Detection of Out-of-Norm Behaviors in Event-Triggered Virtual Networks

The DECOS system architecture integrates time-triggered and event-triggered control for combining the benefits of both paradigms. For applications based on event-triggered control, this architecture establishes event-triggered virtual networks as overlay networks on top of an underlying time-triggered physical network. In the scope of this paper, we show that the underlying time-triggered network significantly improves the accuracy of the error detection mechanisms in comparison to event-triggered architectures used today (e.g., based on controller area network in the automotive industry). We introduce a framework with detectors for out-of-norm behavior, which are distributed across the nodes of the distributed real-time system. The detectors produce diagnostic messages augmented with information about the location and time of the detection event. Due to the fault isolation and the global time base of the underlying time-triggered network, the additional spatial and temporal information (besides the value domain) is available in the diagnostic messages and forms a meaningful input to a subsequent analysis process. The proposed framework manages the inherently imprecise temporal specifications of event-triggered application subsystems by correlating diagnostic messages along value, space and time. Thereby, a discrimination between a correct behavior of the computer system (e.g., triggered by rare conditions in the environment) and different fault classes (e.g., design faults, physical faults) becomes feasible. Also, the paper describes an implementation of the framework based on the time-triggered protocol and explains the specification of the detectors and the analysis process using timed automata.