Abstract:  Cavitation phenomenon seriously restricted the development of axial piston pump to the direction of high speed and high pressure. It was necessary to study the cavitation state detection and intelligent fault diagnosis of piston pump.Therefore, combining the advantages of deep learning network and nonlinear classifier, a cavitation state recognition method (detection model) of plunger pump based on CBLRE (CNN + BiLSTM + RELM) model was proposed. Firstly, the one-dimensional original vibration signals under different cavitation states of the piston pump were enhanced and standardized. Then, the convolutional neural networks(CNN) network was used to automatically extract the features from shallow to abstract signals and carry out feature dimension reduction. Bidirectional long short-term memory(Bi-LSTM) network was used to learn the time dependence of feature sequences. Regularized extreme learning machine(RELM)nonlinear classifier was used to classify,and the cavitation state detection and intelligent fault diagnosis of piston pump were realized. Finally, in order to test the performance of CBLRE model, an experimental platform was built. The CBLRE model was compared with other models,and its performance under different working conditions was also compared. The experimental results show that the model proposed has stable structure, short training time,and good generalization performance under different loads. The recognition rate of cavitation state can reach more than 99%.The results verify the effectiveness of the cavitation state identification method of piston pump. In addition, cavitation phenomenon and other faults of piston pump can be identified effectively.

Key words:  positive dispcacement pump; axial piston pump;cavitation phenomenon;convolutional neural networks(CNN);bi-directional long short-term memory (Bi-LSTM) network;regularized extreme learning machine (RELM);deep learning network;nonlinear classifier