might be a QE thing

[jtrofatter]

Hey guys! So recently i was asked to make some adjustments to a program to shorten the run time. To do this i only rearranged the order that the same points are taken, and reduced the clearance distances around the part. While doing this shouldnt impact the measured results at all, we were asked to conduct a 30 piece correlation study to verify. The operator who conducted the study had misloaded a handful of parts making it appear as though there was a large deviation in results from the normal program to the reduced time program. We verified the issue was misload by having a different operator load the parts in question and see if the results were the same or if they conformed to the historical results of the parts. My suggestion was to conduct the study again ensuring the parts are loaded correctly while ran, but the QE said this could not be done. Has anyone ever dealt with a situation similar? any tips at all on how to move forward with this?

[RandomJerk]

Curious, did you have the same part loading issues with the other program? If no there's a good possibility something in the program changed enough to cause it, if yes you need to point out to the QE that the program is acceptable, as shown by the re-run of the samples, but the verification uncovered an improvement opportunity with the fixture. Then improve it or don't.

[bfire85]

It's called a Gage R&R. To make sure measurements are repeatable and reproducible, different operators need to load and measure parts. If one operator cannot load the part correctly, then your fixturing is not robust enough. Start by gathering input from the operators and/or engineers on potentially how you can make the fixture more robust.

[edrake]

"Could not be done", as in your new program is forevermore invalid, or some other silliness?

Your program may benefit from a little 'alignment robustification', as far as repeatability and misloading goes. But that may not do you any good if it misloaded so grossly that there is shanking, or datum targets are out of position...

[jtrofatter]

Not necessarily forevermore invalid, but invalid until we can conduct a larger additional GR&R. the profiles are all 3D best fit, so a couple thou difference in the results from a slight tilt here or there ends up causing some issues once the best fit algorithm (Calypso) gets ahold of it. Just feels like he's calling the whole process into question, when the issue is due to the process not being followed as written.

[BKulpa]

If this is that much of an issue you should consider more robust fixturing.

[jtrofatter]

Therein lies the problem. We are constrained by a customer program, which i have reached out to them about making the alignment more robust and they said they nailed it no way it can be better. Given the geometry of the part (medical implant), if i make the fixturing more robust i lose access to features i have to take points on. Basically the fixturing of the part involves using a machinist square and a level to ensure we are clocked to the center of the part and sitting level in the fixture. It was observed that the operator loading the part was not making sure he was clocked and level, just eyeballing it and getting pretty dang close. Unfortunately pretty dang close caused a difference in the results by like 0.0028" for one profile. All the gage R&R's and correlation studies between machines we have already done shows we are more than capable of repeating within 0.0003 or less if the part is loaded using the correct process. Because of those studies, that is why the QE says we can't just repeat the study after re-training the operator who obtained the questionable results. He says this challenges the results of the previous GR&R's, but if we re-train the operator on the correct method of fixturing the part, and he then obtains results repeatable to the original results, would that not mean the GR&R data is unchallenged?

[edrake]

Has the Operator in question been involved in the previous Gage R&Rs? Did those have the same fixuring complexity?

I'm also curious whether the misloading can be properly compensated for with a more robust Alignment procedure (I get that the customer provided it). Is it a matter of the part cannot physically be probed properly, or does the program assume a perfect setup, and level/rotate/translate based on that assumption without any 'refining and updating'. Often times, a non-critical/non-datum feature ends up being a much better feature to start off with, just because it is easier to reliably hit the targets without refining - then updating from that as a 'launchpad' to more reliably find your actual datums.

I tend to be a bit anal about my alignments, pretty much updating the 'rough alignment' with each of the first 3 to 6 hits. Then armed with a better assessment of the actual position of the loaded part, proceed to measure actual datum features/targets (making sure you're at least in the ball park's parking lot before you try to get into that ball park). This has helped mitigate many of the potential misload issues we have. It obviously adds a bit of time, and that is what you're concerned with conserving, but if you can add maybe ten seconds or so to a program for robustness while also shaving off some minutes as you were (and reducing need for diagnosing a misload/reinspection), maybe it would help?

[jtrofatter]

The operator in question just joined the team barely 90 days ago with no previous CMM experience and wasnt a part of the previous studies. He's doing a really great job, just still getting the hang of it. Most of the reason i'm unable to make the alignment more robust is in addition to it being customer supplied, we also have to maintain correlation between ourselves and the customer. If i make the alignment more robust, the best fit profiles (held to no datums) could show difference between ourselves and the customer. Even though the difference the new operator is causing in the part, it's not enough to cause a difference in pass/fail, but its just enough difference to make a QE think that taking the points in a different order will change the measured value of those points. Thats why we had a different operator re-run the parts that showed significant deviation from original results to see if the deviation was reproducible, which is wasnt. if any other operator runs the same parts they see correlating results to the original run.