An extended hazard regression model for accelerated life testing with time varying coefficients

Accelerated life testing (ALT) is a method for testing units in a more severe environment than the normal operating environment to collect failures in a short period of test time. Inference about the reliability of units at normal conditions can be made from test results obtained at the ALT conditions. In this dissertation, we propose an Extended Linear Hazard Regression (ELHR) model to analyze failure data for ALT. The ELHR model is a generalized model that combines the advantages of the Extended Hazard Regression (EHR) model, the Proportional Hazards model (PH) with time-varying coefficients and the Accelerated Failure Time (AFT) model.; Since some of the ELHR model's covariate coefficients are within its baseline hazard rate, partial likelihood method is not appropriate for parameter estimation and the baseline hazard rate function must be a parametric form. We develop and maximize a likelihood function that estimates all the parameters of the ELHR model. After parameters are estimated, the point estimate of reliability at a given time can be obtained. We also develop methods to estimate the confidence intervals of the estimated reliability and give a hypothesis test for checking the significance of the parameters in the ELHR model.; To validate the model, we fit the ELHR model using simulated data sets and an experimental data set analyzing time-dependent-dielectric-breakdowns (TDDB) of thermal oxides on 6H-SiC. The results show that the ELHR model provides more accurate reliability estimates and confidence intervals of the estimated reliability without being limited by the underlying assumptions of the PH model and the AFT model.