Study On Monitoring Method Of The Hydraulic Pump Health Condition Based On Deep Learning

The hydraulic transmission has the advantages of high motion accuracy,fast response speed,large power-to-mass ratio and wide speed range.Therefore,it is widely used in many important fields such as plastic machinery,engineering machinery,metallurgical industry,aerospace,machine tools,ships,etc.The hydraulic system plays a core control or transmission role in various equipment.The hydraulic pump is an important part of the hydraulic system and is the power source of the entire hydraulic transmission system.Its working performance directly affects the working state and working efficiency of the entire hydraulic system.Therefore,it is necessary to perform real-time condition monitoring and fast and accurate fault diagnosis of the hydraulic pump.In recent years,the research on deep learning is getting deeper and deeper,Convolutional neural network(CNN)is an important deep learning algorithm,which is widely used in the field of image processing and has a good ability to recognize and classify images.Therefore,this paper proposes to use the CNN to classify and identify the time-frequency image of the vibration signal of the hydraulic pump,so as to monitor the health of the hydraulic pump.The parameters of CNN have great influence on image recognition.In this paper,the influence of layer number,iteration number,batch size,number and size of convolution kernel on the image recognition results is studied,and the parameters of the best parameters of CNN for the identification of the health state of hydraulic pump are found out by the final recognition accuracy.The vibration signals under different fault states of the axis piston pump and different volumetric efficiency of the hydraulic gear pump are collected through fault diagnosis experiments and the accelerated life test bench.Then the vibration signals are analyzed and processed by three time-frequency methods,namely short-time Fourier transform,wavelet transform and Wigner-Ville.Three time-frequency images are obtained.According to the recognition results of the images generated by three time-frequency variation methods,the optimal time-frequency transformation method is determined.The images are divided into training set and test set,and are input into the convolutional neural network for parameter tuning for fault diagnosis and health status monitoring.Through many times of training,the optimal parameters are determined,and the fault of the axial piston hydraulic pump and the health status of the gear pump are identified.Good results are obtained.