Machine-learning based Li-Ion Cell state prediction using Impedance spectroscopy

Accurate and reliable monitoring of battery state parameters is crucial for ensuring optimal battery performance, safety, and lifetime. Existing methods have limitations, such as requiring modeling of each degradation mechanism involved or relying on direct measurement techniques that impose restrictions on field studies or end-user use. In this paper, we propose a machine learning-based approach that combines the strengths of electrochemical impedance spectroscopy (EIS) and machine learning algorithms to predict battery state parameters. We have developed an efficient prediction system that can learn from EIS data and accurately predict battery state parameters. Our approach is trained on an open dataset comprising of over 30,000 spectra, generated using an automated measurement technique that outperforms current machine learning-based models, particularly in terms of generalization across different cells and measurement setups. • We show an approach using EIS and CNNs that outperforms previous studies in predicting Li-Ion battery state parameters. • We present a new open dataset of over 30,000 EIS spectra, providing a valuable resource for the research community. the largest of its type. • We demonstrate the effectiveness of our method in predicting capacity, particularly in terms of generalization across different cells and measurement setups. • We provide rich new open source tools for working with and simulating EIS datasets, as well as for exploring machine learning algorithms on them.