Power Estimation for High-Performance Lithium-Ion Batteries

This thesis addresses the challenges associated with power estimation for lithium-ion batteries from both experimental and modeling perspectives. From a safety perspective, accurately determining whether the battery pack can provide the necessary power for a given task is crucial for applications with high power demand. This critical information is computed by the Battery Management System (BMS) by employing State of Power (SOP) estimation algorithms, responsible for providing the maximum power allowed within safety constraints of temperature, current, and voltage. In addition, constraints of State of Charge (SOC) might be imposed to avoid overcharging or overdischarging the battery. This thesis addresses two research gaps: an experimental gap and a implementation gap, as presented below. The experimental aspect of this thesis investigates the influence of SOC, temperature, stack pressure, and load history on power deliver. The challenge is that the experiments employ high-power pulses to measure the impact of each factor, which might introduce errors due to accelerated degradation. This thesis addresses this challenge by proposing a novel experimental methodology using Analysis of Variance (ANOVA) and Design of Experiments (DOE). The individual effects of factors were investigated, as well as the interaction between factors. This research shows that the cell-to-cell variations are more significant than the cell degradation for the Lithium Cobalt Oxide (LCO) pouch cell tested. Furthermore, temperature and SOC affect power output with statistical significance, in contrast to stack pressure and load history. The second aspect of this research addresses the real-time implementation of SOP algorithms. Two algorithms were investigated using software-in-the-loop (SIL), and processor-in-the-loop (PIL) environments deployed in an Arm® Cortex®-M4F microprocessor. The first algorithm is based on Model Predictive Control (MPC). Power estimation is computed as an optimal, dynamic power profile within a prediction horizon of 30 seconds. The challenge is the high computational cost to solve the Quadratic Programming (QP) problem formulated within the MPC framework. This research expands upon the literature by proposing an alternative objective function for the QP problem, which reduced the computational cost enabling the real-time implementation in the low-cost Cortex®-M4F microprocessor. Finally, this research also enhanced the real-time deployment of a Constant Current (CC) SOP algorithm. By employing a secant root-finding method instead of the traditional bisection method, the execution time of the algorithm was reduced by two-thirds.

Grantor dc:publisher

Oxford Brookes University

Author dc:creator

• Schommer, Adriano

Contributors dc:contributor

• Collier, Gordana
• Morrey, Denise

Statement dc:rights

• All rights reserved

Language dc:language

en

DOI dc:identifier

https://doi.org/10.24384/cyd2-pz34

OAI identifier oai:identifier

tle:74d20161-b820-476a-8421-ac99c5b5e3b5:d6bd9758-527a-46cd-bfe2-c433766e8fca:1