Research On Fault Diagnosis Of Methanol Diesel Dual Fuel Engine Based On Application Of PCA-BP Neural Network

With continuous development of transportation industry,the number of automobiles has increased,the demand for fuel has increased,and alternative fuel engines have been steadily developed.The methanol/diesel dual-fuel engine has been continuously researched and explored by local and foreign scholars.The methanol/diesel dual-fuel engine has new type of failures due to the newly added methanol supply system,and there are many engine problems caused by the methanol system failure.In order to support the promotion of methanol/diesel dual-fuel engines,it is an important task to develop diagnostic software for methanol/diesel dual-fuel engines.The development of engine fault diagnosis software based on PCA-BP neural network requires three aspects:methanol/diesel dual-fuel engine bench test,engine fault diagnosis model establishment and fault diagnosis software development.The main tasks are as follows:（1）Methanol/diesel dual-fuel engine bench test.The methanol/diesel dual-fuel engine is modified based on the Changchai 4G33TC engine,and installed a methanol supply system to achieve methanol injection in the intake manifold.Beijing Huahai S1 control unit is selected as the main controller of the methanol injection system.A methanol/diesel dual-fuel engine bench was built for testing the function of methanol/diesel dual fuel engine.Test the methanol/diesel dual-fuel engine in regular state,single-cylinder methanol injection stop,blockage of the intake pipe,blockage of the methanol injector and leakage of the methanol fuel supply pipe,etc.Collect the data of methanol/diesel dual-fuel engine bench during normal operation and when different kinds of failures occur.Test data is provided for testing the established fault diagnosis model.（2）Research on fault diagnosis method of methanol/diesel dual-fuel engine.The operating parameter-fault data sample set is established based on the historical operating parameters and fault data obtained from the test of the methanol/diesel dual-fuel engine.Establish a sample set of operating parameters-fault data,and select a normalized processing method to process the sample data.First,use principal component analysis to simplify the input parameters,then build a neural network-based engine fault diagnosis model.Data obtained from the experiment are used to training the engine fault diagnosis model.（3）The overall development and verification of the fault diagnosis software was realized.Based on the data of the methanol/diesel dual-fuel engine operating parameters stored in the database established from the test,the good interface characteristics of LabVIEW and the powerful computing power of MATLAB are combined.The PCA subsystem and the BP neural network subsystem are connected in LabVIEW to design and develop PCA-BP network diagnostic software.The number of network training steps required to meet the accuracy requirements is 11,the training time is 0.622s,the output error is 9.4×10^-3,which shows the diagnosis accuracy is high.Use the trained PCA-BP neural network fault diagnosis model to diagnose the selected inlet blockage,single-cylinder methanol injection stop,and methanol injector blockage.The accuracy rates can reach 99.46%,97.31%,and 99.58%.The feasibility and accuracy of the engine fault diagnosis model are tested.The real-time display and fault diagnosis functions of the diagnostic software are tested,and the practicability and effectiveness of fault diagnosis have been well verified.