Olamide Taiwo Blessing, Gebretsadik Angesom, Abbas Hawraa H, Khishe Mohammad, Fissha Yewuhalashet, Kahraman Esma, Rabbani Ahsan, Abiodun Akinlabi Adams
Department of Mining Engineering, Federal University of Technology, Akure, Nigeria.
Department of Resources Engineering, Graduate School of Engineering, Division of Sustainable Resources Engineering, Hokkaido University, Sapporo, 001-0015, Japan.
Heliyon. 2024 Jun 19;10(12):e33099. doi: 10.1016/j.heliyon.2024.e33099. eCollection 2024 Jun 30.
Maximizing the use of explosives is crucial for optimizing blasting operations, significantly influencing productivity and cost-effectiveness in mining activities. This work explores the incorporation of machine learning methods to predict powder factor, a crucial measure for assessing the effectiveness of explosive deployment, using important rock characteristics. The goal is to enhance the accuracy of powder factor prediction by employing machine learning methods, namely decision tree models and artificial neural networks. The analysis finds key rock factors that have a substantial impact on the powder factor, hence enabling more accurate planning and execution of blasting operations. The analysis uses data from 180 blast rounds carried out at a dolomite mine in south-south Nigeria. It incorporates measures such as root mean square error (RSME), mean absolute error (MAE), R-squared (R), and variance accounted for (VAF) to determine the best models for predicting powder factor. The results indicate that the decision tree model (MD4) outperforms alternative approaches, such as artificial neural networks and Gaussian Process Regression (GPR). In addition, the research presents an efficient artificial neural network equation (MD2) for estimating the values of optimum powder factor, demonstrating outstanding blasting fragmentation. In conclusion, this research provides significant information for improving the accuracy of powder factor prediction, which is especially advantageous for small-scale blasting operations.
最大限度地利用炸药对于优化爆破作业至关重要，这对采矿活动的生产率和成本效益有重大影响。这项工作探索了结合机器学习方法，利用重要的岩石特性来预测炮孔装药量系数，这是评估炸药部署效果的关键指标。目标是通过采用机器学习方法，即决策树模型和人工神经网络，提高炮孔装药量系数预测的准确性。分析发现了对炮孔装药量系数有重大影响的关键岩石因素，从而能够更准确地规划和执行爆破作业。分析使用了在尼日利亚西南部一个白云石矿进行的180次爆破的数据。它采用均方根误差（RSME）、平均绝对误差（MAE）、决定系数（R）和解释方差（VAF）等指标来确定预测炮孔装药量系数的最佳模型。结果表明，决策树模型（MD4）优于其他方法，如人工神经网络和高斯过程回归（GPR）。此外，该研究提出了一个用于估计最佳炮孔装药量系数值的高效人工神经网络方程（MD2），显示出出色的爆破破碎效果。总之，这项研究为提高炮孔装药量系数预测的准确性提供了重要信息，这对小规模爆破作业尤其有利。
