(Invited) Artistic: An AI and Modeling-Supported Digital Infrastructure for Battery Manufacturing Optimization

Abstract

Energy, power, lifetime and safety of Lithium ion batteries (LIBs) are determined in part by the internal microstructure of the electrodes. That microstructure is characterized by the spatial location and the resulting interfaces between the active material particles, the conductive additive, the binder and the pores. The microstructure is determined by the manufacturing process of the electrodes, which encompasses, in its traditional wet form, multiple steps and numerous parameters. Therefore, its optimization is difficult and can be time consuming and costly if carried out only on the basis of a trial and error approach.In this lecture I discuss a digital infrastructure for accelerated optimization of the manufacturing process of LIBs that we are developing since few years: ARTISTIC.1 Such digital infrastructure is supported on a hybrid approach encompassing a physics-based multiscale modeling workflow, Artificial Intelligence (AI)/machine learning models and high throughput experimental characterizations.2 ARTISTIC simulates different steps along the battery cells manufacturing process, such as the electrode slurry preparation, the coating, the drying, the calendering and the electrolyte infiltration. For that purpose, it uses a combination of sequentially-coupled 3D-resolved physical models based on experimentally-validated Coarse Grained Molecular Dynamics, Discrete Element Method and Lattice Boltzmann Method. This allows predicting the impact of the manufacturing process parameters on the final electrode microstructure in three dimensions. The predicted electrode microstructures are then injected in a continuum performance simulator capturing the influence of the electrode microstructure on the solid electrolyte interphase formation (for anodes) and the electrochemical response (of anodes vs. lithium, cathodes vs. lithium and the full cells). Machine learning models are used to unravel manufacturing parameters interdependencies from the physical models’ predictions and experimental data, and to suggest manufacturing process optimizations. The predictive capabilities and chemistry-neutrality of the ARTISTIC digital infrastructure are illustrated with results for different LIB electrode formulations and chemistries (NMC, LFP, graphite, silicon/graphite blends) and also for organic active materials for sodium ion batteries. Finally, I discuss the free online battery manufacturing simulation services (online caculator and associated databases) offered by our project and that you can use interactively from your Internet Browser.3