A novelty detection scheme for rolling bearing based on multiscale fuzzy distribution entropy and hybrid kernel convex hull approximation

Novelty detection as an effective condition detection method is of great necessity for rolling bearing. In this paper, a novelty detection scheme based on multiscale fuzzy distribution entropy (MFDE) and hybrid kernel convex hull approximation (HKCHA) is proposed. First, MFDE is presented as a new entropy to measure complexity, reflect non-linear and non-consistent characteristics, and extract underlying features of vibration signals. Then, the first several MFDE values containing fault information are fed to a novel classifier, based on HKCHA, to detect rolling bearing working status more accurately. Finally, two cases concerning about rolling bearing are conducted respectively to validate the feasibility of proposed scheme, whose results indicate that MFDE is superior to other entropies and HKCHA surpasses other existing detection methods with higher detection accuracy and recall rate. (C) 2020 Elsevier Ltd. All rights reserved.