Postdoctoral Scholar University of California, Berkeley
A Digital Twin Approach for Dynamic Risk Assessment Under Sequential Earthquake Hazards
Co-Authors: Luis Cossio (University of California, Berkeley), Luis Ceferino (University of California, Berkeley), Ziqi Wang (University of California, Berkeley), and Michael Lindsey (University of California, Berkeley)
Abstract: Abstract(250-word max.)
Strong earthquakes can cause widespread damage to electrical distribution infrastructure, leading to cascading blackouts and distress for utilities and community members. Sequential extreme events, such as earthquake mainshocks followed by aftershocks, further complicate decision-making for utilities by forcing repeated assessments and repairs, prolonging outages, and increasing restoration challenges. To address these decisions, utilities require real-time tools that combine physics-based modeling with monitoring of power infrastructure to assess damage and provide critical information for repair prioritization during disaster response. This talk presents a digital twin framework that updates the nonlinear properties of power infrastructure after each hazard sequence using time-history response data, Bayesian inference, and structural models. We demonstrate the framework using a complex cable-connected pole system subjected to a sequence of earthquakes with no intermediate repairs. First, we developed a high-fidelity finite element model of a cable-connected pole system that captures large cable deformations and material nonlinearities caused by pole damage. Second, we implement Bayesian inference to update the nonlinear parameters of the digital cable-connected pole system using recorded time-history data after each earthquake to track the strength degradation of power poles that may trigger repairs. Such a digital twin can provide utilities with real-time information on property degradation and damage states of individual poles, supporting more informed prioritization of inspections and repairs during sequential earthquake events.
Postdoctoral Scholar Florida A&M University-Florida State University College of Engineering.
A Scalable Computational Framework for Regional Scale Flood Fragility Assessment of Bridge Networks Using Open-Source Data
Co-Authors: Neetesh Sharma (Florida A&M University-Florida State University) and Eren Erman Ozguven (Florida A&M University-Florida State University)
Abstract: Flooding poses a persistent and growing threat to transportation networks, yet evaluating flood vulnerability at the regional scale remains challenging due to the data and computational demands of detailed hydraulic and structural models. This paper presents a computationally efficient framework for assessing flood fragility across large bridge inventories using only widely available open datasets and class-based structural representations. The approach integrates simplified estimation of flood intensity measures with a joint depth velocity hazard formulation and a reliability-based probabilistic model to characterize component-level failure of representative bridge classes. Designed for regional deployment, the framework enables systematic evaluation of hundreds of structures. A case study of the bridge network in Duval County, Florida, demonstrates the method's ability to distinguish class-dependent flood responses, reveal nonlinear increases in failure probability beyond critical inundation thresholds, and quantify the influence of key parameters, such as deck freeboard and self-weight, through sensitivity analysis. The results show that regional flood fragility can be assessed in a transparent and computationally tractable manner, supporting large-scale screening, prioritization, and resilience planning in data-constrained environments.
PhD Student Stanford University
Probabilistic Connection Fracture Assessment of Tall Welded Steel Moment Frame Buildings during the 1989 Loma Prieta Earthquake
Co-Authors: Gregory Deierlein (Stanford University) and Francisco Galvis (ResiQuant Inc.)
Abstract: The brittle performance of welded beam-to-column connections in welded steel moment frames (WSMFs), documented after the 1994 Northridge earthquake, highlighted their vulnerability to low-cycle fatigue fracture and raised questions regarding their damage potential during prior strong-motion events. This study develops a probabilistic framework to quantify the likelihood of connection fractures in a set of WSMF building archetypes representative of structures in San Francisco affected by the 1989 Loma Prieta earthquake. The proposed approach integrates (1) site- and orientation-dependent estimates of Loma Prieta ground motion intensities, enabling implementation within SimCenter tools such as R2D to generate directionality-consistent intensity measures, and (2) nonlinear response history analyses of detailed WSMF models to estimate probabilities of beam-to-column weld fracture in conjunction with FEMA-352 damage indices. Uncertainties in ground motion intensity and fracture model parameters are propagated using a Monte Carlo simulation framework. Application to seven buildings indicates that, despite moderate shaking levels, connection fractures may have occurred, with aggregated floor damage indices reaching values up to 0.5. Fracture demands tend to concentrate on stories experiencing the largest drift demands, although variability across buildings suggests additional influence from structural irregularities, member sizes, and bay spans. The methodology enables visualization of connection-level fracture probabilities and floor damage indices, supporting targeted inspection strategies and resilience-informed decision-making. The workflow is compatible with the R2D environment and can be deployed to enhance ground motion characterization and structural response estimation by incorporating additional earthquake features and detailed numerical models that explicitly capture connection fracture behavior.
PhD Student Oklahoma State University
Quantifying the Effects of Near-Fault Ground Motion Directionality on Regional Seismic Risk Assessment using Physics-Based Fault-Rupture Simulations
Co-Author: Maha Kenawy (Oklahoma State University)
Abstract: The variation of the intensity of ground shaking with changes in orientation (ground motion directionality) can substantially influence the seismic risks to civil structures. This is especially the case at near-fault locations where ground motions tend to be highly polarized—one horizontal component has a substantially higher intensity than the other component. Although ground motion directionality is well documented, its spatial variation and influence on near-fault structural demands are difficult to assess due to the sparsity of near-fault ground motion recordings. In this study, we use physics-based earthquake simulations with broadband and spatially dense ground motions, in combination with nonlinear simulations of building structures, to assess the effects of ground motion directionality on near-fault seismic structural demands. We leverage a database of fifty Hayward fault rupture simulations to evaluate the spatial variability of the maximum direction of ground motion intensity across the San Francisco Bay Area, and its systematic dependence on the fault rupture characteristics and site conditions. Our structural simulations identify important trends of the spatial variation of site-specific and direction-dependent seismic demands on low-rise and high-rise building structures influenced by the earthquake source, path, and site effects, such as the hypocenter location, directivity, fling pulses, and soil properties. We use our simulations to develop and evaluate a spatial Gaussian process model that describes the variation of directionality features in regions of high seismic hazard. Our findings can be used to improve the characterization of ground motion and structural demand directionality in regional-scale seismic risk assessment applications.
PhD Student UC Berkeley
Phase Diagrams for Multi-Scale Regional Structural Risk Under Natural Hazards
Co-Author: Ziqi Wang (UC Berkeley)
Abstract: Citywide damage patterns that extend far beyond individual structures under major natural hazards demand regional and city-scale risk assessment for resilient natural hazard engineering. Yet much of the field still prioritizes ever-finer resolution and higher-fidelity simulations, with limited attention to the universal patterns or first-principles explanations that govern collective structural responses and that detailed simulations alone may not reveal. In recent work, we demonstrated that regional-scale structural damage exhibits phase-transition behavior: as hazard intensity increases, the collective state can shift abruptly from largely safe to largely damaged, whereas increasing structural-property diversity smooths this shift into a continuous transition. Framed in the language of statistical physics, these behaviors define a phase diagram in the space of hazard intensity and regional structural diversity and provide actionable diagnostics, underscoring the importance of phase-awareness for balanced city-scale hazard risk assessment.
Here we extend this framework through numerical coarse-graining to test how collective risk phases evolve as the spatial scale of interest increases. Starting from building-level simulation ensembles, we aggregate local damage states into progressively larger spatial units using a decision-relevant coarse-graining rule, and we track how damage-fraction distributions, dominant collective states, and inferred phase boundaries change with scale. We find that the phase diagram’s qualitative structure, i.e., the emergence of distinct phases and the transition boundaries between them, persists across scales, suggesting that phase-diagram representations capture scale-consistent, emergent features of regional damage. This scale-aware phase representation formalizes how collective-risk phases are renormalized by aggregation, linking parcel-level damage patterns to district- and city-level decisions.
Postdoctoral Scholar Texas A&M University
Multi-State Spatial Fragility Modeling Using a Warped Heteroscedastic Gaussian Process
Co-Author: Maria Koliou (Texas A&M University)
Abstract: This study presents a novel probabilistic methodology for spatial fragility modeling based on a warped, heteroscedastic Gaussian Process. Unlike conventional approaches that treat damage probabilities as conditionally independent and deterministic point estimates, the proposed methodology models fragility as a spatially and structurally correlated random field, where each fragility probability is treated as a random variable governed by two layers of uncertainty: aleatory and epistemic. A latent fragility index is derived and spatially modeled over coordinates, damage states, and structural archetypes, and a composite multi-output kernel encodes correlations across these dimensions. Fragility probabilities are then obtained via a nonlinear warping of the latent Gaussian Process, resulting in a tractable multivariate distribution. Application to over 15,000 buildings in the San Francisco Bay Area under a regional seismic hazard demonstrates the method’s ability to generate spatially coherent fragility maps with quantified uncertainty. Ordinality across damage states is preserved not only in expectation, but throughout the full joint distribution, and correlation among spatially proximate and structurally similar buildings is explicitly modeled. Results illustrate the methodology’s ability to produce spatially coherent, uncertainty-aware fragility predictions within an analytically tractable and computationally efficient framework.
PhD Student Stanford University
Evaluating the Impact of Seismic Retrofit Programs using VLM-Enhanced Building Inventories
Co-Authors: Iro Armeni (Stanford University) and Gregory Deierlein (Stanford University)
Abstract: One- and two-story wood housing generally performs well in terms of life safety under earthquake loading. However, certain seismic vulnerabilities, including crawlspace (cripple wall) foundations, living spaces over garages, and masonry chimneys have led to high financial loss and human displacement in past earthquakes. Building-level analysis has demonstrated significant performance differences depending on such features; however, most regional-level models of seismic risk use one fragility function, depending on era, to represent all wood-frame housing. Moving to more detailed models of wood-frame housing to capture the impact of specific vulnerabilities and retrofit measures at the regional scale requires sufficiently detailed building inventories. However, existing building inventory data rarely describes such features. This study addresses that gap by leveraging street view imagery alongside GPT-5 to systematically collect relevant building features. Results show that GPT-5 can reliably identify visible features (i.e., number of stories, living space over garage) and that accuracy can be improved using model reasoning and chain-of-thought prompting. While foundation types are inherently difficult to classify using only exterior building views, probability-based model predictions and neighborhood context can produce more detailed, accurate results. Following the methodology used in the SimCenter’s BRAILS++ tool, VLM results are used in conjunction with existing national and local inventory data to more readily understand the amount of seismic risk in a case study city. Furthermore, these building inventories are paired with detailed structural models using the SimCenter’s R2D and Pelicun tools to assess the costs and benefits of specific seismic retrofit programs aimed at reducing risk.