Explainable Machine Learning-Based Predictive Thermal Management and Risk-Aware Decision Support System for Lithium-Ion Batteries Under Multi C-Rate Conditions

Effective thermal management is essential for the safety and lifetime of lithium-ion batteries. This study presents a proof-of-concept framework combining electrochemical–thermal simulation with machine learning for temperature prediction, thermal-state classification and uncertainty-aware decision support for 26650 cells with phase change material (PCM) cooling. A Newman–Tiedemann–Gu–Kim model, calibrated at 0.5C–1.5C (average voltage and heat-flux accuracies of 94% and 82%), was used to simulate six PCMs and a no-PCM baseline at 2C–4C and three ambient temperatures (53 cases). With hyperparameters selected on training data only, Gaussian Process Regression (GPR) predicted the final temperature with a held-out RMSE of 0.56 K (R2 = 0.993) and a nested cross-validated RMSE of 0.76 K, clearly outperforming Random Forest (RF), but did not generalize reliably to materials absent from the training data. Ambient temperature and C-rate were the dominant predictors. An RF classifier of the thermal state (Cold, Phase-Change, Overheating) reached only 58.5% cross-validated accuracy, whereas assigning the state from the GPR predictions reached 92.8%. Monte Carlo input perturbation, predicted temperature and C-rate were combined in a rule-based scheme assigning Normal, Monitor, Pre-Cool and Emergency actions. Larger datasets and experimental validation at high C-rates are required before deployment.