Search

This work presents the design of an intelligent strategy for the automatic diagnosis of processes by means of Artificial Neural Networks (ANNs) prediction and fuzzy classification. To design the diagnosis strategy, historical information of the process is used. The classification is implemented as a tool for the diffuse grouping of patterns. ANNs of multilayer configuration were trained to predict the functional states of the process. The outputs in the prediction stage are the entries of the classifier. In the proposed diagnostic scheme, the functional states are presented to operators of the processes, as future information to generate the predictive-preventive actions before the transition to a fault state. The proposed strategy was implemented on a conventional control system; and on a medicinal air production system. In this last system, the predictor of functional states presented filtration capacity in the observation of relative humidity, which is considered as an additional gain in the design, as it is a variable of difficult prediction in tropical climates. It can be concluded that the proposed methodology is able to make appropriate diagnoses of both industrial processes. [ABSTRACT FROM AUTHOR]