AI Sets New Benchmark in Materials Analysis with Data From 10,000 Steel Samples

With around 5,000 types of steel, nuances in the production process are decisive. To create new properties or ensure consistent quality, steels are analyzed using a range of imaging methods. Professor Frank Mücklich and his research team have built extensive expertise in this field over many years. Using their microscopic analysis data, they trained an AI system capable of detecting even the smallest changes in steel. This AI can now serve as a standard in industrial laboratories for analyzing metallic and ceramic materials. To achieve this, the Saarbrücken researchers are working with Imagic, a Swiss company specializing in image databases.

Understanding Microstructure to Improve Material Quality

In the production of steel and other metals, every manufacturing step influences the internal structure, referred to by materials scientists as the “microstructure.” This structure changes depending on chemical composition, rolling processes or heat treatments. “The microstructure of steel is extremely complex and varies greatly depending on the desired properties. Under the microscope or in computed tomography, even the smallest differences must be detected and correctly classified. Our AI-based method now performs this automatically,” explains Frank Mücklich, Professor of Functional Materials at Saarland University.

Years of Interdisciplinary Research Behind the Breakthrough

Extensive research was required to train the artificial intelligence so that it not only recognizes different patterns in material microstructures but also analyzes them objectively. “Several doctoral theses at my chair have contributed to this work, all with an interdisciplinary approach. We brought in scientists from the Max Planck Institute for Informatics and the German Research Center for Artificial Intelligence, who transferred their machine learning and AI methods to materials science,” says Mücklich, who also heads the Steinbeis Research Center for Materials Engineering (MECS). Through long-term collaboration between this transfer institute and the Saarland steel manufacturer Dillinger, the researchers analyzed around 10,000 material samples from different steels on micro, nano and atomic scales and stored the results in a comprehensive database.

Database of 10,000 Steel Samples Enables Reliable AI Training

To enable industrial companies to perform their analyses independently on the basis of this database, the Steinbeis Transfer Institute MECS has now entered into a strategic partnership with the Swiss company Imagic Bildverarbeitung AG. The company develops software for microscopy, image analysis and image data management. “We provide what is known as the ground truth, meaning verified and reliable data suitable for training artificial intelligence and achieving correct results. So far, these material data relate to steel grades and various metals, but we also intend to extend this to all other metals and ceramics,” explains Frank Mücklich.

Strengthening Materials Expertise on Campus Saarbrücken

The materials researcher aims to keep the expertise surrounding imaging methods for materials on the Saarbrücken campus to provide highly qualified jobs for his graduates. “At the Steinbeis Research Center MECS, which we spun off from the university 15 years ago, several of my former doctoral students are already working and bringing their research expertise with them,” says Professor Mücklich.

From Research to Industry Practice Through Award-Winning Work

One of those graduates, Dominik Britz—now deputy director of the transfer institute—received several research awards for his doctoral thesis on AI-based quality assessment of steel, including the Georg Sachs Prize of the German Materials Society. He emphasizes the importance of transferring research results quickly into industrial practice. “With our AI-supported method, we want to make analyses safer and faster, similar to medical imaging procedures. Our data foundation can serve as a standard for how material samples should be evaluated in the future. This will not only support the development of new steel grades and metals but also help detect material defects at an early stage,” explains Dominik Britz.