Abstract

Lithium-ion batteries are an increasingly popular source of power for many electric applications. Applications range from electric cars, driven by thousands of people every day, to existing and future air vehicles, such as unmanned aircraft vehicles (UAVs) and urban air mobility (UAM) drones. Therefore, robust modeling approaches are essential to ensure high reliability levels by monitoring battery state-of-charge (SOC) and forecasting the remaining useful life (RUL). Building principled-based models is challenging due to the complex electrochemistry that governs battery operation, which would entail computationally expensive models not suited for prognosis and health management applications. Alternatively, reduced-order models can be used and have the advantage of capturing the overall behavior of battery discharge, although they suffer from simplifications and residual discrepancy. We propose a hybrid solution for Li-ion battery discharge and aging prediction that directly implements models based on first-principle within modern recurrent neural networks. While reduced-order models describe part of the voltage discharge under constant or variable loading conditions, data-driven kernels reduce the gap between predictions and observations. We developed and validated our approach using the NASA Prognostics Data Repository Battery dataset, which contains experimental discharge data on Li-ion batteries obtained in a controlled environment. Our hybrid model tracks aging parameters connected to the residual capacity of the battery. In addition, we use a Bayesian approach to merge fleet-wide data in the form of priors with battery-specific discharge cycles, where the battery capacity is fully available (complete data) or only partially available (censored data). The model's predictive capability is monitored throughout battery usage. This way, our proposed approach indicates when significant updates to the hybrid model are needed. Our Bayesian implementation of the hybrid variational physics-informed neural network can reliably predict the battery's future residual capacity, even in cases where previous battery usage history is unknown.

Notes

If this is your thesis or dissertation, and want to learn how to access it or for more information about readership statistics, contact us at STARS@ucf.edu

Graduation Date

2022

Semester

Spring

Advisor

Viana, Felipe

Degree

Doctor of Philosophy (Ph.D.)

College

College of Engineering and Computer Science

Department

Mechanical and Aerospace Engineering

Degree Program

Mechanical Engineering

Format

application/pdf

Identifier

CFE0008982; DP0026315

URL

https://purls.library.ucf.edu/go/DP0026315

Language

English

Release Date

May 2022

Length of Campus-only Access

None

Access Status

Doctoral Dissertation (Open Access)

Share

COinS