Sample size for destructive testing

cg21

New Member
#1
Hello all,

I'm in need of guidance on proving that a sample of 1 (test is VERY expensive and destructive) is enough to conclude unbiased and objective results. I have tried the risk based approach on based on confidence and reliability, but I'm still obtaining a large sample size.

Does anyone have any other ideas for statistical tests to prove that the results from 1 test represent my population? By the way, it is assumed that my population is homogeneous, as the product is in the form of a powder.

Thank you! Any suggestions are welcome!
 

noetsi

Fortran must die
#3
If you could prove that every test is identical to every other test, including external events, you could do it. I am not sure that is realistic. If you are simply trying to show something will blow up, as with atomic bombs, and it does then one test is enough. But even nuclear weapons were tested multiple times for various effects like height.

I think you could compare your test parameters to possible parameters that could exist. You can create simulations, this was done with nuclear weapons when they could not longer be tested, but they are only correct if the assumptions about the simulation are.
 

Miner

TS Contributor
#4
Are the results in the form of a continuous measurement, or a binary pass/fail? Do you have any historical information on the variation of the process?

If (and that is a big if) you have historical data that indicates that all of your process variation is at a comfortable distance from your specification AND your single test is within that variation, you can justify a single test in most cases.

Distribution Plot.jpg