Synchronization between run-time and design-time view of context-aware automotive system architectures

In contrast to current automotive system architectures, future architectures will continuously gain knowledge at run-time with the help of machine learning techniques. For making this knowledge visible to the developer, synchronization mechanisms between run-time systems and design-time models have to be introduced. In this paper, we propose a framework for continuously updating design-time models with learned knowledge from the run-time system. The core of our framework is a system architecture with a gateway which retrieves data from different car functions. On this gateway, feature selection algorithms are implemented in order to select a data subset that describes the nominal behavior for the usage of car functions. Considering the resulting nominal model as baseline, the run-time system is able to detect violations of the nominal behavior. For assessing these violations from the perspective of a developer, we connect the run-time system with the design-time models by means of a semi-automatic feedback loop. For the evaluation, we test our approach on an exemplary scenario based on the function of the window lever by using real car data.