A guide to equivalent circuit fitting for impedance analysis and battery state estimation

In this study we define a comprehensive method for analyzing electrochemical impedance spectra of lithium batteries using equivalent circuit models, and for information extraction on state-of-charge and state-of-health from impedance data by means of machine learning methods. Estimation of circuit parameters typically implies a non-linear optimization problem. A detailed method for estimating initial values of the optimization algorithm is described, emphasizing short computation times and efficient convergence to global minimum. Parameters identifiability is investigated through an analysis of the injectivity of the model, Cramer–Rao lower bound, profile likelihood, and sensitivity analysis. An exploratory data analysis is presented to estimate the degree of correlation between impedance spectra (or circuit parameters) and battery state-of-charge or state-of-health, prior to the implementation of any machine learning algorithm. A publicly available dataset of impedance spectra of five lithium-polymer batteries is used to test the whole procedure. Estimation of state-of-charge and state-of-health is performed by implementing Gaussian process regression.