Scanning electron microscopy datasets for local fibre volume fraction determination in non-crimp glass-fibre reinforced composites

The fatigue damage evolution depends on the local fibre volume fraction as observed in the co-submitted publication [1]. Conventionally, fibre volume fractions are determined as an averaged overall fibre volume fraction determined from small cuts of the laminate. Alternatively, automatically stitching of scanning electron microscopy (SEM) images can make high-resolution scans of large cross-section area with large contrast between the polymer and glass-fibre phase. Therefore, local distribution of the fibre volume fraction can be characterised automatically using such scan-data. The two datasets presented here cover two large Field of Views scanning electron microscopy (SEM) images. The two images is generated from between 1200 and 1800 high-resolution scan pictures which have been stitched into two high-resolution tif-files. The resolution corresponds to between 700 and 5000 pixels covering each fibre. The datasets are coming from two different non-crimp fabric glass fibre reinforced epoxy composites typically used in the wind turbine industry. Depending on the regions analysed, fibre volume fraction in the range of 50–85% is found. The maximum local fibre volume fraction is found averaging the local fibre volume fraction over 5 × 5 fibre diameter (80 × 80 µm2) areas. The local fibre volume fraction has been used in the analysis performed in [1].


a b s t r a c t
The fatigue damage evolution depends on the local fibre volume fraction as observed in the co-submitted publication [1] . Conventionally, fibre volume fractions are determined as an averaged overall fibre volume fraction determined from small cuts of the laminate. Alternatively, automatically stitching of scanning electron microscopy (SEM) images can make highresolution scans of large cross-section area with large contrast between the polymer and glass-fibre phase. Therefore, local distribution of the fibre volume fraction can be characterised automatically using such scan-data. The two datasets presented here cover two large Field of Views scanning electron microscopy (SEM) images. The two images is generated from between 1200 and 1800 high-resolution scan pictures which have been stitched into two high-resolution tif-files. The resolution corresponds to between 70 0 and 50 0 0 pixels covering each fibre. The datasets are coming from two different non-crimp fabric glass fibre reinforced epoxy composites typically used in the wind turbine industry. Depending on the regions analysed, fibre volume fraction in the range of 50-85% is found. The maximum local fibre volume fraction is found averaging the local fibre volume fraction over 5 × 5 fibre diameter (80 × 80 μm 2 ) areas. The local fibre volume fraction has been used in the analysis performed in [1]

Value of the Data
• The high-resolution large field of view SEM scanning data is used to determine the local fibre volume fraction distribution in two different non-crim p fabric based glass-fibre composites. The data is used to characterise local fibre volume fraction in conventional non-crimp fabrics, which values subsequently is used in reference [1] . • The industry and academia can use the provided datasets for studying fibre volume fraction variations in non-crimp fabric-based composites. • The two datasets can be used as a benchmark dataset for developing segmentation and analysis tools for local fibre volume fraction and fibre diameter determination. The datasets can also be used for investigating variation in the fibre volume fractions at different locations in the fabric, e.g., close to the backing fibre bundle.

Data Description
For each of the two cases, five files are made available at the Zenodo repository, see [4] . Those two file-set contains: • Tif-file: The stitched SEM scanned image which is used in the fibre volume fraction analysis   • Zip-file: Collection of the individual SEM scanned images and meta-data files behind the stitched SEM image.
The full cross-section and a zoom-in on the scanning electron microscope images for the two cases are shown in Figs. 1 and 2 . The images were acquired using a Tescan VEGA3 SEM with the settings as listed in Table 1 . The composite samples were cut orthogonal to the dominating fibre orientation, with the cutting surface, subsequently polished and applied with an approximately 10 nm thin layer of carbon using a Bal-Tec SCD 005 Sputter Coater. The material samples are cut from fatigue test samples used in reference [2] and [3] . The two cases will here be denoted as Cases 1 and 2, respectively. The images were taken with a pixel size of 527.25 nm and 195.31 nm for Figs. 1 and 2 , respectively, using a source magnification of 692x at a high tension of 15 or 20 kV using a backscatter (BSE) detector.

Experimental Design, Materials and Methods
The scan parameter for the scanning electron microscopy (SEM) images, see Table 1 , were carefully selected to generate the best possible contrast difference between fibres and resin. The SEM images were acquired and afterward stitched together with the "image snapper" function in the Tescan software VegaTC to generate SEM images of large regions with high resolution. The SEM images were processed with a MATLAB script, where the data were loaded with the function imread and binarised using an Otsu threshold value determined by the Matlab-function otsuthresh based on a histogram of a central 10 0 0 × 10 0 0 pixel subsection as shown in Figs. 3 and 4 for Cases 1 and 2, respectively.
The binarisation shown to the left in Figs. 3 and 4 can now be used for calculating the fibre volume fraction simply by finding the ratio between the number of binarised pixels above the  Otsu threshold (fibers) over the total amount of pixels inside the region. For the total scanned area shown in Figs. 1 and 2 , the overall fibre volume fraction is found to be V f = 0 . 598 and V f = 0 . 525 , respectively. A value that in Table 2 is compared with the values measured by a back-calculated or burn-of experiment reported in the two references [2] and [3] , respectively. For Case 1, it should be noted that not the total layer of the lower biax ply is included in the SEM scan, which may result in the slightly larger overall fibre volume fraction found compared to the value reported in reference [2] . Figs. 5 and 6 show manually segmented UD fibre bundles from inside the SEM scanned region. In Case 1, 8 different unidirectional bundles are segmented, while it for Case 2 includes a major part of two unidirectional bundles. The unidirectional bundles were manually segmented using the Region of Interest drawing tool in the Image Segmenter toolbox in Matlab. Inside each segmented region, the fibre volume fraction was calculated in the same way as for the overall fibre volume fraction. The values are reported in Table 2 together with their standard deviations.
Figs. 7 and 8 show the variation of the fibre volume fraction calculated by using a moving area averaging of a 5 × 5 fibre diameters sized area which for a fibre diameter equal to D f = 16 μm corresponds to approximately 80 μm × 80 μm . This was done using the Matlab function conv2 and was applied over the full scanned region as shown in Figs. 7 and 8 . From this, a local fibre volume fraction is determined. In addition to the full scanned region, an image of a region with regions of high fibre volume fraction is presented. From the contour plots, regions with local fibre volume fraction up to around V f ≈ 0 . 85 for both cases were identified.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships which have or could be perceived to have influenced the work reported in this article.