Iryna Zenyuk1,Ying Huang1,Jermone Quenum2,3,Daniela Uschizima2,3,4
University of California, Irvine1,Lawrence Berkeley National Laboratory2,University of California, Berkeley3,University of California, San Francisco4
Iryna Zenyuk1,Ying Huang1,Jermone Quenum2,3,Daniela Uschizima2,3,4
University of California, Irvine1,Lawrence Berkeley National Laboratory2,University of California, Berkeley3,University of California, San Francisco4
This scientific study presents a novel computational approach to inspecting and quantifying the durability of batteries using high-resolution X-ray data obtained from the Argonne National Laboratory Advanced Photon Source (ANL APS) and Lawrence Berkeley National Laboratory Advanced Light Source (LBNL ALS). The focus of the investigation was on lithium metal batteries (LMB), which are prone to lithium dendrite formation, impacting battery efficiency and posing safety hazards. The study employed in-situ and operando imaging techniques, including 3D microtomography, to detect battery defects and monitor the dynamics of lithium plating during cycling experiments. A multiclass semantic segmentation method based on the Iterative Residual U-net architecture was proposed for accurately identifying lithium plating dynamics. The computations were performed using the NERSC Perlmutter high-performance systems, exploiting both CPU, GPU, and large memory nodes. The results demonstrated the successful application of the proposed computational tool, batteryNET, for semantic segmentation and classification of different phases in the battery after cycles of charge and discharge. The analysis included volume quantification of different phases, spatial correlation among different components at different points during the cycling and renderings of the semantic segmentation results, showcasing the effectiveness of the approach. Ongoing and future work includes exploring larger datasets and new semantic segmentation algorithms based on vision transformers and other transformer-based deep learning architectures for the detection of battery defects using semantic segmentation techniques.<br/><br/>References<br/>[1] Ushizima, Huang, Quenum, Perlmutter, Parkinson, Zenyuk, “Lithium Metal Battery Characterization using X-ray Imaging and Machine Learning”, American Physical Society Meeting, USA 2022.<br/>[2] Huang, Perlmutter, Su, Quenum, Shevchenko, Parkinson, Zenyuk, Ushizima, “Detecting lithium plating dynamics in a solid-state battery with operando X-ray computed tomography using machine learning”. Nature Partner Journal Computational Materials 2023.<br/>[3] Quenum, Perlmutter, Huang, Zenyuk, Ushizima, “Lithium Metal Battery Quality Control via Transformer-CNN Segmentation”, Journal of Imaging 2023.