FMRG: Manufacturing USA: Cyber: Privacy-Preserving Tiny Machine Learning Edge Analytics to Enable AI-Commons for Secure Manufacturing

目前基于全球供应链的制造系统运行良好，以低成本生产大批量产品。然而，它们很容易受到扰动（例如，COVID-19）的影响，即使是对成功至关重要的流程，如新产品开发和生产升级和优化，也可能过于漫长和昂贵。人工智能（AI）已经彻底改变了许多行业，可以成为提高制造业生产力和敏捷性的重要工具，但制造业的进展缓慢。该未来制造研究基金（FMRG）美国制造：网络制造项目是对分布式AI/ML技术的一次根本性的重新构想，通过建立一个AI- commons，通过安全、分布式机器学习（ML）、激励信息共享以及持续的质量改进和培训来连接多个站点和公司，从而改变制造业的未来。这解决了当前人工智能在制造业中的一个关键缺陷；培训数据仅限于公司内部数据。由于人工智能算法的能力随着数据的增加而增强，安全的数据共享和聚合有可能为所有制造商提供更好的人工智能解决方案。该项目还将研究在工厂车间执行所得算法所需的边缘计算硬件。该团队将与常青藤技术社区学院和垂直整合项目密切合作，该项目帮助普渡大学、哈佛大学和塔斯基吉大学的学生从事行业定义的制造业人工智能项目。该项目的目标将通过实现以下四个目标来实现：(1)共同优化用于制造的微型机器学习（TinyML）硬件和软件；(2)为鼓励和激励知识共享的ai commons设计隐私和保密政策；(3)对部分基础制造工艺进行数据聚合和预测技术成本建模论证；(4)在制造业课程中引入人工智能，并与劳动力发展相结合。一个关键的重点是人工智能系统的部署，以获得必要的数据，以优化ML算法，用于制药、食品加工、作业车间加工和混合复合材料等常见制造过程。项目团队与印第安纳州沃巴什中心地带创新网络（WHIN）地区的小型和大型制造商建立了牢固的合作伙伴关系。TinyML设备将部署在整个地区，在那里，数据和人工智能的共享可以显著改善运营。本项目由土木工程部联合资助。机械和制造创新、计算机和网络系统司、工程教育和中心司以及社会、行为和经济科学理事会的多学科活动办公室。该奖项反映了美国国家科学基金会的法定使命，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。

Current manufacturing systems based on a global supply chain work well, producing high-volume products at low cost. However, they are fragile to perturbations (e.g., COVID-19) and even processes that are essential to success, such as new product development and manufacturing ramp-up and optimization, can be prohibitively lengthy and expensive. Artificial Intelligence (AI) has revolutionized many industries and can be an important tool to improve the productivity and agility of manufacturing, but progress has been slow in manufacturing. This Future Manufacturing Research Grant (FMRG) Manufacturing USA: CyberManufacturing project is a fundamental reimagination of distributed AI/ML techniques to transform the future of manufacturing by establishing an AI-Commons that bridges multiple sites and companies using secure, distributed machine learning (ML), incentivized information sharing, and continuous quality improvement and training. This addresses a critical shortcoming of the current approach to AI in manufacturing; the limitation of training data to in-company data. Since AI algorithms increase in power with more data, secure data sharing and aggregation has the potential to provide vastly better AI solutions to all manufacturers. The project will also research the edge computation hardware needed to execute the resulting algorithms on the factory floor. The team will work closely with Ivy Tech Community College and the Vertically Integrated Projects program, which helps students from Purdue, Harvard, and Tuskegee University work on industry-defined manufacturing AI projects. This project’s goal will be achieved by fulfilling the following four objectives: (1) Co-optimization of Tiny Machine Learning (TinyML) hardware and software for manufacturing; (2) Design of privacy and confidentiality policies for an AI-Commons that encourages and incentivizes knowledge sharing; (3) Demonstration of data aggregation and predictive technical cost modeling for some foundational manufacturing processes; and (4) Introduction of AI in manufacturing curricula and integration with workforce development. A key focus is on AI system deployment to obtain necessary data for optimization of ML algorithms for common manufacturing processes in pharmaceutical, food processing, job shop machining, and hybrid composite materials. The project team has developed a strong partnership with small and large manufacturers in the Wabash Heartland Innovation Network (WHIN) region in Indiana. TinyML devices will be deployed throughout this region, where the sharing of data and AI could improve operations significantly.This project is jointly funded by the Division of Civil. Mechanical and Manufacturing Innovation, the Division of Computer and Network Systems, the Division of Engineering Education and Centers and the Office of Multidisciplinary Activities of the Directorate for Social, Behavioral and Economic Sciences.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.