Hierarchical Data Fusion-Based Health State Assessment of Nuclear Power Plant Operation Under Human-Cyber-Physical System Architecture

Human-cyber-physical system has emerged as a pivotal tool for intelligent production throughout the whole lifecycle. In this paper, a health state assessment approach based on hierarchical data fusion is proposed under human-cyber-physical system architecture. The human-cyber-physical system emphasizes the flexibility of dynamic operating states and the evolvability of the enabling models. First of all, a multiple hierarchy consisting of equipment, systems and function is established as a physical basis for human-cyber-physical system. Regarding the equipment-level assessment characterized by extensive input parameters, a dynamic weighting method is exploited to highlight the abnormal parameters. Thus, the degraded device can be significantly reflected in the system-level assessment by fusing equipment parameters. To evaluate the operation function of a plant, a Nearest Class Mean classifiers-based Random Forests algorithm is explored considering the small-scale and imbalanced instances in different health levels. However, the conventional fixed and changeable models threaten the accuracy and rationality of the human-cyber-physical system. We introduce incremental learning to reform Random Forests that can update promptly with new data and knowledge. Finally, the proposed scheme and technologies are applied in a nuclear power plant to demonstrate their effectiveness. Note to Practitioners-The modern production desires to be further intelligent for improvement of economic benefits but with the consistent requirements of high safety and reliability. The trades-off between efficiency and safety bring the greater challenges to the operation and maintenance for production. This paper proposes an evolvable human-cyber-physical system scheme that not only integrates more intelligent elements but also attaches some solution to the abovementioned problem. The scheme is implemented on the multi-hierarchy health states assessment which is the core in production maintenance against lagging. This equipment-system-function fusion and assessment provide a unified platform for multi-departments to collaborate and self-manage. The novel weighting method abandons the conventional subjective evaluation criteria with high time consumption. Computers are convenient to possess sensing data efficiently and highlight the anomaly obviously. In particular, it is mentioned that the intelligent models in human-cyber-physical systems are always fixed and unable to alter with the physical system changes. Incremental learning, a promising branch of machine learning, is introduced to reform the assessment model to be stainable developed with new data adding. The scheme takes a nuclear plant as example because it is more corresponding to these issues. The future studies can extend the scheme to other scenarios.