Research On Fault Diagnosis Method Of Hydraulic System Based On Deep Learning

Because hydraulic system has the characteristics of large capacity,smooth rotation and high output power,it is widely used in mechanical equipment.As the core of the hydraulic system,the reliability and comprehensive performance of hydraulic system power source is the key factors to ensure the stability and efficiency of a hydraulic system.Once the power source of the hydraulic system fails,the whole hydraulic system will be paralyzed,thus the whole hydraulic system will be paralyzed.It will cause the whole mechanical equipment to fail or stop.Hydraulic oil is not only the working medium of transmitting power in the power source of the hydraulic system,but also the lubricant of various hydraulic components.Its performance status will directly affect the reliability,stability,system efficiency and past life of the system.After the hydraulic oil is polluted,the change of oil viscosity will affect the load of the hydraulic system,which will change the working pressure of the hydraulic system,and then affect the current signal of the motor.In addition,the contamination of hydraulic oil will lead to improper sealing at the joint of the suction pipeline of gear pump,which will cause cavitation in the suction air of the hydraulic pump,thus affecting the working stability and service life of the hydraulic pump.Therefore,from the point of view of fault diagnosis,this paper focuses on the failure of the hydraulic system caused by the damage of gear pump and motor under the condition of hydraulic oil pollution.Then the collected data samples of gear pump-motor unit operation are studied by using a BP neural network,single hidden layer and multiple hidden layer MLP model,convolution neural network and other methods.Firstly,it can independently complete the control and data acquisition of the experimental platform on the self-built hydraulic experimental platform,and design VIof the electromagnetic proportional control module and data acquisition module in LabVIEW software,then complete the acquisition and storage of the running data samples of gear pump-motor unit.Secondly,a three-layer BP neural network model is constructed based on the classification characteristics of BP neural network.The effect of fault classification,fault verification and fault diagnosis is successfully realized through the operation of feature extraction and normalization of data samples.According to the diagnosis results,this method of fault classification can achieve the purpose of fault diagnosis.Then,aiming at the shortcomings of BP neural network,such as difficult selection of feature parameters,complex network structure and easy to fall into local optimum,a fault diagnosis method based on multi-hidden layer MLP model is proposed.This method can directly input all the extracted eigenvalues into the MLP model for training,and then make the network learn more deeply,so as to ensure that the output of the model is reliable and the accuracy of fault diagnosis is improved.Finally,in order to make the convolution neural network recognize one-dimensional data samples,this paper adopts a fault diagnosis method based on one-dimensional convolution neural network.Unlike the previous two chapters,the data samples of the algorithm do not need to be extracted artificially,but can be extracted automatically through the training of the neural network,thus realizing the self-learning process of the neural network.The diagnostic results show that the method can still achieve high accuracy in the training data sample set,which verifies the feasibility of using a convolutional neural network in power source fault diagnosis of hydraulic system.