A Stochastic Microstructure Reconstruction-Based Mechanical and Transport Modeling Approach for Learning the Microstructure-Property Relationship of Li-Ion Battery Graphite Anodes

Abstract

Understanding the relationships between microstructural characteristics and multiphysics properties is key to designing battery electrode materials for desired properties. Significant efforts have been made to achieve a quantitative understanding of the relationship between mass transport properties and Li-ion microstructure characteristics [1-3] in literature, but it is still challenging to predict the coupled mechanical and electrochemical behaviors based on microstructure characteristics.We seek to further this understanding through an investigation of how deformation affects the transport properties of the Li-ion battery graphite anode microstructure, which features packed particles and irregular pore networks. We propose a statistical Microstructure Characterization and Reconstruction (MCR) approach to characterize statistical microstructure features from 2D microscopic images and then reconstruct 3D stochastic microstructures. MCR is an effective tool for material property prediction [4] and microstructure-mediated material design [5]. The proposed MCR approach generates microstructure designs that are beyond the scope of empirical data.By exploring the input microstructure feature space, 3D microstructure samples are reconstructed and simulated to investigate relationships between the microstructural characteristics and properties of Li-ion battery graphite anodes. The selection of microstructure characteristics is informed by an external open access battery microstructures library. For each microstructure reconstruction, compression and transport simulations are conducted to determine the Young’s modulus and to understand the transport properties (e.g. diffusivity) of the undeformed and deformed microstructures. Convergence studies are conducted to establish a Representative Volume Element (RVE) size. With the simulation dataset, the microstructure-property relationship is examined. A feature selection algorithm is used to examine the size of the effects of microstructural characteristics on multiphysics properties. Machine learning models are established to predict the microstructure-property relationship.