I have run a set of Gage R&R tests recently.

One of the things that was immediately apparent was that some operators allowed greater variation in their results sets than others. Clearly more training is required.

My question is, how do I determine what is/is not acceptable? What statistical technique would be best?

The most obvious answer to me is to have an "expert" (or at least a "acceptable operator") perform the measurements and determine the variation in those measurements. Then you can compare the variation for the "expert" with the variation for an operator. If there is a sgnificant difference (e.g. using an F test), then the operator is not as good as the expert and could use some more training and/or practice.


Tim F

Hello,

The gauge R&R result tells you if you need training, and who needs the training, and how much improvement is required, there is no need for other statistical calculations to take place.

Regarding your question, the GRR result will tell you what’s acceptable. It provides you with a result in percentage format, this result tells you how much influence the results (actual product results not GRR results) are being affected by the measurement activity.

Basically if you have part that had a tolerance of 0,010”, when you did a GRR on the measurement solution the result indicated 10%, that means of the 0,010” tolerance you have measurement uncertainty of 0,001”.

The tolerances for GRR are laid out in the MSA book, but I tend to use: 0 – 20% = measurement solution has no undue influence in the results, 20 – 30% 0 the measurement solution chosen needs improving/ is not the best solution and needs looking at as the product results may be biased by the measurement solution. 30% + = the measurement solution is interacting the results and thus the data collected for the part cannot be used.

Remember, GRR is telling you how much influence the data being capture (from the part) is good and true data that represents the actual part, and not biased because the measurement solution is poor. Its basically addressing the old problem of “Oh the parts not wrong, it’s the gauge, or it’s the inspector, or your not using the thimble correctly”


Regards

Lee.moffatt

I agree that SPC isn't appropriate in this case. SPC is for processes that are capable, and you've described one that is not in control.

I also agree that correction needs to take place via comparing the techniques against a known "good" technique. I didn't catch what types of instruments are involved with this issue, but in another thread a suggestion came up to ask the instrument supplier if a representative to be sent out for training, from Mitutoyo/whomever.

Whomever does the training, a period of careful monitoring should take place to see if old habits have been broken and proper methods are consistently applied. This could be done by watching charts, but I'd worry about what mis-inspections are taking place in the meantime. How much impact has taken place on inspections already? This question should be considered if the inspections are critical.

The problem should be attended aggressively and soon.

I agree that SPC isn't appropriate in this case. SPC is for processes that are capable, and you've described one that is not in control.

There is no way to determine whether or not a process is "capable" without some form of statistical analysis, even if it's a relatively informal statistical analysis. SPC is not just for capable processes, because you have no way of determining capability without SPC.

In this case, the GR&R seems to have done what it was supposed to do--it identified the source of a measurement problem, or at least a potential problem. MSA results should never be characterized as "passing" or "failing," because any analysis that helps to identify problems is successful.

I also agree that correction needs to take place via comparing the techniques against a known "good" technique. I didn't catch what types of instruments are involved with this issue, but in another thread a suggestion came up to ask the instrument supplier if a representative to be sent out for training, from Mitutoyo/whomever.

Yes. :agree1: It's important to pre-measure GR&R parts so that the study not only takes into account comparisons between operators, but comparison of results to actual values. It's possible (although not likely) for three operators to all measure parts incorrectly, but consistently. Simply understanding consistency isn't always good enough.

There is no way to determine whether or not a process is "capable" without some form of statistical analysis, even if it's a relatively informal statistical analysis. SPC is not just for capable processes, because you have no way of determining capability without SPC...It's possible (although not likely) for three operators to all measure parts incorrectly, but consistently. Simply understanding consistency isn't always good enough.Yes. :agree1:

When it's known that correction is needed, it seems sensible to me to approach that before taking statistical data. Since the questions included "how do I know what is/is not acceptable?" I think it's worthwhile to bring in an expert for evaluation and then monitor performance after the first round of correction.

Of course I could be wrong. I just don't view SPC as worthwhile on something that's already defined as broken.

That said, one sort of statistic isn't the same as the next. If there are a number of error types then a Pareto or histogram could be helpful. I'm not an SPC guru, but I think people sometimes confine their thinking to X-Bar & R, P charts and such when they aren't appropriate for the need. It's hard to know if Soutern Cross was referring to those methods, but I wouldn't use them here, yet.

Also, seems to me that part of a GR&R study is to understand how sensitive the results are to varying technique. As you said, certainly there could be no variation but an across-the-board misuse of tools. Sounds to me like it's time for what we used to call a "stand down".

Of course I could be wrong. I just don't view SPC as worthwhile on something that's already defined as broken.


It's possible to know that variation exists without knowing anything about the nature of the variation. If your car won't start, it's helpful to know that it won't start, but without knowing why it won't start, you won't be able to remedy the problem. Of course, when solutions are self-evident, there's no point in invoking unnecessarily complex methods.

I agree. Truth is, we actually know almost nothing about this matter; what kind of gages, how often and how much variation, and how it was decided it's a training problem.

Thank you all. This has been very informative.

The instrument is a rheometer, it's used to measure viscosity and pseudoplasticity of polymer solutions. The variability is going to be due to the operator not placing the sample correctly between the cone and plate. If the cone is not completely covered to the edge, viscosity is low. If residual material is left on the side of the cone, viscosity is high. Once the sample is loaded, the machine takes care of the rest.

I'm afraid there is really no way to get a 'master' sample with a set viscosity and pseudoplasticity. I have a silicone viscosity standard, but it is not pseudoplastic.

From the GRR test, I've been able to get down to ~20%, but some operators are still giving >30%. I believe that this is due to sloppy loading of the sample, I really can't think of any other explanation that would be possible.

It sounds like I need to train all the operators, run GRR tests and conduct further training or weeding out of anyone who fails to get below 30%.

What is the industry standard for passing grade? 75? 80? 100? Please advice.
Basically, our previous passing grade is 100% for all Operators and Inspectors, that’s why most of our customers would want that idea having an Operators and Inspectors with 100% mark. Now we have a newly installed HR Manager, and he would want the 100% passing grade be reduced to 80%, and he’s claiming 80% is the industry standard. Can anybody confirm so that when an auditor or customer comes in to audit us, we could provide justification in changing the passing grade.
Best regards,
Raffy:cool:

In general it seems that, IMO, HR is looking at this as a training comfirmation and not a process control validation. The question seems to me to be how much variation can your process withstand. If 30% variation in the results is acceptable then it might not be worth the fight with HR. BUT, if 5% reading variation causes scrap, then there is a significant and different issue. Process requirements are usually outside the scope of HR.

If the GR&R is an attribute calculation, we use 80% agreement as the cutoff for "good" results. If the GR&R is variable, then the amount of error allowed is up to 10%, with management sign off allowed up to 30% of the total tolerance taken with the gage error.