Literature-Informed Stochastic Scenario Analysis of Thermal-Runaway Risk During eVTOL Battery Fast Charging

This is the authors' abstract. We don't add key points for this paper.

Risk management is essential for the safety of air mobility systems as electric aircraft traffic grows. This paper presents a Monte Carlo scenario-analysis framework for thermal-runaway initiation risk in electric vertical take-off and landing (eVTOL) battery systems during fast charging, built on a seven-input multiplicative risk-factor architecture, with extrapolation beyond the thermal-onset calibration range handled through explicit flat-cap and linear-extrapolation scenarios. The framework quantifies sensitivity to an induced input-dependence structure tested in both risk-increasing and risk-reducing directions, overlap between the two cycle-aging pathways, C-rate model-form uncertainty under alternative direct-factor and thermal-onset assumptions, and an interaction-aware Sobol/Spearman analysis. Under the independent baseline, cycle count and initial state of charge were the most influential inputs, while temperature contributed negligibly; limited rank checks across the tested dependence and model-form scenarios indicated that cycle count and state of charge remained influential, though not stably ordered relative to each other. Because the risk-amplification functions are semi-empirical constructions informed by cell-level evidence that does not directly validate eVTOL fast-charging initiation risk, and the automotive incident rate used as an absolute-scale reference could not be independently reconstructed, results are reported primarily as a dimensionless risk multiplier relative to an unstressed reference case, invariant to that reference value, rather than as a validated absolute probability; no regulatory threshold comparison or compliance classification is made. Mitigation strategies are evaluated jointly with proxy operational costs rather than as a risk-only optimization. The absence of eVTOL-specific validation and incident data remains the principal limitation.