A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Reliability Analysis of Structures Subjected to Seismic Excitation Using a Deep Learning-Based Surrogate Model
Reliability of structures subjected to seismic loading is a persistent and challenging task as it requires performing the dynamic analysis of structures and repeating these analyses multiple times with variables sampled from predefined distributions via the Monte Carlo simulation method. Thus, this study proposed a surrogate model as an alternative for the Finite Element Method in forecasting the structure’s response to earthquakes. The proposed method consists of four main steps: (i) using experimental data from the literature to calibrate a reliable numerical model, (ii) leveraging the numerical model to generate sufficient data for data-driven method, (iii) elaborating a deep learning architecture dedicated to forecasting structure’s response to earthquake using the lastest advent of Deep Learning in handling time-series data, namely, self-attention mechanisms, and (iv) estimating the reliability of the structure of interest. The efficiency and effectiveness of the proposed approach are demonstrated in detail through an example of a 3D steel frame prone to ground motions, showing a reduction up to 180 fold in computational time compared to the plain Monte Carlo simulation using the conventional Finite Element Method, with less than 5% error.
Reliability Analysis of Structures Subjected to Seismic Excitation Using a Deep Learning-Based Surrogate Model
Reliability of structures subjected to seismic loading is a persistent and challenging task as it requires performing the dynamic analysis of structures and repeating these analyses multiple times with variables sampled from predefined distributions via the Monte Carlo simulation method. Thus, this study proposed a surrogate model as an alternative for the Finite Element Method in forecasting the structure’s response to earthquakes. The proposed method consists of four main steps: (i) using experimental data from the literature to calibrate a reliable numerical model, (ii) leveraging the numerical model to generate sufficient data for data-driven method, (iii) elaborating a deep learning architecture dedicated to forecasting structure’s response to earthquake using the lastest advent of Deep Learning in handling time-series data, namely, self-attention mechanisms, and (iv) estimating the reliability of the structure of interest. The efficiency and effectiveness of the proposed approach are demonstrated in detail through an example of a 3D steel frame prone to ground motions, showing a reduction up to 180 fold in computational time compared to the plain Monte Carlo simulation using the conventional Finite Element Method, with less than 5% error.
Reliability Analysis of Structures Subjected to Seismic Excitation Using a Deep Learning-Based Surrogate Model
Lecture Notes in Civil Engineering
Ha-Minh, Cuong (editor) / Tang, Anh Minh (editor) / Bui, Tinh Quoc (editor) / Vu, Xuan Hong (editor) / Huynh, Dat Vu Khoa (editor) / Ha, Manh-Hung (author) / Nguyen, Trong-Phu (author) / Hoang, Duc-Minh (author) / Dang, Viet-Hung (author)
CIGOS 2021, Emerging Technologies and Applications for Green Infrastructure ; Chapter: 194 ; 1917-1926
2021-10-28
10 pages
Article/Chapter (Book)
Electronic Resource
English
Unanchored Structures Subjected to Seismic Excitation
| British Library Conference Proceedings | 2000
| British Library Conference Proceedings | 1999
Displacements of Multiblock Geotechnical Structures Subjected to Seismic Excitation
| Online Contents | 2007
Displacements of Multiblock Geotechnical Structures Subjected to Seismic Excitation
| British Library Online Contents | 2007