Multivariable control of ill-conditioned distillation columns utilizing process knowledge

In distillation, a reliable model of the column is generally considered as a prerequisite for the design of efficient two-product control by multivariable methods. However, such models are often very hard to obtain. In fact, even very small identification errors may introduce features which are in conflict with physical knowledge, and which make the model useless. Instead of focusing on the development of consistent models, this work is concerned with the utilization of physical process knowledge directly for multivariable control, even if a reliable input-output model is lacking. Such knowledge is, for example, the sign of the RGA-values and an estimate of the input-directionality. It is shown that such structural information of the process can form an entity of control-relevant knowledge that is sufficiently rich for the design of a multivariable SVD controller.