Thermal Inspection Dataset for Defect Segmentation in CFRP Laminates

This dataset contains thermal image data and corresponding annotations from a pulsed thermography (PT) inspection conducted on a carbon fiber-reinforced polymer (CFRP) laminate. The dataset is designed to support research in defect segmentation, early defect detection, and related applications in composite material inspections.

Dataset Composition:

Images:
A total of 1,034 thermal images captured using a midwave infrared (MWIR) camera at a resolution of 640 × 512 pixels and a frame rate of 55 Hz. These images were recorded during PT inspections following a short, intense heat pulse.

Annotations:
Each of the thermal images has a corresponding expert-annotated image, resulting in 1,034 segmentation masks. The annotations were created using the VGG Image Annotator, a tool developed by the Visual Geometry Group at the University of Oxford, and represent the ground truth data for model training and validation.

Sample Details:

Material:
The sample is a unidirectional carbon/PEEK laminate with a fiber volume fraction of 61%. The stacking lay-up of the laminate is [02/902]6, and its dimensions are 100 × 100 mm.

Defect Characteristics:
Artificial defects in the form of Kapton tape inserts were introduced into the laminate during the moulding process. These defects vary in size and depth:

Nominal Depths:
D1: 0.13 mm
D2: 0.26 mm
D3: 0.39 mm
Nominal Sizes:
2 × 2 mm
3 × 3 mm
4 × 4 mm

Inspection Methodology:
A MWIR camera captured thermal profiles as the surface of the laminate cooled after heat application. Images display the temporal evolution of surface defects, highlighting changes over time. This temporal variation is critical for identifying and segmenting defects accurately.

Applications:
Defect segmentation in composite materials.
Development and validation of machine learning models for thermal image analysis.
Early detection and tracking of defect progression.

Ground Truth and Segmentation Masks:
The annotated dataset includes:
Segmentation masks created from the annotations, used for model training and validation.

If you use this dataset in your research, please acknowledge the authors and cite the associated publication.
For more details or to access the dataset, please contact the authors.

Associated publication:
Garcia Vargas, I., & Fernandes, H. (2025). Spatial and temporal deep learning algorithms for defect segmentation in infrared thermographic imaging of carbon fibre-reinforced polymers. Nondestructive Testing and Evaluation, 1–21. https://doi.org/10.1080/10589759.2025.2457593