Fault Diagnosis Of Aero-engine Rotor System Based On EMD And SVM

As a kind of complex rotating machine system with rigorous technical requiements,aero-engine is consisted of a large number of components which are coupled with each other and the work processes are different.It works for long time In the bad external environmental conditions including temperature,pressure and so on.Therefore many kinds of fault states occur,and they will seriously influence the normal flight of aircraft.Rotor system is the core structural part of aero-engine,its working performance directly affect the operation safety and reliability of the whole system.In order to improve the safety performaty of aero-engine,it is urgently demanded to diagnose the fault types of rotor system timely and accurately.Abundant state information of rotor system is often hidden in the vibration signals.Admittedly,it is an effective means to make fault diagnosis through processiLg vibration signals and extracting fault features.In view of this situation,the research contents of this paper include:(1)Vibration signals of a certain type of turboprop rotor system are decomposed by ensemble empirical mode decomposition(EEMD)m,and energy values,singular values,multiscale permutation entropies and time domain non-dimensional features are respectively calculated.The algorithm of neighborhood rough set(NRS)is applied to reduce the attribute dimension of the feature set,and the fault states of turboprop rotor system can be classified by support vector machine(SVM),The experimental results indicate that the classification accuracy is 97.5%,and the effect is able to meet the practical engineering requirements.(2)Due to the higher rotating speed of micro turbojet,the non-linear and non-stationary trend of vibration signals is further enhanced.The EEMD method is not fit for processing the complex random signals.Thus a new fault diagnosis model which combines improved complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN),distance evaluation technique,genetic algorithm and SVM is established.The original feature set is consisted of multi feature domains and sensitive features are able to be selected.Firstly,the improved CEEMDAN method is used to deal with the vibration signals of rotor system.The energy values,singular values and envelope sample entropies of the first several intrinsic mode functions which include the main fault information are separately extracted.Then the sensitivity factors of fault features are evaluated through the distance evaluation technique and sensitive feature set can be selected.The genetic algorithm is utilized to optimize the parameter indexes of the SVM classifier.Ultimately,the fault modes of turbojet rotor system can be recognized through the trained SVM.The experimental results confirm that the established model can accurately recognize the fault states with different rotating speeds.The fault recognition accuracy reaches 99.4737%,and satisfactory results are obtained.(3)On the basis of the above two sections,the graphical user interfaces(GUI)of aero-engine rotor fault diagnosis system is developed by MATLAB.It improves the human-computer interaction and simplicity of the whole fault diagnosis process,and makes the operation more convenient and intuitive.