Loading...
PHYSICS-INSPIRED AND CONTROL-ORIENTED MODELING OF LITHIUM BATTERIES FOR ACCURATE STATE-OF-CHARGE PREDICTION AND FAST-CHARGING
Citations
Altmetric:
Genre
Thesis/Dissertation
Date
2025-05
Advisor
Committee member
Group
Department
Mechanical Engineering
Permanent link to this record
Collections
Research Projects
Organizational Units
Journal Issue
DOI
https://doi.org/10.34944/a40n-mk42
Abstract
This dissertation presents a physics-inspired, data-driven framework for lithium-ion battery modeling and control, designed to enhance state prediction accuracy and optimize fast-charging strategies across diverse operating conditions. It addresses key challenges in battery management systems for critical applications such as electric vehicles, focusing on improving performance, adaptability, and safety under extreme temperatures, high charge/discharge currents, and varying states of charge. Conventional methods often fail to fully utilize battery capacity, particularly in low-temperature conditions, resulting in reduced performance and limited range. The research proposes a novel methodology combining data-driven techniques with physics-based insights to overcome these issues. A key innovation is the development of PhITEDD (Physics-Informed Temperature Dependent Explicit Data-Driven), a framework that enhances model accuracy, interpretability, and generalizability while reducing reliance on proprietary knowledge.
The PhITEDD framework combines physics-inspired features that connect the model to underlying battery processes, a Monte Carlo search algorithm for exploring extensive feature spaces, and an automated hyperparameter tuning mechanism. This approach strikes an optimal balance between model accuracy and complexity by quantifying individual feature contributions, enabling robust state prediction across a wide range of operating conditions. Its digital twin for state-of-charge forecasting achieves prediction errors below 1% using experimental drive cycle data while maintaining high accuracy for unseen aggressive drive cycles and varying operating conditions.
A key innovation of the framework is its temperature-dependent recalibration method, which adjusts model coefficients to optimize performance under new operating conditions, ensuring consistency across the full SOC (0%-100%) and temperature (−20°C to 40°C) range. The framework also investigates the impact of data sampling rates on model accuracy, providing practical guidelines for optimization. These advancements collectively enhance the interpretability, efficiency, and practicality of lithium-ion battery models, supporting improved battery utilization and extended lifespan.
The dissertation also addresses the fast-charging optimization problem using a direct data-driven control method. This strategy learns the battery’s Jacobian from input/output data to optimize the charging current profile, minimizing charging time while adhering to safety constraints such as maximum cell temperature and voltage. The data was generated using a full-order electrochemical Doyle-Fuller-Newman model integrated with a thermal model. The optimal solution comprises a hybrid charging strategy that charges a 5Ah NMC-811 cylindrical cell 66% faster than the standard CCCV method while ensuring safety limits like 4.2V and 57°C. This approach closely aligns with actual battery mechanisms.
By improving efficiency and safety, the proposed methodology has significant implications for the performance and adaptability of batteries in electric vehicles and other critical applications. The work reduces dependency on proprietary models, enhancing the accessibility and applicability of battery modeling tools. In summary, this dissertation advances Li-ion battery research by integrating physics-informed and data-driven methods, resulting in innovative modeling and control strategies for accurate SOC prediction and optimal fast charging under complex operating conditions.
Description
Citation
Citation to related work
Has part
ADA compliance
For Americans with Disabilities Act (ADA) accommodation, including help with reading this content, please contact scholarshare@temple.edu