Machine learning approach to the possible synergy between co-doped elements in the case of LiFePO4/C

Abstract

This study investigates the synergistic effects produced by the co-doping of several components in the LFP/C structure. To execute this work, a dataset was initially created from the existing literature, encompassing information on doped LFP structures by a singular element. Numerous intrinsic and extrinsic characteristics, such as atomic number, valence, relative variations in atomic and ionic radii of Fe and Li, electronegativity, molar percentage of dopant, and C-rate, were evaluated. The optimal selection of features leading to satisfactory model training was achieved by analyzing the Pearson correlation coefficient factors. Subsequently, two machine learning algorithms (i.e., Random Forest and Gaussian Process Regression) were trained using the optimized feature set. The two models were evaluated, and the model with superior predictive power was chosen for further study. An analysis of the synergistic effect of two co-dopants was conducted by comparing the actual specific discharge capacities with the expected values derived from the superimposition of the machine learning predictions. Ultimately, experimental validation was conducted by synthesizing several unique LiYxNdyFe1-x-yPO4/C (Nd = 0.06, 0.02 <Y<0.08) samples using solid-state methods. The synthesized powders underwent relevant testing, including SEM, TEM, CV, EIS, and GD. Finally, based on the best ML scheme developed and experimental results, another ML scheme was developed to analyze the possible synergic effects that co-dopants may exhibit regarding the specific discharge capacity of co-doped LFP structures.