(245m) Dynamic Simulation of a Cryogenic Air Separation Unit

For the production of large quantity of air gases in industry, cryogenic distillation is the most economically viable process. In such a process, air is liquefied and separated in a series of distillation columns, namely, high pressure column, low pressure column, crude argon column, and pure argon column, into three pure components, oxygen, nitrogen, and argon. Over more than 100 years of process improvements, the design of such an Air Separation Unit (ASU) is typically a highly integrated process. Many of the product and intermediate streams are used as heating or cooling agents of the distillation columns, and the temperature difference is often just a few degrees. For example, the reboiler of the low pressure column and the condenser of the high pressure column are integrated together, and the liquid effluent from the high pressure column is used as the cooling agent in the condenser of the crude argon column.

For such a highly integrated process, although the steady state design has been optimized for many years, the corresponding dynamic behavior is much less studied due to its complexity, even though it is very important in operating such an ASU in a non-steady state manner. This has become increasingly more important in recent years with deregulated energy market. Because an ASU is a high energy consumer, it is desirable that it can be ramped according to the fluctuating electricity price while the product specifications are still preserved.

In our effort, a dynamic model is set up for a full ASU with first principles, with the parameter values taken from an industrial air separation plant. By solving such a model in an equation orientated simulator, the study of different dynamic scenarios can be carried out, which can be used for operation improvements, validation of new control schemes, including advanced control, controller tuning, and operator training. In this presentation, the model equations of different modules in such an ASU will be discussed, as well as the strategies used to converge such a large scale model. Simulation results will also be presented.