AI-driven diagnostic digital twin demonstration for condition monitoring in nuclear piping systems

Globally, advanced nuclear reactors are becoming a key part of the clean energy transition, offering reliable, low-carbon power with enhanced safety and efficiency. Their compact design and ability to reduce greenhouse gas emissions make them ideal for deployment in remote areas and disaster recovery zones. However, reducing maintenance costs while ensuring safety remains essential for their widespread adoption. The use of Digital Twin (DT) technology can facilitate autonomous operations by enhancing system health monitoring and minimizing maintenance-related costs. One critical area of focus in nuclear DT frameworks is the condition monitoring of safety-critical systems, such as piping networks, which are susceptible to flow-accelerated corrosion and erosion, leading to pipe-wall thinning. Without timely intervention, such degradation can result in unplanned shutdowns and costly repairs. Conventional non-destructive testing (NDT) methods, though effective, are labor-intensive and often require halting operations, resulting in financial losses. By integrating AI-driven condition monitoring within the digital twin, sensor data can be analyzed to detect early signs of structural degradation, allowing for proactive maintenance without the need for costly NDT procedures. This study develops a diagnostic digital twin framework for monitoring the integrity of nuclear piping systems. Using AI-based techniques, such as Multi-Layer Perceptrons (MLPs) and Convolutional Neural Networks (CNNs), degradation patterns like pipe-wall thinning are simulated and analyzed. A laboratory demonstration is conducted using both simulated and empirical data from a piping system subjected to operational vibrations. The performance of deep learning algorithms is evaluated based on diagnostic accuracy and computational efficiency, highlighting the potential for scalable degradation detection in nuclear piping systems.