Increasing the reliability of a self-optimizing railway guidance system

Recent advances in information processing enable new kinds of technical systems, called self-optimizing systems. These systems are able to adapt their objectives and their behavior according to the current situation and influences autonomously. This behavior adaptation is non-deterministic and hence self-optimization is a risk to the system, e.g. if the result of the self-optimization process does not match the suddenly changed situation. In contrary, self-optimization could be used to increase the dependability by pursuing objectives like reliability and availability. In our preceding publications we introduced the so called multi-level dependability concept to cope with this new kind of systems (cf. [6]). This concept comprises the monitoring of the system behavior, the classification of the current situation, and the selection of the appropriate measure, if reliability limits are exceeded. In this paper we present for the first time experimental results. The dependability concept is implemented in the selfoptimizing active guidance system of a railway vehicle. The test drives illustrate clearly that the proposed concept is able to cope with, e.g., sensor failures, and is able to increase the reliability and availability of the active guidance module.