Scholarlink

<p>Assessment of pile bearing capacity reliability is critical for foundation safety; however, incorporating advanced predictive models into reliability frameworks remains limited. This study focused on integrating hybrid ANN-predicted bearing capacity into the limit-state function for first- and second-order reliability methods (FORM and SORM). Hybrid-ANN models were developed using Imperialistic Competitive Algorithm (ICA), Ant Colony Optimization, Antlion Optimization (ALO), and Particle Swarm Optimization. The ANN-ICA model demonstrated superior convergence with lower Root Mean Square Error (RMSE) in both training (0.0088) and testing (0.0111) phases. The hybrid ANN-ALO model exhibited faster convergence but poorer predictive performance, with RMSE values of 0.0339 in training and 0.0329 in testing. The comprehensive measure (COM) was used to rank models based on multiple performance criteria. ANN-ICA ranked first with the lowest COM (0.034). The piles exhibited a moderate reliability level under an average loading of 500kN, with a reliability index (β) of 3.911 and 5.33 for 300&#xa0;mm and 400&#xa0;mm piles, respectively, using FORM, and 3.803 and 5.244 using SORM. Based on the Mutual Information sensitivity analysis, soil parameters exerted greater influence on the model output (33.43%, 20.51%, and 16.64% for unit weight, friction angle, and undrained cohesion, respectively).</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

Direct Integration of Hybrid ANN Models into FORM/SORM for Reliability Analysis of Pile Foundations

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Karthikeyan M,
  • Manish Kumar

摘要

Assessment of pile bearing capacity reliability is critical for foundation safety; however, incorporating advanced predictive models into reliability frameworks remains limited. This study focused on integrating hybrid ANN-predicted bearing capacity into the limit-state function for first- and second-order reliability methods (FORM and SORM). Hybrid-ANN models were developed using Imperialistic Competitive Algorithm (ICA), Ant Colony Optimization, Antlion Optimization (ALO), and Particle Swarm Optimization. The ANN-ICA model demonstrated superior convergence with lower Root Mean Square Error (RMSE) in both training (0.0088) and testing (0.0111) phases. The hybrid ANN-ALO model exhibited faster convergence but poorer predictive performance, with RMSE values of 0.0339 in training and 0.0329 in testing. The comprehensive measure (COM) was used to rank models based on multiple performance criteria. ANN-ICA ranked first with the lowest COM (0.034). The piles exhibited a moderate reliability level under an average loading of 500kN, with a reliability index (β) of 3.911 and 5.33 for 300 mm and 400 mm piles, respectively, using FORM, and 3.803 and 5.244 using SORM. Based on the Mutual Information sensitivity analysis, soil parameters exerted greater influence on the model output (33.43%, 20.51%, and 16.64% for unit weight, friction angle, and undrained cohesion, respectively).