Scholarlink

<p>In this study, we model a solution surface using an ensemble machine learning approach to simultaneously predict and optimise rock fragmentation and ground vibration during blasting operations at the Jwaneng Diamond Mine in Botswana. Using 120 production blasts and ten input parameters, we trained several models: ANN, RF, ANN–RF and other hybrid baselines PSO–ANN, PSO–ELM, ANN–SVR, PSO–XGBoost and GA–ANN . The ANN-RF ensemble achieved the best test performance (<i>R</i>2 = 0.956, RMSE = 0.315, MAE = 0.250, and VAF = 95.5 for fragmentation; <i>R</i>2 = 0.930, RMSE = 0.380, MAE = 0.302, VAF = 92.5 for PPV). Tree-SHAP analysis identified powder factor and burden as dominant fragmentation drivers, and burden, charge per delay, and distance as key vibration controls. A projected gradient-descent search on the learned solution surface produced feasible blast settings that increase fragmentation toward approximately 84% while reducing PPV to nearly 0.12 mm/s. This approach allows blasting engineers to interactively vary input parameters within defined constraints to obtain optimal environmental and operational outcomes. The input parameters include spacing, charge per delay, burden, stemming length, powder factor, hole depth, hole diameter, rock factor, distance from the blast point to the monitoring point and blastability index, offering a comprehensive AI-driven framework for precision blast design in open-pit mining.</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

Simultaneous prediction and optimisation of rock fragmentation and ground vibration using an ANN–RF ensemble in open-pit blasting

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月27日

查看全文
  • Onalethata Saubi,
  • Rodrigo S. Jamisola Jr.,
  • Raymond S. Suglo,
  • Oduetse Matsebe

摘要

In this study, we model a solution surface using an ensemble machine learning approach to simultaneously predict and optimise rock fragmentation and ground vibration during blasting operations at the Jwaneng Diamond Mine in Botswana. Using 120 production blasts and ten input parameters, we trained several models: ANN, RF, ANN–RF and other hybrid baselines PSO–ANN, PSO–ELM, ANN–SVR, PSO–XGBoost and GA–ANN . The ANN-RF ensemble achieved the best test performance (R2 = 0.956, RMSE = 0.315, MAE = 0.250, and VAF = 95.5 for fragmentation; R2 = 0.930, RMSE = 0.380, MAE = 0.302, VAF = 92.5 for PPV). Tree-SHAP analysis identified powder factor and burden as dominant fragmentation drivers, and burden, charge per delay, and distance as key vibration controls. A projected gradient-descent search on the learned solution surface produced feasible blast settings that increase fragmentation toward approximately 84% while reducing PPV to nearly 0.12 mm/s. This approach allows blasting engineers to interactively vary input parameters within defined constraints to obtain optimal environmental and operational outcomes. The input parameters include spacing, charge per delay, burden, stemming length, powder factor, hole depth, hole diameter, rock factor, distance from the blast point to the monitoring point and blastability index, offering a comprehensive AI-driven framework for precision blast design in open-pit mining.