System identification and artificial intelligent (AI) modelling of the molten salt electrolysis process for prediction of the anode effect

Abstract

NdFeB magnets are widely used in various applications including electric and hybrid vehicles, wind turbines, and computer hard drives. They contain approximately 31–32 wt% Rare Earth Elements (REEs), mainly neodymium (Nd) and praseodymium (Pr), and are produced by molten salt electrolysis using fluoride electrolytes. However, anode passivation or anode effect may occur, generating greenhouse gases if insufficient amounts of metal oxides are available in the system. Therefore, in this study, a dynamic model of the electrochemical process was developed to estimate the system variables and predict the anode effect using several system identification methods. The Transfer Function (TF) estimation, Auto-Regressive with Extra inputs (ARX), Hammerstein-Weiner (HW), and Artificial Neural Network (ANN) models were used, and their results were compared based on the occurrence of the anode effect. The best model achieved an average accuracy of 96% in predicting the process output.