Scholarlink

<p>This study introduces an optimized CatBoost model for predicting the mechanical properties—specifically unconfined compressive strength (UCS) and deformation modulus—of cement-stabilized soils enhanced with calcium or sodium bentonite additives. Systematic laboratory experiments were conducted on 300 specimens with variations in bentonite type, content ratios (0–10%), cement dosages (3–20%), and curing ages (7–90 days). These experiments enabled us to establish quantitative structure–property relationships that connect the composite’s chemical composition with its macroscopic performance. The machine learning (ML) framework demonstrates exceptional predictive accuracy (R² &gt;0.94) by capturing non-linear interactions between bentonite’s cation exchange characteristics, cement hydration kinetics, and time-dependent hardening processes. Shapley Additive exPlanations (SHAP) analysis reveals that bentonite type contributes 18.7% to strength prediction, while cement content dominates with 39.2% influence via Calcium Silicate Hydrate (CSH) gel formation. This computational approach bridges material composition parameters with engineering performance, providing a predictive tool that advances traditional empirical models through explicit consideration of bentonite’s ionic characteristics and curing-phase transformations.</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

Predictive Modeling of Cement-Bentonite Soil Composites: A CatBoost Approach Integrating Compositional and Curing Parameters

  • 期刊论文

  • 发表日期：

    2025年12月22日

查看全文
  • Xiaohan Zhang,
  • Zheyuan Feng,
  • Linjian Ma,
  • Yu Zhou,
  • Hansheng Geng,
  • Jicheng Liu

摘要

This study introduces an optimized CatBoost model for predicting the mechanical properties—specifically unconfined compressive strength (UCS) and deformation modulus—of cement-stabilized soils enhanced with calcium or sodium bentonite additives. Systematic laboratory experiments were conducted on 300 specimens with variations in bentonite type, content ratios (0–10%), cement dosages (3–20%), and curing ages (7–90 days). These experiments enabled us to establish quantitative structure–property relationships that connect the composite’s chemical composition with its macroscopic performance. The machine learning (ML) framework demonstrates exceptional predictive accuracy (R² >0.94) by capturing non-linear interactions between bentonite’s cation exchange characteristics, cement hydration kinetics, and time-dependent hardening processes. Shapley Additive exPlanations (SHAP) analysis reveals that bentonite type contributes 18.7% to strength prediction, while cement content dominates with 39.2% influence via Calcium Silicate Hydrate (CSH) gel formation. This computational approach bridges material composition parameters with engineering performance, providing a predictive tool that advances traditional empirical models through explicit consideration of bentonite’s ionic characteristics and curing-phase transformations.