Collaborative Research: Fusion of Siloed Data for Multistage Manufacturing Systems: Integrative Product Quality and Machine Health Management

随着产品变得越来越复杂，需要涉及多个阶段的现代制造工艺是非常常见的。一个阶段的输出工件是下一个阶段的输入工件。在多阶段制造系统中，不仅机器的健康状况与进出产品质量相互作用，而且出厂产品质量也影响下游机器的机器健康状况和产品质量。由于产品质量和机器健康状况在多个阶段之间存在复杂的相互作用，为了有效的系统监控和操作控制，在每台机器上单独考虑产品质量问题和机器状况退化问题是不够的（甚至是误导）。随着传感技术和信息技术的快速发展，在多级制造系统中的多个机器处的大量产品质量数据和机器健康状况数据被容易地获取和访问。该项目建立了一系列数据驱动的方法，通过对产品质量和机器健康数据的综合建模，实现多级制造系统的有效监控和运行。所开发的方法进行了测试和验证，在实验室测试平台和真实的生产系统与工业合作者。该项目通过促进制造、计算、传感和数据分析的跨学科研究来促进劳动力培训，并通过新课程开发和各种推广活动为学生提供独特的培训机会。研究了描述不同机器之间的空间相互作用和每个机器的时间退化的综合数学框架。在该框架的核心，一个灵活的非齐次隐马尔可夫模型被用来描述机器的时间退化。模型中考虑了产品质量和机器健康状况之间的相互作用，并引入了外部因素。围绕综合模型设计，四个相互关联的研究任务包括：（i）学习质量交互和局部异常指标，（ii）学习机器退化模型和故障预测，（iii）系统级操作优化的随机控制，以及（iv）测试和验证。该项目通过利用不断增长的可用质量数据和机器健康状况数据，使其在控制方面更加集成，从而为现代制造工厂提供了额外的能力。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

As products become more intricate, it is very common that modern manufacturing processes which involve multiple stages are required. The output workpiece of one stage is the input workpiece of the next stage. In a multistage manufacturing system, not only does the health condition of a machine interact with incoming and outgoing product quality, but the outgoing product quality also impacts the machine health condition and product quality at downstream machines. Due to the complicated interactions among product quality and machine health conditions across multiple stages, for effective system monitoring and operational control, it is insufficient (even misleading) to consider product quality issues and machine condition degradation issues separately at each individual machine. With the fast progress of sensing and information technology, a large amount of product quality data and machine health condition data at multiple machines in a multistage manufacturing system are easily acquired and accessed. This project establishes a series of data-driven methodologies to achieve efficient monitoring and operation of multistage manufacturing systems through integrative modeling of product quality and machine health data. The developed methodologies are tested and validated in a laboratory testbed and on real production systems with industrial collaborators. This project contributes to workforce training by promoting the interdisciplinary research of manufacturing, computing, sensing, and data analytics and provides unique training opportunities for students through new curriculum development and various outreach activities.An integrated mathematical framework to describe spatial interactions among different machines and the temporal degradation of each machine is investigated. At the core of the framework, a flexible non-homogeneous hidden Markov model is used to describe the machine temporal degradation. The interactions between product quality and the machine health condition are considered by incorporating exogenous factors into the model. Designed around the integrative model, four interrelated research tasks include: (i) Learning quality interactions and local anomaly indicators, (ii) Learning machine degradation model and failure prognosis, (iii) Stochastic control for system-level operation optimization, and (iv) Testing and validation. The project provides added capabilities for a modern manufacturing factory by making it more integrated in control through the exploitation of ever-growing available quality data and machine health condition data.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.