A Hybrid Modelling Approach Coupling Physics-Based Simulation and Deep Learning for Battery Electrode Manufacturing Simulations

The future projection concerning Lithium-ion Battery indicates the necessity to find ways to accelerate the optimization of the electrode's manufacturing process. The traditional solvent-based manufacturing process includes the optimization of several steps and their complex relationships to produce an optimized electrode. In state of the art, this optimization is heavily based on trial-and-error approaches, resulting in high resource utilization, high scrap rates, high energy consumption, and high emissions. Previous research efforts have combined empirical and traditional computational methods to understand these complex relationships to further simplify the optimization process. However, these methods themselves are demanding in terms of resources such as materials or computational effort. In this work, we proposed a novel supervised Deep Learning (DL)-based workflow that couples with physics-based cathode manufacturing simulation frameworks previously reported by us¹^,² further reducing computational cost in synthetic electrode microstructure production, and showing generalizability to various electrode chemistries. The trained DL model, integrates well into the physics-based simulation framework and shows good performance in reducing simulation time and forecasting Nickel Manganese Cobalt Oxide (in the ratio of 1:1:1), Lithium Iron Phosphate, and All-Solid-State Battery based cathode slurry microstructures with 3D spatial resolution. The proposed novel hybrid approach, combining DL with physics-based manufacturing simulations, can successfully predict how slurry formulation and solid content influence its microstructure and experimental properties, such as density and viscosity as a function of the shear-rate. The resulting slurry microstructures are in turn used to inform a physics-based simulation pipeline of the drying and the calendering to predict electrode microstructures (defined by the spatial arrangement of the materials particles and pores) and their associated functional properties, such as the porosity and electrical conductivities. Our hybrid electrode manufacturing simulation approach promises to streamline lab-scale electrode manufacturing and reduce errors, waste, and resource consumption³.<br/><br/>References<br/>[1] M. Duquesnoy, C. Liu, D. Z. Dominguez, V. Kumar, E. Ayerbe, and A. A. Franco, “Machine learning-assisted multi-objective optimization of battery manufacturing from synthetic data generated by physics-based simulations,” <i>Energy Storage Materials</i>, vol. 56, pp. 50–61, Feb. 2023, doi: 10.1016/j.ensm.2022.12.040.<br/>[2] C. Liu, T. Lombardo, J. Xu, A. C. Ngandjong, and A. A. Franco, “An experimentally-validated 3D electrochemical model revealing electrode manufacturing parameters’ effects on battery performance,” <i>Energy Storage Materials</i>, vol. 54, pp. 156–163, Jan. 2023, doi: 10.1016/j.ensm.2022.10.035.<br/>[3] U. Vijay, D.E. Galvez-Aranda, F.M. Zanotto, T. L. Dinh, M. Alabdali, M. Asch, and A. A. Franco, “A Hybrid Modelling Approach Coupling Physics-based Simulation and Deep Learning for Electrode Manufacturing Simulations,” (to be submitted 2024).