Entwurf robuster modellbasierter Fehlerisolationsfilter

Increasing demands regarding speed, accuracy, energy-efficiency, and costs drive the complexity of technical system. This comes along with a rising number of potential faults in a system creating a need for advanced fault diagnosis methods. To this end, this thesis is focused on the design of robust model-based fault isolation filters for linear time-invariant plants. These allow to reliably localize faults based on quantitative system models. First, observer-based fault isolation filters are considered. Duality to non-interacting controller design is established and systematically exploited by transferring well-known methods (Falb and Wolovich’s noninteracting controller design as well as the parametric eigenstructure assignment approach) to the problem of fault isolation. Furthermore, the class of systems, which are tractable using fault isolation observers, is extended and the methods are generalized to non-square systems with additional measurements. This gives rise to additional degrees of freedom in the design. Since robustness with respect to both exogenous disturbances and uncertain parameters is of utmost importance for practical applications, both aspects are taken into account in this thesis. Optimization-based methods are deduced with the aid of linear matrix inequalities to optimize robustness of the fault isolation observers employing the free design parameters. Lastly, fault isolation filters with a generalized structure are discussed. Compared to the observer-based approaches they enable a unifying design framework. The method is optimization-based and allows to design the filters by means of tuning only few intuitive parameters.