Maximum L-Kurtosis deconvolution and frequency-domain filtering algorithm for bearing fault diagnosis

Blind deconvolution technique can effectively recover the periodic fault impulses. However, deconvolution algorithms may not perform well for fault detection if the objective criteria are sensitive to outliers. Aiming at solving the limitation, a novel maximum L-Kurtosis deconvolution (MLKD) algorithm is proposed. Firstly, the L-Kurtosis is introduced, and the results show that it remains approximately unchanged for different outliers. This characteristic of the L-Kurtosis is the keystone for the proposed deconvolution algorithm. Secondly, an iterative filter is designed to deconvolve the unknown fault signals by maximizing the L-Kurtosis. Additionally, a frequency-domain filtering (FDF) algorithm is further established to reduce the effect of noise component. Finally, based on the quantitative indexes, the proposed MLKD and FDF algorithms are effectively validated and compared with the stat-of-the-art algorithms through simulated and experimental signals. Overall, results show that the L-Kurtosis-based blind deconvolution algorithm, combined with the frequency-domain filtering technique, has a noticeable advantage over comparative algorithms.