Creating 3D Digital Twins from 2D Images with optimization and AI approaches

In this seminar, we will show how statistical digital twins of granular materials can be created with GeoDict. When 3D µCT images are available, the material geometry can be analyzed directly, and the resulting structural information can be used to generate a statistical 3D digital twin. 

When only 2D images are available, a different method is needed. In this case, 3D microstructure models are generated iteratively until their 2D cross sections reproduce the material characteristics observed in the original images. 

In GeoDict 2027, this process is automated in two different GeoApps. The first applies a classical optimization approach to determine suitable input parameters for a microstructure generator. The second uses generative adversarial networks (GANs), in which a generator AI is trained to produce 3D microstructures whose 2D cross sections cannot be distinguished from the original images by a discriminator AI.

We will present both approaches and compare their advantages and disadvantages, as well as their potential and limitations for creating statistical 3D digital twins from 2D image data.
 

Register now