Dataset: Impact of electrolyte impurities and SEI composition on battery safety

Li-ion batteries have a potential risk of thermal runaway. Current safety evaluations in academia and industry rely on experiments or semi-empirical simulations. This limits the understanding of processes leading to or occurring during thermal runaway, and how chemical species and impurities can impact them. The limited (quantitative) understanding in turn hinders a holistic safety assessment and optimisation of countermeasures through design or operation. The here presented thermal-runaway model contains a detailed degradation reaction network, which allows the impact of chemical species and impurities on thermal runaway to be studied. We set a particular focus on water impurities and solid-electrolyte interphase (SEI) properties, as both are known to impact life-time of batteries. SEI composition and thickness change during ageing, which is shown here to impact battery safety significantly. The model can reproduce reported experimental behaviour: aged cells are more safe, as they start self-heating, i.e. heat production without an external heat source, at 15–20 °C higher temperatures than fresh cells. Our model suggests a thick inorganic, and thus less reactive SEI as the underlying cause. Furthermore, we could show that extensive electrode drying to remove water impurities before building battery cells will not significantly improve safety characteristics. In contrast, electrodes not subjected to any drying procedure cause an earlier start of the self-heating phase, i.e. have a higher risk of thermal runaway. These insights into the sensitivity to thermal runaway allow robust methods to be tailored for its prevention, from controlling battery and SEI properties during production to adjusting safety assessment for effects of ageing.

The published source data file contains the underlying data of all figures of the manuscript of the publication “Impact of electrolyte impurities and SEI composition on battery safety”. The source data of each figure is presented in a single sheet named after the corresponding figure. Legends are given if required. Numerical details: The model was implemented in MATLAB R2022a and differential equations were solved with the ode15s solver. The calculations were performed on an i7-9750H CPU with 16 GB RAM. For detailed information on the modelling approach see the Method section in the manuscript and the Supplementary Information.