Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Physics-Guided Machine Learning for Structural Metamodeling and Fragility Analysis
Seismic response modeling of steel structures is crucial to enhance community wide resilience under earthquake hazards. This research aims to develop a structural fragility framework utilizing machine learning tools guided by physics. The computational cost of analyzing high fidelity models of archetype steel frames can increase exponentially with numerous iterations associated with design optimization, stochastic loading, uncertainty analysis. To reduce computational demand, structural metamodeling can provide a low-order surrogate model for a higher-order system. Purely data-driven ML metamodeling requires extensive training data and lacks interpretability, reducing its reliability. Physics-informed Machine Learning (PiML) metamodeling integrates physics knowledge into the construction of the ML model, providing physical bounds to the solution space. The first step of this framework focuses on the ML aided selection of ground motions for analysis and the generation of a structural response dataset on an archetype structure. In the second step, the construction and training of a PiML metamodel is shown to capture underlying nonlinear behaviors of the structure. The third and final step utilizes both the selected GMs and trained PiML metamodel in a probabilistic structural seismic response prediction, with applications for estimating seismic fragility. This approach is assessed based on its efficiency and accuracy of seismic response prediction with comparisons to existing physics-based methods and can expand the range of archetype structures that are traditionally considered for ML-aided fragility analysis.
Physics-Guided Machine Learning for Structural Metamodeling and Fragility Analysis
Seismic response modeling of steel structures is crucial to enhance community wide resilience under earthquake hazards. This research aims to develop a structural fragility framework utilizing machine learning tools guided by physics. The computational cost of analyzing high fidelity models of archetype steel frames can increase exponentially with numerous iterations associated with design optimization, stochastic loading, uncertainty analysis. To reduce computational demand, structural metamodeling can provide a low-order surrogate model for a higher-order system. Purely data-driven ML metamodeling requires extensive training data and lacks interpretability, reducing its reliability. Physics-informed Machine Learning (PiML) metamodeling integrates physics knowledge into the construction of the ML model, providing physical bounds to the solution space. The first step of this framework focuses on the ML aided selection of ground motions for analysis and the generation of a structural response dataset on an archetype structure. In the second step, the construction and training of a PiML metamodel is shown to capture underlying nonlinear behaviors of the structure. The third and final step utilizes both the selected GMs and trained PiML metamodel in a probabilistic structural seismic response prediction, with applications for estimating seismic fragility. This approach is assessed based on its efficiency and accuracy of seismic response prediction with comparisons to existing physics-based methods and can expand the range of archetype structures that are traditionally considered for ML-aided fragility analysis.
Physics-Guided Machine Learning for Structural Metamodeling and Fragility Analysis
Lecture Notes in Civil Engineering
Mazzolani, Federico M. (Herausgeber:in) / Piluso, Vincenzo (Herausgeber:in) / Nastri, Elide (Herausgeber:in) / Formisano, Antonio (Herausgeber:in) / Bond, Robert Bailey (Autor:in) / Ren, Pu (Autor:in) / Sun, Hao (Autor:in) / Hajjar, Jerome F. (Autor:in)
International Conference on the Behaviour of Steel Structures in Seismic Areas ; 2024 ; Salerno, Italy
03.07.2024
12 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Metamodeling and Machine Learning
| Springer Verlag | 2021
| Taylor & Francis Verlag | 2021
Physics-informed machine learning for metamodeling thermal comfort in non-air-conditioned buildings
| Springer Verlag | 2023
| British Library Conference Proceedings | 1994
Seismic Fragility Analysis of Structural Systems
| Online Contents | 2006