Detection of no-model input/output combination in transfer matrix in closed-loop MIMO systems1

A method to detect input/output (IO) combinations with no-model or poor model in the transfer matrix of a closed-loop MIMO system is proposed. Traditional approaches to IO selection are not adequate when used to detect no-model IO combination of a closed-loop identification process. The feedback effect, controller action and the characteristics of the excitation signal employed during the pre-identification stage cause this limitation. In this proposal the detection of no-model or poor model IO combinations is made based on regularity of low values of polynomial coefficients of parametric identification models. This information is gathered during the pre-identification stage. Improvement in model estimation is obtained once these “null” combinations are zeroed, before the identification process takes place. A study case involving identification of a 2 x 2 MIMO system is discussed.