What is the major at college for control engineers?

ME? ChE? or ECE?

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On Mon, 28 Jan 2008 04:34:44 -0800, workaholic wrote:

Control theory is either a very versatile sub-discipline, or a homeless one.
You can study (and even get a doctorate in!) control theory in the context of all of the above disciplines, plus mathematics, economics (no, really), biophysics, and probably others as well.
If you're pondering getting a degree in control systems, I'd suggest that you think about just what you want to do. Whatever else you do, keep in mind that the "system" in "control system" is an important part -- the more you understand about mechanics, electronics _and_ software (plus chemistry, if appropriate), the more you'll be able to wrap your brain around the whole problem to come up with good, effective solutions.
If you want to be "the servo guy" at an aerospace or related company, then ECE is the way to go -- few people in those places understands what control theory is, so whoever designs the boards (or sometimes the software) will be the one closing the loops, regardless of how good they are. With an ME at such a place you'll find that you're working for the EE who designs the loops.
If you want to design big machines with motors and nice rectangular boxes that sequence and control the motors, then you (probably) want to get an ME. You won't get to actually _write_ the control rules (unless you buy your boxes from Peter), but you'll get to tune them, and curse at their limitations.
If you want to be an economic adviser to the white house, drive a big shiny car and wear gold cufflinks -- get a law degree, and forget about control systems :-). Or maybe get an economics degree with lots of control classes.
For a job at a university, or slaving away in a back room at The Mathworks, a mathematics degree is ideal.
--
Tim Wescott
Control systems and communications consulting
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Tim has good advice. For the most part control theory is control theory and will be the same what ever way you go. There are small differences in what is stressed but not enough to determine your decision. As Tim said, you want to do should determine your direction. What ever way you go ( ME, ChE or ECE ) make sure they have a good program for system identifcation and modeling. I don't think system identification and modeling is stressed enough. Just about everything else depends on system identification. Calculating gains for PIDs, Kalman filter, H inf filters, observers, compensating for dead time etc,
Peter Nachtwey
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

You lucked out on this question. The prior advice was great.
You may find that there are interesting differences in how control is applied in each group. The great news, already stated, is that staying in whatever field you are in will provide some great opportunities for understanding and applying control theory.
I've found it fascinating to look at the time scale of each group. My EE's often talk about loops in the kilohertz and megahertz range or shorter. My ME's worry about things in the fractional hertz to the kilohertz range, and as a ChE, I worry about loops in the hertz to cycles per hour range, or slower. Managers tend to look at loops in the hours to years range, with increasing length as they move up the ladder. An economist looks at loops in the weeks to years range.
Michael
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Thanks for all, and how about be a researcher studying the pure mathematic problem unrelevant to the control object? Can one achieve that without any knowledge of a specific industry plant, and focus on formulas and theories?
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

I don't think that is a very good idea unless you just want to teach. So much has been done already that hasn't been widely implemented. What I think is lacking are people that can implement or the advanced control methods. The are plenty of good algorithms out there already that I can't implement because they take too much processing power or they add more complexity than what the average joe can handle.
Peter Nachtwey
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
wrote:

I don't think that is a very good idea unless you just want to teach. So much has been done already that hasn't been widely implemented. What I think is lacking are people that can implement or the advanced control methods. The are plenty of good algorithms out there already that I can't implement because they take too much processing power or they add more complexity than what the average joe can handle.
Peter Nachtwey
Really interesting discussion.
IMO the problem with a lot of the theory is that it won't necessarily work in an environment of uncertainty. In particular with industrial processes, non-stochastic noise is a huge killer. My biggest struggle is with disturbances that emanate from events rather than nice, stochastic noise. Operator fiddle, pumps trip, units come on and off. You only need to pull up some trends for the typical plant to see how bumpy things are.
That makes the application of formal methods a lot harder, because the question then isn't 'will it work', but 'what is the domain within which it'll work?' Couple that with the fact that non-stochastic noise is so hard to quantify and you have the reality of applying automation to plants. Everything works some of the time and the job becomes one of patching up the problems, and keeping the users happy.
I'm also not sure that more computing power will help with some of this. At least until the machines approach or surpass the human mind. And when that happens we'll have to deal with their psychoses and neuroses, and be prepared to switch them off when they go feral.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
wrote:

I agree and the answer to is minimize the uncertainty. That is whas system identification is all about. Once the system model is known then calculating the gains is easy.
In particular with industrial processes,

I don't think you are using the word stochasitic right but I understand what you are trying to say. System identification can yield a sum of square error or other measure between the actual plant and the estimated model. I use the sum of squared error to provide an indication as to how good the estimated model is.

Disturbances happen all the time. I don't understand the point. If a properly simulate a disturbance then there will be bumps even in an ideal simulation. That is what a disturbance is. However, with a good model I can predict what the worse case response. In some cases one can predict the change in control output like when one puts cold material into an overn. The rate of cold material entering the oven can be used to predict a bias to compensate for the extra energy required to heat the extra cold material.
A related topic is feed forwards. I am a big believer in feed forwards. Feed forwards are derived from the inverse of the model. The allows one to calculate the feed forward gains that will provide a open loop gain of 1. Essentially the control output is predicted from the SP and its derivatives. It the predictions are close there will be little error.

Look at Eggi's problem. He has a model and uncertainty factors. The gains for his ideal model can be calculated. This is easy. It is then possible to varry the model 1000 times or more and plot the closed loop poles and see if any get dangerously close to the RHP. I know I can place the poles were they will not come close so my tuning will always be stable. Usually I am upset if there is over shoot because motion customers don't like over shoot.
Couple that with the fact that non-stochastic noise is so hard

Patching is bad, it is not the way to solve the problem. Knowledge of the system is good. I hope workaholic pays attention because I think system identification is the key to success.

We talk about this al the time. Real time system identification is ultimate goal but it will take a lot of processing power. I don't see why this isn't possible now with with slow process systems. You wouldn't believe what is possible when the theory is properly applied, but the key to it all is system identification so one can predict how a system will respond...
Peter Nachtwey
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On Tue, 29 Jan 2008 00:43:11 -0800, workaholic wrote:

I'm with Peter on this (mostly). There is some applied research going on, but a whole bunch that's esoteric, inaccessible, requires too much computing power (today at least), or some combination of the three.
If you do want this, then get a degree in control theory from a mathematics department, plan on getting a PhD, and expect that it'll be more difficult to get a job in industry (although I do know a woman who got a job recently with a fresh PhD in control, and who appears to be doing well).
Were I to win the lottery*, I'd go back to school for a PhD in control systems, with an aim toward finding ways of taking some of the nifty but oh-so-abstract theories in nonlinear systems stability and turning them into easily visualized forms, much the same way that Bode and Evans took oh-so-abstract polynomial transfer functions and turned them into something that could be visualized on a Bode plot or a root-locus plot.
I don't know if I'd _succeed_, but that's what I'd _try_.
* Don't hold your breath -- I never buy tickets.
--
Tim Wescott
Control systems and communications consulting
  Click to see the full signature.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Good goal, but why would YOU need to go get a PhD in control to do that? By now you should have PhD from the school of hard knocks. Who needs a lottery?
I don't think the visualization needs to be much more than a non- linear line. I think the hard part is identifying the non-linear features. Once they are known the rest is much easier.
Peter Nachtwey
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
On Tue, 29 Jan 2008 13:13:02 -0800, pnachtwey wrote:

Because I want the faculty position that funds it. In an age where you need a BS just to be qualified to sweep the floor, I don't think I'll get that without piling it higher and deeper.

--
Tim Wescott
Control systems and communications consulting
  Click to see the full signature.
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.