Predictive Insight into Tailings Flowability at Their Disposal Using Operating Data-Driven Artificial Neural Network (ANN) Technique

• Published in: Minerals (Basel) 2024-08, Vol.14 (8), p.737
• Main Authors: Herrera, Nelson, Mollehuara, Raul, Gonzalez, María Sinche, Okkonen, Jarkko
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Artificial intelligence; Artificial neural networks; artificial neural networks (ANNs); Categories; Complex variables; Datasets; Decision making; Deep learning; flowability; Forecasts and trends; Laboratories; Mathematical models; Methods; Mine tailings; Mine wastes; Mineral industry; mineral waste disposal; Mining industry; Moisture content; Neural networks; Occupational safety; Ores; Particle size; Particle size distribution; Permeability; prediction accuracy; Refuse and refuse disposal; Size distribution; Storage capacity; Storage conditions; Stress concentration; Tailings; tailings management; Turbidity; Variables; Viscoelasticity; Waste disposal; Waste management; Water content; Yield strength; Yield stress
• ISSN: 2075-163X; 2075-163X
• DOI: 10.3390/min14080737

This study investigates the application of artificial neural networks (ANNs) in predicting the flowability of mining tailings based on operational variables. As the mining industry seeks to enhance operations with complex ores, the constant improvement and optimization of mineral waste management are crucial. The flowability of tailings was investigated with data driven by properties such as particle-size distribution, water content, compaction capacity, and viscoelastic characteristics that can directly affect stacking, water recovery capabilities, and stability at disposal, influencing storage capacity, operational continuity, and work safety. There was a strong correlation between water content and tailings flowability, emphasising its importance in operational transport and deposition. Three ANN models were evaluated to predict tailings flowability across three and five categories, where a model based on thickening operational variables, including yield stress and turbidity, demonstrated the highest accuracy, achieving up to 94.4% in three categories and 88.9% in five categories. Key variables such as flocculant dosage, water content, yield stress, and solid concentration were identified as crucial for prediction accuracy The findings suggest that ANN models, even with limited datasets, can provide reliable flowability predictions, supporting tailings management and operational decision-making.