{"id":{"repo_id":"uiuc","oai_identifier":"oai:www.ideals.illinois.edu:2142/129676"},"canonical_url":"https://search.dev.ndltd.org/etd/uiuc/oai:www.ideals.illinois.edu:2142/129676","repository":{"repo_id":"uiuc","name":"University of Illinois - Urbana-Champaign","base_url":"https://www.ideals.illinois.edu/oai-pmh"},"display":{"title":"Autonomous post-earthquake damage assessment augmented by graphics-based digital twin and deep learning","abstract":"Submission published under a 24 month embargo labeled 'Closed Access', the embargo will last until 2027-05-01","abstract_html":"Submission published under a 24 month embargo labeled 'Closed Access', the embargo will last until 2027-05-01","abstract_has_math":false,"creators":["Zhai, Guanghao"],"institution":"University of Illinois Urbana-Champaign","degree_name":"Ph.D.","degree_level":"Dissertation","degree_discipline":"Civil Engineering","degree_department":null,"school":null,"contributors":["Spencer, Billie F.","Golparvar Fard, Mani","Alipour, Mohamad","Yan, Jinhui","Ramirez, Julio A."],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-04-07","date_published":"2025-04-07","updated_at":"2026-07-22T22:25:05Z","subjects":["Structural Health Monitoring","Computer Vision","Deep Learning","Finite Element Method","Digital Twin","Seismic Damage Assessment."],"languages":["en","eng"],"rights":["Copyright 2025 Guanghao Zhai"],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/2142/129676","outbound_label":"Handle","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Spencer, Billie F.","Golparvar Fard, Mani","Alipour, Mohamad","Yan, Jinhui","Ramirez, Julio A."]},{"key":"dc:creator","label":"Author","values":["Zhai, Guanghao"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2025-04-07","2025-05"]},{"key":"dc:type","label":"Dc Type","values":["text","Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Civil Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Dissertation"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Ph.D."]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of Illinois Urbana-Champaign"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Structural Health Monitoring","Computer Vision","Deep Learning","Finite Element Method","Digital Twin","Seismic Damage Assessment."]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["en","eng"]},{"key":"dc:rights","label":"Dc Rights","values":["Copyright 2025 Guanghao Zhai"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://hdl.handle.net/2142/129676"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Submission published under a 24 month embargo labeled 'Closed Access', the embargo will last until 2027-05-01","The student, Guanghao Zhai, accepted the attached license on 2025-03-29 at 10:35.","The student, Guanghao Zhai, submitted this Dissertation for approval on 2025-03-29 at 10:55.","This Dissertation was approved for publication on 2025-04-07 at 11:09.","DSpace SAF Submission Ingestion Package generated from Vireo submission #21699 on 2025-10-19 at 19:52:33","Rapid assessment of the safety of structures following an earthquake is critical for making decisions regarding needed repairs and resumed operations. Tremendous effort has been devoted to developing automated post-earthquake inspection techniques that leverage Unmanned Aerial Vehicles (UAV) and deep learning techniques. While advances have been made in autonomous data acquisition and damage identification, challenges persist in interpreting identified damage in relation to structural conditions and implementing comprehensive damage assessment frameworks. These limitations highlight the necessity of completing the inspection process to establish a fully integrated approach. To address this gap, digital twin technology and synthetic data present promising solutions to several key challenges, namely: (i) the scarcity of high-quality training data due to the infrequency of earthquakes; (ii) the subjective, unclear, and error-prone interpretation of identified damage when assessing an overall structural condition; (iii) the computationally demanding process of creating digital twins for entire structures; and (iv) limited access to earthquake-damaged regions due to safety and logistical concerns that constrain testing and the validation of developed approaches. To address these challenges, this dissertation develops an autonomous post-earthquake damage estimation framework with four key research tasks: (1) synthetic rendering of 2D images of structural damage for data augmentation, (2) development and evaluation of a damage state estimation framework based on a graphic-based digital twin, (3) an accurate and efficient structural response simulation using the data/physics-driven method, and (4) a comprehensive structural condition assessment of a full three-story building based within a highly realistic synthetic environment. Task 1 develops a synthetic data augmentation method to improve pixel-wise crack damage identification and address the lack of training data. To validate this method, a dataset of cracked bridge girders was used, which includes limited and unique features. The procedure generates synthetic textures to replicate observed damage patterns and augment the original dataset, creating an augmented dataset with greater diversity. A Fully Convolutional Network (FCN) is used to identify fatigue cracks. Comparisons show that models trained with the augmented dataset outperformed those trained with real-world data alone, demonstrating the method’s effectiveness for crack detection in conditions with limited data and the potential for synthetic data to represent realistic damage patterns. With the power of creating realistic damage, Task 2 develops a framework called the Bidirectional Graphics-based Digital Twin (Bi-GBDT) to connect structural conditions with realistic visual damage patterns, as seen in post-earthquake photos. The GBDT is established from Finite Element (FE) models and its realistic graphic representation is transferred from simulated behavior (e.g., deformation, stress, strain, and damage indices) to predict the behavior of and damage to the target structure, referred to as “forward prediction”. The data generated from GBDT subjected to the prescribed earthquake are paired and used to train the deep learning networks to predict damage measures and damage states from visual damage patterns, known as backward prediction. The developed framework is validated on both forward and backward prediction based on a reinforced concrete shear wall experiment. Forward prediction is validated by comparing the experimentally obtained shear wall response (e.g., drift and damage patterns) with those derived from FE model results. The backward prediction is validated by evaluating the performance of predicting damage measures and damage states that is achieved by using deep learning networks trained on datasets generated by GBDT. Task 3 extends the scale of the Bi-GBDT framework from an individual structural component (e.g., a reinforced concrete shear wall) to entire structures (e.g., a multi-story building) by achieving more accurate and efficient simulations. Traditional methods, such as the FE method, often face challenges regarding efficiency, while data-driven approaches, such as deep neural networks, lack robustness and generalization. To address these limitations, this study develops a “data/physics-driven simulation framework” that integrates the strengths of both approaches for an accurate and efficient structural response simulation for large-scale structures. Complex components are modeled using data-driven techniques, while simpler elements rely on physics-driven methods. To further enhance the accuracy of carrying out challenging tasks involving complex elements, an advanced neural network, AttSNN, is developed, integrating conditional augmentation and attention mechanisms. A case study of a three-story frame/shear-wall building demonstrates the framework’s effectiveness in balancing accuracy and efficiency. The autonomous damage assessment process advances significantly with the establishment of an efficient and accurate relationship between visual damage and damage states. Task 4 develops an end-to-end damage assessment framework based on the Bi-GBDT, integrating advancements from previous tasks into a comprehensive virtual implementation that is tested in a realistic synthetic environment. This methodology includes the automated generation of damaged GBDTs, the creation of a synthetic environment that incorporates environmental complexities, and a simulation of UAV flight trajectories for data collection. Advanced deep learning networks are adapted to process the collected data, enabling precise damage identification and damage state estimation. The completion of these four tasks will establish a comprehensive framework for autonomously quantifying and assessing seismic damage, thereby bridging the gap between single-object analyses and scalable, real-world damage assessments while providing a robust and innovative tool for structural evaluation of post-disaster scenarios. The anticipated outcome provides a strong foundation for developing automated strategies for post-earthquake structural assessment."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Autonomous post-earthquake damage assessment augmented by graphics-based digital twin and deep learning"]}]}],"canonical_facts":{"dc:contributor":["Spencer, Billie F.","Golparvar Fard, Mani","Alipour, Mohamad","Yan, Jinhui","Ramirez, Julio A."],"dc:creator":["Zhai, Guanghao"],"dc:date":["2025-04-07","2025-05"],"dc:description":["Submission published under a 24 month embargo labeled 'Closed Access', the embargo will last until 2027-05-01","The student, Guanghao Zhai, accepted the attached license on 2025-03-29 at 10:35.","The student, Guanghao Zhai, submitted this Dissertation for approval on 2025-03-29 at 10:55.","This Dissertation was approved for publication on 2025-04-07 at 11:09.","DSpace SAF Submission Ingestion Package generated from Vireo submission #21699 on 2025-10-19 at 19:52:33","Rapid assessment of the safety of structures following an earthquake is critical for making decisions regarding needed repairs and resumed operations. Tremendous effort has been devoted to developing automated post-earthquake inspection techniques that leverage Unmanned Aerial Vehicles (UAV) and deep learning techniques. While advances have been made in autonomous data acquisition and damage identification, challenges persist in interpreting identified damage in relation to structural conditions and implementing comprehensive damage assessment frameworks. These limitations highlight the necessity of completing the inspection process to establish a fully integrated approach. To address this gap, digital twin technology and synthetic data present promising solutions to several key challenges, namely: (i) the scarcity of high-quality training data due to the infrequency of earthquakes; (ii) the subjective, unclear, and error-prone interpretation of identified damage when assessing an overall structural condition; (iii) the computationally demanding process of creating digital twins for entire structures; and (iv) limited access to earthquake-damaged regions due to safety and logistical concerns that constrain testing and the validation of developed approaches. To address these challenges, this dissertation develops an autonomous post-earthquake damage estimation framework with four key research tasks: (1) synthetic rendering of 2D images of structural damage for data augmentation, (2) development and evaluation of a damage state estimation framework based on a graphic-based digital twin, (3) an accurate and efficient structural response simulation using the data/physics-driven method, and (4) a comprehensive structural condition assessment of a full three-story building based within a highly realistic synthetic environment. Task 1 develops a synthetic data augmentation method to improve pixel-wise crack damage identification and address the lack of training data. To validate this method, a dataset of cracked bridge girders was used, which includes limited and unique features. The procedure generates synthetic textures to replicate observed damage patterns and augment the original dataset, creating an augmented dataset with greater diversity. A Fully Convolutional Network (FCN) is used to identify fatigue cracks. Comparisons show that models trained with the augmented dataset outperformed those trained with real-world data alone, demonstrating the method’s effectiveness for crack detection in conditions with limited data and the potential for synthetic data to represent realistic damage patterns. With the power of creating realistic damage, Task 2 develops a framework called the Bidirectional Graphics-based Digital Twin (Bi-GBDT) to connect structural conditions with realistic visual damage patterns, as seen in post-earthquake photos. The GBDT is established from Finite Element (FE) models and its realistic graphic representation is transferred from simulated behavior (e.g., deformation, stress, strain, and damage indices) to predict the behavior of and damage to the target structure, referred to as “forward prediction”. The data generated from GBDT subjected to the prescribed earthquake are paired and used to train the deep learning networks to predict damage measures and damage states from visual damage patterns, known as backward prediction. The developed framework is validated on both forward and backward prediction based on a reinforced concrete shear wall experiment. Forward prediction is validated by comparing the experimentally obtained shear wall response (e.g., drift and damage patterns) with those derived from FE model results. The backward prediction is validated by evaluating the performance of predicting damage measures and damage states that is achieved by using deep learning networks trained on datasets generated by GBDT. Task 3 extends the scale of the Bi-GBDT framework from an individual structural component (e.g., a reinforced concrete shear wall) to entire structures (e.g., a multi-story building) by achieving more accurate and efficient simulations. Traditional methods, such as the FE method, often face challenges regarding efficiency, while data-driven approaches, such as deep neural networks, lack robustness and generalization. To address these limitations, this study develops a “data/physics-driven simulation framework” that integrates the strengths of both approaches for an accurate and efficient structural response simulation for large-scale structures. Complex components are modeled using data-driven techniques, while simpler elements rely on physics-driven methods. To further enhance the accuracy of carrying out challenging tasks involving complex elements, an advanced neural network, AttSNN, is developed, integrating conditional augmentation and attention mechanisms. A case study of a three-story frame/shear-wall building demonstrates the framework’s effectiveness in balancing accuracy and efficiency. The autonomous damage assessment process advances significantly with the establishment of an efficient and accurate relationship between visual damage and damage states. Task 4 develops an end-to-end damage assessment framework based on the Bi-GBDT, integrating advancements from previous tasks into a comprehensive virtual implementation that is tested in a realistic synthetic environment. This methodology includes the automated generation of damaged GBDTs, the creation of a synthetic environment that incorporates environmental complexities, and a simulation of UAV flight trajectories for data collection. Advanced deep learning networks are adapted to process the collected data, enabling precise damage identification and damage state estimation. The completion of these four tasks will establish a comprehensive framework for autonomously quantifying and assessing seismic damage, thereby bridging the gap between single-object analyses and scalable, real-world damage assessments while providing a robust and innovative tool for structural evaluation of post-disaster scenarios. The anticipated outcome provides a strong foundation for developing automated strategies for post-earthquake structural assessment."],"dc:format":["application/pdf"],"dc:identifier":["https://hdl.handle.net/2142/129676"],"dc:language":["en","eng"],"dc:rights":["Copyright 2025 Guanghao Zhai"],"dc:subject":["Structural Health Monitoring","Computer Vision","Deep Learning","Finite Element Method","Digital Twin","Seismic Damage Assessment."],"dc:title":["Autonomous post-earthquake damage assessment augmented by graphics-based digital twin and deep learning"],"dc:type":["text","Thesis"],"thesis:degree_discipline":["Civil Engineering"],"thesis:degree_level":["Dissertation"],"thesis:degree_name":["Ph.D."],"thesis:institution_name":["University of Illinois Urbana-Champaign"]},"updated_at":"2026-07-22T22:25:05Z"}