Robust State Estimation of Complex Systems

To ensure product quality and guarantee the reliability of technical systems and processes, one needs access to critical information about dynamic system events. However, it is usually impossible to directly monitor all relevant system quantities, due to the technical limitations of available sensor technology and the restricted presence of suitable measurement sites. Moreover, supplying the number of sensors required to directly measure all quantities would often prove too expensive. One way out of this dilemma is to use model-based state estimations. At the Fraunhofer ITWM, we use robust methods from the H ∞ $H_{\infty}$ theory and particle filter methods, along with the traditionally applied Kalman filters, to accomplish this task. Two challenges associated with these methods are modeling the process- and measurement uncertainties and achieving real-time capability for the underlying system simulation. Here, we illustrate how state estimators are used at the Fraunhofer ITWM in the following applications: run-out compensation in connection with contact-free torque observations; online monitoring of torsional vibrations in power plant turbo generator sets; and—from the biological/medical field—analysis of plasma-leucine measurements with particle filter methods in a study involving diabetes patients.