Online monitoring of dynamic systems is a major prerequisite to enable continuous operability. In model based monitoring, this is achieved by comparing the partially observable system dynamics with a prediction using a-priori knowledge of the plant and its uncertainties. This paper presents a novel approach where monitoring is done by incrementally propagating bounded uncertainties of the inaccurately known plant dynamics as well as unbounded uncertainties arising from stochastic disturbances. Partial updates of the propagated covariances are done utilizing measurements to prevent divergence of the prediction. The presented approach allows continuous monitoring of the compliance with probabilistic system requirements and hence is able to increase safety and availability of the monitored system. Such algorithms are especially of importance for the highly emerging market of small to mid-size UAVs, where the need for smaller and cheaper and hence often less accurate and reliable components meets legal probabilistic safety requirements. Evaluations of the implemented algorithm by monitoring of a hypothetical plant as well as the complex task of monitoring relative position control performance of two aircraft flying in close formation showed excellent performance and demonstrated the wide applicability of the presented algorithm.     «

Online monitoring of dynamic systems is a major prerequisite to enable continuous operability. In model based monitoring, this is achieved by comparing the partially observable system dynamics with a prediction using a-priori knowledge of the plant and its uncertainties. This paper presents a novel approach where monitoring is done by incrementally propagating bounded uncertainties of the inaccurately known plant dynamics as well as unbounded uncertainties arising from stochastic disturbances. P...     »