Bearing fault identification based on stacking modified composite multiscale dispersion entropy and optimised support vector machine

Bearing vibration signals have strong non-linear characteristics and contain significant interference noise, however, entropy is an effective means of rationalising the complexity of time-series data and provides a feasible solution for fault feature extraction, classification, and recognition. A novel stacking modified composite multiscale dispersion entropy (SMCMDE) algorithm is proposed to overcome the deficiency of multiscale dispersion entropy. Firstly, the signal is preprocessed by filtering. Then, the stacking operation was applied to the filtered signal, and the standard deviation of the coarse-grained sequence was calculated to replace the original mean operation for correction. The analysis of simulation signals shows that the proposed algorithm has certain advantages over other entropy algorithms. Subsequently, an intelligent bearing fault identification method based on SMCMDE and equilibrium optimiser-based support vector machine (EO-SVM) is proposed. The results from tests on bearings show that the proposed method can not only extract the sensitive features of bearing faults, but also has higher accuracy and stability than other methods.