iSixSigma

Gage RR (MSA) for destructive testing

Six Sigma – iSixSigma Forums Old Forums General Gage RR (MSA) for destructive testing

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
    Posts
  • #29603

    Ernest
    Participant

    Any ideas for measuring variability in a measuring system which is destructive?.Β  I want to measure variability in bottles’ release (opening) torque, but as I open one, I cannot re-measure that one.Β  Any ideas??Β  Thanks.

    0
    #76199

    Jay Fleischmann
    Participant

    i have been struggling with this for around two months and have not yet found an acceptable method.Β  minitab has a nested r&r you can do but even that sets it up so the operator measures two different parts and wants you to assume they are the same part.Β  so there is no way to discern true operator variation from part to part variation.Β  the only thing i considered was randomly splitting the samples into two or more subgroups before the r&r is performed and then perform either a t-test or one-way anova (depending on how many groups you split the parts into)Β and hopefully, be able to say that all the groups are statistically similar (p>.05).Β  anyone else’s thoughts are much appreciated as i am in the same boat with destructive r&r.
    Β 

    0
    #76202

    Tommy
    Member

    build a standard “bottle”, and then a tool to apply caps with a torque wrench to the desired nominal. The use you guage to open them. The readings won’t be real world but with enough repetitions you should be able to ascertain operator error.Β Of course insure the torque wrench is calibrated and passes msa.

    0
    #76214

    Chris Seider
    Participant

    This question is a tough one for practical use.Β  But here are some thoughts.Β 
    Β 
    You may do a nested Gage R&R but when selecting samples for each operator to test be sure to take out of similar lots for analyzing sample 1 for each operator to minimize the effect of difference of the product/sample/etc. analyzed by the operators.
    Another technique I use is to do a crossed Gage R&R and take samples from the same lot (try to get consecutive parts, similar location, etc. to minimize impact of product varying) and realize that the operator*sample interaction is not really valid but that the total Gage R&R includes this (and this is your worst case).Β 
    Another point that is lost by many–be sure to pick different products across 80% of the single product range and even if not able to to use % tolerance which is lost on many I’ve seen.Β 
    The worst case to do is what I often see from corporate folks who learn about Gage R&R and then misapply.Β  Test 3 different products effectively which widely different targets, do a gage R&R on those 3, find the % contribution is small and say the Gage is great!Β  The statement of greatness of the gage is so trivial because the only thing learned is the gage can distinguish one product from another–not the variation of one product’s measurements relative to its region of operation (tolerance [usl-lsl]).

    0
    #76219

    Khandekar
    Participant

    I believe the third edititon of MSA manual from AIAGΒ discusses this issue in detail. Has anyone had a look?

    0
    #76603

    KavSS
    Participant

    I just checked the 3rd edition, and they mention a method (“Split Specimens”) very briefly (less than a page on it), but don’t go into detail.Β 

    0
    #76606

    Mike Carnell
    Participant

    Ernest,
    I do not want to be real negative about theis but you are really screwed. You have a mess on your hands. Trying to measure torque is a bad enough problem but when you lay the destructive testing on top of it you are playing a 3D chess game.
    First you need to decide what you want to measure – break away torque or running torque. They will be different. I would assume if it is torque on a bottle cap that break away is the customer CTQ. You need to make that decision. Could be both.
    Understand that torque is an inherently poor measurement. Unless something has happened the idea of calibration is an issue since there were not any torque standards (it is usually a calulation using a lever arm and a mass standard).
    There are simulated hard and soft joints that can be used to do torque studies. You would have to decide which simulates your situation most acurately.
    You can run the study on the simulated joints. If you do that you probably want to do some type of hypothesis test that proves the distribution of actual product and the distribution of the simulated joints represent the same distribution.
    Good Luck.Β 

    0
Viewing 7 posts - 1 through 7 (of 7 total)

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