Research On Remaining Service Life Prediction Of External Gear Pump Based On Graph Neural Network

The external gear pump has the advantages of simple and compact structure,strong self-suction and large speed range,which is widely used in the field of construction machinery.In order to ensure that the gear pump can work in the normal state,it is necessary to predict the life of the gear pump.The traditional life prediction method is full cycle life test,which is lack of timeliness and waste of resources.Therefore,this paper uses accelerated life test method to predict the remaining service life of gear pump.Firstly,the wear mechanism and failure form of gear pump are explored,and the test method is determined as step accelerated stress test.According to the standard of "accelerated life test method for products GB / T 34986-2017",the accelerated life test bench is designed and built,and the basic data of gear pump is collected based on Lab VIEW.Secondly,the vibration signal with rich life information is modulated and demodulated.Based on the improved variational mode decomposition(VMD)method and particle swarm optimization algorithm,the original vibration signal collected by the test-bed is denoised,and the original signal is restored to the greatest extent.Compared with the traditional VMD and empirical mode decomposition(EMD)algorithm,the improved VMD method is proved to be superior in life prediction.Then,the characteristic parameters of accelerated test data are extracted from time domain,frequency domain and time-frequency domain,and the influence of each characteristic parameter on the degradation trend of gear pump is analyzed and calculated.The characteristic parameters that can characterize the remaining life of gear pump are selected for the life prediction of gear pump.Based on the graph neural network(GNN)to train the characteristic parameters,the graph neural network itself contains rich characteristics,through the training of neural network parameters,the establishment of gear pump degradation evaluation model,and compared with fuzzy neural network,Bayesian network,proved the superiority of graph neural network.Finally,the Weibull distribution two parameter estimation model is constructed with the sum of square error as the objective function,and the samples are expanded based on Monte Carlo simulation to realize the reliability evaluation of external gear pump.In this paper,the life prediction method of external gear pump is deeply explored,and the reliability evaluation of FBW-F304 external gear pump is completed,which provides a new idea for the life prediction of external gear pump.