Please enable JavaScript to view this site.

GeoDict User Guide 2025

Identify Fibers (AI)

The final goal of all FiberFind tools is to obtain accurate statistical parameters for advanced material design, that can be used e.g., as input data for FiberGeo.

There are two distinct approaches for the identification of individual fibers available in FiberFind. The first approach uses classical image processing methods, the second one is an Artificial Intelligence (AI) approach. Please refer to the AI approach to identify binder and fibers for more details on the theory.

  1. The identification of fibers with classical image processing methods is optimized for low density fibrous materials. FiberFind identifies fibers by creating a skeleton, that preserves the topology of the fibers. This skeleton is used to determine the center lines of the fibers.
  2. The AI approach uses a neural network that has been trained with great amounts of input data. The training data for ground truth is created with FiberGeo. Therefore, this approach is capable of identifying individual fibers in a structure, as long as the input data for the training was close enough to the fiber properties in the real 3D-scan.

The start of the computation is a three-dimensional model obtained from segmented computer tomography (or FIB/SEM) scans of the material. Fiber identification provides advanced structure statistics, such as fiber length, number of fibers and better local fiber orientation, as well as object-based image manipulation (for example, changing the diameter of fibers directly from the imported CT-scan images).

Note-Important

Important! Since FiberFind ships with a limited number of neural networks for fiber and binder identification, currently some constraints on the input data exist:

  • Fiber diameters should be 4 to 10 voxels for binder identification and 8 to 40 voxels for fiber identification.
    If fiber diameters differ from this value, ImportGeo-Vol can be used to resample the voxel length. During the import of an image, the 3D Image Processing functionality Image Size→Scale can be used for scaling the 3D image with a different voxel length. Details can be found in the ImportGeo-Vol user guide.
  • If fiber diameters in the structure are very different, and therefore, after resampling, not all diameters lie in the range listed above, we suggest separating the fibers and applying FiberFind - Identify Fibers (AI) for each kind of fiber.
    To separate the fibers, Estimate Fiber Diameters in FiberFind can be used. Run the fiber diameter estimation and choose the number of fiber types in the Result Plots to fit to your structure. On the tab Result Visualization, you can then load a .gdt file with the fibers separated according to different fiber types. See Estimate fiber diameters for detailed instructions and an example.
  • Some networks require fibers to be oriented mainly in the x-y plane. Please see the individual network description in the dialog. If necessary, the structure can be rotated to fulfill this condition.
    Changing the coordinate axis is done easily through ProcessGeo → Permute functionality. Details can be found in the ProcessGeo user guide.
    Changing the coordinate axis or applying complex rotations to the structure by defining Euler angles is also possible through ImportGeo-Vol. Details are explained in the ImportGeo-Vol user guide.
  • Fiber identification works for circular and elliptical fibers right now, not for other cross sections such as hollow or trilobal fibers.

Note-Info

Note! Besides this user guide, several tutorials showing the possibilities of the AI based FiberFind approach for different application cases are available on the GeoDict Learning Center.

One tutorial is dealing with the analysis of four samples of different carbon paper gas diffusion layers used for a proton-exchange membrane in a fuel cell (Digital analysis of fibers and binder content of a carbon paper GDL). Synchrotron scan data for the sample is provided with the tutorial. The tutorial contains a detailed description of the import and preprocessing of the data, the binder and fiber identification using the AI approach, and postprocessing options for the results.

A second tutorial deals with the FiberFind identification of fibers of a composite material, a glass-fiber reinforced polymer, as well as the mechanical analysis of the digital 3D model with ElastoDict (Digital analysis of a GFRP and generation of statistical Digital Twin). Input data and a detailed description about preprocessing of the data is provided with the tutorial.

©2025 created by Math2Market GmbH / Imprint / Privacy Policy