EAGER: Interpretable and Generalizable AI for Smart Manufacturing

这一早期概念探索性研究资助(AGER)奖旨在概念化和研究通用机器学习框架和相关软件工具，该框架和相关软件工具需要对从全面运营的商业微电子制造设施获得的制造数据进行分类，并使用机器学习方法从这些数据中得出可靠的控制行动。与制造业相关的机器学习方法的研究一直受挫，因为缺乏实现它所需的大量经行业验证的数据。该项目将探索新的机器学习方法的潜力，以揭示包含在这些数据中的隐含知识，以提高产量和生产率。该项目解决了机器学习(ML)在制造系统中应用的三个最关键的障碍：(1)缺乏研究和开发适合于制造业衍生数据的机器学习架构所需的大量数据，(2)缺乏针对制造业特定数据进行汇总和分类的ML方法，以产生针对特定工艺、机器或操作的培训ML系统的数据集，以及缺乏为这些数据设计并可以使用这些数据进行推理的ML架构，以及(3)制造工程师不愿相信“黑箱”方法。该项目是与希捷科技合作，以解决所有三个障碍。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This EArly-concept Grant for Exploratory Research (EAGER) award is to conceptualize and research a generalized machine learning framework and the associated software tools needed to categorize manufacturing data acquired from a full-scale, operating commercial microelectronics fabrication facility and derive reliable control actions from that data using machine learning methods. Research on manufacturing-relevant machine learning methods has been frustrated by a lack of access to the large amount of industry-validated data needed to enable it. The project will explore the potential of new machine learning methods to reveal the implicit knowledge incorporated in that data to improve yield and productivity.The project addresses the three most critical impediments to the application of machine learning (ML) in manufacturing systems: (1) a lack of access to the massive amounts of data needed to research and develop machine learning architectures that are suited to manufacturing-derived data, (2) a lack of manufacturing-specific ML methods for aggregating and classifying that data to produce datasets tailored to training ML systems for specific processes, machines or operations and a lack of ML architectures that have been designed for and can make inferences using that data, and (3) a reluctance of manufacturing engineers to trust “black box” methods. The project is a collaboration with Seagate Technology to address all three impediments.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.