PID Control

Hi, Jake,

You can check

Thomas ATMEL Semic> I am on the Systems Design Team with Jim, and I am currently in the

Hi Jake,

I found

was very helpful .

Dave

The article at

has a lot of practical info. Try the home page for more related articles.

I wrote an article about how to tune a PID controller for Circuit Cellar in 2000

It's available here in html

If you have specific questions, this newsgroup is a good resource for you.

Thank you to everyone for all the help. If there is still any help you can provide that would be great.

Look on

for PID discussion on sci.electronics.design, I recall that it's been discussed many times in the last couple of years, at least once in great detail.

-----

Are you looking to understand PID control from an acedemic standpoint or do you actually want to control something ? If so can I recommend that you spare a thought for Fuzzy control instead. From a software point of view, similar results can be achieved a lot easier using Fuzzy !

Dave.

Fuzzy control? Can you elaborate or post a link to some info? I've seen a lot of info on fuzzy login over the years, but never a "Fuzzy Control" system that would be a replacement for a PID control.

--Steve

That's because Fuzzy control can't replace PID. PID is theoretically more efficient and more accurate than fuzzy control. The difference is that fuzzy control often does not need to be tuned and is relatively insensitive to drift in the system parameters. The software price you pay for this is maintaining a set of control parameter sets, instead of just one. If you have a LOT of processor horsepower and want to make a control that applies easily to a wide variety of systems, fuzzy control may be the one for you. Otherwise, stick with PID. You don't need a PID loop to be theoretically tuned to get good results.

In article ,

I wrote an article that appeared in Circuit Cellar Ink about how to tune a PID controller that might help. It was pretty introductory.

You can still read it here:

When you have questions, fire away and we'll all try and help.

This is a good starting point !

of links there.

Another good source is an excellent book called "Using Fuzzy Logic" by Jun Yan et al. ISBN 0-13-102732-8 Although the book is about 10 yrs old now, it concentrates on how you actually USE fuzzy logic in a practical sense, rather than as an academic exercise.

A google search of "fuzzy logic control" throws up about a gazzillion hits too. Have fun !

Dave H.

Eeerm exsqueeze me - but it can ! Any algorithm can be replaced by a rule-set. It's just a matter of the right rules !!

Wrong!

You can create a rule set that mimics PID. But it requires 10 to 100 times the processing power to do the same job. If you were to run it on the same processor, it would take many times longer, and therefore can't do the same job.

People who say that fuzzy logic can substitute for PID fail to take the algorithmic complexity into account. Fuzzy logic by its nature solves the problem a dozen or more times in different ways, then merges the result into its final answer. PID requires one integer algebraic expression over three stored values, getting the multipliers right is the hard part.

The advantage of fuzzy logic is that it doesn't need to be tuned (if the rule set is right). However, it also gives sub-optimal performance on a much bigger core. PID gives the linear theoretical optimum when properly tuned and can be done on a much simpler processor for the same performance. Fuzzy can only match the optimum if the rules are set up and hand tuned like PID. At that point, you are using 10 times the processor to do the same job with the same or greater difficulty (setting up a basic fuzzy control is easy, setting up an optimal one is NOT).

If you don't want to tune, I would recommend adaptive PID. Adaptive programming is also an artificial intelligence technique (like fuzzy logic), however it doesn't have the overhead. The technique is also called self-tuning PID. The basic method is to set up a PID loop that is nowhere near optimal, but guaranteed to be stable. The program then uses a simple mathematical model of the system to adjust the parameters over time. It is actually related to neural networks. The beauty of the technique is that the adjustment doesn't have to happen in real time, so you can run the adjustment loop as slowly as you need to. The real time part is a traditional PID loop, which as previously noted is very simple and fast. The result is that you start out with a really sloppy PID loop, but as time goes on it gets better and better until it is performing at near optimal levels. Depending on your system, how good your model is, and how you use your system, and adaptive PID can "catch on" within a dozen or so cycles.

I can't really blame them for ignoring the horsepower requirements, computer science curriculums stopped emphasizing optimization and algorithmic speed about 15 years ago. The attitude seems to be that any amount of computational effort will be covered by advances in hardware.

It's the age old analogue v. digital arguement. While I agree that analogue gives infinite variance and can produce better results on simpler technology - I don't see too many people rushing out to a buy a film camera or a VHS recorder !!! Fact is, hardware technology IS taking up the slack! The PID controller I set up, did need some hefty processing power. I used a 68HC12 but it cost about $50, runs at 8MHz and comes with fuzzy commands as part of the language set. All I am saying is that there are alternatives out there. Fuzzy also works well with self tuning - either ANN's or Genetic Algorithms !

Guess you pays your money and takes your choice.

Correct me if I'm wrong, but a PID controller can be either analogue or digital, and based on the previous replies, either would be more "efficient" than a fuzzy controller.