Simulation Study On Typical Fault Diagnosis Of Mechanical Fuel Pump Of A Certain Aeroengine

As an important part of fuel control system,aero-engine fuel pump’s health directly affects the reliability and service life of the engine.When the fuel pump fails and can not be maintained immediately and accurately,it is likely to affect the normal operation of the whole fuel system and even the aircraft sately.At present,in the process of aero-engine fault diagnosis in China,the method of artificial experience diagnosis is used more often,that is,technicians rely on rich practical experience and certain theoretical knowledge,with the help of certain simple tools,using methods such as watching,listening,and touching to check and locate the fault,analyze the causes of faults,and then put forward troubleshooting measures.However,a large amount of time is often spent on the diagnosis and maintenance due to the uneven level of personnel technology and immature diagnostic technology,which largely affects the safety and economic value of aircraft use.Therefore,it is necessary to carry out some researchs on improving diagnosis methods to make it more scientific and efficient,and in addition,these researchs are of relatively highly engineering application value.To obtain the fault model of the fuel pump,after the analysis of the working principle of the fuel pump and the variation law of each parameter,the paper also uses the AMESim software and the Simulink module under the Matlab platform to establish the digital twin body of the fuel pump.And on the basis of this method,the simulation model and fault data of this type of fuel pump in typical fault mode are obtained by adding faults to the simulation model.The practice shows that the fitting degree of the fuel pump simulation model established in this paper meets the requirements of the working parameter variation curve output and the data recorded by the FDR under normal working mode.Moreover,the characteristics of the working parameter variation curve under fault mode obtained by fault injection method are basically consistent with the theoretical analysis results.This method provides a good way to solve the problem of diagnosis with little or no fault data.In the paper,Artificial Intelligence is used to establish a fuel pump fault diagnosis model,and diagnostic parameters are selected and normalized.And two typical faults and fault combination of one-way valve jam and membrane box leakage are diagnosed.The results show that the fault diagnosis models based on artificial intelligence such as random forest,support vector machine and neural network have good adaptability and are suitable for the fault diagnosis of fuel pump studied in this paper when the fault samples are small and difficult to obtain.Moreover,compared with the diagnosis results of the three fault diagnosis models,the diagnosis model based on LSTM deep neural network has higher diagnostic accuracy.