Optimal allocations of stress levels and test units in accelerated life tests

Accelerated life testing (ALT) is a widely used approach for reliability demonstration and prediction of reliability of components or systems at normal operating conditions using data obtained at accelerated condition. The accuracy of reliability prediction depends on the stress types, stress levels and corresponding sample sizes. In this dissertation, we present efficient and practical approaches for determining the appropriate stress levels and number of test units at every stress level so as to obtain the most accurate reliability estimates at normal operating conditions.; Extensive research exists that deals with the accelerated failure time based (AFT-based) ALT plans. The proportional hazard (PH) is a much more realistic assumption in many cases in the reliability field; whereas PH-based test plans do not exist in the literature. In this research, we develop the first analytical ALT plans that utilize the PH model for reliability prediction. The plans determine the stress levels and the number of test units allocated to each level in an effort to minimize the variance of reliability estimate at normal operating conditions with constant stress application and Type-I censoring. Optimal test plans for simple step-stress and multiple-stress applications are also developed under the PH assumption.; The extended linear hazard regression (ELHR) model is a generalized model that encompasses the PH, the AFT and the extended hazard regression (EHR) models as special cases. In this research, we provide the first and most general study that applies the ELHR model in the ALT design. The test plans thus derived do not depend on the validity of the hazard rate proportionality assumption of the PH-based plans or the failure time proportionality assumption of the AFT-based plans. We establish the methodology for planning the optimal ALT plans under various stress applications, i.e. constant stress, step-stress and multiple-stress applications. Two types of baseline hazard functions are considered in the statistical model development—the quadratic baseline hazard and Hjorth's generalized baseline hazard.; We conduct an accelerated life test in the laboratory to investigate the time-dependent-dielectric-breakdown (TDDB) failure on n-type 4H-SiC MOS capacitors. The preliminary baseline experiment provides initial estimates of the ELHR parameters which are then used as input to the optimization problem of the test plans.