Abstract

Today's software is expected to be able to work autonomously in an unpredictable environment, avoiding failure and achieving satisfactory performance. Self-adaptive systems try to cope with these challenging issues, autonomously adapting their behaviour to a dynamic environment to fulfil the objectives of their stakeholders. This implies that the software needs multiple ways to accomplish its purpose, enough knowledge of its construction, decision criteria for the selection of specific behaviours and the capability to make effective changes at runtime. The engineering of such systems is still challenging research in software engineering methods and techniques, as recently pointed out by the research community. The objective of this thesis is twofold: First, to capture and detail at design time the specific knowledge and decision criteria needed for a system to guide adaptation at run-time. Second, to create systems which are aware of their high-level requirements, by explicitly representing them as run-time objects, thus enabling it to act according to them and to monitor their satisfaction. To deliver on this aim, we provide conceptual models and process guidelines to model at design time the knowledge necessary to enable self-adaptation in a dynamic environment, extending the agent-oriented software engineering methodology Tropos. The resulting framework, called Tropos4AS, offers a detailed specification of goal achievement, of the relationships with the environment, of possible failures and recovery activities. A claim underlying the approach is that the concepts of agent, goal, and goal model, used to capture the system's requirements, should be preserved explicitly along the whole development process, from requirements analysis to the design and run-time, thus reducing the conceptual gaps between the software development phases, and providing a representation of the high-level requirements at run-time. A direct, tool-supported mapping from goal models to an implementation in a Belief-Desire-Intention agent architecture, and an operational semantics for goal model satisfaction at run-time, complement this work. The framework is evaluated through application to research case studies and through an empirical study with subjects, assessing the usability and the comprehensibility of the modelling concepts.