From Deep Learning to Standardization: The Field of Tension of Modern Microstructure Analysis
This year's meeting of the DGM working group "Quantitative Microstructural Analysis" offered a multifaceted insight into current projects and developments from science and industry. The opening session already showed how broad the field of quantitative microstructure analysis is today: from data-based AI methods and software-supported image analysis to standardized measurement methods, a wide range was covered.

AI methods: The Challenge of Ground Truth and Scalable Applications
In his presentation, Mr. Bachmann (MECS, Saarbrücken) highlighted the challenges of developing scalable AI systems for microstructure analysis on an industrial scale. A central topic was the definition of the so-called "ground truth", which is essential for the training of AI models - but is associated with considerable effort.

The MECS takes a multimodal approach: Light optical microscopy (LOM), scanning electron microscopy (SEM), and electron backscatter diffraction (EBSD) images of the same sample locations are combined to create a reliable reference system.

Based on this data, a web application was developed that automatically recognizes grain sizes and structural components in steel samples from LOM images. The results have already been validated in several round robin tests. An online round robin test for the classification of highly complex phases in bainitic steels is currently underway - an invitation to actively participate in the network.

With a different focus, Mr. Kaip (GFaI) presented the iFrakto project, which is used for automated fracture surface analysis using AI. The system classifies microscopic fracture types using combined electron microscopic image data (SE, BSE, topography) and will in the future also record macroscopic features. An intelligent user interface is also planned - a contribution to the practical support of fractographic investigations.

Mr. Müller (MECS) presented a technically innovative approach: Using generative AI, high-resolution SEM images are generated from LOM images. First comparisons with real SEM data show a high degree of agreement. In the future, the model will be extended to other microstructures and EBSD data, with the long-term goal of replacing SEM with LOM-based modality transfer in certain applications.

Decarburization and Dendrite Analysis: Round Robin Tests Planned
Dr. Korpala (MiViA) presented an AI-based method for determining the depth of decarburization. By detecting and classifying the structural components in decarburization layers, an imaginary carbon profile is reconstructed from which the decarburization depth can be derived. Several specialized AI models are used for this purpose. The analysis of carburization processes is also possible with this approach.

Another topic was the practical implementation of the standardized decarburization depth determination according to DIN EN ISO 3887. Mr. Nützel (Schaeffler Technologies) pointed out that there are often uncertainties in the application - for example in determining the surface position or the carbide content. A planned interlaboratory test within the working group will systematically investigate these aspects.

Mr. Lison (BMW Group) presented a GFaI software solution for the analysis of dendrite arm spacing in aluminum castings. Automated image analysis is still in its infancy - there are challenges in particular in the case of weakly contrasting dendrite boundaries and structural changes due to heat treatment. The goal is to capture all dendrites in a single image, not just along individual lines. A round-robin test is also planned, the planning of which will be coordinated in a joint online meeting with the decarburization depth test.

Working Method and Structure: Hybrid and Thematically Open
In the concluding discussion, the participants spoke in favor of holding hybrid meetings – especially since the current meeting did not receive enough submissions for a face-to-face event. Therefore, a hybrid meeting is planned for spring 2026 at MiViA or at the Institut für Umformtechnik in Freiberg.

A proposal to reflect the increasing focus on AI methods in the name of the working group was discussed. However, the participants decided against this: AI is understood as a methodological tool within microstructure analysis – not as an exclusive feature. AI also plays an increasing role in other DGM expert committees, without influencing the name.

Information on all expert committees can be found on the DGM website at https://dgm.de/en/network.
If you are interested in participating in one of the committees, please send a short email to fachgremiendgm.de.