Scholarlink

<p>Velocity imaging is essential for identifying geological anomalies and enabling proactive safety monitoring in mines. However, traditional tomographic methods face challenges of low computational efficiency and machine learning-based methods have&#xa0;difficulty in handling complex mine geology. To overcome these limitations, this study proposes a tomography approach by combining the fast marching method (FMM) and convolutional neural networks (CNNs). A forward P-wave travel time database was first constructed based on FMM and 3D velocity models, where the reciprocity theory was adopted to efficiently generate the travel time dataset. A U-Net model tailored for sparse and highly uneven ray path coverage was built and then trained on extensive, geology-constrained synthetic data to construct an end-to-end mapping from travel times to a 3D velocity model. Synthetic tests indicate that even in the presence of 20 ms Gaussian noise on travel times, most inverted velocity errors are below 100&#xa0;m/s, demonstrating strong noise robustness. Once trained, this model produces velocity imaging results in approximately 300&#xa0;ms, enabling real-time velocity imaging. In the engineering application, Empirical Bayesian Kriging&#xa0;(EBK) is used to interpolate measured microseismic travel times to transform spatially discrete and irregularly distributed field travel time observations into a standard, CNN-ready travel time dataset. The travel time interpolation errors (0.009&#xa0;s–0.034&#xa0;s) are smaller than most sensor-to-node travel times (0.1494&#xa0;s–0.5138&#xa0;s), demonstrating high accuracy of the interpolation method. The inverted velocity model reveals a low-velocity belt along the orebody, with several surrounding&#xa0;high-velocity zones. This distribution aligns well with the mining plan and indicates that the proposed method is effective for delineating underground structures in mines.</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

Three-Dimensional Microseismic Tomography of Mines Based on Fast Marching Method and Convolutional Neural Networks

  • 期刊论文

  • 发表日期：

    2026年6月3日

查看全文
  • Xueyi Shang,
  • Zheli Wu,
  • Zhaopeng Liu,
  • Yi Wang,
  • Jie Yang

摘要

Velocity imaging is essential for identifying geological anomalies and enabling proactive safety monitoring in mines. However, traditional tomographic methods face challenges of low computational efficiency and machine learning-based methods have difficulty in handling complex mine geology. To overcome these limitations, this study proposes a tomography approach by combining the fast marching method (FMM) and convolutional neural networks (CNNs). A forward P-wave travel time database was first constructed based on FMM and 3D velocity models, where the reciprocity theory was adopted to efficiently generate the travel time dataset. A U-Net model tailored for sparse and highly uneven ray path coverage was built and then trained on extensive, geology-constrained synthetic data to construct an end-to-end mapping from travel times to a 3D velocity model. Synthetic tests indicate that even in the presence of 20 ms Gaussian noise on travel times, most inverted velocity errors are below 100 m/s, demonstrating strong noise robustness. Once trained, this model produces velocity imaging results in approximately 300 ms, enabling real-time velocity imaging. In the engineering application, Empirical Bayesian Kriging (EBK) is used to interpolate measured microseismic travel times to transform spatially discrete and irregularly distributed field travel time observations into a standard, CNN-ready travel time dataset. The travel time interpolation errors (0.009 s–0.034 s) are smaller than most sensor-to-node travel times (0.1494 s–0.5138 s), demonstrating high accuracy of the interpolation method. The inverted velocity model reveals a low-velocity belt along the orebody, with several surrounding high-velocity zones. This distribution aligns well with the mining plan and indicates that the proposed method is effective for delineating underground structures in mines.