.