ALT analytical solution vs. SimuMatic

[MarkT]

I'm designing an ALT using the same approach that is discussed in Reliability Edge V8 iss 1.

I have conducted a prior life test so I am reasonably confident in the failure probabilities used for the test plan generation, although the life data is based on a test conduced at a fixed stress level that is above the nominal usage stress (but below the maximum stress level that is to be used in the ALT), so I do expect a degree of error. However, regardless of the type of test plan I select, the bounds ratio between the analytical approach and SimuMatic differs considerably, usually by a factor of two depending on what type of test plan I select, which is a far greater discrepancy than I expected.

I cannot find any errors in the data used for the two approaches, but considering the fact that the analytical solution assumes that the log transformation of BX life is normally distributed I'm arriving at the conclusion that I should rely more on the bounds ratio that I have calculated from the SimuMatic results, and that in my case the log transformation of BX life is non-normally distributed.

I appreciate that it is difficult (read impossible!) to provide definitive guidance without reviewing my data, but does the conclusion that I have drawn sound reasonable?

Thanks,
Mark

[MarkT]

Returning to this problem after a good nights sleep the error became obvious, but in the spirit of assisting others to avoid the trap that I fell into I thought I'd post the solution, even though the root cause was my stupidity!

When a test plan is generated, the probability of failure at the low and high stress levels have to be the probability of failure over the ALT duration at those stress levels, not the probability of failure over the preliminary test duration, which I seleted...fool!

This may sound obvious in hindsight, and it probably is, but for some stupid reason it didn't occur to me at the time to adjust the durations over which the probabilities were estimated.

Thanks,
Mark