John,

Thanks for the response.

By dither, I am trying to explain how the motor is not stationary at an
idle or rest condition.  When stopped at 0 Velocity with the drives and
linuxCNC enabled, the motor constantly seeks a position between linear
scale graduations (0.01mm in this case).  I can see end of the ball screw
turning and the DRO readout in Axis will change between say 0.0004" and
.0008".  The higher the Pgain the faster the movement.  I've tried some
Igain, but doesn't straighten it out.

I have not tried the deadband equal to one or two, but certainly will do
that in the morning.  That was also a a forum suggestion from PCW, however
he also said w/ a coarse scale you might not want to loose the resolution.
This was partially why I thought the 1um scales might be a good idea
because I don't think I'd care about dithering between .00004 and .00008"
or setting a deadband (maybe of 2 or a tad more here?).

The 7I87 is remote isolated eight channel +-10V analog input card that is a
RS422 smart serial interface.  I've read on the forum w/ a resistor it can
be used to +-20V, which is good because my tachs are 9.5V/1000rpm w/ a max
speed of 2400rpm.  Also, PCW thought this solution still might seek scale
graduations (dither), only at a slower frequency.  If there was a consensus
this would work, I'd be thrilled to try it, but PCW didn't seem to think
this was the answer.  At one point he suggested encoders on the ends of the
ball screws (hmmm, sounds familiar).

To use rotary encoders on the motor, I'm assuming I need to leave the tachs
for the old West Amp drives?  Also, the encoders should have finer
resolution than the linear scales, I think 4:1 for motor encoder to linear
is what I heard?  Does that seem right?

Lastly, to beat a dead horse, I thought the higher resolution linear scales
would be easier to mount because of the existing bracketry.  To add the
rotary encoders, I need to add mounts to the end of the ball screws or
backs of the motors.  Maybe I'm being dumb here...

Thanks again,
Mark

On Fri, May 29, 2015, at 09:23 PM, Mark Johnsen wrote:
> I have made huge strides on my Supermax conversion from an Anilam Crusador
> M to a 5i25/7i77.  I was very happy to have the encoder resolution and
> motor/encoder directions setup so that I can start tuning.  I was able to
> follow John Thornton's Velocity tuning instructions, but I have the
> almighty dither...

What exactly do you mean by dither?  Axis moving back and forth when the
commanded position is constant?

Have you tried deadband equal to one or two counts of the scale?

> Specs for the system are West Amp Servo's w/ tach feedback and 0.01mm
> resolution glass scales connected to the 7i77.
>
> Has anyone been able to tune this system and remove the dither without an
> additional high-res encoder added to the system?
>
> I've read as much as I can, mostly on the forum and here are the options
> I've found.
>
> 1.  Try to tune out the dither.  PCW mentioned maybe add a touch of
> pid-bias or FF3 (need to add FF3 to PID comp).   NOTE:  I need to go and
> work this a little more, but not optimistic.

> 2.  7i87 for tach feedback, but PCW says you're still dead reckoning and w/
> time you'll drift.  (ie no velocity when idle).  And, I didn't see any more
> about this after about January 2013.  I wish this would work because it
> would be rather inexpensive for hardware.

Is the 7i87 an analog input module?

With only velocity feedback from the motors, tuning is likely to be tricky.
It might work though.  And if the tachs are already there it is probably the
least expensive solution.


> 3.  External encoders on ball screws or back of motors.  I've heard people
> mention it, but haven't actually read about someone being successful with
> it.

I think the best approach if you have linear scales is to add rotary encoders
on the motors and use both.  Use dual PID loops, one loop getting its feedback
from the scale, and the other getting its feedback from the motor encoder.
Sum the two PID outputs together.

The scale PID should have I-gain only, for steady-state accuracy in spite
of things like screw lash and thermal expansion of the screw.  The encoder
PID should have P-gain and D-gain for good dynamic response and stability.

FF gains can be applied to either PID loop, since the same command goes
to both.

This has been done successfully.  I used to have a blog posting describing
how we did it on Stuart Stevenson's big horizontal boring mill.  (Ten foot long
X axis.)  But my blog has been down since I upgraded my server.  Sorry
about that.

> 4.  Higher resolution linear scales.

I don't think that is the answer.  Any lash at all between motor and scale
will make it very difficult to tune.

> Ok - I'm going for the jugular here, I found on Ebay some 1um renishaw
> linear encoders and scales from the same guy.  I'm thinking I can get a
> discount and get 3 axes of 1um linear scales for $400.  It's more than
> adding rotary encoders, but I think the futzing around and simplicity of
> the final solution is compelling.  Also, I can remove (and sell) the glass
> scales, put the renishaw's in place and wire in.
>
> Suggestions?

I would add motor encoders before buying higher grade linear scales.

> Thank you,
> Mark


  John Kasunich
  jmkasunich@...
------------------------------------------------------------------------------
_______________________________________________
Emc-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/emc-users

Reply via email to