Optimized weights spectrum autocorrelation: A new and promising method for fault characteristic frequency identification for rotating Machine fault diagnosis

Abstract

Since fault characteristic frequencies (FCFs) and their harmonics are closely connected with specific fault types of rotating machines, identification of FCFs and their harmonics is a very crucial step for signal processing-based rotating machine fault diagnosis. Nowadays, Hilbert transform (HT) based square envelope spectrum (SES) and spectral coherence (SC) based SES are two main tools for FCF identification. Since the HT demodulates signals by calculating envelope signals in the time domain and the SC is based on the temporal instantaneous autocorrelation function of demodulated signals, the temporal waveform of a demodulated signal must be purified by using fault signature extraction methods, e.g., fast kurtogram, blind deconvolution, and their variants, etc. However, it has been extensively reported that these fault signature extraction methods are prone to be affected by interference components such as impulsive noise. Unlike the HT and SC which demodulate from the time domain, this paper aims to demodulate signals from the frequency domain to identify FCFs and their harmonics for rotating machine fault diagnosis. Firstly, it is demonstrated that Fourier spectrum autocorrelation is possible to indicate FCFs and their harmonics. Nevertheless, a troublesome problem is that interference spectral lines and noise spectral lines of real-world signals in the frequency domain will severely affect the performance of the Fourier spectrum autocorrelation. To solve such problem, this paper introduces a recently developed optimized weights spectrum (OWS) and innovatively designs an adaptive threshold method to respectively eliminate an influence of interference spectral lines and noise spectral lines. Thus, a new FCF identification method named optimized weights spectrum autocorrelation (OWSAC) is accordingly proposed. One main merit of the proposed OWSAC is that it does not need any fault signature extraction methods to do signal preprocessing. Two experimental case studies respectively on incipient bearing and gearbox diagnosis validate the effectiveness of the proposed OWSAC. The proposed OWSAC can achieve satisfactory performance respectively in two case studies and it is superior to five methods including HT-based SES, SC-based SES, optimized SES, fast kurtogram guided SES, and Fourier spectrum autocorrelation.