Thanks to a novel machine learning algorithm, scientists can now look for potential designs for the microstructure of lithium-ion batteries and fuel cells, prior to running three-dimensional (3D) simulations that help them make modifications to enhance performance.
Enhancements could involve boosting the power of hydrogen fuel cells that run data centers, increasing the time taken between charges for electric vehicles, and making cellular phones charge faster. The study has been recently published in the npj Computational Materials.
Powering the Future
Fuel cells make use of clean hydrogen fuel to generate electricity and heat; this hydrogen fuel can be produced by solar and wind energy. Likewise, lithium-ion batteries, like those found in electric cars, laptops, and smartphones, are a preferred type of energy storage.
Moreover, the performance of both fuel cells and lithium-ion batteries is closely associated with their microstructure: how the pores, or holes, within their electrodes are arranged and shaped can influence the amount of power generated by fuel cells, and the rapid charge-and-discharge cycle of batteries.
But since the micrometer-scale pores are very small, their particular sizes and shapes can be hard to examine at a sufficiently high resolution to associate them with the overall performance of the cell.
Scientists from Imperial College London have now used machine learning methods that not only allow them to examine these pores virtually but also help them to run 3D simulations to estimate the performance of the cell on the basis of their microstructure.
The scientists employed an innovative machine learning method, known as “deep convolutional generative adversarial networks” (DC-GANs). Algorithms like these can learn to create 3D image data of the microstructure on the basis of the training data acquired from nano-scale imaging performed by synchrotrons—a type of particle accelerator resembling the size of a football stadium.
Our technique is helping us zoom right in on batteries and cells to see which properties affect overall performance. Developing image-based machine learning techniques like this could unlock new ways of analysing images at this scale.
Andrea Gayon-Lombardo, Study Lead Author, Department of Earth Science and Engineering, Imperial College London
When 3D simulations are performed to estimate the performance of the cell, the team required data that is sufficiently large to be considered statistically representative of the entire cell. At present, it is hard to acquire large quantities of microstructural image data at the needed resolution.
But the study authors discovered that their code could be trained to produce either relatively larger datasets that contain all the same characteristics or intentionally produce structures that, according to models, would lead to better-performing batteries.
Our team’s findings will help researchers from the energy community to design and manufacture optimised electrodes for improved cell performance. It’s an exciting time for both the energy storage and machine learning communities, so we’re delighted to be exploring the interface of these two disciplines.
Dr Sam Cooper, Project supervisor, Dyson School of Design Engineering, Imperial College London
By limiting their machine learning algorithm to create only results that are presently viable to manufacture, the scientists are hoping to use their method from designing to manufacturing more improved electrodes for sophisticated cells.
The Engineering and Physical Sciences Research Council funded the study.
Gayon-Lombardo, A., et al. (2020) Pores for thought: generative adversarial networks for stochastic reconstruction of 3D multi-phase electrode microstructures with periodic boundaries. npj Computational Materials. doi.org/10.1038/s41524-020-0340-7.