A novel method to identify weak fault features in strong composite noise by enhanced cyclic correlation entropy with truncated nuclear norm LRSD

Rolling bearings are utilized extensively to support rotating parts. However, under long-term operating conditions, bearings are prone to failure. Moreover, their initial failure is very weak and would be buried in strong Gaussian and impulse noise. When impulse noise and strong Gaussian noise coexist, the majority of signal processing methods degrade to varying degrees. To solve this issue, the enhanced cyclic correlation entropy (CCE) algorithm based on truncated nuclear norm (TNN) low-rank sparse decomposition (LRSD) is proposed to accurately identify weak fault characteristics under strong composite noise situation. Firstly, the Gaussian kernel function is introduced on the basis of cyclostationary analysis to obtain the two-dimensional cyclic correlation entropy spectrum (CCES) matrix to suppress the interference of impulse noise. Then, considering the low-rank and sparse properties of the noise-contaminated signals within the two-dimensional f-α matrix, the spectral matrix is further noise-reduced and feature-enhanced by a sparse decomposition method based on the improved TNN. Finally, a full-band integration process is performed to obtain the enhanced envelope spectrum (EES). Validation is then carried out using simulated fault signal and two bearing datasets, and the analysis results demonstrate that, unlike most existing fault diagnosis methods which are severely impaired or rendered ineffective by mixed noise, the proposed method can still reliably extract fault features while significantly mitigating noise interference.