Sliding time–frequency synchronous average based on autocorrelation function for extracting fault feature of bearings

Making weak repetitive pulses clearly appear in time–frequency distribution is essential for detecting early failure of bearings. However, this operation is a challenging issue in fault diagnosis. To resolve this problem, a signal enhancement method called sliding time–frequency synchronous average based on autocorrelation function (STFSA-ACF) is proposed in this paper, based on three ways of signal enhancement. In the method, the autocorrelation function is first utilized to enhance the repetitive components of signals. The time–frequency representation of the autocorrelation function result is obtained by short-time Fourier transform. Furthermore, an improved version of time synchronous average called the sliding time–frequency synchronous average is developed to make the weak repetitive pulses more visible. In this method, a window sliding in the time–frequency plane is introduced to intercept the signal, and the time synchronous average is employed to process the intercepted section. The aforementioned operations construct the STFSA-ACF. Finally, the gamma transform is used to improve the contrast of generated STFSA-ACF. A series of numerically simulated signals are generated to validate the proposed algorithm. Besides, this method is employed to process part signals of two sets of public data. Performance of the proposed STFSA-ACF has been compared with popular methods such as fast Kurtogram, maximum correlated kurtosis deconvolution, and adaptive maximum second-order cyclostationarity blind deconvolution. Comparison results indicate that the STFSA-ACF has the best performance in terms of making weak repetitive pulses more visible.