A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Efficient Seismic Fragility Assessment Through Active Learning and Gaussian Process Regression
https://orcid.org/http://orcid.org/0000-0002-2078-0038
https://orcid.org/http://orcid.org/0000-0001-7470-9459
Seismic fragility models quantify the damage and collapse exceedance probabilities of civil engineering structures under varying levels of seismic hazards. Fragility assessment plays an important role in both probabilistic seismic risk assessment and performance-based seismic design. Developing accurate and robust seismic fragility models is computationally demanding, as numerous nonlinear time history analyses (NLTHAs) are needed to capture all sources of uncertainties embedded in earthquake loads, structural geometry, material properties, design details, etc. In this regard, this study leverages active learning (AL) and Gaussian process regression (GPR) to efficiently develop seismic fragility models without conducting exhaustive NLTHAs. In particular, the GPR predicts the mean and variance of structural responses conditioned on input features (i.e. structural parameters and seismic intensity measures), from which fragility curves are constructed by convolving the probabilistic seismic demand models with capacity limit state models. Furthermore, the AL algorithm recursively selects the optimal set of motion-structure samples to carry out the least number of NLTHAs for training against the GPR-based fragility model. The accuracy and efficiency of the proposed AL-GPR scheme are demonstrated using a benchmark highway bridge class. First, the GPR-based fragility model shows superior damage/failure exceedance probability inference when compared with conventional approaches. Besides, the seismic fragility model trained on a minimal subset of AL-selected NLTHAs achieves comparable performance as the original model using 1950 samples. This research develops an advanced machine learning technique to efficiently and reliably assess the seismic fragility of structures, which tackles one crucial computational challenge to facilitate high-resolution regional seismic risk assessment of existing structures and performance-based seismic design of new structures.
Efficient Seismic Fragility Assessment Through Active Learning and Gaussian Process Regression
https://orcid.org/http://orcid.org/0000-0002-2078-0038
https://orcid.org/http://orcid.org/0000-0001-7470-9459
Seismic fragility models quantify the damage and collapse exceedance probabilities of civil engineering structures under varying levels of seismic hazards. Fragility assessment plays an important role in both probabilistic seismic risk assessment and performance-based seismic design. Developing accurate and robust seismic fragility models is computationally demanding, as numerous nonlinear time history analyses (NLTHAs) are needed to capture all sources of uncertainties embedded in earthquake loads, structural geometry, material properties, design details, etc. In this regard, this study leverages active learning (AL) and Gaussian process regression (GPR) to efficiently develop seismic fragility models without conducting exhaustive NLTHAs. In particular, the GPR predicts the mean and variance of structural responses conditioned on input features (i.e. structural parameters and seismic intensity measures), from which fragility curves are constructed by convolving the probabilistic seismic demand models with capacity limit state models. Furthermore, the AL algorithm recursively selects the optimal set of motion-structure samples to carry out the least number of NLTHAs for training against the GPR-based fragility model. The accuracy and efficiency of the proposed AL-GPR scheme are demonstrated using a benchmark highway bridge class. First, the GPR-based fragility model shows superior damage/failure exceedance probability inference when compared with conventional approaches. Besides, the seismic fragility model trained on a minimal subset of AL-selected NLTHAs achieves comparable performance as the original model using 1950 samples. This research develops an advanced machine learning technique to efficiently and reliably assess the seismic fragility of structures, which tackles one crucial computational challenge to facilitate high-resolution regional seismic risk assessment of existing structures and performance-based seismic design of new structures.
Efficient Seismic Fragility Assessment Through Active Learning and Gaussian Process Regression
https://orcid.org/http://orcid.org/0000-0002-2078-0038
https://orcid.org/http://orcid.org/0000-0001-7470-9459
Seismic fragility models quantify the damage and collapse exceedance probabilities of civil engineering structures under varying levels of seismic hazards. Fragility assessment plays an important role in both probabilistic seismic risk assessment and performance-based seismic design. Developing accurate and robust seismic fragility models is computationally demanding, as numerous nonlinear time history analyses (NLTHAs) are needed to capture all sources of uncertainties embedded in earthquake loads, structural geometry, material properties, design details, etc. In this regard, this study leverages active learning (AL) and Gaussian process regression (GPR) to efficiently develop seismic fragility models without conducting exhaustive NLTHAs. In particular, the GPR predicts the mean and variance of structural responses conditioned on input features (i.e. structural parameters and seismic intensity measures), from which fragility curves are constructed by convolving the probabilistic seismic demand models with capacity limit state models. Furthermore, the AL algorithm recursively selects the optimal set of motion-structure samples to carry out the least number of NLTHAs for training against the GPR-based fragility model. The accuracy and efficiency of the proposed AL-GPR scheme are demonstrated using a benchmark highway bridge class. First, the GPR-based fragility model shows superior damage/failure exceedance probability inference when compared with conventional approaches. Besides, the seismic fragility model trained on a minimal subset of AL-selected NLTHAs achieves comparable performance as the original model using 1950 samples. This research develops an advanced machine learning technique to efficiently and reliably assess the seismic fragility of structures, which tackles one crucial computational challenge to facilitate high-resolution regional seismic risk assessment of existing structures and performance-based seismic design of new structures.
Efficient Seismic Fragility Assessment Through Active Learning and Gaussian Process Regression
Lecture Notes in Civil Engineering
Desjardins, Serge (editor) / Poitras, Gérard J. (editor) / El Damatty, Ashraf (editor) / Elshaer, Ahmed (editor) / Ning, Chunxiao (author) / Xie, Yazhou (author)
Canadian Society of Civil Engineering Annual Conference ; 2023 ; Moncton, NB, Canada
2024-09-03
13 pages
Article/Chapter (Book)
Electronic Resource
English
Seismic fragility assessment , Active learning , Gaussian process regression , Nonlinear time history analysis , Computational efficiency Engineering , Building Construction and Design , Geoengineering, Foundations, Hydraulics , Transportation Technology and Traffic Engineering , Environment, general
Gaussian process regression for seismic fragility assessment of building portfolios
| TIBKAT | 2019
Regional-scale seismic fragility assessment based on Gaussian process regression
| BASE | 2019
Gaussian process regression for seismic fragility assessment of building portfolios
| BASE | 2020