Regensburg 2025 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 28: Poster: Machine Learning, Data Science
DY 28.6: Poster
Wednesday, March 19, 2025, 15:00–18:00, P4
Advanced Framework for State of Health Estimation Using Equivalent Circuit Models and Machine Learning — •Limei Jin1,2, Franz Bereck2, Josef Granwehr2, Rüdiger-A. Eichel2, and Christoph Scheurer1,2 — 1Fritz-Haber-Institut der MPG, Berlin — 2IET-1, Forschungszentrum Jülich
Traditional Electrochemical Impedance Spectroscopy (EIS) techniques for characterizing a battery's behavior face several limitations, including time-consuming data collection, assumptions of system linearity, and difficulties in accurately assessing State of Charge (SoC) and State of Health (SoH). To address these challenges, we developed a robust framework for estimating SoH within a low-dimensional latent space using an autoencoder applied to raw time-domain battery data. This methodology combines synthetic training data from equivalent circuit models with machine learning techniques, specifically utilizing Chebyshev-based parameter space expansion to vary models on the SoC and SoH scale. Thereby, our framework effectively captures dynamic aging patterns while ensuring efficient data generation with minimal experimental input. Additionally, we introduced a stochastic pulse load profile to the models, which overcomes limitations of conventional frequency-based EIS measurements to better reflect real-world battery usage. This approach was initially validated on coin cell batteries in the lab, requiring only three standard spectroscopy experiments to train the framework. It will be extended to larger batteries, such as LFP batteries commonly used in automotive applications, offering scalable solutions for real-time monitoring and enhanced longevity.
Keywords: Equivalent Circuit Models; Latent Space; State of Health; Stochastic Pulse; Electrochemical Impedance Spectroscopy