Prediction bounds in accelerated life testing: Weibull models with inverse power relationship

We consider the problem of obtaining lower prediction bounds for a future observation from a Weibull population at design (field use) stress level, using Type II censored accelerated life test data from higher than design stress levels. The scale parameter of the life distribution is assumed to have an inverse power relationship with the stress level. We derive an easily obtainable predictive density for a future observation at the design stress level. The use of the lower percentile points of this predictive density as the lower prediction bounds for a single future observation is investigated using Monte Carlo simulation. Results show that a modified version of the proposed predictive density provides reasonable lower prediction bounds.