Fuzzy rule-based classifier for Fault Prediction in a Thermoelectric Unit

Abstract

Pattern recognition from data is a potential alternative for the extraction of knowledge about processes and it may be useful for predicting failures, control and support decision making, among others. The knowledge extracted can be used to implement models based on Artificial Intelligence such as Fuzzy Inference Systems (FIS). Tools from Information Technology (IT) and automation techniques can also be used in data-based approaches to enable the storage and handling of large amounts of historical process data. This paper presents the implementation of a fuzzy inference system for fault prediction in a gas turbine of a thermoelectric unit. The first step comprised the pattern recognition through the clustering of multivariate time series obtained from the Plant Information Management System (PIMS). The second step comprised the development of a FIS using a data-based approach to define the membership functions and rules. The results showed the potential of the fuzzy model to predict the probability of failure during the start of the turbine this presenting a feasible alternative to support decision-making at operational level.

Keywords

• fuzzy inference systems
• pattern recognition
• multivariate time series
• fault prediction.