A review of data-driven fault diagnosis method for nuclear power plant

Fault diagnosis plays a pivotal role in the operation and maintenance of nuclear power plants, serving as a cornerstone for ensuring facility safety and operational reliability. This review systematically examines state-of-the-art fault diagnosis methodologies implemented in the nuclear power industry, categorizing them into five distinct approaches: expert system, graph theory, machine learning, deep learning and reinforcement learning. Through comparative analysis of their theoretical foundations and implementation mechanisms, this study specifically evaluates their efficacy in addressing two critical challenges: the coolant circulation system and age-related equipment degradation and predictive maintenance. The synthesis of recent advancements reveals an emerging paradigm shift: the synergistic integration of data-intensive AI algorithms with physical mechanism demonstrates remarkable potential for developing high-fidelity diagnostic models. To propel this field forward, subsequent research must prioritize: validation under nuclear safety constraints, development of explainable AI frameworks and solutions balancing diagnostic accuracy with operational security.