Probing direction error compensation

[adatesman]

Finally have my head around the error I'm seeing in Wall Following probing a 2.2190" gage ring, and looks to be kinematic with the probe itself. Which is to be expected with low-end probes utilizing 3 ball/pin switches at 120 degrees spacing.

Looks like this is something that can be compensated for in some controls, and I ran across a Mach3 plugin that handled it last night which would probe a ring gage every how many degrees and calculate compensation, so can be done. But not seeing such a thing in CNC12. At least, not in the Probing or Digitizing chapters of the documentation. I did not run a search in the PDF, but Google didn't turn up much in a quick search.

Looking over the Custom Macros doc, it looks east enough to write a macro that probes in X and A axis rotary, so I can avoid the kinematics error with my current project completely.

But it would be nice to know that Wall Following had a way to compensate for kinematic probe error, as right now I'm looking at +/-0.002" error over 360 degrees on top of backlash compensation. This pic is the average error of 3 Wall Following runs (not 10 as the pic says; bailed on that after hearing from Drewtronics), after homing, bore probing, setting zero, then running Wall Following. 1mm ruby stylus, centered 0.0003" TIR.

3 lobes are expected with this sort of 120 degree NC switch probe. Not sure why it dips so far on the right, but is consistent between 3 runs so easy enough to compensate for. Green ring is the nominal 2.2190" diameter, red upper and lower error bands.

Also, 99% sure this is the last of my questions about probing. And frankly, this may actually be a feature request.

[frijoli]

Do the lobes follow the probe? Meaning if you rotate the probe 90°, what happens to the graph?

[martyscncgarage]

What probe do you have?
Besides the tip being perfectly concentric, is the shaft of the probe perfectly parallel with the spindle? Put a tenths indicator on the probe near where it screws into the probe body and rotate it.

[adatesman]

Sorry, Work got in the way and not been able to get back to this.

Probe is Drewtronics, and lots of posts on the Taig FB group as I worked through this. Probe repeats 0 and 180, is off (but repeats) 90 and 270.

Stylus is 1mm ruby, China sourced. Backlash measured and compensated for. Spindle taper 0.0004" TIR with a Mititoyo indicator I've had since new 20+ years ago and little used. Probe has 0.375" mount, so measured TIR of a 0.3760" pin, something like 0.0008" TIR (did not clean taper or collet). Dialed the ruby in to 0.0005" TIR, without reclamping the probe, as not needed. Shaft concentricity is of no significance so long as the ruby is round and shaft not contact before the ruby. Also, looked into ring gage centerline being out from the spindle, and in order to get the amount of I'm seeing error the table has to be out ~0.0625" per inch from the ways. That would be noticed, and table to ways is under 0.002", IIRC

Bore measurements repeat with the LED front and back, and repeat left and right. But values 0 and 180 give 1.2019" and 90 and 270 give 2.2170".

Chart above is error with the CNC12 Wall Following routine, with the bore probing giving the correct 2.2190 result and LED facing front.

Clearly kinematic error in the probe, and have discussed with Drewtronics before posting here.

I see other controls have compensation to deal approach angle of the probe.

CNC12 does not appear to allow this, which means that probing is questionable depending on direction.

I can work around this easy enough for my project, through custom macros that rotate the part and probe in only one direction.

But long term, probing in CNC12 does not seem to handle kinematic error of a probe, which means Wall Following and all of the other routines are not to be trusted. That is disappointing, as there is a Mach plugin for this, Mach is cheap, and CNC12 is worlds past that.

More than happy to pass along the Mathematica programs I've developed to look into this, if there's interest (screenshot above). Also happy to run more tests (I'm like 100+ hours into it at this point, so may as well).

At this point, I'm back to where I started with writing custom probing with KFlop, and not so happy about it as swapping to Acorn was in theory going to avoid that. Started the conversion June-ish? Not going to go back at this point as CNC12 is *so much nicer*, but disappointed probing is unreliable and no one seems to have noticed.

[adatesman]

That said, tonight's project is using a different stylus to determine if error I'm seeing is due to bending of the stylus.

Will log and document, but I doubt that's the issue. Have dug deeply into kinematic error, and my screenshot is similar in form to what Renishaw puts out for their 3 switch probes. They see an order of magnitude less error, but Drewtronics is ~$165 and Renishaw $2600. Mach can compensate for this, so unclear why CNC12 doesn't.

[frijoli]

That said, tonight's project is using a different stylus to determine if error I'm seeing is due to bending of the stylus.

Will log and document, but I doubt that's the issue. Have dug deeply into kinematic error, and my screenshot is similar in form to what Renishaw puts out for their 3 switch probes. They see an order of magnitude less error, but Drewtronics is ~$165 and Renishaw $2600. Mach can compensate for this, so unclear why CNC12 doesn't.

What happens if you rotate the probe 90 degrees?

[martyscncgarage]

That said, tonight's project is using a different stylus to determine if error I'm seeing is due to bending of the stylus.

Will log and document, but I doubt that's the issue. Have dug deeply into kinematic error, and my screenshot is similar in form to what Renishaw puts out for their 3 switch probes. They see an order of magnitude less error, but Drewtronics is ~$165 and Renishaw $2600. Mach can compensate for this, so unclear why CNC12 doesn't.

I would think you would want a more quality probe for digitizing. DP4 is less than a Renishaw.
That said, provide the documentation & examples for Mach's compensation you refer to. (videos, pictures, instructions, explanations etc.)

Marty

[Muzzer]

Given that the error is apparently consistent and isn't rotationally symmetrical, I think you can conclude that there is some physical inconsistency at work within the probe itself. In the interests of science (and if you could be bothered), you might be tempted to dismantle it and attempt a root cause analysis.

There are no black arts at play here, so the cause is likely to be discernable if you know what you are looking at. Given how these things are constructed, the most likely explanation seems (to me) to be that the spring isn't exerting an even force on each of the three pins, possibly due to it not being closed and ground consistently. But that's just a guess. Perhaps you should attempt to measure the force in each of the 3 primary directions for starters.

You'll note that Drewtronics doesn't seem able to show any specs for the probe beyond a rather broad claim of "~0.001" accuracy", whatever that means. As we know trilobe probes aren't symmetrically consistent, there's more to an accuracy spec than that. There may be a clue there for you.

There seems to be a tendency to believe that backlash compensation and probing compensation are some sort of magic solution. When your machine is displaying backlash of +/-0.003", the solution might be a bit more involved than trying to bring some form of fancy software compensation to play.

[martyscncgarage]

The new Centroid KP-3 being currently tested, should be much better than the Drew probe, and other low cost Kinematic probes.
Just be patient while we get through testing....
Marty

[adatesman]

What happens if you rotate the probe 90 degrees?

Sorry for the delay, string of 60-80 hour work weeks.

This is the Bore Probing routine in a 2.2190" master ring gage, 3 times each with the LED facing Front, Left, Back, Right. I think backlash comp was not enabled at the time. Was enabled a couple days later with similar results. It repeats well, but the kinematic error puts Front/Back off from Left/Right.