Modelling for intelligent fault diagnosis-integrating numerical and qualitative techniques

Successful implementation of intelligent systems for supervision and fault diagnosis of complex industrial plant, depends on an adequate representation of the fundamental mechanisms governing plant behaviour. The search for suitable representation tools has produced a number of methods, including the equation based models familiar to control engineers, as well as qualitative models from the field of artificial intelligence. Equation based modelling techniques are aimed at representing dynamic characteristics accurately, whereas qualitative models seek to reproduce the grasp of fundamentals and the overview of the system that engineers and operators possess. The paper presents a modelling framework that draws upon both philosophies. The basic ideas of a qualitative modelling paradigm, qualitative transfer functions (Feray 89), are adapted to include numerical input-output models in the describing relationships between system variables. This methodology has been developed with the view to providing intelligent supervision of a distillation process