Making the investigation of huge data archives possible in an industrial context an intuitive way of finding non-typical patterns in a time series haystack

Modern nuclear power plants are equipped with a vast variety of sensors and measurement devices. Vibrations, temperatures, pressures, flow rates are just the tip of the iceberg representing the huge database composed of the recorded measurements. However, only storing the data is of no value to the information-centric society and the real value lies in the ability to properly utilize the gathered data. In this paper, we propose a knowledge discovery process designed to identify non-typical or anomalous patterns in time series data. The foundations of all the data mining tasks employed in this discovery process are based on the construction of a proper definition of non-typical pattern. Building on this definition, the proposed approach develops and implements techniques for identifying, labelling and comparing the sub-sections of the time series data that are of interest for the study. Extensive evaluations on artificial data show the effectiveness and intuitiveness of the proposed knowledge discovery process.