Advanced Control - Peter Ponders PID

This is meant for those that have been through college and know how to use Laplace transforms and differential equations.

I am making a series of YouTube videos. They will not be flashy but they will be very informative about closed loop control. The reason why I am making these videos is that I have many decades of experience at supporting customers with many different types of application. I am basically self taught. I had to learn because others wouldn't. I just wanted to sell motion controllers.

I have been very dissatisfied with books written by professors that repeat the same old examples using the same old techniques trying to do almost useless stuff like root locus and other things that prove a system is stable. In real life that isn't good enough. Customers want a system that just isn't stable, they want responses that are perfect. This means that about 90% of a control text book is useless.

The other part about control books that I don't like is that they involve a lot of matrix math. Matrix math is good in some places but it is generally USELESS in a motion controller because a motion controller must be doing motion control not indexing through arrays. The other thing I don't like about matrix math is that the parameters are generally put into the matrix math blender and just become a number in an element or two. This does not lead to understanding of what and how parameters affect the response. I prefer symbolic math. I can see what parameters affect the response and how. It is no wonder people graduate from college knowing only what the definition for P, I and D gains but NOT what they are really doing which is placing closed loop poles.

Basically, I have a extreme contempt for most universities and colleges. Most do not have professors that really know any more than what someone wrote in a book before them. They cheat their students by wasting their time on 90% of useless information that I have never used in all my 33+ years of doing motion control software. There are a few good professors out there but not enough.

Have you ever looked at a control book and noticed that there are a lot of problems where you are given Laplace transform transfer functions? Where do they come from? All that stuff you learn is useless UNTIL you can generate your own open loop transfer function. Mechanical engineers should be able to provide the transfer functions to the control engineer, if they really designed instead of evolved or kludge the design, but they don't so I had to learn myself.

In my NOT SO HUMBLE opinion there are two things that are important.
1 System identification. If you don't know what you are controlling you can only guess and tweak gains.
2. Pole placement. Giver the open loop transfer function from the system identification you can place the poles where you will get the desired response. The controller gains can then calculated from the desired closed loop pole locations and the open loop transfer function.

This is for very advanced control

3. Model based control. This is a more advanced topic and isn't available on PLCs but it can make a huge difference in performance.

All the rest except Bode and Pole Zero plots are just distractions.

If you have got this far and are still interested then go to YouTube and search for Peter Ponders PID or my name. I will add more videos. I plan to cover how to control many different type of plants and control methods, not just PID. If you go to my channel you will see the video with the most views is the Sliding Mode Control video with Smith Predictor for controlling a temperature system with a second order plus dead time plant.

Edit, a current thread is asking what is the affect of scan time. I will make a video on that too.

Thanks Peter! I am in dire need of a refresher!

I am definitely in for this.

+1 I will be checking this out. I'll have to get it into my youtube rotation!

Peter Nachtwey said:

I have been very dissatisfied with books written by professors that repeat the same old examples using the same old techniques trying to do almost useless stuff like root locus and other things that prove a system is stable. In real life that isn't good enough. Customers want a system that just isn't stable, they want responses that are perfect. This means that about 90% of a control text book is useless.

There is no sense in trying to make a system response perfect if you don't understand how to make the system stable.

I prefer symbolic math. I can see what parameters affect the response and how. It is no wonder people graduate from college knowing only what the definition for P, I and D gains but NOT what they are really doing which is placing closed loop poles.

Again, if you don't understand the basics of root locus, placing closed loop poles mean nothing.

Basically, I have a extreme contempt for most universities and colleges. Most do not have professors that really know any more than what someone wrote in a book before them. They cheat their students by wasting their time on 90% of useless information that I have never used in all my 33+ years of doing motion control software. There are a few good professors out there but not enough.

You have so much contempt that it is clouding your judgement. You bash them when it is uncalled for and it makes you look less credible. You come across as trying to build yourself up by putting other people down. What they teach in universities is the basics that can be applied to many different areas, not just motion control.

Have you ever looked at a control book and noticed that there are a lot of problems where you are given Laplace transform transfer functions? Where do they come from? All that stuff you learn is useless UNTIL you can generate your own open loop transfer function.

By the same token, until you learn Laplace transforms, open loop transfer functions are useless. Both are needed, and both are taught in universities.


In my NOT SO HUMBLE opinion there are two things that are important.
1 System identification. If you don't know what you are controlling you can only guess and tweak gains.
2. Pole placement. Giver the open loop transfer function from the system identification you can place the poles where you will get the desired response. The controller gains can then calculated from the desired closed loop pole locations and the open loop transfer function.

Click to expand...

I appreciate what you are trying to do. However, I do think you will be more effective if you spend your time explaining what you think you audience needs to learn, rather than spending the time spewing contempt bashing universities and professors.

proof said:

I appreciate what you are trying to do. However, I do think you will be more effective if you spend your time explaining what you think you audience needs to learn, rather than spending the time spewing contempt bashing universities and professors.

Click to expand...

proof, Peter has 6300+ posts on this forum, and now he is telling us he is creating YouTube videos to further share his knowledge with the forum/world. Most who have been around here would say that he is one of the resident experts on applying the mathematics of Control System theory to real-world applications. He has spent/is spending a substantial amount of time explaining to the audience what he feels they need to learn because there are gaps between theoretical and real-world.

His feelings about Universities and Professors remind me of this quote during my 1st semester of College:

"Those who can't do TEACH , those who can't teach WRITE TEXTBOOKS." - Intro To Electrical Engineering Professor, when ranting about text books.

Let's not get away from the point of this thread. Peter has made a wonderful progression from posting on this forum to enhancing the content of his knowledge and making it available on YouTube. His feelings about higher education were what drove him to do so. I'm OK with that.

Every time I see a post from Peter I read it carefully, He is the most proficient in motion control in this forum. So please Peter continues with your labor and do not forget to tell us when videos are available.

Peter, what texts would you recommend for those of us that are also self-taught (and enjoy continued learning)?

Thank you, Peter. As a "mostly" self-taught controls engineer, this thread:

http://www.plctalk.net/qanda/showthread.php?t=27057

Has always been my "goto", when applying PID tuning.

After doing this for nearly 25 years (since I was 18), I am always eager to learn more!

JeremyM said:

Peter, what texts would you recommend for those of us that are also self-taught (and enjoy continued learning)?

Click to expand...

As I pointed out above, system identification is a must but it requires programming. I have simple and advanced video about system identification on my YouTube channel. Most people will simply buy auto tuning software that will return the plant gains and time constants etc. However, if you can program at all it isn't hard to learn how to write your own. Look at my simple example.

There are plenty of videos on YouTube but they are more basic and text book example. My goal is to show the subtleties of different systems the text books don't address but are gotchas in real applications.

Look at the first two videos on system identification. One is very simple and I provided a simple example. The other is much more complex.

Brush up on differential equations, Runge-Kutta and Laplace transforms.

If I were a teacher I wouldn't give a passing grade to anybody that couldn't at least write the differential equation for their system

BTW, I use a mixture of Laplace transforms, state space and differential equations. Laplace transforms and state space are simple but they assume the plant is linear. Differential equations allow one to model non-linear plants. Differential equations with Runge-Kutta rule!!

Brian Douglas has a YouTube channel that isn't as "challenging" as mine but has more polish. The problem I have with his channel is that he doesn't answer the basic question we should all have "Why Bother". If can can't use or never will use it why bother?

proof said:

I appreciate what you are trying to do. However, I do think you will be more effective if you spend your time explaining what you think you audience needs to learn, rather than spending the time spewing contempt bashing universities and professors.

Click to expand...

I have said that system identification, pole placement and model control are important. The model control is very advanced.

I want ALL the control students to ask the question "WHY BOTHER?". "How would I use what your are about to tell me in a real applicatoin?" The students DON'T KNOW WHAT IS IMPORTANT. I want to point that what is important and what is not. The students are the customer. They should get their money's worth.
College costs too much now.

Peter,
As I've stated in other threads & others have mentioned here, your willingness to share your knowledge is much appreciated. Kudos!

The other night I watched the movie The Man Who Knew Infinity about the mathematician Srinivasa Ramanujan. I sometimes get the sense that Peter is The Man Who Knows Control Theory. The difference being that Ramanujan seemed to be a true savant where Peter has combined his schooling from University and experience and sheer will to discover, develop and present what he knows about real world closed loop control.

I think the videos are good but as Peter said, this is advanced material. There is a lot assumed of the audience, especially in assuming that you can imagine physical systems as equations in your head. I'm hoping to get some time to work out some of my own systems using the techniques. I'm also hoping to get a refresher course in Laplace Transforms when my son hits them in his math class in the spring.

At least I Have some experience with Runge-Kutta from school so that should just need light dusting.

My college text has matrices and Root Locus and Ziegler-Nichols. Only part of a chapter on pole placement ;-( So that would be a NOT-Recommendation for Feedback Control of Dynamic Systems 2nd Edition by Franklin, Powell & Emami-Naeini. I suppose you'd get some of the terminology and concepts out of it but old texts tend to be really expensive. Looks like it is up to the 7th Edition - Table of Contents - and is a lot longer but I'm not going to pop for the price...

Norm, you should look at the Why PID2 video. I made this video for people that do hydraulic servo control. I will be giving a presentation on this topic at the IMTS show in Chicago next week on Wednesday. Basically, I prove that when you are tuning a hydraulic system the errors decay at the same rate using a PI or PID controller. There is no advantage to using a derivative gain UNTIL the second derivative gain is added. Without the second derivative gain the rate at which the error decay is limited by the damping factor x natural frequency. This means the poor control guy has no control over how errors decay. The control guy can tweak the PID gains all he wants. He can make it response worse but he can't get past the limit imposed by the natural frequency and damping factor. Unfortunately the control guy usually must work with what he is given. I am sure the people reading this can relate to this. The control guy gets blamed when it is really the hydraulic or mechanical "engineers" fault. The second derivative gain allows the control guy to increase the gains without getting oscillations. The second derivative gain has often been a "get out of jail free card".

The MTS people knew this back in the 1980s. The problem is that there wasn't enough processing power back then nor did the feedback devices have enough resolution to take advantage of the advanced control theory. Back then MTS ( Temposonic ) had a TDC controller that only had a 80188 or 80186 on it just like we did. There wasn't anything either one of us could do back then. We had to wait for the technology to catch up.

Thanks for the IMTS reminder. I planned on going to see you and the other hydraulic presentations. Now I have to remember to sign up.

The first motion controller I worked with was the TDC. I was told it was an electric controller first (the EDC) and then was converted for use with hydraulics. I didn't really know anything about hydraulics back then. I was just the best with a laptop where I worked and got nominated. I do remember any time I tried to add D gain I regretted it. These were drilling machines and if it was just through holes I think we just used P. Spotfacing and chamfers required I as well.

Then I got to help start up the first Rexroth HNC-100 controller in the states with the engineer from Germany that flew over. I'm not sure what processor was in that but I have to believe it was faster than the TDC. The units of gain were super screwy and part of the programming was still in German at the time. That's the first time I used SSI and it had to be grey code. Not knowing anything, I once tried to change the signal direction by changing the Data+ and Data- lines. Because it was grey code it worked in a small area of travel but when it hit a certain position, I made a 30,000 lb machine jump. I was lucky it was made of tough Brazilian cast iron. Rexroth took a different approach in the HNC which was to only turn on the integrator when you got close to the final position. You actually had controls to turn on the integrator all the time if you wanted as well. I don't remember if it had double derivative gain or not. If it did, it was probably called something non-intuitive. Surprisingly, it did have what Delta would call gain scheduling to account for non-linear valves or systems. You couldn't use a complicated curve but you could have the gain follow the "kink" in some of the valve curves.