Re: Freqmod test result & Q

[Chris Wagner <clwagner-at-eecs.wsu.edu>]

On 29 Jun, rehenry-at-up.net wrote:
> 
> Chris
> 
> Hey thanks for the scope report.  This kind of stuff is great info. You
> might build a little r/c network to connect to the step or clock pin and
> you could get a curve trace of voltage based on the frequency.  This
> would give you some sort of parallel graph with the stuff you can graph
> from within the PC.
> 
> Now you did it.  I wrote the pantograph example as a thought experiment
> or explanation and you want to prove it can't be done.

Hi Ray,

No, I really was asking because I did not know if my assumptions about
the effect of the control-loop variables was remotely theoretically
correct.  I like to try to understand the configuration parameters. 
The pantograph example is exactly the kind of behavior one wants for
open loop steppers.

> Any electrical
> system will fall a bit short of the mechanical.  My last exposure to the
> calculus was in the 60's so you are without a doubt way ahead of me with
> the maths behind these things.  When you get if figured out, you can
> explain to me how this stuff like ff0 works.  Gain of 0 may be a problem
> but you might try what you are with a gain of 1.  Otherwise my thoughts
> are below.
> 
> High gain. 1000 as shipped may be okay but you could up it some.
> 
> High acceleration.  I sent off an ini for a rotary axis the other day
> with accel set to 150 and it seems to work okay.  You will need to set
> deadband even with nothing out there so you don't get hunting between
> adjacent steps.  Give yourself a bit more than half the distance between
> steps for this and you should be fine.
> 
> When you get to the amps and motors for this, you will need to oversize
> these components so that you get the quick response from motors driving
> a mass.
> 
> Hope this helps.  Get back with your results.

Everything helps at this stage!  I have a freqmod setup (just P not
zero) that runs.  It is clear from my tests and other list posts that
tuning freqmod to run steppers can be difficult, and tuning is crucial
to get good motor behavior at the higher step speeds.  My current setup
(by the scope and numbers) should run cutting feeds just fine, but it
looks like the rapid feed could be rough - time to try some motors.

>> 
>> I see two primary types of "jitter" on the step clocks. 1) The time
>> quantization for the [task] section "period" variable.  This is an

I think that it should have been [EMCMOT] not [TASK] section period.

>> expected effect in a digital timing generator.  2) Some jitter about
>> the task_period, presumably due to interrupt latency variation.  I have
>> not run motors yet, so I'm just commenting about what I see on the
>> scope.  Both of these effects should also be present to some degree in a
>> steppermod configuration.

There is a third type of jitter - the algorithmic jitter (both freqmod
and steppermod have this effect, perhaps it is easier to tune
steppermod?).  This jitter is at a longer period then the real-time
timers and interrupt latency jitter I reported previously.  I have not
figured out which variable controls this period.

I bet this algorithm jitter is the one that has caused the problems
reported here on the list. This causes the step clock to oscillate
about some nominal running frequency.  For rapid feeds the step clocks
look pretty rough, which could easily cause the rough operation
reported by others.  A tuning issue again?


Thanks for the comments, Ray.

>> 
>> My question, ideally should P=I=D=0, PosFF=1,VelFF=AccFF=0
>> provide the pantograph-like behavior?
>> 

-- 
Recreational Calculus - Just For Fun!

Chris Wagner
clwagner-at-eecs.wsu.edu