Optimization of an Artificial Neural Network Using Four Novel Metaheuristic Algorithms for the Prediction of Rock Fragmentation in Mine Blasting

• Published in: Journal of the Institution of Engineers (India): Series D Vol. 106; no. 2; pp. 1261 - 1280
• Main Authors: Rabbani, Ahsan, Kumar, Divesh Ranjan, Fissha, Yewuhalashet, Bhavani, Nallamilli P. G., Ahirwar, Sunil Kumar, Sharma, Sushila, Saraswat, Bhupendra Kumar, Ikeda, Hajime, Adachi, Tsuyoshi
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 01.08.2025; Springer Nature B.V
• Subjects: Algorithms; Artificial neural networks; Blasting; Civil engineering; Construction; Cost control; Cost effectiveness; Detonation; Efficiency; Energy consumption; Engineering; Explosives; Feature selection; Fragmentation; Heuristic methods; Limestone; Mining; Optimization; Original Contribution; Parameter sensitivity; Performance indices; Rock masses; Sensitivity analysis; Time series; ANN; TLBO; ALO; ICA; Rock Blasting; PSO; Optimization; Fragmentation
• ISSN: 2250-2122; 2250-2130
• DOI: 10.1007/s40033-024-00781-x

Rock fragmentation is a critical process in mining operations, with blasting being one of the most common and effective methods employed to achieve the desired results. The primary goal of blasting operations is to deliver optimum rock fragmentation while avoiding adverse environmental effects like back-breaks, fly rock, and ground vibration. The process involves the controlled detonation of explosives within a rock mass, resulting in the generation of smaller rock fragments. This fragmentation is essential for facilitating the extraction, transportation, and processing of valuable minerals, and it plays a pivotal role in optimizing the overall productivity and cost-effectiveness of mining operations. The current work attempted to optimize the ANN utilizing four innovative MOAs: PSO, ICA, TLBO, and ALO, to predict rock fragmentation in mine blasting. A dataset consisting of 219 blasting events with 10 influencing parameters was considered from a limestone mine blasting site located in India. All optimized models were evaluated using the following performance indices: R 2 , WMAPE, NS, RMSE, VAF, PI, WI, and MAE. The top models were chosen based on the performance indices results, accuracy matrix, ranking analysis, scatter plot, and convergence curve of the optimized models. It was found that the ANN-ICA model outperforms other optimized models for predicting rock fragmentation. This is because, during the training process, the ICA algorithm updates the weights and biases of the ANN better than the PSO, TLBO, and ALO. From the sensitivity analysis of all influencing parameters, it was found that CS emerged as a highly influential factor and MCPD as the second most influential parameter.