Preserving Feed Composition Effects Downstream of An Open Level in a DMCplus Controller

The FS-1 Fractionator is a natural gas fractionation unit which separates natural gas liquids into ethane, propane, butane and C5P. The process consists of three distillation towers in series: Deethanizer (DC2), Depropanizer (DC3) and Debutanizer (DC4). The DC3 is fed by three different streams: the DC2 bottoms and two feed pipelines, all characterized by having fast and frequent changes to both flow and composition. The FS-1 Fractionator does not have significant feedstock or product storage, and plant feed rates are closely coupled to product flow rates to product pipelines. The product compositions and the product flows must be controlled to specified targets. The capability of meeting the production targets depends strongly on the compositions of the feed streams and the ability of the DMC controller to manipulate the ratio between the various feed sources.

A new DMC controller was commissioned for just the DC3 and DC4 towers. A basic design consideration was whether to open the DC3 level between DC3 and DC4 towers. The advantage is that the delay and oscillation created by the level PID loop is removed and sharper, clearer models can be identified for the downstream DC4 C's. However, opening the level breaks the key relationship between the DC4 CVs and the feed compositions. The DC4 butane product flow would have models with the DC3 bottom flow, not the three different feed flows and their compositions.

The project team overcame this trade-off by opening the DC3 level, and re-building the feed composition effects around the open level. At steady-state, the problem can be treated as a mass/component balance. All propane that is added to the column must be removed as propane product or as an impurity in the butane product and so on. Steady state predictor models are created between the feed flows and compositions, with unity models to the product flows (ignore the effect of impurities). The DC3 bottom flow has a dynamic-only model to the butane product flow. The models are identified from open-loop data, but assembled in the way that we know they will perform closed loop. The DC3 bottom flow is dynamically independent, but the need to control the DC3 bottom level within constraints means that it is not truly independent at steady state. We must account for the dynamic behaviour of the DC3 bottom flow, but we can connect the feedstock flows directly to the product flows at steady state, because the DC3 bottom level must remain in control.