Machine learning is being rapidly implemented on a wide variety of problems in the energy sciences due to the ease with which it is able to represent the kind of highly complex, multidimensional, non-linear problems characteristic of electrochemistry. In the ESE group, machine learning methods are being applied to image analysis (such as classification of phases), microstructural generation, cell lifetime prediction, and grid demand forecasting. High quality and abundant data is typically necessary for these problems, which is well catered for by the ESE group’s depth experimental research.
As part of the ESE group, Sam’s group at Dyson School of Design Engineering is leading this topic of research. Sam’s group principally focuses on the use of simulations and machine learning methods to characterise and design materials for energy storage and conversion applications. This includes the collection and analysis of 3D images, isotopic labelling data and impedance spectra, as well as a variety of other experimental techniques. These methods have been applied to a wide range of technologies, including lithium-ion batteries, fuel cells and supercapacitors.
Machine learning for electrode characterisation and optimisation
Lead researchers: Miss Andrea Gayon-Lombardo and Mr Steven Kench
PI’s: Dr Samuel Copper and Prof Nigel Brandon
- EPSRC Faraday Institution Multi-Scale Modeling project (EP/S003053/1, grant number FIRG003)
- CONACYT-SENER fund
Li-ion electrode performance is strongly influenced by microstructural properties such as particle size distribution and surface area, which in turn determine electrochemical behaviour. Understanding these structure/property relationships is thus a promising path for the development of next-generation batteries. Our project aims to address this challenge through the use of machine learning methods, with the goal of developing a toolset for electrode characterisation and optimisation.
Recent work has focussed on generative adversarial nets (GANs), which are used to synthesise visually indistinguishable images compared to a training dataset. Given an electrode micrograph, this allows the generation of large volumes of realistic electrode microstructure. An extension of this concept allows the synthesis of periodic microstructures, which is useful in the context of electrochemical simulation where periodic boundary conditions reduce the required sample volume.
Current studies also include 3D microstructural reconstruction using 2D micrographs, such that the advantages of high resolution, large field of view 2D imaging techniques can be translated into 3D datasets. Future work will focus on how a range of established machine learning methods for image processing can be applied in the field of material science and electrochemistry.
- Earth Science Analytics AS: Dr Lukas Mosser
Gayon-Lombardo A, Mosser L, Brandon NP, Cooper S, 2020, Pores for thought: Generative adversarial networks for stochastic reconstruction of 3D multi-phase electrode microstructures with periodic boundaries, npj Computational Materials, Vol: 6, Pages: 1-11. (GitHub repository: https://github.com/agayonlombardo/pores4thought)
Recent publications, 2018 - to date
Mistry A, Franco AA, Cooper SJ, Roberts SA, and Viswanathan V, 2021, How Machine Learning Will Revolutionize Electrochemical Sciences, ACS Energy Lett., Vol: 16,, Pages: 1422–1431
Gayon-Lombardo A, Mosser L, Brandon NP, Cooper S, 2020, Pores for thought: Generative adversarial networks for stochastic reconstruction of 3D multi-phase electrode microstructures with periodic boundaries, npj Computational Materials, Vol: 6, Pages: 1-11
Characterisation and design of battery electrode microstructures using simulation and machine learning
Invited talk ARTISTIC conference, July 1, 2020
Check out the new publication: Pores for thought