Commissioning a distillation column simulator

Energy saving has become an extremely important issue in the chemical process industries. Distillation columns, in particular, consume huge amounts of energy. One way of minimizing the energy consumption is improved control, which enables operation closer to certain constraints.

It has been estimated that 75% of the cost associated with an advanced control project typically goes into model development (Gevers, 2005). Hence, efficient modeling and system identification techniques suited for industrial use and tailored for control design applications are needed. This task is especially difficult for “ill-conditioned” MIMO systems such as distillation columns. Even for a linear system, this ill-conditioning makes the system behavior resemble that of a (strongly) nonlinear system. Because of this, identification, modeling and control of ill-conditioned systems are demanding tasks.

In an industrial environment, the identification usually has to be carried out while the plant is in normal operation. It is then essential to keep the variation of inputs and outputs within specified limits and to limit the duration of the identification experiments. However, this also limits the information available for system identification. Thus, there is a trade-off between how much one is prepared to “pay” for the information and the information needed for system identification.

These kinds of identification issues can be investigated by means of a pilot-scale distillation column at Åbo Akademi. However, in order to enable more effective identification studies, a distillation column simulator has been constructed using MathWork’s Simulink as programming environment. Because the simulator is to be used in conjunction with the real distillation column, it is desired that the behavior of the simulator is close to that of the real column. A significant number of previously performed identification experiments with the distillation column are available for the tuning of the simulator.

In order facilitate the simulator tuning, we study the effects of column and mixture property parameters appearing in the simulator model on observable static and dynamic properties. In particular, we want to find out which parameters most strongly affect these properties in various input-output relationships. Besides the tuning issue, this information also has more general interest.