RV (Rotary Vector) reducers are widely used as high-efficiency and high-precision transmission devices

and intelligent diagnosis of them is essential to ensure the operation of equipment. However

its complex structure and time-varying working conditions make it a great challenge to diagnose its faults. Therefore

a fault diagnosis method of RV reducer based on multi-source graph structure mining is proposed. Firstly

a multi-source graph structure feature set by mining the non-linear correlation of three-axis data features is constructed; then a node-embedded dual convolutional graph neural network based on the graph classification task to improve the accuracy of fault diagnosis under time-varying operating conditions is constructed; finally

a homemade test bench is used to validate the method. The results show that the present method can effectively mine the graph features from the three-axis sensor data and has a great improvement in the diagnostic performance and accuracy.