Unpaired Fast- and Slow-Acquisition MicroCT Scans of Carbon-Fibre Composites

October 2021 tomographic images of carbon-fibre-reinforced composites with Toray T700S fibres in a Sicomin SR8500 KTA313 epoxy matrix, acquired at the TOMCAT Beamline at the Swiss Light Source. The datasets accompany a paper on deep-learning image enhancement and fibre segmentation and are used to train 3DCycleGAN and UNetID algorithms. The collection includes multiple scans with varying voxel sizes (400-1600 nm³), exposure times (0.5-250 ms), and total acquisition times (0.25-500 seconds).

Use Cases

• Train a 3DCycleGAN enhancement algorithm based on unpaired fast- and slow-acquisition microCT scans.
• Train a UNetID segmentation algorithm for fibre identification based on tomographic images of composites.
• Benchmark deep learning models for 3D image processing on real-world materials science data.
• Study the effect of acquisition parameters (voxel size, exposure time) on image quality and segmentation accuracy.

Strengths

• Includes multiple scans of the same sample with different acquisition parameters (e.g., 800 nm³ and 1600 nm³ voxel sizes).
• Accompanies specific, open-source deep learning algorithms (3DCycleGAN, UNetID) with provided pre-trained models.
• Data provenance is clearly documented, specifying the beamline (TOMCAT), date (October 2021), and material composition.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

TOMCAT Beamline at the Swiss Light Source

Collection Method

Acquired via computed tomography (microCT) scanning.

Time Range

October 2021