Modelling of field experience data

Abstract

Nowadays sensors networks produce tons of time-series data. Data mining techniques are potentially able to extract knowledge from data. However, sensors network time-series require special treatment in its analysis due to such problems as large volumes of data, noise, incompleteness, and so on. The main idea of this thesis is implementation of data mining techniques in the domain of steam and gas turbine operational data. The goal is to find patterns which could indicate a higher wearing out of the turbine components. Thus, we conduct the whole data mining procedure from extraction of data to interpretation of the results and proposal on the future work directions. We perform preprocessing of time-series, segmentation, clustering and transformation of data to find indicative patterns. We get two new features which potentially can contribute to scrap rate prediction and survival analysis models. We assume that this approach can be extended and used in other levels of research.