PID vs floating control

I agree that things should be a simple as possible. KISS applies. That is why we have auto tuning and ONE slider bar that adjusts all the closed loop gains. How the gains are calculated is hidden from the user. Adjusting one slider bar is MUCH simpler than tweaking gains.



The operators don't need to be experts. The people making the systems are or should be and should hide the complexity in their software or firmware.


As far as errors in modeling go, the model doesn't need to be perfect. In fact it is NEVER perfect. It it was then closed loop control wouldn't be necessary. Feed forwards are based by the open loop model. If I can estimate the model within 5% that means the closed loop part only needs to have enough error to generate the last 5%. On a system with feed forwards but only proportional control, this reduces the error by a factor of 20. If my model is only able to estimate the control output with 50% then at least the closed loop part only needs to make up the last 50 reducing the error by 2 which is a lot better than nothing.

Each of these factors will vary continuously in the typical treatment plant, with 2:1 ratios on a good day and 4:1 or 5:1 common occurrences.

Click to expand...

If so then how does floating control deal with these changes when the integrator is so slow? The must be more to the control than just floating control. Didn't you have a proprietary formula that you use in addition to floating control that is more of an estimate like a feed forward?

Peter Nachtwey said:

The operators don't need to be experts. The people making the systems are or should be and should hide the complexity in their software or firmware.

Click to expand...

And I should have a herd of unicorns in my front yard farting rainbows. The likelihood of both of these is about equal. I deal with the market as found, not how it should be.

Peter Nachtwey said:

... If so then how does floating control deal with these changes when the integrator is so slow? The must be more to the control than just floating control.

Click to expand...

One key is to include time delays between successive error calculations to allow the basin to approach a new steady-state value before perturbating it with another control change. In previous posts on this thread, I've listed some of the other modifications, such as biased gain, proportional speed, etc.

Peter Nachtwey said:

... Didn't you have a proprietary formula that you use in addition to floating control that is more of an estimate like a feed forward?

Click to expand...

As previously indicated feed-forward is only applicable if the appropriate real-time raw data is available, and it rarely is.

I go into great detail on my methodology in my book. It isn't proprietary. I simply based it on what I saw when I followed good operators around and observed how they adjusted their aeration systems.

Peter does in fact have a unicorn farm.

Wow, perturbating is a word, archaic, but a word nonetheless

Not unicorns, mathemagic.
Notice that someone recently found this forum searching for info on Smith Predictor.
It is kind of like, "if you build it, they will come".

I've been following this thread and would like to give my two cents.


More than 15 years ago, when I was about to finish my undergraduate studies, we learned about Fuzzy logic controllers and our professor said that "classical PID controllers" are about to become obsolete in near future. That didn't happen. For a few years there were some kind of competition in writing articles about how fuzzy controllers or even custom made control algorithms are superior to PID etc. But clearly, PID still dominates the industry and will in the future. The PID is not a silver bullet, but in most cases if properly tuned it can solve the problem.
In my experience (mostly control of technological porcesses such as pressure, temperature, level etc, where setpoints are constant 99% of the time) PID gets the job done, even though in some cases a special attention should be devoted to how control deviation (so called error signal) is formed. This is especially true if there are more actuators that control one process variable and there must be some kind of leveling to avoid fighting and race between them.


After more than 16 years of practical experience, I can say that the best controller is the one that achieve the satisfactory level of performance i.e. that is good enough. There are a lot of people there who don't know how to tune the control loop and end up with the performance "as is" after many trial and errors.


Many students finish studies without actually understand the "s - Laplace domain" what poles and zeros really are and similar. Of course, real world systems are not produced with a transfer function stamped on it, so it is easy to make Matlab/Scilab or any other simulation software model and to use some built in optimization routine to find the PID gains.



The real key is to learn the basics of system identification and to learn how to make process models that at least, reasonably good represents real world system near operation point.
Peter has mastered this and I learned from him and actually understand more than I was able in college or by myself playing with different simulation software. He made many Mathcad sheets with different forms of PID (I-PD, PIDD and many more that can be found on his YT channel).



It is bad to make general conclusions because they're usually not true. I work in a coal power plant and on one unit we have more than 130 control loops, from very basic ones to really complex ones, but in all of them PID (or PI) controller is the final software block that provides control output wired to actual control element (motorized valve actuator, pneumatic drive, VSD, etc.).


Mos of the operators and even maintenance engineers have no clue about poles, zeros or PID gains. They're only interested if the overall performance is good enough. And if some controller gives good enough performance, than it is indeed good enough solution.


I don't think that general debates such as PID vs Fuzzy or PID vs floating control are justified. It boils down to a person that actually makes the feedback closed and to his knowledge and experience whether the control performance would be good or not.



Peter has actually implemented a lot of control theory structures and concepts in actual controllers that are functioning in a real world (Autotuning, Kalman filter, double derivative gains, Feedforwards and many). RMCTools are very superior software and RMC controllers that have these implementations actually running inside are really state of the art, but that is only my personal opinion.

As the other thread about the Smith predictor shows it is not easy to implement any more advanced control topic in a PLC.


My conclusion is this. If the controller solves the intended automation task with acceptable level of performance, then it is the right structure and algorithm for the job. Whether it is the best, that mainly depends on the person who is doing the tuning.

Pandiani said:

... My conclusion is this. If the controller solves the intended automation task with acceptable level of performance, then it is the right structure and algorithm for the job. Whether it is the best, that mainly depends on the person who is doing the tuning.

Click to expand...

I agree 100%.

Tom Jenkins said:

I agree 100%.

Click to expand...

I know someone that won't agree.

Phrog30 said:

I know someone that won't agree.

Click to expand...

I agree if it is good enough. What I don't like is misleading people to think there is something special about floating control relative to a PID when floating control is just a PID without the P and D and a very slow integrator.

Peter Nachtwey said:

I agree if it is good enough.

Click to expand...

That's the point chief.

Are you practicing for a career in politics, cause I think you've hit a point where sense has left the room.

Phrog30 said:

That's the point chief.

Are you practicing for a career in politics, cause I think you've hit a point where sense has left the room.

Click to expand...

I doubt anyone who could explain what a transfer function was has ever been in a position of political power... possibly why the world has so many problems.

At least the Nukes would have an optimal trajectory if Peter was in charge.

OT, some history

Saffa said:

I doubt anyone who could explain what a transfer function was has ever been in a position of political power... possibly why the world has so many problems.

Click to expand...

Yes, engineers know that everything is a compromise. The trick is to have an accurate evaluation or cost function. Then the goal is to minimize costs or undesirability. in short, engineers can't be politicians. Politicians just buy votes with other people's money

The off topic part.

At least the Nukes would have an optimal trajectory if Peter was in charge.

Click to expand...

Yes, but they are probably already optimal. What is tricky is intercepting high speed objects. Here is an example of matching positions and velocities when the target position and velocity is unknown.

This video was made by Temposonic MDT Transducers. They chose our controller because they know our controller will make the Temposonic rods look good.
https://deltamotion.com/peter/Videos/MTS Sensors - Balloon Demo.mp4

The person is moving the dart randomly. The object to to try to break the ballon.
The ballon allows the dart to barely make contact with the ballon but not put enough pressure on the ballon to pop it even though the darts speed is changing rapidly and randomly. From the Temposonic point of view, they are showing off their feed back update rate and resolution.

This is only a one dimensional. However, we could easily do this in 3 dimensions. Also, it doesn't take much to change the sign of the algorithm to intercept and object instead of follow the target. The algorithm isn't that difficult but it requires super fast feedback, resolution and update rates when moving a speeds MUCH higher than what is possible manually.

We programmed that demo from over 2000 miles away. Temposonic is in NC and we are in WA.

Back in 1987 I almost left Delta since I was not an owner then, yet . I got an interview at the Northrop Research & Technical Center. My interviewer looked like a long hair hippie that sat cross legged on his couch. He was definitely smart. At least they didn't waste my time with a clueless human resources type. My interviewer asked me about how I would avoid anti aircraft missiles. I mentioned a cost function that maximized distance and bearing rate. If the bearing rate is alway changing then there will be no collision if you only consider points. However, distance must be considered too. That interview was fun.

It was a good opportunity. I would have had access to Cray computers, the fastest at that time. The pay was MUCH better. The environment was great with a lot of other nerds to bounce ideas off of. The site was a nice business park. Obviously I decided to stay at Delta instead of being a cog in a wheel.

I also had an interview at Northrup's Hawthorne site in the middle of downtown LA. This was not a good place to be. The interviewer want me to help with voice activated controls for fighter jets. I told her it was never going to work. Although speech recognition has come a long way it would not work for fighter jets. I played realtime games on the internet. I know my hands are much faster than what I could speak. I tried to explain that to the interviewer. She didn't like my answer. Obviously that interview did not go well but I didn't care because the idea is all wrong.

That was 1987. At that time there was a bounty to get engineers.

That was a pretty neat demonstration with the balloon. A good example of where very accurate control is required, and where "good enough" probably doesn't cut it.

Voice controlled fighter planes sound like a terrible idea.. definitely dodged a bullet there, although the pilot probably wouldn't have if they'd gone ahead.

People, I certainly understand that the question is " in a timely manner"))). But can’t you please tell me where I can find your systems data (Set Point change response, Disturbance response)?

MaxK said:

People, I certainly understand that the question is " in a timely manner"))). But can’t you please tell me where I can find your systems data (Set Point change response, Disturbance response)?

Click to expand...

i don't understand your question There is no data except for what is in post #1.
This thread was started in response to another thread where floating control was said to be better than PID control.

I should have put links to the other threads in my first post so people reading this thread would understand the context.