limits of analog electronic pid?

[chris_leyson]

Nice one Steve, thanks for posting the schematic It's a classic and coppercone2 spotted the the jfet limiter, a nice touch  This controller has been around for quite a while and it should really be part of the "Art of Electronics". There is also a separate controller for the LED current and a few subtle linearity tweaks which I had forgotten about.

[chris_leyson]

Nice one Steve, thanks for posting the schematic It's a classic and coppercone2 spotted the the jfet limiter, a nice touch  This controller has been around for quite a while and it should really be part of the "Art of Electronics". There is also a separate controller for the LED current and a few subtle linearity tweaks which I had forgotten about. I think think the 0.1u cap and 20k resistor are for RF decoupling the interconection cable to the galvo. The feedback from the photo diodes is low current and high impedance, so electrostacic screening is important, more useful material for the "Art of Electronics".

[LaserSteve]

Mike Thanos, I believe, is the one responsible for that amp.

Steve

[coppercone2]

do you ever have circuits that turn off various unnecessary mathematical operations to make it more stable and less noisy when it gets real close and the process is not being disturbed? i.e. temperature control.

this has more to do with general control theory, but these various factors have something to do with dynamic range right? then only turn em on when a particular disturbance is expected?
say for a refrigerator

[LaserSteve]

Yes, adaptive control is used in many systems when you get close to the setpoint.   I have one very expensive Omron controller at work that does that.

Steve

[Kleinstein]

do you ever have circuits that turn off various unnecessary mathematical operations to make it more stable and less noisy when it gets real close and the process is not being disturbed? i.e. temperature control.

this has more to do with general control theory, but these various factors have something to do with dynamic range right? then only turn em on when a particular disturbance is expected?
say for a refrigerator

Usually one needs all 3 terms of the PID regulator together and can not just turn away one part. If for some reason one want's to use a regulator without the D term to reduce higher frequency noise, this usually required to adjust the other terms as well. With a correctly tuned PID regulator (including HF roll-off, see below) there is hardly a need to turn of the D term. Analog implementations rarely use extensions to the pure PID. That secret box in the schematics might be a kind of approximate anti windup - it's possible analog, but often tricky. So extensions beyond pure PID are often a reason to go digital. Digital control can include other targets like reduced switching.

One point to observe is that a true pure PID regulator is not possible, as the D term would have diverging high frequency gain. So a real regulator (analog or digital) will always have another high frequency roll off of some kind. In digital implementations this is often the sampling rate and AA filtering. So a real PID regulator will have an additional 4 th parameter to tune.  A lot of classical control theory about a pure PID is  thus ignoring real life limitations.

[coppercone2]

what should I search for to learn about the extra tuning required because of the D-term in an analog pid?

Roll off pid brings up too much stuff about quad copters

Also, does anyone know anything about the higher order integrals and where they are useful. I heard stuff about flow control. I am interested in precision constant flow control with fluid or gas pumps or valves/regulators that deliver flow from a tank.

Does anyone have like diagrams that show what kind of sensor readings would need what kind of math to deal with? Like example of things like ramp perturbation or whatever that causes situations which require more complex circuits

[IanB]

what should I search for to learn about the extra tuning required because of the D-term in an analog pid?

I think the problem is the same in digital implementations. The problem is that the D term amplifies the rate of change of the input signal. Higher frequencies (including noise) have higher rates of change in the signal, so they tend to introduce unstable responses. A typical mitigation is to filter the input to the D term (or both P and D terms) using a first order lag. A first order lag is a low pass filter with a soft roll off, so it has a mild and non-aggressive effect on the overall control function.

[IanB]

Also, does anyone know anything about the higher order integrals and where they are useful. I heard stuff about flow control. I am interested in precision constant flow control with fluid or gas pumps or valves/regulators that deliver flow from a tank.

A rule of thumb is that if the process is non-integrating then the controller can support a lot of integral action. However, if the process is integrating you should keep integral control action to a minimum. Thus, flow loops can have a lot of "I" and moderate "P", while level controllers will tend to have significant "P" and almost no "I".

The amount of "P" in a flow loop depends heavily on the scan rate of a digital controller. The faster the scan rate the more proportional action you can have. (The scan interval is a dead time, and too much proportional action in conjunction with dead times leads to instability.)

[coppercone2]

what if its an analog sensor connected to an analog loop? What is scan rate equal to? Is it the stable zone of operation for the D term frequency response in an analog controller?

[IanB]

what if its an analog sensor connected to an analog loop? What is scan rate equal to? Is it the stable zone of operation for the D term frequency response in an analog controller?

Scan rate is an artifact of digital controllers. Analog loops do not have a scan rate.

Theory says that for a process with a first order response it will be stable for any value of proportional action. This is in the mathematical, ideal case. The real world is not ideal.

Flow controllers tend to be noisy. Put your ear against a pipe with water flowing through it. You will likely hear a hissing noise and feel a rumbling vibration. Many flow instruments work by measuring the pressure drop across a restriction in the flow line. All that noise and rumbling will appear in the pressure transducers. So if you wind up the gain too much you will amplify the noise and cause the output to jitter. Some input filtering will be needed, and you will need to decrease the gain while increasing the integral action. Integral action has an averaging effect, so it will inherently have noise immunity. Since a flow loop is a non-integrating process you can increase the integral action a lot to get tight control.

Derivative action is not really needed in flow loops. PI controllers will be fine.