题 目：Artificially Intelligent Systems for Structural Health Monitoring
（用于结构健康监测的人工智能系统）
时 间：2021年06月18日10：00－11：00
地 点：腾讯会议ID： 363 237 883
会议直播：https://meeting.tencent.com/l/Bk5qlvGKN1st
报告人：梁霄 助理教授（美国布法罗大学）
欢迎广大师生参加
土木与交通学院
2021年06月13日
报告人简介：
梁霄博士目前是美国纽约布法罗大学土木、结构与环境工程系助理教授。他于2010年在湖南大学获得土木工程学士学位后，分别于2011年和2016年在美国加州大学伯克利分校获得土木工程学硕士和博士学位。他的研究侧重于通过高级数据分析、基于模型和机器学习进行健康监测和自主检查（例如使用无人机）。他特别感兴趣的方向是如何将这个领域的发展用于基础设施系统和建筑物中，以便能够实时地评估和量化结构的状况，从而增强其在服务期和极端事件下的可持续性和复原力。他的研究兴趣还包括基于性能的抗危险性方法、非线性结构动力学、地震工程以及未来再制造的人机协作。
Dr. Xiao Liang is currently an Assistant Professor in the Department of Civil, Structural & Environmental Engineering at the University at Buffalo. He obtained his Ph.D. (2016) and M.S. (2011) degrees in Civil Engineering both from the University of California, Berkeley, after he completed his B.S. degree in Civil Engineering at Hunan University in 2010. His research focuses on health monitoring and autonomous inspection (e.g., using drones) through advanced data analytics, model-based and machine learning. He is particularly interested in such developments for infrastructure systems and buildings to enable assessing and quantifying the condition of structures in near real-time, aiming to enhance their sustainability and resilience under service and extreme events. His research interests also include performance-based methodologies for hazard resilience, nonlinear structural dynamics, earthquake engineering, and human-robot collaboration for future remanufacturing.
报告摘要：
民用基础设施的快速状况评估是极端事件后恢复的一个组成部分。通过考虑人工检查的时间成本限制、可靠性和生命安全问题等几个因素，结构健康监测(SHM) 中的自动化越来越受到激励。过去几年人工智能 (AI) 研究取得了令人瞩目的进展，使 SHM 中的自动化概念更接近现实。
本次演讲将简要介绍自主性在快速状态评估中的重要性。演讲由三个主要部分组成。第一部分将从不同的角度重点介绍SHM 中的可视检测。演讲的第二部分将讨论基于振动的SHM，并对基于性能的地震工程概念进行展望。随后，演讲将介绍semantic damage segmentation的概念作为大规模SHM的潜在解决方案的可能性。演讲的最后一部分将专门讨论AI辅助SHM 的风险和可靠性。基于贝叶斯深度学习的方法将被用于前两部分讨论的视觉和基于振动的模型不确定性量化。除了量化之外，还将通过不同的展示探索不确定性的潜在好处。
Rapid condition assessment of civil infrastructure is an integral part of recovery after extreme events. By taking into account several factors such as time-cost constraints, reliability, and life-safety concerns of human-based inspections, there has been a growing incentive for automation in structural health monitoring (SHM). The impressive progress of artificial intelligence (AI) research in the past few years has made the concept of automation in SHM closer to reality. In this presentation, a brief introduction is given regarding the importance of autonomy in rapid condition-assessments. The presentation is comprised of three main sections. The first part will focus on visual inspections in SHM from different perspectives. The second part of the talk will be on vibration-based SHM with an outlook to the concepts of performance-based earthquake engineering. Later, the concept of semantic damage segmentation (SDS) is explained as a potential solution for large scale SHM. The last part of the presentation will be dedicated to the risk and reliability of AI-assisted SHM. Bayesian Deep learning is proposed for model uncertainty quantification in both vision and vibration-based applications discussed in the first two sections. Beyond quantification, the potential benefits of uncertainty are explored with different showcases.
