Digital Twin-Assisted Fault Diagnosis of Rotating Machinery Without Measured Fault Data

• Published in: IEEE transactions on instrumentation and measurement Vol. 73; pp. 1 - 10
• Main Authors: Xia, Jingyan, Huang, Ruyi, Li, Jipu, Chen, Zhuyun, Li, Weihua
• Format: Journal Article
• Language: English
• Published: IEEE 2024
• Subjects: Data models; Digital twin (DT); Fault diagnosis; Feature extraction; gearbox; Machinery; Maintenance; Monitoring; Training; variational mode decomposition (VMD); virtual fault data
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2024.3417592

Timely and accurate data-driven fault diagnosis approaches are essential for ensuring the reliable operation and efficient maintenance of rotating machinery. However, practical applications face challenges in obtaining sufficient fault samples in advance, making it difficult to construct effective fault identification models. The digital twin (DT) methodology offers a potential solution to overcome this obstacle by providing a training dataset through simulation techniques. Currently, one of the most critical challenges is how to effectively leverage virtual fault diagnostic knowledge to boost the development of DT-assisted fault diagnosis methods. Thus, this article develops a DT-assisted intelligent fault diagnosis approach for rotating machinery without any measured fault data. First, the virtual model of a monitored device is developed, capable of generating virtual vibration signals with multiple health states under varying working conditions. Second, valuable information from the virtual fault data is effectively and automatically mined using the variational mode decomposition (VMD) technique and kurtosis calculation. Finally, an intelligent fault diagnosis model is constructed using the extracted virtual fault information. A case study involving a classical gearbox is conducted to verify the feasibility of the proposed DT-assisted approach. The diagnostic results indicate that this method can detect occurring faults and identify the fault types reliably, demonstrating that a new application paradigm can be applied for fault diagnosis in real-world scenarios even without any measured fault data.