Industrial processes alarm prediction using non-supervised classification

Abstract

In this work an alarm prediction system is proposed. Its main aims are to contribute to the establishment of predictive industrial maintenance guidelines and to produce a management decision support tool. The proposed system obtains readings from many sensors that are installed in the industrial plant, extract its characteristics and evaluates the equipment’s health. The diagnosis and prognosis implies in a classification of the industrial plant’s operational condition. Classification and regression trees are applied in this paper. A measurement sample from 73 sensors installed in a hydroelectric power plant is utilized to test and validate the proposed methodology. The measurements were obtained in a 15 months period.