|
Abstract
|
The maintenance tasks of rotating machines are essential to achieving robust health management throughout the asset’s life cycle. However, the conventional widely-used health monitoring methods have shortcomings such as the reliance on the selection of the preset parameters to extract the fault feature. Also, the strong noise interference caused by factors such as transmission path hinders the practical application of many fault feature extraction methods. To overcome these gaps, the digital twin notion is introduced and a new digital twin architecture called the Ramanujan Digital Twin (RDT) is designed. The Ramanujan Periodicity Transform (RPT) model is employed to isolate the potential fault feature. For each frame in the whole life cycle of the rotating machine, the simulation model with high-fidelity to the potential fault features is constructed based on prior knowledge and the real-time operating condition regarding the asset. Once the simulation-induced virtual sample with high-fidelity is obtained, the RPT will be used to provide guidance information about the potential fault. With this information, the potential fault feature can be extracted without preset parameter selection and a health indicator (HI) can be constructed to perform multiple service end tasks including health monitoring and early fault prediction. The effectiveness and the robustness of the RDT are validated through 2 experimental cases.
|