For Discussion: Academia vs Industry

Hello All of you Control Experts out there, As many of you know, since I have been posting in this fine newsgroup for quite a while now, I am doing a Masters/Electrical Engineers Degree at a US
university, I will soon be graduating and going back to my country, nevertheless for what I have been following in this discussion group I have the following question: According to most of you PID control is what is mostly used in industry, actually many books like the Ogata or Niese indeed verify this statement; and also I have barely seen any discussion about Hinf control, Robust Control, not even LQG or LQR which I particularly have found easy to implement at least in simulation (Simulink); if this is the case then why bother students to go trough all the math (note that I do not have a problem with the math as long as it is useful) used in Robust Control, Adaptive Control, Nonlinear Control etc... when classical methods and PID will be what will get the job done. Also if this is the case, then why haven't you people from the industry created a conscience in the universities that offer the control track that PID is enough for most of the cases.
Also, a little bit off topic I was reading in the IEEE Spectrum that in the last decade there has been a noticeable reduction in the amount of papers that people from the industry generate and therefore the academia is the one generating papers which are usually not only unreadable but most of the irrelevant for practical purposes; any ideas why is this happening?.
I would love to hear all your opinions out there
Thanks
Mariano
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

problem
offer
Universities teach you to *think* first, and give you the tools second. Academia in general has no clue what sort of problems (control problems, that is ;-) you might come up against in your chosen field and so teach you the whole lot. You become a more "well-rounded" ginger-beer that way.
It just so happens that PID and non-linear control are the most widely used in industry, but that doesn't mean you shouldn't know about the other options available.

the
one
As it is, people in industry have to work silly hours to pay the bills so they don't have much time to write Papers (or read them! ;-). If they *do* they obviously don't have enough paying work to do.
This is one reason that most Papers are given by people in R&D fields or are junior ("expendable") engineers. Nobody ever gets paid to do it...
A small aside: The above comment holds true for writing Standards as well! I don't know what it's like in the IEC, but Standards Australia are a private company whose main role is to make money selling Standards written by others for free!! (My pet hate at the moment)
Now I come to think of it - I don't get paid to type this either.. (wish I did! ;-)
Cameron:-)
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
PID is a simple enough method to use, with a long history in industry. It has been implemented on just about every modern control system in some form. It works. You don't need a doctorate to understand it. It doesn't require a lot of data storage, and is applicable in most situations.
The newer techniques might be great, but unless some manufacturer is willing to stick their neck out and make it a standard algorithm, they won't get used. We generally cannot implement non standard control schemes on PLC's and DCS systems without a lot of extra effort.
We could go a bit further and ask why we should change control schemes. How much better than PID will the new techniques do? The effort to improve must be exceeded by the benefits of the improvement. Basically, PID controls are mature, well known, easy to implement, and they work. That makes a very hard thing to compete against.
Michael

for
US
have
problem
offer
the
one
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

problem
offer
For the most part all I have used is a PID with feed forwards in motion applications. There is nothing tricky about PIDs and feed forwards but I still feel that most people on the plant floor regard this as black magic. Yet the people on the plant floor still manage to keep the system going which is the most important thing. Also, the basic structure of the PID algorithm does not change depending on the order of the plant being controlled. Just the gains need to change. This is the PIDs strong point and weak point. Personally, I think PIDs are not as good as some other techniques. For instance, I think that Sliding Mode Control is much better and simpler than a PID for normal temperature control applications. Yet no one knows or says anything about it. There was a thread asking about SMC last month and no one replied. The problem with SMC is that the switching equation varies depending on the number of poles the plant has. The dead beat controller is totally dependent on the plant model and the desired transfer function. This bring up another point. No matter what controller one decides to use, the key is knowing the plant model. I don't think this gets covered enough either.

the
one
If I find something new that works better, I will keep it a secret for a competitive advantage. I am skeptical of papers written by grad students that are just trying to get a higher degree. Most techniques are not tested against a wide variety of plants. Most, perhaps all, of the suggest techniques don't offer better performance that what I am already using and are harder to implement.
In the end we are limited by the skill level of the users that must keep the controllers working or by the inability of the controller designers to write control algorithms that just 'automagically' work. I am working hard on the latter.
One more opinion. ZN tuning is a waste. It doesn't provide for the fastest settling time or minimize overshoot. Slow settling times and over shoot both potentially waste money.
Peter Nachtwey

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Peter,
ZN isn't supposed to provide the fastest settling time or to minimize overshoot. It is based on quarter wave damping. I'm sure there is a great reason for that. I would guess that it was developed for the old pneumatic controllers, rather than for electronic ones. It's also fairly easy to apply and it works ok. A lot of systems do just fine with things varying around a bit.
Some of the newer tuning methods will do a much better job for a system. If these newer methods require extensive data (such as timing data) to perform, then they can be harder to put together. If they come out in the form I've seen on a few control papers, it would be a micracle for a neohpyte or even a moderately experienced contro person to figure out what they were talking about.
Michael
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

great
pneumatic
So why do I see ZN still get mentioned so many times on this forum and in text books? ZN still cost money and one can do better.

system.
Getting this data is easy on a Rockwell PLC. The RSLogix has a trending capability that graphs analog values. The best part is that the trend can save this data in a .dbf file which can be imported into Excel. Many systems can get this data. Excel has a LINEST function that can be used to compute the coefficients for a to second, third or more order model in the z domain. For temperature control a second order model is usually enough. From this data one can compute the PID gains and openloop compensator, what I call feed forward and process people call bias..

what
I agree about the neophyte. However, I think a moderately experienced control person can figure this stuff out if the procedure were explained in simple terms.
Check this out. It is about temperature control using SMC. The .htm file was developed using MathCad. http://deltacompsys.com/out/Temp%20SMC.htm
This blows away PIDs with ZN. It is also much simpler.
The trick is to find the model, but this is easy using Excel's LINEST function although usually I use MathCad. Even if one doesn't have the model, sliding mode control is pretty forgiving and one can use trial and error until perfect results are obtained.
Peter Nachtwey
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Sounds like "tuning" to me.
Bruce
Peter Nachtwey wrote:

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

BTW, what is the latest in the Solaia vs. Rockwell saga?
Andrey Romanenko
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Ziegler and Nichols proposed quarter-amplitude damping for proportional control as a useful compromise between a high-gain system with excessive cycling and a low-gain system with excessive offset. They also make the point that the response of a controlled system has to be decided on the merits of the particular installation, and that there is no such thing as a "one-size-fits-all" formula to get suitable tuning values.
Another reason why tweaking of settings is always necessary - but one the academic fraternity seems to find difficult to accept. Modelling of industrial processes in the "wet" industries is always going to be an approximation, when even well-used phenomena like heat transfer in an exchanger is subject to experimental "correlations" and empirical formulae.
Bruce.
Herman Family wrote:

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Why do you say academics find this difficult to understand? Of course, when you teach intro control systems, you generally assume that the plant is known and constant, or at least measurable and constant. If you take more than one or two control courses, you invariably find that things are not that simple, and you learn approaches like adaptive control and model estimation techniques. I would think that most of the advanced control techniques were developed academically.
To assert that academics are unaware that plants change or have nonlinearities is kind of like saying that academics in math don't understand multidimensional calculus because its not taught in calc I.
Scott
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Model predictive control for one assumes a good model. This model must be based either on a purely analytical approach to the plant - but this is often either not very accurate because of all the unknowns, or based on empirical data anyway - or on information derived from tests on the actual operational plant.
Commercial reality, whether we like it or not, is that it is difficult enough finding time during plant commissioning to tune a handful of PI or PID controllers. At least with the PID approach you can throw in some numbers and get adequate control that can be driven to get things operational. But once the plant is up to full rates it is very difficult to persuade the management to play around with it. Fine-tuning a PI or PID control loop is relatively straightforward. But if a controller has more than 2 or 3 knobs to adjust, or a model has more than 2 or 3 parameters, the process of setting it up can get out of hand.
The testing required to develop a reasonable model that is better than the one derived from an open-loop response test as used by Ziegler/Nichols is pretty hard to organise - especially when there is a need to cover a wide operating range.
The other major difficulty with "advanced" control methods that is often overlooked by academia is that the plant has to be operated and maintained. It is difficult enough for relatively highly skilled operators to start up a boiler using the standard cross-limiting combustion control scheme: more advanced techniques may well be totally outside the capacity of operators to comprehend. And the effect of having a transmitter go out of calibration needs to be taken into account as well - although smart transmitters make this unnecessary for calibration checks, problems with impulse tubing or other installation difficulties can also make readings go crazy.
I would agree with you that most of the advanced control techniques were developed academically. On the other hand, most of the techniques that are widely used in practice were developed empirically - starting from the Watt governor. I would suggest that the empirical approach was used because there was a perceived need and a solution was found that met that need, without a whole lot of academic analysis. These solutions were developed in the workplace and included the elements needed to allow them to be put into operation.
Bruce.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Many adaptive control strategies do not assume a good model, and in fact are designed to simplify design in response to a bad model, nonlinear, or very high order plant.

some

difficult

No problem with this. The "practical" approach will be every bit as cumbersone as the "academic" approach for a case like this.

Or when your plant is nonlinear.

often

You keep using this "overlooked by academia" phrase, and its just not quite true. Yes, there's no substitute for field experience, but nowhere have I ever been told in an academic environment, or have I told a student in such an environment, that the plant does not need to be operated and maintained. Does this necessarily come up in a basic course on control theory? No. This is why we call it "theory".

were

We give the students the tools we think they need to do the jobs we think they're going to be doing. Can you learn to twiddle the knobs on the D path and I path to make a system behave nicely? Of course, and a person who can do that reliably is a valuable technician. Somewhere, that jump between technician and engineer happens. When a person understands that fiddling the D knob deals with one kind of error, and twiddling the I knob does something else, he's moving in the direction of engineering. When he understands pole placement, group delay, stability, he's pretty much there. The examples and strategies taught in the academic environment are examples developed to teach these concepts, because we think control engineers that understand them are more valuable than those who don't. We then go on to teach the student how some practical things, such as backlash, dead zones, time delays, time variations in plant parameters, quantization problems, and other nonlinearities, can impact their designs. Those students who move in a more practical direction pick up the more practical skills readily enough, because they realize that what they are doing is a practical example of dealing with these nonlinearities--they learn why their "paper" strategies need to be tweaked, and they start to recognize when their academic approaches are unlikely to help them. We can argue about whether this is the right approach or not, but there it is. It's easy to take what we teach the students in Control I and II is the "academic" approach to control engineering, and that academic engineers can't grasp real situations, but its simply not true. It's just a method of teaching, not a measure of what a control engineer in an academic environment can understand.
Scott
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Academics get brownie points for doing research. Double chocolate brownie points require that the research is into something new and original. Research into effective ways of applying the techniques developed a couple of academic generations ago will deliver stale crumbs.
Bruce.
Mariano I Lizarraga wrote:

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Mariano,
In most cases there is no reason to implement the more adcanced methods. I know that you can find LQG control of papermachines (drying of paper) and making of mechanical pulp(I know at least of one master thesis made on this subject a few years ago). A small decrease in variance means that you can save millions of dollars(you can move the SP close the the desired value).
Advanced control methods have also been applied to the control of large (oil/diesel/gass)engines.
I know at least one large company that use MPC(modell predictive control). According to my source it is very suitable for complex chemical processes.
A reason why you will not see much discussion is that a lot of this is considered secret and strategical investments for most companies.
Niclas

for
US
have
problem
offer
the
one
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Hello All, I want to thank you all for your very useful insights, I will share some of your insights with my control systems teachers.
Thanks
Mariano
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Looks as if you have opened up a lively discussion on industrial control.
I have been a control engineer since 1970 and seen many papers, discussions, etc. on control schemes. The most striking was from a colleague of mine.
He explained that PID control was the same as driving a car down a road with the front windscreen blcked out and only able to see through the rear view mirrors.
I believe that this is the most revealing quotation I have ever heard on the subject of control. If we are going to control anything then we must know where we are and wher we want to go. Knowing this we can drive the system accordingly - just the same as the car. I have adopted this technique in batch control with quite dramatic improvements in quality and throughput.
We must understand the basic requirements and the means available for control in order to drive the process. All theoretical techniques are superflous if they ignore the basic requirements.
This does not mean that control theory is superfluous. One of the most important aspects of advanced control is that of state variable estimation techniques, which I was introduced to in 1968 and cannot see that the academic world has pushed it much further forward.
Anyway enough from an 'old engineer', I wait for the 'new engineers' to catch up.
AJ
Mariano I Lizarraga wrote:

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Polytechforum.com is a website by engineers for engineers. It is not affiliated with any of manufacturers or vendors discussed here. All logos and trade names are the property of their respective owners.