Robust residual generation for diagnosis including a reference model for residual behavior

Abstract

The main goal when synthesizing robust residual generators for diagnosis and supervision is to attenuate influence from model uncertainty on the residual while keeping fault detection performance. Without considering structural constraints of the model, it is easy to form unrealistic performance demands, and it is shown by examples how this can lead to a design with unnecessary poor robustness properties. Guided by the examples, a new design algorithm for robust residual generators is developed. A key step is the design of a reference model incorporating structural properties, and it is shown that this leads to a consistent optimization problem avoiding unrealistic demands. The available design freedom in the developed method is explicit and intuitive, and the whole algorithm fits into the framework of standard robust H ∞ -filtering relying on established and efficient methods. The designer of a diagnosis system is thus provided with a method where it is easy to specify desired behavior without violating structural requirements.