Abstract
Lithium-ion batteries (LIBs) represent the cornerstone of contemporary energy storage technologies, finding widespread application in electric vehicles, portable electronics, and large-scale grid systems. Despite their dominance, the conventional constant current constant voltage (CC–CV) charging protocol encounters substantial challenges at high states of charge (SOC), including pronounced polarization, elevated temperature rise, and extended charging durations. These issues become particularly critical within the 80–100% SOC window, where a disproportionate amount of time is required to complete the charging process. To overcome these limitations, this study introduces a hybrid charging framework that synergistically integrates an auxiliary hardware circuit with a physics-informed deep learning model. The auxiliary circuit, which can be seamlessly inserted between the charger and the battery without necessitating structural modifications to the charger, enables real time identification of the optimal pulse frequency and duty cycle exploiting a novel strategy. Concurrently, a one-dimensional convolutional neural network (1D-CNN) utilizes short relaxation intervals to provide accurate SOC estimation. Through this dual approach, the proposed method alleviates polarization and thermal stress, reduces the charging duration in the 80–100% SOC range by approximately 25%, and enhances open circuit voltage (OCV) prediction accuracy to below 1 mV. The effectiveness of the proposed approach was validated experimentally on NCR18650PF cells at 25 °C, confirming its potential for practical deployment in fast-charging applications.