A major challenge in the analysis of the FIB-SEM images is the binarization of the grey scale images into the solid and pore phases (5). This problem arises because the SEM detects the solid particles inside the pores which are not in the plane of the image. This adds to the complexity of the segmentation by introducing an additional dependency to not only distinguish between the pore and solid phase but also, distinguish between solid phase in the plane of the image and solid phase in the background. In order to tackle this problem, advanced segmentation algorithms have been developed (5,6). However, these algorithms lack consistency across datasets and often require manual intervention which limits their robustness.
Binarization can be improved by enhancing the sample preparation for the image acquisition. This has been done by penetrating the pores with filling material by embedding the sample in an epoxy based resin (7,8) or silicon based resin (9), platinum vapor deposition (10) and atomic layer deposition (ALD) of zinc oxide (11). ALD and platinum vapor deposition result in partial filling of the pores but provide a higher contrast from the carbon rich backbone. However, due to the partial filling manual corrections and additional image filtering operations are required post segmentation of the images. Epoxy and silicon based resin embedding results in complete filling of the pores but provides a very low contrast for the images. Ghosh et al. (8) recently showed that images obtained using the backscattered electrons (BSE) provide a better contrast than the secondary electron (SE) images (8,9) for a PEMFC CL embedded with an epoxy based resin. However, Ghosh et al. (8) only examined and compared 2D images of the CL. Comparison of a full 3D reconstruction of the PEMFC CL using FIB-SEM with and without epoxy embedding has yet to be performed.
In the present article, 3D reconstructions of a conventional high surface area CL are performed using FIB-SEM images of the same sample treated with and without epoxy embedding to analyze the differences in image analysis and structure for both the modes of sample preparation. Figure 1 below shows the raw images obtained using SE mode for the sample without epoxy embedding and BSE mode for the sample with epoxy embedding. Multiple stacks of images are processed from both the datasets to ensure global validity of the results. Image processing operations are performed to further enhance the raw images before segmentation.
Image analysis reveals that a simple thresholding algorithm, such as Otsu, is sufficient to accurately segment the images for the sample with epoxy embedding due to the lack of background features. Comparison of the reconstructions for the two modes suggests an increase in porosity and chord length function (12) for the sample imaged using epoxy embedding due to a more accurate segmentation. Transport simulations are being carried out to compute the effective transport properties.
References
1. S. Thiele et al., Nano Research, 4(9), 849 (2011).
2. M. Sabharwal et al., Fuel Cells (2016).
3. X. Zhang et al., Int. J. Hydrogen Energy, 39(30), 17222 (2014).
4. H. Ostadi et al., Journal of Membrane Science, 351(1), 69 (2010).
5. M. Salzer et al., Materials Characterization, 95, 36 (2014).
6. M. Salzer et al., Materials Characterization, 69, 115 (2012).
7. H. Iwai et al., Journal of Power Sources, 195(4), 955 (2010).
8. S. Ghosh et al., International Journal of Hydrogen Energy, 40(45), 15663 (2015).
9. M. Ender et al., Electrochemistry Communications, 13(2), 166 (2011).
10. S. K. Eswara-Moorthy, P. Balasubramanian, W. van Mierlo, J. Bernhard, M. Marinaro, M. Wohlfahrt-Mehrens, L. Jörissen and U. Kaiser. Microscopy and Microanalysis, 20(05), 1576 (2014).
11. S. Vierrath, F. Güder, A. Menzel, M. Hagner, R. Zengerle, M. Zacharias and S. Thiele. Journal of Power Sources, 285, 413 (2015).
12. L. M. Pant, M. Sabharwal, S. Mitra and M. Secanell. ECS Transactions, 69(17), 105 (2015).