Abstract
The rapid integration of high-density lithium-ion battery packs in autonomous vehicles requires rigorous real-time safety monitoring systems capable of predicting catastrophic failure modes. Thermal runaway represents one of the most critical safety hazards, characterized by uncontrollable exothermic chain reactions that propagate across battery modules. Traditional computational fluid dynamics and finite element models offer high spatial fidelity but are computationally prohibitive for real-time onboard battery management systems. Conversely, standard data-driven machine learning models often fail to generalize under unseen dynamic operational regimes due to a lack of physical constraints. In this study, we propose a Physics-Informed Neural Network (PINN) framework specifically engineered to forecast spatial-temporal temperature distributions and thermal runaway initiation in lithium-ion battery modules under aggressive autonomous driving load profiles. By embedding transient 3D thermal conduction equations and Arrhenius-based exothermic reaction kinetics directly into the neural network loss function, our model accurately reconstructs internal cell dynamics from sparse sensor telemetry. Experimental validation against thermal abuse tests on nickel-manganese-cobalt (NMC-811) module configurations demonstrates that the PINN achieves a mean absolute error below 0.82°C during nominal operation and accurately predicts the onset of thermal runaway 42 seconds prior to severe venting events. Crucially, the model yields an execution speedup of over two orders of magnitude compared to traditional numerical solvers, enabling real-time, physics-compliant predictive safety diagnostics for next-generation autonomous transportation platforms.
