Diagnostics and Surveillance Methods in Nuclear Systems for Real-Time Applications

This paper reviews some process signal analysis and representation methods that can be used during reactor operation such that they are suitable for real-time applicatons. All listed methods have been tested on data from operating plant. The objective is to detect and interpret changes in the plant or core status at an early stage, such that appropriate measures can be taken immediately. The methods that are discussed and demonstrated in the paper can be divided into two categories. The first is the use of fast and intelligent computing methods such as neural networks and fast wavelet transform, in combination with a diagnostic unfolding procedure which would be computationally rather demanding with traditional methods. The second type is based on direct representation of the system state through visualization of large complex data, showing the space–time behavior of the system. This latter is not associated with any unfolding procedure, it uses only a moderate signal preprocessing for filtering out redundant information, but otherwise showing the process status directly. Such methods have been made possible with the development of powerful computer visualization techniques. The potentials represented by this second alternative do not seem to have been explored fully yet in reactor diagnostics. Methods corresponding to both categories will be demonstrated and discussed in the paper.