1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.
Dismiss Notice
You must be a registered member in order to post messages and view/download attached files in this forum.
Click here to register.

Gage R bias and Cgk conflict

Discussion in 'Gage R&R and MSA - Measurement Systems Analysis' started by Joaquin, Dec 10, 2020.

  1. Joaquin

    Joaquin Member

    Joined:
    Dec 10, 2020
    Messages:
    6
    Likes Received:
    0
    Trophy Points:
    1
    In a gage repeatability study (Type 1 gage study) of a testing machine, the p-value of the t-test is 0.000 which means that there is bias, but the specification limits are so wide that the Cgk values are above 20. Could someone elaborate on what should be done in this case?
     
  2. Miner

    Miner Moderator Staff Member

    Joined:
    Jul 30, 2015
    Messages:
    576
    Likes Received:
    492
    Trophy Points:
    62
    Location:
    Greater Milwaukee USA
    The p-value is just indicating that the bias exists and is statistically significant. You now have to make a determination whether this bias has any practical impact on your measurements. Compare this against your calibration standards. If the bias is within these standards, you need not do anything. If it exceeds the standards, you should either adjust the gage to address the bias, use the known bias to correct your measurements, or open your calibration standards if the importance is nil.
     
    Andy Nichols likes this.
  3. Joaquin

    Joaquin Member

    Joined:
    Dec 10, 2020
    Messages:
    6
    Likes Received:
    0
    Trophy Points:
    1
    My confussion comes when using the specification limits and mixing them with a repeatability study--in the previous MSA I did, I didn't use this Type 1 Gage Study at Minitab.

    Taking the definition from Minitab page:
    To assess a gage's repeatability, Minitab calculates the Cg metric to compare the study variation (the spread of the gage's measurements) with a percentage of the tolerance. Values of Cg greater than 1.33 indicate the spread of the gage's measurements is adequately narrow in relation to your tolerance range.
    For example, with the default values of K and L, a Cg metric of 2 indicates that 20% of your tolerance range will cover the entire spread of measurements twice over. This Cg value indicates the gage's effectiveness within this tolerance range.

    For me, this Cg seems an estimation of the capability of the measurement system, but not the repeatability. Basically because I can have a large variation in the measurement system (it is not repeatable) and at the same time obtain an aceptable Cg as soon as my specification limits are wide enough.

    After this, Minitab mixes this Cg concept with bias in the Cgk:
    Minitab also calculates the capability metric Cgk to assess repeatability and bias together. Cgk compares the Study Variation to the tolerance, but it also considers whether the measurements are "on target". Cgk decreases as the difference between the gage's mean measurement and the reference value increases. A Cgk value of 1.33 is a common benchmark value to denote a capable gage – one that is both precise (good repeatability) and accurate (low bias).

    So at the end, I can have an acceptable Cgk in a measurement system that has bias (confirmed by the hypothesis test) and large variation in the measures (not repeatable) and conclude the opposite if I have specification limits wide enough to achieve a Cgk>1.33.


    https://support.minitab.com/en-us/m...-gage-studies-and-measures/type-1-gage-study/
     
  4. Miner

    Miner Moderator Staff Member

    Joined:
    Jul 30, 2015
    Messages:
    576
    Likes Received:
    492
    Trophy Points:
    62
    Location:
    Greater Milwaukee USA
    Can you attach your data and the Minitab analysis? It is much easier for me to respond with something concrete to look at.
     
  5. Joaquin

    Joaquin Member

    Joined:
    Dec 10, 2020
    Messages:
    6
    Likes Received:
    0
    Trophy Points:
    1
    I have this case at hand, it is not that extreme but still same reasoning.

    Data
    0.80
    1.50
    0.95
    1.35
    0.50
    0.80
    0.95
    1.05
    1.70
    1.85
    1.05
    0.55
    1.95
    1.15
    1.35
    1.65
    0.45
    0.30
    0.70
    1.90
    0.40
    0.85
    2.00
    1.05
    0.80

    upload_2020-12-10_22-1-4.png
     
  6. Miner

    Miner Moderator Staff Member

    Joined:
    Jul 30, 2015
    Messages:
    576
    Likes Received:
    492
    Trophy Points:
    62
    Location:
    Greater Milwaukee USA
    Assuming that the tolerance of 100 is correct, your repeatability plus bias of 3.38% looks very good. I don't think you have anything to be concerned about. You should cover your bases by documenting your justification for accepting this gage.
     
  7. Joaquin

    Joaquin Member

    Joined:
    Dec 10, 2020
    Messages:
    6
    Likes Received:
    0
    Trophy Points:
    1
    My concern is not about the capability of the measurement system to spot a part above the specification limits--which obviously is perfectly able, but for the quality of the data when performing a SPC in later stages, that could mask the variation or mean shifts of the process. Wouldn't be more suitable to compare the gage variation with the parts variation for cases with "high" (in this case is in the order of microns) specification limits?
     
  8. Joaquin

    Joaquin Member

    Joined:
    Dec 10, 2020
    Messages:
    6
    Likes Received:
    0
    Trophy Points:
    1
    Reviewing the APQP MSA manual, it stabilises that when the measurement system is also used for Process Control, and not just to define conformance with tolerance (Product Control), the variation of the measurement system (in this case repeatability) should be compared to process variation and not with tolerance range:
    %EV = Repeatability std dev/Process std dev
    <10% acceptable
    10%-30% may be acceptable for some applications
    >30% unacceptable
     
  9. Miner

    Miner Moderator Staff Member

    Joined:
    Jul 30, 2015
    Messages:
    576
    Likes Received:
    492
    Trophy Points:
    62
    Location:
    Greater Milwaukee USA
    What is your process variation?
     
  10. Joaquin

    Joaquin Member

    Joined:
    Dec 10, 2020
    Messages:
    6
    Likes Received:
    0
    Trophy Points:
    1
    Hi Miner,

    The equipment variation was around 150% of the process variation which is what I suspected, but in this case the large specification limits hide this.

    Given these results, the testing equipment is considered valid for spotting a part above specs but obviously not valid for process control. Also long term bias and repeatability should be analysed over time to make sure these issues do not go out of control in the future.