Optimal Weighted Multi-Scale Entropy-Energy Ratio Feature for Rolling Bearing Degradation Assessment

Abstract Extracting effective fea tures from the full lifecycle da ta of mechanica l equipment components and constructing a high-qua lity degrada tion assessment model are crucia l in da ta-driven rolling bearing performance eva lua tion. Traditiona l single-fea ture methods often fa il to capture the complex characteristics and subtle changes of faults comprehensively, leading to inadequa te sensitivity and accuracy in early degrada tion detection. To address this issue, an adaptive optima l weighted multi-sca le entropy-energy ra tio feature extraction method is proposed and combined with a degrada tion model for eva lua tion. This method optimizes the rela tionship between multi-sca le permuta tion entropy (MPE) and root mean square (RMS) through weighted parameters, integra ting the advantages of both to enhance fault detection capabilit ies. Additiona lly, a weight parameter ca lculation method is introduced to ensure tha t fea tures ma inta in good degrada tion expression under varying conditions. The fea tures are then input into a common degrada tion model to genera te the rolling bearing degrada tion curve. The effectiveness of the proposed method is va lida ted using da tasets from the
NSFI/UCR Intelligent Ma intenance Center (IMS) and Xi'an Jiaotong University-Changxing
Shengyang Technology (XJTU-SY). Compared to fea tures before optimization, the proposed method significantly improves early degrada tion point detection, monotonicity, correla tion, and robustness.