Research On Fault Diagnosis Of Control Surface In Aircraft Based On Machine Learning Methods

Reliability of control system is an important guarantee for safe flight of aircraft,and the three main control surface systems – aileron,elevator and rudder,could directly influence the performance of control system,therefore,it is of great significance to conduct condition monitoring and fault diagnosis on aircraft control surface system.Taking advantage of machine learning technology,which driven by process data,this paper conducts deeply studies on fault diagnosis characterized by system signals and develops the intelligent fault diagnosis system.The main work of this thesis is as follows:Firstly,this thesis conducts fault analysis of aircraft control surface system,and introduces the common fault types of actuator and sensor,and chooses the rudder system as detailed object in fault diagnosis.Due to methods based on process data do not rely on precise mathematical model,and easy to realize fusion of intelligent algorithms,machine learning which is driven by data is selected to research on fault diagnosis.Secondly,due to response signal of system interfered by noise,this thesis improves the singular value decomposition method,and then proposes a novel method-singular value morphology filtering algorithm to pre-treat the signal from aircraft rudder system model.Simulation results indicate that the method can not only remove the noise components better,but also keep the features of original signal.Thirdly,because empirical mode decomposition leads to serious mode mixing in dealing with noisy signal,this thesis introduces the ensemble empirical mode decomposition.Further,to deal with end effect problem,a novel EPSVR-EEMD(Envelope Point Support Vector Regression – Ensemble Empirical Mode Decomposition)feature extraction method is presented in this research.Simulation indicates that the method can better extract signal components in real physics,and obtain the corresponding energy features.Finally,due to manual intervention needed after misdiagnosis with normal support vector machine,the model of support vector machine based on posterior probability is constructed.Meantime,to deal with parameters problem of diagnosis model,this thesis puts forwards a generalized diagnosis model,and utilizes particle swarm optimization algorithm to solve the problem;experiment results show that the fault types can be identified accuracy and ordered,which validates the efficiency of the proposed method.In the end,based on the presented work above,a fault diagnosis system with friendly graph interfaces is designed and developed for control surface system,which offers a good platform for applying the method proposed in this thesis to system fault diagnosis in practice.