Files and links (1)
PDFOpen Access (License Unspecified), Open Access
Dissertation
As-damaged numerical modeling of reinforced concrete shear walls using graph theory and generative artificial intelligence
Doctor of Philosophy (Ph.D.), Drexel University
15 Dec 2025
DOI:
https://doi.org/10.17918/00011233
Abstract
The assessment of reinforced concrete (RC) shear walls after seismic events remains largely qualitative, relying on visual judgment rather than measurable indicators of damage. Current guidelines, such as ASCE 41-23, ACI 369.1-22, and FEMA 306, provide descriptive classifications of cracking to infer stiffness degradation but lack quantitative procedures for modeling the as-damaged condition of structural members. This subjectivity introduces uncertainty into post-event evaluations and hinders the development of physics-based models that can accurately predict the performance of damaged structures under subsequent loading. Addressing this limitation, this research presents a unified and quantitative framework for as-damaged modeling of RC shear walls through the integration of generative artificial intelligence, graph-based crack quantification, regression analysis, and nonlinear structural simulation.The research begins with the development of a denoising diffusion probabilistic model (DDPM) trained on experimental crack images to generate physically realistic and statistically diverse synthetic crack patterns. The proposed crack-to-graph conversion framework then transforms crack images into mathematical graph representations, enabling the extraction of measurable topological and weighted features that objectively describe surface damage. A regression-based modeling approach is established to relate the extracted graph features to stiffness reduction factors ([lambda]-values), producing a quantitative relationship between crack morphology and mechanical degradation. Finally, the derived [lambda]-values are incorporated into nonlinear OpenSees simulations to represent the as-damaged condition of RC shear walls and to validate the predictive capability of the proposed framework under cyclic and multi-hazard loading. The results confirm that the proposed methodology can successfully link visual damage characteristics to structural performance parameters in a physically interpretable manner. The generated crack images demonstrated strong statistical similarity to experimental data, the graph features correlated consistently with stiffness degradation trends, and the [lambda]-based OpenSees models accurately reproduced experimental hysteretic responses. Collectively, these outcomes establish a reproducible and quantitative pathway from surface crack observation to analytical modeling, transforming post-event evaluation of RC structures from qualitative assessment to physics-informed prediction.
Metrics
36 File views/ downloads
20 Record Views
Details
- Title
- As-damaged numerical modeling of reinforced concrete shear walls using graph theory and generative artificial intelligence
- Creators
- Pedram Bazrafshan
- Contributors
- Arvin Ebrahimkhanlou (Advisor)
- Awarding Institution
- Drexel University
- Degree Awarded
- Doctor of Philosophy (Ph.D.)
- Publisher
- Drexel University
- Number of pages
- ix, 104 pages
- Resource Type
- Dissertation
- Language
- English
- Academic Unit
- Civil (and Architectural) Engineering [Historical]; College of Engineering (1970-2026); Drexel University
- Other Identifier
- 991022146706804721