From: Stan Hilliard
Date: 19 Jul 2003
Time: 12:59:12
Chris,
Assuming numerical test data ("variables data").
You can have a much lower sample size if you can measure failures -- as opposed to pass/fail data. For example, you can test each wheel to failure and record the force that broke it, or you could record the RPM at which it failed.
The sampling plan's probability statement will take the form: "if the fraction of wheels nonconforming to the specified breakpoint=X is 0.001, the sampling plan provides only Pa=5% probability of acceptance of the lot."
Alternatively, you could specify in terms of the mean of the variable, but I dont think that applies to your application.
With variables data the sample size required to meet the same probability requirement as the attribute plan will be much smaller and a sequential variables plan will have an even smaller sample size (as described by an "Average Sample Number"=ASN curve"
TP414 will calculate fixed-n and sequential sampling plans for variables data.
The assumption is that the variable is normally distributed. If it is not, you would need to transform the data prior to applying it to the sampling plan that TP414 calculates.
These concepts are described in the tutorial, and in the description of TP414.
Tutorial: www.samplingplans.com/modern3.htm
TP414: www.samplingplans.com/programtp414.htm