Abstract

Understanding leaching kinetics is complex due to the variety of interrelated phenomena that occur at the microscale. Given these difficulties in modeling leaching kinetics, machine learning techniques can be used as highly valuable tools to surrogate these phenomena. In the present work, a hybrid model was developed for the heap leaching process, combining phenomenological aspects of fluid flow in porous media (macroscale phenomena) with artificial neural networks (ANN) that replace dissolution kinetics for copper oxide ores in acid media, thereby enabling the prediction of copper recovery and acid consumption in the process. The training results used different laboratory-scale results, obtaining a value for the loss function closer to 4.41·10-3. Differences observed can be attributed to mineralogical differences that were not properly quantified, which the ANN was unable to capture. The column-scale validation showed a moderate fit, with opportunities to improve performance: Mean Absolute Percentage Error (MAPE) ranged from 26.55 to 41.39, Mean Absolute Error (MAE) ranged from 3.52 to 7.04, and Root Mean Square Error (RMSE) ranged from 3.74 to 7.62. Discrepancies between model predictions and experimental data can be attributed to mineralogical variability. Overall, the proposed hybrid modeling approach shows promising potential for replacing complex mechanistic formulations. However, its generalization capability is strongly dependent on the quality and diversity of the experimental dataset. The application’s perspective as a supporting tool offers an opportunity to facilitate more informed decision-making, e.g., for heap leaching operations in the irrigation schedule, thereby contributing directly to sustainable and economically viable mineral processing practices and reducing experimental cost and time.