Collaborative Research: Process Monitoring and Control in Autocorrelated Multistage Manufacturing Processes

随着现代制造业变得越来越复杂，通常会发现一个涉及多个生产阶段的生产过程，例如制药、化学工业、半导体和汽车制造业。在多阶段制造过程中，数据流中的三种类型的相关性（阶段之间、质量特征之间和时间之间）在变化建模、分析和控制方面引入了重大挑战。该项目旨在开发有效的方法来监测、控制和优化自相关的多阶段过程，以提高过程输出的质量。这是一个具有挑战性的问题，因为多阶段过程的复杂性和观测的自相关性使得输出和输入变量之间的关系非常复杂。这将通过变量均值和方差传播到后续阶段的新模型进行研究，使用动态状态空间模型，能够识别变化源传播和过程的监测/诊断。这些方法将与行业伙伴合作开发和验证。该研究为自相关多阶段制造过程中最小化变化传播和快速检测变化点的方法和算法提供了科学基础。本项目的成功完成将为多阶段过程异常行为的检测提供在线监测和诊断方法。其结果将广泛适用于各种行业，以提高生产系统的整体质量和生产率。更广泛的影响还将通过新的课程模块、用于实施的在线软件工具包，以及让代表性不足的本科生和研究生参与研究经验项目，以增强美国工业的人力资源人才。

As modern manufacturing industries become more sophisticated, it is common to find a production process involving multiple stages of production such as those found in pharmaceutical manufacturing, the chemical industry and in semiconductor and auto manufacturing. Three types of correlations in the data streams (among stages, among quality characteristics, and over time) in a multistage manufacturing process introduce significant challenges in variation modeling, analysis, and control. This project aims at developing efficient methodologies for the monitoring, control and optimization of autocorrelated multistage processes in order to improve the quality of the process output. This is a challenging problem due to the complexity of multistage processes and autocorrelations of observations that make the relationship between the output and input variables extremely complicated. This will be investigated through novel models of the propagation of variable means and variances to subsequent stages, using dynamic state space models that enable the identification of the variation source propagation and monitoring/diagnosis of the processes. The methodologies will be developed and validated in collaboration with industry partners.The research contributes to the science base of methods and algorithms to minimize the propagation of variations and quickly detect change points in autocorrelated multistage manufacturing processes. Successful completion of this project will provide online monitoring and diagnosis methods for detecting abnormal behaviors of multistage processes. The results will be broadly applicable in a variety of industries to improve the overall quality and productivity of production systems. Broader impacts will be also generated through new curriculum modules, online software toolkits for implementation, and involving underrepresented undergraduate and graduate students in research experience programs to enhance the human resource talent for U.S. industry.