Scholarlink

<p>This study aims to investigate the degradation mechanism of weakly expansive soil under loading and wetting–drying (W–D) cycles and to establish a corresponding constitutive model. Triaxial shear tests were conducted on specimens subjected to W–D cycles under different hydraulic pathways. The effects of cycle number, W–D cycles amplitude, and external loading on soil strength parameters were analyzed. Furthermore, a modified theoretical model and a machine learning-based constitutive relationship were established. Results indicate that W–D cycling induces soil deterioration, with reductions in peak shear strength and cohesion. Compared to cohesion, the internal friction angle exhibits lower sensitivity to cycling. The degradation effects intensify with decreasing loads and increasing W–D amplitudes. Fracture analysis reveals that structural damage serves as the primary mechanism for mechanical deterioration. Regression relationships between parameters of the Mohr–Coulomb and Duncan-Chang models and influencing factors were established using experimental data. A modified Duncan-Chang model (<i>R</i><sup>2</sup> = 0.90) was proposed after optimizing regression coefficients via a Genetic Algorithm. Concurrently, an artificial neural network (ANN)-based constitutive model (<i>R</i><sup>2</sup> = 0.99) was developed. By integrating global surrogate models (decision tree regressor) and local surrogate models (Local Interpretable Model-Agnostic Explanations), a comprehensive interpretability framework was achieved. Both theoretical and machine learning models effectively captured the stress–strain behavior of weak expansive soils, with the ANN model demonstrating superior accuracy. In addition, through hyperparameter tuning and surrogate modeling, machine learning frameworks can attain enhanced practicality and interpretability.</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

Constitutive Modeling and Mechanical Response of Weak Expansive Soils Under Loaded Wetting–Drying Cycles

  • 期刊论文

  • 发表日期：

    2025年10月31日

查看全文
  • Senwei Wang,
  • Weimin Yang,
  • Meixia Wang,
  • Shijie Ma,
  • Yifan Bai,
  • Cong Tian,
  • Lang Liu

摘要

This study aims to investigate the degradation mechanism of weakly expansive soil under loading and wetting–drying (W–D) cycles and to establish a corresponding constitutive model. Triaxial shear tests were conducted on specimens subjected to W–D cycles under different hydraulic pathways. The effects of cycle number, W–D cycles amplitude, and external loading on soil strength parameters were analyzed. Furthermore, a modified theoretical model and a machine learning-based constitutive relationship were established. Results indicate that W–D cycling induces soil deterioration, with reductions in peak shear strength and cohesion. Compared to cohesion, the internal friction angle exhibits lower sensitivity to cycling. The degradation effects intensify with decreasing loads and increasing W–D amplitudes. Fracture analysis reveals that structural damage serves as the primary mechanism for mechanical deterioration. Regression relationships between parameters of the Mohr–Coulomb and Duncan-Chang models and influencing factors were established using experimental data. A modified Duncan-Chang model (R2 = 0.90) was proposed after optimizing regression coefficients via a Genetic Algorithm. Concurrently, an artificial neural network (ANN)-based constitutive model (R2 = 0.99) was developed. By integrating global surrogate models (decision tree regressor) and local surrogate models (Local Interpretable Model-Agnostic Explanations), a comprehensive interpretability framework was achieved. Both theoretical and machine learning models effectively captured the stress–strain behavior of weak expansive soils, with the ANN model demonstrating superior accuracy. In addition, through hyperparameter tuning and surrogate modeling, machine learning frameworks can attain enhanced practicality and interpretability.