Outlier Detection Methods for Industrial Applications
Silvia Cateni, Valentina Colla, Marco Vannucci · InTech eBooks · 2008
A description of traditional approaches and of the most widely used methods within each category has been provided. As standard outlier detection methods fail to detect outliers in industrial data, the use of artificial intelligence techniques has also been proposed, because it presents the advantage of requiring poor or no a priori assumption on the considered data. A procedure for outlier detection in a database has been proposed which exploits a Fuzzy Inference System in order to evaluate four features for a pattern that characterize its location within the database. The system has been tested on a real industrial application, where outliers can provide indications on malfunctionings or anomalous process conditions. The presented results clearly demonstrate that the Fuzzy Logic-based method outperforms the most widely adopted the traditional methods. Future work on the FIS-based outliers detection strategy will concern the algorithm optimization in order to improve its efficiency and its on-line implementation. Moreover further tests will be performed on different applications.