Fault Diagnosis Of Axial Piston Pump Based On Convolution Neural Network

Hydraulic systems are widely used in the field of construction machinery because of their advantages of high power,high precision and fast response.However,once the failure occurs,it will affect the enterprise benefit,or cause casualties.As the core component of the hydraulic system,the performance of the hydraulic pump will directly affect the operation of the whole hydraulic system.Axial piston pump is widely used in hydraulic system because of its compact structure,long service life and high volumetric efficiency.With the development of science and technology,the structure of axial piston pump is more and more complex,and the fault forms are more and more complex.Therefore,it is necessary to diagnose the fault of axial piston pump,so as to ensure the normal operation of the hydraulic system.Typical faults of axial piston pump include shoe wear,loose shoe,central spring failure and valve plate wear.Because the fault of piston pump is sudden and complex,it is very difficult to use the traditional method of feature extraction and classifier to diagnose the fault of piston pump.The difficulty of fault feature extraction,the limited generalization ability of shallow model and the limited applicability in complex working conditions make the traditional methods have some limitations in piston pump fault diagnosis.In view of the above problems,this paper uses convolution neural network to diagnose axial piston pump.This paper mainly studies the following aspects:(1)Firstly,the structure and working principle of the piston pump are analyzed,and the mechanism of each typical fault of the piston pump and the serious consequences of the fault are analyzed.Then,the characteristic frequency of each working state of the piston pump is analyzed theoretically,and the vibration signal is transformed by Fast Fourier Transform,and the characteristic frequency of the signal is observed from the amplitude frequency diagram.(2)In view of the strong image classification ability of convolutional neural network,this paper transforms the vibration signal of piston pump into the image containing signal features and inputs it into convolutional neural network for fault diagnosis.Because the fault signal of piston pump is non-stationary and has time-varying characteristics,the joint analysis from the time-frequency plane is a powerful means to deal with non-stationary signal,so this paper uses the method of time-frequency analysis to transform the fault signal into time-frequency diagram.Because different time-frequency analysis methods have different sensitivity to fault characteristics of axial piston pump,this paper uses three time-frequency analysis methods to select the most suitable time-frequency analysis method for fault diagnosis of axial piston pump.(3)It is still necessary to extract features manually when using two-dimensional convolution neural network for fault diagnosis,therefore,in this paper,one dimensional convolutional neural network(1DCNN)is used,which does not need to extract data features in advance,and the vibration signal of piston pump can be directly input into the model for fault diagnosis.To solve the problems of traditional 1DCNN,such as insufficient depth and incomplete feature extraction,this paper uses deep one dimensional convolutional neural network(D-1DCNN).D-1DCNN increases the number of convolution layers on the traditional1 DCNN,which makes the signal feature extraction more complete.(4)The fault signal of piston pump is one-dimensional time sequence signal.In order to solve the problem that the feature of fault signal extracted by 1DCNN lacks time dependence,this paper uses the model of 1DCNN combined with long short term memory(LSTM),namely1DCNN-LSTM.Firstly,the feature data is extracted by 1DCNN,and then the extracted feature data is used as the input of LSTM in the form of sequence to extract the time sequence feature.Finally,the fault diagnosis of piston pump is completed by classification.The fault diagnosis of axial piston pump is the key and difficult point in the field of construction machinery at present.This paper takes axial piston pump as the research object,adopts the method of combining theory with experiment,and uses convolution neural network for fault diagnosis.In this paper,several kinds of convolutional neural network models are analyzed and compared,and it is concluded that D-1DCNN is in the dominant position in the fault diagnosis of axial piston pump.