TABLE OF CONTENTS Modern optimal control design methods are generally based on models. Most of the time, an identification step will have to be performed on the process in order to compute such a model and we have seen in Chapters 3 and 4 how to realise the identification in order to maximise our chances of finding a good model for control design. In this chapter, we elaborate on the question of model validation for control. This question concerns the verification of the quality of a given model; the way this model was obtained is irrelevant, so this verification procedure can be applied not only to models obtained by identification, but also to any (LTI) model that the control designer could have at their disposal. The question under consideration is thus:
Can I trustfully use the model that was given to me for control design? Is it close enough to the real process? How can I quantify the error (the distance) between the real process and this model in a way that is representative of the suitability of the model for control design?
If the practitioner is satisfied with the model, he or she can use it for control design. Most of the time, good nominal stability margins and performance will be sought, with the hope that they will remain after implementation of the controller on the actual process. But hope and faith are not sufficient to guarantee closed-loop robustness,...
