Improving Analog Scanning repeatability

[kingsld1]

We are trying to 'validate' the scanning capability of our CMM by doing a Gage R&R on an extremely simple part design. These parts were special made for this purpose. It's a 6" x 4" x 0.25" thick aluminium plate with a 1" nominal diameter machined in it. We had 10 of these parts made with the hole size, X and Y locations randomly varied within +/- 0.005". (Our typical part tolerances are +/-0.005")

Program was done only using scanning for the alignment and the location/ diameter of the hole. The Gage R&R failed in the Y direction (somewhere around 15%). Used the default settings for the scan. Calibrated the scanning probe and lower matrix just prior to running the Gage R&R.

Are there any recommended best practices for this given the part/ all scanning method constraints that I'm under? Any tips/ tricks to improve repeatability? Used PC DMIS 2017 R2 on the original attempt. Have 2018 on the CMM if that would help. Could probably get 2019 if I twist a few arms. I had the scanning training 4 or 5 years ago but never used since getting a scan head fell thru at the time. Another programmer with about 2-3 years experience has had the training but has not used since. I'm sure we lost quite a bit of what we were taught just because we haven't needed to use it.

[JEFMAN]

First, I would create a "ttp" alignment in DCC mode, after the man alignment.
It allows to be more accurate than directly measure the dcc alignment in scanning.
Also, I would decrease the scan speed, increase the points/mm, and filter outliers.

[louisd]

So your location was off, and not the form of the part? to me, that screams a part alignment that needs a more secured 6DOF constrained.
Level, rotate, origin.

[InspectorJester]

Did results improve after this was adjusted?

[kingsld1]

The result improved after correcting the inspection setup. Some of this is hard to remember since it's been 6 to 9 months ago. Management abandoned the effort after we had made a couple of attempts at it.

"Other fish to fry...." / "We never had the capability before so we're not really missing anything". was the theme of the day.

Touch probe validation easily passed. so that was "good enough" at the time.

They are now starting to see where having the scanner could be useful so we're gearing up again.

[InspectorJester]

Typical!
Lol. Glad you were able to figure out the setup, and perhaps a better way to scan along the way!

[kingsld1]

I know the setup is figured out. Jury is still out on scanning.

Did some preliminary testing over Friday and this morning. 7 different ways to evaluate the circle. Used the same alignment for each evaluation and rescanned the circle 7 times, one right after the other all in the same program.

Methods were:
Least Squares,
Max Inscribed (various parameters changed on both),
Least square w/ 3 sigma outlier 50 UPR, Max Inscribed w/ 3 sigma outlier 50 UPR,
Least Squares and Max Inscribed with double the points and half the speed,
and finally a completely clean Default Adaptive Least Square.
Brain Fart on not doing a Default Adaptive Max Inscribed.

Default Adaptive Least Square had the lowest overall range over 7 parts / two operators /two days. (5 w/ one operator, 2 w/ the other) Ranges are X 0.0000652", Y 0.0000403", D 0.0000320". Results are in INCHES. Worst case still had 4 zero's before any other digit.

[louisd]

anything beyond the 5th decimal place is noise, imo. presuming your analog sensor is a leitz LSPX1H-T

[kingsld1]

Agree that it's mostly noise but there is some signal there. The math involved in doing a gage R&R creates some very small numbers. I've done gage R&R's before where if the results were rounded (even at 5th or 6th place) the study failed. If the full results were used then the study passed. It's like not going out enough places on a control chart. Chart can appear to be badly out of control if the results are overly rounded.

[louisd]

agreed if you were trying to validate a measurement method for total tolerance sub 0.001", that maybe 5th decimal reporting is necessary to confirm pass/fail, but if you know your sensor's specified resolution is 1micron (0.00004") at absolute perfect best case (not to mention probe head, machine, TCOE, etc), why report anything smaller? plus you are saying your typical assessment value for a total tolerance is 0.010" heck 4 decimal rounding should be plenty to pass an MSA to 10-15% of 0.010".

[kingsld1]

If I remember correctly we were at somewhere around 13% on one of the axis. Management said that that wasn't good enough/ don't use the scanner. Even put the no scanner use in our CMM procedure. Rather not take any chances in recording the data. If it passes at 7 places I'll take it , if it passes at 4 places I'll take that too. We're a medical device company that has chemical disposables. Each instrument we sell will use in the hundred thousands or better of the disposables. Chemist's are in charge. They are used to doing overkill for the FDA. If it even smells remotely funny it's a no go.

[SingularitY]

I did a study on runout, different than what you are doing I guess, and found the best repeatability using adaptive circle scans with a UPR and gaussian filter, overriding the settings to match the parameters listed for my scanning probe via a Hexagon document (see optionprobe settings), cranking up the point density, and slowing the speed way down. Linear scans and basic scans with constructions gave me the worst results.
You said you are using an aluminum part for the study? What type of tip are you using? Ruby and steel tend to get buildup from aluminum very quickly while scanning. Silicon Nitride tips are what I use because they are resistant to aluminum buildup.

[kingsld1]

It's ruby on aluminium.... Sigh.... But the only use that it's getting is from the r&r study. I'm keeping an eye on it.

[SingularitY]

We had to switch to a 3x40 nitride tip to pass a GR&R for an aluminum part with tight tolerancing on runout. (.0005in) Buildup moves fast. Good luck man!
Keep us updated with your solution.