Parameterization, simulation and analysis of electrode-level physics-based battery model
Open Access
- Author:
- Wojnar, Sara
- Area of Honors:
- Mechanical Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Satadru Dey, Thesis Supervisor
Daniel Humberto Cortes Correales, Thesis Honors Advisor - Keywords:
- Li-ion
electrode-level
battery model
machine learning
aging prediction - Abstract:
- An accurate and a computationally efficient Lithium-ion battery model is essential for analyzing battery performance and controlling batteries in real-time. Many battery models that describe cell-level dynamics can be found in existing literature, while electrode-level models have been explored with less frequency. However, investigating the dynamics at the electrode-level can lead to greater insights to identify dominant aging mechanisms and areas where battery designs can be altered in order to improve battery performance and life. In this context, this thesis presents a modeling framework for battery electrode-level dynamics by combining physics-based battery models with data-driven learning models. The modeling framework intends to capture and identify the effect of several uncertainties arising from model inaccuracies and aging related behaviors. Specifically, a single particle model is utilized to capture the Lithium diffusion phenomena in a positive electrode whereas a Gaussian process regression model is used to capture the inaccuracies of the single particle model. Furthermore, a least squares estimator is used to estimate and predict the changes in positive electrode behavior due to cycling-induced aging. The proposed framework is tested using measured data from Lithium-ion battery cells that underwent 90 cycles of continuous charging and discharging. Applying the data-driven technique to the physics-based electrode-level model better predicted the fresh cell voltage of the electrode. The cycle charging time was used to accurately predict the maximum fresh cell voltage uncertainty using a linear fit, although there was more variation in the maximum cell voltage uncertainty in later cycles.