Artificial Neural Network Prediction of Metastable Zone Widths in Reactive Crystallization of Lithium Carbonate
Siyang Ma, Chao Li, Jie Gao, Yang He, Weiwei Tang, Junbo Gong, Fu Zhou, Zhenguo Gao
Abstract
Metastable zone widths (MSZWs) are one of the crucial parameters in solution crystallization process optimization whose accuracy would determine the crystalline product quality and process robustness. In this paper, the MSZWs of lithium carbonate-reactive crystallization were measured by turbidity technology during the reactive crystallization process of Li2CO3. Three semiempirical models were used to proceed with the prediction of MSZWs, and further, artificial neural networks (ANN) were introduced for the first time to predict MSZWs and compared with semiempirical models. Then, the prediction models were evaluated by the indicators root mean square error (RMSE), R2, mean absolute percentage error (MAPE), and cp. The results indicated that the ANN model has the best prediction accuracy. An orthogonal-dataset-trained ANN model was developed and evaluated, and it showed the highest efficiency and the second-best accuracy. In addition, the effects of process parameters on the MSZWs were investigated and analyzed, including Li2SO4 concentrations, working volumes, agitation speeds, impurities, temperatures, and Na2CO3 feed speeds and concentrations. The results showed that temperatures and concentrations had a strong positive correlation with MSZWs, and temperature is the most sensitive parameter that is recommended for MSZWs-based crystallization process optimization.