Synthesis of an optimal model for robust fault diagnosis in uncertain systems

This paper presents a new robust fault diagnosis algorithm for uncertain linear systems. The model uncertainties are taken into account to synthesise an optimal model for a detection procedure. This model, optimal with respect to a H₂ criterion, is developed for both black-box and grey-box modelling frameworks. A simulation example illustrates the ability of the proposed method to generate a residual less sensitive to model uncertainties than the residual generated from the nominal transfer function. Furthermore, it turns out that grey-box identification allows to reduce the fault-free residual amplitude.