Fast charging and the service life of electric batteries are important prerequisites for the wider application of electric vehicles. For this purpose, the so-called battery management system offers great potential for optimization. Since too little is known about the interrelationships between these two requirements and the internal battery parameters, the battery management systems currently in use "play it safe" and often impose unnecessarily strict safety restrictions on the operation of the batteries.
Together with partners at the Forschungszentrum Jülich and the Fritz Haber Institute in Berlin, we have set ourselves the goal of developing a novel intelligent battery management system that can make better decisions about battery charging cycles using a detailed surrogate model ("digital twin") of the battery and artificial intelligence (AI). Our task in the project is to develop this battery replacement model at the level of the individual battery cells using the Gaussian process method. In doing so, the model is parameterized with parameters of different equivalent circuits and the state of charge of each cell. The model is trained using both experimental spectroscopy data and data from physical equivalent circuit models. In addition, the use of the Gaussian process allows for a determination of the model uncertainty (Uncertainty Quantification), which is required for the AI to function.
In a first test phase, the battery management system equipped with the AI will be implemented on simple hardware. During this process, data will still be collected and stored during operation so that the AI can continue to improve. In the final phase, the AI battery management system will be tested with a fully characterized battery.
The i2Batman project is one of 19 funded projects of the Helmholtz Artificial Intelligence Cooperation Unit.
Contact at the SCC: Dr. Ivan Kondov
(Translated with DeepL.com)