Research On Fault Diagnosis And Life Prediction Technology Of Plunger Pump Based On Graph Neural Network

The plunger pump is the core power component of the switch machine,its running condition is directly related to the normal operation of the hydraulic system of the switch machine.Since the switch machine is exposed to the field operation all year round,the working environment of the plunger pump is characterized by high pressure,high load and large flow rate,etc.,degradation phenomenon and various faults are easy to occur,which will reduce the working efficiency of the plunger pump in light cases,and cause a series of chain reactions in heavy cases,affecting the safe operation of the railway.Therefore,it is of great practical significance to accurately diagnose the fault of the piston pump of the switch machine,monitor its running state and predict its remaining service life.In this thesis,the fault diagnosis algorithm and the remaining service life prediction algorithm of the piston pump are studied.The main research contents include:(1)By analyzing the characteristics of vibration acceleration signals of the piston pump of the switch machine,a fault diagnosis model of the piston pump was designed based on the improved integrated adaptive noise complete integrated empirical mode decomposition algorithm(ICEEMDAN)and the graph convolutional network algorithm based on initial residual and identity mapping(GCNII).The model decomposed the vibration acceleration signal data of the piston pump into multiple dimension signal data by ICEEMDAN method.The time domain and frequency domain characteristics of each dimension signal were calculated respectively and then spliced into feature vectors.In order to facilitate the application of GCNII algorithm,the model uses KNN algorithm to calculate the similarity between the signal feature vectors of each dimension,and generates a fault feature graph according to the similarity,which realizes the transformation from one-dimensional time sequence data to graph structure data,and describes the relationship between different fault feature samples.As the input of GCNII classification model,the fault characteristic diagram can more accurately identify the fault type of the switch machine plunger pump.The diagnosis rate and robustness of the fault diagnosis model are verified by experiments.(2)The remaining service life prediction algorithm of the switch machine plunger pump was studied,and a prediction model of the remaining service life of the plunger pump was proposed,which can be applied to the multi-sensor sampling condition by combining multi-scale parallel one-dimensional convolutional neural network(MP1DCNN)with Bi LSTM and GAT.The model uses MP1DCNN-Bi LSTM module to convert the data of multi-channel one-dimensional vibration signals of the plunger pump into multiple health index sequence to characterize the degradation trend of the plunger pump.Considering the characteristics of the plunger pump in the process of degradation,a K-order lower triangular adjacency matrix representation method was designed to map the causal relationship between the multiple health indicators sequence and convert it into the multiple health indicators degradation feature map.The residual service life of the switch machine plunger pump was predicted by using the GAT module to aggregate the multiple degradation information contained in the feature map.The validity and accuracy of the proposed method are verified by the measured life data of the switch machine plunger pump and the comparison experiment.(3)Based on the fault diagnosis algorithm and the remaining service life prediction algorithm studied in this paper,the management software system of switch machine plunger pump is developed.The system integrates the two algorithms into two modules of fault diagnosis and life prediction.The measured data verify that the requirements of fault diagnosis and life prediction can be fulfilled.