SimCenter Community Roundtables

Title: UQ approaches for computational efficient hazard/vulnerability assessment at the regional scale
Working Group: Uncertainty Quantification (UQ) in Natural Hazards Engineering
Host: Alexandros Taflanidis, University of Notre Dame
Date: September 23, 2024 (12:30pm Pacific; 3:30pm Eastern)
Watch on YouTube

Abstract

Advancements in computational modeling and probabilistic analysis techniques are enabling the assessment of natural hazards and their impact on the built environment (buildings and lifelines) and communities with unprecedented scale and resolution. The outcome of such analysis techniques becomes an invaluable ingredient in guiding emergency responses, assessing societal consequences, simulating the recovery phase, and optimizing disaster policy and design decisions. At the same time, such an attempt at large-scale regional risk assessment presents opportunities for researchers to tackle previously unexplored challenges. Such challenges can be associated, for example, with high-dimensional, spatiotemporally correlated hazard and system response descriptions, scarcity of data needed for the development of models for extreme hazard and corresponding system response, trade-offs between the analysis resolution and computational burden under limited resources, complex interdependency between component and system behaviors, and conflicting influences of hazards on the subcomponents of the system and resulting conflicting policy decision objectives. This setting creates new opportunities (for addressing the aforementioned challenges) for the use of advanced UQ techniques (e.g., surrogate modeling, sensitivity analysis, adaptive sampling, multi-fidelity approaches, low-dimensional latent space projections, multi-objective optimization) and for promoting interdisciplinary efforts to expand the frontiers of the domain.

This roundtable discussion will showcase ongoing UQ efforts in enabling regional-scale modeling and risk assessment, with the intention to create a dialog and foster collaborative efforts between researchers. It will combine a small number of invited presentations and open dialog on this broader topic.

Presenters: Alexandros Taflanidis (University of Notre Dame); Paolo Bocchini (Lehigh University); Jamie Padgett (Rice University); Abdollah Shafieezadeh (The Ohio State University)

Presenter: Alexandros Taflanidis, Professor, University of Notre Dame

Title: On adaptive Monte Carlo simulation strategies for regional-scale uncertainty propagation: challenges in achieving computational efficiency across competing quantities of interest

Abstract: Comprehensive regional risk assessment applications entail a significant computational burden. This presentation focuses on adaptive (intelligent) Monte Carlo (MC) simulation techniques to accommodate the uncertainty propagation in this setting, in support of estimation or regional hazard/risk. Focus is placed on the need of such techniques to establish a compromise across conflicting quantities of interest (QoIs), corresponding to the risk at different assets/geographic-locations within the region of interest. Topics addressed include: the need to find a balanced compromise across the QoIs; the implications of this compromise compared to the idealized scenario that uncertainty propagation was established separately for each QoI; techniques for promoting computational efficiency when compromise needs to be established for millions of QoIs. Discussions are couches within two specific MC-based algorithms: adaptive importance sampling and adaptive Multi-Fidelity Monte Carlo and within a specific application, the real time probabilistic surge estimation during landfalling storms.

Short Bio: Dr. Alexandros Taflanidis is Professor in the Department of Civil and Environmental Engineering and Earth Sciences at the University of Notre Dame. He received his Bachelors (2002) and Masters (2003) from Aristotle University of Thessaloniki, Greece, and his PhD i from the California Institute of Technology (2008). His research focuses on uncertainty quantification and uncertainty-conscious analysis/design, with applications to dynamical system analysis, natural hazard risk mitigation and sustainability/resilience of civil infrastructure systems. He currently leads the NSF SimCenter UQ working group. He received the 2021 ASCE Walter L. Huber Civil Engineering Research Prize for applications of machine learning and computational statistics in the domain of natural hazards engineering.

Presenter: Paolo Bocchini, Professor, Lehigh University

Title: Uncertainty quantification as a mechanism to drive multi-hazard scenario selection

Abstract: Hazard Quantization (HQ) is an effective technique for the selection/generation of extreme event scenarios for catastrophe modeling and regional resilience assessment. HQ selects scenario events that capture the natural variability of regional intensity measure maps, while preserving spatial correlation. Unfortunately, problems arise when applying the basic version of the technique to situations where the extreme event needs to be described by multiple intensity measures, such as peak ground acceleration and multiple spectral accelerations for an earthquake affecting a region with different types of building assets. The same type of challenges arises when dealing with multi-hazard events, such as a hurricane that generates both storm surge and strong winds in a coastal region. A third class of obstacles present themselves when specific locations in the region happen to experience very extreme values of the hazard intensity measure. All these issues are rooted in the fact that locations or intensity measures with large variability bias the whole procedure and render the information coming from the other locations or intensity measures useless. A novel technique that introduces uncertainty quantification in the hazard quantization loop to renormalize variances is capable of mitigating or eliminating these issues. Applications to seismic and weather-related events will be presented.

Short Bio: Dr. Bocchini is Professor and Director of Graduate Programs in the Department of Civil and Environmental Engineering of Lehigh University. His research is related to the application of probabilistic concepts, computational mechanics, operational research, and other analytical and numerical tools to civil engineering problems. Currently, his main areas of focus are catastrophe modeling, resilience assessment, and optimal allocation of resources for the design, retrofit, and recovery of infrastructure systems subjected to extreme events. Dr. Bocchini co-authored the chapter on infrastructure interdependencies in the Objective Resilience Manual of Practice of ASCE, served in a committee of the National Academies to develop guidelines for the U.S. Congress to allocate resources to projects that improve resilience, was in the leadership team of the ASCE Special Project on Effects of Climate Change on the Built Environment, and received a number of awards and recognitions. He is also the founder and Director of the Center for Catastrophe Modeling and Resilience, serves as Associate Editor for the ASCE Journal of Structural Engineering, leads large interdisciplinary teams, and was elected to the rank of Fellow of the ASCE Structural Engineering Institute.

Presenter: Jamie Padgett, Professor, Rice University

Title: : Toward high-fidelity yet efficient structural portfolio representation: The role of surrogate modeling for fragility analysis and influence of model choices on risk and resilience estimates

Abstract: In the context of uncertainty quantification in regional risk or resilience analysis, tradeoffs and decisions are often made with respect to model fidelity and computational efficiency. This presentation highlights two pervasive themes. First, we explore the role of surrogate modeling within the process of deriving fragility functions, as a means of supporting uncertainty quantification across a structural portfolio while rendering efficient yet tailored damage estimation. Second, we probe the impact of model selection and interaction effects across the risk and resilience pipeline. For example, we illustrate how data and knowledge availability regarding structural portfolios and their networked systems, as well as damage or recovery models built with different levels of fidelity, affect uncertain resilience outcomes.

Short Bio: Jamie E. Padgett is the Stanley C. Moore Professor and Department Chair of Civil and Environmental Engineering at Rice University. Padgett is a structural engineer whose research is focused on multi-hazard risk and resilience modeling of structures and infrastructure systems, while understanding their impacts on communities. She has received such honors as the Duke Lifeline Earthquake Engineering Award (2024), TAMEST Edith and Peter O’Donnell Award (2023), and the Executive Leadership in Academic Technology, Engineering and Science (ELATES) Fellowship (2021-2022). Padgett serves in leadership roles within several large national research efforts including the NIST funded Center of Excellence for Risk-based Resilience Planning and the NHERI Cyberinfrastructure “DesignSafe”.

Presenter: Abdollah Shafieezadeh, Professor, Ohio State University

Title: Enhancing System Reliability and Resilience under Natural Hazards: Active Learning and Stochastic Robust Optimization for Critical Infrastructures

Abstract: Analyzing the performance of systems, including reliability and resilience under natural hazard risks, presents significant challenges due to the multitude of uncertainties, high dimensionality, and the computational cost of simulations. This presentation first introduces an adaptive network reliability analysis with applications to power grids. By leveraging active learning techniques and surrogate modeling, this method effectively tackles the complexities of high-dimensional, computationally expensive flow-based models, with a focus on rare event analysis. Its effectiveness is demonstrated through applications across benchmark power grid systems. Next, the presentation briefly discusses decision-making for systems operating under the uncertain conditions of natural hazards. It presents a multi-stage stochastic robust resilience optimization model for smart power distribution systems that captures the interplay between various stages of system response and decision-making, while rigorously accounting for the uncertainties imparted by natural hazards and their effects.

Short Bio: Dr. Shafieezadeh is the Lichtenstein Professor of Structural and Infrastructure Engineering at Ohio State University and leads the Smart and Resilient Communities initiative at the Sustainability Institute at OSU. His expertise and interests are in multi-fidelity computational modeling and integration with data-driven methods, uncertainty quantification techniques, and decision making under uncertainty aiming to enhance climate risk and resilience assessment of civil infrastructure. Shafieezadeh received the Sam Nunn Security Fellowship on the role of technology in public policy as well as a Fulbright Scholarship. He has contributed to public and policy understanding of climate challenges, testified before U.S. Senate on the topic of climate change resilience, and has over 200 publications and presentations. He serves on the Editorial Board of Structural Safety among other journals and is the Associate Editor of ASCE Journal of Structural Engineering.

Title: Advancing Hurricane Regional Simulation: A Community Discussion of Challenges & Opportunities
Working Group: Regional Simulation of Hurricanes
Host: Tracy Kijewski-Correa, University of Notre Dame
Date: September 27, 2024 (9am Pacific; 12pm Eastern)
Watch on YouTube

Abstract

This session will convene leaders in hurricane regional simulation to share lightning talks on some of the latest developments from their initiatives and then engage in a wider discussion with attendees on key challenges and opportunities to further advance the state-of-the-art.

Objectives:

• Promote learning about the different hurricane regional simulation initiatives.
• Identify opportunities and challenges to advance the state-of-the-art.
• Foster dialog within the community of researchers interested in hurricane regional simulation.

Presenters:

David Roueche: David Roueche is an Associate Professor of Structural Engineering at Auburn University, and the Associate Director of the Structural Extreme Events Reconnaissance (StEER) network. David has extensive experience in post-disaster reconnaissance studies, and is leveraging this reconnaissance data with physics-based modeling and experimental testing to improve resilience to natural hazards.

Kooshan Amini:  Kooshan Amini is a Ph.D. student in civil and environmental engineering at Rice University. He works on IN-CORE, a NIST-funded platform for modeling the impact of natural hazards and improving community resilience. Kooshan also leads the Galveston testbed, focusing on real-world applications and analyses within this project.

Kurt Gurley: Dr. Gurley is a Professor in the Department of Civil and Coastal Engineering and the Associate Director of the UF NSF NHERI Wind Hazard Experimental Facility https://ufl.designsafe-ci.org/. Dr. Gurley’s research focus is the modeling of structural vulnerability to hurricane wind damage. He is a co-developer of the Florida Public Hurricane Loss Model.

Rachel Davidson: Rachel Davidson is a civil engineering Professor and a core faculty member in the Disaster Research Center at the University of Delaware. She is also the PI of the NSF-funded Coastal Hazards, Equity, Economic prosperity, and Resilience (CHEER) Hub. Davidson conducts research on regional-scale natural disaster risk modeling and civil infrastructure systems.

Barbaros Cetiner: Barbaros Cetiner is a software developer at the NHERI SimCenter, where he leads the development of the AI-supported BRAILS tool that allows for the automated creation of building and transportation inventories essential to regional-level hurricane simulations.

Teng Wu: Teng Wu is a civil engineering professor at University at Buffalo. Wu’s research interests include simulation of hurricane hazards (wind, rain, surge/wave) and their impacts on coastal communities using high-fidelity (e.g., WRF+CFD), analytical (e.g., reduced Navier-Stokes equations) and machine learning (e.g., knowledge-enhanced neural networks) models.

Seymour M.J. Spence: Seymour M.J. Spence is an Associate Professor at the University of Michigan. Spence specializes in AI-integrated structural engineering. One of his recent contributions in this area is the development of frameworks that combine 3D building modeling, Computational Fluid Dynamics, and AI techniques to estimate hurricane-induced damage, providing high-fidelity, individual-building risk assessments that significantly enhance community resilience against hurricanes.

Title: Liquefaction-Induced Hazards Effects on Buried Utilities
Working Group: Regional Simulations for Lifelines and Transportation
Host: Jon Bray, Univ. of California, Berkeley
Date: October 9, 2024 (10am Pacific; 1pm Eastern)
Watch on YouTube

Abstract

The seismic performance of our buried infrastructure is critical to achieving a resilient nation. Liquefaction-induced ground deformation can damage buried utilities, such as natural gas pipelines and water distribution pipes. Recent advances in developing seismic risk methodologies and open-source software that provide quantitative estimates of seismic risk including the uncertainty range in the risk will be presented. These methodologies and software are required to support risk-informed decision-making by utility companies, their consultants, and regulators. Additionally, the OpenSRA and R2D software provide researchers platforms in which to translate their results into practice.

In the first hour of the webinar, invited speakers will present advances of our understanding of the seismic performance of buried natural gas pipelines affected by liquefaction through the talks listed below after a brief introduction.

Presenters

• Tom O'Rourke, Cornell Univ. - "Liquefaction hazard assessment: PEER CEC OpenSRA project"
• Domniki Asimaki, Caltech - "A Probabilistic, data-informed framework for regional estimation of liquefaction displacements"
• Katerina Ziotopoulou, Univ. of California, Davis - "Nonlinear dynamic analyses of ground and pipeline failures due to lateral spreading"
• Barry Zheng, Slate Geotechnical Consultants, and Jinyan Zhao, Univ. of California, Berkeley and NEHRI SimCenter - "Modeling of liquefaction-induced hazards and effects on buried utilities in OpenSRA and R2D"

Title: Surrogate Modeling of Site, Building, and Bridge System Performance in Regional Earthquake Simulations
Working Group: Regional Simulation of Earthquakes
Host: Gregory Deierlein, J.A. Blume Professor, Stanford University
Date: October 10, 2024, (10am Pacific; 1pm Eastern)
Watch on YouTube

Abstract

By reducing the computational burden that would otherwise be required using detailed analyses, data-driven surrogate models offer unprecedented opportunities to incorporate high-resolution behavior of local site and structural system response in regional simulations. This community roundtable will focus on the question, “How to create useful and usable surrogate models for earthquake engineering – status quo, limitations, and opportunities.” The roundtable will feature three speakers to prompt discussion. Topics to be addressed include (1) illustrative use cases to define key features and applications of surrogate models, (2) a framework for incorporating high-dimension surrogate models in regional simulations, (3) best practices for developing and sharing surrogate models, and (4) assessing the additional uncertainty introduced into regional simulations by surrogate models.

Moderator: Gregory Deierlein, J.A. Blume Professor, Stanford University. Dr. Deierlein is the co-director of the NHERI SimCenter and chair of the SimCenter working group on regional earthquake simulations. His research focuses on seismic design and behavior of structures, computational simulation of buildings and civil infrastructure, performance-based engineering, and informing practices and policies to promote urban resilience.

Speakers:

Henry Burton, Associate Professor, University of California Los Angeles (UCLA).

Title: Surrogate Modeling for Regional Seismic Risk-Based Assessment of Building Inventories

Abstract: This presentation will discuss surrogate modeling strategies for regional seismic performance assessment of large-scale building inventories. To this end, an inventory of approximately 15,000 single and multifamily wood-frame residential buildings in the City of Los Angeles is assessed.  Regional seismic performance is quantified in terms of the probabilistic distribution of earthquake-induced repair costs and functional recovery times. The efficacy of the surrogate modeling approach is benchmarked against the results from a high-fidelity, high-resolution, mechanistic (explicit) assessment.

Bio: Dr. Burton’s research is aimed at reducing the sociotechnical impacts of natural hazard events on communities. His research focus is in structural/earthquake engineering and the application of probability and statistics to challenges associated with the built environment.

Brett Maurer, Associate Professor, University of Washington.

Title: Mechanics-informed machine learning for geospatial modeling of soil liquefaction: global surrogate models for simulation and near-real-time response

Abstract: Data-driven "geospatial" models for predicting soil liquefaction serve an important role in regional-scale applications and are implemented in national, public-facing products. We discuss a new approach to developing such models, wherein first-order geospatial information is trained to mimic the predictions of higher-order geotechnical models. The resulting "surrogate" models benefit from human knowledge of liquefaction mechanics and are geostatistically updated by subsurface tests, which are increasingly accessible in regional and national community datasets. This approach is not limited to liquefaction but can be used to predict other subsurface measurements (e.g., shear-wave velocities) and responses (e.g., coseismic landslides) at regional scales.

Bio: Dr. Maurer’s research focus is geotechnical earthquake engineering, where he uses methods at the intersection of data science and geotechnics to address topics that include soil liquefaction and landslides, paleoseismology, hazard-mitigation economics, in-situ and remote site characterization, and post-earthquake investigation.

Yazhou (Tim) Xie, Assistant Professor, McGill University.

Title: Surrogate Modeling for Efficient Regional Seismic Fragility Assessment of Bridge Infrastructure through Active Learning and Deep Learning

Abstract: This presentation introduces integrated frameworks combining machine learning and surrogate modeling to improve regional seismic fragility assessment of bridge portfolios. The framework utilizes Gaussian process regression and active learning to reduce the need for extensive nonlinear dynamic response analyses, achieving reliable bridge-specific fragility models with just 70 bridge-model-ground-motion samples. Additionally, a deep learning approach using the convolutional variational autoencoder is applied to classify and generate ground motions, enabling consistent fragility estimates for bridges with minimal ground motion inputs. These methods enhance the efficiency in assessing the regional seismic risk of bridge infrastructure.

Bio: Dr. Xie’s research interests lie within the broad areas of resilience, infrastructure systems, and natural hazards. His research goal is to promote hazard resilience and sustainability of infrastructure systems using advanced modeling and smart protection.