An Engineering Approach to Confidence Intervals and Endurance Test Strategies

Abstract

Confidence intervals defined by their lower and upper bounds have been define for Weibull slope and L₁₀ life using Monte Carlo simulations.

Miscellaneous testing Strategies have been explored, namely failing all N specimens or failing only the first specimen out of N (the sudden death approach), this being repeated NR times using NR groups of N specimens.

Statistics about the median duration time of the endurance test are also reported, showing that the sudden death approach is particularly attractive.

Confidence intervals for the L₁₀ can be significantly improved (ratio of two or more) by making use of a 'trick' that assumes that the Weibull slope is known form many previous tests, and to use it.

Recommendations are given for using a (N = 8, NR = 3) strategy instead of (N = 4, NR = 6), in order to reduce the testing time by two without affecting too much the accuracy on L₁₀ (if use id made of the trick).

If the slope needs to be measured, duplicating the (N = 8, NR = 3) tests allows analysis of the results as a (N = 8, NR = 6) case which provides the same testing time and same accuracy of the Weibull slope as in a (N = 4, NR = 6) case, but with a narrower confidence interval on L₁₀.

The zero failure strategy has also been shown to be the one that provides the shortest testing time for checking that the tested specimen life exceeds an expected life within a given probability. The inconvenience of such an approach is that the β and η values are not evaluated.

Presented at the 57th Annual Meeting in Houston, Texas May 19–23, 2002

KEY WORDS:

• Weibull
• Confidence Intervals
• Endurance Test Strategies
• Zero Failure Strategy

Notes

Presented at the 57th Annual Meeting in Houston, Texas May 19–23, 2002