Professional Documents
Culture Documents
Volumetric images of small mortar cylinders under load The data was collected and preprocessed by the Univer-
are acquired by X-ray microtomography. The images are sity of Maine team [6] at Beamline X2B in Brookhaven Na-
binarized at different threshold values, and the connected tional Laboratory using X-ray microtomography [10, 4, 3].
components are extracted at each threshold with a 3D XMT is used to measure the internal X-ray absorption of
space- and time-efficient program. Statistics of volume and a 3D object. It is similar to a medical CAT scan, except
surface area of separated concrete segments and cracks are for the much higher beam intensity and detector resolution.
collected and analyzed. Under increasing load, the con- The high energy beam is generated using synchrotron radi-
crete specimen can undergo size distortion. The largest seg- ation rather than the X-ray tubes used in the conventional
ments and cracks account for 99% of the change. Some seg- CAT scan.
ments and cracks merge. The concrete specimens are small motar cylinders about
4 millimeters in diameter and 4 millimeters in height. The
specimen is a mixture of Type I portland cement, sand and
water. It consists of solids (sand, cement and aggregate) and
1. Introduction voids (cracks and air holes).
The concrete specimen is held on a rotational stage in
a specially designed apparatus (Figure 1(a)). An uniaxial
Concrete is widely used. Understanding why and how compressive load is then applied to the specimen by piezo-
concrete fractures has been of great interest to physicists, electric crystals. The stress is continuously monitored by a
civil engineers and material scientists. Even though the conventional load cell.
application of fracture mechanics to concrete has pro- The X-ray beam is monochromated by a silicon crystal
gressed [1], much remains unknown. While continuum to select the energy of 32 keV. For optimum contrast, this
approaches based on plasticity and linear elastic fracture energy is chosen such that 90% is absorbed by the speci-
mechanics have led to considerable success in predicting men. As in a CAT scan, the rotational stage is designed to
failure in fine-grained materials such as metals, non-linear allow load application with minimum X-ray absorption.
effects have resisted analysis in heterogeneous and quasi- The X2Bs detector has a single-crystal phosphor with
brittle materials such as concrete. The study of microstruc- high spatial resolution. It converts the X-ray to visible light
ture coupled with traditional stress-strain measurements of- (Figure 1(b)). The light is detected by a 10241024 pixel
fers the most promising approach. CCD camera to generate an absorption map. The specimens
In [7], we analyzed thin, warped, interconnected cracks were exposed to the beam at 720 different angles over 180
in sequences of density images of samples of mortar. Now and maps of absorptivity were generated at each angle.
we report the application of image processing techniques to After applying Exxon Direct Fourier Reconstruction to
locate and measure the changes in the separated segments the absorption maps, the result is quantized to generate the
due to cracks, and the overall change properties of concrete final data set, which is a grayscale 3D volume with one byte
specimen under fracture. per voxel.
The resulting measurements will ultimately be used in The same process is applied to the concrete specimen
multiscale modeling of the microstructure for improved un- with 5 or 6 load-and-release cycles, with progressively
derstanding of the macroscopic mechanical properties of greater loads. The last volume is supposed to capture the
concrete fracture. state of the specimen after the peak of the stress-strain
1
curve, before it crumbles with any further load. A new data 3. Data Analysis
set is generated at each stage.
Two volumetric data sets from successive loads, Because of the large volume of data ( vox-
and , were analyzed. The volume of each data set is els), binarization followed by connected component (CC)
800800765 at voxel. Figure 1 (c,d) shows two analysis [8, 9, 2] is the most practical means of investigat-
slice images from these two volumes at corresponding hori- ing the foreground material (mortar, sand, aggregate) and
zontal cross sections, where the specimen was separated background void (cracks and air holes) constituents of the
into segments due to the cracks. specimen.
The grayscale of each voxel corresponds to the X-ray We used a space- and time-efficient CC algorithm with
absorption at that particular point in the specimen. The Find-Union on 1-D runs and path compression [5]. It finds,
brighter the voxel is, the higher its density. At this reso- in each CC, the constituent voxels, number of runs, volume
lution, some concrete microstructures can be distinguished: and surface area (number of free faces). In order to pre-
the dark low density regions are voids (cracks and air holes); serve topological integrity, we used 6-connectivity (face)
the bright high density regions are sand; the gray regions are for foreground (segment), and 26-connectivity (vertex) for
cement and aggregates. background (crack).
The complex photometric, geometric and topological
properties of each specimen are deduced by observing the
cardinality, volume, surface area and surface area to volume
ratio of foreground and background CCs obtained over the
entire range of binarization thresholds. The volume change
of each segment is an indication of distortion under stress.
Since the external force is expected to equal the work re-
quired to stretch the internal surfaces, the surface area in-
crease in segments is also an important parameter. Further-
more, the ratio , a shape invariant, is a useful mea-
(a) sure of compactness and rotundity.
The remaining sections present the observations in de-
tail.
3.1. Density
2
8 6
x 10 x 10
2.5 6
T1 Crack T1 Segment
T2 Crack 5 T2 Segment
2 T1 Crack
Segment Volume
T2 Crack
Number of CCs
4
1.5
3
1
2
0.5 1
0 0
0 50 100 150
Threshold()
200 250 0 50 100 150
Threshold()
200 250 (a) (b) (c)
(a) (b)
10
6
10
10 Figure 3. Sample slice image of binarized at different
thresholds: (a) ; (b)
; (c) .
T1 Segment T1 Segment
T2 Segment
T1 Crack
T2 Segment
T1 Crack
Number of CCs
T2 Crack
Number of CCs
4
10 T2 Crack
5
10
10
2
Table 1. Volume of the largest CCs at
.
Volume
0 0
10 0
10 10
5
10
10 10
0 100 200 300 Total Foreground
CC Volume Threshold()
Largest
(c) (d) segment CC () ( )
2nd Largest
Figure 2. Statistics of , against binarization thres- segment CC ()
hold: (a) Background volume; (b) Number of CCs; (c)
Total Background
Distribution of number CCs at ; (d) Minimum
number of CCs covering 99% of total foreground and
Largest
crack CC (
) ( )
background volume.
2nd Largest
crack CC ( )
or
two dominant segments in and one dominant segment
in ; at
, there are only two dominant cracks in
is caused by the growth of the largest crack CCs. The
growth of the largest CCs accounts for almost 99% of the
and one dominant crack in . enlargement of the entire specimen. Therefore, focusing on
In Figure 3, the results are shown with one slice of the largest segments and cracks will suffice to analyze the
binarized at
, respectively. At , the change in the specimen under load. In particular, the speci-
cracks are disconnected because of binarization; at , mens enlargement can be attributed mostly to the widening
we see the consolidation of aggregates and sand, but the of the largest cracks.
cracks have disappeared; at
, both segments and Under increasing load, the largest two segments and
cracks are identifiable. cracks in have merged in .
3
ratio, indicating both having honeycomb-like struc-
Table 2. Volume, surface area, and surface area to volume tures.
ratio of the largest CCs at threshold 79. We have much work ahead of us. Other than thres-
holding, we would like to explore segmentation techniques
Segment CC Largest 2nd Largest that can provide a better discrimination between segments
Volume
and cracks. We dont yet have any effective measures of
Surface Area
segment and crack shape and topology. Furthermore, cur-
Ratio
rent 3D image registration techniques will have to be ex-
Volume
tended to the compound problem of bringing into corre-
Surface Area
spondence objects exhibiting both global distortion (motion
Ratio
If we are able to discriminate segments and cracks inside the
Ratio
specimen, we can determine which cracks open to the sur-
Volume
face. By tracing these cracks, we can compute what fraction
Surface Area