Research On Vibration Feature Extraction Method For Reciprocating Compressor Fault Diagnosis Based On Empirical Wavelet Transform

According to the non-linear and non-stationary characteristic of reciprocating compressor vibration signal,a fault feature extraction method based on adaptive empirical wavelet transform and mathematical morphology filtering is proposed.The empirical wavelet transform constructs adaptive band-pass filters in the frequency domain so as to construct orthogonal wavelet functions and extract AM-FM components that have compact support in Fourier spectrum,and then uses Hilbert transform to analyse their time-frequency properties.The empirical wavelet transform has strictly mathematical proofs as it is built on the framework of wavelet transform,and the calculation process of EWT is not iterative,so as introduced above the EWT has many advantages such as reasonable mathematical interpretation,clearly modal separation and low computational complexity when decompose the the non-linear and non-stationary signal.However,in the on-site fault diagnosis of reciprocating compressor,there are some problems like too many separated modes and non-monocomponent modes are not suitable for Hilbert transformation.Therefore,this paper bases on the experimental data,improving the empirical wavelet transform so that it can be applied to the fault diagnosis of the reciprocating compressor.Firstly,the spectrum separation method of the empirical wavelet transform is improved.On the basis of the scale space theory,the theoretical analysis and a large amount of data is used to optimize the best parameter of scale transform,and then uses Pearson coefficient and kurtosis to screen modal.Simulation and measured signals show that the improved empirical wavelet transform is a stable and effective modal decomposition method.Secondly,for the problem that the non-monocomponent modes obtained from the measured signals are not suit for Hilbert transform,a method based on mathematical morphology is proposed to filter the modes and extract fault features.This method judges the state of the signal through the energy index,and then adaptively constructs structural elements to perform morphological filtering according to the different states of the signal,making the filtering more targeted.And the fault feature extraction method of pattern spectrum entropy is introduced.When the vibration signal is in different shape features,its pattern spectrum characteristics are different,and the pattern spectrum entropy is an index that can quantitatively describe the pattern spectrum obtained through multi-scale morphological calculation.Using this method to analysis the measured signal,the results show that the pattern spectrum entropy can clearly identify different faults.Finally,a fault feature extraction method based on empirical wavelet transform and morphology filtering of different state is proposed.The method uses an improved empirical wavelet transform to de composition signal,and then calculates the pattern spectrum entropy to identify differebt faults after morphology filtering of different state.The proposed method is used to analyze the reciprocating compressor fault data.The results show that it can effectively extract the fault characteristics and realize the diagnosis of different fault types.