AUTOMATYKA 2006 Tom 10 Zeszyt 3

* Department oI Computer Engineering, Technical University oI Lodz

117
Laurent Babout*
X-Ray Tomography Imaging:
a Necessary Tool for Materials Science
1. Introduction
For many people, X-ray tomography, also known as CT (Computed Tomography) or
CAT-scan (Computed Axial Tomography), is devoted for medical applications to visualize
three-dimensional images of human body. The technique has been firstly commercialized in
the 70s, and different types of generation have been developed (mainly linked with diffe-
rent acquisition procedures) in order to mainly improve the acquisition speed. However,
CT scans are less popular than Magnetic Resonance Imaging (MRI) which normally provi-
des a far better contrast resolution with similar spatial resolution, and above all, which is
harmless to patient (X-ray radiation dose may increase the chance of malignancy, especially
in a fetus). Nowadays, the number of applications of X-ray tomography in other field of
research than in medicine is constantly increasing, such as in paleontology, archeology or in
materials science. In the past 10 years, the technique has known a real breakthrough, with
the development of worldwide synchrotron radiation tomography apparatus which can pro-
vide 3D images with a spatial resolution 100200 times higher than conventional medical
CT systems. The commercialization of industrial and laboratory based system with micro-
scale resolution has also steeply increased and X-ray tomography system is becoming an
inevitable imaging tool to be own by research institutes, in the same respect as well known
secondary and transmission electron microscopes.
Nowadays, X-ray tomography can really compete with classical optical and electronic
microscopy techniques because:
the technique is non invasive and there is no need Ior sample preparation like poli-
shing,
very brittle sample can be easily handled,
in situ mechanical test can be monitored on tomography stage,
the spatial resolution is equivalent to optical microscope (up to 20 magniIication),
and, oI course, three dimensional images are reconstructed.
118 Laurent Babout
This paper presents two examples with industrial implications to attest the state-of-the-
art of the technique from two different types of apparatus mentioned above: synchrotron X-
ray microtomography and laboratory X-ray CT. The respective examples deal with nuclear
graphite and titanium alloy for aeroengine industry.
2. X-ray tomography: basic principle and set-ups
The general concept of X-ray tomography is an extension of classical X-ray radiogra-
phy, and is based on the attenuation of the X-ray beam through the specimen. X-ray radio-
graphy provides only a projection of the sample volume on one single plane. X-ray tomo-
graphy overcomes this disadvantage by combining the information from a series of many
radiographs, each being recorded with different orientation of the sample in front of the
detector. The variation of X-ray attenuation in the volume of the sample can be reconstruc-
ted by combining a sufficient number of radiographs with an appropriate algorithm, such as
filtered back-projection (Feldkamp et al. 1984). The data is obtained in the form of a three-
dimensional array of voxels (analogous to pixels in a two-dimensional digital image), for
which the grey-level of each voxel describes the calculated X-ray attenuation at that posi-
tion. Dedicated imaging software have been developed like Amira or VGStudiomax, with
some of them in shareware such as ImageJ. The image processing toolbox in Matlab also
offers some possibilities to deal with 3D images.
A tomographic scan requires an X-ray source, a rotation stage and an X-ray detector.
This is usually a fluorescent detector which changes the X-rays into visible light, and a set
of optic lenses that transfers the light to a CCD camera. Cone and parallel X-ray sources are
commonly used. The term microtomography is sometimes used to refer to tomography with
an image spatial resolution in the micron range, which can be achieved if the spot size is of
the order of few microns, or via an adequate set of optics in the detector. The main draw-
back of the technique lies in the sample size limitation, which need to stand in the field of
view of the camera. For instance, if a 1024
2
CCD chip is used with a 2 m pixel size, the
maximum sample size in the transversal cross section is ~2 mm.
The two examples presented in this article have been obtained separately on laboratory
tomograph at the Materials Science Centre, University of Manchester (UK) and at the Euro-
pean Synchrotron Radiation Facility (ESRF, Grenoble, France), on the ID19 beam line. The
first tomograph device (setup1 hereafter), shown in Figure 1 uses an X-ray fan beam with
a maximum energy of 225 keV, a focal spot size of around 9 m and a spot angle of 80. The
system is composed with two different detectors equipped with beryllium and aluminum
foils adequate for low and high material density respectively. The main features of the se-
cond tomograph (setup 2 hereafter) are a spectrally and spatially homogeneous, highly co-
herent, high photon flux beam at the sample position, and a tunable photon energy in the
range of 6100 keV, with most of the experiments done in the range 1035 keV(Boller
2005). A state-of-the-art detector, developed at the ESRF, allows radiographs to be acquired
with a tunable spatial resolution in the range of 0.2940 m with a fast readout (20 Mpixel/s)
and very low readout noise (25 e

).
X-Ray Tomography Imaging: a Necessary Tool Ior Materials Science 119
Fig. 1. X-ray tomograph 'system1: a) view oI X-ray Ian beam source and rotation stage;
b) view oI two diIIerent detectors. 1 X-ray source, 2 rotation stage, 3 beryllium Ioil detector,
4 aluminum Ioil detector
Table 1 presents the different set-ups applied to the two examples presented in this
paper. One can see the relative fast scan duration of both systems for the very high spatial
resolution and number of radiographs used.
Table 1
Tomography parameters Ior setup 1 and 2. E energy, R resolution, t exposure time,
A rotation step, Nr number oI radiographs, NI number oI Irame/radiograph, T scan duration
3. Experimental examples
3.1. Example 1: nuclear graphite
Graphite is a major component in gas-cooled reactor because it is a moderator for fast
neutrons, has good thermal and mechanical properties. However, during servicing, oxida-
tion at high temperature (i.e. 600900C) are susceptible to generate microstructural chan-
ges in graphite, such as porosity development, which can lead to the damage, then potential
failure of core component, which can have a major effect on the reactor functioning. There
is then an important need in predicting the behavior of graphite, and the accuracy of the
prediction is necessary linked with the understanding of the microstructure behavior. X-ray
E(keV) R(m) t(s) A() Nr Nf T(min)
Example1 Setup1 30 13 0.16 0.3 600 32 ~50
Example2 Setup2 31 0.7 1.5 0.12 1500 1 ~50
120 Laurent Babout
tomography is then extremely adapted for the purpose of such study because the porosity
network can be easily separated from the graphite during image segmentation and the vo-
xels value in the 3D volume can be directly linked to local density (Babout et al. 2005;
Babout et al. 2006). Example of such possibility is shown on Figure 2 which presents the
microstructure before and after segmentation of a graphite sample before oxidation. Setup 1
was used for the tomography scans. Figure 2a clearly shows the typical porous microstruc-
ture of graphite present in the matrix phase (M) or within inclusions (I). The segmentation
with a simple thresholding procedure is satisfying to select most of the porosity within the
material. Only the smallest pores exhibiting less contrast with the graphite are not segmen-
ted (like MC type).
Fig. 2. X-ray tomography oI graphite microstructure: a) grey level; b) aIter pore segmentation.
M matrix, I inclusion, CC,GEP and MC are diIIerent types oI porosity (Babout 2005)
Because graphite is only composed with one chemical element, namely carbon, it is
possible to correlate the local attenuation in the original 3D volume with the real local den-
sity after step wedge calibration (Babout 2006). This is then very useful to quantify the
variation of density within the material, especially after graphite has been subjected to ther-
mal oxidation. An example of such quantification is shown in Figure 3. This quantification
has been performed using a 3D-density mapping function written in Matlab. The function
considers a cubical region containing an odd number of voxels on each dimension centered
in turn on each voxel. The grey level value of the voxel is replaced by the average value of
the including box (averaging volume). The boundary problem associated with the voxels
that lie nearer to the edge of the volume than the half-width of the box, is solved by assu-
ming that the volume edges are mirrors, so that each plane containing these voxels is dupli-
cated beyond it. This kernel operation of neighborhood averaging, usually used in image
treatment for noise reduction (Russ 1995), allows the block to be smoothed and reveals
more clearly the average density fluctuation throughout the specimen. Naturally, the larger
the averaging volume (box), the more homogeneous the density map. In the present case,
a box size of 286 m (i.e. 22 voxels) was found suitable to quantify density fluctuations
(Babout 2006). An essentially uniform density across the virgin graphite sample is observed
except around the large pores (located at the top of Fig. 3a). Apart from these large pores,
most of the network of porosity has disappeared from the volume under the averaging ope-

X-Ray Tomography Imaging: a Necessary Tool Ior Materials Science 121
ration, as their average equivalent diameter, i.e. 170 m, is lower than the box size. Figure
3b shows the density map with the same kernel size for graphite sample thermally oxidized
at 900C. In contrast to the virgin state, a local microscale variation of density occurs in
the thermally oxidized sample. The variation is due to the differing propensity for thermal
oxidation of the matrix and the inclusions, which have average densities close to 1.0 and
1.6 g/cm
3
respectively. Figure 3b, compared with Figure 2a, reveals that the thermal oxida-
tion process occurs preferentially in the matrix. This is mainly due to open porosity which
works as a channel for carbon diffusion during thermal oxidation at 900C. X-ray tomogra-
phy imaging is the only technique which can perform the quantification of real density va-
riation non-destructively and in three dimensions.
Fig. 3. Density maps oI: a) virgin graphite; b) thermally oxidized graphite (weight loss oI ~23).
Kernel box size: 286 m
3.2. Example 2: titanium alloy
Titanium alloys find many applications in the aeroengine industry for their good tensi-
le and fatigue strength properties at high temperature. They are often fabricated with a la-
mellar microstructure which comprises complicated 3D lamellar colonies ( corresponds
to hexagonal close-packed (hcp) crystal structure) embedded in large grains ( corre-
sponds to body centered cubic (bcc) crystal structure). Further details about this type of
microstructure can be found in (Lutjering 1998). Observations of the interaction between
the 3D fatigue crack and the (+) microstructure is one of the crucial steps to improve the
prediction of the material lifetime using microstructure models. This type of crack tracking
can be performed in situ by mounting a dedicated fatigue machine on the rotation stage of
the tomography system in setup2.
The 3D images shown in Figure 4 present the comparison between the initial stage and
one of two fatigued stages of a notched (+) titanium alloy sample. One can see that the
lamellar microstructure can be satisfyingly observed, with a primary grain boundary (detail
2 in Fig. 4a) separating two grains clearly visualized. Some lamellar colonies situated
within grain are also clearly distinguishable. A closer look shows that white fringes are
situated between lamellae and along grain boundary. Figure 4b also highlights the presen-
ce of cracks (cracks 1 and 2) initiated at a triangular-shape notch. A closer look of the
cracks shows that they are surrounded by sharp white fringes. These fringes also complete
the crack path in some places where no black line is visible. This type of image enhance-
g
.
c
m
-
3
g
.
c
m
-
3
122 Laurent Babout
ment which reveals the microstructure and some sub-resolution parts of the crack path using
X-ray tomography is called edge or phase contrast (Cloetens et al. 1997). A necessary con-
dition to generate such contrast is the spatial coherency of X-ray beam (see section 2). It is
generated by the difference in refractive index between and phases or titanium and air,
and it is amplified simply by increasing the sample detector distance (100 mm to visualize
the microstructure in titanium).
Fig. 4. 3D reconstructed images in three principal directions oI titanium alloy sample: a) virgin state;
b) aIter 29000 cycles oI Iatigue test. 1 notch at sample surIace, 2 grain boundary,
1, 2 two diIIerent lamellar colonies, 3, 4 two diIIerent cracks,
A point where crack initiates, B point where crack deIlects
The phase contrast is particularly interesting to be generated when no attenuation con-
trast between two different materials or phases exists, such as between aluminum and sili-
con (Buffiere et al. 1999), or in the present case between and titanium phases. It is also
a very powerful mean to detect unresolved cracks (i.e. cracks which present a width smaller
than the spatial resolution). Moreover, the difference in refractive index between the diffe-
rent phases in titanium alloy and air gives rise to different intensity of the phase contrast:
One can verify from Figure 4b that the supposed lower difference in refractive index be-
tween and phases results in a weaker phase contrast than the one between cracks and
titanium (n~1 for air and n~2.2 for titanium).
The comparison between Figure 4a and Figure 4b reveals very useful information. One
can see in Fig. 4b that one crack (detail 3) has propagated fairly linearly within a colony
(detail 1) between points A and B, then deviated at a boundary with a neighboring colony
(detail 2). The computation of an oblique slice in the crack plane also reveals more intere-
sting features of the crack development in interaction with the microstructure: sinusoidal
shape of the crack front (between points B and C in Fig. 5a) and apparently crack branches
propagating between lamellae (points 3a to 3d in Fig. 5a). The 3D rendering of this crack is
displayed on Figure 5b with isosurfaces of the notch and grain boundary. The figure also
shows the isosurface of the intense phase contrast fringes generated by the unresolved
a) b)
X-Ray Tomography Imaging: a Necessary Tool Ior Materials Science 123
cracks (detail 5). The different features have been segmented using the region growth pro-
cedure. This alternative segmentation method to classical thresholding, constructs regions
by starting from some user-provided voxels called seeds. Each seed grows and produces
different 3D region. The progression is controlled by selecting a grey level interval surroun-
ding the intensity value of the seed. More details about the algorithm can be found in (Niko-
laidis et al. 2001). It is worth noticing that the difficult segmentation of the grain bounda-
ry has bee facilitated once again by the white fringes which contain the region growing
within the boundary. Figure 5b confirms the observations seen in the other figures. It also
highlights that the crack 4 hardly propagates through the grain boundary.
The images presented in this example show for the first time evidence of interaction
between crack and microstructure in 3D. Indeed, the visualization of the deflection of the
crack crossing over the boundary between two lamellar colonies is hardly achievable with
standard 2D microscopic means. The detection of the branches is even more complex be-
cause the probability to visualize such event at the polished surface of a sample is small.
Fig. 5. Oblique slice- 29K cycles(a); 3D rendering oI crack (b). 1 notch, 2 grain boundary
(dark grey), 3, 4 two diIIerent cracks (grey), 3a, 3d possible branches oI crack (3),
5 phase contrast corresponding to unresolved crack (light grey)
4. Conclusion
The present paper has shown, thanks to two different examples on effect of thermal
oxidation of nuclear graphite and fatigue testing of titanium alloy, the wide capability of
X-ray tomography imaging. The non-destructiveness and three dimensional images have
proven to provide unique information that are not accessible using conventional 2D micro-
scopy techniques. The maximum spatial resolution available at the present time is 0.3m.
However, researchers are currently working on improving systems to reach a spatial resolu-
tion up to 50 nm without sample size limitation, which will guarantee a serious alternative
to electronic microscopy and give rise to new research avenues at the nanoscale.
a) b)
124 Laurent Babout
References
|1| Babout L., Marrow T.J., Mummery P.M., Withers P.J.: Mapping the evolution of densitv in 3D of
thermallv oxidised graphite for nuclear applications. Scripta Mater., 54, (5), 2006, 829834
|2| Babout L., Mummery P.M., Marrow T.J., Tzelepi A., Withers P.J.: The effect of thermal oxidation
on polvcrvstalline graphite studied bv X-rav tomographv. Carbon, 43, 2005, 765774
|3| Boller E.: ID19 beamline webpage, ESRF. URL. http://www.esrI.Ir/UsersAndScience/Experi-
ments/Imaging/ID19/. Last Update: 20/07/2005
|4| BuIIiere J.-Y., Maire E., Cloetens P., Lormand G., Eougeres R.: Characteri:ation of internal da-
mage in a MMCp using X-rav svnchrotron phase contrast microtomographv. Acta Materialia,
47, (5), 1999, 16131625
|5| Cloetens P., Pateyron-Salome M., BuIIiere J.Y., Peix G., Baruchel J., Peyrin E., Schlenker M.:
Observation of microstructure and damage in materials bv phase sensitive radiographv and to-
mographv. Journal oI Applied Physics, 81, 1997, 58785886
|6| Eeldkamp L.A., Davis L.C., Kress J.W.: Practical cone-beam algorithm. Journal oI Optical Socie-
ty oI America A, 1, (6), 1984, 612619
|7| Lutjering G.: Influence of processing on microstructure and mechanical properties of () tita-
nium allovs. Materials Science and Engineering A, 243, (12), 1998, 3245
|8| Nikolaidis N., Pita I.: 3-D Image Processing Algorithms. Wiley-Interscience, New York, 2001
|9| Russ G.C.: The imaging processing handbook. CRC Press, Boca Raton, 1995