Scholarlink

<p>Core drilling is an important method for geological exploration. The rate of penetration (ROP) significantly affects the drilling expenses in core drilling engineering. Optimising the drilling operational parameters can increase the ROP and reduce its overall cost. Drilling operational parameters can be optimised to maximise the ROP. ROP models are mostly based on rock core properties; however, acquiring rock core properties takes time. To improve the ROP in real-time, a real-time ROP optimisation model based on rock core images is proposed. In this model, a convolutional neural network (CNN) based on ResNet-50 was used to extract features from rock core images. These features were then input into the first artificial neural network (ANN), and both the outputs of the first ANN and the drilling parameters were input into the second ANN model. This model outperformed the model with only rock core images or drilling parameter inputs. The coefficient of determination in the testing dataset was 0.9479, and the root mean square error was 0.1286&#xa0;m/h.</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 Hybrid Rate of Penetration Prediction Model Based on Rock Core Images and Drilling Parameters

  • 期刊论文

  • 发表日期：

    2025年12月21日

查看全文
  • Xili Xi,
  • Naipeng Liu,
  • Huaping Wang,
  • Zhenkun Wu,
  • Hui Gao,
  • Longchen Duan,
  • Yule Hu

摘要

Core drilling is an important method for geological exploration. The rate of penetration (ROP) significantly affects the drilling expenses in core drilling engineering. Optimising the drilling operational parameters can increase the ROP and reduce its overall cost. Drilling operational parameters can be optimised to maximise the ROP. ROP models are mostly based on rock core properties; however, acquiring rock core properties takes time. To improve the ROP in real-time, a real-time ROP optimisation model based on rock core images is proposed. In this model, a convolutional neural network (CNN) based on ResNet-50 was used to extract features from rock core images. These features were then input into the first artificial neural network (ANN), and both the outputs of the first ANN and the drilling parameters were input into the second ANN model. This model outperformed the model with only rock core images or drilling parameter inputs. The coefficient of determination in the testing dataset was 0.9479, and the root mean square error was 0.1286 m/h.