Scholarlink

<p>Machine learning has emerged as a practical tool in geotechnical engineering, fueled by the large amount of experimental datasets accumulated over decades. This study develops a machine learning framework to (i) identify a density measure that most effectively normalizes fines effects in silty sands and (ii) predict cyclic resistance (CSR–N) curves. A literature dataset of 530 cyclic triaxial tests at σ′<sub>3</sub> = 100&#xa0;kPa on moist‑tamped silty sands was compiled. Feature selection consistently identified relative compaction (R), log N, and void-ratio range (e<sub>range</sub>) as the most informative predictors. Gaussian Process Regression provided the best cross‑validated performance (R<sup>2</sup> = 0.88; RMSE = 0.049; MAE = 0.033). Validation indicates larger material-specific errors of up to 25%, likely due to differences in grain shape or fabric among sands. The findings suggest that relative compaction outperforms relative density and intergranular void ratio as a fines‑normalizing metric for silty sands and that GPR offers a transparent, probabilistic approach for CSR–N prediction.</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

A machine learning framework for evaluating cyclic resistance of sands with non-plastic fines

  • 期刊论文

  • 发表日期：

    2026年5月13日

查看全文
  • Arslan Mushtaq,
  • Muhammad Salman,
  • Muhammad Noman,
  • Muhammad Faizan,
  • Afaq Ahmed

摘要

Machine learning has emerged as a practical tool in geotechnical engineering, fueled by the large amount of experimental datasets accumulated over decades. This study develops a machine learning framework to (i) identify a density measure that most effectively normalizes fines effects in silty sands and (ii) predict cyclic resistance (CSR–N) curves. A literature dataset of 530 cyclic triaxial tests at σ′3 = 100 kPa on moist‑tamped silty sands was compiled. Feature selection consistently identified relative compaction (R), log N, and void-ratio range (erange) as the most informative predictors. Gaussian Process Regression provided the best cross‑validated performance (R2 = 0.88; RMSE = 0.049; MAE = 0.033). Validation indicates larger material-specific errors of up to 25%, likely due to differences in grain shape or fabric among sands. The findings suggest that relative compaction outperforms relative density and intergranular void ratio as a fines‑normalizing metric for silty sands and that GPR offers a transparent, probabilistic approach for CSR–N prediction.