Scholarlink

<p>Accurate estimation of the ultimate bearing capacity of shallow foundations is essential for safe geotechnical design. Probabilistic analysis and data-driven modelling often suffer from limited datasets because field investigations and laboratory testing are costly and time-consuming. This study proposed a structured framework for generating large synthetic geotechnical datasets using a tabular diffusion-based generative model to address data scarcity in bearing capacity analysis. The model was trained using significant soil parameters, including cohesion, angle of internal friction, unit weight, and ultimate bearing capacity. Synthetic datasets containing 50,000, 100,000, and 150,000 samples were generated. For each generated sample, the ultimate bearing capacity was independently evaluated using a classical bearing-capacity formulation to ensure consistency with established soil-mechanics principles. The agreement between generated and analytically evaluated parameters was evaluated using parity analysis and the coefficient of determination, which indicated strong consistency while preserving realistic variability. Reliability analysis was also performed using a Monte Carlo simulation to estimate failure probability and reliability index. The results showed stable and convergent reliability estimates across all dataset sizes. The generated datasets improved the statistical representation of soil properties and provided a robust foundation for probabilistic analysis and future data-driven modelling in geotechnical engineering.</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

Generative Diffusion Model-based Synthetic Geotechnical Data Generation for Reliability Analysis of Ultimate Bearing Capacity of Shallow Foundations

  • 期刊论文

  • 发表日期：

    2026年4月15日

查看全文
  • Rudranarayan Rudranarayan,
  • Pijush Samui,
  • Sunita Kumari

摘要

Accurate estimation of the ultimate bearing capacity of shallow foundations is essential for safe geotechnical design. Probabilistic analysis and data-driven modelling often suffer from limited datasets because field investigations and laboratory testing are costly and time-consuming. This study proposed a structured framework for generating large synthetic geotechnical datasets using a tabular diffusion-based generative model to address data scarcity in bearing capacity analysis. The model was trained using significant soil parameters, including cohesion, angle of internal friction, unit weight, and ultimate bearing capacity. Synthetic datasets containing 50,000, 100,000, and 150,000 samples were generated. For each generated sample, the ultimate bearing capacity was independently evaluated using a classical bearing-capacity formulation to ensure consistency with established soil-mechanics principles. The agreement between generated and analytically evaluated parameters was evaluated using parity analysis and the coefficient of determination, which indicated strong consistency while preserving realistic variability. Reliability analysis was also performed using a Monte Carlo simulation to estimate failure probability and reliability index. The results showed stable and convergent reliability estimates across all dataset sizes. The generated datasets improved the statistical representation of soil properties and provided a robust foundation for probabilistic analysis and future data-driven modelling in geotechnical engineering.