Sprt

[irishavis]

I am having some difficulty with the concepts in the Sequential Probabilty Ration Test (SPRT). I understand all of the basic ideas. I have an issue with what to choose for the ideal discrimination ratio. The larger the discrimination ratio, the smaller amount of test time is required to accept the hypothesis that the reliability has improved. I undertsand this is because the father apart the two reliability levels are, the easier it is for the test to discriminate which level the true component relaibility is leaning toward.

But with high ratios, the required test time just seems too small. Is there an upper limit for discrimination ratios at which point this test becomes impractical or next to irrelevant. I have a hard time explaining to management that my input for the test was an upper limit mean time between failure of say 50 hours and the test requires only 30 hours with no failures (because of a high discr ratio) to prove the test successfull. Appreciate any thoughts.

[Pantelis]

Well whats your actual requirment? Use values close to what you are trying to demonstrate.

[irishavis]

Thanks for the reply. I think I came to a solution for my problem. It is not a situation where I am trying reach a requirement. I just need to show that a design change improved the reliability of a certain failure mode. I have a component that demonstrates a specific failure mode MTBF of 10 hours. There was a design improvement for that specific failure mode. The lower limit for the SPRT would be 10 hours (the new design needs to have improved the reliability). The MTBF for the failure mode on the the new design is likely MUCH greater than 10 hours (maybe 1000 hrs).

If alpha = beta = 10% and ratio = 2.0, SPRT requires 43.9 hours with no failure to prove that the failure mode MTBF of the new design is 20 hours or greater. But that doesn't say much if the new design MTBF is 1000 hours. So I can increase the ratio to prove new design is at least 4, 5, 6... times as good as the old design, but then the minimum test time keeps decreasing.

However, I just realized though that I can make the lower limit whatever I want. It does have to be the original MTBF. I can make the lower limit 100 hours, use a ratio of 2 (for example) and prove the new design is at least 200 hours and the that test for the minimal testing time.

[Pantelis]

Sounds like you are set then ... good.