Rolling bearing performance degradation assessment based on deep belief network and improved support vector data description. (1st December 2022)

• Record Type: Journal Article
• Authors: Pan, Yuna; Cheng, Daolai; Wei, Tingting; Jia, Yuchen
• Is Part Of: Mechanical systems and signal processing. Volume 181(2022)
• Publication Date: 2022-12-01
• Subjects: Performance degradation assessment -- Rolling bearing -- Deep belief network -- Support vector data description -- Sparrow search algorithm; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2022.109458
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: DBN without the classification output layer is selected as the feature self-extraction model to avoid the participation of human experience, and save time and effort to a certain extent; SVDD is selected as the degradation index to build the model, and SSA is used to solve the problem that its parameters are not easy to determine, and expounding the principle of SSA-SVDD; Two sets of bearing whole-life cycle experimental data were selected to verify the effectiveness of the method. The degradation index constructed by the method in this paper can effectively evaluate the performance degradation process of rolling bearings under different situations, and accurately detect the occurrence of early weak faults of rolling bearings. The model can be constructed with the data under normal conditions, which overcomes the difficulty of obtaining fault samples during the operation of actual bearing equipment, and has a good guiding significance for the performance monitoring of bearing equipment. Abstract: Rolling bearing performance degradation assessment (PDA) based on vibration signal is critical to intelligent maintenance. However, there are two problems in existing methods: 1) degradation signals are needed as building model, 2) feature extraction is dependent on experience and operation parameters. Aiming at these challenges, a PDA method based on deep belief network (DBN) and improved support vector data description (SVDD) is proposed in this paper. The normalized amplitude spectrum under normal state is used as the training sample of DBN without the classified output layer, which is used as the automatic feature extraction model. Then, the SVDD which requires appropriate penalty parameter C and kernel parameter σ is employed to fuse these features into performance indicator (PI), and the sparrow search algorithm (SSA) is introduced into the parameters optimization process of SVDD. The results applied in two experiment datasets show that the proposed method can detect the occurrence of incipient degradation and reflect the whole bearing performance degradation. And the advantages over other methods have been shown. In addition, the anti-noise capability, compared with RMS, is discussed, and the results show that the proposed method is better than RMS to some extent, which is worth being researched further in principle.