Abstract:  Axial piston pump was an important part of the hydraulic power system. Due to the serious hazards caused by the failure of the axial piston pump, it was necessary to carry out fault diagnosis of the axial piston pump. However, a large number of engineering practices was showed that the axial piston pump often presents composite faults in different forms at different parts at the same time. Because the multi-component coupling modulation characteristics and characteristic parameters of the composite fault data of the axial piston pump were difficult to determine, a compound fault diagnosis method for the axial piston pump based on Gramian angular difference field - deep residual network (GADF-ResNet) was proposed. Firstly, the original vibration signal of the axial piston pump was converted into a two-dimensional array by Gramian angular difference field (GADF), and the array was stored in the form of a grayscale image to obtain feature samples, which were divided into training sets and test sets, and marked in the form of multiple labels. Then, it was inputted the samples into deep residual network (ResNet)and determined the best network structure and parameters through forward propagation and back propagation. Finally, feasibility and robustness of the model was verified through test sets and experiments. The experimental results show that the compound fault identification accuracy of the axial piston pump based on GADFResnet can reach more than 87%. The results show that this method can effectively identify the compound faults of axial piston pump.
Key words:  hydraulic transmission system; positive displacement pump; compound fault; Gramian angular difference field (GADF); deep residual network (ResNet); multilabel