A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Sequential sampling method using Gaussian process regression for estimating extreme structural response
https://orcid.org/0000-0001-7531-3595
https://orcid.org/0000-0002-7390-6474
https://orcid.org/0000-0003-1973-949X
Abstract A methodology for estimating extreme response statistics for marine structures, that takes both the long-term variability of the metocean environment and the short-term variability of response into account is presented. The proposed methodology uses Gaussian process regression to estimate parameters of the short-term response distribution, based on output from computationally expensive hydrodynamic simulations. We present an adaptive design strategy for sequential updating of the model, focusing on the metocean conditions that contribute the most to the long-term extreme. With this approach, only a limited number of hydrodynamic simulations are needed. The suggested approach is demonstrated on the problem of estimating the 25-year extreme vertical bending moment on a ship. We show that a relatively small number of iterations (full hydrodynamic simulations) are needed to converge toward the “exact” results obtained by running a large number of simulations covering the entire range of sea states. The results suggest that the proposed method can be used as an alternative to contour-based methods or other methods that consider a few sea states using accurate numerical simulations, with little or no added complexity or computational effort.
Highlights A response emulator based on Gaussian Process Regression is established. Long- and short-term variability of the response are accounted for. Active learning approach is applied for optimal selection of sea states to consider. Extreme ship responses are analyzed.
Sequential sampling method using Gaussian process regression for estimating extreme structural response
https://orcid.org/0000-0001-7531-3595
https://orcid.org/0000-0002-7390-6474
https://orcid.org/0000-0003-1973-949X
Abstract A methodology for estimating extreme response statistics for marine structures, that takes both the long-term variability of the metocean environment and the short-term variability of response into account is presented. The proposed methodology uses Gaussian process regression to estimate parameters of the short-term response distribution, based on output from computationally expensive hydrodynamic simulations. We present an adaptive design strategy for sequential updating of the model, focusing on the metocean conditions that contribute the most to the long-term extreme. With this approach, only a limited number of hydrodynamic simulations are needed. The suggested approach is demonstrated on the problem of estimating the 25-year extreme vertical bending moment on a ship. We show that a relatively small number of iterations (full hydrodynamic simulations) are needed to converge toward the “exact” results obtained by running a large number of simulations covering the entire range of sea states. The results suggest that the proposed method can be used as an alternative to contour-based methods or other methods that consider a few sea states using accurate numerical simulations, with little or no added complexity or computational effort.
Highlights A response emulator based on Gaussian Process Regression is established. Long- and short-term variability of the response are accounted for. Active learning approach is applied for optimal selection of sea states to consider. Extreme ship responses are analyzed.
Sequential sampling method using Gaussian process regression for estimating extreme structural response
https://orcid.org/0000-0001-7531-3595
https://orcid.org/0000-0002-7390-6474
https://orcid.org/0000-0003-1973-949X
Abstract A methodology for estimating extreme response statistics for marine structures, that takes both the long-term variability of the metocean environment and the short-term variability of response into account is presented. The proposed methodology uses Gaussian process regression to estimate parameters of the short-term response distribution, based on output from computationally expensive hydrodynamic simulations. We present an adaptive design strategy for sequential updating of the model, focusing on the metocean conditions that contribute the most to the long-term extreme. With this approach, only a limited number of hydrodynamic simulations are needed. The suggested approach is demonstrated on the problem of estimating the 25-year extreme vertical bending moment on a ship. We show that a relatively small number of iterations (full hydrodynamic simulations) are needed to converge toward the “exact” results obtained by running a large number of simulations covering the entire range of sea states. The results suggest that the proposed method can be used as an alternative to contour-based methods or other methods that consider a few sea states using accurate numerical simulations, with little or no added complexity or computational effort.
Highlights A response emulator based on Gaussian Process Regression is established. Long- and short-term variability of the response are accounted for. Active learning approach is applied for optimal selection of sea states to consider. Extreme ship responses are analyzed.
Sequential sampling method using Gaussian process regression for estimating extreme structural response
Gramstad, Odin (author) / Agrell, Christian (author) / Bitner-Gregersen, Elzbieta (author) / Guo, Bingjie (author) / Ruth, Eivind (author) / Vanem, Erik (author)
Marine Structures ; 72
2020-04-20
Article (Journal)
Electronic Resource
English
| Elsevier | 2023
Estimating Compressive Strength of High Performance Concrete with Gaussian Process Regression Model
| DOAJ | 2016