Functional State Estimation Methodology Based on Fuzzy Clustering for Complex Process Monitoring

In order to prevent faults, many methodologies have been proposed for process monitoring. When it is difficult to obtain a classical model, the use of fuzzy clustering techniques allow the identification of classes that can be associated to the process functional states (normal, alarms, faults). This paper presents a methodology for predicting functional states in order to prevent critical situations. Using the process historical data and combining fuzzy clustering techniques with Markov's chains theory a fuzzy transition matrix is constructed. This matrix called WFT is used later online in order to predict the next functional state of the monitored process. The methodology was tested in a boiler subsystem, and a water treatment plant.