CAREER: Transfer Learning Based Quality Improvement in Spatially-Temporally Complex Systems

该学院早期职业发展（Career）奖的研究目标是开发基于“迁移学习”的方法，用于产品种类多、生命周期短的制造系统（称为时空复杂系统）的质量改进。这种系统通常存在于半导体和可再生能源制造业中。方法论上的创新——迁移学习——是指利用在一个过程（或一个过程的前几代）的质量控制中获得的知识来控制其他过程（或新一代）的质量的能力。将开发一套统计严谨、计算高效的基于迁移学习的方法，以实现各种质量控制目标，如过程建模、监测、根本原因诊断、传感器不确定性建模和传感器放置。与研究相结合的是一项雄心勃勃的教育计划，旨在为未来的劳动力提供新的科学和工程知识以及提高质量的相关技能。如果成功，本研究的结果将通过从其他过程和过去几代的有效知识转移，显著加快制造系统中每个过程质量改进的学习曲线。这将使强大的，实时的，甚至是主动的质量控制决策跟上快速的产品扩散和世代变化。通过在半导体和太阳能制造业的验证和应用，本研究将提供突破性的技术，显著提高这些行业的质量，生产力和成本效益。研究方法也可转移到其他系统的研究，如卫生保健服务，人类大脑和生物系统。将开展一系列广泛的教育活动，包括新课程开发、工业培训课程、本科生和少数民族学生的参与，以及K-12学生和教师的外展。将通过强有力的国际合作建立研究和教育项目，以实现全球影响。

The research objective of this Faculty Early Career Development (CAREER) award is to develop "transfer learning" based methodologies for quality improvement of manufacturing systems featured by a high product variety and short life cycles (called spatially-temporally complex systems). Such systems typically exist in semiconductor and renewable-energy manufacturing industries. The methodological innovation - transfer learning - refers to the capability of leveraging the knowledge gained during quality control of one process (or past generations of a process) for quality control of other processes (or a new generation). A body of statistically rigorous and computationally efficient transfer learning based methods will be developed to achieve various quality control objectives such as process modeling, monitoring, root cause diagnosis, sensor uncertainty modeling, and sensor placement. Integrated with the research is an ambitious education plan aiming at equipping future workforce with new science and engineering knowledge and associated skills in quality improvement. If successful, the results of this research will significantly expedite the learning curve in quality improvement of each process in a manufacturing system through effective knowledge transfer from other processes and past generations. This will enable robust, real-time and even proactive quality control decision making to keep up with rapid product proliferation and generation changes. Through validation and application in semiconductor and solar energy manufacturing, this research will provide breakthrough technologies to significantly improve the quality, productivity, and cost-effectiveness of these industries. The research methodologies are also transferrable to the study of other systems such as health care delivery, human brain, and biological systems. A broad array of educational activities will be pursued, including new course development, industrial training sessions, undergraduate and minority students involvement, and K-12 students and teachers outreach. Research and education programs will be established through strong international collaboration to achieve global impact.