• creep
• finite element analysis
• predictive maintenance
• safety
• long term operation
• ageing
• artificial intelligence
• neural network
• autoencoder
• anomaly detection
• monitoring
• pressure vessel
• station black out

Addressing the service life and operational safety of Nuclear Power Plants (NPPs), this thesis comprehensively investigates the aging, degradation based on predictive monitoring in the systems, structures, and components (SSCs) of NPPs. A multipronged approach, utilizing both high performance thermophysical simulation (FEA) and artificial intelligence methods, is employed to study and predict the ongoing and future integrity of NPPs, especially focusing on the reactor pressure vessel (RPV) and primary pipes, ensuring their safe long-term operation (LTO).
A cornerstone of this investigation is the application of the inverse space marching method to evaluate the thermal loads and reconstruct temperature and stress profiles in various sections of the pipes. This method, based on the monitoring of accessible boundaries, allow the analysis of temperature at inaccessible surfaces, facilitating an accurate assessment of thermal gradients.
Moreover, accurate analysis of a reactor pressure vessel (RPV) under a simulated station blackout (SBO) event is conducted. Coupling between MELCOR and Finite Element Method (FEM) codes is performed, demonstrating a significant reduction in RPV's thermomechanical properties at high temperatures. The findings highlight the vulnerability of aged RPVs at the end of life, underscoring the necessity of timely and effective intervention measures.
Monitoring and rapid response mechanisms are crucial for the safety of nuclear power plants. Sensors installed on Safety Systems and Components (SSCs) constantly generate multivariate time series data, creating a pool of significant information into the operational state of the power plants. Traditional methods of monitoring often rely on threshold-based approaches that may not capture the complex relationships between variables in multivariate systems. To address this gap, our methodology incorporates a suite of machine learning algorithms designed specifically for anomaly detection, including Autoencoder, Minimum Covariance Determinant (MCD), Local Outlier Factor (LOF), Isolation Forest, and One-Class Support Vector Machine (OCSVM). Among them, the Autoencoder stood out, detecting an anomaly approximately 39 minutes prior to RPS and ECCS safety system activations. The proposed method is not only scalable but also adaptable to different configurations and sensor setups. Preliminary results indicate that our approach outperforms conventional monitoring techniques.
This study offers a consistent, holistic, and innovative approach to ensuring the LTO of NPPs, based on cutting-edge computational high-performance techniques to ensure the ongoing safety, integrity, and efficiency of nuclear plants in the LTO framework. The findings not only underline the critical aspects influencing the degradation and ageing of nuclear plant components but also propose effective strategies for ensuring the long-term operation and safety of nuclear power plants. The multifaceted approach adopted in this research offers significant contributions to the body of knowledge, facilitating informed decision-making for the extended and safe operation of aged nuclear plants.