The Optimum Design for Accelerated Degradation Experiments with Linear Random Coefficient Model

• Main Authors: Zu-Liang Lin, 林祖樑
• Other Authors: Shuen-Lin Jeng
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/73970823279579672416

碩士 === 東海大學 === 統計學系 === 95 === ================================== [ ABSTRACT ] ================================== Traditional accelerated life tests (ALT's) record failure and censoring times of devices subjected to elevated stress. For highly reliability devices, ALT's provide little information because of few failures, even at very high levels of stress. On the contrary, accelerated degradation tests (ADT's) take measurements along the experiment. These data provide valuable information on the failure mechanisms of the devices. This is the reason why ADT's are getting more popular in industry today. In our study, we focus on linear degradation model with random coefficient, arises in some simple wear process. Some other nonlinear degradation relationships may be transformed into linear forms, then they also can be solved by our approach. The main purpose of our study is to decide the position of the lower stress of an two-level ADT design by minimizing the asymptotic variance of a particular percentile of the device's lifetime distribution at the use-condition. The distribution of random coefficient is modelled by lognormal and gamma, respectively. Joint modelling of degradation measures and lifetime data is considered in parameter estimation by using maximum likelihood procedure. We name it as accelerated life-degradation test (ALDT). There are two major results in this study. The first one is to provide the optimal lower stress level in design stage of an ADT under certain range of distribution parameters. The second one is to discover that by using joint modelling, there is more efficient in quantile estimation. That is, an ALDT will give us a better plan than by ALT and ADT. We will also discuss the robustness when the model distribution is misspecified.