Balancing accuracy and simplicity in battery modelling
Muiz Sheikh defended his PhD thesis at the Department of Electrical Engineering on 30 June.
Electric cars and bikes, homes, factories, and even entire power grids increasingly rely on lithium-ion batteries to store and release energy. All of those batteries use a battery management system (BMS) that determines how fast the battery charges and when it should rest, ensuring it remains safe, efficient, and long-lasting. To make these decisions, the BMS needs accurate information about what is happening inside the battery, which is provided by a mathematical model. However, it is challenging to develop models that both capture all relevant chemical details and run fast enough on small chips. In his research, Muiz Sheikh develops a modelling framework that finds the optimal balance between accuracy and simplicity in models for practical applications.
Muiz Sheikh鈥檚 framework makes it possible to develop simple models that contain only a few mathematical terms while still covering a wide range of battery operating conditions, from fully charged to empty and from cold to hot temperatures. These models, known as linear parameter-varying (LPV) models, are derived from battery measurement data. Sheikh鈥檚 research shows that these models perform remarkably well. For example, they remain accurate across a temperature range of 0 to 40 掳C and even when the battery is nearly empty. Most models tend to drift under these conditions, but the models developed in this research do not.
SOC estimation
LPV models can be used for a variety of important tasks, such as state-of-charge (SOC) estimation. SOC essentially functions as the battery's fuel gauge. Although it can be inferred from measured battery data, the accuracy depends heavily on the methods used. Notably, the methods presented in this research deliver reliable SOC estimates even when the measurements are of poor quality. They also perform well for lithium iron phosphate (LFP) batteries, for which SOC estimation is notoriously difficult.
Aging batteries
Another important task is monitoring battery health as it ages. This research also demonstrates a method for inferring descriptions of a battery's health from its everyday operating data, similar to how a blood test reveals the state of a human body. This approach show how far a battery has progressed through its lifetime.
Open-source tools
To ensure real-world applicability beyond academia, all methods have been packaged into two open-source Python tools: PyBatteryID and PyBatterySE. Both are freely available on GitHub. They have been thoroughly tested on a wide variety of battery cells and consistently outperform existing approaches.
Title of PhD thesis: . Supervisors: Dr. Tijs Donkers and Prof. Henk Jan Bergveld.