S3E12
Episode 12 (April 4, 2021)
Wei Li
MIT
Hongbo Zhao
MIT
Steven Kench
Imperial College London
Physics-guided neural network framework for multiphysics modeling of battery
Abstract of Talk 1
Physic-guided data-driven methods can bridge the gap between the data-driven approaches and the physics-based approaches, which aim to implement the already-known physics into the data-driven approaches. In this presentation, we will focus on applications in multiphysics modeling of battery. We start with establishing a neural network-based computational framework to characterize the finite deformation of elastic plates, which in classic theories is described by the Föppl--von Kármán (FvK) equations with a set of boundary conditions (BCs). The accuracy of the modeling framework is carefully examined by applying it to four different loading cases. We then extended this framework to incorporate the fluid dynamics. This brought new challenges in dealing with complex systems with more PDEs and BCs and initial conditions (ICs). The advantages and limitations of the framework will be discussed in detail based on these applications.
Biosketch of Speaker 1
Dr. Wei Li is currently a postdoctoral associate in the Department of Mechanical Engineering at Massachusetts Institute of Technology. He received his Ph. D. in Engineering from School of Vehicle and Mobility, Tsinghua University, in 2019. His research interests include mechanical testing and modeling of materials and structures, multiphysics modeling of lithium-ion batteries, and physics-informed data-driven applications.
Data-driven learning of battery physics
Abstract of Talk 2
A number of tools including microscopy, diffraction, spectroscopy, and impedance exist to probe and measure from multi-scale electrochemical systems such as batteries. With these datasets, there is an opportunity to extract more information such as the physics governing the battery behavior via data-driven methods. In this talk, I will present my work on using in-situ X-ray diffraction (XRD) and ex-situ microscopy data to determine that the mechanism for the observed fictitious phase separation is an autocatalytic reaction (previously attributed to diffusion-induced heterogeneity). The reaction kinetics inferred from XRD data agrees well with electrochemical measurements. I will also introduce the potential of learning quantitative models from microscopic images.
Biosketch of Speaker 2
Hongbo Zhao is a PhD candidate from the department of chemical engineering at MIT, working under Professor Martin Bazant. He received his bachelor degree from Tsinghua University in 2015. The topic of his PhD research is data-driven modeling of lithium intercalation materials. His research combines theory, numerical simulations, and data-driven methods to understand a variety of experimental characterizations and elucidate the fundamentals of electrochemical systems. He is also broadly interested in applied mathematics, modeling and computation of complex systems and their engineering applications.
Dimensionality expansion: Generating 3D electrode microstructures from a 2D slice
Abstract of Talk 3
In the field of computational materials design, 3D microstructural datasets are crucial for understanding structure-performance relationships through physical modelling. However, 3D imaging can be slow and often has limited resolution compared to its 2D counterparts. In this talk, I will present a novel machine learning network architecture, SliceGAN, which can use a single representative cross-sectional image to synthesise realistic 3D volumes. Furthermore, I will briefly discuss the possibility of extending this tool to enable microstructural optimisation using conditional generative adversarial.
Biosketch of Speaker 3
Steve Kench is a PhD student in the the Dyson School of Engineering at Imperial College London. He graduated from the University of Oxford in 2019 with First Class Honors in Material Science and a Masters degree in chemo-mechanical modelling of superalloy deformation. Steve then started a PhD fellowship with the Faraday Institution, focussing on the application of machine learning methods for battery electrode design. He recently published an article in Nature Machine Intelligence describing a method for synthesising 3D microstructures from 2D slices with generative adversarial networks. He is interested in image processing, tool development and materials optimisation.
Guest Host: Juner Zhu
Dr. Juner Zhu holds the position of Postdoctoral Associate at MIT and Executive Director of the MIT Industrial Battery Consortium. He earned his Ph.D. degree from the Department of Mechanical Engineering at MIT. His Ph.D. thesis entitled “Mechanical Failure of Lithium-ion Batteries” provides a comprehensive study on the mechanical modeling of battery component materials, cells, and modules. He is working jointly in the Departments of Mechanical Engineering and Chemical Engineering at MIT. This allowed him to extend his research interest in multiphysics modeling with data-driven methods such as inverse methods, PDE-constrained optimization, and physics-informed neural networks. Dr. Zhu has considerable industrial experience from his work as a material engineer at Ford Motor Company in 2016 and as a battery analyst at Apple in 2018. He got his B.S. and M.S. degrees from Tsinghua University, where he graduated with the top honor for Tsinghua graduate students in 2012.