Qualitative analysis for decision making in supervision of industrial continuous processes

Abstract

Supervision systems aim to provide operators with help to interpret the actual behaviour of the process in order that they can watch it over more efficiently. They are not conceived so as to replace completely the humans in charge of the plant, but rather to detect early abnormal situations and to explain them so that operators can take right and efficient decisions. At the end the operators have still to decide whether the proposed diagnosis or action is valid or not; as a consequence the supervision system must be able to justify at any time its reasonings with explanations. In our approach, the model of a process is expressed as a graph whose nodes are relevant variables and whose arcs represent the causal relations between them. It is not only a good representation of the process behaviour, but also an efficient knowledge base used by the simulation and the fault diagnosis system to deduce the effects of actions, disturbances or faults on the behaviour. The simulation and the error calculus are purely numerical treatments, and the qualitative analysis concerns diagnosis: the simulation results are qualitatively interpreted with regard to the actual behaviour, in order to decide whether the situation is normal or not. The fault detection and isolation are considered as decision-making processes in a vague context. The use of fuzzy sets theory is an adequate method to manage with the imprecision of the modelling and of the data and then to elaborate an efficient numerical-symbolic interface between low level information (data supplied by sensors) and high level concepts (normal or faulty situation).