iSixSigma

discrimination of the measurement system?

Six Sigma – iSixSigma Forums Old Forums General discrimination of the measurement system?

Viewing 25 posts - 1 through 25 (of 25 total)
  • Author
    Posts
  • #46218

    Allthingsidiot
    Participant

    Any  idea  on  how  to determine  the discrimination  of  the measure,ment  system?What  is  the  increment of  time?
    thanks  and  regards

    0
    #152421

    Darth
    Participant

    I know Mini and JMP provide a measure of discrimination.  It usually is in the form of how many increments your measurement system can detect.  If you get a discrimination of 2 it means you can tell good and bad but no further scale is possible.  The rule of thumb is for the value to be 4 or more so at least you can make some distinction between items.

    0
    #152423

    Allthingsidiot
    Participant

    Thank You.Please  elaborate  more to  clarify the  concept.when  to  use  it ?At  which  control  chart (XmR ?).Appreciate the  enlightenment? 

    0
    #152424

    Jim Shelor
    Participant

    To get the discrimination of your measurement system you use the MSA process.  This has nothing to do with the type of control chart you are using.
    You perform the MSA, and you divide the sigma (parts) by the sigma (gage) and multiply by 1.14, then truncate the answer to an integer.
    This tells you what is called the number of distinct categories.  That is within the variation of your parts, how many groups can be discriminared by your gage.
    A number of distinct categories of 4 is considered adequate for some applications.
    If you need good discrimination for fine control, a number of distinct categoriest of 10 is considered excellent.
    Jim Shelor

    0
    #152430

    Darth
    Participant

    As Jim explained, this has to do with MSA and not control charts.  It is probably a good idea to do an MSA before collecting data and plotting on a control chart.  If your measurement system has poor discrimination then the signals that might come from a control chart will have you needlessly chasing variation that can’t even be measured by your measurement system.  If your measurement system can only discriminate whole integers, reacting to points which might vary by the second decimal point, is just noise.  A nice conceptual analogy I like to use is time….one measurement system with really poor discrimination can only measure daytime and nighttime.  One slightly better might be able to discriminate by hour and an even better one down to minutes.  If you want to be able to identify time by seconds, none of the above measurement systems have sufficient discrimination.  If you want to be able to determine morning, afternoon and evening, one of the above will be sufficient.

    0
    #152456

    Whitehurst
    Participant

    Discrimination is the resolution of the measurement system to detect small changes in parts.
    Recommended discrimination : 1/10 of  6 times of the measurement standard deviation.
    ” Number of distinct categories ” will give you this measurement. Generally it has to be more than four or five to have adequate discrimination.
     
    regards
    joe

    0
    #152453

    The Force
    Member

    You can perofrm this in Minitab or JMP and see the number of distinct categories (ndc). The idela is that it should be >4.

    0
    #152586

    accrington
    Participant

    Hi Jim
    You multiply by 1.41, not 1.14. Typo?

    0
    #152618

    Jim Shelor
    Participant

    Yep, typo.  I hate it when I do that!

    0
    #152623

    Perryman
    Participant

    Hi Jim,
    I tried to reply to your e-mail but it bounced back.  Do you have an alternate?

    0
    #152628

    Jim Shelor
    Participant

    OOPS!  I meant the TO line, not the subject line.

    0
    #152627

    Jim Shelor
    Participant

    No alternate.
    That happens to most people because they spell my last name wrong.
    Copy and paste [email protected] into the subject line of the message.

    0
    #152760

    Tierradentro
    Participant

    You can use a technique called an ISOPLOT which is a special type of regression analysis (Scatter Diagram).  This is a technique developed by the late Dorian Shainin and is an excellent tool.
    The Discrimination is figured by determining the Product Variation and dividing it by the Measurement Variation.  One must acheive 6 or greater.  10 is best.

    0
    #152762

    Mikel
    Member

    ISOPLOT is a crappy tool. Use real math.
    The plot is a good way to visualize a problem, but not to analyze a problem.
    People using Shainin tools are stuck in the 1970’s.
     

    0
    #152766

    GB
    Participant

    Heay maaaan,
    that ISOPLOT… It’s really groovy maaaaan.
    maaaan, I’ve got the munchies
    peace.

    0
    #152773

    Mike Carnell
    Participant

    hbgb b^2,
    If you get some time could you give me a call tomorrow?
    Regards

    0
    #152845

    Tierradentro
    Participant

    The Isoplot works fine for me and our processes despite your ignorant comments.
    and your solution is?  I noticed you criticize often but have no solutions.  Please entertain us with your answer to the discrimination issue.

    0
    #154364

    Bob Porter
    Participant

    Hi Jim.  I understand your response on discrimination but am wondering why multiply by 1.14?  Thanks,  bob

    0
    #154369

    Jim Shelor
    Participant

    Bob,
    There are two errors in my original post.  I said:
    You perform the MSA, and you divide the sigma (parts) by the sigma (gage) and multiply by 1.14, then truncate the answer to an integer.
    That sentence should have said:
    You perform the MSA, and you divide the sigma (parts) by the sigma (measurement) and multiply by 1.41, then truncate the answer to an integer.
    I do not know where the 1.41 comes from, but now that you asked, I am going to have to do a little study to see if I can find out.  It seems like we are picking up a sqrt(2) from somewhere.
    Regards,
    Jim Shelor

    0
    #154370

    Allthingsidiot O
    Participant

    It  should  be  6

    0
    #154373

    Jim Shelor
    Participant

    Bob,
    The link below will take you to the derivation of the NDC equation.
    http://www.minitab.com/support/docs/Answers/Root2.pdf
    It shows where the 1.41 comes from.
    I hope this helps.
    Jim Shelor

    0
    #154374

    Mikel
    Member

    Go read Wheeler’s evaluating the measurement system – it will explain it all.

    0
    #154375

    Bob Porter
    Participant

    Thanks much, your response was very helpful.  bob

    0
    #154377

    Bob Porter
    Participant

    Thanks, I will check it out.  I got another posting with a web site also.  bob

    0
    #154386

    Allthingsidiot O
    Participant

    Excellent  Elaboration

    0
Viewing 25 posts - 1 through 25 (of 25 total)

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