Estimation of lithium-ion battery electrochemical properties from equivalent circuit model parameters using machine learning

Equivalent Circuit Models (ECMs) are widely used in Battery Management Systems (BMSs) for their computational efficiency, yet they often lack precision in predicting battery internal states. In contrast, Physics-Based Models (PBMs) offer detailed insights into battery behavior but are computationally intensive. The advent of cloud computing provides a promising solution to improve battery state estimation by means of a PBM-based battery digital twin. However, tracking the variation of PBM parameters during battery life is a non-trivial problem. This work proposes an approach based on a neural network, which leverages the existing BMS capability to estimate ECM parameters online by correlating them to critical electrochemical parameters of a Pseudo-Two-Dimensional PBM. The neural network is trained and validated on a synthetic dataset generated using a Pseudo-Two-Dimensional PBM to explore various degrees of battery degradation. The achieved error on electrolyte transport properties estimation is below 1 % for 99.9 % of the tested cases. Errors on electrode solid phase diffusivity are below 1 % for 94.1 % and 97.4 % of cases for negative and positive electrodes, respectively. Similarly, errors below 5 % are achieved for 93.4 % of cases in negative electrode intercalation kinetics and 90.5 % in positive electrode intercalation kinetics. Therefore, this methodology represents a practical solution to leverage Equivalent Circuit Models and Physics-Based Models synergistically, thereby enhancing battery state estimation capabilities using cloud computing approaches.