An advanced statistical method to analyze condition monitoring data collected from nuclear plant systems

Condition monitoring data are routinely collected from various nuclear plant systems to ensure they are operating within an acceptable envelope, and to detect any potential onset of degradation in the system condition. The condition monitoring includes periodic monitoring of not only physical variables, such as temperature and vibration, but also chemical properties of lubricants, oils, and other control fluids. The time series of such monitoring data tend to exhibit non-stationary nature and complex correlation structure, as they consist of fluctuations of different time scales and noise. Since standard text-book methods of stationary time series analysis are not applicable to such data sets, the paper presents an advanced method of Empirical Mode Decomposition (EMD) to filter out the noise and identify the long-term trend, i.e., a likely indicator of degradation, in condition monitoring data. The proposed method is verified by a simulation example and then applied to a real data set obtained from an operating nuclear plant.