From: Stan Hilliard
Date: 24 Feb 1999
Time: 12:51:19
True, but not a serious problem.
Wearout causes the weibull slope (b) to increase, which decreases the variance, which reduces the actual sample size(test time/cycles) required for acceptance.
The following relationship between weibull slope and variance is from page 293 of Kapur and Lambertson, "Reliability in Engineering Design":
SLOPE VARIANCE/(SCALE)
0.5 20.000
1.0 1.000 (exponential)
2.0 0.215
3.0 0.105
4.0 0.081
In the wearout case, TP781 - using the exponential distribution - leads always to a safely inflated sample size.
Infant mortality failures cause the weibull slope (b) to decrease, which increases the variance, increases the sample size required.
In the infant mortality case, TP781 - using the exponential distribution - will understate the sample size. I think that this is not dangereous, however, because the infant mortality failures will prevent erroneous acceptance.
In conclusion, with both wearout and infant mortality, I think it is safe to develop the initial test plan using the exponential distribution and TP781.
I plan to post a more thorough analysis at www.samplingplans.com/wearout.htm during the next week.
Sincerely, Stan Hilliard