Workshop Structure
Day 1: Battery Aging Data, Health Indicators & SOH Fundamentals
- Battery Health Fundamentals: Capacity, cycle life, degradation, State of Health (SOH), End-of-Life (EOL), and Remaining Useful Life (RUL).
- Battery Aging Mechanisms: Capacity fade, internal resistance growth, temperature effects, and cycling degradation.
- Battery Cycling Data: Understanding voltage, current, capacity, temperature, charge/discharge cycles, and timestamps.
- Battery Datasets: Introduction to publicly available battery-aging datasets such as NASA, CALCE, and Oxford datasets.
- Data Preparation: Cleaning, restructuring, handling missing data, and cycle-wise battery-data organization using Python.
- Health Indicator Extraction: Capacity retention, voltage characteristics, charge/discharge duration, temperature statistics, and cycle-based indicators.
- SOH Calculation: Creating SOH labels from battery capacity and degradation data.
- Exploratory Analysis: Visualizing capacity fade and identifying battery-aging patterns.
Hands-On
Build a Battery Aging & SOH Dataset
Participants will load a real battery-cycling dataset, extract degradation indicators, calculate SOH, and visualize battery capacity fade across charge-discharge cycles.
Tools: Python, Google Colab, Pandas, NumPy, Matplotlib
Day 2: Machine Learning & Explainable AI for SOH Prediction
- SOH Prediction as an AI Problem: Converting battery degradation data into supervised machine-learning datasets.
- Feature Engineering: Creating predictive features from voltage, current, temperature, capacity, and cycle information.
- Feature Selection: Identifying the most informative battery-health indicators.
- Baseline Models: Linear Regression and Random Forest for SOH estimation.
- Advanced Machine Learning: XGBoost-based battery-health prediction.
- Model Training & Testing: Train-test splitting and cross-validation for battery datasets.
- Model Evaluation: MAE, RMSE, R², and prediction-error interpretation.
- Predicted vs Actual SOH: Evaluating model reliability across battery cycles.
- Explainable AI: Using SHAP to identify which battery parameters drive SOH predictions.
- Model Generalization: Challenges when applying an AI model across different batteries, chemistries, temperatures, and operating conditions.
Hands-On
Develop an Explainable AI Model for Battery SOH Prediction
Participants will train Random Forest/XGBoost models, predict battery SOH, evaluate prediction accuracy, and use SHAP to interpret important degradation indicators.
Tools: Scikit-learn, XGBoost, SHAP, Pandas, Google Colab
Day 3: Deep Learning, Degradation Forecasting & Remaining Useful Life Prediction
- From SOH to RUL: Understanding how degradation trajectories are converted into remaining-life estimates.
- Time-Series Battery Data: Sequential behavior of capacity and degradation measurements.
- Degradation Trend Modeling: Learning battery-aging trajectories from historical cycles.
- Deep Learning for Battery Prognostics: Introduction to LSTM networks for sequential battery data.
- LSTM Data Preparation: Time windows, sequence generation, scaling, and training datasets.
- Battery Degradation Forecasting: Predicting future SOH and capacity decline.
- End-of-Life Threshold: Defining battery EOL using capacity-retention criteria.
- RUL Estimation: Predicting the number of remaining operating cycles before EOL.
- ML vs Deep Learning: Comparing conventional machine-learning and LSTM-based approaches.
- Prediction Uncertainty: Understanding why RUL predictions change across batteries and operating conditions.
- Research Interpretation: Translating AI predictions into battery-management, EV, and energy-storage applications.
Hands-On
Build an AI Workflow for Battery RUL Prediction
Participants will prepare sequential battery data, train an LSTM-based degradation model, forecast future SOH, identify the predicted EOL point, and estimate Remaining Useful Life.
Tools: TensorFlow/Keras, Scikit-learn, Pandas, NumPy, Google Colab
