PID loop - long residence time 1

[Extasis]

Hi there,

I am trying to control a mass flowmeter to input Chemical A into a series of reactors, each with a residence time of 30 mins. I need to dope enough of Chemical A to remove environmentally harmful Chemical B from the reactors (need <1ppm). My analysis sampling point for Chemical B is after Reactor 2, so total residence time is 1 hr (after 2 reactors x 30 mins each).

I am using a PID controller to control the flow of Chemical A into Reactor 1 based on the ppm level of Chemical B out of Reactor 2. It takes approximately 10 mins to get a ppm reading of Chemical B via the online analyzer.

Once I get a ppm reading for Chemical B, I set the flowrate for Chemical A dosing. The problem is that if I change the flow of Chemical A, I do not see the effect of this change (lowered ppm reading of Chemical B) for at least 1 hr.

My knowledge of controllers is very limited; I realize that this problem may be very basic and so I appreciate any help/ideas I can get!

Thanks in advance!

 

[JLSeagull]

Look up a loop tuning - dead time compensaton technique called the Smith Predictor. Take an example like a conveyor where stuff goes onto the belt, time passes, then stuff comes off the belt. In this case the response is a purely dead time issue.

 

[itdepends]

Your best bet would be to include a sample point to analyse the concentration of Chemical B entering reactor 1. Assuming you have some idea of the relative proportions required (how much A to neutralise X amount of B) you can set up a feed forward loop and adjust A based on your analysis of Chemical B exiting reactor 2.

 

[sreid]

Delays in control loops may be basic but are usually difficult to deal with. A delay has constant gain but the phase linearly increases with frequency.

As JLseagull states, a Smith Predictor is a classic way to deal with delays

(see

http://www.manufacturing.net/ctl/article/CA188333?pubdate=5/1/1996

for example).

The model of the "Plant" has to be pretty good though for the Smith Predictor to work well as pointed out in the above article.

 

[FrancisL]

Can the amount of chemical A that you need be predicted by something else, like the flow of some stuff somewhere?
If so add some feed forward to the loop so the flow of A is changed before the analyser sees the result.
The PID can then fine tune.

 

[JimCasey]

You get this same effect frequrntly when trying to control temparatures. The time delay causes the controller to overshoot and oscillate. One trick that helps is to cascade the corrective signal of chemical A based on the flow in of chemical B, then you will only need to make fine adjustments on the discharge end.

In my world this is known as cascade control. I have not heard of Smith Predictor before, but I will be sure to school myself on that topic immediately. May be similar.

 

[FrancisL]

Feed Forward is not the same as cascade at all.
Go Google feedforward cascade

 

[Extasis]

Thanks everyone for your help! Very much appreciated.

 

[JLSeagull]

If any confusion exists, cascade control uses two PID controllers. The controller ouput from a slow loop is used as the remote setpoint for a fast loop. Temperature is a slow loop compared to flow, thus the output of the temperature controller is the setpoint for the flow controller. This technique can smooth the size and shorten the period of the cycle swings.

 

[m777182]

Can you afford to make a relliable model of the process?
m777182

 

[Extasis]

yes, in fact we are likely going to do just that - process modelling. thanks

 

[m777182]

Try to follow Itdepends idea AND process modelling.
m777182

 

[CARF]

DOWNLOAD PRACTICAL PROCESS CONTROL BOOK @

http://carf.ariesplace.com/pressure_drop_calculator/download/Practical Process Control.pdf

Good luck,


PS. From practice I know deadtime like these (distance/velocity-lags) are very difficult to control.

A by-pass line, by-passing the big tanks, directly inserting into the reaction tank should be considered.

 

[itsmoked]

I would agree with m777 that you probably want to go with modeling instead of PID.

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com

 

[fjh502]

I agree with JLseagull that a cascade control would help you, you can try.