Elsevier

Journal of Power Sources

Volume 483, 31 January 2021, 229148
Journal of Power Sources

Mapping the architecture of single lithium ion electrode particles in 3D, using electron backscatter diffraction and machine learning segmentation

https://doi.org/10.1016/j.jpowsour.2020.229148Get rights and content
Under a Creative Commons license
open access

Highlights

  • Mapping the 3D architecture of NMC particles with FIB–EBSD.

  • Grain-wise segmentation of FIB–EBSD data for NMC particles, using neural networks.

  • Computation of size and shape characteristics for each grain.

  • Copula models for multivariate distributions of grain characteristics.

  • Efficient description of the grain architecture of NMC particles.

Abstract

Accurately quantifying the architecture of lithium ion electrode particles in 3D is critical to understanding sub-particle lithium transport, rate limitations, and degradation mechanisms within lithium ion batteries. Most commercial positive electrode materials consist of polycrystalline particles, where intra-particle grains have a range of morphologies and orientations. Here, focused ion beam slicing in sequence with electron backscatter diffraction is used to accurately quantify intra-particle grain morphologies in 3D. The intra-particle grains are identified using convolution neural network segmentation and distinctly labeled. Efficient morphological characterization of the grain architectures is achieved. Bivariate probability density maps are developed to show correlative relationships between morphological grain descriptors. The implication of morphological features on cell performance, as well as the extension of this dataset to guide artificial generation of realistic particle architectures for 3D multi-physics models, is discussed.

Keywords

Convolutional neural network
Statistical image analysis
Model fitting
Copula
Lithium ion battery
Electron backscatter diffraction

Cited by (0)