Model-Based Monitoring of Piecewise Continuous Behaviors using Dynamic Uncertainty Space Partitioning

Abstract

Monitoring gains importance for many technical systems such as robots, production lines or anti lock brakes. A monitoring system for technical systems must be able to deal with incomplete knowledge of the supervised system, to process noisy observations and to react within predefined time windows. This paper presents a new approach to monitoring technical systems based on imprecise models. Our approach repeatedly partitions the uncertainty space of an imprecise model and checks the derived model's state for consistency with the measurements. Inconsistent partitions are then refuted resulting in a smaller uncertainty space and a faster failure detection. This paper further focuses on the extension of our basic approach to monitoring systems that exhibit both continuous and discrete behaviors. Our monitoring system has been implemented using COTS components and has been demonstrated in online monitoring of a non-trivial heating system.