B
And here in lies one of my biggest issues with an R&R. By selecting non-random parts, particularly those outside know control or spec limits, you artificially increase the calculated control limits. Conversely, selecting 10 random parts believed to be within known control or spec limits drives the calculated control limits smaller. The last is particularly troublesome when the resultant value is less than the know accuracy of the measuring device.
Another pit fall occurs all too often here, where our QE’s hand us a tray full of parts mistakenly believing they are both in spec, and created from a process already in control. We do an R&R and, lo and behold, the parts were neither. …Obviously (so say the QE’s), “the measuring device was flawed.. …We’ve been making them for 10 years…”
Then they say: “We can’t use this (data), your measurement uncertainty is 50% of the tolerance.” To which I reply: “Duh! The spec is unreasonable, and we’ve once again proved the manufacturer specs for the accuracy of the gage are correct.”
I’ve been approaching the problem from the front view. I find out what the measurement uncertainty is with an MSA. Then multiply sigma by 36. I give this number back to the designers and say, “If we want a 6 sigma manufacturing process, the spec limits better by 36 sigmas of the measurement uncertainty.” The difference on our care-o-meter is 0.0002 vs 0.000020 inches.
Another pit fall occurs all too often here, where our QE’s hand us a tray full of parts mistakenly believing they are both in spec, and created from a process already in control. We do an R&R and, lo and behold, the parts were neither. …Obviously (so say the QE’s), “the measuring device was flawed.. …We’ve been making them for 10 years…”
Then they say: “We can’t use this (data), your measurement uncertainty is 50% of the tolerance.” To which I reply: “Duh! The spec is unreasonable, and we’ve once again proved the manufacturer specs for the accuracy of the gage are correct.”
I’ve been approaching the problem from the front view. I find out what the measurement uncertainty is with an MSA. Then multiply sigma by 36. I give this number back to the designers and say, “If we want a 6 sigma manufacturing process, the spec limits better by 36 sigmas of the measurement uncertainty.” The difference on our care-o-meter is 0.0002 vs 0.000020 inches.