When not to use PID-controllers

[ecwai]

Hello. How do you know when you cannot apply PID-controllers to a system? Have you had an experience where you first thought that a PID-controller might work but eventually find it impossible? How would you decide?

 

[JLSeagull]

PID works just fine for most applications. Other control functions exist too. Nobody addressed fuzzy logic or bang-bang control. Consider some of the other tricks of the trade.

Use derivative. If it causes problems reduce the amount or turn it off. This is an advantage of electronic controls over pneumatic controls - the derivative can be turned off with electronic controls. In the pneumatic days the derivative function was a hardware module. Once installed, typically the amount could be turned down to the bottom setting but not off.

 

[PNachtwey]

1. A PID works well for only a small number of plants. There are plants with two poles.
2. A PI works well for plants with 1 pole.

No one has answered why a derivative gain should be off. Telmatrix, could you provide an example? I have motion systems where the shaft was bent an I could see the effect of the bent shaft on the controller output.

skogsgurra, I know when not to apply a PID. I would put it another way. I know when to use a PI or PID or I-PD. I know when to use feed forwards. I know when to use observers, Kalman filters etc. What I need is specifics.

For the most part some form of PID along with feed forwards will do. The feed forwards should be able to predict the control output within 5%, The PID , or some of of it, should be able to correct the last 5% with little problem.

There are also cases where an observer or Kalman filter will provide better feedback for the controller.

I don't think this is a matter of PID vs something else but more spme form of PID + something else.

Peter Nachtwey

 

[telmatrix]

Hi PNachtwey,

You asked:
No one has answered why a derivative gain should be off. Telmatrix, could you provide an example? I have motion systems where the shaft was bent an I could see the effect of the bent shaft on the controller output.

I never said explicitly that derivative gain should be off, just judiously applied, but yeah, perhaps indeed should off for the reasons given.

You seem to be a textbook guy, so an example outside of the noise reason, it is easy to visualize. The "D" term is used to anticipate the future, that is, reduce overshoot, hence if your system is sufficiently damped, the derivative gain should be off as it is not even needed.

You also said:
1. A PID works well for only a small number of plants. There are plants with two poles.

Ans: It appears that you may be too engrossed in that textbook theory. See my first post in this thread. PID is quite versatile in fact, and there are many different permutations. By the time one throws in the motor, load(s), semiconductors, etc., most systems have far more than just 2 poles, it is really a question where the dominant poles lie, and how stability can be managed, and managed cost effectively, right?

2. A PI works well for plants with 1 pole.

Ans: Sure, no dispute here, but will work in some more complex scenarios as well. BTW... A "plant" with "1" pole? What real life plant has only "1" pole? Even the simplest DCPM motor alone has 2 poles (either imaginary or real - depends on motor construction). This does not include the rest of the system the motor resides.

 

[PNachtwey]

You seem to be a textbook guy, so an example outside of the noise reason, it is easy to visualize.

Not at first. I have learned over the years that if it doesn't work in theory it doesn't work in the real world. Actually, I get involved with a lot of motion systems, Most are hydraulic. Derivative gains are used all the time.

BTW... A "plant" with "1" pole? What real life plant has only "1" pole?

None, in reality. There may be systems where only one pole is dominant. Only those systems can be tuned using a PI controller and only if one doesn't excite the higher order poles.

The point I have moving towards is that PI and PID controllers only have enough gains to place the poles for the simplest of systems and one must be able do use the derivative gains to tune the system with two or more poles properly.

The OP asked when not to use a PID controller. One answer is when the plant can't be reasonably be modeled as having two dominant poles. A PID controller may for for a system such as...

(Ka*omega^2)/(s*(s^2+2*zeta*omega*s+omega^2)) where
Ka is the actuator gain
zeta is the damping factor
omega is the natural frequency

but only the damping factor is high. Even then one can't get a nice critically damped response except at one theoretical spot. A second derivative gain is required because the actuator has three poles. Calculating the acceleration from position feedback is a little more work than calculating the velocity but it can be done.

 

[telmatrix]

We are both on the same page here. Good discussion.

BTW, you said:
Calculating the acceleration from position feedback is a little more work than calculating the velocity but it can be done.

Ah... A double derivative. Now that can be "noisy".