Fault Diagnosis Of High Speed Electrically Driven Feed Pumps Based On Deep Learning

As a common mechanical industrial equipment,high-speed electrically driven feed pumps are widely used in various fields of national production,which has made a great contribution to industrial production.Whether its operation status is good or not directly affects the production status of modern industry.Therefore,it is necessary for users to know its running state at all times.Once the high-speed electrically driven feed pumps have abnormal,it will cause property loss to the factory,and even cause life impact to enterprise employees.Therefore,it is necessary to take measures to diagnose the fault.In recent years,the research of deep learning is increasingly in-depth,and the application field of deep learning is also expanding,from the initial use in voice image,to now widely used in the field of fault diagnosis.However,there are few fault diagnosis methods applied to pumps.Therefore,this paper proposes to use the stacked auto-encoder network,stacked denoising auto-encoder network and generative adversarial network in deep learning to diagnose the fault of high-speed electrically driven feed pumps.Firstly,the time-domain signals measured by power plants are used,and the stacked auto encoder network is used to classify and identify the vibration level faults of feed water pumps.Aiming at the problem that the fault data samples are less accumulated in the actual production process,this paper proposes to generate part of the fault data through the generative adversarial network to solve the problem of low accuracy of fault diagnosis caused by unbalanced sample categories.Then,in view of the actual monitoring,the vibration signals of the pump are collected from the bearing,so the vibration caused by the bearing fault is considered,so the bearing fault diagnosis is carried out.The bearing fault data of the bearing database of Western Reserve University is used to simulate the fault of the water pump bearing in the actual production.Considering the influence of external factors and noise on the measured data,a method of stacked denoising auto-coder network is proposed.At the same time,the influence of the number of hidden layer nodes,learning rate,noise figure and other parameters on the accuracy rate is studied,and the optimal network structure is determined.Finally,aiming at three common faults of high-speed electrically driven feed pumps on site,including rotor unbalance,rotor misalignment and foundation looseness.All kinds of fault data collected by sensors are extracted by empirical mode decomposition,and by generating data of generative adversarial network,the data are input into stacked denoising auto-encoder network model,so as to realize the judgment of fault type and achieve good results.