PID constants: Proportional band or Gain, reset or integral time?

AFAIAA, proportional band and reset rate (repeats / minute) are more or less legacy approaches.

If you go for scan frequency OTOO 100Hz, just ensure that your numeric precision is sufficient to avoid flat spots in control action. With normal IEEE floating pt, that's probably not an problem but if you're using small micros and reduced numeric formats, it could cause you problems. For large processes, integral times of many hours occur from time to time.

It depends on what the majority of your customers applications are. If the applications are in process control then I would stick with the ISA form with integrator and derivative time constants. Motion controllers control systems with time constants in the milliseconds so time constants in minutes isn't used. Motion controller usually have independent gains.

I agree with Bruce, proportional bands and the term reset should be discarded.

What do you want to support? I would go with the ISA because there is plenty of documentation on it. Auto tuning software will work with your PID controller if you stick with the standards.

I think the sample time depends on the procress being controlled. Temperature system often get away with a second or more. Motion controllers usually are at 1K Hz or faster.

If this is a product then keep it standard otherwise you will end up doing a lot of tech support for the odd 'features'. You will get tech support calls. What do you want to support?

Other questions you could have asked are:

1. Do all the gains work on the error or can the higher order gains work on just the PV?
2. What form of PID are you going to use? Position or Velocity. I hate those terms. I prefer 'full value' or incremental respectively.
3. Are you going to give the customer a choice?
4. Does the customer need a choice for the targeted application or must the controller handle anything and everhthing.

Peter Nachtwey

Peter Nachtwey

IEEE single-precision floating point can run out of useful bits surprisingly fast when you're oversampling. Think about the integral gain you'll have if you're sampling at 100Hz and have an integrator time constant that runs over a good part of an hour.

Double check. _Always_ double check. If you're putting it into the hands of a customer, _triple_ check.

My book goes into sizing data paths in the chapter on implementing control systems in software. I dunno if it's worth getting a copy just for that, but if you're interested take a look:

look:

I bought several copies of Tim's book and think it is excellent for the material it covers. Beyond personal study, it provides me with a reference platform that I can give to the mechanical and software engineers so we all "sing from the same sheet". Use words that same way, and it is understandable by any engineer.

Ray