Control of fast reaction


Thread Starter

Syamsul Shukor

I would like some info on controlling of a fast reaction ie. residence time in a few seconds. What's the best control strategy to overcome the time delay of the instrumentations on the system. Adaptive? Gain scheduling? The process is a neutralisation process.
If the time delay is from the instrumentation, we
can try to use Smith Predictor or Feedforward
Control. Adaptive and Gain scheduling deal with
nonlinearity but not time delays.


Dr. Siu O'Young, P.Eng.
Associate Professor
Instrumentation, Control and Automation
Faculty of Engineering and Applied Science
Memorial University of Newfoundland
St. John's, Newfoundland
Canada A1B 3X5

Voice (709) 737 8345
Fax (709) 737 4042
email: [email protected]
office: Engineering EN3074
This problem is certainly one of the worst in Process Control and Instrumentation.
I doubt the Smith predictor will help. I had that before but very slow (12 hours). Whatever. Because of the assymptotic (even reversal) of the dynamic, you should approach using feedforward c/w an approximation of the reaction in the feedforward signal. If things would match correctly there is nothing left to correct from the controller. As this may be hypothetical, then the controller may do the rest. Watch, because you may need second order derivative, and that one is more than nasty in numerical maths.

Being very expert in numerical approximation of functions, I can assure you that your reaction would approximate near perfect, may be exactly.
you may just have it in your chemestry book.

Having modeled your feedforward signal, if not perfect, the controller will see a small linear residue very slow in comparison to the native dynamic. This, does not contradict with the eventual need of second order drivative.

Good luck.
[email protected]
Simplest strategy is PID control and for this reason is probably the best. IF you can measure a signal that is representative of the dead time then modifying the Integral value of the controller with a factor of this value is the way to go. If your controller has D also, you will also need to modify the D term. Both I and D have time in them, gain does not, so you would NOT modify the gain of the controller to adjust for changing dead time in the process.

Dahlin, Smith Predictors, etc. will work also, but are a LOT more complex for a very small if any improvement. They also suffer from robustness problems.

PID does not care if your process is slow or fast, with the proper tuning AND the proper sample interval. General rule for sampling is 4 to 10 time the dead time of the loop.

Want more info and in depth explanation with formulas? See:

John Gerry