TABLE OF CONTENTS We have seen in Chapters 3 and 4 how the experiment carried out on the process should be designed to collect the best data for building a good model ? for control design. Here, we have seen that the same data, collected using the same rules of good practice, can also be used
before control design, to assess the quality of the identified model ? (or that of any given model G nom) with respect to the control design objective. This quality is quantified and represented by a single number, the worst-case ?- gap, or by its frequency-by-frequency counterpart, the worst-case chordal distance;
after control design, but prior to the implementation of the controller, to make sure that no significant loss of stability or performance will happen in comparison with those achieved with the nominal model.
The practical procedure consists of the following steps:
Collect data on the process, if possible in closed loop with a controller K id that is not too different from the optimal, yet to be designed, one (hence, if necessary, finely tune the possible current PID controllers before collecting the data). Use for instance a PRBS as reference signal to excite the system.
Use this data to identify an unbiased and, preferably, not overparametrised model 
) of the process, taking into account the recommendations of Chapter 4. The covariance matrix P ? of the parameters must be estimated during the identification procedure;
