Ensuring correct self-reconfiguration in safety-critical applications by verified result checking

Abstract

The application of Organic Computing techniques, which often involves nature-inspired algorithms, to safety-critical systems depends mainly on whether behavioral guarantees can be provided or not. Self-x algorithms sometimes return incorrect results. For example, if learning or similar methods are involved. To allow the use of such algorithms even if they sometimes return incorrect results, adequate techniques are needed. Verified result checking which is presented in this paper provides an approach to ensure during runtime that only valid results are applied to the system, despite the uncertainty provided by the use of self-x algorithms. It allows to give formal proofs and thereby to give guarantees about the system's behavior. This approach enables verification at design time, independently of the self-x algorithms used at runtime. In this paper an architecture for Organic Computing systems which accommodates these concepts is presented. First the components of the architecture and integration of a result checker are described. The systematic development of a result checker based on a system model is shown for the application domain of self-organizing resource-flow systems. Further its formal verification and an implementation are roughly sketched.