Multifidelity Uncertainty Quantification in Battery Performance for eVTOL Flights Under Material and Loading Uncertainties

This study addresses safety concerns within the rapidly evolving Electric Vertical Takeoff and Landing (eVTOL) aircraft domain, focusing on efficient tools to quantify uncertainties in lithium-ion battery behavior - a critical aspect of eVTOL. One major issue with quantifying uncertainty is the prohibitive computational cost associated with many queries of an expensive-to-evaluate computational model. This work employs three physics-based battery models models of varying fidelity and cost to estimate the mean and the variance of the selected quantities of interest through a multifidelity method to reduce the computation cost. By combining information from multiple cheaper, lower-fidelity models through the Multifidelity Monte Carlo method, we significantly reduce the number of high-fidelity samples required for a prescribed mean-squared error, consequently reducing computational costs down to a tractable level. The proposed methodology is applied to estimate the mean and the variance of the battery temperature and voltage, accounting for uncertainties in flight conditions and materials. The first example focuses on a 580-second flight and is benchmarked against a standard Monte Carlo sampling technique. Results indicate a notable fourfold speed-up using the Multifidelity Monte Carlo method compared to the standard Monte Carlo method for the same mean-squared error for the voltage estimate. To showcase the method's generality, the multifidelity method is then applied to a longer flight of 3580 seconds for estimating the mean and the variance and utilizing these statistics to approximately estimate the probability of the flight completion. This demonstrates the adaptability of the methodology to various power profiles and considered uncertainties, with potential extensions to any battery chemistry. In conclusion, the presented multifidelity method offers a robust approach to enhance eVTOL safety by efficiently estimating uncertainties in battery behavior.