Identify Fibers by Single Fiber Comparison
Validation: FiberFind
To validate the fiber identification algorithm in FiberFind, three representative nonwoven models were generated using GeoDict’s fiber generator (FiberGeo). Since these synthetic structures are created parametrically, the ground truth is inherently known, including the exact fiber ID assigned to each voxel.
FiberFind was then applied to the voxelized geometries to identify and segment individual fibers based solely on the voxel data. The identified fibers were compared directly to the ground truth information from the original FiberGeo models.
The agreement between FiberFind and the FiberGeo ground truth was nearly perfect. Only a very small number of identification errors were observed, occurring predominantly at the domain boundaries where fibers are partially truncated. Overall, the results demonstrate the high accuracy and robustness of FiberFind for reliable single-fiber identification in complex nonwoven structures.
References
A. Grießer, R. Westerteiger, E. Glatt, H. Hagen, and A. Wiegmann, “Identification and analysis of fibers in ultra-large micro-CT scans of nonwoven textiles using deep learning,” Journal of the Textile Institute, vol. 114, no. 11, pp. 1647–1657, Nov. 2022, doi: doi.org/10.1080/00405000.2022.2145429
| Validation 1 | Validation 2 | Validation 3 | ||||
| Number of fibers | 222 | 408 | 340 | |||
| Absolute | Relative | Absolute | Relative | Absolute | Relative | |
| Boundary, contact & merge errors | 16 | 7.2% | 13 | 3.2% | 5 | 1.5% |
