Question about modeling carbon fibers in a catalyst layer from CT images

[Shunsuke Takai]

Dear GeoDict Support Team,

Thank you very much for your continued support.

I am currently using GeoDict to investigate the structure of a fibrous catalyst layer for a fuel cell. My goal is to quantitatively characterize the carbon-fiber structure, including properties such as fiber length, fiber orientation, tortuosity, pore structure, and other structural parameters, and ultimately to reconstruct and model the catalyst layer in GeoDict.

However, I am currently having difficulty reconstructing only the carbon fibers from my CT images. The grayscale contrast in the CT images is not very clear, which makes it difficult to distinguish and segment the carbon fibers accurately.

The catalyst layer consists mainly of carbon fibers, with Nafion and palladium supported on the fibers. The approximate dimensions of the carbon fibers are:

* Fiber diameter: approximately 500 nm
* Fiber length: approximately 3–5 µm

Ideally, I would like to reconstruct the catalyst layer as a three-dimensional fibrous structure similar to the example image attached to this message, so that the individual carbon fibers and the pore space between them can be identified and analyzed.

I have also attached some of my current CT images of the actual catalyst layer for reference.

Could you please advise me on the following points?

1. Is it possible to extract and identify individual carbon fibers from CT images with this level of image contrast and fiber size using GeoDict?
2. Are there any recommended preprocessing or segmentation procedures in ImportGeo-Vol for improving the identification of the carbon fibers?
3. Would FiberFind, including its AI-based fiber identification functions, be applicable to a catalyst layer with fibers approximately 500 nm in diameter and 3–5 µm in length?
4. If the CT images themselves are not sufficient for identifying individual fibers, is there another recommended approach for reconstructing a statistically representative catalyst-layer model in FiberGeo using information obtained from CT and/or SEM images?
5. Are there any previous examples, tutorials, publications, or similar applications in which a catalyst layer or another structure consisting of very short and fine carbon fibers was reconstructed from CT images and modeled in GeoDict?

My main objective is to understand the actual structure of the carbon-fiber catalyst layer as quantitatively as possible and then use this information to construct a representative model in GeoDict.

I would greatly appreciate any advice regarding an appropriate workflow or relevant examples for this type of structure.

Thank you very much for your time and support.

Best regards,
Shunsuke Takai

[Jonas Schabernack]

Dear Shunsuke,

Thank you for your detailed message. Please find below our answers to your five questions.

1. Extracting individual fibers with the current image contrast
The key question here is the resolution (voxel length) of your CT images. To reliably resolve a feature, it should consist of at least 2–3 voxels, preferably more. For your ~500 nm fiber diameter, that means you would need a voxel size of at least 250 nm or less. In our experience this is typically not achievable with CT.
If the resolution were sufficient, GeoDict would be able to extract and identify individual fibers from such data, so this is primarily an imaging limitation rather than a GeoDict one. Could you let us know what voxel size your images have? That would allow us to give you a more definitive assessment.

2. Preprocessing / segmentation in ImportGeo-Vol
We don't have specific tips for carbon fibers, but for low-contrast data we would suggest trying the Enhance Contrast tool: https://geodict-userguide.math2market.de/2026/imageprocessing_enhancecontrast.html
That said, if the resolution is insufficient to resolve the fibers, no amount of image processing will make them visible. Contrast enhancement can only help with features that are actually present in the data.

3. Applicability of FiberFind / FiberFind-AI
FiberFind-AI requires a minimum of about 8 voxels per fiber diameter. For your ~500 nm fibers, that would correspond to a resolution of roughly 60 nm or better, again beyond what CT typically offers. If you do have (or can acquire) data at such a resolution, you may be limited to analyzing a smaller crop of the structure, but for these fiber dimensions that should still be perfectly acceptable.

4. Alternative approach: reconstructing a representative model in FiberGeo
If direct reconstruction from CT is not feasible, a practical route can be to derive the key structural parameters like fiber diameter, length distribution, and some indication of curvature from higher-resolution SEM images and use those as input for generating a statistically representative fibrous structure in FiberGeo. The main caveat is that SEM does not provide 3D information, so fiber orientation distributions and realistic curvature are harder to capture accurately.
For reference, we presented a very similar case at our last GeoDict Innovation Conference: fibers were generated based on measurements taken from SEM images. You can watch the relevant part of the video starting around minute 30: https://www.youtube.com/watch?v=d0pbNFC5bqY

5. Examples, tutorials and publications

• The Innovation Conference video mentioned above (from ~min 30): https://www.youtube.com/watch?v=d0pbNFC5bqY
• This tutorial on digital analysis of fibers and binder content in carbon paper GDLs: https://www.math2market.com/service-support/learning-center/digital-analysis-of-fibers-and-binder-content-of-four-carbon-paper-gdls.html

To summarize: with typical CT resolution, individual 500 nm fibers will likely remain below the detection limit. The most promising path forward is to combine what your CT data can still provide (e.g., overall porosity and larger-scale structure) with SEM-based measurements of fiber dimensions and then build a representative model in FiberGeo. Alternatively, you can try to obtain FIB-SEM measurements of your sample which allow you to get a high-resolution 3D image of your catalyst layer structure.

Best regards,
Jonas Schabernack