D
So far for one-sided tolerance non-replicable measurement system GRR analysis, the often used method I saw is to use the formula GRR%=GRR/TV by using iid samples.
The disadvantage of using this method is that it can’t differentiate the variance within a subgroup from the Equipment and appraiser variance, therefore makes the GRR% value much bigger.
I think that Tolerance should be used to decide if the GRR meets the requirements of product quality control especially for non-replicable measurement system.
Suppose the result data were normally distributed.
For example:
The sample hardness specification is > 60, so it is a one-sided tolerance.
Using three operators, take ten groups of samples, every group has nine samples that their variances were supposed negligible, and measure them.
Suppose the average of the above gathered data is 70, because we presuppose it’s normal distribution, we can predicate that the chance of the hardness value being 80 is as small as it being 60. Because 60 is the lower limit, I think we can use the specification 70+-10 as a specification substitute for the one-sided tolerance specification, therefore we can use the tolerance (80-60) divided by 6 to replace TV to get a better-looking GRR% result.
I don’t know how others think of this method. You comments are appreciated.
The disadvantage of using this method is that it can’t differentiate the variance within a subgroup from the Equipment and appraiser variance, therefore makes the GRR% value much bigger.
I think that Tolerance should be used to decide if the GRR meets the requirements of product quality control especially for non-replicable measurement system.
Suppose the result data were normally distributed.
For example:
The sample hardness specification is > 60, so it is a one-sided tolerance.
Using three operators, take ten groups of samples, every group has nine samples that their variances were supposed negligible, and measure them.
Suppose the average of the above gathered data is 70, because we presuppose it’s normal distribution, we can predicate that the chance of the hardness value being 80 is as small as it being 60. Because 60 is the lower limit, I think we can use the specification 70+-10 as a specification substitute for the one-sided tolerance specification, therefore we can use the tolerance (80-60) divided by 6 to replace TV to get a better-looking GRR% result.
I don’t know how others think of this method. You comments are appreciated.