iSixSigma

gage validation

Six Sigma – iSixSigma Forums Old Forums General gage validation

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
    Posts
  • #29315

    Rouge BB
    Member

    I am installing an automated gage system. I would like to be able to validate it statistically. I have a number of different product styles that will be “inspected” by this system. From analysis so far the samples are homogenous. What I would like to do is reinspect the product after the automated system and analyze the agreement or disagreement somehow. Does anyone have any ideas on statistically sound methods to accomplish this task?

    0
    #74754

    Gabriel
    Participant

    Is it continous or attribute data? If it is just an automatic measuring instrument (for example, a dial gage that automatically inspects 100% of the product in the line, and then sorts good and bad) then the clasical bias, linearity, stability and repeatability should work. You may need tha help of master parts (production parts that are calibrated in the metrology room at a level better enough than the requiered accuracy of the instrument)

    0
    #74755

    Rouge BB
    Member

    The data will be continuous. Although this will not be a destructive test, the configuration of the equipment will prevent the sample from being tested twice by the same instrument. The equipment configuration will also prevent a “master” sample from being intentionally loaded into the system. All verification will need to be done after the fact. As a additional note, the equipment will inspecting for numerous types of “defects”.

    0
    #74757

    Joy Cowling
    Participant

    Without actually seeing the system, this is my suggestion.
    If you really can’t re-test individual parts, then run the test like a destructive test.  Each batch should have homogenous samples.  Then treat the testing equipment as your operators.  Since this is an automated test machine, your reproduceability number will provide information about the machines agreeing with each other (instead of operators).
    My brain is on slow motion this morning. If this isn’t clear, let me know and I’ll try to clarify.
    Joy

    0
    #74812

    Gabriel
    Participant

    Sorry, I have no experience then to help you.
    But an idea has just came across my mind. Imagine that you specify that the bias of the instrument should be less than 10% of the tolerance, every time (stability) and all along the measuring range (linearity). Then the allowed error (true velue – measured value) would be 0+/-10%Tol. Now, if you can read the variable data shown by the instrument and after that you can measure the piece linked to the previous reading, you can establish this error for each part (and each characteristic to be checked) (the measurement after the equipment shold be very accurate compared with 10%Tol you want to demonstrate). After performing this test with many pieces (let’s say 100 at least) covering all the measuring range, you can calculate the average error and the standard deviation of this error and require that the average error +/- 4sigma be included in the 0 +/-10%Tol range (i.e. Cpk better than 1.33). This would be like making a process study on the “Measuring Process” and requiring a mimimum capability. Of course you can choose other allowance than +/-10%Tol and other requiered capability other than Cpk=1.33.
    Again, this is just an idea. Do you think this can work?

    0
    #74813

    Rouge BB
    Member

    Interesting idea,
    It may work. I have to look at that one a little closer. Thanks or the input.

    0
Viewing 6 posts - 1 through 6 (of 6 total)

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