Adaptive Maximum Cyclostationarity Blind Deconvolution and Its Application on Incipient Fault Diagnosis of Bearing

For the detection of incipient bearing faults, maximum second-order cyclostationarity blind deconvolution (CYCBD) can effectively overcome noise interference and recover periodic pulses from the original signal. However, in practical applications, the main drawback of CYCBD is that the key parameters need to be selected manually by relying on prior knowledge, the uppermost important of which is the cyclic frequency or filter length. This limits the superiority of CYCBD to a certain extent. To solve the above problems, an adaptive CYCBD is proposed in this article. Firstly, the optimal demodulation band is selected by combining Autogram, and the adaptive estimation of cyclic frequency is achieved by utilizing a square envelope autocorrelation function. Secondly, a new objective function, i.e., the comprehensive performance efficiency ratio (CPER), is constructed, and the adaptive principle of filter length is further proposed by combining the equal-step search strategy. Finally, the effectiveness and advantages of ACYCBD are demonstrated by simulation and experiment.