Improved design of experiments for identification of MIMO systems

A problem in the identification of MIMO systems is that the system outputs in an identification experiment may be strongly correlated if the inputs are perturbed by uncorrelated signals, as is standard practice. Such a correlation reduces identifiability.A set of methods to design input perturbations that minimize the sample correlation between the outputs has previously been proposed. These methods require an initial model for the design. In this paper, a data-based design method is proposed. Data are obtained from preliminary experiments with the system to be identified. Besides being preferable from a practical point of view, the data-based approach makes it easier to handle some numerical issues that gave problems in the model-based approach.A design method that minimizes the input or output peak value subject to desired output variances with no output correlation is presented. A model of an ill-conditioned distillation column is used to illustrate the method.