a Hybrid framework for Remaining Useful Life and State of Health prediction of eVTOL lithium-ion batteries.

Abstract

This thesis investigates the prognostics for eVTOL lithium-ion batteries by comparing a GPR, an MCMC and a hybrid framework consisting of the two to estimate the State Of Health (SOH) and Remaining Useful Life (RUL). We shall use the public eVTOL dataset of Sony-Murata 18650 VTC-6 battery cells. Performance was then assessed using prediction accuracy metrics and uncertainty quantifi- cation metrics. On average, the GPR yields the least errors in terms of both categories of SOH and RUL prediction. And it performs substantially better in terms of RUL prediction. However, in the face of big aleatoric uncertainty, it is outperformed by the other methods. We conclude that in the face of abundant data, the GPR is best used for prediction, but when the SOH data shows big outliers, a model-based method or hybrid method is best used.