From: Stan Hilliard
Date: 06 Nov 1999
Time: 01:03:13
Greetings Yeon,
I would use a sampling plan for the mean to sample bulk material. I have a page discribing this kind of application ysing TP414 at:
www.samplingplans.com/outputvariablesmean.htm
A main concern with sampling bulk materials is non-homogeniety within the bulk. Therefore it is recommended to test individual samples from different locations in the bulk without blending them. Their range is a measure of non-homogeniety.
Sampling plans for the mean can be one-sided or two-sided. I suspect the C3A would be one-sided with an upper limit. (<8%) The AQL mean and RQL mean could be 1 to 2 standard deviations apart. For example if sigma=0.50%, you could set RQL=8.5% and AQL=7.5% C3A.
I think Mil-Std-105D is a bad idea. It requires the sampling of individual items from a lot containing N items. This doesn't work with bulk material - which doesn't have items. It also uses pass/fail data which usually requires a higher sample size than a measured variables plan. By the way, Mil-Std-105D is not based on 95% probability of acceptance.
Sincerely, Stan Hilliard