An Improved Data Privacy Diagnostic Framework for Multiple Machinery Components Data Based on Swarm Learning Algorithm

• Published in: IEEE transactions on instrumentation and measurement Vol. 72; p. 1
• Main Authors: Sun, Shilong, Huang, Haodong, Peng, Tengyi, Wang, Dong
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.01.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Condition monitoring; Data collection; Data exchange; Data models; Data privacy; data privacy protection; Diagnostic systems; different local diagnosis model; Fault diagnosis; Feature extraction; Federated learning; global fault diagnosis; Machine learning; multiple machinery components; Performance degradation; Privacy; Servers; swarm learning; Training
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2023.3310057

The continuous operation of the equipment degrades the performance of the critical parts, which can cause the equipment to fail and stop at a particular and unexpected moment. Timely diagnosis of the equipment is vital for condition monitoring and maintenance. Due to the small amount of data and data collection limitation, it is difficult to train an efficient diagnosis model for real-time tracking within only one piece of equipment data. This study proposes an improved data privacy diagnostic framework for multiple same type of machinery components, solving the insufficient data, data protection, and multiple machines' fault information exchange. First, the swarm learning framework integrates various data sources to enrich the data contained within a solo diagnosis network. Second, the different training nodes utilize various local diagnosis models to improve data protection efficiency further and realize the interaction of data information. Third, we developed three different local diagnosis models which can mutually exchange the partially faulty information with each other to make up for the inefficiency of model diagnosis caused by single insufficient data. The experimental demonstration is conducted on the bearing fault datasets, proving that the proposed method can be more flexible and reliable in more industrial scenarios.