Simultaneous electrochemical detection of Gd, Dy, and Eu using boron-doped diamond electrode combined with machine learning

Rare earth elements (REE) are currently high-demand mineral commodities for various countries. The electrochemical technique plays a crucial role in determining REE, offering high sensitivity compared to X-ray fluorescence spectrometry (XRF). This study aimed to detect the content of Gd, Dy, and Eu in a mixture without passing through a chemical separation, using the Differential Pulse Voltammetry (DPV) method and Boron Doped Diamond (BDD) working electrode combined with machine learning. A total of 125 variations of Gd, Dy, and Eu mixture solutions were prepared as the training set and measured using the DPV method. By employing the BDD working electrode, the current peak of Eu appeared separately from that of Gd and Dy, at a potential of -0.6 V. Meanwhile, Gd and Dy appeared in a single current peak at a potential of -1.4 V. Eu exhibited a Limit of Detection (LoD) and Limit of Quantification (LoQ) at 3.040 ppm and 9.211 ppm, Gd at 17.201 ppm and 7.475 ppm, as well as Dy at 22.652 ppm and 5.676 ppm, respectively. After algorithm selection and preprocessing in machine learning, the best model obtained was GLMNET for Eu with an R 2 of 0.853, Dy at 0.376, and SVM for Gd with 0.557. These algorithms correctly predicted the closeness of the percentage recovery of the Gd, Dy, and Eu combination to the actual percentage recovery. • Boron-Doped Diamond (BDD) working electrodes enable the distinct detection of Europium (Eu) at -0.6 V, separate from Gadolinium (Gd) and Dysprosium (Dy) which share a peak at -1.4 V when using Differential Pulse Voltammetry (DPV). • The Limit of Detection (LoD) and Limit of Quantification (LoQ) for Eu were 3.040 ppm and 9.211 ppm, respectively. For Gd, the LoD and LoQ were 7.475 ppm and 17.201 ppm, and for Dy, they were 22.652 ppm and 5.676 ppm. • Machine learning algorithms were successfully applied to analyze rare earth element (REE) mixtures, with the selected machine learning algorithms (GLMNET and SVM) improved the R² values, leading to more accurate predictions of %Recovery for the REE mixture.