This page is meant to hold various notes that arose during different studies until the information can be incorporated into other pages or documents on this site.
For labs (such as a vendor) with only one possible technician to run the measurement system, try to run the MSA/GR&R study with at least two technicians. When doing the analysis:
Run the analysis with all technicians.
Run the analysis limited to only the one 'real' technician. (Will get Gauge Repeatability, but not Operator Reproducibility)
When to compare GR&R against Process Variation vs. Tolerance?
Comparison against process variation is recommended when:
using the measurements in an SPC study
when it is important to be able to resolve the width of the variation of the process
when marketing claims are involved
Comparing against tolerance might be acceptable when:
the process is statistically in control and has a high process capability index (Cpk), meaning that the measurements are no-where near the upper and/or lower specification limits
when the measurements will not be used in any sort of SPC effort based on variables data
other considerations:
how much cost and/or effort would be required to improve the measurement system?
what is the risk to the customer of having extra (non-material-related) variation in the data?
have there been previous, good faith efforts to improve the measurement system? if so, has there been demonstrated improvement?
are further improvements technically feasible?
are the results of GR&R borderline (i.e. total GR&R ~= 30%) with low risk?
are there other mitigating factors, such as functional testing in addition to dimensional measurements
Example of when it might be OK to have higher GR&R, to include risk mitigation
If, given the nature of the product, the medical risks associated with any failure of the part concerned are very low and certainly not life-threatening, then an approach based on some calculation of the probability of a false negative result, coupled with some appropriate risk mitigation in the form of an action limit placed well above the existing acceptance limit might be acceptable.
On the assumption that the system is reasonably accurate (i.e. the measurement variability is centered around the true value) and with the worst-case assumption that all of the observed variability in this study is due to the measurement system, move the action limit to a distance from the spec limits (LSL, USL) based on a six-sigma span (based on the total variance).
Interesting article: "An Honest Gauge R&R Study" by Donald Wheeler.