Aero-engine Fault Diagnosis Based On Relevance Vector Machine

Aero-engine is the most significant part of aircraft, so it is called the "heart" ofplane. In order to protect the safety of passenger’s lives, improving efficiency in theuse of the engine, preventing serious economic losses and saving maintenance costs,the fault diagnosis is the effective way to make it operation stability. However, thestructure of the aircraft engine is becoming more and more complicated, the relationsbetween units are getting closer. This makes the state monitoring and fault diagnosisto Aero-engine become more difficult and people have to explore new theories andapproaches constantly to solve the problems.Compared with the traditional method, artificial intelligence has the irreplaceableadvantages to complex nonlinear systems. Michael E. Tipping advanced RelevanceVector Machine (RVM) theory based on Bayesian theory in2001. This method issparser than Support Vector Machine (SVM). Therefore, this paper focuses onresearch about the relevance vector machine application to the aero-engine faultdiagnosis. The main contents are as follows.First of all, it is a introduction of traditional intelligent diagnostic methods suchas neural networks, support vector machines and Least Squares Support VectorMachine(LSSVM). At the end of this part, the analysis of their advantages anddisadvantages has been given through the simulation and applications in theaero-engine fault diagnosis.Secondly, it introduces relevance vector machine theory. This part details theBayesian framework and associated learning algorithm for the RVM, and gives someillustrative examples of its application in the aero-engine gas path fault diagnosis andoil spectral data analysis. Compared with traditional method, RVM obtain fasterlearning speed and higher accuracy. The relevance vector machine is a Bayesiantreatment, its mathematics model hasn’t have regularization coefficient, and its kernelfunctions don’t need to satisfy Mercer’s condition. These make RVM has broadapplication prospects in the field of fault diagnosis.The third part is the research about optimization algorithm. It includes ParticleSwarm Optimization (PSO) algorithm and its improved algorithm which is QuantumParticle Swarm Optimization (QPSO) algorithm. Compared with PSO algorithm,QPSO has faster convergence, and less parameter number than PSO. Therefore QPSOis used to solve the selection problem about RVM’s hyper parameters. At the end of paper, QPSO-RVM algorithm is used to making predictions ofExhaust Gas Temperature(EGT) and Spectrometric oil analysis. Compared with othermethods, the simulation of QPSO-RVM application in aero-engine fault diagnosisshows it improves the accuracy of the relevance vector machine model.