Scholarlink

<p>Deep excavations constructed near existing buildings are prone to stability degradation due to strongly coupled soil–structure interactions, yet the governing failure mechanisms and dominant controlling factors remain insufficiently understood. This study investigates a deep excavation adjacent to existing buildings in Langfang, China, by integrating finite-element limit analysis (FELA) with explainable machine learning. Two-dimensional plane-strain models are developed in OptumG2 to quantify stability and failure modes under varying building loads and geometric configurations, with upper- and lower-bound solutions ensuring numerical robustness. The analyses reveal a clear mechanism transition from localized retaining-wall deformation to global shear sliding involving both the excavation and building foundations as structural loading increases. Threshold effects of building distance and width are identified, beyond which geometric influences become marginal. Parametric results further show that pile length and dewatering depth are the most effective stability-enhancing measures, whereas pile spacing plays a secondary role. An XGBoost model interpreted using SHapley Additive exPlanations (SHAP) is then established based on FELA-derived safety factors. The feature attributions consistently rank building load as the most influential factor, followed by excavation geometry, pile configuration, and dewatering parameters, suggesting that the trained surrogate captures variable influence trends broadly consistent with the mechanism-based observations from the FELA simulations. The proposed FELA-based XGBoost–SHAP framework offers a mechanism-consistent and interpretable approach for excavation assessment and design support within the investigated parameter space.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From Failure Mechanisms to Interpretable Assessment: Stability Degradation of Deep Excavations Near Existing Buildings

  • 期刊论文

  • 发表日期：

    2026年5月7日

查看全文
  • Zheng Liu,
  • Ting Yu,
  • Wei Li,
  • Jie Dou,
  • Shuoliangyuan Liu,
  • Luqiang Ding

摘要

Deep excavations constructed near existing buildings are prone to stability degradation due to strongly coupled soil–structure interactions, yet the governing failure mechanisms and dominant controlling factors remain insufficiently understood. This study investigates a deep excavation adjacent to existing buildings in Langfang, China, by integrating finite-element limit analysis (FELA) with explainable machine learning. Two-dimensional plane-strain models are developed in OptumG2 to quantify stability and failure modes under varying building loads and geometric configurations, with upper- and lower-bound solutions ensuring numerical robustness. The analyses reveal a clear mechanism transition from localized retaining-wall deformation to global shear sliding involving both the excavation and building foundations as structural loading increases. Threshold effects of building distance and width are identified, beyond which geometric influences become marginal. Parametric results further show that pile length and dewatering depth are the most effective stability-enhancing measures, whereas pile spacing plays a secondary role. An XGBoost model interpreted using SHapley Additive exPlanations (SHAP) is then established based on FELA-derived safety factors. The feature attributions consistently rank building load as the most influential factor, followed by excavation geometry, pile configuration, and dewatering parameters, suggesting that the trained surrogate captures variable influence trends broadly consistent with the mechanism-based observations from the FELA simulations. The proposed FELA-based XGBoost–SHAP framework offers a mechanism-consistent and interpretable approach for excavation assessment and design support within the investigated parameter space.