Abstract

Multidimensional sensors can deliver vast and rich information about the industrial processes operation. At industrial level, they are becoming widely available for supervising tasks. However, their use at control level is not very widespread, since there are no standard methodologies for including the information provided by this type of sensors into existing control systems. This paper describes the traditional approach to include multidimensional information into conventional control systems, and proposes a new structure based on pattern recognition. The latter makes use of artificial neural networks and finite state machines as a framework for designing the control system. The main characteristics and limitations of both approaches are illustrated by the image based control of an experimental fluidization bed. © 2003 Elsevier Ltd. All rights reserved.