CAREER: Leveraging Existing Knowledge and Artificial Intelligence to Understand the Performance of Civil Infrastructure Under Extreme Hazard Loads

该学院早期职业发展（CAREER）补助金将支持研究，以了解极端负荷下的民用基础设施的物理性能，如地震和风暴，以及材料，结构，系统和社区需求之间的相互作用在这种负荷下。关于常规材料和结构在正常运行和各种危险载荷条件下的性能的现有知识已经积累了多年;由于改进的仪器、实验和观察，越来越多地捕获了关于地震和风暴载荷下性能的数据。新材料和结构设计方面的创新正在被创造出来，以应对这些极端载荷。为了跟上这些创新的步伐，同时继续提供坚固和弹性的结构，需要一种快速可靠的方法来了解新材料和结构设计在这些更极端载荷下的行为。 人工智能（AI）在土木工程领域的融合提供了学习材料，结构和负载特性与结构性能或社区响应之间高度非线性，复杂关系的能力。这项研究将利用人工智能的力量和现有丰富的基于物理的性能数据，转移有关传统的、经过充分研究的结构部件和加载机制的知识，为样本外的情况和几乎没有数据的创新做出性能预测。这可以减少对实验测试和计算费用昂贵的分析评估的依赖，并减轻自然灾害对社区的灾难性影响。同时，该奖项将支持多学科教育和推广计划，STEM in Motion，该计划将专注于为本科和研究生土木工程课程开发技术前沿和积极的学习活动。本项目产生的数据将在自然灾害工程研究基础设施（NHERI）数据库（https：/www.DesignSafe-c.org）中存档并公开提供。该奖项有助于国家科学基金会在国家减少地震灾害计划（NEHRP）中的作用。这项研究将利用现有的实验数据，准确，稳健，快速地预测普通结构的抗震性能，以及特别容易受到危险载荷的结构。借助高度精确的人工智能方法，在已知的材料和载荷约束（例如，在当前考虑的地震载荷下的钢筋混凝土建筑物），在大数据可用的情况下，这些模型中的固有知识可以跨域并且以不同的尺度稳健地转换（例如，材料、部件、系统和载荷），以减少不确定性，并增强对新结构物理行为的近实时理解。由于能够直接从现有数据集中导出这些关系，因此有机会进行创新的材料和结构建模程序，并进行独特的设计，以满足特定的性能要求。具体而言，本研究将调查以下内容：（1）建立数据集之间的相关性所需的数据集特征或特性，（2）算法开发所需的数学构造和基于物理的约束，以减少迁移学习问题中样本偏差的影响，（3）当模型本身不容易解释时，AI模型的验证和确认，以及（4）多学科、以团队为中心的环境对工程专业学生保持基本工程原理和学习新概念的影响。该奖项反映了NSF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。

This Faculty Early Career Development (CAREER) grant will support research to understand the physical performance of civil infrastructure under extreme loads, such as earthquakes and windstorms, and the interactions among materials, structures, systems, and community needs under such loading. Existing knowledge regarding the performance of conventional materials and structures under normal operating and various hazardous loading conditions has been amassed over years; and data on the performance under earthquake and windstorm loading increasingly are being captured as a result of improved instrumentation, experimentation, and observations. Innovations in new materials and structural design are being created to respond to these extreme loads. To maintain pace with these innovations, while continuing to provide robust and resilient structures, there is a need for a rapid and reliable approach to understanding the behavior of new materials and structural designs under these more extreme loads. The convergence of artificial intelligence (AI) into the civil engineering domain provides the capability to learn the highly nonlinear, complex relationships between material, structural, and load characteristics and a structure’s performance or community’s response. This research will leverage the power of AI and the existing wealth of physics-based performance data to transfer knowledge concerning conventional, well-studied, structural components and loading mechanisms to make performance predictions for out-of-sample cases and innovations where little data is available. This can reduce the reliance on experimental testing and computationally expensive analytical evaluations, and mitigate the catastrophic effects of natural disasters on communities. In tandem, this award will support a multidisciplinary educational and outreach plan, STEM in Motion, which will focus on the development of technologically-forward and active learning-focused activities for undergraduate and graduate civil engineering courses. Data generated from this project will be archived and made publicly available in the Natural Hazards Engineering Research Infrastructure (NHERI) Data Depot (https:/www.DesignSafe-c.org). This award contributes to the National Science Foundation's role in the National Earthquake Hazards Reduction Program (NEHRP). This research will employ available experimental data to accurately, robustly, and quickly predict the seismic performance of common structures as well as structures which are exceptionally susceptible to hazardous loads. With a highly accurate AI approach to modeling the behavior of existing structures under well-known material and loading constraints (e.g., reinforced concrete buildings under currently considered seismic loads) where big data is available, the inherent knowledge in these models can be robustly translated across domains and at varying scales (e.g., material, component, system, and load) to reduce uncertainty and lead to enhanced, near-real-time understanding of the physical behavior of new structures. With the ability to derive these relationships directly from existing datasets, the opportunity arises for innovative material and structural modeling procedures and designs uniquely suited for specific performance requirements. Specifically, this research will investigate the following: (1) the dataset features or characteristics necessary to establish relevance between datasets, (2) the mathematical construction and physics-based constraints needed for the algorithmic development to reduce the impact of sample bias in transfer learning problems, (3) verification and validation of the AI model when the model itself is not easily interpretable, and (4) the impact of a multidisciplinary, team-focused environment on the retention of fundamental engineering principles and learning of new concepts by engineering students.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.