iSixSigma

MSA on Destructive Tests

Six Sigma – iSixSigma Forums Old Forums General MSA on Destructive Tests

Viewing 4 posts - 1 through 4 (of 4 total)
  • Author
    Posts
  • #26921

    Wazza
    Member

    Can somebody help me to do an MSA on a destructive test. The test is a tensile test which applies a tensile force to a strip of steel until it fractures. Once fractured it cannot retrieve its original form, and therefore any sort of repetition is not possible. Please help…..

    0
    #65458

    Doug Mader
    Participant

    All you really need is Lenth’s PSE method. It computes an estimate of the standard error of the effects based on the median of the effects rather than on the standard error. You can then use the PSE method to estimate the repeatability. There was a recent article (I am sorry I don’t remember the author) on computational stuff for the PSE method.

    0
    #65459

    Doug Mader
    Participant

    The article was in JQT.

    0
    #65462

    Joseph Conklin
    Participant

    There are at least two approaches one can take.

    One approach is a Latin Square design. Take a sheet of material and segment as follows:

    ______________________
    1 | 2 | 3 | 4 |
    —|—–|—–|—–|
    5 | 6 | 7 | 8 |
    —|—–|—–|—–|
    9 | 10 | 11 | 12 |
    —|—–|—–|—–|
    13 | 14 | 15 | 16 |
    ___|_____|_____|_____|

    The 4×4 scheme shown here is for illustration. You can use 3×3, 5×5, etc., depending on what makes sense for
    your application.

    In the 4×4 scheme, the 16 test specimens are tested in random order. The analysis of the data from a Latin Square uses the row and column positions of the test specimens to remove or at least minimize the effect of material variation.

    The root mean square error from the analysis estimates the precision of the test method.

    If one operator is used, one Latin Square layout is enough. If you wish to test the effect of multiple operators, one could try what’s called a Graeco-Latin square, or one could use one Latin Square layout for each operator.

    For more detailed information you may wish to consult the Douglas C. Montgomery book Design of Experiments or
    any similar text. The statistics division website of the American Society for Quality (www.asq.org) can provide you with more references and contacts.

    If a Latin Square approach is not feasible, you may wish to form pairs of test specimens. For example, consider the 4×4 illustration above:
    _____________________
    1 | 2 | 3 | 4 |
    —|—–|—–|—–|
    5 | 6 | 7 | 8 |
    —|—–|—–|—–|
    9 | 10 | 11 | 12 |
    —|—–|—–|—–|
    13 | 14 | 15 | 16 |
    ___|_____|_____|_____|

    Pairs consist of adjacent segments like (1,2) and (3,4)and so on. The pairs are treated as repeated
    measures in the MSA.

    Under this approach the estimate of test precision contains is not pure. It contains some contribution of variation introduced by the fact you are not testing the same specimen.

    The theory goes that if you can’t test the same specimen twice, then test two specimens that are exactly alike once each.

    The assumption under the paired approach is that specimens that are next to each other are as nearly alike as it is possible to get.

    I could say more but I don’t want to go on too long.

    Good luck.

    Joe Conklin
    [email protected]

    0
Viewing 4 posts - 1 through 4 (of 4 total)

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