Multilayer-graph-embedded extreme learning machine for performance degradation prognosis of bearing. (15th February 2023)

• Record Type: Journal Article
• Authors: Shi, Mingkuan; Ding, Chuancang; Que, Hongbo; Wu, Chengpan; Shi, Juanjuan; Shen, Changqing; Huang, Weiguo; Zhu, Zhongkui
• Is Part Of: Measurement. Volume 207(2023)
• Publication Date: 2023-02-15
• Subjects: Performance degradation -- Multilayer-Graph-embedded Extreme Learning Machine -- Feature fusion -- Local information and non-local information -- Information mining; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2022.112299
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: The GEELM-AE that can preserves local structure information and non-local structure information of the input data is proposed. The MGEELM model used for performance degradation prognosis of bearing is proposed. The proposed model combines the advantages of ELM and deep learning algorithms. Multiple bearing experimental datasets are analyzed to validate the method. Abstract: As a key component in electromechanical systems, the health condition monitoring of rolling bearings is crucial for the safe operation of the whole system. For this purpose, the prediction of rolling bearing performance degradation is indispensable. To improve the accuracy of Extreme Learning Machine (ELM) based algorithms for the bearing performance degradation prediction, a novel Graph embedded ELM autoencoder (GEELM-AE) is first constructed via combining the graph embedding framework and then a performance degradation prediction method of Multilayer-Graph-embedded ELM (MGEELM) with a deep framework is developed by stacking multiple GEELM-AEs. The MGEELM algorithm not only extracts the abstract features in the data by virtue of its deep structure, but also maintains the local structural information and non-local structural information of the data during feature extraction. Accordingly, the performance degradation trend of rolling bearings can be accurately predicted with the proposed MGEELM. Moreover, the MGEELM algorithm does not require reverse fine-tuning, which greatly reduces the training time and improves the prediction efficiency. The advantages of the proposed method are validated by rolling bearing life-cycle vibration data. In the experimental analysis, the four performance indicators of the proposed method obviously outperform the comparative methods, which indicates the superior ability of the proposed method in tracking the bearing operating condition evolution.