The Extraction Of Signal Feature Based On Spectral Kurtosis And Its Application In The Fault Diagnosis Of Crucial Components In Transmission System

Localized fault in machine components including bearings and gears tend to result inshocks and arouse transient impulse response in vibration signal which is the dynamicforms of the machine, and the characteristic waveform changed with the evolution of fault.Thus, the extraction of signal transient feature which shows the localized fault in machinecomponents has always been the most crucial problem in machinery fault diagnosis. Thisresearch is supported financially by the Natural Science Foundation of Jiangsu Province(No. BK2010225) and also by the State Key Laboratory for Manufacturing SystemsEngineering of Xi’an Jiaotong University (No. sklms2011006). With the aim of machineryfault diagnosis, with the research target of bearings and gears, which are the crucialcomponents of transmission system, this paper proposes the method based on spectralkurtosis to extract the signal transient feature. The theoretical research and applicationresearch are studied in depth, respectively.Firstly, the failure mechanism and the characteristics of vibration signal of bearingsand gears in the transmission system are analyzed respectively, which provides thetheoretical support of rationality and necessity about the extraction of signal feature. Toensure that the theoretical research works are validated by the experiment analysis, thebearings and gears are tested under the localized fault condition and the vibration signalsare collected.The theory, the definition and the properties of spectral kurtosis and the detectiontheory of signal transient feature with spectral kurtosis are introduced systematically.Besides, the traditional algorithms of spectral kurtosis, including the Short Time Fourier ransform-based spectral kurtosis, the fast kurtogram and the adaptive spectral kurtosistechnique based on window superposition are introduced, and further the characteristicsand shortcomings of these algorithms are explained respectively through the simulationanalysis.This paper proposes an adaptive spectral kurtosis filtering based on Morlet waveletsin view of the shortcomings of traditional algorithms of spectral kurtosis. The Morletwavelets are used as a filter bank whose center frequency is defined by the waveletcorrelation filtering. Different bandwidth filter in the filter bank is used to select theoptimal filter for extracting the signal transient feature as the one that maximizes the SK.The correctness and effectiveness of the proposed technique are verified through thesimulation analysis and combined with the adaptive spectral kurtosis technique based onwindow superposition.By considering the practical application in machinery fault diagnosis, the proposedtechnique is applied in the fault diagnosis of crucial components in transmission system.For the detection of bearings fault, gearbox fault and automobile transmission fault, theproposed technique is applied in the extraction of the transients impulse response that showthe localized fault of bearing outer race, inner race, rolling and the gears, which furtherproves the correctness in theory and the effectiveness, applicability in practical.In this paper, it is confirmed that the technique of adaptive spectral kurtosis based onMorlet wavelet is effective for extracting the signal transient feature through the researchon the extraction of characteristic signal based on spectral kurtosis, which is of certaintheoretical and practical value for machinery fault diagnosis.