Probing direction error compensation

[adatesman]

Sorry, reply above was inappropriate. The assumption of not knowing what you're doing is commonplace on hobby-target forums, and that is not the case with me. BTDT, know what I'm talking about.

I've been around high end machinery a long while. Repeatable kinematic error can be compensated for.

In for a penny, in for a pound, over the past 2 days I've probed ~7500 points on this ring gage at 5 minutes per 140 points. And am ~38 hours into my work week at this point, which started Sunday. Which leaves me tired and snippy. Apologies.

All I care about is the ability to compensate for kinematic error. Which seemingly Mach has plugins for, anad Centroid does not handle. Only reason I switched to Centroid from KFLOP is I don't have time to write probing routines. And yet, here I am, needing to work around things and write custom macros. Annoys me is all.

Apologies for being snippy. was undeserved.

[adatesman]

FWIW, I just finished running those ~7500 points and will now start doing analysis. ~140 points per run at 0.050" stepover, 10 times for spindle/probe position, rotated 90 degrees.

Since I have the programs to look into error written, I'm curious how Centroid's new probe does with it. Hopefully better, at this point I have a workaround to eliminate error and will not be buying one.

[adatesman]

At this point, I am confident that kinematic compensation would be appropriate in this case and am moving on with my project as I can do it a different way (probe one direction only and rotate the rotary). Ultimately I would like compensation to be added, as it's easy enough to manually orient probe/collet/spindle on hobby machines, and given the error I'm seeing here it would seem the hobby-level probes people are using with this hobby-level control will not actually meet the accuracy implied in the sales pitch. Compensation would likely address this, given what I'm seeing with this Drewtronics probe.

This chart represents 10 Wall Following runs in succession using the stock Drewtronics stylus (rather than the much thinner 1mm Chinese one from the earlier tests, to eliminate possible stylus bending issues), with 0.050" stepover on a 2.2190" master ring gage. Backlash compensated for, but similar results with it not. TIR was measured at under 0.0002". Drewtronics claims +/-0.001", and that seems to be the case, provided orientation is kept the same. Which make a strong case for software compensation for the kinematic issues.

Green band is nominal it should be reading at, red band is average of the 10 runs, inner/outer blue bands are the extreme/worst case minimum/maximum of those 10 runs. I did it with the LED front, left, back, right, then front again to see if it repeated as spanned 2 days (each set of runs took 90 minutes, and little free time).

EDIT: numbers look off, poking at it.

[martyscncgarage]

While backlash compensation works to a certain extent, it can't remove the mechanical backlash in a machine. That's why it's always good practice to bore a hole instead of interpolate it.

How much mechanical backlash does your machine have? Ballscrews? I'm wondering if this is playing into your results at all?

[cncsnw]

To give us an apples-to-apples comparison with the table on page 11 of Centroid's DP4 manual, what was the stylus length in your tests?

Theirs was 43mm, and with it they report the DP4 showed a maximum variation of 0.0011" (vs. about 0.006" for the Drewtronics). But of course, if you put a 240mm-long stylus in the DP4, then presumably you would see 0.006" variation.

Am I correct in understanding that the shape of your measured-error plot follows the probe orientation (i.e. some particular angle on your plot corresponds to the LED on the side of the probe, regardless of whether you do the test with that side of the probe facing X+, Y+, X- or Y-)?

[frijoli]

At this point, I am confident that kinematic compensation would be appropriate

Compensation is just that. It is NEVER desirable. There may be times when there are few options; whether that is mechanical, feasible, or financial is a different discussion.
I would consider large backlash to be unacceptable in a ball-screw as there are replacements usually available. If none are available or outside the budget of the project then compensation "may" be appropriate.
In my opinion, kinematic error in a 165 dollar probe is not something I would compensate with a software adjustment/process. This is my opinion, and mine alone. If I had a probe that was .0001" out and wanted to adjust, yes, I would look at compensation from a software standpoint.

I am not knocking your position, but I am saying I don't think the addition of this option should be very high on Centroid's priority list. Again, my opinion.

[Gary Campbell]

This appears to be an effort to find a software fix for cheap hardware. Usually an exercise in futility. Software should enhance hardware, not cover for its deficiencies.

[Muzzer]

I took a different approach. I won a Renishaw MPS1 probe (https://www.renishaw.com/mtpsupport/en/mp1--32438) for £200 ($260) off ebay and direct wired it into my Acorn. Admittedly, the detection current may not be exactly as per the Renishaw spec but with a repeatability spec of 1um (40 microinch to you boys) when used "correctly", I doubt that's a major concern.

It seems to be in pretty good condition, so I believe it is still at least an order of magnitude better than my machine (likely more), despite my wanton abuse of this fine instrument. I've taken care to set it up as recommended and check it periodically. What testing I have done seems to indicate a fair degree of repeatability.

Whether or not I align the probe the same way each time (I generally do, as the cable wants to emerge in one direction, towards the controller) will surely have an immeasurably small impact on the end result in the context of my application.

I wouldn't waste weeks of my life bu99ering about on forums talking about software compensation and making experimental measurements (to what end?). I suppose that might be seen as a passtime in itself and there is no law against it but speaking personally as an engineer I'm looking to use my machine, not to polish a turd.

[frijoli]

So as not to cause any headache, heart ache, heart burn. I want to make sure adatesman realizes we aren't trying to belittle or talk down. We appreciate your thought process, and your methodology, but our opinion of the goal and the reason is different.

[cncsnw]

I second what Clay said.

The data and analysis that adatesman presented was sound and informative. It is remarkable that the Drewtronics probe is as repeatable as it is, given the significant asymmetry of its pre-travel.

Direction error compensation in the software would allow Centroid to turn a probe they do not sell, with 0.006" variation, into a probe with 0.001" variation, at the expense of adding considerable complexity to their software. Since Centroid already has a solution for those who need a probe with 0.001" variation, I won't be holding my breath watching for the software compensation.

If, someday, someone buys a DP4 (or a bunch of them) and installs them on machines truly capable of measuring to within 0.0001", and asks for a way to reduce the DP4 variation to suit, and is prepared to pay a premium to get that capability, then it might happen.

In the mean time, note that it is technically possible for you to add directional compensation yourself, by editing cncm\system\probe_move.cnc. Using the inital XY position saved at the start of the move, and the XY position after the slow probe trip, determine the angle; then add the X and Y components of your compensation amount for that angle to the values returned in #34006 and #34007. See approximately lines 140-141 of probe_move.cnc.