The operational security of nuclear energy systems has always been a focus of attention and research. The failure of components of equipment may cause a shutdown or serious accidents, or the inaccuracy of sensors may lead to deviation from the optimal operation, which seriously affects the safety and economy of the reactor. Condition monitoring is one of effective methods to ensure the safe operation of systems. Condition monitoring, on or off-line, is a type of maintenance inspection where an operational asset is monitored and the data obtained is analyzed to detect signs of degradation, diagnose cause of faults. The purpose of monitoring is to implement fault detection and isolation. Correct diagnosis will lead to less unplanned maintenance and short downtime, which can also avoid harmful and sometimes devastating consequences and reduce economic losses.

Nuclear power systems are relatively conservative from design to operation. Compared with other industries, the development of application technology is correspondingly backward. Safety and reliability are the first priority of concern, since once an accident occurs, it will likely cause extremely serious consequences. As digitization and automation have a high development, an increasing number of emerging technologies and intelligent methods are studied in the field of nuclear power, it still has limitations in practical application though.

At present, most nuclear power DCS only provide single signal alarm and related protection functions, lack of effective high-dimensional parameter centralized monitoring function, and can not give early warning in time at the early stage of failure. Therefore, we hope to find a method that can detect the occurrence of faults in time with low computational complexity and high applicability. Through research and application verification, we finally choose the principal component analysis method to establish the condition monitoring model and verify the accuracy of the model. Combined with the modified reconstruction-based contribution analysis, the abnormal signal can be identified.