Вы находитесь на странице: 1из 10

Journal of Colloid and Interface Science 280 (2004) 202211

www.elsevier.com/locate/jcis

The 3D structure of real polymer foams


Matthew D. Montminy a , Allen R. Tannenbaum b , Christopher W. Macosko a,
a Department of Chemical Engineering and Materials Science, University of Minnesota, 421 Washington Ave. SE, Minneapolis, MN 55455-0132, USA
b Schools of Electrical & Computer Engineering and Biomedical Engineering, Georgia Institute of Technology, 777 Atlantic Dr. NW, Atlanta,
GA 30332-0250, USA
Received 17 September 2003; accepted 26 July 2004
Available online 2 September 2004

Abstract
The intricate structure of polymeric foams may be examined using 3D imaging techniques such as MRI or X-ray tomography followed
by image processing. Using a new 3D image processing technique, six images of polyurethane foams were analyzed to create computerized
3D models of the samples. Measurements on these models yielded distributions of many microstructural features, including strut length and
window and cell shape distributions. Nearly 8000 struts, 4000 windows, and 376 cells were detected and measured in six polyurethane foam
samples. When compared against previous theories and studies, these measurements showed that the structure of real polymeric foams differs
significantly from both equilibrium models and aqueous foams. For example, previous studies of aqueous foams showed that about 70% of
foam windows were pentagons. In the polymeric sample studied here, only 55% of windows were pentagonal.
2004 Elsevier Inc. All rights reserved.

Keywords: Foam; Microstructure; Cells; Characterization; Visualization; 3D image processing; Minimal surface

1. Introduction ins tetrakaidecahedral [6] cell shapes, tetrahedral cell in-


tersections, and Plateaus laws for the formation of soap
The physical properties and potential applications of froths [2,710]. Other studies simplified foam shapes fur-
foams result from the chemistry and physical properties of ther, using cubic unit cells to describe foam behavior [4,11].
the bulk material and the cell structure of the foams. The While these relations allowed the creation of many prac-
effect of foam chemistry and processing conditions on phys- tically useful and fairly accurate engineering correlations,
ical properties has been researched extensively, and many they are abstractions which do not necessarily accurately de-
useful empirical structureproperty relationships have been scribe real polymeric or metallic foams.
developed, including many foam density/property relation- Indeed, solid foams are very different from aqueous
ships [15]. However, relatively little experimental work has foams because the base materials have very different ma-
been done to relate foam microstructure with macroscopic terial properties. Depending on composition and process-
physical properties. This is in part because there are few ing temperature, polymeric and metallic foams are usually
good structure characterization methods. formed from materials which are significantly more viscous
Nonetheless, many theoretical models relating the physi- than water, and, during their formation, heat transfer, phase
cal structure and mechanical performance of polymer foams transition, and viscoelastic effects may come into play. Be-
have been created using rules formulated by observing soap cause of the nature of the base materials and cooling and
froths. These models and discussions relied on assumptions phase transition effects, cell nucleation and growth in these
such as monodispersity, rhombic dodecahedral or Kelv- foams is very different than in liquid foams. Most impor-
tantly, solid foams are not equilibrium structures. Depending
* Corresponding author. on processing conditions, their high viscosity and low melt-
E-mail address: macosko@cems.umn.edu (C.W. Macosko). ing points results in quenching of the foaming process well
0021-9797/$ see front matter 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.jcis.2004.07.032
M.D. Montminy et al. / Journal of Colloid and Interface Science 280 (2004) 202211 203

before a stable state in the liquid phase is reached. This into the model, and can be manipulated to allow for the var-
at least partially invalidates the use of the observations of ious styles of measurements.
Plateau [7] in describing the structure of solid foams. To Such models can be created using 3D imaging techniques
further advance our knowledge of foam structureproperty such as magnetic resonance imaging (MRI) or X-ray com-
relationships, better investigation of the structures of real puterized tomography (CT) coupled with 3D image process-
solid foams is needed. ing [1927]. However, the difficulties involved in 3D image
Foam measurements are generally taken using visual in- analysis, including image noise and memory and processing
spection, photography, or optical microscopy. In particu- constraints, have prevented previous models from tackling
lar, optical microscopy techniques can be used to measure large datasets and achieving widespread use in industry. Pan-
many features, including window thickness, cell diameters grle et al. [25] did not model any cells, but estimated cell
and strut lengths [12,13]. Confocal microscopy can also be density from strut and window statistics. Kose and Kashi-
used to examine foam structures [14]. However, because wagi [23,24] detected and measured 8 cells in their gelatin
of the complicated three-dimensional structures of foams, foam sample. Monnereau and Vignes-Adler [21] detected
measuring foam features manually using these techniques and modeled 9 bubbles within wet and dry foam samples.
is tedious, time consuming, and often requires destruction While Cenens et al. [19] modeled and examined a much
of the sample. In addition, some distinguishing features of more significant number of cells, they did so using a water-
foams, including cell volume, are extremely difficult to mea- shed segmentation method which estimated strut locations
sure at all using traditional techniques. As a result, there based on the approximated locations of cell centers. This
are few experimental studies which correlate microstructure approach does not necessarily capture the true structure of
and macroscopic properties. Cell size and anisotropy [1] and foams, since it implicitly assumes that all cell borders will
open-celled content [2] are known to affect some physical lie halfway between cell centers. This assumption is ques-
properties, but because of the lack of complete structural tionable when analyzing polydisperse foams.
data, predictive modeling is not possible. In this study, computer models of real, open-celled, poly-
Image processing can be used in order to help ameliorate disperse polyurethane foams were created. These models are
this dilemma. Previous studies using 2D image processing true to the actual structure of the foam specimens modeled,
to automate foam measurement collected data in a fraction with struts, vertices, windows, and cells detected directly
of the time required for physical inspection. In most earlier from their locations in 3D X-ray tomography images of
studies using automated image analysis to extract structural real foams. Such computerized foam models may become
information; however, only one or a few structural mea- invaluable in examining the complicated relationships be-
surements were compared to physical properties [1517]. tween the physical properties and structures of solid foams,
Furthermore, the use of different standards for even sim- or in detecting subtle changes in foam structure resulting
ple measurements such as cell diameter [18] makes it dif- from changes in foam formulation.
ficult to compare the results of similar studies to develop
a body of structureproperty knowledge. For these reasons,
a method for more complete characterization of foams is 2. Materials and methods
needed.
Creating an accurate 3D scale model of an actual foam The structures of six polyurethane foam samples were
sample in computer memory provides a method for more investigated in detail over the course of this research [28].
complete characterization. Then, many spatial measure- Background information on each samples is provided in
ments such as strut length and cell size can easily be made Table 1. Samples 1a and 1b were high density foams ob-
on the scale model rather than measuring the actual sample. tained from the same foam sample provided by Air Products
If sufficiently large and reliable computerized 3D models of and Chemicals (Allentown, PA). Samples 2a and 2b were
real foams can be created, researchers can obtain measure- taken from a low-density foam sample from Air Products.
ments from the 3D model rather than measuring features of Both Samples 1 and 2 were made on a slabstock production
the original foam sample. Measurement criteria can be built machine. Sample 3 is FoamEx 40, a commercial structural

Table 1
Polyurethane foam samples analyzed in this study
Sample Original volume Mass density Approx. cell Description Source
# size (voxels) (lb/cu ft) density (cells/cm)
1a 293 293 300 2.1 20 Open-celled slabstock foam Air Products
1b 226 226 300 2.1 20 Open-celled slabstock foam Air Products
2a 294 312 300 1.1 14 Open-celled slabstock foam Air Products
2b 275 284 300 1.1 14 Open-celled slabstock foam Air Products
3 245 236 128 6 Reticulated slabstock foam FoamEx 40
4 213 223 100 8 Handmade foam bun Created in lab
204 M.D. Montminy et al. / Journal of Colloid and Interface Science 280 (2004) 202211

Fig. 1. Cross-section of the 3D image of the foam sample. Fig. 2. Surface rendering of the foam sample. Notice that this sample is very
large and contains about 100 cells. In the foam visualizations presented in
this paper, fog is used to convey depth. Lighter, washed-out sections of
foam, and Sample 4 was created in our laboratory. Samples the image are farther away from the viewer than darker sections.
1, 2, and 4 were made using standard flexible polyurethane
formulations similar to those described in Refs. [1], [14],
image volume, which may have up to 256 different colors,
and [25]. All images presented in this section are from Sam-
into a binary volume image. In this binary volume image,
ple 1a.
voxels with a value of 0 (black) represent the foam struc-
A 3D image of this foam was obtained via micro X- ture, while those with a value of 1 (white) represent void
ray computerized tomography (CT), in which a 3D image space. This binary volume image is the volume model. The
is mathematically reconstructed from a set of 2D X-ray creation of this model is important because segmenting the
shadow images [29]. CT was performed on these samples image creates a real distinction between the locations in the
using a SkyScan (Aartselaar, Belgium) 1072 desktop micro- image where the foam structure exists and those where void
tomograph, and a typical foam image cross-section from the space exists. This distinction is necessary for the detection
sample is presented as Fig. 1. The black voxels within this and analysis of shapes within the image and the creation of
image represent the locations of foam struts, while the white the next two models. A 3D surface rendering of this volume
areas represent void space. The best spot resolution currently model is shown in Fig. 2.
available on this instrument is about 5 m [30], so this imag- From the volume model, a stick figure model of the foam
ing method is only practical for the in-depth observation of sample is created. The stick figure model is used to extract
samples with average cell diameters of over 100 m. most of the useful structural information present in the foam
3D images obtained using X-ray CT or MRI contain a image. The stick figure model contains the locations of the
great deal of information about foam structure, but extract- struts, windows, and cells which comprise the cellular struc-
ing this information from these images is difficult. A 3D ture of the foam. The transition of the thick struts from the
image processing algorithm [31] and an image processing volume model into a stick figure is performed using an im-
software package called FoamView [28] were developed age processing technique called volume thinning.
specifically to extract structural information from 3D foam During volume thinning, dark voxels are successively re-
images.1 moved from the thick foam structure until only a set of
A brief overview of the image analysis process is pro- strut centerlines remains. The volume thinning algorithm
vided here. Once a 3D foam image has been acquired and used [32] guarantees that the resulting skeleton preserves
converted to the appropriate data format, the image analy- the connectivity of the original foam sample and is as close
sis process can begin. The first step in the process involves to the actual centerline as possible. From this thinned vol-
determining which voxels within the image volume repre- ume image, a stick figure is created. Noise in the image and
sent foamed material and which voxels represent void space. thick, difficult to resolve vertices cause some artifacts to ap-
This image processing method is called segmentation. The pear within the stick figure, but the stick figure can be refined
segmentation process converts the original 8-bit grayscale automatically or interactively using our package FoamView.
A result for this sample is presented as Fig. 3.
1 FoamView, a users manual, and a 3D model of the sample described This resulting stick figure model consists of lists of the lo-
here are available online as Supplementary Material. cations of vertices of the foam structure and the struts which
M.D. Montminy et al. / Journal of Colloid and Interface Science 280 (2004) 202211 205

Fig. 3. Refined stick figure model of the sample. This model contains over
100 complete cells. In this visualization, the foam struts are shown as blue Fig. 4. Visual comparison of the surface and stick figure models for the
cylinders, vertices are small red spheres, and detected cells show up as large sample.
transparent blue spheres. Detected windows are shaded green.

connect them. From this information, foam cells and win-


dows are detected. Since the locations of all of these features
are stored in computer memory, struts, strut intersection an-
gles, windows, and cells are all measured directly using this
model.
This image analysis approach yields models which are
derived directly from the actual strut structure apparent in the
original 3D image, ensuring faithfulness to the actual foam
structure. These models correspond well with surface ren-
derings showing the surface of the foam in the original 3D
image, as shown in Figs. 4 and 5 and inspection of the results
in 3D using the FoamView software.2
Thanks to the high level of automation present within the
image analysis process, this method can quickly provide data
on foam samples. Image acquisition at 1024 1024 512
voxel resolution requires about 3 h of instrument time us-
ing current X-ray CT technology, and image processing took Fig. 5. Close-up of the foam structure showing the correlation between the
from 24 h for the above samples, depending on sample detected strut, vertex, and cell locations and the original foam volume. The
size. Most of this time (all but about 20 min for each sam- large, blue spheres in the image indicate the centers of detected foam cells.
ple) was spent on user-assisted detection of foam features Some struts (top middle and bottom left) were not detected because they are
incomplete struts on the edge of the image.
using FoamView. The user can view and edit the 3D stick
figure structure while comparing it to the original 3D image
in FoamView, improving on the automated result and ensur- ter filtered by the human eye than by computer algorithms,
ing that stick figure model corresponds well to the original especially in 3D.
3D image of the foam structure. While this prevents exact re-
producibility of results, it allows models to be created from
3. Results and discussion
fairly poor foam images. Noise and other artifacts can be bet-
As shown in Table 2, between the six samples imaged and
2 The original foam volume and stick figure model can be compared analyzed over the course this research, 376 cells were de-
interactively using the FoamView software and sample datasets available tected. Nearly 8000 foam struts and 4000 windows were also
online as Supplementary Material. detected and measured. Note that the cell densities listed
206 M.D. Montminy et al. / Journal of Colloid and Interface Science 280 (2004) 202211

Table 2
Features measured in the samples analyzed in this study
Sample Approx. cell # features detected Avg. cell Std. dev. of cell
# density (cells/cm) Struts Vertices Windows Cells volume (mm3 ) volume (mm3 )

1a 20 2066 1111 1063 106 0.19 0.15


1b 20 1249 682 628 60 0.21 0.17
2a 14 1328 740 631 54 0.31 0.29
2b 14 1120 620 559 58 0.32 0.37
3 6 1176 651 578 54 4.15 1.8
4 8 1024 568 494 44 3.21 0.57
Total 7963 4372 3953 376

in the table are linear cell densities estimated by counting the Table 3
number of cells in a straight line on the surface of the sample Measurements and statistics describing the cellular structure of the sample
before processing, while the cell volumes listed were mea- Basic measurements
sured using the 3D microscopy technique described above. Image resolution 293 293 300 voxels
Each cell volume is roughly proportional to the cube of the Voxel size 12.27 12.27 12.27 m
reciprocal of the corresponding linear cell density, resulting Sample volume 47.30 mm3
in the higher average cell volumes found in Samples 3 and 4. Solid volume 3.18 mm3
Solid fraction 0.0671
Because of its complexity and polydisperse cell size dis- Anisotropy ratio 1.292
tribution, only the results for Sample 1a will be discussed
in detail here. See Ref. [28] for detailed information on the Window shape distribution # features detected
remaining samples. 9 Triangles 2066 Struts
251 Quadrilaterals 1111 Vertices
587 Pentagons 1063 Windows
3.1. Overview of sample measurements and statistics 205 Hexagons 106 Cells
11 Heptagons +
The resulting 3D model consists of a set of vertex loca- Cell structure statistics
tions and lists of the vertices composing foam struts, win-
Strut Interior Window Cell Cell Cell
dows, and cells. This information is used to calculate many length angles area volume surface area isoparametric
statistics and perform many measurements of foam features. (mm) (deg) (mm2 ) (mm3 ) (mm2 ) quotient
Table 3 shows selected measurements and statistics describ- Mean 0.281 106.72 0.133 0.193 1.749 0.646
ing the structure of Sample 1a, our highlighted foam sample. Std. dev. 0.104 17.78 0.080 0.145 0.834 0.075
As shown in Table 3, the vertex location and connectivity Min 0.079 20.70 0.006 0.012 0.323 0.407
information from the stick figure model generated from this Max 0.716 164.57 0.520 0.937 5.151 0.823
sample image provides a wealth of information. Structural Cell shape information
unit measurements, including strut lengths, interior angles, Mean # of windows per cell 13.01
window areas, and cell volumes and surface areas are avail-
able. Shape information, including cell and window shapes,
cell isoparametric quotients, and the amount of anisotropy
present within the structure can also be calculated. The
anisotropy ratio describes the elongation of the foam in the
rise direction versus that in the other major directions. The
isoparametric quotient, 36V 2 /A3 , is a ratio of volume to
surface area which indicates how spherical a cell is; a sphere
has an isoparametric quotient of 1. Together, all this infor-
mation provides a detailed snapshot of the foam samples
microstructure.

3.2. Strut lengths

2066 individual struts were detected and measured within


this sample. The mean length of foam struts detected within
the sample was 0.28 mm. A strut length distribution his-
togram for the sample is presented as Fig. 6. Notice that
the strut lengths fall in a right-skewed distribution, a com- Fig. 6. Strut length distribution for the foam sample.
M.D. Montminy et al. / Journal of Colloid and Interface Science 280 (2004) 202211 207

Fig. 7. Strut intersection angle distribution for the foam sample. Fig. 8. Window area distribution for the foam sample.

3.4. Window size and shape


mon distribution shape for statistics describing natural sys-
tems. The shape of this strut length distribution corre-
Many practical characteristics of foams, including their
sponds well with those detected in previous foam structure
ability to be used as filters or catalyst supports, depend upon
studies [19,25] as well as with the strut length distribu-
the sizes of a foams windows. The distribution of areas for
tions of other samples analyzed over the course of this re-
the 1063 windows detected within the sample is provided as
search [28].
Fig. 8.
Window area is calculated by partitioning each window
3.3. Strut intersection angles into a set of triangles, and calculating the area of these tri-
angles. Each of these triangles has vertices at two adjacent
When a load is placed on a foam, it places a stress on window vertices plus the windows centroid. This is neces-
both the struts of the foam and the vertices at which they sary since the vertices of real foam windows rarely lie within
intersect. Therefore, analysis of the angles at which struts a plane.
intersect may provide important insight into foam behavior. Notice that the window area distribution is another right-
Strut intersection angles have been previously examined and skewed distribution, but that this distribution is more signif-
measured using image processing approaches by Cenens et icantly skewed than the strut length distribution presented
al. [19], Kose [23], and Pangrle et al. [25]. Fig. 7 shows the above. This suggests that there is more diversity in window
distribution of the 5273 interior angles at which the foam sizes than in strut lengths within the sample. The differ-
struts meet within the sample. The angles appear to fall in ence between the strut length and window area distributions
a nearly normal distribution. The mean strut intersection an- can be explained by two effects. First, window areas are
gle of 106.7 is close to but significantly smaller than the roughly proportional with the square of strut lengths, broad-
expected tetrahedral angle of 109.5 at which four edges ening the distribution. Second, shape may play a significant
meeting at a corner should intersect in an equilibrium struc- factor in window size. When struts of like size are used to
ture, as noted by Plateau [7]. All of the other samples in create both quadrilateral and hexagonal windows, two dif-
Tables 1 and 2 also had average intersection angles signifi- ferent windows sizes may be created, resulting in a broader
cantly smaller than 109.5 [28]. There are possible reasons distribution. Finally, anisotropy in the rise direction will also
for this variation from the theoretical rule. First, as noted widen the window size distribution.
in the introduction, foamed polymers are not equilibrium
structures. Structural foams are frozen before being able to 3.5. Window shape
reach complete equilibrium. Second, there are stresses on
the shape of real foam systems. This sample was a slabstock Window shape has previously been used to describe foam
polyurethane foam, and, as such, has cells that were elon- cells and how they divide space [6,23,33,34]. Kelvin [6] pro-
gated in the rise direction during the rising process. Surface posed the tetrakaidecahedron (a 14-sided shape consisting
tension is not the main driving force at work in determin- of six quadrilaterals and eight hexagons) as a shape which
ing cell shape for polymer foams. In addition, we observed would minimize surface area within a monodisperse foam
that in real foams, more than four edges may meet at a ver- and should therefore be favored at equilibrium. Weire and
tex. A few occurrences of this phenomenon within a large Phelan [34] suggested that a tessellation of equally sized
sample may significantly depress the mean intersection an- cells with pentagonal and hexagonal sides could more effi-
gle. ciently subdivide space. However, Matzke [33] showed that
208 M.D. Montminy et al. / Journal of Colloid and Interface Science 280 (2004) 202211

Table 4
Comparison of window shapes for the foam Sample 1 with window shapes expected by other models and experiments
# cells Window shape distribution (%)
analyzed Triangles Quadrilaterals Pentagons Hexagons Heptagons +
Kelvin model [6] 0 43 0 57 0
WeirePhelan model [34] 0 0 89 11 0
Matzke experiment [33] 600 0 11 67 22 <1
Kose experiment [23] 8 0 9 70 21 0
Sample 1a 106 1 24 55 19 1
Note that only central bubbles from Matzkes experiments (those in the center of his jars, unaffected by edge effects) are listed here.

in reality, even monodisperse liquid foams consist of cells


and windows of many varied shapes.
Nine triangles, 251 quadrilaterals, 587 pentagons, 205
hexagons, and 11 heptagons were detected in the sample.
Most of the windows in our sample have the expected
quadrilateral, pentagonal, and hexagonal shapes, with the
pentagonal window being the most common shape.
Table 4 provides a comparison of the distribution of win-
dow shapes found in this study to those found in area mini-
mizing models and other foam characterization experiments.
Note that the foam studied here has a much more diverse
window shape distribution than those hypothesized in area-
minimizing models. This is largely due to our samples poly-
dispersity. In addition, notice that the cell shape distributions
of the Kose [23] and Matzke [33] experimental results, both Fig. 9. Cell volume distribution for the foam sample.
performed on liquid foams, are very similar to each other,
but quite different than the window shapes within the poly-
mer foam examined in this study. These results highlight the Table 5
Distribution of cell shapes for the foam sample
facts that polymeric foams do not follow ideal theoretical
structures. The structures found in this foam are more con- # sides # occurrences Frequency (%)
sistent with recent studies of random soap froths found in 7 3 3
8 4 4
the important work of Kraynik [35,36]. Modeling and study
9 4 4
of foams with these less ordered, more realistic polydisperse 10 7 7
structures may allow a better understanding of the correla- 11 19 18
tion between foam physical properties and microstructure. 12 12 11
13 13 12
14 14 13
3.6. Cell size and shape 15 10 9
16 4 4
17 9 8
The polydispersity of cell sizes within this foam is most 18 5 5
apparent when the foam cell volume distribution is exam-
21 1 1
ined. This distribution, presented in Fig. 9, is skewed very
heavily to the right, indicating a high degree of polydisper- 24 1 1
sity of cell sizes. In fact, of the 106 cells detected, the 10 Total: 106 cells
largest cells make up 25% of the sample volume. Statistics Average # sides: 13.01
like this one may give a good way to quantify polydisper-
sity from an engineering perspective, since a small number
of large gaps in the foam structure (cells that are too large) Along with polydispersity of cell sizes comes diversity
can drastically impact a foams performance. in cell shapes. When very large and fairly small cells share
Cell volumes are calculated using a method analogous to borders in a sample, the large cells can end up with very large
the window area calculation method described above. Each numbers of windows, as shown in Table 5. In this sample,
cell is split into a set of pyramids with the top of each pyra- most cells detected had 1115 sides, but cells composed of
mid at the centroid of the cell, and the base of each pyramid 18, 21, and 24 windows were also detected.
being a window. The volume of all of these pyramids for The average cell detected in the sample was 13-sided,
each cell are added together to calculate the cell volume. which is somewhere between a rhombic dodecahedron and
M.D. Montminy et al. / Journal of Colloid and Interface Science 280 (2004) 202211 209

Table 6
Comparison between the highlighted sample results (Sample 1a) and those of an adjacent sample of the same foam (Sample 1b)
Sample Cells Mean strut Mean interior Mean window Mean cell Mean cell surface Mean # windows
# detected length (mm) angle (deg) area (mm2 ) volume (mm3 ) area (mm2 ) per cell
1a 106 0.281 106.7 0.133 0.193 1.749 13.01
1b 60 0.281 106.1 0.134 0.209 1.84 13.37

Difference (%) 0.0 0.6 0.8 8.3 5.2 2.8

Table 7
Comparison between two adjacent samples of another piece of polyurethane foam (Samples 2a and 2b)
Sample Cells Mean strut Mean interior Mean window Mean cell Mean cell surface Mean # windows
# detected length (mm) angle (deg) area (mm2 ) volume (mm3 ) area (mm2 ) per cell
2a 54 0.319 107.1 0.178 0.308 2.371 12.98
2b 58 0.330 107.5 0.192 0.319 2.318 12.40

Difference (%) 3.4 0.4 7.9 3.6 2.2 4.5

Kelvins tetrakaidecahedron, two theoretically proposed and up, as this sample was, it provides a relatively good measure
popular approximations of foam cell shapes. This suggests of anisotropy within the system, with more elongated foam
that these shapes may provide reasonable approximations of systems exhibiting higher anisotropy ratios.
actual foam shapes, but the diversity of cell shapes within
this sample show that these truly are simplified models of the 3.8. Reproducibility
actual structure of real polymer foams. Polydisperse foams
naturally have more diverse cell shapes. Comparison of Sample 1as statistical results to those of
an adjacent sample from the same foam (Sample 1b) show
3.7. Other quantitative measurements good agreement, as shown in Table 6. Results for Samples
2a and 2b are also provided in Table 7. Notice that in both
Other quantitative measurements can be made using the cases, the statistical results for the two samples agree within
model information collected from the 3D foam image, al- less than 10%, and that the largest errors occur for cell-based
though these measurements tend to be less accurate than measurements, where the number of samples averaged is
those listed above due to image limitations. Solid fraction small (60 and 106 cells vs thousands of struts, angles, and
and average strut thickness can be calculated by counting windows detected).
the dark pixels from the segmented 3D image. However, as Ensuring reproducibility of image processing results de-
shown in Fig. 1, the original images are somewhat blurry, pends upon careful inspection of the model and comparison
and it is difficult to decide exactly where the surface of the of the model to the original 3D image by the person using the
strut begins. This is inconsequential when determining strut image analysis software. Because of the time and expense
lengths because of the connectivity of the foam structure, involved in obtaining and creating 3D images of foams, and
but causes trouble when strut thickness is examined. Un- performing multiple experiments to ensure reproducibility,
fortunately, this issue cannot be resolved by using higher this technique is more difficult and expensive than for other
scanning resolution because bigger, more detailed pictures experimental techniques. In this study, each sample was sent
would slow down the main image processing method dis- to an external laboratory for imaging, and imaging of each
cussed. sample cost about $500.
When a slabstock foam rises quickly, its cells are of-
ten stretched in the direction in which the foam rises. This 3.9. Method and model limitations
anisotropy or elongation is important to the final physical
properties of the foam, since the structure is strengthened While this method offers a new way to create models of
in the direction of elongation, much like the structure of an foams, it has some limitations. First, while the image analy-
egg. The anisotropy ratio displayed in Table 2 gives a rough sis process may be faster than traditional microscopy and
measure of elongation within the system. It is determined examination of foams, it is still relatively time intensive. Al-
by calculating the sum of absolute values of the dot prod- though the process is semi-automated, analysis of the six
ucts of the vectors describing each strut with the x, y, and samples examined over the course of this work required
z unit vectors, then dividing the largest of these values by 24 h each. Most of this time was spent on user-assisted
the smallest. That is, the x, y, and z component of each strut analysis of the 3D images. While this time frame is rea-
within the model is calculated, and the sum of each direc- sonable for researchers, it prevents the use of this method
tional component for all struts is calculated. If the sample is for real time quality analysis and control of manufactured
placed into the imaging device with the rise direction facing foams. In addition, processing difficulty scales with the cube
210 M.D. Montminy et al. / Journal of Colloid and Interface Science 280 (2004) 202211

of the image size, so large images can quickly become too els of real polymeric foams could be used to (1) model foam
slow and difficult to process. At the time of this research, the performance at many levels, including modeling of foam
293 293 300 voxel size of Sample 1a stretched the capa- compression and fluid flow through foams, and (2) analyze
bilities of the image processing hardware used, 733 MHz PC many elements of foam structure at once to develop ro-
with an nVidia GeForce 2 graphics card with its own graph- bust structureproperty, structurechemistry, and structure
ics processor. processing relationships.
In addition, 3D imaging equipment has limited resolu- Reliable computerized models of real foams have many
tion. Current micro X-ray CT has a resolution of about potential uses. There is a significant industrial demand for
5 m [30], while recent MRI experiments [26,27] obtained improved characterization techniques for quality control and
resolutions of 8 m. Due to this constraint, for reasonably ac- new product development in the polymeric foams industry.
curate measurements to be obtained, struts must be at least This method allows for the study of foams at a microstruc-
100 m long and about 20 m thick. This precludes the ture level, and may perhaps allow the detection of subtle
analysis of microcellular polymers, which have cell diam- structural changes due to changes in foam formulations, such
eters of 10 m or less [37]. as the addition of catalysts or surfactants. Study of the effects
Because of the limited resolution of 3D imaging equip- of the microstructural properties of foams on their macro-
ment, some foam features cannot be accurately detected or scopic physical properties may assist in the development
measured. As described above, strut thickness is difficult to of new foams for various applications. Academic studies
judge because of the weak CT signal generated by foam of foam compression, fluid flows through foams, and cell
struts and the small length to thickness ratio of the struts. nucleation would directly benefit from 3D models of real
Struts are often only 5 voxels thick in a 3D foam image, foams. Theories developed using ideal foam models could
making strut width measurement quite inaccurate. Similar be tested on real foams with irregular shapes and in nonideal
problems would plague attempts at quantifying strut surface conditions.
area or foam surface roughness. Relatively thick vertices However, 3D characterization tools must be developed
are common in polydisperse foams, and some judgment is further if they are to be used for many industrial purposes.
required in locating the center of these vertices. This in- Developing robust structureproperty relationships will re-
troduces some imprecision into the measurement of strut quire the analysis of hundreds of samples, which would still
intersection angles, as small movements to a vertex can sig- be a daunting task using this method, since sample analy-
nificantly impact these angles. sis requires about 24 h for image analysis, plus instrument
Thin windows are also difficult to detect and measure time for image acquisition. Quality control applications also
at this resolution, preventing analysis of window thickness require a quick turnover time to ensure that a minimum
and percentage of open-celled content. Since the thickness amount of product is wasted, making this method useful as
or existence of windows can drastically affect the physi- a quality control check for small batch processes but infea-
cal properties of a foam [3,5], this method would have to sible for large-scale continuous processes. Future research
be combined with window measurement techniques such as will hopefully yield faster image acquisition and faster, more
traditional microscopy or air flow measurements to create a automated image analysis algorithms, providing for quicker
complete picture of closed-celled foam structure. analysis of samples to serve these industrial uses.

4. Conclusions Acknowledgments

The software and foam structural data presented in this The authors thank Dr. Mark Listemann, Air Products and
paper show that in-depth 3D characterization and modeling Chemicals, and Dr. Xiao Dong Zhang, Dow Chemical, for
of real structural foams is possible and can yield interesting providing foam samples and input for this paper. We thank
results. For the polydisperse polyurethane foam discussed Andy Kraynik for his very helpful comments. This paper
in this paper, feature size distributions provided significant is based upon work supported by a National Science Foun-
insight into the samples structure. dation Graduate Research Fellowship. This work was also
This research shows that real polymeric foams may have supported in part by grants from the National Science Foun-
not only polydisperse cell sizes, but rich, diverse arrays of dation, the Air Force Office of Research, the Army Research
cell and window shapes. Since polymeric foams are not equi- Office, and MURI grant.
librium structures, they exhibit disordered structures which
are quite different from the perfectly ordered, monodisperse
tetrakaidecahedral or cubic unit cells previously used to Supplementary material
model the physical properties of foams.
The development of models of real polymeric foams is The online version of this article contains additional sup-
an enabling technology which will facilitate future research plementary material.
into the structure and physical performance of foams. Mod- Please visit DOI: 10.1016/j.jcis.2004.07.032.
M.D. Montminy et al. / Journal of Colloid and Interface Science 280 (2004) 202211 211

References [19] J. Cenens, R. Huis, B. Chauvaux, J.M. Dereppe, C. Gratin, F. Mayer,


in: V. Kumar, K.A. Seeler (Eds.), Cellular and Microcellular Materials,
[1] A. Cunningham, N.C. Hilyard, in: A. Cunningham, N.C. Hilyard vol. 53, ASME, New York, 1994, p. 29.
(Eds.), Low Density Cellular Plastics: Physical Basis of Behaviour, [20] G. Woolf, Chem. Eng. London 558 (1994) 10.
Chapman & Hall, London, 1994, p. 1. [21] C. Monnereau, M. Vignes-Adler, J. Colloid Interface Sci. 202 (1998)
[2] R.E. Jones, G. Fesman, J. Cell. Plast. 1 (1) (1965) 200. 45.
[22] R. Garcia-Gonzales, C. Monnereau, J.-F. Thovert, P.M. Adler, M.
[3] H.F. Mark, Encyclopedia of Polymer Science and Engineering, Wiley,
Vignes-Adler, Colloids Surf. 151 (1999) 497.
New York, 1985.
[23] K. Kose, J. Magn. Res. Ser. A 118 (1996) 195.
[4] T.G. Nieh, K. Higashi, J. Wadsworth, Mater. Sci. Eng. A 283 (2000)
[24] K. Kashiwagi, K. Kose, NMR Biomed. 10 (1997) 13.
105.
[25] B.J. Pangrle, X.D. Zhang, C.W. Macosko, B.E. Hammer, N.P. Bidault,
[5] F.A. Shutov, in: D. Klempner, K.C. Frisch (Eds.), Polymeric Foams,
M.L. Listemann, R.E. Stevens, in: Polyurethanes Expo 98 Conference
Hanser, Munich, 1991, p. 17.
Proceedings, September 1720, 1998, p. 230.
[6] W. Thomson, Philos. Mag. Ser. 5 24 (151) (1887) 503514.
[26] M. Szayna, R. Voelkel, Solid State Nucl. Magn. 15 (1999) 99.
[7] J. Plateau, Statique Experimentale et Theorique des Liquids Soumis
[27] M. Szayna, L. Zedler, R. Voelkel, Angew. Chem. Int. Ed. 38 (17)
aux Seules Forces Moleculaires, Gauthier-Villars, Paris, 1873.
(1999) 2551.
[8] A.N. Gent, A.G. Thomas, J. Appl. Polym. Sci. 1 (1) (1959) 107. [28] M.D. Montminy, Ph.D. thesis, University of MinnesotaTwin Cities,
[9] W.L. Ko, J. Cell. Plast. 1 (1965) 45. Minneapolis, 2001.
[10] W.E. Warren, A.M. Kraynik, J. Appl. Mech. 55 (1998) 341346. [29] P. Anderson, G.R. Davis, J.C. Elliott, Eur. Microsc. Anal. (March
[11] S.H. Goods, C.L. Neuschwanger, C.C. Henderson, D.M. Skala, J. 1994) 3537.
Appl. Polym. Sci. 68 (1997) 1045. [30] Micro Photonics website, http://www.microphotonics.com/skymto.
[12] M.B. Rhodes, J. Elast. Plast. 12 (1980) 201. html, accessed 7/7/2003.
[13] M. Richardson, D. Nandra, Cell. Polym. 5 (6) (1986) 423. [31] M.D. Montminy, A.R. Tannenbaum, C.W. Macosko, J. Cell. Plast. 37
[14] R. Hamza, X.D. Zhang, C.W. Macosko, R. Stevens, M. Listemann, (2001) 501515.
in: K.C. Khemani (Ed.), Polymeric Foams: Science and Technology, [32] K. Palagyi, A. Kuba, Graph. Model. Image Proc. 61 (1999) 199.
ACS, Washington, DC, 1997, p. 165. [33] E.B. Matzke, Am. J. Bot. 33 (1946) 58.
[15] K.M. Lewis, I. Kijak, K.B. Reuter, J.B. Szabat, J. Cell. Plast. 32 (1996) [34] D. Weaire, R. Phelan, Philos. Mag. Lett. 69 (2) (1994) 107.
235. [35] A.M. Kraynik, D.A. Reinalt, F. van Swol, Phys. Rev. E 67 (31) (2003)
[16] P. Chaffanjon, G. Verhelst, Cell. Polym. 11 (1) (1992) 1. 031403.
[17] M.B. Rhodes, B. Khaykin, Langmuir 2 (1986) 643. [36] A.M. Kraynik, MRS Bull. 28 (4) (2003) 275278.
[18] M.B. Rhodes, B. Khaykin, J. Therm. Ins. 14 (1991) 175. [37] V. Kumar, Cell. Polym. 12 (3) (1993) 207.

Вам также может понравиться