CRII:OAC:A Data-Driven Closed-Loop Platform for Optimal Design of Deployable Pin-Jointed Structures

可展开销连接(DPJ)结构由于其重量轻、可折叠、高刚度等特点，在航空航天机械工程、土木工程、机器人和生物医学材料等领域表现出了极大的研究和开发兴趣。DPJ结构由受压构件(通常为杆件或压杆)和张拉构件(通常为拉索或钢索)组成，通过销钉连接。在运行中，这些DPJ结构有各种性能要求，如保持高水平的表面精度，以及实现所需的刚度和可调的固有频率。期望性能的损失通常会导致DPJ结构的故障甚至崩溃。然而，由于与本构模型相关的问题和缺乏设计工具，很难获得具有期望性能的DPJ结构的优化设计。这个项目通过创建一个数据驱动的闭环平台来解决这些问题，以优化DPJ结构的设计。将设计和建造一系列数据驱动的工具，以创建所建议的闭环系统平台：(1)将创建一种确定DJP结构初始平衡构型的新的随机方法；(2)将开发一种基于机器学习和先进无损检测的DPJ结构的新的计算建模技术。该项目将满足结构工程方面的迫切需求，并提供对DPJ结构的设计和计算建模的更深层次的了解。该项目的成果可以帮助提高建筑、航天器、军事装备和高科技设备等领域中一类结构的性能、安全性和寿命。该项目将通过开发一系列数据驱动工具来促进数值和高性能科学计算，并扩大实体和结构力学的建模和模拟能力，以补充建立下一代先进网络基础设施生态系统的努力。该项目还将有助于升级关于结构计算建模的课程。在这个项目中，工程专业的学生将被招募和指导。对学生的培训将包括结构设计、计算建模、算法开发和实验测试。传统的结构设计是开环协议，遵循设计-建模-验证程序。该项目的主要目标是创建一个数据驱动的闭环平台，用于DPJ结构的优化设计。这个框架不变的平台在提供闭环结构设计协议方面是新的。在该平台中，实验结果不仅用于模型验证，还将提供训练和测试数据，以进一步提高数据驱动计算模型的性能。然后通过使用计算模型来指导初始结构设计来闭合环路。为实现这一目标，将开展两项工作。第一个任务是开发一种随机方法来寻找形式。传统的DPJ结构找形方法需要对构件进行分组，这高度依赖于结构的几何简单性。为了解决这一问题，将设计和研究一种新的方法，称为随机固定节点位置法。这种方法的主要优点是它不使用成员分组，也不需要任何几何简单性。这些特点将使该方法成为设计大规模、复杂和不规则的DPJ结构的有力工具。第二个任务是开发一种数据驱动的计算建模技术。本构模型技术经常过度简化DPJ结构，导致不能反映结构的重要力学特性。由于DPJ结构在其他固体或结构中不常见的特殊特性，最近开发的计算建模技术很少适用于DPJ结构。该项目将开发一种新的基于机器学习和无损检测的DPJ结构计算建模技术。这项技术将绕过传统的本构模型，并在处理DPJ结构方面提供良好的性能。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。

Deployable pin-jointed (DPJ) structures, due to their being lightweight, foldable, and having high stiffness, have shown high research and development interest in state-of-the-art applications in many fields, such as aerospace and mechanical engineering, civil engineering, robotics and bio-medical materials. A DPJ structure is composed of members in compression (usually bars or struts) and tension (usually cables or tendons), connected by pin joints. In operation, these DPJ structures have various performance requirements, such as maintaining a high level of surface accuracy, and achieving desired stiffness and tunable natural frequencies. Loss of desired performance often yields malfunction or even breakdown of a DPJ structure. However, optimal design of DPJ structures with the desired performance is hard to obtain, due to issues related to constitutive modeling and lack of design tools. This project addresses these problems by creating a data-driven closed-loop platform for optimal design of DPJ structures. A series of data-driven tools, that create the proposed closed-loop platform, will be designed and built: (1) a novel stochastic method for determining an initial equilibrium configuration of a DJP structure will be created; and, (2) a new computational modeling technique for DPJ structures, based on machine learning and advanced nondestructive testing, will be developed. The project will address an urgent need in structural engineering, and provide a deeper understanding of the design and computational modeling of DPJ structures. The results obtained from this project can help enhance the performance, safety and longevity of a class of structures in various areas, including architectures, spacecraft, military equipment and high-tech devices. The project will complement efforts to build the next-generation advanced cyberinfrastructure ecosystem by developing a series of data-driven tools to facilitate numerical and high-performance scientific computing, and expand modeling and simulation capabilities for mechanics of solid and structures. The project will also help upgrade the curriculum on computational modeling of structures. Engineering students will be recruited and mentored in this project. The training for students will include structural design, computational modeling, algorithm development and experimental testing.Traditional structural design is an open-loop protocol, in which a design-modeling-validation procedure is followed. The main objective of this project is to create a data-driven closed-loop platform for optimal design of DPJ structures. This frame-invariant platform is new in providing a closed-loop structural design protocol. In this platform, experimental results will not only be used for model validation, but also in turn serve to provide training and testing data to further improve performance of a data-driven computational model. The loop will then be closed by using the computational model to guide initial structural design. Toward this goal, two tasks will be carried out. The first task is to develop a stochastic approach to form finding. Traditional methods for form finding of DPJ structures require member grouping, which relies highly on the geometric simplicity of the structure. To resolve this issue, a new method, called the stochastic fixed nodal position method, will be designed and investigated. The key benefit of this method is that it does not use member grouping or require any geometric simplicity. These features will allow the method to serve as a powerful tool in design of large-scale, complex, and irregular DPJ structures. The second task is to develop a data-driven computational modeling technique. Constitutive modeling techniques often over-simplify a DPJ structure, which results in the failure to reflect important mechanical properties of the structure. Very few recently developed techniques for computational modeling are suitable to DPJ structures, due to their special characteristics that are not commonly seen in other solids or structures. This project will develop a novel computational modeling technique based on machine learning and non-destructive testing for DPJ structures. This technique will bypass traditional constitutive modeling and provide good performance in handling DPJ structures.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.