The work reported in this paper focuses on the design problem of integrated control and diagnosis in the case of time-variant systems. The developed methods take the interactions between control and diagnosis into account. The main purpose is to improve fault detection performances in a closed-loop framework by minimizing the model uncertainties effect on the residual generation. The first approach consists in choosing a controller among an a-priori set. This controller is computed by means of the optimal solution of a performance criterion, which takes the model uncertainties effects into account. The second approach proposes an augmented criterion combining control and diagnosis goals. These approaches are illustrated and compared through an example of simulation.