Lithium-ion battery state of health estimation using intelligent methods

Abstract

In electric vehicle applications, detecting Li-ion battery degradation is essential to ensure safety and reliability. A key approach to assessing battery health is monitoring the internal impedance and capacity over the battery's lifetime, which provides insight into the State of Health (SOH) and indicates whether the battery has reached its End of Life (EOL). This study proposes an intelligent SOH estimation algorithm utilizing Feed-forward and Recurrent Neural Networks, trained with the Levenberg-Marquardt function, to predict battery SOH under various aging conditions. The methodology begins with life cycle and Electrochemical Impedance Spectroscopy (EIS) tests to establish the charge-discharge characteristics and create an Equivalent Circuit Model that represents the dynamic properties and degradation indicators of an 18650 Li-ion battery. Key model parameters, such as internal resistance, are extracted per cycle to track aging progression. Finally, the SOH estimation models, developed in SIMULINK, utilize internal impedance and capacity metrics to predict SOH under various aging scenarios. Results in SIMULINK demonstrate that both networks provide accurate SOH estimations; however, the Recurrent Neural Network achieves faster convergence, reaching accurate predictions within 10 epochs. This improved convergence speed, along with high measurement accuracy and reliability, underscores the Recurrent Neural Network's suitability for real-time SOH monitoring in electric vehicle applications.