Nuclear power pumps are critical equipment for the safe operation of nuclear power plants, and their fault diagnosis technologies face challenges such as adaptability to extreme environments, interference from dynamic operating conditions, and the coupling of multiple faults. Soft sensing technology provides a new approach to overcoming the limitations of traditional diagnostic methods by constructing mathematical mapping models between observable variables and target states. This paper systematically reviews the research progress in this field: traditional machine learning methods demonstrate high efficiency in small-sample scenarios but exhibit limited generalization capability; deep learning models significantly improve the accuracy of complex fault identification through end-to-end feature learning; transfer learning and hybrid strategies effectively address the challenge of cross-condition adaptability. The study also reveals current technical bottlenecks, including insufficient dynamic response capability of models, high data dependency, and a lack of interpretability. Future research should focus on the innovation of intelligent algorithms, the construction of edge-cloud collaborative validation platforms, and the formulation of industry standards, in order to promote the comprehensive implementation of soft sensing technology from theory to engineering and provide core support for the safe operation and maintenance of nuclear power systems.
Zhang et al. (Tue,) studied this question.