The  Solid  Electrolyte  Interphase (SEI)  is  formed  on  the  electrode  as  the  reduction  product  of  electrolyte. The better understanding of SEI configurations helps to adjust its properties for optimum battery performance and safety.  The complexity in capturing the physicochemical properties of the evolving SEI by the traditional modeling approaches requires the implementation of virtual material design based on data driven approach.  This work aims to develop a workflow based virtual material design to characterize the better SEI configurations with respect to its properties to distinguish them based on reduction species of electrolyte contributing to target SEI configuration.  In virtual material design, Variational Auto Encoders (VAEs) are emerging as powerful tools to investigate complex, high-dimensional, and non-convex design spaces.  This work combines a VAE with a regressor model to study the Solid Electrolyte Interphase (SEI) formation on the electrode in a battery,  computed by Kinetic Monte Carlo simulations.  The encoder of the VAE maps the two dimensional SEI configurations to a lower dimensional, continuous latent space and the additional regressor learns to map that latent space to the SEI properties.  From this  latent  space,  simple  vector  operations,  the  VAE  decoder,  and  the  regressor  allow  tuning towards optimal SEI configurations and properties.

Aknowledgements:

This work is supported by the “Cluster of Excellence” POLiS of the Deutsche Forschungsgemeinschaft and by the FestBatt competence cluster of the BMBF.