Scholarlink

<p>Lithology identification stands as a core objective in geological exploration. Traditional methods suffer from strong subjectivity and low efficiency. This paper innovatively proposes an intelligent identification model integrating an enhanced AlexNet with ensemble learning: First, sound and triaxial vibration signals from drilling were acquired and optimized using a 2-second observation window. Subsequently, a novel AlexNet-RCBAM base model incorporating Residual structure (R) and Convolutional Block Attention Modules (CBAM) was constructed, demonstrating significantly superior accuracy compared to traditional machine learning algorithms. Finally, demonstrating significantly superior accuracy the EM-AlexNet-RCBAM model. The ensemble model achieved a remarkable identification accuracy of 97.43% on the test set, representing an improvement of 5% to 19% over individual models. Critically, it maintained an accuracy of 92.06% even under altered drilling conditions. This research effectively addresses the limitations of conventional methods, namely their reliance on expert experience and poor adaptability, delivering an efficient intelligent solution for real-time lithology determination of drilling strata.</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

Ensemble deep learning for lithology identification using drilling sound and vibration signals

  • 期刊论文

  • 发表日期：

    2025年10月22日

查看全文
  • Jie Zhang,
  • Sheng Wang,
  • Chengchao Xie,
  • Zheng Zhang,
  • Kun Lai,
  • Shiyi Xu,
  • Jun Bai

摘要

Lithology identification stands as a core objective in geological exploration. Traditional methods suffer from strong subjectivity and low efficiency. This paper innovatively proposes an intelligent identification model integrating an enhanced AlexNet with ensemble learning: First, sound and triaxial vibration signals from drilling were acquired and optimized using a 2-second observation window. Subsequently, a novel AlexNet-RCBAM base model incorporating Residual structure (R) and Convolutional Block Attention Modules (CBAM) was constructed, demonstrating significantly superior accuracy compared to traditional machine learning algorithms. Finally, demonstrating significantly superior accuracy the EM-AlexNet-RCBAM model. The ensemble model achieved a remarkable identification accuracy of 97.43% on the test set, representing an improvement of 5% to 19% over individual models. Critically, it maintained an accuracy of 92.06% even under altered drilling conditions. This research effectively addresses the limitations of conventional methods, namely their reliance on expert experience and poor adaptability, delivering an efficient intelligent solution for real-time lithology determination of drilling strata.