Degradation evaluation of slewing bearing using HMM and improved GRU.

• Life-cycle fatigue tests of slewing bearings are conducted. • A hybrid adaptive method reduces noise of signal effectively. • Deep learning methods have more accurate predictions than shallow networks. • Optimized Gated Recurrent Unit improves effect of degradation evaluation. Degradation process assessment from normal to failure condition of slewing bearing is viewed as a part of health monitoring in condition-based maintenance (CBM). The algorithm integrating Hidden Markov Model (HMM) and improved Gated Recurrent Unit (GRU) network is proposed to establish the component's health indictor and evaluate performance degradation. As a deep learning network, GRU network has more powerful approximate ability than machine learning methods in time series prognosis problems. The research on accelerated life experiments of a certain type of slewing bearing was carried out to verify the superiority of proposed method. Firstly, the signal preprocessing includes raw signal de-noising combining Hilbert transform with Robust Local Mean Decomposition (RLMD) and feature extraction in time and frequency domains. Then, the life health indictor is established using extracted signal features through the HMM model to complete the incipient degradation recognition. Finally, an improved method Moth Flame Optimization-based GRU (MGRU) is applied to predict the health indictor and residual life of slewing bearing. Experiments comparing with several algorithms show that the proposed methods can effectively evaluate the health condition of the slewing bearing. [ABSTRACT FROM AUTHOR]