Determination of Timed Transitions in Identified Discrete-Event Models for Fault Detection

Model-based fault detection compares modeled and observed behavior to decide whether a system operates properly or not. The key issue in this paper is to model large-scale Discrete Event Systems (DESs) with little a-priori knowledge. For this class of systems a new approach to black-box determination of timed transitions for timed automata is proposed. The method identifies a set of time guards leading to an advantageous trade-off between the fault detection errors: false alarms and missed detections. A real industrial system is applied for evaluation of time guard determination. It is shown that applying the proposed method results in a better trade-off between the fault detection errors than using common methods (e.g. Min/Max, Normal Distribution).