Structural analysis of advanced polymeric foams by means of high resolution X-ray

computed tomography
,
M. Nacucchi , F. De Pascalis, M. Scatto, L. Capodieci, and R. Albertoni

Citation: AIP Conference Proceedings 1749, 020009 (2016); doi: 10.1063/1.4954492

View online: http://dx.doi.org/10.1063/1.4954492
View Table of Contents: http://aip.scitation.org/toc/apc/1749/1
Published by the American Institute of Physics
Structural Analysis of Advanced Polymeric Foams by Means
of High Resolution X-ray Computed Tomography
M. Nacucchi1, a) F. De Pascalis1, M. Scatto2 , L. Capodieci1 and R. Albertoni3
1
ENEA Italian National Agency for New Technologies, Energy and Sustainable Economic Development, Research
Centre of Brindisi, S. S. 7 Appia – km 706, I-72100 Brindisi, Italy.
2
Nadir Srl, Via Torino 155b, c/o Università Cà Foscari Venezia, I-30172 Mestre (VE), Italy.
3
Sitael SpA, Space Propulsion Business Unit, Via A. Gherardesca 5, I-56121 Ospedaletto (PI), Italy
a)
Corresponding author: [email protected]

Abstract. Advanced polymeric foams with enhanced thermal insulation and mechanical properties are used in a wide
range of industrial applications. The properties of a foam strongly depend upon its cell structure. Traditionally, their
microstructure has been studied using 2D imaging systems based on optical or electron microscopy, with the obvious
disadvantage that only the surface of the sample can be analysed. To overcome this shortcoming, the adoption of X-ray
micro-tomography imaging is here suggested to allow for a complete 3D, non-destructive analysis of advanced polymeric
foams. Unlike metallic foams, the resolution of the reconstructed structural features is hampered by the low contrast in
the images due to weak X-ray absorption in the polymer. In this work an advanced methodology based on high-resolution
and low-contrast techniques is used to perform quantitative analyses on both closed and open cells foams. Local
structural features of individual cells such as equivalent diameter, sphericity, anisotropy and orientation are statistically
evaluated. In addition, thickness and length of the struts are determined, underlining the key role played by the achieved
resolution. In perspective, the quantitative description of these structural features will be used to evaluate the results of in
situ mechanical and thermal test on foam samples.

INTRODUCTION
Foams are defined as materials containing gaseous voids surrounded by a denser matrix, liquid or solid. Thanks
to their advantageous properties, such as low density (50y70 kg/m3) and reduced thermal conductivity (0.020y 0.050
W/K m), polymeric foams are widely used in a large number of cross-industry applications (e.g., insulation, cushion,
absorbents and weight-bearing structures) for a total market value of $ 2B in 2015 [1-3]. Foams of high porosity
with interconnected pores have also been used as tissue engineering scaffolds for cell attachment and growth [4].
Important macroscopic properties such as mass density, thermal conductivity and mechanical properties are related
to the morphological features of the internal structure of the foams. Traditionally, optical or electron microscopy is
used to analyze the microstructure of polymeric foams, with the obvious disadvantage that only the surface of the
sample can be investigated. The sample needs to be cut in order to characterize its internal structures, but its surface
is inevitably damaged by the microtome knife. X-ray micro-tomography (X-ray micro-CT) imaging could allow for
a fully three dimensional, non-destructive analysis of polymeric foams. Successful examples of metal foams
characterization using X-ray micro-tomography are already reported in the literature since several years [5-7].
However, unlike metallic foams, the weak X-ray absorption in the polymer leads to a significant drop of the contrast
in the images, negatively affecting the resolution of the reconstructed structural features and requiring different
strategies to perform a three-dimensional quantitative analysis [8, 9]. The aim of this paper is to highlight the
success of a quantitative analysis carried out on two different types of foams, a polyurethane nanocomposite-based
insulating material and a flexible polyurethane foam. The choice of these two examples allows to comprise the two
extreme cases of polymeric foams: a more rigid closed cell foam and a flexible open cell. For each of the selected

NanoItaly 2015
AIP Conf. Proc. 1749, 020009-1–020009-8; doi: 10.1063/1.4954492
Published by AIP Publishing. 978-0-7354-1406-8/$30.00

020009-1
foams, local structural features of individual cells, e.g. equivalent diameter, anisotropy and orientation, or global
features, such as average strut thickness, have been measured and statistically evaluated.

MATERIALS AND METHODS

In the current work, the first case study is derived from an innovative polyurethane (PU) nanocomposite matrix.
In particular, several nanoplatelets, such as grapheme oxide, have been dispersed in a rigid polyurethane foam, in
order to improve its thermal insulation and compressive strength. Platelet-like nanostructures are also expected to
reduce the gas diffusion of PU foams (since they can increase the pathways for gas diffusion) and the thermal aging
phenomena. Rigid polyurethane foams have closed cell structures. In general, their thermal conductivity depends on
the foam density, cell size and orientation, ratio of close-to-open cell contents and on the thermal conductivity of the
blowing agents and filling materials [10]. This foam can be used to produce insulating panels. The second case study
is a flexible PU foam used for cushioning of furniture. The matrix incorporates organic modified layered double
hydroxides (LDH) fillers having an antimicrobial activity to control the undesirable growth of microorganisms in the
nanocomposite material. The open-cell structure is the dominating one. In this case, it is interesting to characterize
the solid component of the foam formed by an interconnected network of struts, by evaluating the strut length, strut
thickness and their connectivity. These parameters are known to have impact on mechanical and thermal properties
of the foam [11].

Experimental Set-up
The non-destructive volume characterization of foams was performed by a GE Phoenix Nanotom s CT system
equipped with a 180 kV/15 W nanofocus X-ray tube and a 2300 x 2300 pixel Hamamatsu flat panel detector. A
molybdenum target, suitable for weak absorbing specimens, was used. Due to their very low absorption,
polyurethane foams are likely to become transparent to X-rays, if acquisition parameters are not carefully selected.
The accelerating voltage of the X-ray tube was 49 kV. The beam current value was equal to 190 PA. The integration
time used was 1.000 s and each projection was obtained by averaging five (rigid foam) or four (flexible foam) 2D X-
ray images, in order to improve the signal-to-noise ratio. The number of projections was 1750 for the rigid foam and
2000 for the flexible one. In both the cases, the total acquisition time for the complete scan was approximately three
hours. The sample preparation took into account the need of obtaining a statistically significant number of cells in
the analyzed volume, without compromising the ability to achieve high resolution. Cubic shape samples having an
edge length of about 10 mm were used. The geometric magnification, M, is given by the ratio between the focus-
detector-distance (FDD) and the focus-object-distance (FOD):

M FOD FDD . (1)

High magnification values can then be obtained approaching the sample to the focus. In turn, the voxel size, V,
i.e. the effective pixel pitch, and thereby the achieved resolution depends upon the geometric magnification, namely:

V P M, (2)

where P is the physical pixel pitch of the detector, equal to 50 Pm in the current case. The following table
summarizes the geometric acquisition parameters of the scans

TABLE 1. Geometric acquisition parameters of the analyzed samples.

Foam type Sample identifier FOD (mm) FDD (mm) M Pm)
V (P
Rigid PU nanocomposite DC GO 16 263 16.4 3.0
Flexible PU nanocomposite 30 S 15 LDH 20 250 12.5 4.0
The 3D initial volume is a set of 2D gray level slices derived from reconstruction. The volume reconstruction was
carried out with the proprietary application software Phoenix datos|x 2 reconstruction. The electron micrographs
were acquired by a ZEISS Merlin scanning electron microscope.

020009-2
 3D Image Analysis
The 3D visualization and analysis software Avizo 8 Fire Edition of Visualization Sciences Group (a FEI
Company) was used for the image processing of the datasets. Some computational parts relating to anisotropy and
orientation of cells were accomplished by a home-made “C++” application based on VTK library [12].

Pore Reconstruction and Analysis

In order to characterize the cell foam features, the following processing steps were performed: de-noising of
grey-scale images by median filter, segmentation by thresholding, pore separation and labeling by watershed
algorithm, removing all the pores touching the borders of the image (border killing) and, finally, calculation of cell
features. Figure 1 shows the application of the above-mentioned steps in the case of the DC-GO foam.

(a) (b) (c)

(d) (e) (f)

FIGURE 1. Processing steps for segmentation and labeling: a) grey level image of a slice; b) binary image after thresholding; c)
inverted distance transform; d) watershed separation; e) labeling; f) border killing.

Note that, after the detection, the pores in the foam seem to be in contact with their neighbors (fig. 1b). This type of
output cannot be avoided either because of the noise or a too low resolution to distinguish the finest cell walls.
However, images obtained by using a scanning electron microscope clearly demonstrate that the foam has a closed-
cell structure, as shown in fig. 2. For this reason, in order to perform a proper analysis, the pore boundaries have to
be resolved and separated. As a consequence, the position of void boundaries has been reconstructed by an image
processing algorithm. The principle of the objects separation is to compute watershed lines on a distance map. The
watershed algorithm simulates the flooding from the center points of the cells; it then expands the regions according
to an “altitude” map (the inverted distance transform), until the regions reach at the watershed lines [13 - 15]. At this
point, each cell can be indexed starting from the integer value 1, in the so-called labeling operation. Finally, a
morphological operation to remove the cut cells in close proximity with the edges of the image is required. The label
image file so obtained was then used as input for the analysis process needed to characterize the cells. The software
module that computes their elongation and orientation interfaces with Avizo and is able to visualize a series of
glyphs (arrows) placed in the barycenter of each cell and oriented along the direction of the eigenvector associated

020009-3
with the maximum eigenvalue. The eigenvalues and eigenvectors are determined from the covariance matrix for
each cell [6, 16]. The arrows are scaled or colored according to a selected parameter, e.g. cell volume or scalar
anisotropy. The achievable information is integrated by a stereographic projection of the vectors representing the
axes of the spheroids approximating the cells.

(a) (b)
FIGURE 2. Secondary electron images of DC-GO foam sample: a) large field of view image obtained at low magnification; a
particular of the previous image observed at higher magnification.

Strut Analysis

In the second foam sample, the flexible one, the open-cell structure is the dominant one. In order to estimate the
mean thickness of the foam struts, the struts were separated from the pores, using a careful segmentation by
thresholding. All the pixels in the strut zones were set to 1 (foreground) in the output binary image, whereas all
remaining pixels were set to 0 (background). Then, a distance transformation converted the binary image into a
grey-values image in which all the foreground pixels have a value corresponding to the distance to the nearest
background pixel. To get the better approximation of the Euclidian distance, the chamfer distance transformation
was used [17]. Finally, the medial axis of the foam struts was calculated by means of a volume thinning method
(skeletonization) [18].
The skeleton of an object is theoretically defined as the locus of center voxels in the object. It is a binary image
whose foreground is centered with respect to the distance map [19, 20]. As a consequence, it is possible to associate
the values of the half ligaments’ thickness to the medial axis lines as a result of a multiplication point by point of the
skeleton by the distance map. Therefore, the average thickness of the foam struts can be calculated with the
following formula:

2r ¦ v ,
I
Pt (3)
Nv
where r is the CT nominal resolution, 6Iv is the sum of voxel intensities in the distance map along the strut
centerlines and Nv is the number of voxels in the binary skeleton image.

RESULTS AND DISCUSSION

Figures 3a and 3b show a 3D reconstruction of the DC-GO foam sample and its pores, respectively. The volume
of interest contains 1500 x 1500 x 1500 voxels in a cube of side length of 4.5 mm. The number of analysed cells was
2616.
Information about cell size and shape can be obtained directly from its volume. The equivalent diameter is a
measure of the object size. For a given particle, it is defined as the diameter of the spherical particle having the same
volume. The sphericity, i.e. roundness, is a measure of how spherical an object is and it is defined as the ratio
between the area of a spherical surface having its radius equal to half of the equivalent diameter to the actual area of
the cell. By definition, a sphere has sphericity equal to unity.

020009-4
 (a) (b)

(c) (d)
FIGURE 3. a) Foam 3D reconstruction; b) Pore segmentation; c) 3D Rendering of major axis of some pores; d) Stereographic
projection in the XY plane

Normally, the shape of a cell is not a perfect sphere but is closer to that of an elongated spheroid. This observation
allows for the definition of cell anisotropy. Each cell is replaced by an equivalent ellipsoid and its covariance matrix
is determined. The eigenvalues, O, of this covariance matrix are computed to obtain the anisotropy of a cell, namely:

An 1  Omin Omax . (4)

Note that a sphere has anisotropy equal to zero. Table 2 summarises the obtained results and Fig. 4 shows the
statistical distributions of equivalent diameter and anisotropy of the cells.

TABLE 2. Numerical characteristics of the statistical distributions of sample DC-GO

Equivalent Diameter (mm) Sphericity Anisotropy
Average 0.326 0.867 0.39
Standard Deviation 0.084 0.025 0.11

The anisotropy was analysed also from a global standpoint. Figures 3c and 3d show the rendering of the glyphs
placed in the barycentre cells along with the stereographic projection of the eigenvector of the main principal axis on
the XY plane. The sample appears to be essentially isotropic.

020009-5
 (a) (b)
FIGURE 4. Statistical distributions of: a) equivalent diameter of the cells; b) their anisotropy.

Figures 5a and 5b show a 3D reconstruction of the 30-S-15-LDH foam sample and a small portion of it at higher
magnification. The process of skeletonization was performed as described in the previous section.

(a) (b)
FIGURE 5. Sample 30-S-15-LDH: 3D reconstruction; b) reconstruction of the highlighted portion with a cube.

The average thickness computed by using equation (3) was 24 Pm. As for the study of connectivity each 3D
skeleton’s voxel can be classified into three different categories depending on its 26 neighbors:
x End-point voxel: if it has less than 2 neighbors;
x Junction voxel: if it has more than 2 neighbors;
x Segment voxel: if it has exactly 2 neighbors.
Finally, the voxel skeleton was converted into a spatial graph by using the “Trace Lines” Avizo module, as
shown in Fig. 6. A spatial graph consists of nodes and segments, in which nodes are junction and end-points, while
segments are the curved lines connecting the nodes [21]. From this graph, the strut lengths distribution and
connectivity information can be inferred providing a complete description of the sample lattice structure. The graph
was composed of 14535 segments having a mean length of 265 Pm - giving an overall length of about 3.8 m - and
8394 nodes. In Fig. 7, the histograms of the lengths of the struts and connectivity of the nodes are presented. About
63% of the nodes intersect 4 struts, whereas nearly 23% of them only 3.

020009-6
 (a) (b)
FIGURE 6. a) 3D skeleton; b) network graph.

(a) (b)
FIGURE 7. a) Strut length distribution; histogram of the connectivity of the nodes.

CONCLUSIONS
The current work attests that micro-CT is very effective in providing high resolution 3D information about
different foams, independently of their cell topology (open or closed). 3D image processing methods applied to CT
data allowed for a complete characterization of the foam structure at both local and global level. This, in turn, is
essential for a better understanding of the correlation between the foam structure and its physical properties. This
correlation is expected to enable polymer processing companies to optimize their products and production processes.
Often, nanofillers are added to the base polymer in order to improve the properties of the foam and expand the range
of its industrial applications. Although the nanofillers cannot be detected directly with the micro-CT due to their
nanometer size, if some changes to the foam structure are induced, then they can be measured by the micro-CT. The
internal structure of two nanocomposite foams was investigated. This information will be used to better understand
their behaviour in compression resistance and thermal conductivity tests. In perspective, quantitative description of
the structure in conjunction with in-situ compression or Peltier stages could provide an immediate interpretation of
how foam properties change under different loading.

ACKNOWLEDGMENTS
The authors wish to thank Eng. F. Felline and Eng. C. Rosato of Consorzio CETMA for the kind preparation and
supply of samples. Some research activities described in this paper were carried out as part of the following funded
research project, in the framework of PON 2007-2013 (National Operating Program “Research and

020009-7
Competitiveness” 2007-2013 - Grant: Italian Ministry of Education, Universities and Research): SILVER-
Nanoantimicrobial technologies and processes for the controlled modification of textile and other goods.

REFERENCES
1. N. J. Mills “Polymer Foams Handbook. Engineering and Biomechanics Applications and Design Guide”
Butterworth-Heinemann (Oxford, 2007).
2. D. Klempner in “Polymeric Foams and Foam Technology”, edited by D. Klempner and V. Sendijarevic,
(Hanser, Munich, 2004).
3. Paul Ashford et al. IPCC/TEAP Special Report: Safeguarding the Ozone Layer and the Global Climate
System, 315 – 348 (2015) available at https://www.ipcc.ch/pdf/special-reports/sroc/sroc07.pdf
4. Mikos, A.G, Temenoff, J.S. “Formation of highly porous biodegradable scaffolds for tissue engineering.” J.
Biotechnol. 3(2),114 - 119 (2000)
5. A. A Malcolm, A. A., Spowage, A. C., Shacklock A. P., Sim, L. M., Liu T. and Kennedy, A. R.
“Characterisation of ultra-light aluminium foams using micro-CT” SIMTech technical reports 7(1), 44-49
(2006).
6. Dillard, T., N’Guyen, F., Maire, E., Salvo, L., Forest, S., Bienvenu, Y. et al., “3D quantitative image analysis
of open-cell nickel foams under tension and compression loading using X-ray microtomography” Philos. Mag.
85(19), 2147-2175 (2005).
7. Buyens, F., Legoupil, S., Vabre, A., “Metallic foams characterization using X-ray microtomography” in CD-
Proceedings of International Symposium on Digital Industrial Radiology and Computed Tomography, Lyon,
2007.
8. Michaeli, W., Schrickte, L., Berdel K., “Structural Analysis of Polymeric foam” in Proceedings of the SkyScan
User Meeting, Gent, 2009, pp. 51-53.
9. Kastner, J., Kickinger, R., Salaberger, D., “Structural analysis of polymeric foams by sub-Pm X-ray computed
tomography” in Proceedings of the 10th European Conference on Non-Destructive Testing, Moskow, 2010,
(Curran Associates, New York, 2011) Vol. 3, pp. 1993-2000.
10. Van Krevelen D.W., Properties of Polymers (Elsevier, New York, 1990).
11. Liebscher, A. and Redenbach, C. "Statistical Analysis of the local strut thickness of open cell foams", Image
Analysis & Stereology 32,1-12, (2013).
12. Schroeder, W., Martin, K., Lorensen, B., The Visualization Toolkit: An Object-Oriented Approach to 3D
Graphics (Kitware Inc. 2013) ISBN: 1-930934-07-6
13. R. C. Gonzalez and R. E. Woods, Digital Image Processing, (Prentice Hall, Upper Saddle River, New Jersey,
2002) ISBN 0-201-18075-8.
14. Zhang, Q., Lee, P. D., Singh, R., Wu, G., Lindley, T. C., “Micro-CT characterization of structural features and
deformation behaviour of fly ash/aluminium syntactic foam” Acta Materialia 57(10), 3003-3011 (2009).
15. Schladitz, K., Redenbach, C., Sych, T., Godehardt, M. “Microstructural characterisation of open foams using
3d images” Published report of the Fraunhofer ITWM 148 (2008).
16. Benouali, A. H., Froyen, L., Dillard, T., Forest, S., N’Guyen, F., “Investigation of the influence of cell shape
anisotropy on the mechanical performance of closed cell aluminium foams using micro-computed
tomography” J. Mater. Sci. 40, 5801-5811 (2005).
17. Borgefors, G., “Distance Transformations in arbitrary dimensions” Computer Vision, Graphics and Image
Processing 27(3), 321-345 (1984)
18. Pudney, C., “Distance-Ordered Homotopic Thinning: A Skeletonization Algorithm for 3D Digital Images”
Computer Vision and Image Understanding 72(3), 404-413 (1998).
19. Lee, T.C., Kashyap, R.L., “Building Skeleton Models via 3-D Medial Surface/Axis Thinning Algorithms”
Graphical Models and Image Processing 56(6), 462-478 (1994).
20. Tran, S., Shih, L., "Efficient 3D binary image skeletonization" Computational Systems Bioinformatics
Conference - Workshops”, Stanford, 2005, pp. 364-372, doi:10.1109/CSBW.2005.57.
21. Fouard, C., Malandain, G., Prohaska, S., Westerhoff, M. “Blockwise processing applied to brain microvascular
network study” IEEE Transaction on medical imaging 25(10), 1319-1328 (2006).

020009-8