Machine learning methods in equipment life management

Spanish nuclear power plants have accumulated extensive operational experience, approaching 40 years of service, and must continuously maintain the highest standards of safety and reliability in their systems. In this context, the development of tools capable of accurately assessing equipment health and predicting Remaining Useful Life (RUL) is of great value. By leveraging information generated through online condition monitoring, degradation can be detected and failures prevented, thereby ensuring safe and efficient plant operation.

Proposed methodology

This study presents the methodology used to develop condition-based maintenance models aimed at diagnosing equipment degradation status, estimating time to failure, and providing quantitative support for maintenance decision-making. The methodology is framed within a broader research project focused on developing artificial intelligence applications and techniques to improve life management and maintenance strategies for nuclear power plants under Long-Term Operation. It can be summarized in the following steps: first, the variables most strongly related to equipment degradation are identified and processed to facilitate information extraction. Next, different statistical distributions are fitted to obtain the reliability model, enabling estimation of the probability of failure at any given time. Finally, the evolution of the degradation indicator is modeled over successive operating cycles. Combined with the reliability model, this allows projection of future failure probabilities and determination of the point at which maintenance intervention will be required.

Application case

In addition, a case study is presented in which specific techniques are applied to a dataset representative of the degradation behavior of the equipment under analysis. The results obtained in the case study demonstrate the validity of the proposed methodology in generating tools that support more realistic and informed decision-making.