KI4MaterialModeling

Simulations have become an indispensable part of modern material development, predicting properties by mapping material structures. Digital models are essential in simulation chains, but precise separation of material phases on different length scales is crucial for accurate property simulations. Current methods involve computed tomography CT scans and computer-generated models, both with limitations leading to deviations in simulations. In CT scans there is always a trade-off between resolution and the field of view (size of the scanned volume) and unavoidable image artefacts lead to the fact that the segmentation of the material phases is subject to uncertainty. The computer-generated models require a high number of parameters for precise modeling, whereby the aforementioned challenges (especially uncertainty in the extraction of geometric dimensions) remain. In KI4MaterialModeling, together with the Math2Market, the gap between the two mentioned methods shall be closed by using artificial intelligence (AI) methods. The AI-based workflow comprises: (1) generation of synthetic CT data to train AI, (2) AI for enhancing the trade-off between sample size and resolution in CT scans, (3) AI for object recognition in CT data. This workflow improves materials analysis and development efficiency.

Applications include:

• • Supporting segmentation and object recognition in real CT data to arrive at high-quality geometry model
• • Automated extraction of structural informationfrom CT data for synthetic model generation
• • Combining CT data with synthetic data to create a hybrid model with large volume at high resolution

The aim of the project is to close the gap between synthetically generated geometry models and geometry
models derived from CT data the systematic use of AI methods.