Machine learning 3D-resolved prediction of electrolyte infiltration in battery porous electrodes

Electrolyte infiltration is one of the critical steps of the manufacturing process of lithium ion batteries (LIB). We present here an innovative machine learning (ML) model, based on the multi-layers perceptron (MLP) approach, to fast and accurately predict electrolyte flow in three dimensions, as well as wetting degree and time for LIB electrodes. The ML model is trained on a database generated using a 3D-resolved physical model based on the Lattice Boltzmann Method (LBM) and a NMC electrode mesostructure obtained by X-ray micro-computer tomography. The trained ML model is able to predict the electrode filling process, with ultralow computational cost and with high accuracy. Also, systematic sensitivity analysis was carried out to unravel the spatial relationship between electrode mesostructure parameters and predicted infiltration process characteristics. This paves the way towards massive computational screening of electrode mesostructures/electrolyte pairs to unravel their impact on the cell wetting and optimize the infiltration conditions.

Automated summary

An innovative machine learning model based on MLP approach has been proposed to predict electrolyte flow and wetting degree in LIB electrodes accurately and quickly. The trained ML model, along with systematic sensitivity analysis, provides a new way to optimize electrode wetting and infiltration conditions.