A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Estimating hurricane-induced vertical surge and wave loads on elevated coastal buildings based on CFD simulations and ensemble learning
https://orcid.org/0000-0002-9543-9766
Abstract This study focuses on developing Machine Learning models based on CFD simulations to estimate hurricane induced vertical wave and surge forces on elevated coastal buildings caused by regular waves. In particular, this study aims at addressing the existing gap in designing elevated coastal buildings, which stems from lack of a generalizable equation for estimating vertical hurricane wave and surge forces for a variety of building geometries and wave conditions. To this end, based on a validated Computational Fluid Dynamics (CFD) model, six key variables that affect the wave forces on buildings are considered, which include building dimensions (length and width of rectangular plan buildings) and wave characteristics. In addition, the effect of building length in relation to wave length on the resulting vertical forces is explicitly considered, while the effect of building width is considered implicitly. Subsequently, a Latin Hypercube Sampling (LHS) strategy is implemented for generating 256 random samples, each representing a unique building geometry with a rectangular plan and wave condition scenario. A pipeline is then developed for generating CFD models corresponding to each sample, which are then deployed on a high-performance computing (HPC) cluster, creating a dataset of vertical surge and wave forces. Based on the key design variables, an ensemble of Machine Learning (ML) models is developed to predict vertical wave forces. For the purpose of design and risk assessment, different force levels may be of interest. Therefore, different models were trained to predict multiple levels of forces on elevated buildings, including maximum, average, 25th, 75th, and 99th percentiles of peak forces. The developed model for each of these force levels is then investigated to identify the main building geometry and wave conditions features that affect the resulting forces, which can inform the decisions of the designers in the early-stage phases. For further assistance in the early-stage design, based on a subset of identified important features, a simple expression for each force level is provided to narrow down the design space to top performing early design candidates, while the main estimators can be used for the final design of elevated buildings. Results show that based on CFD simulations, highly accurate and generalizable predictive models with error rates of about 4% on the test set can be developed.
Highlights Estimating the vertical surge and wave forces on elevated coastal buildings under regular waves. Conducting 256 CFD simulations derived from an enhanced Design of Experiment method. Capturing the effect of building geometry and wave properties on the resulting forces. Using Ensemble Learning method to provide a highly accurate and generalizable estimate of surge and wave forces. Identifying new important features that influence the resulting surge and wave forces specific to elevated coastal buildings.
Estimating hurricane-induced vertical surge and wave loads on elevated coastal buildings based on CFD simulations and ensemble learning
https://orcid.org/0000-0002-9543-9766
Abstract This study focuses on developing Machine Learning models based on CFD simulations to estimate hurricane induced vertical wave and surge forces on elevated coastal buildings caused by regular waves. In particular, this study aims at addressing the existing gap in designing elevated coastal buildings, which stems from lack of a generalizable equation for estimating vertical hurricane wave and surge forces for a variety of building geometries and wave conditions. To this end, based on a validated Computational Fluid Dynamics (CFD) model, six key variables that affect the wave forces on buildings are considered, which include building dimensions (length and width of rectangular plan buildings) and wave characteristics. In addition, the effect of building length in relation to wave length on the resulting vertical forces is explicitly considered, while the effect of building width is considered implicitly. Subsequently, a Latin Hypercube Sampling (LHS) strategy is implemented for generating 256 random samples, each representing a unique building geometry with a rectangular plan and wave condition scenario. A pipeline is then developed for generating CFD models corresponding to each sample, which are then deployed on a high-performance computing (HPC) cluster, creating a dataset of vertical surge and wave forces. Based on the key design variables, an ensemble of Machine Learning (ML) models is developed to predict vertical wave forces. For the purpose of design and risk assessment, different force levels may be of interest. Therefore, different models were trained to predict multiple levels of forces on elevated buildings, including maximum, average, 25th, 75th, and 99th percentiles of peak forces. The developed model for each of these force levels is then investigated to identify the main building geometry and wave conditions features that affect the resulting forces, which can inform the decisions of the designers in the early-stage phases. For further assistance in the early-stage design, based on a subset of identified important features, a simple expression for each force level is provided to narrow down the design space to top performing early design candidates, while the main estimators can be used for the final design of elevated buildings. Results show that based on CFD simulations, highly accurate and generalizable predictive models with error rates of about 4% on the test set can be developed.
Highlights Estimating the vertical surge and wave forces on elevated coastal buildings under regular waves. Conducting 256 CFD simulations derived from an enhanced Design of Experiment method. Capturing the effect of building geometry and wave properties on the resulting forces. Using Ensemble Learning method to provide a highly accurate and generalizable estimate of surge and wave forces. Identifying new important features that influence the resulting surge and wave forces specific to elevated coastal buildings.
Estimating hurricane-induced vertical surge and wave loads on elevated coastal buildings based on CFD simulations and ensemble learning
https://orcid.org/0000-0002-9543-9766
Abstract This study focuses on developing Machine Learning models based on CFD simulations to estimate hurricane induced vertical wave and surge forces on elevated coastal buildings caused by regular waves. In particular, this study aims at addressing the existing gap in designing elevated coastal buildings, which stems from lack of a generalizable equation for estimating vertical hurricane wave and surge forces for a variety of building geometries and wave conditions. To this end, based on a validated Computational Fluid Dynamics (CFD) model, six key variables that affect the wave forces on buildings are considered, which include building dimensions (length and width of rectangular plan buildings) and wave characteristics. In addition, the effect of building length in relation to wave length on the resulting vertical forces is explicitly considered, while the effect of building width is considered implicitly. Subsequently, a Latin Hypercube Sampling (LHS) strategy is implemented for generating 256 random samples, each representing a unique building geometry with a rectangular plan and wave condition scenario. A pipeline is then developed for generating CFD models corresponding to each sample, which are then deployed on a high-performance computing (HPC) cluster, creating a dataset of vertical surge and wave forces. Based on the key design variables, an ensemble of Machine Learning (ML) models is developed to predict vertical wave forces. For the purpose of design and risk assessment, different force levels may be of interest. Therefore, different models were trained to predict multiple levels of forces on elevated buildings, including maximum, average, 25th, 75th, and 99th percentiles of peak forces. The developed model for each of these force levels is then investigated to identify the main building geometry and wave conditions features that affect the resulting forces, which can inform the decisions of the designers in the early-stage phases. For further assistance in the early-stage design, based on a subset of identified important features, a simple expression for each force level is provided to narrow down the design space to top performing early design candidates, while the main estimators can be used for the final design of elevated buildings. Results show that based on CFD simulations, highly accurate and generalizable predictive models with error rates of about 4% on the test set can be developed.
Highlights Estimating the vertical surge and wave forces on elevated coastal buildings under regular waves. Conducting 256 CFD simulations derived from an enhanced Design of Experiment method. Capturing the effect of building geometry and wave properties on the resulting forces. Using Ensemble Learning method to provide a highly accurate and generalizable estimate of surge and wave forces. Identifying new important features that influence the resulting surge and wave forces specific to elevated coastal buildings.
Estimating hurricane-induced vertical surge and wave loads on elevated coastal buildings based on CFD simulations and ensemble learning
Moeini, Mohammad (author) / Brown, Nathan (author) / Memari, Ali M. (author)
Coastal Engineering ; 183
2023-04-20
Article (Journal)
Electronic Resource
English
Wave and storm surge simulations for Hurricane Katrina using coupled process based models
| Springer Verlag | 2008
A Conceptual Hurricane Surge Protection Plan for Coastal Louisiana
| British Library Conference Proceedings | 2007