Machine learning for microstructural optimisation - S. Kench

In the field of cathode optimisation, 3D microstructural datasets are important in order to understand structure-performance relationships through physical modelling. However, 3D imaging techniques can be slow and often have limited resolution compared to their 2D counterparts. We propose a novel machine learning method, SliceGAN, which can use a single representative cross-sectional image to synthesise realistic 3D volumes. Furthermore, through the use of conditional GANs we are able to extend this approach and synthesise microstructures associated with different manufacturing parameters. This tool, alongside a PyBamm DFN model, can be implemented within a bayesian optimisation framework to enable high speed optimisation of accessible capacity. To submit your questions for our Q&A please tweet us at @batterymodel, email us at battery@, contact us on Slack (link provided in presentation e-mail) or simply leave a comment in the field below.