Fault Diagnosis Of Aero-engine Using Spectral Clustering Algorithm

As the power of the flight system is mainly provided by the aero-engine,which plays an important role in the safe flight of the aircraft,it is of great significance to ensure the smooth running of aero-engine to flight safety.Aero-engine fault diagnosis based on data fusion faces two problems: one is that the fault diagnosis model based on supervised learning cannot be fully and effectively trained because of some factors including incomplete fault labels caused by the variety of the aero-engine faults and the complexity of fault data;the other is because of the working conditions of aero-engine is extreme,the aero-engine parameters measured by airborne sensors are greatly affected by noise.The main research object of this paper is part of parameters that recorded by QAR(Quick Access Recorder).As the aero-engine fault data is complex,fault diagnosis model based on supervised learning cannot learn all fault types.This paper puts forward an aero-engine data processing method using clustering algorithm,which can classify aero-engine QAR data under different health status into different categories to support the engineers and the scientists.In addition,in order to solve the problem that the measurement data of airborne sensors in QAR system is disturbed by noise greatly,and the data recorded by QAR is characterized by high dimensions,large capacity,and irregular data set shape,this paper introduces the empirical wavelet so as to do noise filtering,which help solve the problemof large noise interference of QAR data.To solve the difficulty that the dimensions of QAR feature data is too high diffusion maps algorithm has been introduced to do data reduction of QAR feature data.Besides,the paper has introduced the spectral clustering algorithm in order to solve the problem that the QAR feature data set has irregular geometry.As for the defect that the spectral clustering algorithm can not self-confirm the number of cluster categories in the data cluster,the method of selecting the cluster center point based on the density peak and the BIC criterion are introduced,and the number of data clusters and the cluster center point are selected and solved.The clustering center point selection method based on the density peak and BIC criterion are applied to the Nystrom extension of the spectral clustering algorithm to optimize the Nystrom extension,and the problem that the spectral clustering algorithm is not suitable for processing large-capacity data sets has been solved as well.