…as told by Walt Driedger, P. Eng.:
Walter Driedger has my vote for being one of the best automation engineers in the world. He is consistently correct in what he posts to forums like the automation list at control.com and the ISA Control mailing list. He’s a graduate of the University of Calgary, and a professional engineer in the Province of Alberta. He’s a Senior Member of ISA, a fact which that organization ought to be extremely proud of. Many of his longer writings on control can be found on his website: www.driedger.ca
Walt gave me permission to use the following post, because it is absolutely the best distillation of the sum total of human knowledge on the behavior of control loops that I have ever seen. I’m trying to figure out where I could put this in the magazine…maybe in Ask the Experts, if Bela will permit.
(I’ve left out the name of the questioner pending receipt of his permission to be named.–WB)
2006 August 12
From: [name withheld] To: The ISA Control News Group
I have some question regarding loop tuning.
Dear [name withheld],
Controller tuning is one of all time FAQs at this site. What I’m going to tell you will be hard to take: It will contradict everything you learned at school. My answers to your questions are interleaved between the questions…
Q 1- Do I have to know the time constant, dead time and process gain? If yes how do I extract these values from process data?
A 1 – No, you don’t need to know any of them. It’s called ‘tuning’ not ‘calculating’ because it is an empirical process. Forget the formulas you learned in school.
Q 2- What is meant by process static gain and dynamic behavior? How do these affect the stability of a control loop?
A 2 – Forget about static gain, dynamic gain and loop stability. These are useful concepts for understanding how things work but not used in actual process control.
Q 3- How can I measure the performance of a control loop?
A 3 – A loop is performing best when the process is under control and doing what you want. For some loops this means minimum integral error squared, for others it means absolutely no overshoot, and for still others it means minimum average valve movement. For some it means putting the loop in manual and leaving it alone! (Operators believe that applies to most of them.)
Q 4- If I have a nonlinear transmitter, how is this going to affect loop stability?
A 4 – All process control loops are nonlinear. That is why the math you learned in school is useless. The loop has to be tuned in the high gain region so that it is never unstable there. That means it will be detuned in the low gain region. There are sophisticated ways of dealing with this but you shouldn’t bother with them until you fully understand the simple methods.
Q 5- How can I determine the degree of loop stability?
A 5 – Don’t worry about it; nobody cares. The only thing that matters is that the loop is stable in the high gain regions. You will find that out by looking at it.
Q 6- What are the factors that influence loop stability?
A 6 – Hysteresis, non-linearity, dead time, dead band and lots of other nonlinear things. Some of them are maintenance issues, some are process issues that you can’t avoid. For example, a feedback loop based on reactor product and controlling reagent feed, when there is a two minute transport time through the reactor, is a devil. It is dealt with using a combination of feed forward, and restricted range feedback trim; a highly non-linear setup.
Q 7- Is there a rule of thumb that I can follow if I don’t have process data?
A 7 – That’s always. You never have decent process data. And even if you did, it would be useless since the process is always non-linear and time varying. Get some books on tuning and forget about any calculations based process data. This link directs you to an excellent one for only $18.
CONTROLLER TUNING AND
CONTROL LOOP PERFORMANCE,
A PRIMER, SECOND EDITION
PID WITHOUT THE MATH
By David W. St. Clair (Retired DuPont Engineer)
A copy of the original paper “Optimum Settings for Automatic Controllers”
can be had for free at my web site www.driedger.ca but it may be difficult to use as it written for some rather ancient devices and uses terms no longer in common use. But if you had nothing else you would be far ahead of anyone trying to calculate values.