Robust partial fault isolation for linear systems using observers

This paper deals with observer-based fault isolation in linear systems. Only a single observer is implemented to achieve fault isolation, which results in a small implementation effort. If it is known that not all potential faults can occur at the same time, it suffices to achieve partial instead of full fault isolation. In this paper we show how existing methods for full fault isolation can be extended to partial fault isolation based on a parametric eigenstructure assignment approach and elaborate on the arising degrees of freedom. Furthermore we show that under a certain condition there is a simple (bi-)linear relationship between the observer parameterizations for full and partial fault isolation. This gives rise to an efficient algorithm optimizing robustness of partial fault isolation observers with respect to disturbances. The method is successfully applied in both simulations and a lab experiment with a gantry crane.