Digital Twin for Lithium-Ion Batteries Using Singular Value Decomposition

Abstract

Real-time thermal management of lithium-ion batteries remains a critical challenge in electric vehicle applications. This study explores the development of a real-time electrothermal digital twin to efficiently predict thermal and electrical behaviors, thereby improving battery thermal management systems. Our approach integrates numerical thermal modeling, model order reduction techniques, and equivalent circuit models. We begin with full numerical model simulations of a cylindrical battery cell to capture thermal gradients and heat generation patterns. To reduce computational complexity while preserving accuracy, singular value decomposition is applied to the simulation data, identifying dominant thermal modes and enabling model order reduction. Subsequently, a state-space model is developed that accurately reproduces the thermal dynamics at the system level. This reduced-order model is then integrated with an equivalent circuit model to create an electrothermal digital twin. The resulting digital twin demonstrates considerable computational efficiency, reducing simulation time from hours to seconds compared to the full-order model approaches, while maintaining high accuracy in predicting both thermal and electrical behaviors. The model's performance is validated using the New European Driving Cycle, a standardized test representing typical driving conditions. The validation results indicate strong agreement between the digital twin predictions and the full thermal model, with a maximum error of 0.12% for the average cell temperature. The model achieves real-time performance, with an average computation time of 0.012 s per time step, making it suitable for integration into advanced battery management systems.

Keywords: Digital twin Lithium-ion batteries Model order reduction Numerical modeling Singular value decomposition State-space model