Elements: Curating and Disseminating Solid Mechanics Based Benchmark Datasets

从汽车安全到外科手术计划，预测材料在载荷作用下的力学行为是工程设计中必不可少的一步。然而，目前预测机械行为的方法通常非常缓慢，单个模拟可能需要数小时甚至数月才能运行。由于这一限制，最近的研究利用机器学习技术来快速预测工程系统的行为，并直接从数据中创建预测模型。尽管这些新兴技术具有巨大的造福社会的潜力，但目前从机械数据中创建机器学习模型的策略仍然是临时的。这个研究项目的目标是通过向更广泛的社区传播大型基于力学的数据集和相关的科学资源来改变这种状况。此外，该项目为更广泛的力学社区内的文化转变提供了资源，转向数据和代码传播实践，这将使力学研究界能够更有效地取得集体研究进展。这项工作的目标是创建和传播开源基准数据集，重点关注两大类机械挑战：非线性材料行为（例如工程复合材料的断裂）和非线性结构行为（例如工程超材料的大变形行为和屈曲）。结合数据集传播，本工作定义并传播了围绕新兴和未解决领域的核心挑战问题，这些领域的方法创新将显著推动固体力学领域向前发展，例如：(1)非线性力学的全场预测，(2)力学预测的复杂几何表示，(3)力学特定的分布泛化，以及(4)力学问题的不确定性量化。最后，该项目包括一个“开放获取力学数据集网站和教育资源”组件，用于整理来自更广泛的研究界的开放获取力学数据集。该奖项由美国国家科学基金会高级网络基础设施办公室颁发，并由美国国家科学基金会民用、机械和制造创新部门联合支持。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。

From automobile safety to surgical planning, predicting the mechanical behavior of materials under loading is an essential step in engineering design. However, current approaches to predicting mechanical behavior are often very slow, where single simulations can take hours or even months to run. Motivated by this limitation, recent research has leveraged machine learning techniques to rapidly predict the behavior of engineered systems and create predictive models directly from data. Though these emerging techniques have enormous potential to benefit society, current strategies for creating machine learning models from mechanical data remain ad hoc. The goal of this research project is to change this through disseminating large mechanics-based datasets and associated scientific resources to the broader community. In addition, this project provides resources for a cultural shift within the broader mechanics community towards data and code dissemination practices that will allow the mechanics research community to make collective research progress more efficiently. The objective of this work is to create and disseminate open source benchmark datasets that focus on two broad classes of mechanical challenge: nonlinear material behavior (e.g., fracture of engineered composites), and nonlinear structural behavior (e.g., large deformation behavior and buckling of engineered metamaterials). In conjunction with dataset dissemination, this work defines and disseminates core challenge problems around emerging and unresolved areas where methodological innovation would significantly move the solid mechanics field forward such as: (1) full-field prediction for nonlinear mechanics, (2) representation of complex geometries for mechanical prediction, (3) mechanics-specific out of distribution generalization, and (4) uncertainty quantification in mechanical problems. Finally, this project includes an “Open Access Mechanics Dataset Website and Educational Resource” component to collate open access mechanics datasets from the broader research community.This award by the NSF Office of Advanced Cyberinfrastructure is jointly supported by the NSF Division of Civil, Mechanical and Manufacturing Innovation.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.