PSAM 16 Conference Paper Overview
Welcome to the PSAM 16 Conference paper and speaker overview page.
Lead Author: Koushik Araseethota Manjunatha Co-author(s): Vivek Agarwal, vivek.agarwal@inl.gov
Randall D. Reese, randall.reese@inl.gov
| Federated Transfer Learning for Scalable Condition based Monitoring of Nuclear Power Plant Components |
|---|
| Condition-based monitoring (CBM) techniques are widely being adopted for maintenance activities in nuclear power plants. Asset operational data are collected by smart sensors mounted on and around the components. The sensed data is often gathered and processed by a monitoring and diagnostic center to garner various component fault signatures. These fault signatures are subsequently used as input to train predictive machine learning (ML) models for the specific component. Development of ML models require a significant amount of healthy and fault data. As faults are rare events, it is highly unlikely that all the potential fault modes are captured for a single component. Moreover, new components without historical data cannot contribute to ML model development. Additionally, fault signatures extracted from a single component cannot be robust enough to handle unseen fault patterns in same or different components. Privacy, security, legal, and commercial concerns often prevent data sharing across different plant systems. This research presents federated transfer learning (FTL) to scale ML models for CBM across components or plant systems by combining federated learning (FL) and transfer learning (TL) approaches as shown in Figure 1. FL enables developing local models at the component level across different units that are securely shared to a centralized server to aggregate into a global model. TL enables application of the developed aggregated global model to different but related systems within the same plant site, or to the same system at different plant sites. FTL was demonstrated for the circulating water system from two nuclear plant sites (representing three units) to predict the health of circulating water pumps. The FTL framework was verified using a multi-kernel adaptive support vector machine and an artificial neural network. The results show significant improvement in prediction performance while reducing over-fitting issues. |