Real-time AI-enabled digital twin for battery health estimation and fast charging using partial-discharge data
Mohammad Qasem, Jeff Stubblefield, Moath Qandil, Yazan Yassin, Mariana Haddadin, Mahesh Krishnamurthy
Abstract
Digital twin technology has emerged as a promising approach for integrating multi-physics models in real-time to optimize the operation of electric vehicles (EVs) and electric vertical take-off and landing (eVTOLs), particularly in terms of battery performance. However, the mitigation of dynamic lithium plating and solid electrolyte interphase (SEI) growth during fast charging remains unaddressed in current studies. This paper proposes an AI-enabled digital twin that uses partial-discharge data, data from incomplete discharge cycles, for real-time battery-health estimation and couples this insight with an age-aware fast-charging controller that adaptively controls the charging current to mitigate lithium plating and SEI growth. The experimental results demonstrated the framework’s robustness across varying ambient temperatures and initial state of charge (SoC) conditions. A novel real-time estimation model within the framework achieved a root mean square error (RMSE) of less than 0.5% and 0.4% for both battery capacity and internal resistance. Additionally, the proposed framework preserved battery capacity of 87.6% at 25 ° C compared to 81.4% and 64.3% for MCC-CV and CC-CV, respectively, representing relative improvements of +7.6% and +36.2% over MCC-CV and CC-CV, respectively. This approach helped mitigate battery side reactions during fast charging, while it reduced the time required to reach 80% SoC to less than 25 min, which was 28.6% faster than MCC-CV (35 min) and 35.9% faster than CC-CV (39 min) after 200 cycles. These results support practical deployment in embedded BMS and EV/eVTOL charging to enhance safety, reduce plating risk, and extend service life.