Investigating the Effectiveness of Machine Learning Paradigms for Supporting Engineering Designers in Rapidly Evolving Digital Manufacturing

研究机器学习范式在快速发展的数字化制造中支持工程设计师的有效性

• 批准号: 2309250
• 负责人: Christopher McComb
• 金额: $ 42.47万
• 依托单位: Carnegie-Mellon University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-10-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2309250&HistoricalAwards=false
• 关键词: Investigating; Effectiveness; Machine; Learning; Paradigms

Manufacturing technology is advancing at an unprecedented rate as it become increasingly digital. Taking advantage of new manufacturing technology is critical for maintaining national security and prosperity, but even dedicated experts and leading-edge companies struggle to keep pace with manufacturing's rapid advancement. This makes it difficult for engineers to learn about the latest manufacturing technology and design products that take full advantage of new fabrication processes as they become available. Fortunately, many new manufacturing technologies rely on digital models that produce abundant design data to which machine learning can be applied to derive design knowledge. While design and manufacturing datasets may be useful for that reason, they are also highly variable in both number and quality of solutions. This work investigates how the size and quality of these datasets relate to the accuracy and usefulness of machine learning insights, and how this impacts the support provided to engineering designers. In this work, we focus on additive manufacturing (also called "3D printing") as a representative digital manufacturing technology that is rapidly evolving and growing, and which is projected to contribute substantially to the nation?s future manufacturing portfolio. Studies conducted with engineering students as part of this work will be used to provide skill training as well as collect data, helping prepare them for the manufacturing workforce.The research will combine machine learning, additive manufacturing, and explainable artificial intelligence to evaluate the use of automated design feedback derived from existing crowdsourced additive manufacturing design challenges. First, part designs will be mined from open, online repositories as well as through curated repositories established in this work via in-class design challenges. Next, a machine learning pipeline will be implemented to extract design patterns from curated digital repositories. This will make it possible to test the effect of repository size on the accuracy of design feedback and of repository size on the granularity of feedback. Finally, a user validation study will be conducted in which students will undertake a design task specific to additive manufacturing technology. Feedback with varying characteristics will be provided to some participants by extending the machine learning pipeline developed previously with explainable capabilities. Specific technical deliverables will include (1) a novel dataset of voxelized part designs, (2) a deeper understanding of the impact of repository size and quality on usefulness of machine-generated feedback, and (3) empirical evidence of the impact of real-time additive manufacturing feedback on the solutions generated by engineering designers.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.

制造业技术正以前所未有的速度发展，因为它变得越来越数字化。利用新的制造技术对维护国家安全和繁荣至关重要，但即使是敬业的专家和尖端公司也难以跟上制造业的快速发展。这使得工程师很难了解最新的制造技术和设计产品，这些产品充分利用了新的制造工艺。幸运的是，许多新的制造技术依赖于产生大量设计数据的数字模型，机器学习可以应用于这些数据来获取设计知识。尽管设计和制造数据集可能因此而有用，但它们在解决方案的数量和质量方面也存在很大差异。这项工作调查了这些数据集的大小和质量如何与机器学习见解的准确性和有用性相关，以及这如何影响提供给工程设计人员的支持。在这项工作中，我们关注的是加法制造(又称3D打印)作为一种具有代表性的数字制造技术，它正在迅速发展和发展，并预计将对国家做出重大贡献？S未来的制造组合。作为这项工作的一部分，与工程专业学生进行的研究将用于提供技能培训以及收集数据，帮助他们为制造劳动力做好准备。该研究将结合机器学习、加法制造和可解释人工智能，以评估从现有众包加法制造设计挑战中获得的自动化设计反馈的使用情况。首先，部件设计将从开放的在线存储库以及通过本工作中通过课堂设计挑战建立的经过管理的存储库来挖掘。接下来，将实施一条机器学习管道，以从经过管理的数字存储库中提取设计模式。这将使测试存储库大小对设计反馈准确性的影响以及存储库大小对反馈粒度的影响成为可能。最后，将进行用户验证研究，在该研究中，学生将承担特定于添加剂制造技术的设计任务。通过扩展以前开发的具有可解释能力的机器学习管道，将向一些参与者提供具有不同特征的反馈。具体的技术成果将包括(1)新的体素部件设计数据集，(2)更深入地了解存储库大小和质量对机器生成反馈的有用性的影响，以及(3)实时加法制造反馈对工程设计者生成的解决方案的影响的经验证据。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。

项目成果

A Data-Driven Approach for Multi-Lattice Transitions

多晶格转变的数据驱动方法

• 发表时间: 2022
• 期刊: 2022 International Solid Freeform Fabrication Symposium
• 作者: Baldwin, Martha;Meisel, Nicholas A.;McComb, Christopher
• 通讯作者: McComb, Christopher

Capturing Local Temperature Evolution During Additive Manufacturing Through Fourier Neural Operators

通过傅里叶神经算子捕获增材制造过程中的局部温度演变

• DOI: 10.1115/detc2023-117055
• 发表时间: 2023
• 期刊: ASME 2023 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference
• 作者: Chen, Jiangce;Xu, Wenzhuo;Baldwin, Martha;Nijhuis, Björn;Boogaard, Ton van;Gutiérrez, Noelia Grande;Narra, Sneha Prabha;McComb, Christopher
• 通讯作者: McComb, Christopher

Accelerating Thermal Simulations in Additive Manufacturing by Training Physics-Informed Neural Networks With Randomly Synthesized Data

通过使用随机合成数据训练物理信息神经网络来加速增材制造中的热模拟

• DOI: 10.1115/1.4062852
• 发表时间: 2024
• 期刊: Journal of Computing and Information Science in Engineering
• 影响因子: 3.1
• 作者: Chen, Jiangce;Pierce, Justin;Williams, Glen;Simpson, Timothy W.;Meisel, Nicholas;Prabha Narra, Sneha;McComb, Christopher
• 通讯作者: McComb, Christopher

AddLat2D the 2D Lattice Generator

• DOI: 10.1016/j.simpa.2023.100567
• 发表时间: 2023-08
• 期刊: Softw. Impacts
• 作者: M. Baldwin;N. Meisel;Christopher McComb

Hybrid Geometry/Property Autoencoders for Multi-Lattice Transitions

用于多晶格转变的混合几何/属性自动编码器

• 发表时间: 2023
• 期刊: 2023 International Solid Freeform Fabrication Symposium
• 作者: Baldwin, M;Meisel, N. A.;McComb, C.
• 通讯作者: McComb, C.