Rolling element bearing weak fault diagnosis based on optimal wavelet scale cyclic frequency extraction

Abstract

Rolling element bearing fault characteristic information is within the second-order cyclic stationary signal. However, it is susceptible to noise interference. In this article, a new method is proposed for rolling element bearing early fault characteristic extraction according to the cyclic periodogram method. The wavelet transform coefficients are processed and analyzed using the cyclostationary theory. As a result, the implicit cyclic characteristics are contained in wavelet transform coefficients. Therefore, using the modulus or envelope of wavelet transform coefficients instead of the calculation of the cyclic statistics can avoid the window function length selection while maintaining the computation rate. In addition, the calculation of correlated kurtosis is introduced into frequency domain to select optimal wavelet scales. The larger the correlated kurtosis, the stronger the cycle impact characteristic in wavelet coefficients. Calculating the cyclic frequency in the optimal wavelet scale range can accurately extract the weak fault characteristic information. The data processing results demonstrated that the proposed method outperforms existing cyclostationary signal analysis methods in weak fault feature extraction for rolling element bearing.