Abstract: In view of the limited fault characterization ability of traditional shallow models, and the extraction of signal features relying too much on expert experience, a fault diagnosis method for axial piston pump based on deep one-dimension convolution neural network (D-1DCNN) was proposed. Firstly, the vibration signals of piston pump under five states: normal, loose shoe, shoe wear, central spring failure and valve plate wear were collected. The signals were made into a sample set and got labeled. The sample set was divided into training samples and test samples. Then the samples were input into D-1DCNN for feature extraction of training sample signals. The D-1DCNN specific model was obtained by forward propagation and back propagation, the SoftMax classifier was used to classify the test samples, and the parameters in the network model were adjusted to obtain the accuracy of piston pump fault diagnosis. Finally, the simulation comparison of the bearing fault signals of Western Reserve University was carried out. The research results show that the method can achieve 100% diagnostic accuracy for axial piston pump fault diagnosis, using D-1DCNN for fault diagnosis can realize intelligent diagnosis of piston pump without manual design or extraction of features; the method also has good fault diagnosis effect for different objects, and has a certain universality.

Key words:  axial piston pump; fault diagnosis; deep one-dimension convolution neural network (D-1DCNN); deep learning;SoftMax

XU Chang-ling, HUANG Jia-hai, LAN Yuan, et al. Fault diagnosis of axial piston pump based on D-1DCNN［J］.Journal of Mechanical & Electrical Engineering, 2021,38(11):1494-1500.