Some study on identification of ill-conditioned processes for control

The identification of ill-conditioned processes is considered. For this class of processes, good estimates of individual transfer functions do not guarantee good closed-loop behaviour; it is essential to have good estimation of the very small differences between certain transfer functions, or, the so called low gain direction. A model of a high purity distillation column is used as an example. We study how different identification methods and test approaches will affect model quality for use in control. Also the influence of model structures, namely, MIMO, MISO and SISO parameterizations will be studied. Finally a simple open loop test approach will be proposed. The general conclusions are that a good test design has a stronger impact on the model quality for control than the choice of identification method and model structure; and a SISO structure is not suitable for the problem.