Machine learning-based predictive model for estimating and improving lithium-Ion battery lifespan and aging mechanisms in electric vehicles.

Abstract

This project presents the development of a machine learning-based predictive model to estimate the lifespan and understand aging mechanisms of lithium-ion batteries (LIBs) in electric vehicles (EVs). With growing reliance on LIBs for sustainable mobility, accurate prediction of battery degradation is critical for optimizing performance, enhancing safety, and reducing costs. The study utilizes a real-world dataset from NASA, incorporating key electrochemical indicators such as internal resistance (Re), charge transfer resistance (Rct), capacity, and ambient temperature. A new feature, degradation_feature, was engineered by combining Re and Rct to better capture battery wear behavior. Four machine learning models were trained and evaluated: Random Forest Regressor, Gradient Boosting, XGBoost, and Support Vector Regression. Among these, the Random Forest Regressor showed the best performance, achieving a near-perfect R² score of 0.999 on the test set, with low mean squared error and mean absolute error values. The model was deployed using Flask and integrated into a user-friendly web interface, allowing real-time lifespan predictions in both cycles and years based on user inputs such as Re, Rct, average daily mileage, and mileage per cycle. Visual graphs and interactive tabs (Overview, Details, Lifespan) improved usability. The system not only predicts Remaining Useful Life (RUL) but also contextualizes it for everyday EV users. However, challenges included limited public data, the need for specialized measurement tools, and difficulty in generalizing across battery chemistries. Future work includes integrating the model into real-time Battery Management Systems (BMS), relaxing input constraints, expanding datasets to capture edge cases, and applying online learning for adaptive performance.