Lithium-Ion-Batteries are the key technology for the electrification of the mobility industry and the transition to a sustainable transportation sector. Because of their high energy and power density in combination with a long cycle life, they have taken on a dominant role in the field of electrochemical energy storage systems. In order to optimize the operation of the batteries and to ensure safe and efficient charge and discharge cycles, it is, however, crucial to determine the condition of the battery at all times. While battery parameters as voltage, current, and temperature are measurable, conditions like the state-of-charge or the state-of-health are not directly measurable. This work presents new condition determination techniques for lithium-ion-batteries based on machine learning algorithms. For the development of all models, the whole process from data selection, over data preprocessing to model evaluation is taken into account. At first, the determination of the state-of-charge is optimized by means of artificial neural networks and by creating artificial data points for a robust model training. The lowest mean absolute error for the test dataset could be obtainend with a value of 0,44% based on a convolutional neural network. For the estimation of the state-of-health, a single model, in this case a gaussian process regression, for three different types of lithium-ion-batteries is developed. The mean absolute error for the testing dataset is 1,33%. Both models, for the state-of-charge and for the state-of-health estimation, aim to find a function to determine the condition based on the battery parameters voltage, current, and temperature. Lastly, the change of the impedance over the cycle life of batteries is analyzed. The aging experiments result in over 25000 impedance spectra over the degradation of lithium-ion-batteries. The state-of-charge with the highest correlation between capacity and impedance is identified at a state-of-charge of 10%. After selecting the impedance at different measurement frequencies based on their correlation to the capacity, a support vector regression model is trained with the impedance values to estimate the state-of-health. The model shows a mean absolute error of 0,86% and validates the correlation results. In comparison to the other model for the state-of-health estimation, the impedance is used instead of the battery parameters voltage, current, and temperature. The proposed approaches demonstrate the high potential of data-driven models to estimate the condition of a battery and can significantly improve the condition monitoring of the battery management system. This is crucial for further advancements for stationary and mobile applications, and therefore for the overall move to a sustainable mobility and efficient energy storage solutions in the future.     «

Lithium-Ion-Batteries are the key technology for the electrification of the mobility industry and the transition to a sustainable transportation sector. Because of their high energy and power density in combination with a long cycle life, they have taken on a dominant role in the field of electrochemical energy storage systems. In order to optimize the operation of the batteries and to ensure safe and efficient charge and discharge cycles, it is, however, crucial to determine the condition of th...     »