EI、Scopus 收录

 引用本文: 彭宗桓, 盛建龙, 叶祖洋, 李硕. 基于t-SNE-AMARS-MPM的空间变异性边坡可靠度分析及大变形破坏模式研究. 力学学报, 待出版.
Peng Zonghuan, Sheng Jianlong, Ye Zuyang, Li Shuo. Reliability analysis and investigation of large deformation failure modes in spatially variable slope using t-sne-amars-rmpm. Chinese Journal of Theoretical and Applied Mechanics, in press.
 Citation: Peng Zonghuan, Sheng Jianlong, Ye Zuyang, Li Shuo. Reliability analysis and investigation of large deformation failure modes in spatially variable slope using t-sne-amars-rmpm. Chinese Journal of Theoretical and Applied Mechanics, in press.

## RELIABILITY ANALYSIS AND INVESTIGATION OF LARGE DEFORMATION FAILURE MODES IN SPATIALLY VARIABLE SLOPE USING T-SNE-AMARS-RMPM

• 摘要: 在考虑参数空间变异性对大变形边坡可靠度评估的影响, 特别是在小失效概率的情况下, 传统分析方法常因耗时冗长或难以评估大变形失稳破坏而受限. 针对这一问题, 提出一种结合t分布-随机邻近嵌入、主动学习多元自适应回归样条法和物质点法的新方法, 并深入探讨了t分布-随机邻近嵌入对代理模型的优化作用. 首先, 利用Cholesky分解法离散边坡参数随机场, 并通过确定性分析获取训练样本. 随后, 采用主动学习函数优化代理模型, 高效求解边坡安全系数并按升序排列. 最后, 使用物质点法依次模拟并获取失效样本. 以一个两层不排水土坡为例, 验证了该方法的有效性. 结果表明, 所提方法相较于传统随机物质点法, 计算量仅为其1.64%, 显著提高了计算效率. 同时, 通过对失效样本的深入剖析, 识别出边坡失稳具有4种不同的破坏模式, 这些破坏模式的演化过程与土体参数的空间分布紧密相关. 尤为值得注意的是, 多层渐进破坏模式作为其中破坏过程最为复杂的类型, 其对周边环境的威胁尤为显著, 需要重点防范. 这为边坡工程的风险评估和加固提供了重要依据.

Abstract: The reliability analysis of slopes is often hindered by time-consuming or challenging evaluations of large deformation, particularly when dealing with the intricate effects of spatial variability in soil parameters on large deformations. To overcome these obstacles, this study introduces an innovative methodology that seamlessly integrates t-distributed stochastic neighbor embedding (t-SNE), active learning multiple adaptive regression spline (AMARS), and the material point method (MPM). At the core of this novel approach, Cholesky decomposition serves as a crucial tool for discretizing the complex random fields of slope parameters, thereby facilitating subsequent deterministic analysis to generate essential training samples. These samples serve as the foundation upon which the MARS model is constructed and further refined through employing an active learning function and construct AMARS model to ensure optimization and adaptability. Subsequently, leveraging Monte Carlo simulation (MCS), augmented by AMARS model delivers reliable estimates of slope stability. This integration provides a robust framework for quantifying uncertainties and predicting the likelihood of slope failures under varying conditions. In order to gain deeper insights into failure mechanisms, meticulous examination using material point method (MPM) is employed to analyze failure samples and unravel intricate dynamic evolution processes associated with diverse failure modes. This comprehensive analysis, demonstrated through a two-layer cohesive soil slope example not only enhances our theoretical understanding but also offers practical insights for real-world applications. Remarkably, results demonstrate that our proposed approach significantly outperforms random material point method (RMPM) with an impressive computational cost reduction rate of 1.64%. It is notably noteworthy that the multi-layer progressive failure mode, being the most intricate and complex of all failure processes, poses a significantly considerable threat to the adjacent environment, thereby necessitating urgent and heightened attention to ensure safety and mitigate potential hazards. This comprehensive study provides a crucial basis for the rigorous assessment and reinforcement of slope stability risks.

/

• 分享
• 用微信扫码二维码

分享至好友和朋友圈