Reliability verification methods for artificial intelligence components in instrumentation and control systems at nuclear power plants

The rapid advancement of artificial intelligence (AI) technologies has created new opportunities to improve the safety, efficiency, and operational support of nuclear power plants (NPPs). However, the distinctive characteristics of AI components (e.g., high complexity, lack of formal specifications, and non-deterministic behavior) pose significant challenges in ensuring their reliability. This study proposes a comprehensive framework for verifying the reliability of AI models used in instrumentation and control systems in NPPs. The framework integrates both qualitative and quantitative evaluation methods. The qualitative evaluation assesses the safety, transparency, and procedural reliability of AI model development, and the quantitative evaluation focuses on the performance and robustness testing of the final AI model. Additionally, a case study demonstrates the practical application of the proposed framework to real-world model evaluation. This research aims to serve as a foundational reference, fostering understanding and consensus among system engineers, operators, and regulators regarding the reliability of AI technologies in nuclear power applications.