Exploitation
Taiwo Blessing Oamide; Adebayo Babatunde; Toluwase Daniel Olaiya
Abstract
This study developed and assessed several artificial intelligence (AI) models for predicting blast-induced toe volume in small-scale dolomite mines located in the Akoko Edo Local Government Area, Edo State, Nigeria. Seven predictive models were constructed: Adaptive Boosting (AdaBoost), Random Forest ...
Read More
This study developed and assessed several artificial intelligence (AI) models for predicting blast-induced toe volume in small-scale dolomite mines located in the Akoko Edo Local Government Area, Edo State, Nigeria. Seven predictive models were constructed: Adaptive Boosting (AdaBoost), Random Forest (RF), Gaussian Process Regression (GPR), Support Vector Regression (SVR), a conventional Artificial Neural Network (ANN), and two Imperialist Competitive Algorithm-optimized ANNs (ICA-ANNs). The models were trained using eight input parameters including uniaxial compressive strength (UCS), spacing (S), burden (B), sub-drill (SB), drill hole length (DHL), stiffness ratio (SR), maximum instantaneous charge (MIC), and powder factor (K) with blast toe volume (TV) as the target output. Input data were collected through a combination of field measurements and laboratory analyses. Among all the models evaluated, the ICA-ANN with an 8-7-1 architecture achieved the highest predictive accuracy. It outperformed AdaBoost by 9.17%, SVR by 7.20%, GPR by 5.56%, RF by 4.75%, a standard ANN (8-5-1) by 0.78%, and a standard ANN (8-7-1) by 0.28%, based on mean squared error (MSE) and coefficient of determination (R²) metrics. Furthermore, the ICA-ANN model was applied to optimize blast design parameters. The optimal values obtained were: spacing = 1.0 m, burden = 0.8 m, sub-drill = 0.6 m, MIC = 0.72 kg, and powder factor = 0.65 kg/m³. These optimized parameters reduced the blast toe volume by 20.05%, from 209.50 m³ to 154.87 m³. The results highlight the robustness and efficiency of the ICA-ANN model for blast design optimization. By improving fragmentation quality and minimizing residual toe volume, the approach offers a practical pathway for enhancing both productivity and cost-effectiveness in small-scale mining operations.
