iSixSigma

Two DISTINCT types of GRR

Six Sigma – iSixSigma Forums Old Forums General Two DISTINCT types of GRR

Viewing 16 posts - 1 through 16 (of 16 total)
  • Author
    Posts
  • #39088

    Schuette
    Participant

    There are (at least) two distinct types of GR&R.  One might be characterized as “Gage Only R&R”, and is an atempt to verify performance of the gage itself.  The other, let’s call it “Measurement Process R&R” is the usual GR&R where you are verifying performance of entire measurement process, including the gage, the operators, the samples, etc.
    Here is the main difference: WITHIN-PART variability such as might be found in 1) coating thickness from one one section of the piece to another, 2) the OD of out-of -round parts, etc.
    The gage itself might be fine for these measurements when within-part variation is small.  In cases where the within part variation is large however, it tends to result in an unacceptably high %GR&R.
    Radically increasing the number of measurements per part will result in a lower %GR&R, but I was wondering if anyone out there has company procedures or guidelines that deal with this.  I know that you can do a Nested ANOVA to assess the components of variation, i.e. part-to-part, within part, operator-to-operator, and test variability.  Even with this information however, how would then proceed to determine if the measurement method meets your R&R requirements?
    Does your group/company have guidelines to deal with excessive within-art variation?
    Jim

    0
    #118204

    FTSBB
    Participant

    May I suggest looking at your R chart – specifically, the R-bar and UCL, as these are hints at your “average” test-retest error and the upper limit you “expect” the measurements to vary before you suspect assignable cause.  See if this “error” is acceptable compared to your spec range.  Not bullet-proof, but should give you a realistic picture.
    I have taken a similar approach using a gage to measure powder paint top coating thickness.  On the same part, the thickness can vary by 20%+, even on locations less than 1/2″ from one another!  This is not unique; I’ve seen other processes that were similar and documented cases of top coatings varying “significantly” in the short-term.  The trick to proving out the gage was to have a method of measuring the EXACT same point during the MSA study – actually circled a point on the part for each operator to measure.

    0
    #118208

    Jim P.
    Participant

    Thanks for the response but I already know that these particular coated parts have excessive within-part variability – 59% of Tolerance GR&R for the parts vs. <10% when within-part variation is removed.
    What you suggested is what I had referred to as a “gage only R&R”.  We have already done that and have proven that gage itself is fine.  However, that still doesn’t provide any confidence that the “measurement process R&R” is suitable – in fact we know it is not.  And without that, the knowledge that the “gage only R&R” is OK doesn’t provide any comfort that we can adequately assess production parts.
    Tha AIAG solution of measuring the parts in the same place each time is not a solution unless you only want to qualify the gage itself.  Their proposal is a cop-out and does not address the real concern, which is the significant contribution of within-part variation.
    Fortunately we have just discovered a major cause of the within-part variation of coating thickness, and a work order has been submitted.  However, I think that will only solve some of the problem, not all of it.
    Jim
     
     

    0
    #118210

    Anonymous
    Guest

    Jim P,
    Just a thought for future reference. The Anova method assumes meaurements are random and independent, this is rarely the case within samples. My apologies for poking my nose in.
    Andy

    0
    #118211

    FTSBB
    Participant

    In that case, it’s not the gage’s fault – your parts simply vary quite a bit.  I think the root of your discussion lies in understanding the components of variation – sounds like you want to know if the parts are good or bad over a widely varying part.  I’d look into a COV study of the process – sounds like you’ve already started this. 
    The gage I described earlier is used in an SPC plan.  The key to balancing this data is useful subgrouping.  The operators take multiple measurements at each of the gun nozzle locations.  The SPC chart they view reports an average thickness for each nozzle.  This helps limit over-reaction to thin points at one nozzle for one reading that is OOC.  You’ll have to develop that knowledge and what causes variation in your process.

    0
    #118212

    Jim P.
    Participant

    Andy,
    Thanks for the response.  No apologies needed. I can certainly engineer a Nested ANOVA to assess the within part variation, and can probably do so for a regular ANOVA as well (although I haven’t thought about it as yet).  For this situation, I would prefer the Nested method.
    It’s amazing that no one has come up with a reasonable method for Gage R&R where you know that the gage alone is good for measuring parts in a certain way (marking the spot to measure), but not good for day-to-day use in production when within part variation is significant.  On the other hand, maybe it is not so amazing when you consider the AIAG doesn’t have a clue on how to treat this problem.
     
    Jim

    0
    #118214

    Anonymous
    Guest

    Jim,
    I’m not familiar with the standard but I can understand the sentiment many people assume random independence. Unfortunately, it is not always a safe assumption with web processes and the like.
    I might be wrong but I don’t believe Nested Anova will help either, because although it is true that you can partition the sources of variation, in terms of position, if measurements are correlated then two things can happen. You might detect spurious interactions (not real) and the effect will be over-estimated, because the procedure will believe it actually has more information (samples carrying unique information) than it really does. Perhaps one of the statisticians can correct me!
    All I know is Anova is not my preferred method, not because I’m indolent, but because it’s given me wrong answers in the past.
    My approach has always been to make up data first, before doing an actual test, or even the experiement, so that I have a reasonable idea what to expect. I started doing this in 1983 when Dorina Shainin kicked my but in front of the whole class for not using rational subgrouping. Take a page out of my notebook and always invent data to test before actually testing data.
    Cheers,
    Andy

    0
    #118215

    Jim P.
    Participant

    Yes, I know it’s not the gage’s fault. I thought I had made that clear. I guess not.
    I prefer a Nested ANOVA to determine the components of variation such as part-to-part, spray booth, spray head, operator, test location (within part variation). But don’t think I will even take the time to do that.  Within part thickness variability is not high on the priority list of our customers right now – meeting our internal GR&R rquirement however is.  We have discovered the major cause of thick spots and a work order has been put in to fix that. That however is only part of the problem I imagine. Becasue we are using a wireless gage which will send results directly to a PC, we can always use 20 measurements per part to get a pretty good estimate of the true average. However, we would rather do n=5.
    Later we will be doing much more with our powder-paint thickness testing as well.  You mentioned your situation with that.  We have multiple spray heads in our power paint process as well. Oh well, something else to look forward to.
    You use an SPC chart for each spray head. Have you considered a group chart?  Just curious. We can do that very easily with our software (InfinityQS).  A group chart might be a lot less effort for your operators.
    Jim
     
     
     

    0
    #118217

    FTSBB
    Participant

    Group chart?

    0
    #118275

    Jim P.
    Participant

    Grouping on a control chart is done in situations where measurements from several locations are required to ensure uniformity.  Group charts plot and analyze multiple data streams on one chart. Two points are plotted on the X or XBar chart – the maximum and the minimum. Each is labeled as to which process stream or position was responsible for that point.
    I tried to find a short paper on-line for you but couldn’t find the one I was looking for.  InfinityQS software provides this capability.  It is probably not a well known technique.
    You can read about it in a book, “Innovative Control Charting”, by Stephen A. Wise and Douglas C. Fair, ASQ Quality Press, 1998.
    Jim
     

    0
    #118294

    FTSBB
    Participant

    Interesting.  Sounds like a fancy way of displaying a range chart… I’ll see if I can find a copy of the ASQ article.
    Currently using WinSPC for this process.  It works fine, but the workforce is not too keen on ‘puters.  Not to put them down, but I think they paid closer attention and dilligence to gathering data when recording the measurements by hand.  They at least had to turn in a sheet at the end of the day.  Also, I’m not the direct process owner, just when things are going haywire (usually from scanning the wrong part # in WinSPC!).  Failure of the “system”, huh?

    0
    #118297

    FTSBB
    Participant

    Found a pseudo-reference…
    http://sunsite.univie.ac.at/textbooks/statistics/stquacon.html#common
    … just in case anyone else is secretly watching!  Not much of a description from this link, but I get the jist.  Thanks for the tip.

    0
    #118298

    Jim P.
    Participant

    It wasn’t an ASQ article that I was looking for. I coudln’t find the one I wanted, but I did find this one – not quite as good, but it gives you the idea.  Go to:
    http://infinityqs.com/nav-4-2.asp
    and scroll down until you find the article, “Group: Tracking Multiple Frozen Food Ingredients on a Single Chart”.  Although a lot different than multiple Power Paint spray heads, it is exactly the same idea.  However I’m not sure if this was a “Group Short Run Target Chart” or simply a Group Target Chart”.  I know that you would use the short run version due to the potential of widely different standard deviations. I don’t know how this chart was processed.
    Have a nice weekend.
    Jim

    0
    #159263

    Mohammad
    Participant

    I am final year student. I need GR&R information.

    0
    #159277

    Guna
    Participant

    I suggest u to read some reference books if u need more detail on Gauge R n R. Anyway for yr basic understanding, the defination for gauge R n R is :
     Gage Repeatability 
    The variation obtained from one gage and one operator when measuring the same part several times.
     
    Machine Variation
     
    Gage Reproducibility
     

    The difference in the average of the measurements made by
    different operators using the same gage when measuring the same part.
     

    Operator-to-Operator Variation
     
     

    0
    #159278

    Omashi Sabachi
    Participant

    Well-said

    0
Viewing 16 posts - 1 through 16 (of 16 total)

The forum ‘General’ is closed to new topics and replies.