Mineral Processing
Mojtaba Saeidi; Vahideh Shojaei; Hamid Khoshdast; Alireza Gholami
Abstract
Effective blast design is crucial for downstream comminution efficiency. However, modeling its impact on size-based production remains challenging due to complex, non-linear rock–fragmentation interactions. This study developed an integrated mine-to-mill framework using optimized Artificial Neural ...
Read More
Effective blast design is crucial for downstream comminution efficiency. However, modeling its impact on size-based production remains challenging due to complex, non-linear rock–fragmentation interactions. This study developed an integrated mine-to-mill framework using optimized Artificial Neural Networks (ANNs) to predict total and size-fractioned production (0–40 mm, 40–80 mm) from blasting parameters at the Sar-Asiab Limestone Complex, Iran. The dataset comprises 55 real industrial observations after preprocessing. Two optimization algorithms, Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), were employed to enhance ANN performance. The PSO-optimized ANN demonstrated significantly superior generalization, achieving a testing RMSE of 1341.8 (vs. 1549.5 for GA), an MAE of 1108.9 (vs. 1260.4), and an R² of 0.9487 (vs. 0.8937). Sensitivity-based feature importance analysis quantified the dominant predictors: explosive quantity (Emulite) contributed 18.4%, the number of boosters 14.9%, and key geometric parameters (mean hole depth, diameter, and spacing) collectively accounted for over 41%. In contrast, initiation system variables such as cortex length showed high linear correlation (r = –0.91) but minimal predictive importance (1.5%). These results underscore that explosive energy and its distribution, not just initiation configuration, are the primary drivers of size-based output. This data-driven framework provides a practical tool for aligning blast design with processing needs, directly supporting energy reduction and production stability in open-pit mining.
