How to get a sample that covers Process Variation for GRR (Gage R&R) study

[Beiquan - 2010]

How do I get the sample for GRR to assure that covers process variation?

Beiquan

[D.Scott]

Welcome to the Cove.

The best way I know is to "hand pick" your 10 samples from a production run. You will need to measure a significant number to choose from as you should include as much of the variation as possible to ensure linearity. Be sure you number the 10 pieces to be used in the R&R.

I hope I understood your question and this helps.

Dave

[Beiquan - 2010]

Thanks for your reply and you understood me correctly.

I picked about 1/3 sample near to USL , about 1/3 near to LSL , about 1/3 near to average, and I calculate the GRR result <30% , but GRR result was 78% if I picked 10 samples randomly

So I want to know which method is right.

Beiquan

[D.Scott]

You made your sample selection correctly.

I am not sure I understand the 30% V 78% though. Are you using 10 identified samples and comparing the results of 3 operators measuring the same parts 3 times?

Once the samples are selected and identified (random or chosen) there shouldn't be a big difference in the study results. Obviously there may be differences but the measurement variance should be similar. If an operator measures the same part 3 times the results should be very close unless there is a change in how it is measured.

Is it possible that because of the configuration of the part, the spot you measure isn't always the same? This could cause a difference as you describe. Try marking the part at the spot you want to measure. This will ensure it is measured as close as possible to the same spot each time.

Another thing to consider is if there is really that much variation you might have to find a different way to measure the part. The current method may prove unsuitable.

One last thought - are you using the total tolerance method or part-to-part? I am not an expert but that may cause the difference. Be sure you use the same method and tolerance in both sets of your tests.

Dave

[Beiquan - 2010]

Very thanks , I am using 10 identified samples and comparing the results of 3 operators measuring the same parts 3 times , I am using part-to-part method , And our operator were trained again and again , so I think measure method is ok .

I’m looking for a good method to collect data of 10 samples ,and I’ll consider your opinion seriously !

It’s time to go to bed in china, it’s a funny thing that I don’t know I should say good morning or good night ! In any case, very nice to talk with you.

Beiquan

[D.Scott]

I think you will find using the part to part method is what is throwing off your results. I assume you are using a Gage R&R that allows you to enter the total tolerance and use that method instead. I would give that a try and see if it doesn't show more accurate results.

At my age, it is always time to go to bed here. Rest well.

Dave

[Grizz1345]

I have found the results of a GR&R to be very sensitive to the allowable tolerance. The tighter the tolerance the greater the effect of a small variance on the outcome of a study.

[Miner]

You should determine the reason for performing the MSA prior to planning the sample collection. There are two basic reasons to perform an MSA.

Reason #1) to determine whether the gage can distinguish a good part from a bad part for acceptance measurements. You need to compare the measurement error to the tolerance using P/T Ratio. In this case, your sample selection method is the appropriate method.

Reason #2) to determine whether the gage can distinguish one part from another part for statistical measurements such as used for process control (SPC) or process improvement studies (DOE, t-tests, etc.). You need to compare the measurement error to the process variation using %GRR. In this situation, your sample selection criteria will most likely give you false results depending on your actual process variation. You should select sample using the same criteria that you would use to collect samples for a capability study because the standard deviation of your samples must match the standard deviation of the short-term process capability.

If you have a capable process, your method would result in an artificially high %GRR because you have provided samples with a standard deviation greater than the actual process variation.