Session: 15-03-02: Machine Learning and Deep Learning in Safety, Reliability, and Maintenance II

Paper Number: 167277

Bayesian Neural Network for Estimating RUL of eVTOL Batteries While Quantifying Aleatoric and Epistemic Uncertainty 

Electric vertical takeoff and landing (eVTOL) aircraft represent a transformative step in sustainable urban transportation, promising to reduce commuting times and enhance productivity. However, eVTOL batteries face more stringent design requirements than those in electric vehicles (EVs). For instance, eVTOLs require significantly higher power output and charge/discharge rates than EVs, as they operate under high-power transient conditions during takeoff and landing. The high-power demands and frequent charging cycles reduce battery lifespan compared to EVs. A key challenge in battery prognostics is estimating the remaining useful life (RUL), which is typically done by measuring a health index such as battery state-of-health (SOH). SOH is defined as the ratio between the initial capacity and the current capacity of the battery. While SOH prediction has been widely studied in conventional EVs, eVTOL batteries introduce unique challenges due to their high-power demands, aggressive charge-discharge cycles, and stringent safety requirements. Furthermore, most existing research in battery prognosis relies on deterministic SOH prediction models, which provide a single-point estimate of battery degradation over time. However, this approach overlooks the inherent stochastic nature of battery aging, particularly in eVTOLs, where variable mission profiles and extreme operating conditions influence degradation mechanisms. A robust prognostics framework should incorporate uncertainty-aware models to improve reliability and risk assessment in battery management. Some recent studies have started incorporating uncertainty in battery RUL estimation, but they often treat uncertainty as a black-box output without distinguishing its sources. Two key types of uncertainty should be accounted for: Aleatoric uncertainty arises from inherent randomness in the system, such as sensor noise, operational variability, and external environmental conditions. Epistemic uncertainty comes from a lack of knowledge about the degradation phenomenon, which is particularly significant in eVTOL batteries since their aging behavior under real-world flight conditions is poorly understood. Failing to separate and quantify these uncertainties limits the interpretability of predictive models and hinders the development of trustworthy battery health monitoring systems.

This study investigates the following research question: Can the RUL of eVTOL’s battery be predicted under variable mission profiles considering uncertainty? To address this, three specific aims are established: (1) generate degradation features to enhance predictive model performance, (2) develop a data-driven aging model tailored for eVTOL battery degradation, and (3) quantify both aleatoric and epistemic uncertainties, followed by a sensitivity analysis to identify influential factors affecting battery health. A Bayesian neural network framework is proposed to estimate the SOH of eVTOL batteries, taking advantage of its probabilistic modeling capabilities to capture inherent uncertainties in degradation patterns. The Sobol index is also employed for sensitivity analysis, identifying critical features that impact battery aging.

The model is validated using the measurements recorded from Sony-Murata 18650 VTC-6 cell lithium-ion batteries used to complete mission profiles for the VAHANA eVTOL. The proposed method outperforms traditional prognostics techniques by improving SOH prediction accuracy while providing valuable uncertainty quantification. Moreover, the insights gained from this approach contribute to risk evaluation and decision-making in eVTOL battery management, enhancing overall safety and reliability in advanced air mobility systems.

Presenting Author: Camilo Lopez-Salazar Texas Tech University

Presenting Author Biography: ..

Authors:

Camilo Lopez-Salazar Texas Tech University
Stephen Ekwaro-Osire Texas Tech Univ
Onur Can Kalay Texas Tech University