News

Improving Fuel Cell Efficiency Through Machine Learning Techniques

June 27, 2020 by Gary Elinoff

Better fuel cells and lithium-ion batteries will be made possible by greater knowledge of electrode microstructure.

The performance of both fuel cells and lithium-ion batteries (LiBs) is closely related to how the pores or holes inside their electrodes are arranged and shaped.  This microstructure will affect how quickly the LiBs charge and discharge, and how much electricity a fuel cell can produce.

The pores are quite small, in the range of micrometers. Because of their tiny scale, studying them to the resolution that would be necessary to relate them to device performance has proven difficult.

Investigators from London’s Imperial College have devised a way to study the pores virtually. Scientists have been able to run three-dimensional simulations through the application of machine learning and, in this manner, gain knowledge of microstructure, and through this knowledge, predict performance. 

 

Deep Learning and Particle Accelerators

The researchers employed a machine learning technique called deep convolutional generative adversarial networks (DC-GANs) to generate three-dimensional image data of the microstructure. This method involved applying training data garnered from the use of a synchrotron

A synchrotron is a circular particle accelerator that accelerates charged particles until they approach the speed of light, producing very bright light, called synchrotron light. Investigators can use this light to study matter as small as atoms and molecules. 

According to lead author Andrea Gayon-Lombardo, of Imperial’s Department of Earth Science and Engineering, “Our technique helps 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.” 

 

The structure of the machine learning algorithm and the approach it uses to learn the essence of microstructural data.

A diagram detailing the structure of the machine learning algorithm and the approach used to learn the "essence" of the microstructural data. Image credited to Imperial College London
 

Difficulties Involved in Obtaining Sufficient Amounts of Data

The process of running 3D simulations to predict cell performance requires a large body of data. This is necessary for the data set to be statistically representative of the whole cell. 

At present, obtaining enough microstructural image data is difficult. However, the authors of the study were able to train their code to generate much larger datasets with all the same properties. Alternatively, they could purposefully generate structures that would result in better-performing batteries, as per the model’s suggestion.

As per project supervisor, Dr. Sam Cooper, of Imperial’s Dyson School of Design Engineering, said: “Our team’s findings will help researchers from the energy community to design and manufacture optimized 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.”

 

Research Goals

The researchers intend to apply their methods to the manufacture of optimized electrodes for improved cells. In pursuit of this goal, they limited the scope of their algorithm only to produce results that are buildable using current manufacturing.
 

Valuable Insights Into Microstructures 

Studying the microstructure of LiB elements is an ongoing area of research. We recently reported on the work of scientists at the U.S. Department of Energy’s SLAC National Accelerator Laboratory at Stanford University. These researchers employed X-ray tomography data, as well as machine learning. The Stanford work provided valuable insight into the breakdown of LiB cathodes.

Progress in understanding how energy storage and generation devices function on an atomic structural level is necessary for the quest for a LiB battery that will enable the widespread adaptation of electric vehicles. It will also make possible the smaller, more energy-dense batteries required for the further advancement of remote IoT edge devices, where frequent battery replacement or recharging would impede progress.