A digital twin framework for rapid performance-based post-earthquake assessment using computer vision and building information models

After an earthquake occurs, timely community recovery depends on the capacity to ensure that buildings in the affected region are safe to reoccupy. Currently, post-earthquake building inspection is hindered by several factors. Mobilizing a qualified inspection team can take weeks, and conducting the inspections can take months. Inspectors face hazardous conditions in the post-disaster environment. Finally, inspections are subjective; inspectors frequently lack access to construction documents and must rely on their own judgement in the field. To address issues with speed and safety, prior work in computer vision has developed methods for automatically detecting structural damage in photographs taken by automated unmanned aerial vehicles (UAVs), reducing the danger to humans. However, simply recognizing damage on the building exterior is insufficient; to assess the overall safety of the building, damage must be localized to individual components and interpreted in context of that component’s function in the structural system. This dissertation presents a comprehensive digital twin framework for rapid post-earthquake building safety assessment that uses a building information model (BIM) as a reference frame to tie images collected by UAVs to a performance-based structural analysis. After an earthquake occurs, a UAV survey collects images of the damaged building. The BIM guides selection of an optimal set of UAV images containing structural components of interest, which are input to a semantic segmentation algorithm for automatic damage identification. The BIM is overlaid on each image to associate any detected damage with specific building components, which are then classified into discrete damage states consistent with predetermined component fragility models. 3D point cloud change detection is also applied to classify nonstructural damage states. The damage state classifications are used to update a probability model of maximum earthquake demands on the building. Bayes’ Theorem, combined with a machine learning-based surrogate model, is applied to predict demands on both exterior visible components and interior nonvisible components. The result is a complete probabilistic characterization of earthquake demands throughout the building, which are compared against a target performance objective to assess whether the building is safe to reoccupy. An additional contribution of this research is the method for validating the framework. A synthetic environment is developed in a computer graphics program where structural materials and damage can be photorealistically rendered. Within this synthetic environment, a 3D model of the building is linked with a finite element analysis to create a physics-based graphics model (PBGM). Damage is applied to the PBGM in manner consistent the with building’s earthquake response, and a simulated UAV survey is performed of this synthetic building model. New methods for generating damage in synthetic environments are developed that allow severe structural damage to be represented. The PBGM is used to validate the digital twin framework. A PBGM is created of a reinforced concrete moment frame building in Urbana, Illinois, and the assessment methodology is applied. Damage is successfully localized and classified using the BIM, and those classifications are used to update probability models of component earthquake demands. The updated probability models better predict ground truth demand values compared to prior models that do not incorporate observed damage. Based on the probability of exceeding a target performance objective, the safety state of the example structure is correctly classified. Ultimately, this method will enable more rapid, automated post-earthquake inspections and will help ensure rapid community recovery after an earthquake.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Civil Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Levine, Nathaniel Moses

Contributors dc:contributor

• Spencer, Billie F
• Popovics, John S
• Golparvar-Fard, Mani
• Ramirez, Julio A.

• Structural engineering
• earthquake engineering
• computer vision
• building information model
• structural health monitoring
• automated inspection
• performance-based engineering

Statement dc:rights

• Copyright 2022 Nathaniel Levine

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/116032

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F116032