Knowledge Bank

The Battery Management System (BMS) performs various tasks in order to ensure the safe and reliable operation of lithium-ion battery packs in electric vehicles (EVs). One of the primary functions of the BMS is to perform state estimation to describe the condition of the cells. Of these states, state of charge (SOC) defines how much capacity is left in a cell, and more in generally in the pack. Due to the complex electrochemical characteristics of batteries, the SOC cannot be directly measured. As a result, it must be estimated with respect to the current, voltage, and temperature of the cell to fully capture the state. The most general form of SOC estimation is Coulomb counting, however this method lacks the complexity required for intense systems, such as EVs. As such, a model based approach using the battery equivalent circuit model (ECM) of the cell is better. Using this model, a Kalman filter may be implemented for better approximation, accounting for the effects of current, voltage, and temperature on the cell while being highly adaptable to changes as well. The goal of this research is to quantify the trade offs between model complexity and SOC estimation accuracy by optimizing the ECM for SOC estimation. This study will focus specifically on an algorithm for a Samsung INR21700-50G battery cell. The impact of first- and second-order ECMs alongside polynomial approximations for ECM parameters used in the linear and extended Kalman filters will be analyzed. To do so, the ECM parameters will be approximated as polynomials through Matlab and validated against experimental data in Matlab, Simulink, and Gt-Suite. Multiple profiles will be tested to model multiple operating conditions of the cell. This will help determine which combination performs best for the required uses by comparing the root mean square (RMS) error and a statistical analysis of the terminal voltage prediction alongside simulation time. Currently, the results show using an extended Kalman filter with an first-order EMC and an OCV approximation in the fifth-order with respect to SOC and second-order in temperature provides the best results compared to the other polynomials. These results also show a comparative analysis on how effective each iteration of the Kalman filter is when given the set of profiles. Specifically, these results show the large difference in accuracy when assuming linearity and when allowing for nonlinear approximations. Implementing the developed algorithm into a BMS will allow for accurate and quick estimation of SOC for the cell. These results will push forward more effective modes of SOC estimation and help provide a gateway for estimating SOC on a module and pack level.