Scholarlink

<p>This study examines the resilient modulus for cohesive soils using laboratory experiments and computational modeling, aiming to create reliable predictive models for resilient modulus. A dataset of 1,023 laboratory-tested samples was employed to develop three separate models utilizing multiple linear regression (MLR), artificial neural networks (ANN), and genetic algorithm approaches (GA). The input variables comprised gradational and physical properties (percentage passing the #200 sieve, plasticity index), compaction characteristics (optimum moisture content, degree of saturation), and confining pressure, deviator stress, with output variable being the resilient modulus. The ANN architecture showed enhanced predictive accuracy, obtaining a coefficient of determination (<i>R</i><sup>2</sup>) of 0.95 in both training and validation phases, exceeding the performance of MLR (<i>R</i><sup>2</sup> = 0.62) and GA-based models (<i>R</i><sup>2</sup> = 0.69). The sensitivity analysis showed that the degree of saturation (Sr) is the most significant parameter. Empirical models were developed from ANN and GA frameworks, providing efficient methods for estimating (MR) in situations where extensive laboratory testing is not feasible. Twenty subgrade soil samples were tested to find the resilient modulus et al.-Balqa Applied University laboratory and compared with ANN &amp; GA models. ANN achieved a more accurate prediction (<i>R</i><sup>2</sup> = 0.94) than GA (<i>R</i><sup>2</sup> = 0.73).</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

Development of Empirical Equations for Estimating Subgrade Resilient Modulus Using Machine Learning Methods and Experimental Validation

  • 期刊论文

  • 发表日期：

    2025年10月2日

查看全文
  • Shadi Hanandeh,
  • Saeed Alzahrani,
  • Zaid Alajlan,
  • Ahmad Hanandeh,
  • Frank Aneke

摘要

This study examines the resilient modulus for cohesive soils using laboratory experiments and computational modeling, aiming to create reliable predictive models for resilient modulus. A dataset of 1,023 laboratory-tested samples was employed to develop three separate models utilizing multiple linear regression (MLR), artificial neural networks (ANN), and genetic algorithm approaches (GA). The input variables comprised gradational and physical properties (percentage passing the #200 sieve, plasticity index), compaction characteristics (optimum moisture content, degree of saturation), and confining pressure, deviator stress, with output variable being the resilient modulus. The ANN architecture showed enhanced predictive accuracy, obtaining a coefficient of determination (R2) of 0.95 in both training and validation phases, exceeding the performance of MLR (R2 = 0.62) and GA-based models (R2 = 0.69). The sensitivity analysis showed that the degree of saturation (Sr) is the most significant parameter. Empirical models were developed from ANN and GA frameworks, providing efficient methods for estimating (MR) in situations where extensive laboratory testing is not feasible. Twenty subgrade soil samples were tested to find the resilient modulus et al.-Balqa Applied University laboratory and compared with ANN & GA models. ANN achieved a more accurate prediction (R2 = 0.94) than GA (R2 = 0.73).