La thèse s’inscrit dans le domaine de la modélisation et de la validation des systèmes modernes complexes. Les systèmes actuels sont en fait d’une complexité sans cesse croissante et formés de plus en plus de composants de natures différentes. Ceci rend leur processus de conception et de validation coûteux et difficile. Il semble être la simple façon permettant de faire face à cette hétérogénéité et à cette complexité est l’approche orientée composant. Suivant cette approche, le système est une entité formée par un ensemble des composants interconnectés. Les composants définissent une interface qui permet d’abstraire leur modèle interne (boîte noire), ce qui favorise la modularité et la réutilisation des composants. L’interaction entre ces composants se fait conformément à un ensemble des règles pré-établies, permettant ainsi d’avoir une vision globale de comportement du système. La conception ainsi que la validation des systèmes modernes reste alors problématique à cause de la nécessité de prendre en compte l’hétérogénéité des différents composants. Dans ce cadre, dans un premier temps, nous définirons un cadre formel générique dans lequel une large famille de formalismes de description de systèmes à base d’états peut être naturellement capturée. Ainsi, nous allons définir un ensemble de règles de composition permettant de mettre en correspondance les différents composants et ainsi de constituer un modèle global du système à concevoir. Dans un second temps, nous proposerons une approche de test d’intégration qui permet de valider le comportement d’un système complexe sous l’hypothèse que chaque composant est testé et validé. Cette approche vise à générer automatiquement des cas de test en s’appuyant sur un modèle global décrit dans notre framework du système sous test. / In spite of several decades of research, assuring the quality of software systems still represents a major and serious problem nowadays for the industry with respect to both results and costs. This thesis comes within the scope of a proposal centered on a generic unified framework for both complex software systems modeling and testing. The contribution of this paper is then twofold: first, it defines a unified framework for modelling generic components, as well as a formalization of integration rules to combine their behaviour. This is based on a coalgebraic definition of components, which is a categorical representation allowing the unification of a large family of formalisms for specifying state-based systems. Second, it studies compositional conformance testing i.e. checking whether an implementation made from correct interacting components combined with integration operators conforms to its specification

Abstract Due to increasing amount of concurrency, systems have become difficult to design and analyse. In this effort, formal verification, which means proving the correctness of a system, has turned out to be useful. Unfortunately, the application domain of the formal verification methods is often indefinite, tools are typically unavailable, and most of the techniques do not suit especially well for the verification of software systems. These are the questions addressed in the thesis. A typical approach to modelling systems and specifications is to consider them parameterised by the restrictions of the execution environment, which results in an (infinite) family of finite-state verification tasks. The thesis introduces a novel approach to the verification of such infinite specification-system families represented as labelled transition systems (LTSs). The key idea is to exploit the algebraic properties of the correctness relation. They allow the correctness of large system instances to be derived from that of smaller ones and, in the best case, an infinite family of finite-state verification tasks to be reduced to a finite one, which can then be solved using existing tools. The main contribution of the thesis is an algorithm that automates the reduction method. A specification and a system are given as parameterised LTSs and the allowed parameter values are encoded using first order logic. Parameters are sets and relations over these sets, which are typically used to denote, respectively, identities of replicated components and relationships between them. Because the number of parameters is not limited and they can be nested as well, one can express multiply parameterised systems with a parameterised substructure, which is an essential property from the viewpoint of modelling software systems. The algorithm terminates on all inputs, so its application domain is explicit in this sense. Other proposed parameterised verification methods do not have both these features. Moreover, some of the earlier results on the verification of parameterised systems are obtained as a special case of the results presented here. Finally, several natural and significant extensions to the formalism are considered, and it is shown that the problem becomes undecidable in each of the cases. Therefore, the algorithm cannot be significantly extended in any direction without simultaneously restricting some other aspect.