Abstract The purpose of accelerated life testing (ALT) is to shorten the time in which reliability evaluation is possible. The usual assumption is that this shortening occurs by a linear contraction of the time scale, where the acceleration factor is the scaling term. Since most extrapolations from ALT are based on this assumption it is important to understand how this assumption can go wrong and what might be happening if it does. In this paper a strategy for the design and graphical analysis of ALT is described which allows one to explore both the linear contraction assumption, and assumptions about the particular functional form of the acceleration factor. One analytical example is then discussed in which statistical methods based on this approach provide a far more powerful test of the assumptions going into the extrapolation than comparable statistical methods based on the standard approach of searching for curvature in plots of the stress function against log quantiles. This approach is also shown to allow differentiation between standard acceleration models, and acceleration models involving competing risk.
[1]
W. Meeker.
Accelerated Testing: Statistical Models, Test Plans, and Data Analyses
,
1991
.
[2]
D. Cox,et al.
Analysis of Survival Data.
,
1985
.
[3]
D. J. Klinger,et al.
Failure time and rate constant of degradation: An argument for the inverse relationship
,
1992
.
[4]
Michael J. LuValle,et al.
Acceleration transforms and statistical kinetic models
,
1988
.
[5]
J. Kalbfleisch,et al.
The Statistical Analysis of Failure Time Data
,
1980
.
[6]
B. Efron.
The jackknife, the bootstrap, and other resampling plans
,
1987
.
[7]
Michael J. LuValle,et al.
A note on experiment design for accelerated life tests
,
1990
.
[8]
R. L. Hartman,et al.
Methodology of accelerated aging
,
1985,
AT&T Technical Journal.
[9]
E. Kaplan,et al.
Nonparametric Estimation from Incomplete Observations
,
1958
.