FMSG: Cyber: Federated Deep Learning for Future Ubiquitous Distributed Additive Manufacturing

FMSG：网络：面向未来无处不在的分布式增材制造的联合深度学习

• 批准号: 2134689
• 负责人: Jia Liu
• 金额: $ 49.88万
• 依托单位: Auburn University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-09-01 至 2024-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2134689&HistoricalAwards=false
• 关键词: FMSG; Cyber; Federated; Deep; Learning

Distributed additive manufacturing has promising potential to connect and coordinate individual manufacturers for efficient, on-demand production. It can leverage the freeform fabrication of numerous additive manufacturers to form a flexible and robust supply chain and achieve reconfigurable mass customization in the future. However, product quality, consistency and privacy concerns among those distributed manufacturers pose a grand challenge to fully unleashing the potential of distributed additive manufacturing. This Future Manufacturing Seed Grant (FMSG) CyberManufacturing project will support fundamental research to provide needed knowledge for developing a unified algorithmic and training framework. The new framework, named FEDMDL, will lay a solid foundation to enable consistent and reliable production in a privacy-preserving, insight-sharing manufacturing network. This will further promote the adoption of additively manufactured parts in various industries, such as aerospace, automobile, healthcare, and will boost the participation of small-and-medium-sized manufacturers in the national supply chain. Therefore, results from this research will benefit the competitive advantages of US manufacturing and economy. This research provides manufacturing companies with the synergy of novel machine learning and federated computing techniques. The multi-disciplinary approach will help broaden the participation of underrepresented groups in research and positively impact engineering education. The unified algorithmic and training framework, FEDMDL, will chart a new theoretical path to enabling reliable production, consistent quality, and privacy-preserving data sharing in distributed additive manufacturing. FEDMDL will synthesize the fundamental physics of additive manufacturing processes into deep learning algorithms and train the new models on a federated learning cyberinfrastructure. In this seed grant, FEDMDL will be prototyped with fatigue performance assessment of additively manufactured metals in a distributed manufacturing network. The research team will: (1) conduct fatigue testing and defect characterization to understand material-defect-geometry-loading-fatigue relationships; (2) develop fracture-mechanics-centric deep learning models to approximate multi-physics multiscale processes and predict the fatigue performance of complex geometries under multiaxial loading; (3) design a cross-silo, additive-manufacturing-aware federated learning cyberinfrastructure to train the deep learning models with collective insights from the sparse, siloed datasets across manufacturers; and (4) evaluate the framework by deploying it in a real-world distributed additive manufacturing network. This work will result in an experimentally validated, generalizable algorithmic and training framework to catalyze research and applications in quality modeling, qualification, and control for future distributed additive manufacturing with collective intelligence.This project is jointly funded by the Division of Civil. Mechanical and Manufacturing Innovation, the Established Program to Stimulate Competitive Research (EPSCoR), and the Division of Electrical, Communications, and Cyber Systems.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.

分布式增材制造在连接和协调单个制造商以实现高效的按需生产方面具有很大的潜力。它可以利用众多增材制造商的自由制造，形成灵活而稳健的供应链，并在未来实现可重构的大规模定制。然而，在这些分布式制造商中，产品质量、一致性和隐私问题对充分释放分布式增材制造的潜力构成了巨大挑战。该未来制造种子基金（FMSG）网络制造项目将支持基础研究，为开发统一的算法和培训框架提供所需的知识。名为FEDMDL的新框架将为在隐私保护、见解共享的制造网络中实现一致和可靠的生产奠定坚实的基础。这将进一步促进增材制造零部件在航空航天、汽车、医疗等各行业的应用，并将促进中小制造商参与国家供应链。因此，本研究的结果将有利于美国制造业和经济的竞争优势。这项研究为制造公司提供了新型机器学习和联邦计算技术的协同作用。多学科方法将有助于扩大代表性不足的群体在研究中的参与，并对工程教育产生积极影响。统一的算法和培训框架FEDMDL将为分布式增材制造中实现可靠的生产、一致的质量和保护隐私的数据共享开辟一条新的理论路径。FEDMDL将把增材制造过程的基本物理原理综合到深度学习算法中，并在联邦学习网络基础设施上训练新模型。在这笔种子基金中，FEDMDL将在分布式制造网络中进行增材制造金属的疲劳性能评估原型。研究团队将：(1)进行疲劳测试和缺陷表征，了解材料-缺陷-几何-载荷-疲劳关系；(2)建立以断裂力学为中心的深度学习模型，以近似多物理场多尺度过程，预测复杂几何形状在多轴载荷下的疲劳性能；(3)设计一个跨竖井、感知增材制造的联邦学习网络基础设施，利用来自各制造商稀疏、竖井数据集的集体见解来训练深度学习模型；(4)通过在真实世界的分布式增材制造网络中部署该框架来评估该框架。这项工作将产生一个经过实验验证的、可推广的算法和培训框架，以催化在质量建模、鉴定和控制方面的研究和应用，为未来的分布式增材制造提供集体智能。本项目由土木工程部联合资助。机械和制造创新，刺激竞争研究的既定计划（EPSCoR），以及电气，通信和网络系统部门。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。

项目成果

SFS: Smart OS Scheduling for Serverless Functions

• DOI: 10.1109/sc41404.2022.00047
• 发表时间: 2022-09
• 期刊: SC22: International Conference for High Performance Computing, Networking, Storage and Analysis
• 作者: Yuqi Fu;Li Liu;Haoliang Wang;Yue Cheng;Songqing Chen

Defect Criticality Analysis on Fatigue Life of L-PBF 17-4 PH Stainless Steel via Machine Learning

• DOI: 10.1016/j.ijfatigue.2022.107018
• 发表时间: 2022-05
• 期刊: International Journal of Fatigue
• 影响因子: 6
• 作者: Anyi Li;Shaharyar Baig;Jia Liu;Shuai Shao;N. Shamsaei

FedAT: A High-Performance and Communication-Efficient Federated Learning System with Asynchronous Tiers

• DOI: 10.1145/3458817.3476211
• 发表时间: 2020-10
• 期刊: SC21: International Conference for High Performance Computing, Networking, Storage and Analysis
• 作者: Zheng Chai;Yujing Chen;Ali Anwar;Liang Zhao;Yue Cheng;H. Rangwala

Defects Classification via Hierarchical Graph Convolutional Network in L-PBF Additive Manufacturing

L-PBF 增材制造中通过分层图卷积网络进行缺陷分类

• 发表时间: 2022
• 期刊: 2022 International Solid Freeform Fabrication Symposium
• 作者: Li, Anyi;Liu, Jia;Shao, Shuai;Shamsaei, Nima
• 通讯作者: Shamsaei, Nima