Partner: Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V.
Field: materials development
How can steel be prevented from crumbling under the stress of high temperatures and corrosion? A team of scientists from Germany’s leading applied research organisation, Fraunhofer-Gesellschaft, is seeking an answer to this question. They need to use advanced technology to analyse the microscopic layers of protective coatings on steel.
They were looking for a strong partner with the expertise and experience to apply it in the industry. Therefore, they joined forces with the Czech National Competence Centre in High Performance Computing (HPC).
“We are researching new protective coatings that can be applied to steel to enhance its performance. Such research includes identifying and subsequently optimising coating parameters by modifying the production and technological process. The coating properties can be detected by scanning electron microscope images of the samples. This is an image analysis problem, and for this purpose, we have teamed up with experts from NCC (IT4Innovations),” says Fraunhofer research team leader Vladislav Kolarik.
Challenge: How to understand the “invisible” parts of the steel surface?
Imagine the steel surface under a microscope, composed of layers as thin as a hair, full of tiny pores and cracks. These details determine how well a material can withstand high temperatures and corrosion. Manual analysis of these coatings is enormously time-consuming and often fails to identify fine details. How can such a challenge be solved?
The solution offers a combination of AI, supercomputers, and a digital twin
Scientists at IT4Innovations National Supercomputing Center have found the answer by combining three modern technologies – synthetic data, artificial intelligence, and the computing power of supercomputers.
“We needed training data to help the models identify all the details. But to label hundreds of images manually would take months,” explains Petr Strakoš, one of the IT4I team scientists. “Therefore, we created synthetic images that faithfully simulate real structures and can be generated in huge volumes.”
The IT4I team used the Blender software to create a virtual 3D copy of the steel surface images. These included all the key elements of the coatings – alloy layers, pores, cracks, and oxides. The advantage was that they could quickly change parameters – such as layer thickness and pore concentration – and generate various images for training.
“It was not just about creating images,” Strakoš adds. “We also had to ensure the model understood what to look for. Linking synthetic data with artificial intelligence was key to success.”
Supercomputers accelerated the whole process
A conventional computer would take weeks to train a model, but with a supercomputer, thousands of images can be processed in a few days.
“Computing power is essential when you want to achieve both speed and accuracy,” explained Tomas Karasek, Head of the National Competence Centre in HPC.
Read about more successful collaborations.
Don’t miss any news. Follow NCC Czechia on social media.
