Multi-Fidelity Surrogate Modeling for Lithium-Ion Battery Pack Thermal Runaway Propagation Considering Propagation Occurrence in Adaptive Sampling

Thermal runaway propagation in large-scale lithium-ion battery packs poses critical safety risks, but high-fidelity physics-based simulation of propagation across pack design parameters is computationally prohibitive for design exploration. This paper develops a multi-fidelity surrogate modeling framework that combines simulation data at different fidelity levels to predict thermal runaway propagation behavior while explicitly accounting for whether propagation occurs. An adaptive sampling strategy concentrates training data in regions where propagation is uncertain or design parameters strongly influence outcomes. The approach enables efficient evaluation of battery pack design alternatives and supports safer, more resilient pack architectures with substantially reduced computational cost compared with high-fidelity simulation alone.