Scholarlink

<p>Soil compaction is an extensively utilized technique for stabilizing soils by increasing their density through mechanical energy. Soil compaction is typically evaluated using two key parameters: OMC and MDD. Laboratory tests, known as Proctor tests, are used to obtain these two vital parameters. The modified Proctor test is just a variation of the standard Proctor test, often used for projects involving heavy traffic loads. Conducting the Proctor test is necessary, but it requires large volumes of soil and is time-consuming, especially for determining modified Proctor parameters. Therefore, predicting modified MDD and OMC from physical soil properties can be advantageous for preliminary soil assessment. This study aims to predict modified Proctor compaction parameters using several multivariable and machine-learning models, including LR, MLR, NLR, PQ, IN, FQ, M5P tree, and ANN. The study compiled 1715 datasets from previous studies, covering six independent soil properties: G, S, F, LL, PL, and PI. The models were assessed employing several statistical metrics, such as R<sup>2</sup>, RMSE, MAE, WMAPE, SI, a20-index, VAF, and OBJ. Based on the analysis, the ANN model consistently provided superior predictions for both OMC and MDD, with R<sup>2</sup> values of 0.9551 for training and 0.9830 for testing for OMC, and 0.9400 for training and 0.9833 for testing for MDD. Finally, sensitivity analysis was conducted to determine the parameter most influenced by MDD and OMC. Gravel content was the dominant factor influencing MDD, while OMC was primarily affected by the plasticity index.</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 techniques and multivariable mathematical models for predicting modified soil compaction parameters based on particle size and consistency limits

  • 期刊论文

  • 发表日期：

    2025年1月8日

查看全文
  • Bashdar Omer

摘要

Soil compaction is an extensively utilized technique for stabilizing soils by increasing their density through mechanical energy. Soil compaction is typically evaluated using two key parameters: OMC and MDD. Laboratory tests, known as Proctor tests, are used to obtain these two vital parameters. The modified Proctor test is just a variation of the standard Proctor test, often used for projects involving heavy traffic loads. Conducting the Proctor test is necessary, but it requires large volumes of soil and is time-consuming, especially for determining modified Proctor parameters. Therefore, predicting modified MDD and OMC from physical soil properties can be advantageous for preliminary soil assessment. This study aims to predict modified Proctor compaction parameters using several multivariable and machine-learning models, including LR, MLR, NLR, PQ, IN, FQ, M5P tree, and ANN. The study compiled 1715 datasets from previous studies, covering six independent soil properties: G, S, F, LL, PL, and PI. The models were assessed employing several statistical metrics, such as R2, RMSE, MAE, WMAPE, SI, a20-index, VAF, and OBJ. Based on the analysis, the ANN model consistently provided superior predictions for both OMC and MDD, with R2 values of 0.9551 for training and 0.9830 for testing for OMC, and 0.9400 for training and 0.9833 for testing for MDD. Finally, sensitivity analysis was conducted to determine the parameter most influenced by MDD and OMC. Gravel content was the dominant factor influencing MDD, while OMC was primarily affected by the plasticity index.