Reliability Analyses of Soil Slopes with Multiple Spatially Varying Parameters Using Multi-Input Convolutional Neural Networks
文献类型：期刊论文
| 作者 | Liu, Yibiao1,2; Ren, Weizhong1,2 |
| 刊名 | INTERNATIONAL JOURNAL OF GEOMECHANICS
 |
| 出版日期 | 2023-07-01 |
| 卷号 | 23期号:7页码:12 |
| ISSN号 | 1532-3641 |
| DOI | 10.1061/IJGNAI.GMENG-8234 |
| 英文摘要 | Constructing meta-models and selecting a suitable deterministic analysis method are important to improve the computational efficiency and accuracy of the nonintrusive reliability analysis of a spatially varying soil slope. However, existing meta-models are not applicable to the slopes considering multiple parameters with high spatial variability. Moreover, it is difficult to identify the failure modes when the spatial variability is high by using deterministic analysis methods based on slip surface search. Therefore, a nonintrusive stochastic strength reduction finite-element method (SRFEM) is developed based on the multi-input convolution neural networks (CNNs) and ABAQUS 2016. The SRFEM developed based on ABAQUS is adopted as the deterministic analysis method to avoid the uncertain search for the critical slip surfaces of slopes with high spatial variability. A multi-input CNN is proposed to construct the meta-model to avoid the "curse of dimensionality" and replace the overmuch times of time-consuming finite-element simulations. It can fit the relationships between multiple spatially varying parameters and the factor of safety by processing different parameters with different streams of CNNs. Two illustrative examples show that the proposed method can accurately identify the failure modes of slopes with different degrees of spatial variability. The agreement of the reliability results based on the proposed method and the general random finite-element method (RFEM) shows the high accuracy of the proposed method. The time cost of the proposed method can be reduced to 6.0 x 10(-3) times that of the general RFEM, verifying the high computational efficiency of the proposed method. The multi-input CNN also shows higher fitting accuracy and better interpretability than the single-stream CNN and the support vector machines (SVMs). The generalization ability, accuracy, and efficiency of the proposed method show its potential to carry out the reliability analyses of slopes with multiple spatially varying parameters. |
| 资助项目 | Science and Technology Project of Hubei Province, China[2015BKA223] |
| WOS研究方向 | Engineering |
| 语种 | 英语 |
| WOS记录号 | WOS:001026635600026 |
| 出版者 | ASCE-AMER SOC CIVIL ENGINEERS |
| 源URL | [http://119.78.100.198/handle/2S6PX9GI/39088]  |
| 专题 | 中科院武汉岩土力学所 |
| 通讯作者 | Liu, Yibiao |
| 作者单位 | 1.Chinese Acad Sci, Inst Rock & Soil Mech, Wuhan 430071, Peoples R China 2.Univ Chinese Acad Sci, Beijing 100049, Peoples R China |
| 推荐引用方式 GB/T 7714 | Liu, Yibiao,Ren, Weizhong. Reliability Analyses of Soil Slopes with Multiple Spatially Varying Parameters Using Multi-Input Convolutional Neural Networks[J]. INTERNATIONAL JOURNAL OF GEOMECHANICS,2023,23(7):12. |
| APA | Liu, Yibiao,&Ren, Weizhong.(2023).Reliability Analyses of Soil Slopes with Multiple Spatially Varying Parameters Using Multi-Input Convolutional Neural Networks.INTERNATIONAL JOURNAL OF GEOMECHANICS,23(7),12. |
| MLA | Liu, Yibiao,et al."Reliability Analyses of Soil Slopes with Multiple Spatially Varying Parameters Using Multi-Input Convolutional Neural Networks".INTERNATIONAL JOURNAL OF GEOMECHANICS 23.7(2023):12. |
入库方式： OAI收割
来源：武汉岩土力学研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。