Predictive maintenance system for high-end equipment in nuclear power plant under limited degradation knowledge

Abstract

To ensure the safe operation of high-end equipment, the three-stage maintenance strategy comprising unplanned shutdown, temporary shutdown, and scheduled shutdown is currently employed in nuclear power plants. However, this strategy hinders the acquisition of degradation knowledge (run-to-failure data or degradation mechanism), thereby impeding the application of traditional predictive maintenance systems. Hence, the responsibility for determining the maintenance stage primarily lies with experienced field engineers, and an incorrect decision could potentially result in an unplanned shutdown. To this challenge, we integrate the three-stage maintenance strategy and prognosis methods to form a predictive maintenance system for nuclear power plants. The system framework is first established, and prognosis methods, including sensor selection, degradation trend prediction for short-term prognosis, and remaining useful life (RUL) prediction for long-term prognosis, are then developed under limited degradation knowledge. Finally, the system is deployed in the circulating water pump of a nuclear power plant utilizing an Internet of Things (IoT) architecture. An industry case study verifies that the proposed system can provide decision-making support and further achieve predictive maintenance for high-end equipment in nuclear power plants.