Selective Recovery of Zinc from Alkaline Batteries via a Basic Leaching Process and the Use of a Machine Learning-Based Digital Twin for Predictive Purposes

Abstract

Recycling the metals found in spent batteries offers both environmental and economic benefits, especially when extracted and purified using environmentally friendly processes. Two basic leaching agents were tested and compared: ammonium hydroxide (NH4OH) and sodium hydroxide (NaOH). Using NH4OH 4 M at 25 °C, 30.5 ± 0.7 wt. % of zinc (Zn) was dissolved for a solid/liquid (S/L) ratio of 1/10 (g of black mass (BM)/mL of solution); meanwhile, with NaOH 6 M at 70 °C, and an S/L ratio of 1/5 (g of BM/mL of solution), 69.9 ± 2.8 wt. % of the Zn initially present in the BM of alkaline batteries was leached. A virtual representation of the experimental data through digital twins of the alkaline leaching process of the BM was proposed. For this purpose, 90% of the experimental data were used for training a supervised learning procedure involving 600 different artificial neural networks (ANNs) and using up to 12 activation functions. The application was able to choose the most suitable ANN using an ANOVA analysis. After the training step, the network was tested by predicting the outputs of inputs that were not used in the training process, to avoid overfitting in a validating process with 10% of the data. The best model was employed for estimating the degree of leaching of different metals that can be obtained from BM, obtaining a data deviation of less than 10% for highly concentrated compounds such as Zn.