Some (me) still use them. Some (Peter) don't.
But Peter doesn't post here much any more, so I guess the answer is "yes".
I rarely use root locus plots, and when I do they're either really simple
ones that I have memorized, or I generate them on a computer. I often do
frequency-response design, for which it is convenient to generate a Bode
plot and a Nyquist plot at the same time, as they each present different
and valuable views of the same information.
In a way I still do. I use pole placement so I work the problem
backwards. Instead of increasing the ONE controller gain through a
range and watch where the roots/poles go I move the poles and
calculate the gains. When I was learning control I found the root
locus as used is useless. First who moves just one gain? A
controller has 2,3 or 4 grains to set. Now how would you do the root
locus moving all those gains around. All the gains must work
together. Tweaking one at a time is not very efficient.
I find Nyquist plots to be equally useless for the same reason. If I
had to adjust one proportional gain then OK but figuring out the curve
with 3 or 4 gains is nuts.
There are better ways. I have symbolic formulas for calculating
controller gains for many different types of systems. The allow me to
place the poles where I want which is usually on the negative real
axis in the s plane.
Sometimes I use IMC when talking to people about temperature control.
If I were teaching I would focus more on system identification and
modeling mention the ancient history only briefly.
I am still in contact with one of the other guys that posted here in
the past. I meet him on-line as a undergrad and now he has his
master's degree and teaches. I have asked him why he teaches the same
old control in the same old way. There is a lot of inertia when it
comes to education. I think a lot is a waste of time.
I look from time to time but signal to noise ratio is too low if you
know what I mean. There are no moderators removing spam. I get e-
mails from people from time to time and that keeps me occupied. I have
made a SOPDT temperature simulator that is on-line on the interenet.
I can demonstrate many different forms of PID, SMC and Smithd
Predictor. I recently did a webinar using gotomeeting for the
In a way this group is better like this because there aren't any nuts
spreading bad information.
I'm guessing that the exam you mention is aimed mainly at practitioners at
the process end of control, ie. big tanks and pumps. I find that dynamics as
a whole is a regrettably small element of my job, and when a need to apply
it arises, it can almost always be handled using intuition as much as
rigorous analysis, although simulation is one of the tools that I really
A lot of my time is spent on platform matters, and non-regulatory things
like MMI, reporting and alarms. Many of the applications I implement are not
related to continuous loops, they include things like logic, switching and
The vast majority of SISO, approximately linear problems in plants can
usually be left to the techs, in fact one plant I was at recently has a
policy that "all flow controllers will have tuning settings of X, Y, all
level controllers will have settings of Z, W", etc, and it runs smoothly on
those settings that they started up with. If something comes to the
attention of the control engineer, it's likely to be multivariable,
significantly nonlinear, or have some other unfriendly characteristic that
renders many of the traditional techniques inapplicable. The best estimates
you can obtain of plant dynamics are often very rough, and plant behaviour
changes regularly, sometimes to a large degree. Noise is generally quite
non-random, being mostly created by people through things like lineup
switches and feedstock changes.
You get the picture. It's good to go through the bode/nyquist process to get
a feel for what makes loops behave the way that they do, but IME their use
in plants is pretty rare. If I had to take the exam I'd fail for sure.
Thanks a lot.
There are a lot of textbooks available now with a lot of efforts in
explaining the root locus plotting, which can be simply drawn using
Matlab. It seems that the college professors should hold a meeting and
discuss which is more useful for control systems teaching. Tell the
student more about modeling and PID stuff.
I still remember that I have learned PID digital implementation,
control loops in motor control in college, besides the root locus and
Nyquist plot. I think perhaps those content may be reinforced.
Even the professors themselves have no actual experiences in real
control system engineering. My professors are mathematicians and they
surely pay no attention to the real world stuff. Perhaps college have
to hire some real-world engineers to teach the college students.
Anyway, I passed the exam, then I need not revisit the ancient tools
Being able to sketch a simple root locus on a napkin is a good way to get
an intuitive grasp of a system's behavior. Certainly, being able to
interpret a root locus is good, and to really do that you need to push
yourself beyond just sketching them on napkins -- either by doing dozens
by hand, or by playing with hundreds on the computer.
The super-duper whiz-bang optimal control techniques require a lot of
information about the system that you may not have. Ditto pole
placement. So if you use them you need to either do some system
identification and then some shading of what you tell the algorithm
(Peter really likes pole placement, but when pressed he'll say things
like "of course I don't try to place the poles _there_, the system would
end up unstable!"). Robust control techniques take this into account,
but you still have to know how much your parameters vary.
Which is all the long way of saying that out here in the real world, the
tried-and-true techniques still have some currency. So do things like
pole placement (or it wouldn't work so well for Peter) and robust design
techniques. About the only thing that you've mentioned that isn't really
current is plotting root-locus plots by hand -- and if you think while
you do them, those lend you a life-long intuition for how a system
behaves with changing parameters, so knowing the material is hardly a
(Note: I rarely do root-locus plots on paper any more. But when I'd
contemplating changes to a system, I'll often visualize simple ones in my
head, sort of as a check plot on what I'm trying to do.)
I thought the first part was funny.
I have studied Hinf and compared it with Kalman filters. It is just another=
way to compute the corrective gains. There is no magic. One the whole th=
e Hinf gains are just a little bigger than the Kalman gains. So what. I a=
m not impressed.
What would impress me is updating the system transition matrix on-the-fly w=
ith some continuous system identification.
I really prefer to model the system and then use the symbolic formulas to c=
ompute the controller gains as I move the poles around. Usually I move the=
m on the negative real axis to the left until there is some un-modeled limi=
t. A big factor is feed back resolution and how the derivative gains make =
the output look very noisy. I can also place the poles and the low pass fi=
lter pole for the control output. This helps filter out the control output=
noise ( I know it is due to non-linear feedback due to digital quantizing =
) but if you can trust your model it better to use an observer.
OT but still talking old techniques. I have a SOPDT temperature simulator.=
I actually tried ZN for the first time ever to tune the temperature loop.=
The results were bad considering the effort that it takes but the ZN techn=
ique doesn't control so well on long dead times. It also takes a loooonnnn=
g time to adjust the gain to find Ku when the cycle time of the oscillation=
s is 12 to 13 minutes. There is lots of tweaking gains and drinking coffee=
between the cycles but little up front knowledge or effort is required.
Along the same lines of the observer. I find that using a Smith Predictors=
does wonders for temperature systems with a large dead time. I can even u=
se FOPDT tuning on the SOPDT temperature simulator and it works reasonably =