Integral and Proportional Gains


Thread Starter

Mario Bordignon

I am developing a PI Temperature control for a Hot By pass (A kind of by pass line for gas cooler)in PLC-5. Regarding the Process Logic itself, no problems. My questions is about what values should I use in the integral and Proportional Gains. I know that should be a small gain, because is a Temp controls (slow), but I wouldn't like to use my feelings only, i'd like to projet those gains, according the valve size and the process.
Any idea, Does anybody know any site where I can find out a program for that.

Heavner, Lou [FRS/AUS]

You need to know more than the valve size. The valve could be oversized, undersized, etc. You need to know the dynamic process response to change in controller output. You might also like to know how the process changes due to external disturbances if they can be measured. Either you can simulate the process if you have an adequate dynamic model or you test the actual process. Process test can be open loop (bump test) or closed loop (oscillating). Because the process could be nonlinear, you might want to perform your testing with both small and large changes in both directions and at more than one condition in the operating range. Your mission, should you choose to accept it, will be to get operations to let you test the process. So do your homework.
Lou Heavner
Emerson Process Management

Johan Bengtsson

There are several ways of getting good values for
gain and integral time.

Some of them include:
Ziegler-Nichols, gives you generally a somewhat oscillating control, if you use PID it would get quite fast.

Lambda, gives you a slower (but controllable) speed and normally not overshooting at all

A lot more exists, if you really want a longer list you could get one, but I suppose you don't really want that.

Completely regardless of what you do you have to find out the dynamics of the process, either by meauring (generally easiest) or by creating an accurate mathematic model of the process (not

Besides, just because it is a slow process does NOT mean small gain (it does not mean a big gain either) since that actually don't have anything at all to do with the speed of the process
but rather other process parameters. True enough, low gain gives you slow control, but that is the desired closed loop characterisitcs, not the process itself.

A quick walk thru for lambdatuning:
1. Do a step response, measure the process gain (Kp), the dead-time (L) and the time constant (T). This approximates the process as a first order process with a delay, usually good enough.
2. Select the desired closed loop time constant (lambda)
3. Ti = T, Kc = Ti / (Kp * (lambda + L) )
(where Ti is integration time and Kc is the gain)
These are ONLY valid for self regulating processes.
(Other formulas exists for integrating processes).

The desired closed loop time constant is normally selected as being around T up to T*3, provided that T>L otherwise you use L*2 to L*3.
There is no limit upwards (ie you can make the control as slow as you want) there does of course exist a limit downwards somewhere, the exact limit depends on the process I have in some
cases succesfully used T/4 and even lower, but this gives an overshoot.

If you have any problem with this I suggest you get yourself some education in the area, contact me off list if you want a suggestion in this direction.

/Johan Bengtsson

P&L, Innovation in training
Box 252, S-281 23 H{ssleholm SWEDEN
Tel: +46 451 49 460, Fax: +46 451 89 833
E-mail: [email protected]