System Drifts Monitoring using a Neural Supervision System

H.A. Boubacar and S. Lecoeuche (France)

Keywords

: AutoAdaptive Neural Network, UnsupervisedClassification, Non stationary Data, Industrial Process, OnlineSupervision.

Abstract

From our previous works, a specific supervision system, that consists of three stages : modelization, monitoring and diagnostic, has been proposed. The heart of this system is to use an on-line modelization of the functioning modes of the system to be monitored. This task is achieved with the help of the AUDyC neural network especially developed to classify non-stationary data distribution. According to the system behaviour changes, current functioning modes could evolve in various ways. To detect these system deviations, new tools have been developed to analyze online some parameters of the models of functioning modes. The main novelty of our approach consists in the description of new online monitoring techniques based on the AUDyC modelisation. Two new techniques detailed in this paper, are dedicated to the detection of both abrupt changes and slow degradations of a system. At the end, some experiments achieved on a water circulating temperature controller (WCTC) show the performances of monitoring tools.

Important Links:



Go Back