Scholarlink

<p>Calcareous sands often exhibit low bearing capacity and significant particle crushing which limits their use as coastal foundation materials. Stabilization with colloidal silica offers a promising alternative to conventional binders, yet the undrained strength response of treated sand remains highly nonlinear due to complex interactions between bond formation and fabric rearrangement. This study developed a machine learning framework to predict the peak deviatoric stress of these materials under consolidated undrained triaxial loading. The model utilizes four inputs: colloidal silica content, relative density, curing age, and effective confining pressure. Several algorithms were compared to capture nonlinear dependencies among variables. Results show that nonlinear models provide superior accuracy compared to traditional regressions, with artificial neural networks achieving the most consistent performance. Explainable AI analysis identifies effective confining pressure as the governing factor for peak strength, while relative density plays a significant secondary role. Furthermore, the model reveals an optimal colloidal silica dosage near 5%. The proposed framework supports performance oriented design by combining accurate prediction with interpretable trends and uncertainty aware decision support.</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

Machine Learning–Driven Prediction and Interpretation of Peak Undrained Strength in Colloidal Silica–Stabilized Calcareous Sands

  • 期刊论文

  • 发表日期：

    2026年7月10日

查看全文
  • Seyed Vahid Mousavi,
  • Adel Asakereh,
  • Abdolhosein Haddad

摘要

Calcareous sands often exhibit low bearing capacity and significant particle crushing which limits their use as coastal foundation materials. Stabilization with colloidal silica offers a promising alternative to conventional binders, yet the undrained strength response of treated sand remains highly nonlinear due to complex interactions between bond formation and fabric rearrangement. This study developed a machine learning framework to predict the peak deviatoric stress of these materials under consolidated undrained triaxial loading. The model utilizes four inputs: colloidal silica content, relative density, curing age, and effective confining pressure. Several algorithms were compared to capture nonlinear dependencies among variables. Results show that nonlinear models provide superior accuracy compared to traditional regressions, with artificial neural networks achieving the most consistent performance. Explainable AI analysis identifies effective confining pressure as the governing factor for peak strength, while relative density plays a significant secondary role. Furthermore, the model reveals an optimal colloidal silica dosage near 5%. The proposed framework supports performance oriented design by combining accurate prediction with interpretable trends and uncertainty aware decision support.