Resilient ACYCBD with uniform initialization for the bearing fault diagnosis

Abstract Fault diagnosis is of significance for ensuring the safe and reliable operation of machinery equipment. Due to the heavy noise and interference, it is difficult to detect the fault directly from the measured signal. Hence, signal processing techniques that can achieve feature extraction and signal denoising are the most common tools in the field. The adaptive maximum second-order cyclostationarity blind deconvolution (ACYCBD), an advanced blind deconvolution technique, the algorithm shows great superiority in noise reduction of faulty signals. However, ACYCBD uses a single FIR filter to process non-smooth signals, and the length of the filter has a significant impact on the deconvolution results. To address above issues, resilient ACYCBD is proposed. This approach employs a set of uniformly distributed FIR filter banks across the entire frequency spectrum, in contrast to the conventional single FIR filter. Firstly, FIR filters are generated with different cutoff frequencies. Secondly, ICS 2 serves as the objective function guiding the update of each filter coefficient within the filter bank. Additionally, after each iteration, correlation coefficients of the different modes from the filter bank are computed, and the ICS 2 of two modes with the highest correlation coefficients are compared to eliminate redundant modes. Finally, the envelope spectral Gini index is used as an evaluation index to select the corresponding filtered signal under the optimal filter length. The results of the simulation and experimental data from bearing faults indicate the efficacy of the suggested approach in enhancing the accuracy of fault identification.