Distillation column dual temperature control

1 dual temperature control , January 2010, Page 0 Petrocontrol Leader in inferential control technology Distillation column dual temperature control By Y. Zak Friedman, PhD Principal Consultant December 2009 dual temperature control , January 2010, Page 1 Distillation column dual temperature control This editorial takes issue with a claim by Mr. Greg Shinskey that Distillation dual composition control cannot be done by MPC (multi-variable predictive controller), but can be done by ARC (advanced regulatory controllers) effectively [1]. In fact the argument is not about MPC versus ARC but about how to set up a workable basic DCS control structure. Mr. Shinskey s starting point is the dual temperature control structure shown in figure 1, with two temperature controllers, one in the rectifying section, manipulating reflux, the other in the stripping section, manipulating the reboiler.

2 Mr. Shinskey asserts, and I agree, that manipulation of reflux and boil-up has almost the same effect on both temperatures . In other words, increasing steam flow will raise the top temperature almost as much as it raises the bottom temperature , and increasing reflux flow will lower both by similar no amount of clever setpoint adjustment by the MPC can break through that tight coupling . The real issue is not what manipulates the dual temperature setpoints, but stability of the basic DCS configuration of figure 1. Regardless of what drives the temperature setpoints, MPC or ARC, the basic DCS must provide stable dual temperature control before any APC can begin. Having said that this problem is next to impossible to handle, Mr. Shinskey goes through a relative gain analysis to show that some columns can be stable as configured in figure 1, whereas other, high purity columns must incorporate the mass balance structure of figure 2, with drum level control on the reflux and top temperature control on the top product.

3 Why is it then that in my travels I have seen neither the structure of figure 1 nor figure 2 working? It is because of what Mr. Shinskey so eloquently said about tight coupling. Even if one can manage to tune these temperature loops under one condition, they go unstable the next day under different conditions, and operators quickly turn one of the temperature controllers to Manual. And there is no point squeezing too much mileage out of relative gain analysis. Relative gain analysis is a steady state technique, whereas the complexity of this problem is dynamic. High purity Distillation is not normally controlled by a high up tray temperature because the temperature profile is flat. The meaningful tray temperature would have to be located many trays below the top, which renders the dynamics between product draw and tray temperature slow and complex, on top of the already disruptive interactions.

4 I would challenge Mr. Shinskey to write a paper showing a real column having stable dual temperature control , and showing how such an interactive system can be tuned. If you accept that dual temperature control is not the ideal basic structure - how then should we configure DCS for Distillation ? In my experience, either with heat balance or mass balance, only one of the tray temperature controllers can work. Many considerations affect the selection of rectifying versus stripping tray. Which of the products is more important, Which of the temperatures is a better inference of product purity Which of the loops is dynamically simpler to handle dual temperature control , January 2010, Page 2 It goes without saying that this temperature controller must ultimately affect the yield, and hence it isn t desirable to choose a stripping tray temperature for a mass balance control structure.

5 Connecting tray temperature controller to manipulate distillate draw on a mass balance DCS structure is feasible, but loop dynamics becomes slow and complex. Say we have selected the stripping tray temperature on a heat balanced column , and that is a simplified inference of bottom purity. What should we do to control top purity? The simplest way is to set the reflux to feed ratio to a reasonable value known to give good results. I prefer reflux to feed ratio over reflux to product because the latter introduces a mass balance component into a heat balance structure. That gives us the basic DCS structure of figure 3, and now we can discuss APC, and whether MPC or ARC is to be employed. Contrary to Mr. Shinskey statements, it is rather rare in refining to find both top and bottom analyzers on a single column .

6 Perhaps the petrochemical industry is somewhat better in that regard. Not counting main fractionators, I would say that fewer than 10% of refinery Distillation columns have two analyzers. If we consider only reliable analyzers then the number goes down below 5%. APC has to rely on inferential models, and that is not bad because inferential models do not have the dead time of analyzers, and inferential models, if done right, have fewer reliability problems than analyzers. If you are lucky enough to have analyzer as well as inference model, both reliable, the ideal structure is to connect the inference as the primary controller, while setting up a slow inferential bias to reduce the difference between inference and analyzer. That is an ideal arrangement because it takes advantage of both inference dynamics and the slower but more accurate analyzer.

7 And what if two inference models are available, can we attempt control of both top and bottom composition? Being in the inferential modeling business I often come up with both top and bottom models that rely on rectifying and stripping tray temperatures . Connecting both models in closed loop is just as problematic as dual temperature control . My approach for avoiding interaction between the two loops is to use one of the models for manipulating yield, for example connecting it to the reboiler steam. The other, say top model, I would rework as a function of bottom purity and internal reflux, and use that model for setting reflux ratio. Now we have APC design that overcomes the top / bottom interactions, and would work either in MPC or ARC format. Lastly, where do I stand in this argument about whether ARC works better or worse than MPC?

8 In the seventies MPCs were not available and we implemented all APC using ARC plus custom logic. When implemented by a competent engineer it worked well. By and large we continued this way into the eighties. MPCs were already there but the early versions were no better than ARC. MPCs improved only when they started inverting the dynamic control matrix every minute. I have continued to be involved with ARC through inferential modeling. I often supply models in open loop. Then once the inference works there s a push to close the loop, and the cheapest way to do that is ARC. I have thus implemented ARC applications with fairly high complexity: crude units, alkylation unit FCC and a host of smaller Distillation columns . My judgment is that under multiple constraints MPCs work somewhat better, but the most important factor is the engineer.

9 The difference between competent implementation versus mediocre one is much greater than the difference between MPC dual temperature control , January 2010, Page 3 and ARC. Also, both formats require maintenance, and a second important consideration is the availability of local skill to perform MPC or ARC maintenance work. REFERENCES 1. Shinskey, F. G., Multi-variable control of Distillation , parts 1, 2 and 3, ControlGlobal, May, June, July 2009. dual temperature control , January 2010, Page 4 Figure 1. dual temperature control on a heat balance control structure Tray 25 Tray 30 LC FC LC TC PI FC FC TC FC PC LC Tray 6 FC TI TI dual temperature control , January 2010, Page 5 Figure 2. dual temperature control on a mass balance control structure Tray 25 Tray 30 LC FC LC TC PI FC FC TC FC PC LC Tray 6 FC TI TI dual temperature control , January 2010, Page 6 Figure 3.

10 Single temperature control on a heat balance control structure Tray 25 Tray 30 LC FC LC TC PI FC FC TI FC PC LC Ratio FC TI TI RC