Machine Learning Enabled Operando Optical Microscopy for Determination of Lithium Transport in Battery Electrodes

Diffusion coefficients of electrode materials are often determined using galvanostatic (GITT) or potentiostatic intermittent titration technique (PITT), electrochemical impedance spectroscopy (EIS) or cyclic voltammetry (CV). However, these methods require special care, as each of their formal derivations use quite restrictive assumptions. As an alternative, a machine learning model is presented to extend a previously proposed optical method of studying lithium transport by operando microscopy. The herein reported model enables the measurement of concentration-dependent diffusion coefficients in a wide solubility range. For this purpose, a python code is developed that determines concentration profiles from RGB images using support vector regression (SVR), a flexible machine learning tool. To evaluate the diffusion coefficient, an inverse Boltzmann-Matano concept is applied. Representing the diffusion coefficient with generalized Redlich-Kister polynomials, concentration profiles are predicted and fit to the measured data. The method is demonstrated here on the example of delithiation of LiMn₂O₄, but it can, in-principle, be extended to any other battery material showing significant optical response on lithiation, which most of them do.