A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Spectrum-matched ground motion selection method based on Siamese Convolutional Neural Networks
https://orcid.org/0000-0001-6995-2761
https://orcid.org/0000-0002-8410-5584
Abstract Selecting spectrum-matched ground motions is one of the critical problems in the seismic analysis and design of structures. The sum of the square errors (SSE) has been commonly used to quantitatively measure the similarity between the response spectrum of a ground motion and a target spectrum. However, it cannot represent the two-dimensional plane feature of the response spectrum because the response spectrum is treated as a one-dimensional vector in the SSE-based approaches. Considering the high-performance feature extraction of images and small-sample learning for the Siamese Convolutional Neural Networks (SCNNs), a new spectrum-matched ground motion selection method is proposed based on the SCNNs. The types of ground A, B, C, and D acceleration design spectra for the Eurocode 8 are selected as the target spectra without loss of generality. The procedures for the sample image generation, network training, testing, and spectrum-matched ground motion selection are elaborated. Analysis results indicate that using 40 samples for each class can achieve satisfactory training results. The mean response spectrum of the ground motions selected by the proposed method is matched with the target spectrum in all periods. Compared to the SSE-based approaches, the results obtained by the proposed method have a minor standard deviation. The proposed method could be used as an alternative in the ground motion selection for the dynamic analysis of structures.
Highlights Spectrum-matched ground motion selection based on the SCNNs. The inputs are the response spectral images. Sample generation, network training, and testing are studied. The mean spectrum is matched with the target spectrum in all periods. Compared to SSE-based approaches, the results have a minor standard deviation.
Spectrum-matched ground motion selection method based on Siamese Convolutional Neural Networks
https://orcid.org/0000-0001-6995-2761
https://orcid.org/0000-0002-8410-5584
Abstract Selecting spectrum-matched ground motions is one of the critical problems in the seismic analysis and design of structures. The sum of the square errors (SSE) has been commonly used to quantitatively measure the similarity between the response spectrum of a ground motion and a target spectrum. However, it cannot represent the two-dimensional plane feature of the response spectrum because the response spectrum is treated as a one-dimensional vector in the SSE-based approaches. Considering the high-performance feature extraction of images and small-sample learning for the Siamese Convolutional Neural Networks (SCNNs), a new spectrum-matched ground motion selection method is proposed based on the SCNNs. The types of ground A, B, C, and D acceleration design spectra for the Eurocode 8 are selected as the target spectra without loss of generality. The procedures for the sample image generation, network training, testing, and spectrum-matched ground motion selection are elaborated. Analysis results indicate that using 40 samples for each class can achieve satisfactory training results. The mean response spectrum of the ground motions selected by the proposed method is matched with the target spectrum in all periods. Compared to the SSE-based approaches, the results obtained by the proposed method have a minor standard deviation. The proposed method could be used as an alternative in the ground motion selection for the dynamic analysis of structures.
Highlights Spectrum-matched ground motion selection based on the SCNNs. The inputs are the response spectral images. Sample generation, network training, and testing are studied. The mean spectrum is matched with the target spectrum in all periods. Compared to SSE-based approaches, the results have a minor standard deviation.
Spectrum-matched ground motion selection method based on Siamese Convolutional Neural Networks
https://orcid.org/0000-0001-6995-2761
https://orcid.org/0000-0002-8410-5584
Abstract Selecting spectrum-matched ground motions is one of the critical problems in the seismic analysis and design of structures. The sum of the square errors (SSE) has been commonly used to quantitatively measure the similarity between the response spectrum of a ground motion and a target spectrum. However, it cannot represent the two-dimensional plane feature of the response spectrum because the response spectrum is treated as a one-dimensional vector in the SSE-based approaches. Considering the high-performance feature extraction of images and small-sample learning for the Siamese Convolutional Neural Networks (SCNNs), a new spectrum-matched ground motion selection method is proposed based on the SCNNs. The types of ground A, B, C, and D acceleration design spectra for the Eurocode 8 are selected as the target spectra without loss of generality. The procedures for the sample image generation, network training, testing, and spectrum-matched ground motion selection are elaborated. Analysis results indicate that using 40 samples for each class can achieve satisfactory training results. The mean response spectrum of the ground motions selected by the proposed method is matched with the target spectrum in all periods. Compared to the SSE-based approaches, the results obtained by the proposed method have a minor standard deviation. The proposed method could be used as an alternative in the ground motion selection for the dynamic analysis of structures.
Highlights Spectrum-matched ground motion selection based on the SCNNs. The inputs are the response spectral images. Sample generation, network training, and testing are studied. The mean spectrum is matched with the target spectrum in all periods. Compared to SSE-based approaches, the results have a minor standard deviation.
Spectrum-matched ground motion selection method based on Siamese Convolutional Neural Networks
Zhao, Guochen (author) / Xu, Longjun (author) / Zhu, Xingji (author) / Lin, Shibin (author) / Xie, Lili (author)
2022-08-24
Article (Journal)
Electronic Resource
English
Siamese Convolutional Neural Networks to Quantify Crack Pattern Similarity in Masonry Facades
| Taylor & Francis Verlag | 2023
Convolutional autoencoder-based ground motion clustering and selection
| Elsevier | 2025