52
IRUS Total
Downloads
  Altmetric

Machine learning and simulation for the optimisation and characterisation of electrodes in batterie

File Description SizeFormat 
Gayon-Lombardo-A-2021-PhD-Thesis.pdfThesis14.44 MBAdobe PDFView/Open
Title: Machine learning and simulation for the optimisation and characterisation of electrodes in batterie
Authors: Gayon Lombardo, Andrea
Item Type: Thesis or dissertation
Abstract: The performance of electrochemical energy storage (EES) and energy conversion (EC) technologies is closely related to their electrode microstrcuture. Thus, this work focuses on the development of two novel computational models for the characterisation and optimisation of electrodes for three devices: Redox Flow batteries (RFBs), Solid Oxide Fuel Cells (SOFCs), and Lithium-ion batteries (LIBs). The first method introduces a Pore Network Model (PNM) for simulating the coupled charge and mass transport processes within electrodes. This approach is implemented for a vanadium RFB using different commercially available carbon-based electrodes. The results from the PNM show non-uniformity in the concentration and current density distributions within the electrode, which leads to a fast discharge due to regions where mass-transport limitations are predominant. The second approach is based on the stochastic reconstruction of synthetic electrode microstructures. For this purpose, a deep convolutional generative adversarial network (DC-GAN) is implemented for generating three-dimensional n-phase microstructures of a LIB cathode and a SOFC anode. The results show that the generated data is able to represent the morphological properties and two-point correlation function of the real dataset. As a subsequent process, a generation-optimisation closed-loop algorithm is developed using Gaussian Process Regression and Bayesian optimisation for the design of microstructures with customised properties. The results show the ability to perform simultaneous maximisation of correlated properties (specific surface area and relative diffusivity), as well as an optimisation of these properties constrained by constant values of volume fraction. Overall, this work presents a comprehensive analysis of the effect of the electrode microstructure in the performance of different energy storage devices. The introduction of a PNM bridges the gap between volume-averaged continuum models and detailed the pore-scale models. The main advantage of this model is the ability to visually show the concentration and current distributions inside the electrode within a reasonably low computational time. Based on this, this work represents the first visual showcase of how regions limited by low convective flow affect the rate of discharge in an electrode, which is essential for the design of optimum electrode microstructures. The implementation of DC-GANs allows for the first time the fast generation of arbitrarily large synthetic microstructural volumes of n-phases with realistic properties and with periodic boundaries. The fact that the generator constitutes a virtual representation of the real microstructure allows the inclusion of the generator as a function of the input latent space in a closed-loop optimisation process. For the first time, a set of visually realistic microstructures of a LIB cathode with user-specified morphological properties were designed based on the optimisation of the generator’s latent space. The introduction of a closed-loop generation-optimisation approach represents a breakthrough in the design of optimised electrodes since it constitutes a first approach for evaluating the microstructure-performance correlation in a continuous forward and backward process.
Content Version: Open Access
Issue Date: Mar-2021
Date Awarded: Jul-2021
URI: http://hdl.handle.net/10044/1/91548
DOI: https://doi.org/10.25560/91548
Copyright Statement: Creative Commons Attribution NonCommercial Licence
Supervisor: Brandon, Nigel
Neethling, Stephen
Cooper, Samuel
Sponsor/Funder: Consejo Nacional de Ciencia y Tecnología (Mexico)
Secretaría de Energía
Funder's Grant Number: 580490
Department: Earth Science & Engineering
Publisher: Imperial College London
Qualification Level: Doctoral
Qualification Name: Doctor of Philosophy (PhD)
Appears in Collections:Earth Science and Engineering PhD theses



This item is licensed under a Creative Commons License Creative Commons