CoolFace
Datasetpublic

lambolimm/carbon-fiber-defects

🧡 Carbon Fiber Defect Sample Dataset (CFYD + WPSD) This repository presents sample images from two industrial defect datasets β€” Carbon Fiber Yarn Defects Dataset (CFYD) and Woven Preform Surface Defects Dataset (WPSD). These datasets were constructed to support defect detection in carbon fiber manufacturing, covering both yarn-level and preform-level production stages. πŸ“˜ Dataset Overview This dataset project contains two sub-datasets, each corresponding to… See the full description on the dataset page: https://huggingface.co/datasets/lambolimm/carbon-fiber-defects.

sourceHugging Faceupdated 1y agoView on Hugging Face
0likes4downloads
Dataset Card

🧡 Carbon Fiber Defect Sample Dataset (CFYD + WPSD)

This repository presents sample images from two industrial defect datasets β€” Carbon Fiber Yarn Defects Dataset (CFYD) and Woven Preform Surface Defects Dataset (WPSD).

These datasets were constructed to support defect detection in carbon fiber manufacturing, covering both yarn-level and preform-level production stages.


πŸ“˜ Dataset Overview

This dataset project contains two sub-datasets, each corresponding to a distinct stage in the composite preform manufacturing process.


1️⃣ Carbon Fiber Yarn Defects Dataset (CFYD)

The CFYD dataset focuses on the yarn delivery stage of carbon fiber production.

  • β€”Raw Data: 2,108 high-resolution images (4096Γ—3000)
  • β€”Processed Data: Each raw image was divided into 1024Γ—1024 patches
  • β€”Cleaned Data: After discarding blurred or noisy samples, approximately 10,000 valid patches were retained
  • β€”Defect Categories:
  • β€”Fuzz
  • β€”Broken Yarns
  • β€”Fiber Clumps
  • β€”Loose Filaments
  • β€”Normal Category: Images without visible defects
  • β€”Purpose: Enables fine-grained defect localization and recognition in slender fiber structures.

2️⃣ Woven Preform Surface Defects Dataset (WPSD)

The WPSD dataset focuses on the weaving stage of carbon fiber preform manufacturing.

  • β€”Raw Data: 1,574 preform images (4096Γ—3000)
  • β€”Processed Data: Each image was partitioned into 1024Γ—1024 patches
  • β€”Cleaned Data: After manual screening, approximately 5,000 valid patches were retained
  • β€”Defect Categories:
  • β€”Knots
  • β€”Yarn Breakage
  • β€”Floating Yarns
  • β€”Normal Category: Defect-free woven surfaces
  • β€”Note: Defect instances in this dataset are comparatively rarer than in CFYD, reflecting real-world manufacturing distributions.

πŸ—‚οΈ File Organization

The repository is structured as follows:

carbon-fiber-defects/
β”‚
β”œβ”€β”€ CFYD/
β”‚   β”œβ”€β”€ Fuzz/
β”‚   β”‚   └── Fuzz.jpg
β”‚   β”œβ”€β”€ Broken Yarns/
β”‚   β”‚   └── Broken Yarns.jpg
β”‚   β”œβ”€β”€ Fiber Clumps/
β”‚   β”‚   └── Fiber Clumps.jpg
β”‚   └── Loose Filaments/
β”‚       └── Loose Filaments.jpg
β”‚
└── WPSD/
    β”œβ”€β”€ Knots/
    β”‚   └── Knots.jpg
    β”œβ”€β”€ Yarn Breakage/
    β”‚   └── Yarn Breakage.jpg
    └── Floating Yarns/
        └── Floating Yarns.jpg

Each subfolder contains one representative image that illustrates a specific type of surface or yarn defect. All images are resized and standardized for visualization purposes.


⚠️ Important Notes

  • β€”The current dataset contains sample images only and does not represent the full datasets used in the publication.
  • β€”The complete datasets (CFYD & WPSD) used for model training, testing, and evaluation in the paper will be fully released upon official publication.
  • β€”These samples are shared solely for academic reference and visualization.