Knowledge-based dynamic reconfiguration for embedded real-time systems

Abstract

Innovations in many domains like transportation, industrial systems, home automation, healthcare and consumer electronics are driven by embedded computer systems in order to attain an unprecedented quality of control in physical processes. While this enables new services as well as the enhancement of existing capabilities based on changing user demands and technological advancements, these systems increasingly depend on the correct operation of electronic devices. Therefore, fault-tolerance mechanisms must be introduced to ensure a continued provision of services even in the presence of the failure of individual components. These fault-tolerance mechanisms are based on a fault hypothesis and exploit redundancy in the system to detect and mitigate faults and their effects. Typically, the system designers explicitly introduce redundancy at design time by the replication of components or computations, which increases the production cost, energy consumption, weight and size of the system. Alternatively, implicit redundancy can be exploited, which is available as a priori knowledge about the correct system behaviour and the relationship between system properties. However, it requires high engineering effort to identify implicit redundancy, and its availability varies during the runtime of dynamically changing systems. Within this thesis a dynamic reconfiguration framework for embedded real-time systems is presented that automatically identifies redundancy and adapts the configuration of components and their interactions accordingly. This allows to react at runtime to changes in the system or its environment and to recover from service failures, including unanticipated failures that are not covered in the fault hypothesis. State-of-the-art solutions are not capable to provide temporal guarantees for reconfiguration. Also, they do not consider the semantics and accuracy of processed information. The framework provides an architecture for systems with dynamic reconfiguration capability and a modelling language for a knowledge base that describes the semantic relationship of the system-s building blocks and its properties. Algorithms have been developed that search for redundant information in the knowledge base and to determine the information uncertainty. Formal analyses and experiments show the real-time capability of the framework and the effectiveness of semantic matching as well as the determination of uncertainty of information. An evaluation of the probability to find substitutes for failed services demonstrates the increase of reliability when dynamic reconfiguration is used as a never-give-up strategy for the system.