Pattern recognition in multivariate time series - A case study applied to fault detection in a gas turbine

摘要: Advances in information technology, together with the evolution of systems control, automation and instrumentation have enabled recovery, storage manipulation a large amount data from industrial plants. This development has motivated advancement research fault detection, especially based on process history data. Although work been conducted recent years diagnostics gas turbines, few them present use clustering approaches applied to multivariate time series, adopting PCA similarity factor (SPCA) order detect and/or prevent failures. paper presents comprehensive method for pattern recognition associated prediction turbines using series mining techniques. Algorithms comprising appropriate metrics, subsequence matching fuzzy were extracted Plant Information Management System (PIMS) represented by series. A real case study detection turbine was investigated. The results suggest existence safe way start that can be useful support dynamic system monitoring predicting probability failure decision-making at operational level. Real turbine.Comprehensive turbines.Results show efficiency proposed approach level.The whole three step presented is flexible portable.An extended version FCM algorithm suitable applied.