A least-squares approach to analyzing life-stress relationship in step-stress accelerated life tests

Abstract

A new approach is proposed for the data analysis of failure-censored step-stress accelerated life tests with exponential failures, following an assumed log-linear life-stress relationship. This approach, transformed least squares: 1) provides closed-form parameter estimates;, 2) motivates a simple statistic for life-stress relationship verification; and 3) appreciably simplifies the derivation of optimal test plans. The proposed D-statistic has good power and compares reasonably well with likelihood-ratio test statistic. Estimates are unbiased and exact for failure-step tests of any sample size. The derived optimal test plans agree with those derived using maximum likelihood estimation.