Research On Typical Fault Diagnosis Method Of Diesel Engine Fuel Supply System Based On GCN

In the actual production,the operating state of the diesel engine has a huge impact on the safety and reliability of the entire mechanical system,and it is very necessary to perform real-time non-disintegration fault diagnosis for the diesel engine.In recent years,fault diagnosis methods based on deep learning have received widespread attention due to their ability to automatically extract features.But when the number of labeled training samples is small,most existing models are prone to overfitting.In addition,the operating conditions of diesel engines are complex and changeable,and the current research is mainly focused on fault diagnosis under single operating conditions.To solve these problems,the paper selects the fuel supply system with a higher failure rate in diesel engines to carry out failure simulation experiments,builds "end-to-end" fault diagnosis models based on Graph Convolutional Network(GCN),and verifies the diagnosis performance of the methods by using the measured vibration data sets,the details are as follows:(1)In view of the small number of labeled training samples,GCN is introduced into the field of diesel engine fault diagnosis,and a fault diagnosis method based on KNNG(K-nearest neighbor graph)-GCN is established.Firstly,the edge connection between samples is established by calculating the cosine similarity to realize the conversion from vibration data to graph data.Secondly,the double headed weight matrix mechanism and one-dimensional maximum pool layer are used to optimize the model structure of GCN to further suppress the possible over fitting phenomenon of the model.Finally,when the label ratio is only 5%,the diagnostic accuracy of the selected data set can reach 98.95%.(2)In order to improve the diagnostic efficiency of the model,an adjacency matrix construction method adapted to the vibration signal of multiple measuring points of the diesel engine is established,and a fault diagnosis method based on MC(multi-channel)-GCN is obtained.Firstly,select the same time interval from the signals collected by the three measuring points to establish the edge connection,and calculate the unified weight based on the training sample.Secondly,the graph maximum pooling layer is introduced to perform graph-level pooling to achieve effective integration of multichannel features.The analysis results of the selected data sets show that this method is superior to the comparison method in diagnosis accuracy,generalization performance and diagnosis efficiency.(3)For samples under various working conditions,the two graph convolution layers in MC-GCN model are used as common feature extractor and exclusive feature extractor respectively to build a multi condition fault diagnosis model,and a method based on MMD(Maximum Mean Discrepancy)-MC-GCN is established.Among them,MMD is used as a part of the objective function for back propagation to realize the feature alignment of samples under different working conditions in the common feature space.Applied to the selected data set,this method improves the diagnosis accuracy under full load 750r/min and full load 1000r/min by 11.13% and 7.59%.The comprehensive performance has obvious advantages over the single condition fault diagnosis method.In conclusion,the fault diagnosis method for diesel fuel supply system established in this paper can accurately diagnose the weak power fault signal of the system with a small number of training samples.This method can also be extended to early fault early warning of diesel engine or other mechanical systems.