Lithium-Ion Battery Electrode Manufacturing Model Using X-Ray Computed Tomography Characterization Data

Nowadays, in face of the increasing need for lithium-ion batteries (LIBs), how to achieve higher energy densities while maintaining or reducing costs has been widely studied. To achieve optimization of the performance of batteries, it is essential to understand the influence of parameters at each stage of the LIBs manufacturing process on the architectures of the electrodes, which affects the energy, power, lifetime and safety of the LIB cells.<br/>Non-destructive techniques, such as X-ray computed tomography (XCT), are widely used to obtain the microstructure of electrodes, enabling volume-based 3D characterization. However, the characterizing the spatial location of the carbon binder domain (CBD) in the electrode volume is challenging due to the nano-features and the similarities between X-ray attenuation coefficients of carbon and pores. Extensive works have been done recently to improve the characterization of CBD, such as combining separate scans of high-attenuating LiNi_xMn_yCo_1-x-yO₂ (NMC) and low-attenuating CBD¹, using X-Ray holographic nano-tomography², and using convolutional neural network-based image segmentation methods³.<br/>The ARTISTIC project⁴, funded by the European Research Council, aims at developing a digital twin that enables predicting the electrode architectures and their electrochemical performances from the parameters used at each stage of the LIBs manufacturing process. Such a digital twin is supported by a combination of multiscale modeling^5-9and machine learning (ML)¹⁰. In this work, we present our new computational approach to simulate the manufacturing process of cathodes by taking into account CBD and the realistic shape of LiNi_0.33Mn_0.33Co_0.33O₂(NMC111) active material particles obtained by X-ray micro-computational tomography (μ-XCT)¹¹. Our methodology encompasses coarse-grained molecular dynamics (CGMD) to simulate the different manufacturing stages, from the slurry, to the drying and the calendering of the resulting electrodes. The μ-XCT data is segmented through a Random Forest algorithm available in the Image J Trainable Weka segmentation plugin to extract the active material (AM) phase. Then a Watershed-based algorithm is used to separate the individual secondary particles. The input of CGMD model is prepared by substituting the down-sampled voxels of the individual particles by spherical primary AM particles with equivalent volume. The new model captures the variation of the electrode microstructure in the manufacturing process from the particle scale, and the effect of the manufacturing parameters on the electrode heterogeneity. We discuss the changes in porosity, tortuosity factor, electronic conductivity, different phases distribution and pore network during the calendering process and compare them with the μ-XCT results, which show reasonable agreements. In addition, our approach allows capturing the deformation of the secondary particles during the calendering process, as well as their possible variation of orientation.<br/><br/>Reference:<br/>1. X. Lu et al., Nat Commun 2020, 11, 2079.<br/>2. T. Nguyen et al., Adv. Energy Mater. 2021, 11, 2003529.<br/>3. Z. Su et al., npj Comput Mater 2022, 8, 30.<br/>4. "ERC Artistic: Home", can be found under http://www.erc-artistic.eu/<br/>5. T. Lombardo et al., Batteries & Supercaps 2020, 3, 721.<br/>6. T. Lombardo et al., Energy Storage Materials 2021, 43, 337.<br/>7. A. C. Ngandjong et al., Journal of Power Sources 2021, 485, 229320.<br/>8. A. Shodiev et al., Energy Storage Materials 2021, 38, 80.<br/>9. C. Liu et al., Journal of Power Sources 2021, 512, 230486.<br/>10. M. Duquesnoy et al., ArXiv, 2022. arXiv:2205.01621<br/>11. J. Xu et al., ArXiv, 2022. arXiv:2206.03969