Monitoring Piecewise Continuous Behaviors by Refining Semi-Quantative Trackers

We present a model-based monitoring method for dynamic systems that exhibit both discrete and continuous behaviors. MIMIC [Dvorak and Kuipers, 1991] uses qualitative and semiquantitative models to monitor dynamic systems even with incomplete knowledge. Recent advances have improved the quality of semi-quantitative behavior predictions, used observations to refine static envelopes around monotonic functions, and provided a semiquantitative system identification method. Using these, we reformulate and extend MIMIC to handle discontinuous changes between models. Each hypothesis being monitored is embodied as a tracker, which uses the observation stream to refine its behavioral predictions, its underlying model, and the time uncertainty of any discontinuous transitions. keywords: model-based monitoring; model refinement, hybrid systems