MSA manual 3rd page 73 states, “For Process Control situations where the measurement result and decision criteria determine, ‘process stability, direction and compliance with the natural process variation’, the availability of samples over the entire operating range becomes very important. An independent estimate of process variation (process capability study) is recommended when assessing the adequacy of the measurement system for process control. ”

Does this mean I should use different samples to compute GRR (repeatability & Reproducibility) and process variation, and use them to compute %GRR?

Page 74, “When an independent estimate of process variation is not available, OR to determine process direction and continued suitability of the measurement system for process control, the sample parts must be selected from the process and represent the entire production operating range.”

My colleagues and I have some disagreement on how to select samples to perform GRR. And we have the following three methods:

Assumption: We have a cup manufacturing process. The cup height’s spec is 10(+/-1) mm. Take 50 consecutive samples, and the standard deviation of the samples is 0.4.

Method1: Select 10 samples, the 10 samples need to be between 9 mm to 11 mm. We are advised to have a sample as close to 9 mm as possible, and a sample as close to 11 mm as possible. Use these 10 samples to perform GRR.

Method2: Selected 10 samples, the 10 samples need to be between 9.6 mm and 10.4 mm. We are advised to have a sample as close to 9.6 mm as possible, and a sample as close to 10.4mm as possible. Use these 10 samples to perform GRR.

Method3: Randomly select 10 samples from the production line, and use these 10 samples to perform GRR (my prefer).

Which sampling method do you think is correct? Or you have another method?

As no one else more competent has come forward, I will offer my opinion. Remember it may be worth exactly what you are paying for it. I don't use any of your methods and there is good and bad aspects to them all.

Many say you should do MSA before Capability but I try to sort of combine them and although it is not strictly “right”, and I can be sure others might have better approaches, so far this has worked OK for me – no auditor has questioned it deeply.

Sorry but some of my decisions below can not be accurately quantified, sometimes you have to go by instinct.


Initial Gage selection

I use “experience” to determine which gage to use during APQP. If I expect the process capability to be adequate to good, or I am looking at P/T ratio applications, then I will try to use a gage at a minimum of resolution:tolerance 1:10. If however I expect process capability to be to good to very good, then I will try to use a gage at a minimum of resolution:tolerance 1:20.


Parts selection

I then take the 125 parts segregated and measured from preliminary capability study applications and try to select the 10 parts for MSA from them in a manner to very nearly reflect actual distribution.

I list, sort the parts, and very loosely divide the distribution into six equal sections - in effect near equivalent to +/- 3 sigma’s sections.

Select the absolute min and absolute max from the population, these are the only two parts taken from the outer sections i.e. between +/- 2 to 3 sigma.

I then select 1 part each from next sections inwards i.e. between +/- 1 to 2 sigma. Wherever possible these parts should be mid point or very slightly closer to mean within the section.

From the inner two sections, i.e. between +/- 0 to 1 sigma I select 3 parts either side of the mean, fairly equal distributed within the sections. Again some minor bais towards mean is acceptable.


Issues

I accept this doesn’t exactly give you a sample population exactly reflecting the overall process, but its good enough in my experience.

Again if the preliminary capability study turns out to not be a true reflection of ongoing capability you may need to repeat this exercise but so far that has not been a great issue to me.

Hope this helps.

This is also our dillema in selection of parts to be used in MSA.
I think Method 2 is more appropriate for your existing specification, if you're processing only 10 mm. Suggest to use your standard deviation as your control limit for the min and max values.
9.6 mm min and 10.6 mm max.

Given:
Assumption: We have a cup manufacturing process. The cup height’s spec is 10(+/-1) mm. Take 50 consecutive samples, and the standard deviation of the samples is 0.4.


Method2: Selected 10 samples, the 10 samples need to be between 9.6 mm and 10.4 mm. We are advised to have a sample as close to 9.6 mm as possible, and a sample as close to 10.4mm as possible. Use these 10 samples to perform GRR.

The method that you should use to select samples depends on the type of MSA that you intend to perform.

Decide in advance whether the gage will be used for inspection to the tolerance or for process control (SPC).

If the gage is to be used for inspection, part selection is less critical. Many recommend selecting parts from the full range of the tolerance. There is nothing wrong with doing this, but it is not critical that this be done. When the study is evaluated, use the P/T Ratio (% of Tolerance) to assess the gage.

If the gage is to be used for process control (SPC), part selection is extremely critical or the results are invalid. The parts selected MUST have a standard deviation that is equivalent to the standard deviation of the process from which they came. This is why it is important to have some benchmark capability data with which to compare the MSA samples. If the standard deviations are equivalent, use the %GRR to assess the gage.