Fault Feature Extraction Study For Aero Engine Intershaft Bearing

The feature extraction and analysis of aero-engine fault are mainly used for fault identification and diagnosis,especially for the mechanical status and fault of rotor system.It is generally believed that the vibration information contains plenty of mechanical status information(including amplitude,frequency,phase,etc.),which can reflact the mechanical status of the structural system best.As the core parts of aero-engine rotor support system,intershaft bearings' working condition is directly related to the safety of engine.Accurate and effective extraction of fault features of intershaft bearings is significant for bearing fault type,degree and trend analysis,and is also the technical prerequisite for engine condition monitoring and fault diagnosis.The transmission path of fault feature information of intershaft bearing is complex and there are many interference sources,while the actual operation of the engine is more complex,which causes the fault feature information to be submerged in a large amount of noises.To solve this problem,Morlet complex wavelet resonance demodulation method is applied to extract fault features of intershaft bearings.Fast kurtosis image and Crest factor of envelope spectrum(Ec)are used to analyze fault data of intershaft bearings,and some results are achieved in extracting fault features from data.The evaluation index is optimized based on the spectrum characteristics of the inner and outer ring faults of the intershaft bearing.This method can extract the fault features more effectively under various working conditions.To solve the problem that the envelope spectrum of rolling element fault can not be effectively correlated with roller fault characteristic frequency by resonance demodulation of Morlet.A fault feature extraction method based on multi-point Morlet entropy and Locally Linear Embedding(LLE)is proposed in this paper.Combining with the method of resonance demodulation,the fusion information entropy of multi-point acceleration data is extracted.LLE is used feature-extraction and dimension-reduction.Finally,support vector machine(SVM)is used to test the effect of fault feature extraction.It is verified that the method can greatly improve the accuracy of fault ifentification based on the experimental data.