Re: Gain Consistency

[ Follow Ups ] [ Post Followup ] [ Control Consulting ] [ FAQ ]

Posted by Don Snowden on September 21, 2000 at 11:51:39:

In Reply to: Gain Consistency posted by Graeme Taylor on September 19, 2000 at 17:18:23:

This is a common problem with debutanizers (“stabilizers” as well). Many people have experienced the problem you’ve outlined, but few appreciate the significance of the issue (as you have). I’ve worked on several debutanizer applications where I have had to force the gains on the top and bottom “specs” to have EXACTLY the same ratio. You usually don’t have to change the curves more than 5-10% to make this happen. By doing this you force the “near-singularity” to be exactly singular. Thus the controller will not bounce between limits for small disturbances. You then set up the LP to drive the unit to the direction you want (e.g. maximum separation).

The other debutanizers I’ve worked on probably would have demonstrated similar symptoms but they hit a constraint (e.g. flooding limit) before you got into this “region”.

I guess it is an issue of “controllability”. I’ve heard of people spending a lot of time trying to study (e.g. Masters thesis work) this “behavior” (e.g. trying to control top and bottoms specs independently) in debutanizers. It simply boils down to you can’t control both ends of the tower independent after a certain point. For example, look at the following example:

Reflux , Control Temperature , Upper tray Temperature
6 , 300 , 150
6.5 , 300 , 148
7 , 300 , 147
7.5 , 300 , 147
8.0 , 300 , 147.1
8.5 , 300 , 147.2

You get the idea. Once you get to a certain point, then you can’t control both ends. The above data is only representative (not exact numbers) of offline steady-state simulation data. Thus it is not sensitive to any flooding constraints (i.e. it is a real phenomena). Also, if I remember correctly it is not like the top purity was super pure. It may have been like 2% C5s in overhead.

In any case, I think you have come-up with the proper solution. Make the gains exactly singular (down to as many significant digits as you can). There is an Aspentech program available that runs the model through a singular value decomposition that can help pick-up near singular models.

Follow Ups:

Re: Gain Consistency Graeme Taylor 15:32:25 9/28/00 (1)

Re: Gain Consistency Don Snowden 16:26:32 10/02/00 (0)

Post a Followup

Name    :

E-Mail  :

Subject :

Comments:

: This is a common problem with debutanizers (“stabilizers” as well). Many people have experienced the problem you’ve outlined, but few appreciate the significance of the issue (as you have). I’ve worked on several debutanizer applications where I have had to force the gains on the top and bottom “specs” to have EXACTLY the same ratio. You usually don’t have to change the curves more than 5-10% to make this happen. By doing this you force the “near-singularity” to be exactly singular. Thus the controller will not bounce between limits for small disturbances. You then set up the LP to drive the unit to the direction you want (e.g. maximum separation).
: The other debutanizers I’ve worked on probably would have demonstrated similar symptoms but they hit a constraint (e.g. flooding limit) before you got into this “region”.
: I guess it is an issue of “controllability”. I’ve heard of people spending a lot of time trying to study (e.g. Masters thesis work) this “behavior” (e.g. trying to control top and bottoms specs independently) in debutanizers. It simply boils down to you can’t control both ends of the tower independent after a certain point. For example, look at the following example:
: Reflux , Control Temperature , Upper tray Temperature
: 6 , 300 , 150
: 6.5 , 300 , 148
: 7 , 300 , 147
: 7.5 , 300 , 147
: 8.0 , 300 , 147.1
: 8.5 , 300 , 147.2
: You get the idea. Once you get to a certain point, then you can’t control both ends. The above data is only representative (not exact numbers) of offline steady-state simulation data. Thus it is not sensitive to any flooding constraints (i.e. it is a real phenomena). Also, if I remember correctly it is not like the top purity was super pure. It may have been like 2% C5s in overhead.
: In any case, I think you have come-up with the proper solution. Make the gains exactly singular (down to as many significant digits as you can). There is an Aspentech program available that runs the model through a singular value decomposition that can help pick-up near singular models.

Optional Link URL:

Link Title:

Optional Image URL:

[ Follow Ups ] [ Post Followup ] [ Control Consulting ] [ FAQ ]