Reliable and Real-Time Anomaly Detection for Safety-Relevant Systems

Abstract:
Safety-relevant embedded systems, e.g. in automotive applications, often require redundancy and monitors for anomaly or error detection. This paper presents an approach that permits to detect deviations of a deployed system from the possible behavior of a model. In order to satisfy realtime requirements, we use reachability analysis and represent results by a novel data type Affine Arithmetic Cartesian Decision Diagrams (AACDD). The benefits are demonstrated by the analysis of the comparison with a One Class Support Vector Machine approach on the example of a Σ − Δ modulator.