Purpose This paper aims to solve the problem that multiscale dispersion entropy (MDE) is prone to information loss in the process of coarse-grain, which makes it difficult to extract bearing fault information comprehensively. Design/methodology/approach A new fault diagnosis method of rolling bearing using refined composite multiscale peak-to-peak normalized dispersion entropy (RCMPNDE) and sparrow search algorithm optimized probabilistic neural network (SSA-PNN) is proposed. First, coarse-graining employs the peak-to-peak value calculation instead of the segmented mean calculation in the RCMDE algorithm, which can overcome the shortcomings of traditional coarse-graining and highlight the fault characteristics. Then, the influence of the selection of different parameters is reduced through the normalization operation, and the RCMPNDE is formed. Finally, the extracted feature parameters are combined with SSA-PNN for diagnosis recognition to construct the RCMPNDE-SSA-PNN fault diagnosis method. Findings The proposed RCMPNDE-SSA-PNN fault diagnosis method is tested on actual data sets and its outcomes have been compared to those generated by methods built upon MDE, RCMDE and PNN. The comparison results showed that the proposed method can extract the fault feature information of rolling bearings more accurately and improve the accuracy of fault classification. The recognition accuracy reached 98.5% under the conditions of this experiment. Originality/value The RCMPNDE-SSA-PNN method can obtain more accurate fault diagnosis accuracy and provide a new reliable diagnosis method for rolling bearing fault diagnosis. Peer review The peer review history for this article is available at: https://publons.com/publon/10.1108/ILT-09-2024-0332/
Read full abstract