Modeling Accelerated Degradation Data for Product Reliability Improvement and Warranty Analysis

这笔赠款支持研究开发加速降级测试(ADT)的新方法，包括分析产品或系统在各种高应力环境中的降级，以预测产品寿命或系统性能。这项研究集中于当前在工业中使用的测试方法中存在的差距，并致力于将这些当前方法扩展到更大的问题领域，包括描述制造业中更现实的产品退化场景的非标准退化模型。最初的工作是基于真空荧光显示器(VFD)、发光二极管和光纤制造的应用问题。VFD的表现有助于激励这些模式。从阴极发射的电子用于消除真空中的杂质，VFD的光强度实际上增加到某个时间点，然后由于年龄诱导的退化而下降。标准的ADT模型不能描述这种现象。这项研究的具体进展包括：(1)基于物理驱动的退化路径的应力依赖退化模型的一般框架；(2)建立各种复杂(非线性)ADT模型的失效时间公式，以及基于统计重采样方法的不确定性估计；(3)将失效时间数据与单独的退化数据集相结合以改进产品寿命估计；以及(4)基于与产品开发相关的时间和成本约束来寻找最优测试程序(根据获得的信息)。ADT模型包括(非线性)随机系数，以反映测试单元之间的可变性。由于对于这样的随机系数退化模型通常不能获得更简单的方差近似，因此推导了自举重采样过程以确定不确定性。利用降级数据，提出的产品失效时间分布的估计作为评估公司在高可靠性产品的过程改进的关键质量度量，并可以帮助公司经理决策他们的产品保修政策。如果成功，这项研究的结果将对工艺条件的改变、材料的选择、设备的创新、维护计划的修订和其他生产操作的改变产生强烈的影响。从而为企业提高运营效率和盈利能力提供有价值的信息。

This grant supports research for deriving new methods for accelerated degradation testing, or ADT, which involves analyzing product or system degradation in various high stress environments in order to predict product lifetime or system performance. The research concentrates on existing gaps in current test methods used in industry, and works to extend these current methods to a larger domain of problems, including non standard degradation models that describe more realistic product degradation scenarios in manufacturing. Initial work is based on applied problems with vacuum fluorescent displays (VFDs), light-emitting diodes, and fiber optics manufacturing. VFD performance helps to motivate such models. Emitted electrons from its cathode serve to eliminate impurities in the vacuum, and VFD light intensity actually increases up to a certain point of time before it decreases due to age-induced degradation. Standard ADT models cannot characterize this phenomenon. Specific developments of this research include: (1) A general framework for stress dependent degradation models based on physically motivated degradation paths; (2) Building formulas for failure times of various complex (non-linear) ADT models, along with uncertainty estimates based on statistical resampling methods; (3) Combining failure time data with separate sets of degradation data to improve product lifetime estimates; and (4) Finding optimal test procedures (in terms of information gained) based on time and cost constraints associated with product development. The ADT models include (non linear) random coefficients to reflect variability between test units. Bootstrap resampling procedures are derived to ascertain uncertainty because simpler variance approximations are not generally available with such random coefficient degradation models. With degradation data, the proposed estimates of the product failure time distribution serves as a key quality measure for evaluating a company's process improvement in highly reliable products, and can help company managers decide their product warranty policy. If successful, the results of this research can strongly affect process condition changes, material selections, equipment innovations, maintenance schedule revisions and other production operation changes. Thus, the research provides valuable information for companies to improve their operation efficiency and profitability.