Scholarlink

<p>Reliable prediction of soil behavior is crucial for safe and cost-effective geotechnical design, where triaxial strength serves as a key indicator of mechanical performance. This study introduces a Pipefish-based Graph Neural Prediction System (PbGNPS) to model the complex nonlinear behavior of cohesive soils under loading conditions. Experimental data from unconsolidated undrained triaxial tests were collected and preprocessed to enhance data quality and predictive capability. A hybrid learning framework was then developed to train the model for predicting displacement, loading duration, failure area, failure strain, and deviator strain. The proposed system achieved high prediction accuracies of 97.726% for duration, 97.742% for displacement, 97.523% for failure area, 97.475% for failure strain, and 88.884% for deviator strain. The novelty of this work lies in integrating Pipefish optimization with graph neural networks to effectively capture complex soil behavior patterns. Practically, the proposed model reduces reliance on extensive laboratory testing and supports faster, more reliable decision-making in geotechnical design and construction.</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

A data-driven technique for predicting the triaxial behaviour of soils

  • 期刊论文

  • 发表日期：

    2026年5月13日

查看全文
  • Manisha Devidas Sawant,
  • Nilima Santosh Pote,
  • Ramatai Ganesh Pawar,
  • Anandrao R. Deshmukh,
  • Urmila Anil Mahadik,
  • Preeti Gajghate

摘要

Reliable prediction of soil behavior is crucial for safe and cost-effective geotechnical design, where triaxial strength serves as a key indicator of mechanical performance. This study introduces a Pipefish-based Graph Neural Prediction System (PbGNPS) to model the complex nonlinear behavior of cohesive soils under loading conditions. Experimental data from unconsolidated undrained triaxial tests were collected and preprocessed to enhance data quality and predictive capability. A hybrid learning framework was then developed to train the model for predicting displacement, loading duration, failure area, failure strain, and deviator strain. The proposed system achieved high prediction accuracies of 97.726% for duration, 97.742% for displacement, 97.523% for failure area, 97.475% for failure strain, and 88.884% for deviator strain. The novelty of this work lies in integrating Pipefish optimization with graph neural networks to effectively capture complex soil behavior patterns. Practically, the proposed model reduces reliance on extensive laboratory testing and supports faster, more reliable decision-making in geotechnical design and construction.