Scholarlink

<p>Predicting the real three-dimensional (3D) crack parameters in rock-like materials based on acoustic emission (AE) data remains both challenging and crucial. In this study, an AE monitoring system, CCD cameras integrated with SAM, and 3D Generative AI are employed to directly observe and generate real 3D crack in transparent resin specimens under uniaxial loading, while datasets for training and testing neural network extracted and characterized via Data Analyzer and Segmented Zero-Order Mapping. What is more, we develop PQNF, a novel neural network framework that incorporates Cross-modal Alignment Strategy and Dimension-Adaptive Selective State-Space Block. Unlike existing approaches that rely on manually curated datasets or impose prior knowledge of physics, PQNF trains directly with raw experimental data and saves computational resources. PQNF predicts real 3D crack parameters based on AE data, achieving <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({R}^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mrow> <mi>R</mi> </mrow> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> of approximately 0.9170 (location), 0.9404 (orientation), 0.9802 (volume), 0.9805 (surface area), and 0.8486 (thickness), while also enabling the visual reconstruction of the real 3D crack and prediction of instability index based on stress, strain, and crack parameters. This purely data-driven approach offers a powerful tool for crack monitoring and early warning in engineering applications, and it opens new avenues for research in data-driven damage prediction.</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

PQNF: Parameter Quantitative Prediction Neural Network Framework for Real 3D Crack in Rock-Like Material Specimen Based on Acoustic Emission Monitoring Data

  • 期刊论文

  • 发表日期：

    2026年4月2日

查看全文
  • Zinan Du,
  • Enyuan Wang,
  • Liang Yuan,
  • Benyu Su,
  • Zhonghui Li,
  • Mohamed Elgharib Gomah

摘要

Predicting the real three-dimensional (3D) crack parameters in rock-like materials based on acoustic emission (AE) data remains both challenging and crucial. In this study, an AE monitoring system, CCD cameras integrated with SAM, and 3D Generative AI are employed to directly observe and generate real 3D crack in transparent resin specimens under uniaxial loading, while datasets for training and testing neural network extracted and characterized via Data Analyzer and Segmented Zero-Order Mapping. What is more, we develop PQNF, a novel neural network framework that incorporates Cross-modal Alignment Strategy and Dimension-Adaptive Selective State-Space Block. Unlike existing approaches that rely on manually curated datasets or impose prior knowledge of physics, PQNF trains directly with raw experimental data and saves computational resources. PQNF predicts real 3D crack parameters based on AE data, achieving \({R}^{2}\) R 2 of approximately 0.9170 (location), 0.9404 (orientation), 0.9802 (volume), 0.9805 (surface area), and 0.8486 (thickness), while also enabling the visual reconstruction of the real 3D crack and prediction of instability index based on stress, strain, and crack parameters. This purely data-driven approach offers a powerful tool for crack monitoring and early warning in engineering applications, and it opens new avenues for research in data-driven damage prediction.