A technology has emerged that enables artificial intelligence (AI) to identify the causes of fuel cell failures instead of humans.
A research team led by Jeong Chi-young, a senior researcher at the Korea Energy Technology Institute’s Hydrogen Demonstration Research Center, announced on the 19th that they developed a technology to analyze the microstructure of carbon paper, a key material in hydrogen fuel cells, at a speed 100 times faster than before.
Carbon paper is a key material that aids in water discharge and fuel supply in the stacks of hydrogen fuel cells. It is composed of materials such as carbon fiber, binder, and coating agents, and as it is used, the arrangement, structure, and coating state of the materials change, reducing the performance of the fuel cell. Analyzing the microstructure of carbon paper is an essential element for diagnosing the state of the fuel cell.
Until now, there has been no technology to analyze the microstructure of carbon paper in real time. This is because precise analysis requires the procedure of breaking the carbon paper sample and then conducting detailed analysis using an electron microscope.
The research team successfully analyzed the microstructure of carbon paper using X-ray diagnosis and an AI-based image learning model. It became possible to conduct precise analysis merely through X-ray tomography without the need for an electron microscope, allowing for near real-time diagnostics.
The research team extracted 5,000 images from over 200 carbon paper samples and trained a machine learning algorithm on them. The trained model predicted the three-dimensional distribution and arrangement of the components of carbon paper, such as carbon fiber, binder, and coating agents, with an accuracy of over 98%.
Because the model developed by the research team uses X-ray tomography equipment instead of an electron microscope, the time taken for analyzing carbon paper has been shortened from a minimum of two hours to several dozen seconds. Jeong Chi-young noted, “This research has great significance in that it enhances analysis technology using virtual space combined with AI technology and clarifies the relationship between the structure and properties of energy materials, proving its practical applicability. We anticipate it will play a significant role in related fields such as secondary batteries and electrolysis in the future.”
Reference material
Applied Energy (2024), DOI: https://doi.org/10.1016/j.apenergy.2024.124689