Hierarchical Modelling and Diagnosis for Embedded Systems Herv´ e Ressencourt 1,2 Louise Trav´ e-Massuy` es 1 J´ erˆ ome Thomas 2 1 LAAS-CNRS 2 ACTIA 7, Avenue du Colonel Roche 25, Chemin de Pouvourville FRANCE-31077 TOULOUSE Cedex 4 FRANCE-31432 TOULOUSE Cedex 4 hressenc,louise@laas.fr jerome.thomas@actia.fr Abstract Because of the increasing complexity of engineered systems, abstractions and hierarchies in models are receiving great attention. The behaviour of embed- ded systems is commonly characterised by hybrid phenomena in which each operational mode is acti- vated by electronic units: it hence involves hard- ware and software components. The aim of this work is to apply a multimodelling approach on such systems for the diagnosis task. This is illustrated by an example taken from the automotive domain. 1 Introduction The increasing complexity of engineered systems led the Model-Based Reasoning (MBR) communities to focus their research in reasoning tasks - like diagnosis - based on mul- tiple abstraction level models organised through a hierarchy. Abstractions are useful to reduce the computational complex- ity of diagnosis reasoning, to account for observations at qual- itative levels, and to handle systems whose available knowl- edge about components is heterogeneous. Two kinds of hierarchies are commonly used in MBR: structural abstraction [Chittaro and Ranon, 2004] [Mozetic, 1991] [Autio and Reiter, 1998], which aggregates compo- nents to describe the system at different levels of detail and functional abstraction, which abstracts the behaviour accord- ing to the functional and teleological understanding of the system [Chittaro et al., 1993] [Kitamura et al., 2002]. The main idea of a functional description is to bridge from be- havioural to teleological knowledge (knowledge about goals) by exhibiting the functional roles that the structural compo- nents may play in the achievement of the function of the whole system. The objective of this work is to devise a multimodelling cooperation framework for the diagnosis of complex embed- ded hybrid systems controlled by electronic units. A brief overview of existing approaches on functional modelling is first proposed in section 2. In the third section the limits of these approaches are discussed and an extention of Chittaro’s framework [Chittaro et al., 1993] is proposed to model hy- brid physical systems including hardware and software com- ponents. Then, some perspectives are presented for the off- board diagnosis task of automotive systems based on this framework. 2 Knowledge representation for Model Based Reasoning Knowledge representation is a key issue in MBR. Luca Chit- taro and colleagues [Chittaro et al., 1993] write that choices have to be made, especially about ontologies, epistemological types, representational assumptions and aggregation levels. These choices are mainly directed by the goals of the mod- els (design analysis, diagnosis...) and by the requirements of the reasoning task. It is commonly accepted that knowledge about physical systems can be organised through two axes [Lind, 1982]: • The Whole-Part hierarchy relies on different aggrega- tion levels for a same type of knowledge. An entity of this hierarchy is a part of the upper one. For example, structural abstraction has been used for the diagnosis task [Chittaro and Ranon, 2004] [Mozetic, 1991] [Au- tio and Reiter, 1998]. • The Mean-End or functional hierarchy relies on the theological understanding of behavior [Chittaro et al., 1993] [Kitamura et al., 2002]. A functional description hierarchy has to answer three questions: ”Why was the system designed?”, ”What is the system supposed to do to achieve the goal?” and ”How must different parts of the system interact in order to realise the functions?” [Modarres and Chehon, 1999]. 2.1 Functional abstraction hierarchy Several works agree on a model hierarchy consisting in a dis- tinction between four epistemological types : • The Structural knowledge is the knowledge about sys- tem topology. • The Behavioural knowledge describes the physical laws underlying the behaviour of components composing the system. • The Functional knowledge describes the roles compo- nents may play in the process in which they take part. This level is named Base-Function layer in [Kitamura et al., 2002].