AI for Degradation Modeling in Energy Storage Systems

Topics Covered

• Lithium-ion battery aging: calendar aging vs cycle aging
• Major degradation mechanisms: capacity fade, resistance growth and thermal stress
• Understanding charge–discharge curves and battery cycling data
• Battery health indicators: SOC, SOH, RUL and End-of-Life
• Extracting capacity, energy, temperature and cycle-based health features
• Detecting degradation trends, anomalies and the battery “knee point”
• Data quality challenges: missing cycles, noise and irregular measurements
• Data-driven, physics-based and hybrid degradation modeling approaches

Day 1 Output: Cleaned battery health dataset, SOH trajectory and degradation diagnostic dashboard.

Day 2: Machine Learning for Explainable SOH Estimation

Topics Covered

• Converting charge–discharge signals into predictive health features
• Feature engineering from voltage, current, temperature and cycle duration
• Incremental capacity and curve-derived degradation indicators
• Baseline modeling using Linear Regression
• Nonlinear SOH prediction using Random Forest and XGBoost
• Battery-aware train–test splitting and prevention of data leakage
• Model evaluation using MAE, RMSE and R²
• Feature importance and explainable battery AI using SHAP
• Comparing accuracy, interpretability and computational cost

Day 2 Output: Validated SOH prediction model, model-comparison report and SHAP-based degradation interpretation.

Day 3: Remaining Useful Life Prediction and AI-Enabled Battery Prognostics

Topics Covered

• Formulating RUL as a time-series forecasting problem
• Defining End-of-Life and generating cycle-wise RUL labels
• Early-cycle prediction of long-term battery degradation
• Sequence preparation using sliding windows
• LSTM-based modeling of nonlinear degradation trajectories
• Prediction uncertainty and confidence-aware maintenance decisions
• Physics-informed constraints for reliable battery prognostics
• Model generalization across cells and operating conditions
• Deployment pathway for BMS, predictive maintenance and battery digital twins
• Translating model results into research figures and publication workflows

Day 3 Output: Colab-based RUL prediction pipeline, End-of-Life forecast and maintenance decision plot