Data Driven AI Models for Particle Size Prediction in Ore Mining Ball Mills

The challenge of mining industry is in optimizing the crushing of particles within ball mills to achieve optimal particle sizes. Particle size is used for determining the efficiency of the grinding process in ore mining. This study explores the application of various machine-learning techniques and primarily uses historical data to predict particle size at the end of the process. The effectiveness of interpolation and prediction algorithms consisting of ElasticNet, Random Forest Regressor, XGBoost, SVR and KNN was examined. The impact of feed-forward neural networks (FNN), GRU and LSTM was evaluated on how neural networks performed in predicting particle size. This study provides a comparative analysis of the above machine learning methods for ball mill particle size prediction and helps us select the best method for this specific production process. Random Forest outperforms all of the experimental approaches that were used. The data has been augmented using linear interpolation to increase data points by 25%.