Meta-learning-based surrogate modeling and system digital twin for life-cycle management of infrastructure systems
September 27 @ 10:00 am - 11:00 am
Presentation
Meta-learning-based surrogate modeling and system digital twin for life-cycle management of infrastructure systems
Life-cycle management (LCM) is an effective tool to balance structural safety and cost. With recent advances in computational power and artificial intelligence, applications of LCM can be extended from single structures and assets to complex networks and systems. This presentation will introduce two computational methods to enable this transformation. First, meta-learning-based surrogate modeling is a framework to transfer knowledge from existing modeling tasks to new and similar tasks. This framework is suitable for efficient reliability analysis and optimization at a community level. It has been applied to climate change adaptation and fleet management. Second, performance-oriented system digital twin (SDT) is a user-centric method to establish custom SDTs based on user’s modeling needs and available data sources. Based on Bayesian network, the SDT can fuse the data and the complex model of the system. It has been applied to data-driven life-cycle risk assessment of bridge networks in Miami-Dade County. In the future, with further development of these two computational methods, LCM can help us move towards a smart, sustainable, and resilient community.
Speaker
Minghui Cheng, PhD
University of Miami
Dr. Minghui Cheng is an assistant professor jointly appointed by the Department of Civil & Architectural Engineering and School of Architecture at the University of Miami. Previously, he was a postdoctoral associate at Systems Engineering, Cornell University. He obtained his Ph.D. in Structural Engineering from Lehigh University in 2021 and his B.E. in Civil Engineering at Hunan University, China, in 2016. His research focuses on the system digital twin of infrastructure systems and sustainability-informed life-cycle management with an emphasis on surrogate modeling, optimization, and machine learning.