Collaborative Research: Automated Knowledge Discovery in Reliability and Healthcare from Complex Data with Covariates

收集和分析带有协变量（如温度、湿度和辐射水平）的数据是科学和工程中的日常活动。一个重要的例子是锂离子电池等产品设计中使用的加速寿命测试数据。这些数据是通过将测试单元暴露在比正常条件更恶劣的环境中来收集的，以加速故障过程。由此产生的失效次数由概率分布和寿命-应力关系建模。然而，如果选择的概率分布和/或寿命-应力关系不能充分描述潜在的失效过程，则得出的可靠性预测可能会产生误导。在医疗保健系统中也遇到了类似的例子，其中量化重要指标（如住院时间、等待时间和疾病进展）的概率分布以及有影响的协变量对这些指标的影响至关重要。该奖项支持从可靠性和医疗保健领域的复杂数据中自动发现知识的基础研究，这些研究对制造业、医疗保健、能源、运输和航空航天工业领域具有潜在影响。研究小组将努力扩大弱势群体和少数族裔的参与，并对工程教育产生积极影响。这个项目的目的是研究一种新的方法，从复杂的协变量数据中自动发现知识，使用矩阵分析模型。将开发统计工具和优化算法，以便有效地收集这些数据或从大量数据中选择有用的子集以便快速实施。研究结果将有助于在数据产生机制未知或难以使用现有统计工具分析的情况下，为建模和解释这些数据创造新的途径。为此，将通过数学优化探索构建包含协变量的一般相位型分布的自动化建模方法。为了提高数据收集的统计效率，本文将研究一种优化的实验设计方法，并研究可行的计算工具来规划具有相位型模型的加速测试实验。此外，基于最佳实验设计方法的数据选择方法将被开发，以最大限度地利用医疗保健数据。研究结果将通过在实验室进行锂离子电池的加速测试，并与生物医学信息学服务机构合作，在目标医疗保健应用方面进行验证。

Collecting and analyzing data with covariates such as temperature, humidity, and radiation level are everyday activities in science and engineering. An important example is accelerated life testing data used in the design of products such as lithium-ion batteries. Such data are collected by exposing test units to harsher-than-normal conditions to expedite the failure process. The resulting failure times are modeled by a probability distribution and a life-stress relationship. However, if the probability distribution and/or the life-stress relationship selected cannot adequately describe the underlying failure process, the resulting reliability prediction may be misleading. A similar example is also encountered in healthcare systems, where it is crucial to quantify probability distributions of important measures such as the length-of-stay, waiting time, and disease progression, and the effects of influential covariates on these measures. This award supports fundamental research on automated knowledge discovery from complex data in reliability and healthcare with potential impacts in the areas of manufacturing, healthcare, energy, transportation, and aerospace industries. The research team will strive to broaden participation of underrepresented groups and minorities, and positively impact engineering education. The objective of this project is to investigate a new methodology for automated knowledge discovery from complex data with covariates using matrix-analytic models. Statistical tools and optimization algorithms will be developed for efficiently collecting such data or selecting the useful subsets from massive data for quick implementation. The research findings will help create a new avenue for modeling and interpreting such data in situations in which the data-generating mechanisms are unknown or difficult to analyze using existing statistical tools. To this end, an automated modeling methodology to construct general phase-type distributions incorporating covariates will be explored via mathematical optimization. To improve the statistical efficiency of data collection, an optimal experimental design methodology will be investigated, and viable computational tools for planning accelerated testing experiment with phase-type models will be studied. In addition, a data-selection approach based on the optimal experimental design methodology will be developed to maximize the utilization of healthcare data. The research findings will be validated by conducting accelerated tests of lithium-ion battery in the laboratory and collaborating with biomedical informatics services on targeted healthcare applications.