Application Of Empirical Wavelet Transform And Zero Frequency Filter In Engine Fault Diagnosis

As the core component of motor vehicle,the internal structure of engine is extremely complex,and there are many kinds of unbalanced rotating and reciprocating gears and bearings.Most fault signals are non-linear and non-stationary pulse modulation signals,and there is always serious noise pollution,especially in the early fault,so it is difficult to effectively extract the characteristic frequency for fault diagnosis.In order to extract the engine fault information effectively,this paper establishes the idea of two-level analysis to separate the characteristic frequency band and amplify the pulse information to extract the characteristic frequency,and studies the empirical wavelet transform and zero frequency filter algorithm.Firstly,in order to effectively separate the characteristic frequency,the advantages and disadvantages of EMD,WPT and EWT are compared by simulation signal analysis.The results show that there are over decomposition phenomena in both empirical mode decomposition and wavelet packet transform.Redundant components in the original signal are obviously separated from the decomposition components,and there are a lot of frequency overlaps.Compared with empirical mode decomposition and wavelet packet transform,the resolution of empirical wavelet transform is obviously higher.It can decompose each quantity accurately and independently under weak noise.It neither excessive decomposition nor frequency overlap is avoided.However,due to noise peak interference,the division of the induced frequency band is unreasonable,and the phenomenon of frequency aliasing appears,which has some shortcomings.In view of the shortcomings of EWT decomposition,an improved method based on Otsu and wavelet energy spectrum method is proposed.Through Otsu calculation of the wavelet energy spectrum of the signal,the pixel is binarized and the threshold is determined to determine the best threshold.Then the Otsu calculation results of the first EWT decomposition component are screened,and the frequency band corresponding to the component with energy value greater than is reserved,and the energy value less than is selected A new scheme of frequency band division is obtained by combining the corresponding frequency bands of the components.Then EWT decomposition is performed again to extract the characteristic frequency bands.Then,the zero frequency filter algorithm proposed recently in speech signal processing is applied to fault signal processing.By continuously integrating the normalized Hilbert envelope of the time domain signal in the characteristic frequency band,and calculating the residual between the zero frequency filter output and its local mean value,the discontinuous features caused by pulse excitation in the signal are extracted and amplified,and the noise is effectively weakened The characteristic frequency is extracted from the residual signal.On this basis,a two-stage fault diagnosis method based on improved EWT decomposition and zero frequency filter is proposed.Finally,taking a JS110 motorcycle engine as the research object,the acoustic signal processing based on the fault diagnosis method of improved EWT decomposition and zero frequency filtering is carried out to solve the problem of abnormal noise of the right crankcase cover.The characteristic frequency band of 2000～4000Hz is completely separated,and the characteristic frequency 27 Hz and its second and third harmonic are accurately extracted from the residual signal output by zero frequency filter.The wavelet energy spectrum also shows that the residual signal has good performance The results show that the noise is mainly concentrated at 27 Hz,which further proves that the abnormal noise comes from the oil pump drive gear shaft,and verifies the effectiveness of the method.