Simulation-based Bayesian optimal ALT designs for model discrimination

摘要: Accelerated life test (ALT) planning in Bayesian framework is studied this paper with a focus of differentiating competing acceleration models, when there uncertainty as to whether the relationship between log mean and stress variable linear or exhibits some curvature. The proposed criterion based on Hellinger distance measure predictive distributions. optimal stress-factor setup unit allocation are determined at three levels subject test-lab equipment test-duration constraints. Optimal designs validated by their recovery rates, where true, data-generating, model selected under DIC (Deviance Information Criterion) selection rule, comparing performance other plans. Results show that design method has advantage substantially increasing plan׳s ability distinguish among ALT thus providing better guidance which appropriate for follow-on testing phase experiment.