Please forgive me if this question has been answered, I spent some time looking but couldn’t find something directly applicable.
We have an expensive product we need to pressure cycle test to prove it meets requirements.
Our requirement for cycle life is 1000 cycles minimum (95%,95%)
I used a binomial formula N=ln(αlpha)/ln(P) =ln(.05)/ln(.95) and determined that I needed 59 samples and no test failures.
But my manager said no way, use only 30 samples. I responded that for this sample size I calculate that I can only state 90% reliability (ln(.05)/ln(.90)=28 samples) and this is pretty much the same value as I get from table 2 of the RIAC article “Analysis of one shot devices”.
So I do the test with 30 samples and we are at 100,000 with no failures, way past our 1000 cycle requirement. We are at 100 times our requirement and I want to stop the test and write the test report as a pass.
But people here will state that I have no failure data to statistically analyze and my sample size calculation above states I tested only half the amount of samples I needed for the company (95%,95%) standard.
So I need another formula or modification for sample size that takes into account far exceeding requirements. For example, in the formula above I calculated that 28 samples with 0 failures would demonstrate 90% reliability with respect to my 1000 cycle requirement. I think its safe to say that I demonstrated that if I had no failures at 100,000 cycles under this sample size designed to demonstrate 90% reliability, I readily have 95% reliability with respect to my requirement with this sample size based on my results.
To further complicate things, our measure of failure is a leak test. The leak test results are unchanging from the baseline and I suspect that if and when a failure occurs I will have a sudden gross leak (there will be pass/fail data only). There is no data that can be used to model any distribution.