Application Of SVM In Aero-engine Gas Path Fault Diagnosis

With the rapid development of modern aerospace industry, theperformance of the aircraft mostly is determined by the performance ofthe aero-engine. As a core component of the aircraft–aero-engines, theirwork status, and whether the failure case occur, has a critical influence onthe stability and security of the flight. Therefore, regular aviation engineoverhaul and maintenance for aero-engine is very necessary. Theaero-engine fault diagnosis has a very important role on reducingmaintenance cost and improving efficiency of repairing. Aero-enginefault is generally divided into gas path fault, annex fault and rotor fault.According to incomplete statistics,90%of the engine failure is the enginegas path fault, and gas path fault is the most difficult to predict. Sotroubleshooting for the aero-engine gas path components has a veryimportant practical significance.Traditional method of fault diagnosis algorithm is based on Linearmodel. Modeling accuracy is difficult to guarantee with the complexityof the continuous improvement of the engine size. While the neuralnetwork method needs a large number of data samples for training, andeasy to fall into local optimum. The method of data drive doesn’t need theestablishment of mathematical model. Use existing data, without the needfor a large number of samples, and extract data feature and design faultclassifier to complete fault diagnosis problem.In this paper, substantive of aero-engine gas path component faultdiagnosis is to classify the status parameters measured which can reflectperformance of aero-engine. Accuracy of the sensor measurement has acritical influence on the accuracy of the engine gas path components faultdiagnosis. Therefore, this paper also considered the sensor as a part ofaero-engine, and also do research on fault diagnosis on sensor. To solvethe aero-engine sensor fault diagnosis problem, a new approach on sensor fault diagnosis based on time-frequency analysis and the multi-kernelsupport vector machines is proposed. The result of sensor fault diagnosisexperiment shows that this approach can be used in fault diagnosis ofvarious types of sensors more effectively.In this paper, use the existing literature data to do research on faultdiagnosis of gas path components of the double rotor turbojet engines. Toimprove the accuracy of fault diagnosis, we use new machine learningmethod for diagnosis of aero-engine gas path component fault. And do alot of experiments and comparative analysis. Eventually I use a newAdaboost+SVM(Support Vector Machine) algorithm to troubleshootengine gas path components fault. And apply it to a single, dual gas pathcomponent fault diagnosis. And add a certain disturbance to verify thevalidity of the method.