Abstract
Processing and extracting actionable information such as fault or anomaly indicators originating from vibration telemetry is both challenging and critical for an accurate assessment of mechanical system health and subsequent predictive maintenance. In the setting of predictive maintenance for populations of similar assets, the knowledge gained from any single asset should be leveraged to provide improved predictions across the entire population. In this paper, a novel approach to population-level health monitoring is presented adopting a transfer-learning approach. The new methodology is applied to monitor multiple rotating plant assets in a power generation scenario. The focus is on the detection of statistical anomalies as a means of identifying deviations from the typical operating regime from a time series of telemetry data. This is a challenging task because the machine is observed under
different operating regimes. The proposed methodology can effectively transfer information across different assets, automatically identifying segments with common statistical characteristics and using them to enrich the training of the local supervised learning models. The proposed solution leads to a substantial reduction in mean square error relative to a baseline model.
different operating regimes. The proposed methodology can effectively transfer information across different assets, automatically identifying segments with common statistical characteristics and using them to enrich the training of the local supervised learning models. The proposed solution leads to a substantial reduction in mean square error relative to a baseline model.
Original language | English |
---|---|
Article number | e21 |
Number of pages | 19 |
Journal | Data-Centric Engineering |
Volume | 6 |
DOIs | |
Publication status | Published - 17 Mar 2025 |
Funding
This work was also supported by the Engineering and Physical Sciences Research Council under grant EP/R004889/1 ’Delivering Enhanced Through-Life Nuclear Asset Management’
Keywords
- transfer learning
- domain adaptation
- change point detection
- statistical discrepancies
- rotating plant
- power station
- vibration
- condition monitoring