School of Civil and Environmental Engineering
Ph.D. Thesis Defense Announcement
Towards the Next Generation of Data-driven and Physics-informed Procedures for the Seismic Performance Assessment of Geosystems
By Chenying Liu
Advisor:
Dr. Jorge Macedo
Committee Members: Dr. David Frost (CEE), Dr. Zhigang Peng (EAS), Dr. Mahdi Roozbahani (CSE), Dr. Norman Abrahamson (CEE, UC Berkeley), Dr. Jack Baker (CEE, Stanford)
Date and Time: April 11, 2025. 9:00AM - 12:00 PM
Location: SEB122 / https://gatech.zoom.us/j/4205093739
ABSTRACT
Performance-based engineering (PBE) is widely recognized as the leading framework for the rigorous assessment of engineering systems subjected to extreme events. Over the past 60 years, earthquake engineering has evolved from deterministic approaches—which evaluate only a limited number of assumed-to-be-conservative scenarios—toward performance-based methods that explicitly account for uncertainty. While many recent advances in earthquake engineering have been driven by PBE, important challenges remain.
Emerging data-driven methods present a unique opportunity to advance performance-based earthquake engineering, particularly for geotechnical systems. However, their successful application requires the integration of physical principles, as design scenarios frequently involve extreme events—such as the “Maximum Credible Earthquake” specified in seismic codes—where extrapolation beyond available data is often necessary.
This study contributes to the development of next-generation data-driven and physics-informed approaches in geotechnical earthquake engineering. It introduces new conditional ground motion models that enable hazard-consistent estimation of ground motion intensity measures, aligning more closely with performance-based assessment frameworks. In addition, it proposes new nonergodic ground motion models that capture repeatable effects of earthquake sources, recording stations, and wave propagation paths, leading to improved uncertainty quantification for regions like California and Turkey. The work also investigates the implications of the nonergodic paradigm—traditionally applied in seismic demand estimation—for broader seismic risk assessment, including regional liquefaction hazard and the resilience of distributed energy resource systems in specific areas of California.
Further, the study develops machine learning– and deep learning–based semi-empirical models to evaluate the seismic performance of slope systems in subduction zones and buildings in urban areas underlain by liquefiable soils. These models demonstrate more robust median predictions and reduced uncertainty compared to traditional approaches. The study also explores the potential of physics-informed neural networks as efficient surrogates for traditional numerical methods, which, despite decades of development, still face practical and computational limitations. The study concludes with a perspective on the future of performance-based engineering in the context of emerging computational and data-driven paradigms