Development of a digital twin system for inspection UAV in fusion reactors

Abstract

Regular inspection and maintenance of internal structures in nuclear fusion reactors are critical for ensuring safe and stable operation. To overcome the limitations of traditional inspection methods in extreme fusion environments, this study proposes an intelligent Unmanned Aerial Vehicle (UAV) inspection system based on a fivedimensional Digital Twin (DT) model. The system establishes a three-layer architecture (“cognitive-digital twin integration-intelligent decision-making”) to achieve high-fidelity virtual mapping of the UAV and its environment. A trajectory tracking algorithm leveraging Nonlinear Model Predictive Control (NMPC) is designed to integrate dynamic disturbance observers and time-varying safety constraints to resolve trajectory accuracy and obstacle avoidance coupling challenges. A ROS-Unreal Engine co-simulation framework and a three-level digital twin architecture (“L0-L1-L2”) are developed and validated through hardware-in-the-loop (HITL) simulations and synchronous mapping experiments. Results demonstrate robust performance in complex environments, with a Root-Mean-Square-Error (RMSE) of 0.092 m for virtual-real pose mapping and trajectory tracking accuracy within ±0.15 m. The system achieves complete closed-loop verification of “perception-decision-execution” in fusion reactor environments, providing a high-precision, low-risk solution for the intelligent maintenance of nuclear fusion facilities.