EGI Yeon "Hydrogen Fuel Cell 'Malfunction' Rapid Diagnosis with Artificial Intelligence"

The research team led by Dr. Jeong Chi-young at the Hydrogen Demonstration Research Center of the Korea Institute of Energy Research announced on the 19th that they have succeeded in analyzing the microstructure of carbon fiber paper, a core material of hydrogen fuel cells, using virtual space utilization technology and artificial intelligence learning.

Through this, the diagnosis of failure causes in hydrogen fuel cells has become 100 times faster than before.

(From left) Myeong Gwang-sik, Senior Research Fellow; Jeong Chi-young, Center Director; Park Young-je, Student Researcher; Lee Jong-min, Engineer. Provided by Korea Institute of Energy Research

Carbon fiber paper is a key material in the stack of hydrogen fuel cells that helps with water discharge and fuel supply. Composed of materials such as carbon fibers, binders (adhesives), and coatings, the arrangement, structure, and coating condition of carbon fiber paper change during use, which degrades the performance of the fuel cell. For this reason, analyzing the microstructure of carbon fiber paper is an essential element for diagnosing the condition of hydrogen fuel cells.

However, it was previously impossible to analyze the microstructure of carbon fiber paper in real time. To obtain accurate analysis results, it was necessary to break the carbon fiber paper sample and then perform detailed analysis using an electron microscope.

To overcome this limitation, the research team developed a technology that analyzes the microstructure of carbon fiber paper using X-ray diagnostics and an artificial intelligence-based image learning model. By utilizing the developed technology, precise analysis is possible with only X-ray computed tomography without an electron microscope, enabling near real-time condition diagnosis.

During the technology development process, the research team extracted 5,000 images from about 200 carbon fiber paper samples and trained a machine learning algorithm. The trained model predicted the three-dimensional distribution and arrangement of the components of carbon fiber paper, such as carbon fibers, binders, and coatings, with an accuracy of over 98%. According to the research team, this allows for confirming changes in the composition between the initial and current states of the carbon fiber paper, enabling immediate identification of the causes of performance degradation.

While the conventional analysis method, which involves crushing carbon fiber paper samples and using an electron microscope, takes at least two hours, the analysis model developed by the research team can identify the degradation and damage areas and their extent within tens of seconds using only X-ray computed tomography equipment.

Using the data from the developed model, the research team clarified how design factors such as the thickness of carbon fiber paper and binder content affect fuel cell performance, and proposed an optimal design plan that can improve fuel cell efficiency by extracting the ideal design parameters.

Dr. Jeong Chi-young said, “This research is meaningful in that it demonstrated practical applicability by combining analysis technology using virtual space and artificial intelligence technology to elucidate the interrelationship between the structure and properties of energy materials.”

Meanwhile, this research was conducted with the support of the basic project of the Korea Institute of Energy Research. The research results (paper) were also published online last October in ‘Applied Energy,’ a world-renowned journal in the energy field.