TABLE OF CONTENTS In the previous 210 pages, we have proposed a series of modelling tools that may seem somewhat disconnected from each other at first glance. Put together, however, they form a whole procedure for model-based control design consisting of the following steps.
Obtainment of a nominal model. This can be done by means of various modelling techniques. System identification using data collected on the real plant G 0 is just one possible option among many others (including nonlinear physical modelling followed by a step of linearisation, grey-box modelling using parametrised subsystems where the parameters have a physical meaning, etc.). Depending on the chosen method and, in case of prediction-error identification, on the chosen model structure, a full-order (unbiased) or a low-order (biased) model can be obtained. In case of prediction-error identification, it is best to operate in closed loop, provided the current to-be-replaced controller K id is not too different from the to-be-designed controller. Therefore, re-tuning the present PID controllers should not be neglected, even if the ultimate objective is to replace them by some optimal, possibly multivariable, controller; it may also be necessary to iterate. The guidelines of Chapter 4 must be followed in order to avoid nominal closed-loop stability issues caused by some controller singularities.
Model reduction. This is generally an optional step. If direct closed-loop prediction-error identification is used in the first step, it is indeed possible to identify a reduced-order model with an ideally tuned bias error. However,...
