TOWARDS A MULTIMODELING APPROACH OF DYNAMIC SYSTEMS FOR DIAGNOSIS Marc Le Goc and Emilie Masse Laboratoire des Sciences de l'Information et des Systemes - LSIS, UMR CNRS 6168 - Paul Cezanne Aix-Marseille III University, Avenue Escadrille Normandie Niemen, Marseille, France marc.legoc@lsis.org, emilie.masse@lsis.org Keywords: Modeling, Model Based Diagnosis, Dynamic Systems, Conceptual Model Abstract: This paper presents the basis of a multimodeling methodology that uses a CommonKADS conceptual model to interpret the diagnosis knowledge with the aim of representing the system with three models: a structural model describing the relations between the components of the system, a functional model describing the relations between the values the variables of the system can take (i.e. the functions) and a behavioural model describing the states of the system and the discrete events firing the state transitions. The relation between these models is made with the notion of variable: a variable used in a function of the functional model is associated with an element of the structural model and a discrete event is defined as the affectation of a value to a variable. This methodology is presented in this paper with a toy but pedagogic problem: the technical diagnosis of a car. The motivating idea is that using the same level of abstraction that the expert can facilitate the problem solving reasoning. 1 INTRODUCTION This paper is concerned with the design of knowledge based systems to supervise, diagnose and control industrial process. The dynamic aspect of industrial processes poses the difficult problem of the acquisition and the representation of the underlying temporal knowledge which is often mixed with other types of knowledge (Basseville and al, 1996). To solve these problems, we first focus our works on the multi model based diagnosis approach (Chittaro and al, 1993) with the aim of designing models at the same level of abstraction level than the experts. Second, we want that the model formalisms to be adequate to represent the temporal knowledge coming from both from Experts and from the learning algorithms of the Stochastic Approach of (Le Goc et al, 2005). And three, we want the interpretation knowledge to closed to the cognitive tasks the models are made for and we propose to use a generic conceptual models. So, section 2 of this paper positions shortly our approach according to the main modelling approaches for diagnosis. Section 3 presents the basis of our methodology through its application to a toy but pedagogic problem: the technical diagnosis of a car. Finally, section 4 states our conclusions and perspectives. 2 MODELLING APPROACHES The limitations and the problems (Dagues, 2001) of the heuristic approach (Clancey, 1985) has motivated the Model Based Diagnosis approach (MBD) where the knowledge about the system is represented in a unique logical model (Reiter, 1987). The MBD approach use of a unique model of the system to be diagnosed containing the knowledge about both the structure (components and interconnections) and the behavior of the system (relations between the values of the input and the output of the components). This model generally comes from the design model of the system so that it contains a lot of components leading to computational difficulties for the diagnosis task (the number of potential diagnosis is exponential with the number of components). This problem is crucial and has motivate a large amount of works to reduce the size of the space search. But more, this model contains nothing about the evolution of the values of the variables over time and nothing to represent the Proceedings of the 2nd International Conference on Software and Data Technologies (ICSoft'07), 22-25 july 2007, Barcelona, Spain. 1/6