PID questions in finding Kp Ki and Kd

I'm trying to run a PMAC motor in a closed loop. The program used by the
TMS320F2812 implenents a PID control element to control the Iq and Id.

Problem is that I don't know how to set the gains on the Kp Ki Kd.

There is an AVR app note that the motor driver "plays" with the motor to
determine the gain factors. What do I need to stimulate and measure to
figure out the PID gains?

Thanks

 

The no brain solution is Ziegler Nichols Method. Google gives somewhat 418k
hits for it.
If you would like to understand how does this stuff work, you might want to
get a book on Control Systems by Tim Wescott.

Vladimir Vassilevsky
DSP and Mixed Signal Consultant
www.abvolt.com

 

Thanks for the plug. If you do get the book, keep an eye on my web page
-- I taught a class this spring that helped expand the errata, and if I
get enough time free I'm going to pour all of the homework problems and
answers into the addenda.

Ziegler Nichols is a good way to get starting values, but it tends to be
aggressive with the gains and produces systems that often don't have very
good margins. Once you understand Z-N look around for the Astrom-Hagglund
method. It does the data collection step in much the same way as Z-N, but
it uses different conversions to get controller gains that are more
conservative and safer (mostly) in practice.

--
Tim Wescott
Control systems and communications consulting
http://www.wescottdesign.com

Need to learn how to apply control theory in your embedded system?
"Applied Control Theory for Embedded Systems" by Tim Wescott
Elsevier/Newnes, http://www.wescottdesign.com/actfes/actfes.html

 

Thanks,

I've read your articls on PID without a PHD and things are clearing up.

TI has a multistep process to build feeback contolled PMSM with resolver
FB.
1) is to run the motor open loop with and Id and Iq command
2) is to apply the Id and Iq feedback and close the torque control loop
3) is the aply thespeed control look using resolver feedback.

We have run the motor with a DC paramaters for Id and Iq. Id
= 0 since we are running a PMAC motor with resolver feedback. We run these
values into the Ipark- Iclark block to generate the 3 phase SVM signal for
the motor windings.

In the open look it works OK but pulls WAY too much current. We have to
turn the power supply voltage down to keep from saturating the current
sensors (2A). There is no load on the motor as we are just starting out. my
recommendation is to reduce Iq until the current is under control and the
motor still spins openloop. The FW guy is resistant to this for some
reason.

One the open loop is stabilized, the next step in the build procedure is to
connect the clark/park Id and Iq output from the measured 3 phase currents
and for a feedback loop with a difference block subtracting the target
Iq and Id from the measured Iq and Id to generatethe error. That is feed to
the PID block then to the ipark/iclark to generate the corrected SVM
waveforom on the motor.

Using the Z-N method, I should set the Kp = 1 and Kd = Ki = 0? Then make
the Iq command step up and down between values in the operational range of
the motor and look at the feedback iq response. Should be slugish.

Increase Kp until you get some overshoot on the steps. Then begin
oncreasing Kd until the overshoots damp out. Then increase Ki to reduce the
steady state error.

Is this the proper method to apply Z-N for motor controll of the current.

Once the current loop is stabilized, a speed control loop will be added to
provide the Iq target. The speed control Fb will come from the resolver. I
do the same process with this on by stepping the speed command and watching
the speed Fb response?

Does this sound like the proper approach?

thanks.

 

I havent looked at the TI PMSM code, but the normal approach is to have
a current setpoint, and a closed-loop current controller, that then
generates a voltage setpoint to the waveform generator. If you try
running it open-loop, with a current setpoint, it'll saturate PDQ.

What you need to do is bypass the current controller, and provide a
fixed voltage setpoint to the waveform generator.

you might be better off with some inductors rather than a machine to get
the open-loop control working. check your waveform generator is working
nicely (I built some 8th order passive bessel filters that Tektronix
P5200 probes plug into, to measure the fundamental voltage waveform from
a hairy-assed PWM mess, for this purpose).

once you are happy that you can produce the voltage waveforms you intend
to (IOW the WG is working), then add in the current controller. If Id =
0 then Vd = 0, so you dont need the Id controller at all.

And yes, you start with a nice small Iq. My personal preference is to
use inductors, because there arent any rotating bits to worry about, and
you dont have to worry about getting the darned thing started.

Once you have got the current regulator working at a variety of currents
with an L-R load (mostly L), you can do step responses etc. I find it
handy to feed actual Id, Iq out a DAC port, so you can see whats going
on in the SRRF on a scope - there are devices called vector visualisers
that implement Parke, Clarke transforms with analogue ICs, to let you do
this with V,I measurements directly, but if you have a DAC port its way easy

with the current regulator working and a machine attached, organise to
slowly bring it up to speed with open-loop voltage control (if you ramp
the voltage slowly, the unloaded machine will easily follow it), then
kick in the current controller. at this point you can tune it, using
small step-changes in Iq*

you dont need any Kd for the current controller. PI is fine, and thats
one less parameter to mess with.

a crude analysis of ZN is:

- crank up Kp until it gets pissy
- wind it back a bit
- ditto with Ki


but really its not too hard to sit down and write some control-loop
equations, so you can figure out the "ideal" Kp, Ki for whatever
response you desire. Dynamically its just a current controller providing
a voltage setpoint which is driving into an L-R circuit.

In practice its more complex, there are sampling issues etc, but as long
as your sampling rate is fast c.f. the machine rated speed, they wont
affect you too much - and the analyutic approach provides a good
starting point.

 

Terry,

Sincere thanks man.

Its really starting to clear up for me now.

You are correct, the open loop build (phase 2) had ramp generator to run
theta (phase angle) as a user provided frequency (RPM). The theta is used
to feed the Inv_Park and Park transform modules.

Two user inputs are Vd and Vq into the inverse park module and the alpha and
beta outputs of that feed the inverse clark module that provides the
modulation signal to the PWM registers.

So you are correct and in the open loop build, we are simply setting the
motor voltage amplitude across the motor and the currents are what ever they
work out to be.

During this step we look at the measured currents on the line, at the output
of the Clark and at the output of the Park module.

Good Good. This corrolates well with what I've seen in the lab.

In phase 3 we are forming the current loop. In this loop, the Iq and Id
output of the Park module in the feed back loop is compared to a Iq and Id
(Id = 0) reference. The PID block generated a voltage signal appropriate to
regulate the current at a particular value.

I started with a Kp = 1 and Ki = Kd =0, motor just buzzed. Thats where I
left off.
Monday, I plan to increase Kp util its at least kind of functional then keep
going until it goes from being "kinda functional" to disfunctional. then
back off to near the top of the kinda functional range. (Z-N method
recommend halfing the disfunctional Kp value)

Then step change the Iq command by a small amount and watch the response as
I add Ki gain.
leave Kd =0 for now.


Follwing that a similar approach is used for the speed control loop that
ouputs the Iq command required to maintain the speed setpoint at various
loads.

In think this will get me there.

Thanks again Terry,

P.S

If, theres anything above that looks incorrect, or you have a few more
nuggets of info, please let me know.

 

no worries mate. someone might as well get some use out of this
knowledge....

Im not familiar with the TI code, but you might want to think about the
numerical scaling of the gains - is the TI code per-unitised (Kp = 1 =
100%) or is it scaled in V/A etc? if the latter, then 1 might turn out
to be an astonishingly small gain.

actually if its the former, Kp = 1 may also be a very small gain if, for
example, 1PU = 8192.....

not really, you've got the basic idea.

It took me a long time to grok this stuff - didnt really do so until I
went to a seminar with Ramu Krishnan and Joachim Holtz; I asked Prof
Holtz about tuning current loops, and he basically said "how come people
dont understand this stuff" then cranked thru the maths. Recognition
eventually dawned, and nowadays I always calculate my gains before I
start fiddling.

Cheers
Terry