posted on 2021-08-03, 15:52authored byZhiqiang Huo, Eve ZhangEve Zhang, Lei Shu, Xiaowen Liao
Permutation Entropy (PE) has been widely applied as a non-linear statistical indicator to estimate the change of complexity in time series. Though it is conceptually simple and computationally fast, PE encounters a few limitations. For example, the amplitude differences in time series are neglected, and the symbolic sequences generated from equal values are according to their emergence order. In this paper, an Edge Permutation Entropy (EdgePE) measure is proposed to improve the performance of PE, mainly overcoming its lack of ability to differentiate between amplitude differences in motifs that correspond to the same order pattern. The advantage of EdgePE relies on that amplitude change information can be identified and distinguished by the information underlying in the 'edge' distance between data points in the reconstructed embedded vectors. To demonstrate its improvement, the proposed EdgePE is compared with other related improved PE approaches, for analyzing synthetic time series and experimental rolling bearing data sets, respectively. The results indicate that the EdgePE can effectively characterize amplitude changes in time series (e.g., for spike and stuck detection) and improve the accuracy of pattern recognition for rolling bearing fault diagnosis, compared to those of the other related PE measures.
Funding
International and Hong Kong, Macao & Taiwan collaborative innovation platform and major international cooperation projects of colleges in Guangdong Province (No.2015KGJHZ026)
The Natural Science Foundation of Guangdong Province (No.2016A030307029)
History
School
Aeronautical, Automotive, Chemical and Materials Engineering
Department
Aeronautical and Automotive Engineering
Published in
IECON 2019 - 45th Annual Conference of the IEEE Industrial Electronics Society