Thermal imaging dataset from composite material academic samples inspected by pulsed thermography

This paper presents a thermal imaging dataset from composite material samples (carbon and glass fiber reinforced plastic) that were inspected by pulsed thermography with the goal of detecting and characterizing subsurface defective zones (Teflon inserts representing delaminations between plies). The pulsed thermography experiment was applied to 6 academic plates (inspected from both sides) all having the dimensions of 300 mm x 300 mm x 2 mm and same distribution of defects but made of different materials: three plates on carbon fiber-reinforced plastic (CFRP) and three plates made on glass fiber reinforced plastic (GFRP) specimens with three different geometries: planar, curved and trapezoidal. Each plate contains 25 inserts having length/depth ratios between 1.7 and 75. Two FX60 BALCAR photographic flashes (6.2 kJ per flash) were used to generate the heat pulse (2 ms duration), an X6900 FLIR infrared camera using ResearchIR software to record the thermal images and a custom-built software/control unit to synchronize data recording with pulse generation. Finally, the dataset proposed consists of 12 sequences of approximately 2000 images of 512 × 512 pixels each.


a b s t r a c t
This paper presents a thermal imaging dataset from composite material samples (carbon and glass fiber reinforced plastic) that were inspected by pulsed thermography with the goal of detecting and characterizing subsurface defective zones (Teflon inserts representing delaminations between plies). The pulsed thermography experiment was applied to 6 academic plates (inspected from both sides) all having the dimensions of 300 mm x 300 mm x 2 mm and same distribution of defects but made of different materials: three plates on carbon fiber-reinforced plastic (CFRP) and three plates made on glass fiber reinforced plastic (GFRP) specimens with three different geometries: planar, curved and trapezoidal. Each plate contains 25 inserts having length/depth ratios between 1.7 and 75. Two FX60 BALCAR photographic flashes (6.2 kJ per flash) were used to generate the heat pulse (2 ms duration), an X6900 FLIR infrared camera using ResearchIR software to record the thermal images and a custom-built software/control unit to synchronize data recording with pulse generation. Finally, the dataset proposed consists of 12 sequences of approximately 20 0 0 images of 512 × 512 pixels each.
© 2020 The Author(s Wind speed (0 m/s) Distance between IR camera and sample (100 cm) Angle between IR camera lenses and the sample surface (90 °) Distance between flashes and sample (50 cm) Angle between flashes (each) and sample (45 °). The heat pulses of them coinciding at the center of the sample surface.

Description of data collection
The infrared camera recorded the thermal evolution on the surface of the inspected CFRP/GFRP sample for several seconds (approximately 16 to 17 s) at 120 and 145 frames/s sampling rate, from before applying a heat pulse, during the heat pulse and during cooling. The experiment was performed from the two faces of specimen (providing shallow defect depths from the front face and deeper depths from the back face), and each sequence was labeled and saved into an independent file. Data

Value of the Data
• The research community can use the thermal images dataset to evaluate the infrared imaging processing techniques performance. The methods can focus on locating laminar defective regions and on the assessment of the attributes of these regions, such as the shape, size, depth, or specific materials properties of thermograms that compose the dataset. • The research community can use the dataset of thermal images to develop and test processing strategies without having an experimental platform available. • The dataset of raw thermal images contains undesirable effects of low contrast, non-uniform heating, and noise. It is possible to use the dataset to test or develop processing techniques to overcome these unwanted effects.
• Besides, the different f eatures of the dataset (thermograms of two different materials, a variety of sample geometries, and defect sizes and locations) will allow a broader study concerning the scope and limitations of the processing methods applied to the images.

Data Description
The objective of the pulsed thermography experiment is to monitor the surface temperatures of the samples as a function of time and the flow of transient heat generated through an energy stimulus in the samples [3][4][5][6] . The thermal stimulus allows the generation of enough temperature differences to identify sub-surface anomalies if they are present [7][8][9] .
The 12 image sequences provided in this dataset show the evolution of temperature over time on the surface of composite materials tested by pulsed thermography experiment (see Fig. 1 ). Sequences are stored in folders containing '.CSV' files. Table 1 shows the physical properties of the materials. Table 2 presents the infrared camera specifications. Table 3 lists the acquisition conditions that were adjusted in the experiment. Fig. 2 shows the geometry of the CFRP/GFRP samples used and presents the characteristics of their defective zones. Lastly, Table 4 consolidates the dataset files information. Folders and files are label as CFRP/GFRP-samplename _facq-frequencyvalue _s-sideofinspection _Img-frames . This tagging scheme describes the sample name tested, the acquisition frequency used, the sample side of      inspection taken, as well as the number of each frame (for .CSV files) within the sequence, and the whole number of frames from the image sequence (for the folders), respectively.

Experimental Design, Materials, and Methods
The pulsed thermography procedure is composed of three main stages: (1) a CRFP/GFRP sample with internal artificial defects is placed perpendicularly to the IR camera at a fixed distance, the power flashes are placed in reflection mode [1] between the camera and the specimen also at a fixed distance, (2) the acquisition parameter (temperature calibration range, emissivity, integration time) are set in the IR acquisition software, and (3) the inspected sample is heated with the thermal stimulus of the heat sources and simultaneously the temperature evolution of its surface is recorded. These stages are shown in Fig 1 . The acquisition procedure was performed in one session under constant conditions for all the tested specimens ( Table 1 ). The dataset was generated with a pulsed thermography experiment ( Tables 2 and 3 ) on three CFRP samples and three GFRP samples. The CFRP0 06/GFRP0 06 samples have a flat geometry ( Fig. 2 a). The CFRP0 07/GFRP0 07 and CFRP0 08/GFRP0 08 specimens have a curved and trapezoidal geometries, respectively ( Fig. 2 b and c). Each of these samples contain 25 square internal defects (Teflon inserts) with variable area, and depth, but all having the same thickness.
The infrared camera records for several seconds (approximately 16 to 17 seconds) the thermal evolution on the surface of the inspected CFRP/GFRP sample while applying a heat pulse of 2 ms (at Full-width half-maximum) and 12.8 kJ of energy. Each specimen/sample was tested in reflection mode under two conditions. The first condition takes the thermal images on the front surface from the composite material where the defective zones are close to this side (shallower). A second condition takes the thermal images on the back surface where the defects are deeper.
The acquisition of the database required the below materials.