Abstract:
Fault diagnosis has normally been utilized to effectively identify bearing fault patterns and severities in the whole life cycle in modern intelligent agriculture. In this study, an intelligent fault diagnosis was proposed for the rolling element bearing using hierarchical multiscale dispersion entropy. Firstly, a signal complexity assessment called the hierarchical multiscale dispersion entropy (HMDE) was proposed to integrate the theories and ideas of hierarchical decomposition and multiscale analysis using dispersion entropy theory. Secondly, a swarm intelligence optimization named bird swarm (BSA) was employed to determine the important parameters. The HMDE with the optimized parameter was utilized to extract multilevel and multiscale fault features hidden in the raw bearing vibration signal and avoid empiric selection of HMDE parameters. Finally, the multi-dimensional fault feature matrix was constructed and then fed into the support matrix machine (SMM) for the training of the SMM model. The well-trained SMM model was adopted to automatically identify different fault patterns and severities of rolling bearing. A two-group test of bearing accelerated life was carried out to verify the model. Experimental results showed that the diagnostic accuracy reached 99.66% in the first group, whereas, the diagnostic accuracy of seven (i.e., refined composite multiscale dispersion entropy (RCMDE), generalized composite multiscale permutation entropy (GCMPE), generalized refined composite multiscale sample entropy (GRCMSE), hierarchical fuzzy entropy (HFE), hierarchical sample entropy (HSE), modified hierarchical multiscale dispersion entropy (MHMDE) and hierarchical multiscale permutation entropy (HMPE)) were 95.77%, 87.32%, 93.03%, 90.51%, 92.57%, 98.85%, and 97.03%, respectively. In the second group, the diagnostic accuracy reached 99.34%, whereas, the diagnostic accuracy of seven (i.e., RCMDE, GCMPE, GRCMSE, HFE, HSE, MHMDE and HMPE) were 97.17%, 95.83%, 93.17%, 89.83%, 87.83%, 98.17%, and 96.33%, respectively. It was clearly found that the average accuracy in the first group was improved by the percent point of 3.89, 12.34, 6.63, 9.15, 7.09, 0.81 and 2.63, respectively, where that in the second group increased by the percent point of 2.17, 3.51, 6.17, 9.51, 11.51, 1.17, and 3.01, respectively, compared with the seven (RCMDE, GCMPE, GRCMSE, HFE, HSE, MHMDE, and HMPE). The diagnostic accuracy demonstrated that the different fault patterns and severities of rolling bearings were identified in the whole life cycle. In addition, the broader and richer feature information of bearing faults was achieved with a greatly better identification performance, compared with the traditional fault diagnosis using multiscale or hierarchical entropy. The finding can provide a new idea to improve the fault diagnosis accuracy of rolling bearings in the whole life cycle.