Model-Based Reasoning for Self-Adaptive Systems – Theory and Practice

Internal faults but also external events, or misinterpretations of sensor inputs as well as failing actuator actions make developing dependable systems a demanding task. This holds especially in the case where systems heavily interact with their environment. Even in case that the most common faults can be handled, it is very unlikely to capture all possible faults or interaction patterns at development time. As a consequence self-adaptive systems that respond to certain unexpected actions and observations at runtime are required. A pre-requisite for such system behavior is that the system itself has knowledge about itself and its objectives, which can be used for adapting its behavior autonomously. In order to provide a methodology for such systems we propose the use of model-based reasoning as foundation for adaptive systems. Besides lying out the basic principles, which allow for assurance of correctness and completeness of the reasoning results with respect to the underlying system model, we show how these techniques can be used to build self-adaptive mobile robots. In particular the proposed methodology relies on model-based diagnosis in combination with planning. We also discuss modeling issues and show how modeling paradigms influences the outcome of model-based reasoning. Moreover, we introduce some case studies of self-adaptive systems that rely on model-based reasoning concepts in order to show their applicability in practice. The case studies include mobile robots that react on hardware and software failures by applying corrective actions like restarting subsystems or reconfiguration of system parameters.