From: Stan Hilliard shilliard@samplingplans.com
Date: 13 Aug 2001
Time: 13:38:33
Hi Hakan,
No, not yet - nothing has been published describing TSS except for what I have put on samplingplans.com.
I developed the TSS method to support my consulting and teaching of statistics in quality control, but the article that you ask for is not finished. To finish it is on my to-do list. (That will follow my current project, which is to finish upgrading the software.) TSS for variables is implemented in the software program TP414: www.samplingplans.com/software_oc.htm
I always prefer TSS to SPR plans, but SPR is included in the software for the benefit of those users who would rather use a method that is covered broadly in the literature.
Some special benefits and properties of TSS plans are:
(1) TSS sampling plans for variables never exceed the fixed-n sample size for the same oc curve. The matched fixed-n and its decision limit are on the last row of the plan. This engenders credibility by the people starting to use sequential plans. I have found that some people resist using SPR plans when they learn that their sample sizes have no theoretical maximum.
(2) TSS sampling plans for variables can be made into double and multiple plans.
(3) TSS sampling plans for variables can accommodate both known-sigma and unknown-sigma applications.
(4) TSS sampling plans for variables apply to either the percent nonconforming to specifications, or to the lot mean.
In the case of the lot mean, the AQL mean and RQL mean are specified directly without reference to specifications for individual items. This makes the mean plans suitable for non-item based materials -- like chemicals, liquid solutions, powders, pellets, color matching, etc.
"Known-Sigma" TSS plans for the mean and for variables non-conforming to specification limits are verified by simulation to exactly achieve the specified risks and the oc-curve.
About the assumption of "known-sigma" With any "known sigma" plan, good practice calls for checking the sample range or standard deviation to ensure that the standard deviation of the current lot does not differ from the "known" value. TP414 generates decision limits for the range test. If the test rejects the known-sigma because of high variability in the sample, then you switch to a matched unknown-sigma fixed-n or sequential plan to make the acceptance decision.
Sampling plans for unknown standard deviation are ISL plans. For mean plans, if the known standard deviation is rejected, you can perform a t-test that the true mean is equal to or better than the AQL mean. Rule: if the t-test rejects then reject the lot.
Sincerely, Stan Hilliard