TALENT EDUCATION

Pan Feng’s research group has made progress in constructing cross-scale machine learning force field models based on graph theory and AI and revealing the morphological evolution of lithium dendrites


Due to its high energy density, lithium-ion batteries have gained wide applications in electric vehicles and large-scale energy storage, while lithium metal anodes are considered to be one of the ideal for the next generation of anode materials due to their low electrode potential and high theoretical specific capacity. However, under actual working conditions, the dendrite growth problem of lithium metal anode will not only reduce the coulombic efficiency of the battery, but also may bring serious safety risks.

Professor Pan Feng’s research group of the School of Advanced Materials of Shenzhen Graduate School of Peking University and Professor Wang Linwang, Chief Scientist of the Institute of Semiconductors, Chinese Academy of Sciences, developed a set of machine learning force field construction strategies based on graph theory and AI for cross-scale morphology simulation, and applied them to the simulation of lithium dendrite morphology evolution in electrolyte environment. The simulation results reveal the two-stage process of lithium dendrite evolution, and analyze the driving forces behind the morphology evolution in detail. This study points out that surface energy has a significant influence on the morphological evolution of lithium dendrites, which has great reference value for further promoting the development of lithium metal anodes. The relevant research results were recently published in the top international journal Advanced Energy Materials (DOI: 10.1002/aenm.202202892, IF=30).

Based on the energy splitting method, this work realizes the construction of end-to-end machine learning force field model by localizing the total energy of the system in the traditional DFT calculation results. Furthermore, an implicit solvent model is introduced in the dataset generation process, and the in-situ cross-scale molecular dynamics simulation of lithium dendrite morphology in the electrolyte environment is realized by machine learning force field. In order to realize the leap from small-scale dataset to large-scale model application, this paper proposes a set of cross-scale active learning schemes and a cross-scale model accuracy verification method combined with energy splitting method. The above method provides a solution for the study of atomic-precision dynamics of large-scale morphology problems.

Machine learning force field dataset generation and model building

Based on the molecular dynamics simulation results of machine learning force field, this work summarizes the two-stage morphological evolution process of lithium dendrites and observes the kinking phenomenon consistent with the experiment. The kinetic simulation results show that this phenomenon occurs due to the tendency of different crystal domains to slip along grain boundaries. Furthermore, this paper analyzes the different morphological evolution stages in detail, and determines that surface energy and grain boundary energy are the main driving forces of morphology evolution. In addition, the polycrystalline structure ensures the stable existence of lithium dendrites, and how to eliminate the stabilizing effect of polycrystalline domains and grain boundaries on dendrites will be the focus of the next research.

Two-stage morphological changes of lithium dendrites

Zhang Wentao, a doctoral student at the School of Advanced Materials of Peking University Shenzhen Graduate School, is the first author of the paper, and Professor Pan Feng and Professor Wang Linwang are co-corresponding authors. The research was supported by the Guangdong Soft Science Research Program, the Shenzhen Science and Technology Program, the Shenzhen Municipal Development and Reform Commission Scientific Research Project, and the Key Scientific Research Project of the Chinese Academy of Sciences.
 
Special statement: This article is reproduced only for the need to disseminate information, and does not mean to represent the views of this website or confirm the authenticity of its content; If other media, websites or individuals reprint and use from this website, they must retain the “source” indicated on this website and bear their own legal responsibilities such as copyright; If the author does not wish to be reprinted or contact the reprint fee, please contact us.



Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button