U.K. uses machine learning to study microstructures to design better-performing batteries

According to foreign media reports, researchers from Imperial College London have demonstrated how machine learning can help design better-performing lithium-ion batteries and fuel-powered lithium batteries. This new machine learning algorithm allows researchers to explore possible designs for the microstructure of fuel-powered lithium batteries and lithium-ion batteries, and then run 3D simulation models to help researchers make changes that improve battery performance. Performance improvements include faster charging of smartphones, extended range of electric vehicles, and the power of new hydrogen fuel-powered lithium batteries in the new data center. Fuel-powered lithium batteries can use clean hydrogen fuel from wind and solar energy to generate heat and electricity, and lithium-ion batteries in smart phones, laptops and electric cars are also a very popular way of energy storage. The performance of both is closely related to their microstructure: the shape and arrangement of the small holes inside the battery will affect the energy that the fuel-powered lithium battery appears and the charging and discharging speed of the battery. However, since the size of such small holes is on the order of micrometers, which is very small, it may be very difficult to study the specific shape and size of such small holes under the condition of sufficiently high resolution to correlate them with the overall performance of the battery. . Now, researchers at Imperial College have used machine learning technology to help them explore such small holes in a virtual way, and run a 3D simulation model to predict its performance based on the microstructure of the battery. The researchers used a new type of machine learning technology called 'Deep Convolution Generative Adversarial Networks' (DC-GANs), based on training data obtained from nano-scale imaging of synchrotrons (a particle accelerator the size of a football field). Learn to generate 3D image data of the battery's microstructure. The lead author of the study, Andrea Gayon-Lombardo of the Department of Earth Sciences and Engineering at Imperial College London, said: 'Our technology helps us zoom in on batteries and cells to see what characteristics affect overall performance. Development of this image-based machine learning technology New methods can be provided for analyzing images of this size.' When running 3D simulation models to predict battery performance, researchers need a large enough amount of data to statistically represent the entire battery. At present, it is difficult to obtain image data of a large number of microstructures with a resolution up to the standard. However, researchers have found that training codes can be used to generate larger data sets with the same attributes, or intentionally generated structures, which can build models of better-performing batteries. By limiting its algorithm to only the results that can be produced currently, the researchers hope to apply the technology to battery manufacturing to design optimized electrodes for next-generation batteries.