Remaining Useful Life Prediction of Li-NMC Batteries Using Algorithmic Fusion and Cascading Approach

Lithium-NMC batteries in electric vehicles exhibit complex degradation mechanisms, where capacity fade, internal resistance growth, and discharge behavior evolve nonlinearly under varying operating conditions. Accurate remaining useful life prediction necessitates capturing these intricate interdependencies, which traditional models fail to generalize effectively. This study develops a robust machine-learning framework leveraging experimental cycling data under nominal and over-discharge conditions. Key parameters like voltage, discharge time, internal resistance, and state of health were chosen due to their direct correlation with electrochemical aging, resistive losses, and failure progression, ensuring high sensitivity to degradation dynamics. Support Vector Regression and Bayesian-optimized Lasso Regression were employed to model these dependencies, providing precise predictions of key battery health indicators. A hybrid framework integrating these models for remaining useful life estimation achieved $R^2$, MAE, RMSE of 0.9998, 0.093 and 0.138 respectively, significantly outperforming conventional approaches. Rigorous evaluation through K-fold cross-validation and subset stability analysis ensured generalizability across diverse operating conditions. Benchmark comparisons with state-of-the-art methods demonstrated superior predictive accuracy. By addressing critical limitations in traditional degradation modeling, this work provides a scalable, data-driven solution for real-time battery health management, enhancing the reliability and sustainability of electric vehicle applications.