Improving energy efficiency in the mining Industry: an LSTM-ANN predictive model for sieve refusal in grinding mills
نویسندگان
1 Laboratory of Applied Mathematics and Decision-Making Informatics (LMAID), National School of Mines of Rabat (ENSMR), Rabat, Morocco.
2 Laboratory of Applied Mathematics and Decision-Making Informatics (LMAID), National School of Mines of Rabat (ENSMR), Rabat, Morocco.
3 Reminex Research Center, MANAGEM Group, Marrakech.
4 Embedded Systems and Artificial Intelligence Department, MAScIR, Rabat, Morocco.
5 Laboratory of Applied Mathematics and Decision-Making Informatics (LMAID), National School of Mines of Rabat (ENSMR), Rabat, Morocco.
6 Digitalization and Microelectronic Smart Devices, MAScIR, Rabat, Morocco.
7 LISI Laboratory University of Cadi Ayyad, UCA, Marrakech, Morocco.
8 LISI Laboratory University of Cadi Ayyad, UCA, Marrakech, Morocco.
9 Al-Qualsadi Research and Development Team, ENSIAS, Rabat, Morocco.
doi
10.22059/ijmge.2025.383890.595204چکیده
The integration of Artificial Intelligence (AI) within Mine 4.0 has significantly advanced the mining industry by enhancing performance, efficiency, safety, and overall productivity. Despite these advancements, a critical gap exists in predicting sieve refusal, a key parameter affecting grinding mill efficiency and product quality, particularly when accounting for the temporal and nonlinear dependencies inherent in mining data. This study introduces a hybrid predictive model that combines Long Short-Term Memory networks (LSTM) with Artificial Neural Networks (ANN) to predict sieve refusal in grinding mills utilizing actual process and energy data from mining industry databases. The LSTM component captured the temporal dynamics and time-delayed dependencies of variables and power consumption. Concurrently, the ANN component modeled complex, nonlinear relationships among input variables. This hybrid approach effectively addressed the intrinsic characteristics of mining data, which are often overlooked in traditional models. Comparative analyses demonstrated that the proposed LSTM-ANN model significantly outperformed existing advanced regression and deep learning methods, establishing it as a state-of-the-art solution for sieve refusal prediction. The enhanced predictive accuracy provided direct support for operational planning and scheduling, contributing to improved energy efficiency and cost-effectiveness in mining operations. By addressing the underexplored temporal and spatial interrelations between variables and sieve refusal, this research fills a notable gap in the application of deep learning to grinding mill operations. The findings underscore the transformative potential of advanced AI models in optimizing mining practices, aligning with the broader objectives of Mine 4.0 to leverage intelligent data analysis for operational excellence.
