Scholarlink

This study develops data-driven approaches to enhance the Engineering News Record (ENR) dynamic formula for pile bearing capacity estimation using Pile Driving Analyzer (PDA) measurements. Through analysis of 84 PDA test cases incorporating maximum transferred energy (EMX), permanent set (S), and elastic compression (C), Multiple Linear Regression (MLR) and Random Forest (RF) models were developed. The MLR approach yielded a reformulated version of the ENR dynamic formula, in which the coefficient for parameter S was adjusted to 1.09 and for C to 0.67, resulting in improved predictive accuracy (R2 = 0.96, MAPE = 19.26%). Superior performance was observed in the RF model, achieving very accurate training (MAPE = 7.66%) and accurate testing results (R2 = 0.93, MAPE = 10.39%). Comparative analysis confirms both methods effectively predict CAPWAP-derived capacities, with RF exhibiting better generalization through ensemble learning. The results demonstrate that statistical optimization can refine empirical formulas, while machine learning captures complex nonlinear relationships in driving data. These findings provide practical frameworks for improving dynamic formula reliability when PDA records are available, advancing geotechnical practical through hybrid data-driven and fundamental approaches.

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 Prediction of Pile Capacity Based on EMX Value from Dynamic Test

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ali Iskandar,
  • Agustinus Purna Irawan,
  • Aksan Kawanda,
  • Daniel Christianto,
  • Sherlin Angelina

摘要

This study develops data-driven approaches to enhance the Engineering News Record (ENR) dynamic formula for pile bearing capacity estimation using Pile Driving Analyzer (PDA) measurements. Through analysis of 84 PDA test cases incorporating maximum transferred energy (EMX), permanent set (S), and elastic compression (C), Multiple Linear Regression (MLR) and Random Forest (RF) models were developed. The MLR approach yielded a reformulated version of the ENR dynamic formula, in which the coefficient for parameter S was adjusted to 1.09 and for C to 0.67, resulting in improved predictive accuracy (R2 = 0.96, MAPE = 19.26%). Superior performance was observed in the RF model, achieving very accurate training (MAPE = 7.66%) and accurate testing results (R2 = 0.93, MAPE = 10.39%). Comparative analysis confirms both methods effectively predict CAPWAP-derived capacities, with RF exhibiting better generalization through ensemble learning. The results demonstrate that statistical optimization can refine empirical formulas, while machine learning captures complex nonlinear relationships in driving data. These findings provide practical frameworks for improving dynamic formula reliability when PDA records are available, advancing geotechnical practical through hybrid data-driven and fundamental approaches.