Prediction of Back Break Using Sensitivity Analysis and Artificial Neural Networks

• Published in: Journal of the Institution of Engineers (India): Series D Vol. 106; no. 1; pp. 383 - 398
• Main Authors: Kannavena, Sravan Kumar, Pradeep, T., Chandrahas, N. Sri, Prasad, D. U. V. D.
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 01.04.2025; Springer Nature B.V
• Subjects: Algorithms; Artificial intelligence; Artificial neural networks; Back propagation networks; Bench height; Blast holes; Boreholes; Datasets; Decision trees; Drill bits; Engineering; Explosives; Influence; Machine learning; Mines; Neural networks; Original Contribution; Predictions; Regression analysis; Sensitivity analysis; ANN; Sensitivity analysis; Linear regression; Back break
• ISSN: 2250-2122; 2250-2130
• DOI: 10.1007/s40033-024-00653-4

Back break is a negative event produced during blasting operation, which cannot be avoided completely. Large quantity of potential waves released in explosive bore hole, which cross over last row of blast hole. Back break prediction is need of hour, which influences prominently in drilling operations by seizing-up drill bits and escalates mine economics and as well generation of rock boulders. Therefore, in this paper an accurate back break prediction was predicted by using back propagation neural networks (BPNN) and ML techniques like decision tree regressor (DTR) and linear regression (LR) algorithms. To prepare the model dataset for training and testing, 119 blast datasets were collected at JVROCP-II extension project, SCCL. In these analyses the most influential parameters of back break were burden, spacing, stemming length, bench height, number of holes and powder factor. To predict back break a BPNN was used and developed in MATLAB software and compared with ML models such as DTR and LR model in Python. ANN produced a better result in terms of R 2 value as 0.96, and DTR and LR models produced 0.93 and 0.72, respectively. Similarly, in terms of RMSE and VAF ANN produced 0.7 and 94%, which is superior than other two models. ANN gives a better result than DTR and LR techniques, in predicting back break with accuracy of 94%.