control loop evaluation


Thread Starter


hi all,

which criteria (formal..informal..any) do you employ to evaluate the loop performance?
qdr? settling time?.. or a combination?

simply, how do you know that the loop is good? do you do a bump test, just watch it running or something similar?

Peter Nachtwey

> which criteria (formal..informal..any) do you employ to evaluate the loop performance?
----snip ----

It depends on how one changes the set point or target generator.

In process control the set point often changes in jumps.

The IAE used by Expertune works but it weighs early errors too heavily and errors in the future not enough. I like to use the ITAE where the absolute error is multiplied by the time before integrating. This puts little weight on early errors after a change in the set point but penalizes severely error as time goes on.

In motion control the set point is moved smoothly from one position to the other. In this case I prefer SSE or the sum of squared error.

well, as it goes, i currently don't have the luxury of expertune. i have got regular dcs and that's why i was willing to be happy with decay ratio and settling time and the like.

one more question, most of the times (if not always), i have seen loops being tuned with trial & error method plus a little knowledge (that's what i also do,99% of the times). that doesn't involve a bump test (forget IAE, ITAE, even decay). so what credibility does it hold? i (and even you) have seen these loops doing good.
The point of my earlier reply was not to suggest you go out and buy Expertune. It was that this isn't a simple question with a simple answer.

There are literally hundreds of ways in which you can measure the "performance" of a closed loop control system. Which measures are appropriate for an individual loop depend very much on what you want to do with that loop.

In most cases, manual tuning provides you with something that works "well enough". But often it could be much better, maybe its unstable sometimes, maybe its too slow, maybe it oscillates, maybe it reaches a new setpoint OK but can't control process upsets properly. How you define these measurements requires much more than just looking at IAE.

Fundamentally however it boils down to this ... How much wasted money and time is it costing you?

I totally agree with Rob.

Control systems must meet certain tasks under conditions specified previously, called Performance Specifications, which relate to accuracy, stability and speed of response.

Performance specifications of a closed loop system depends on the process to be controlled and it´s establishment is of great importance, as they determine the success or failure of control system.

Therefore a control system can be evaluated based on whether or not satisfactorily meets the performance specifications, which should not be unnecessarily rigid or redundant because this increases the cost of the system.

Performance specifications can be set in the Time domain or in the Frequency domain. There is considerable literature dealing with this issue to deepen.
i need to clarify my last statement, though its not needed actually and its well understood but still.

when i said trial & error method doesn't involve bump test, i meant this is not used for parameter calculations as opposed to methods like
z-n or cohen coon.
If you set up a nice big target and throw enough rocks at it, then some of those rocks are going to hit the target right in the middle. How often you hit the target depends on how big it is and how good the person throwing the rocks is.

Tuning control loops by hand is completely subjective. Some engineers are better at it than others. Some mathematical methods are better at it than others. If you do it often enough you will hit the target. If you're experienced you will hit the target more often. If you have the right tools and training you can hit the target pretty much every time. But a detailed discussion of "good" and "bad" tuning is way beyond the scope of a couple lines in an online forum.

So how do I personally evaluate loop performance? I do some bump tests to collect data and throw that data into a decent control tuning package. The package does the math for me and draws some nice pictures that should make it easier for others to understand what I've done and why I've done it. It also helps me figure out whether the problem is bad loop tuning or damaged plant. And frankly, in most cases its worn out or badly set up equipment and poor loop tuning is only part of the story.
Re: comments from Rob...

Wonderful advice. You cannot compensate for a bad plant design, or broken/worn equipment with a few adjustments of the control system. Fix the plant first.

One of the most common pieces of equipment to give problems is the control valve - it sticks - resulting in poor response to the corrections of the control loop. You cannot "tune-out" control valve stickiness.

This is why I have dedicated my recent efforts to recommending the use of VFD (Variable Frequency Drives) to replace control valves in most liquid flow control applications. See my series of articles in

Dick Caro
CMC Associates
Certified Automation Professional (ISA)
Buy my books at the ISA Bookstore:
Wireless Networks for Industrial Automation
Automation Network Selection
Consumers Guide to Fieldbus Network Equipment for Process Control
Dick is right. It is never a good idea to try to use PID, or APC or multivariable control to compensate for a design failure in the plant. For example, you cannot tune a pH loop when the physical loop lag time is very long and the fluid velocity is high.

Sent from my iPad