An effective fault diagnosis approach for bearing using stacked de-noising auto-encoder with structure adaptive adjustment

• Published in: Measurement : journal of the International Measurement Confederation 2023-06, Vol.214, p.112774, Article 112774
• Main Authors: Chen, Lerui, Ma, Yidan, Hu, Heyu, Khan, Umer Sadiq
• Format: Article
• Language: English
• Subjects: Bearing; Fault diagnosis; SDAE; Structure adaptive adjustment
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2023.112774

•Multi-sensor vibration data are adopted to enrich the bearing fault features.•A novel loss function of SDAE is proposed to improve the network performance.•An adaptive algorithm for SDAE structure adjustment is proposed to improve efficiency. Fault diagnosis of bearing plays an important role in maintaining the stable operation of rotating equipment. However, the existing approaches are not effective enough in multi-working conditions. Especially, the diagnosis network structure was selected by experience, which is not self-adaptive and smart. In addition, the data collected in the industrial environment contain noise, which would interfere with the diagnosis results. To overcome these concerns, an effective approach based on stacked de-noising auto-encoder (SDAE) with structure adaptive adjustment is proposed for bearing fault diagnosis. In the aspect of fault information acquisition, two vibration sensors are adopted to collect the vibration data of bearing in two directions to enrich the fault information. In the aspect of diagnosis network design, a novel comprehensive loss function of SDAE in the process of reverse fine-tuning is proposed, which makes the network optimization in favor of feature classification. Meanwhile, an adaptive optimization algorithm is designed to adaptively adjust the structure of SDAE, which can avoid the blind selection of network structure by experience. The proposed approach is verified by experiments, and the results indicate that the proposed method has the excellent performance in diagnosis accuracy, anti-noise, and generation ability.