Gearbox Fault Diagnosis Based On Translation Invariant Wavelet And Ensemble Local Mean Decomposition

Gearbox is a kind of typical connecting and power-transfering equipment,widely used inmachinery industry. Therefore, the gearbox fault diagnosis is very valuable and very necessary.The research object of this paper is ZS65type gearbox, and gearbox vibration signals iscollected in various conditions by DASP data acquisition system. Then, the ensemble localmean decomposition(referred to as ELMD) method is used for vibration signal analysis inorder to implement fault diagnosis.Before processing and analysis of the collected vibration signal, the first important stepis to remove the noise signal, in order to eliminate the interference of noise signal on theuseful signal. The result of signal denoising will have a direct impact on the final result of theensemble local mean decomposition. The translation invariant wavelet noise reductionmethod is used for signal denoising in this paper to eliminate the phenomenon of "pseudoGibbs’ caused by the traditional wavelet thresholding denoising method. The simulationshows that compared with the traditional wavelet thresholding denoising method, thetranslation invariant wavelet noise reduction method has higher of the noise ratio andsmoothness index and the graphics more smooth.The experiment subject of this paper is ZS65type of three speed gearbox, on whichconducted a series of fault diagnosis experiments. Gearbox vibration signals are analyzed inthe conditions of gearbox normal conditions, geartooth wear failure, bearing outer ring failure,rolling wear failure and cage fracture failure. In order to eliminate aliasing modes which wascaused by the local mean decomposition method in the decomposition process of the vibrationsignal, this paper uses the ensemble local mean decomposition method which based on noise assisted for decomposing the denoising vibration signal, using the ratio of standard deviationof the high-frequency components obtained by wavelet and denoising signal as the standarddeviation of white noise. It decomposes the signal mixed with white noise, getting a series PFcomponent. Then, it chooses one of PF components for analysing the refinement spectrum inorder to extract the fault frequency and realise the fault diagnosis. Experimental results showthat combining the translation invariant wavelet and the ensemble local mean decompositionmethod is applied to gearbox fault diagnosis, getting good results.