Optimal Allocation of Test Items in an Accelerated Life Test under Model Uncertainty

In accelerated life testing, a relationship is usually assumed between the stress and a parameter of the lifetime distribution. However, the true relationship is not usually known, and therefore, the experimenter may wish to provide protections against the likely departures from the assumed relationship. This paper considers an accelerated life test in which two stress levels are involved, and the lifetime of each test item at a stress level is assumed to have an independent, identical, exponential distribution. For the case where a first order relationship is assumed while the true one is quadratic, a procedure is developed for allocating test items to stress levels such that the bias and/or the variance of the estimated(log-transformed) mean lifetime at the use condition is minimized.