Model based fault detection, identification and compensation in industrial processes

In the project of control systems it is usually assumed that the instruments exhibit unlimited levels of reliability and so its functioning never deviates from ideal conditions, i.e, the sensors and actuators never fail.  
The main goal of these control systems, currently designated by fault tolerant, is to keep the process under acceptable limits, minimising then the risks resulting from eventual faults, in order to protect people, equipments and the environment. Therefore, the control systems should be able to detect and identify in due time, the process faults, and, in those circumstances, to undertake the command and promote the correct actions needed to reduce its impact [Afonso 1998].  
The implementation of a fuzzy logic structure of this type demands the strict knowledge of the process. On the other hand, the complexity and the lack of adequate mathematical tools for modelling have been motivating the development of sophisticated methodologies, in which the accuracy and readability of the models assume great importance, mainly in the operation of real systems, as in the present project.    
The fuzzy logic belongs to this category, producing the models that have the capacity to integrate the expert knowledge with experimental data. The models are described as a set of linguistic rules of the IF…THEN type. The readability of these models depends on the structure and on the number of the rules used in the description of the system under study.
A new concept of relevance of a set of rules was recently proposed [Salgado, 1999], that allows the settlement of their relative importance. This concept is defined through a set of intuitive axioms that lead to a set of properties that define the conditions that a general relevance function must satisfy. In parallel, a new methodology has been proposed to organize the information that describes a system, designated by Separation of Linguistic Information Methodology (SLIM) [Salgado 1999].
The application of this methodology has a real impact in the reduction (10 to 20%) of the number of rules, simplifying the process model and its application, which constitutes a very promising result [Salgado et. al., 2001]. It is the main goal of the project to develop a system of fault compensation in real processes, exploring the referred methodology in a pilot plant existent in the Chemical Engineering Department of the Faculty of Sciences and Technology of the University of Coimbra that will be used as a test bed. In this plant, where there are several research projects going on presently, a fault compensation strategy using classical approaches has already been implemented with success. The research team responsible for the present project believes that these new concepts have a real field of application and will have a significant impact in this area of knowledge.