Correlation analysis and predictive performance based on KNN and decision tree with augmented reality for nuclear primary cooling process / Ahmad Azhari Mohamad Nor

Efficient and safe maintenance of nuclear reactors hinges on the optimal condition of critical components, particularly the primary cooling system governed by sensor, pump, and valve conditions. The intricate nature of these systems demands a well-managed maintenance program, prompting the need for a comprehensive characterisation of the primary cooling system. This study addresses the challenges posed by traditional maintenance approaches within nuclear power plants, where lengthy procedures and a laborious working process hinder the timely identification and resolution of potential operational issues. Practical algorithms incorporating k-nearest neighbour and decision tree models with quantifiable performance metrics are proposed to streamline and optimise the maintenance process. The study aims to enhance the monitoring and maintenance of the primary cooling system through a multifaceted approach. Firstly, descriptive and correlation analyses characterise the primary cooling system based on temperature, flow, and conductivity data. These analyses provide nuanced insights into system operational dynamics and efficiency. Subsequently, predictive models employing k-nearest neighbour and decision tree algorithms are constructed and evaluated based on accuracy, precision, and recall metrics. Furthermore, the research introduces an innovative augmented reality application utilising a 3D marker-based technique. This application, integrated with a predictive model, offers real-time visualisation of primary cooling system characteristics through handheld devices. The study presents promising results, showcasing the effectiveness of the predictive model in fault detection and the initial design of an augmented reality application. Results include descriptive and correlation analysis revealing crucial primary cooling system data patterns. Predictive modelling using k-nearest neighbour demonstrates high accuracy, precision, and recall metrics, while decision tree modelling raises considerations for further refinement. The early design of the augmented reality application exhibits potential for real-time data visualisation, providing insights into optimal detection distances and angles. This research lays the foundation for future investigations into integrating methods for stable augmented reality applications in cooling systems. The findings contribute significantly to predictive maintenance practices, offering a comprehensive solution for efficient monitoring, early fault detection, and informed decision-making in nuclear reactor environments.