Download as pdf or txt
Download as pdf or txt
You are on page 1of 23

Road Materials and Pavement Design

ISSN: 1468-0629 (Print) 2164-7402 (Online) Journal homepage: https://www.tandfonline.com/loi/trmp20

Aggregate structure characterisation of asphalt


mixtures using two-dimensional image analysis

Aaron R. Coenen , M. Emin Kutay , Nima Roohi Sefidmazgi & Hussain U.


Bahia

To cite this article: Aaron R. Coenen , M. Emin Kutay , Nima Roohi Sefidmazgi & Hussain
U. Bahia (2012) Aggregate structure characterisation of asphalt mixtures using two-
dimensional image analysis, Road Materials and Pavement Design, 13:3, 433-454, DOI:
10.1080/14680629.2012.711923

To link to this article: https://doi.org/10.1080/14680629.2012.711923

Published online: 11 Sep 2012.

Submit your article to this journal

Article views: 961

View related articles

Citing articles: 51 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=trmp20
Road Materials and Pavement Design
Vol. 13, No. 3, September 2012, 433–454

Aggregate structure characterisation of asphalt mixtures using


two-dimensional image analysis
Aaron R. Coenena , M. Emin Kutayb , Nima Roohi Sefidmazgia and Hussain U. Bahiaa *
a Department of Civil and Environmental Engineering, University of Wisconsin-Madison, 1415 Engineering

Dr., Madison, WI 53706, USA; b Department of Civil and Environmental Engineering, Michigan State
University, 3554 Engineering Bldg, East Lansing, MI 48824, USA

In current practice of mixture design, volumetric properties such as voids and binder content
along with mechanical properties such as modulus or rutting resistance are used as the main
quality indicators. Visualisation is an important tool that has not been widely used in asphalt
mixtures. As part of the Reunion Internationale des Laboratoires et Experts des Materiaux
activities, the aggregate structure has been identified as a possible important mixture char-
acteristic in need of measuring and quantifying. This paper is a report on part of this effort.
Software for processing and analysing two-dimensional images of asphalt concrete mixtures
to provide information about the aggregate structure within a mix was developed. Images with
accompanying volumetrics and gradation information can be processed with the software and
a virtual sieve analysis of aggregates within the image is performed to verify a match with
known measured gradations. Once images were successfully processed, analysis is performed
to determine the number of contact points between aggregates as well as radial distribution
and orientation of each aggregate. Segregation of aggregates within each specimen was also
determined. Mixtures with a broad range of variables were compacted in the laboratory, using
a number of compaction methods of various countries. In addition, mixtures with various nom-
inal maximum aggregate sizes, aggregate type (limestone or gravel) and design ESALs (E-3 or
E-10) were compacted in the US gyratory compactor, using two pressures (600 and 300 kPa)
and two temperature levels (120◦ C and 60◦ C). Results indicate that the aggregate structure
is affected by compaction methods and conditions although volumetrics are very similar. The
results show that a fresh look at evaluating the aggregate structure within mixtures is required.
Keywords: aggregate structure; imaging; asphalt mixtures; compaction; rutting

1. Introduction
Digital imaging techniques for asphalt concrete are effective tools for evaluating the internal
structure. These techniques have been advanced throughout the years by a number of researchers
(Masad, Muhunthan, Shashidhar, & Harman, 1999; Rothenburg, Bogobowicz, & Haas, 1991;
Tashman, Wang, & Thyagarajan, 2007; Yue, Bekking, & Morin, 1995). The capabilities of these
techniques improved significantly with technology and advancing equipment availability. Over
the course of the last 15 years, significant strides have been made towards affordable, readily
available tools capable of providing users with measurements of interest.
The interest in the internal structure of asphalt mixtures is due to the heterogeneity of the
material that consists of multiple components, including the (coarse and fine) aggregates which
provide the main load-bearing skeleton, visco-elasto-plastic asphalt binder and air voids. In

*Corresponding author. Email: bahia@egr.wisc.edu

ISSN 1468-0629 print/ISSN 2164-7402 online


© 2012 Taylor & Francis
http://dx.doi.org/10.1080/14680629.2012.711923
http://www.tandfonline.com
434 A.R. Coenen et al.

the past, obtaining images was treated as a two-step process in which a camera was used
to obtain a photograph of a specimen and then digitised with a flatbed scanner (Yue et al.,
1995). The objective was guided by the concept that physical performance of asphalt con-
crete is significantly influenced by the aggregate structure of the mix, and can therefore be more
accurately and reliably predicted by more precise characterisation of this structure (Masad, Lit-
tle, & Sukhwani, 2004; Masad, Al-Rousan, Bathina, McGahan, & Spiegelman, 2007; Olard
& Perraton, 2010). Discrete or finite element analysis of asphalt concrete came into play in
order to model its components and have clearly defined boundaries between individual com-
ponents within the elemental mesh. However, the assumptions about interactions within the
discrete or finite element mesh began to face scepticism. Underlying assumptions included
modelling asphalt concrete as a two-part composite of elastic aggregate and linear viscoelas-
tic asphalt binder combined to form a homogeneous and isotropic continuum (Rothenburg
et al., 1991). Though asphalt concrete has always been considered to have a more complex
composition, it is the division between aggregate and the rest of the mix that has remained
the focus throughout several studies. The point-to-point contact between individual aggregates
(also referred to as aggregate interlock) is claimed to directly affect physical performance
(Emery et al., 1993). Further investigation of macroscopic properties, including various mix
designs compacted by a variety of laboratory and field methods, continued to show varia-
tion in performance contributed to differences in orientation and spatial distribution of coarse
aggregates.
In the late 1990s, with the adoption of the Superpave Gyratory Compactor (SGC), many
studies were focused on compacted specimens using the SGC in the laboratory along with
field cores to measure the newly defined parameters of aggregate orientation and segrega-
tion (Masad, Muhunthan, Shashidhar, & Harman, 1998; Masad et al., 1999). Work done to
estimate the strain distribution within binder domains of asphalt mixtures, using imaging and
finite element, was accomplished in several studies (Kose, Guler, Bahia, & Massad, 2000;
Masad, Somadevan, Bahia, & Kose, 2001). Imaging results were also used to apply com-
posite theories to predict the bulk modulus (Lakes, Kose, & Bahia, 2002). More recently, a
renewed interest in studying the influence of compaction methods was revisited. The influ-
ence of different European compaction methods on the final structure of compacted asphalt
was investigated by compacting one type of mix design using a variety of compaction equip-
ment (Hunter, Airey, & Collop, 2004). Zhang, Wang, Xu, and Ding (2008) compared the
aggregate orientation and distribution from four gradations and two compaction methods
to determine if the internal aggregate structure could be used to calculate the shear stress
resistance. Although many investigations have focused on using imaging and modifying the
definition of aggregate parameters, little focus has been placed on improving the image
processing stage.
With advancements in technology, obtaining digital images of the internal structure of asphalt
concrete is no longer challenging. However, little improvement has been made to the precision
and accuracy of analysing these images. Past researchers utilised various filtering procedures
and relied on personal judgment to define the aggregates’ boundaries (Masad et al., 1999).
It is important to provide users with tools to evaluate the accuracy of the aggregate bound-
aries obtained from the image by matching actual measured volumetrics. This could lead to a
more objective, standardised and uniform image-based analysis method. In addition, there is
a need for developing terminology to define image-based parameters that quantify the aggre-
gate structure. This could make the imaging technology a viable tool for quality control and
assurance. This paper presents a series of image processing and analysis techniques, that is, a
candidate for a standardised procedure for quantification of aggregate characteristics in asphalt
mixtures. The method presented herein utilises known characteristics of asphalt mixtures such
Road Materials and Pavement Design 435

as the volumetrics and aggregate gradation to “calibrate” the image filtering parameters (e.g.
Gaussian filter, threshold, etc.). This minimises the manual user input and increases the objec-
tivity of the results. After an accurate image processing phase, which results in a segmented
binary image showing individual coarse aggregates, key characteristics of aggregates such
as the contact points, orientation and segregation were computed. These computations were
performed on a variety of asphalt mixtures compacted at different levels to investigate the
effect of compaction parameters (e.g. temperature, pressure level, etc.) on the internal structure
characteristics.

2. Experimental study
This study focused predominantly on the SGC method of compaction as it is the most widely
used method in the USA at that time. In addition, to ensure that the imaging technique developed
can detect the effect of varying compaction conditions, additional methods and equipment were
studied. Images were all processed and analysed with the same software developed in this study.

2.1. Materials and testing conditions


The first part of this study focused on developing software for image analysis and the definition
of aggregate structure indices. The software, after careful testing and evaluation for repeatability
of results, was used for two main experimental studies (to evaluate effect of compaction methods
and conditions in the laboratory on aggregate structure parameters):
In the first study, images obtained from compaction of the one mix design at different laboratories
by the German steel sector, Marshall and Hveem compaction equipment, were analysed. The
mixture was produced by LCPC in France and was used for the construction of a field testing
section. The different laboratories shipped the samples for cutting, scanning and performing image
analysis.
In the second study, the effect of temperature and pressure on samples compacted in the US
gyratory compactor were studied. The samples included six mix designs that cover two nominal
maximum aggregate sizes (NMAS), two gradations (fine and coarse) and two aggregate types
(limestone and gravel). Figure 1 depicts the gradations used. For each combination, the mixtures
were compacted at 60◦ C and 120◦ C paired with two compaction pressures of 300 and 600 kPa,
yielding four combinations of temperature and pressure for each mix design. This level of variation
in both the temperature and the pressure was expected to result in a clear distinction between
aggregate structure measurements for the specimens.

Figure 1. Gradations of mixtures compacted using the gyratory compactor.


436 A.R. Coenen et al.

3. Software description
3.1. Image processing and analysis system sequence
The software developed for this study includes two main functions: processing of an image and
analysing the image for microstructure indices. The sequence used in the software is illustrated
by the flow chart shown in Figure 2. The steps at each stage are listed under the flow line.
There are input requirements that mainly include volumetric properties. The image processing
filter values (e.g. median filter size and Hmax size) are determined based on the volumetrics and
the aggregate gradation. The final output includes three main parameters: (i) number of contact
points, (ii) aggregate orientation spectrum and (iii) aggregate segregation. Details of the software
are explained the next section.

3.2. Image processing and analysis system details


The methodology incorporated into the software confirms appropriate processing of the image
and gives the user confidence that the image is a good representation of the actual mixture before
proceeding to the analysis. The software requires input of the known aggregate gradation to be
compared with the gradation obtained from the image. In addition, the software requires the input
of volumetric information composed of percent air voids, asphalt binder content and a specific
gravity of aggregates and the asphalt binder. The combination of this information allows the
software to perform basic calculations with volumetric relationships resulting in a volumetric
percentage of aggregates (Psv ). The Psv is used to estimate the planar (two-dimensional (2D))
percentage of the aggregate area. The calculations are used to verify the application of appropriate
filter parameters for the identification of boundaries of aggregates.
After filters are applied, the software isolates the individual aggregates (using watershed trans-
formation), allowing the area of each aggregate to be calculated. Then, the total percentage of
the aggregates with respect to the total area of the specimen in the image is computed. This 2D
relationship is compared with the volumetric calculation to ensure that the software is identifying
the mixture accurately based on known parameters. Since the software is identifying individual
aggregates in order to calculate the area fraction, the aggregates in the image are each converted
into the appropriate sieve size based on the image resolution. The image resolution is used to
control the smallest size of aggregate the software should identify for determining boundaries.
This allows the software to separate the aggregates according to the previously identified sieve
sizes (from laboratory gradation) and perform a virtual sieve analysis of the aggregates in the
image. The virtual grain size distribution is plotted on the screen on the same axes as the labora-
tory gradation and both the volumetric and area fractions of aggregates are displayed at this time.
Each of these comparisons allows the user to better determine if the selected threshold and filter
values are appropriate.

Figure 2. Schematic of image processing and analysis software.


Road Materials and Pavement Design 437

Original image Median filtered Hmax filtered

(a) (b) (c)

After watershed transform After thresholding Labeled image


7 16 44
48 56 75 83
17 19 52 82
(d) (e) (f)
1 10 14 23 31
39
46
61

8 3742 57
62 64
34 60
47
6 18 28 69 77
21 30 6366 79
2 35 68 70
59
9 2025 67
13 65
3 29 72
4143 38 76
73
4 11 36 80
33 50
51 71
5 15 27 45 53 81
24 32 49 58 74

Figure 3. Illustration of steps of image processing filters.

The following digital image processing filters are picked based on a series of trial and error and
studies that are appropriate to asphalt mixtures: First, the median filter is applied to remove the
random noise in the image (Figure 3(b)). Then, the regional maxima (Hmax ) filter is applied to elim-
inate the variation in pixel intensity of the aggregates to give a uniform grey value (Figure 3(c)).
The Hmax filter performs this task without changing the intensities of the darker (i.e. lower inten-
sity) regions (Soille, 1999). Next, the image is inverted, that is, all the pixels are subtracted from
255, and watershed transformation is performed (Figure 3(d)). After this process, the image is
divided into unique watershed regions and the boundaries of these regions are determined (Meyer,
1994). The last step is the conversion of the greyscale image into a binary image, using a threshold
value (Figure 3(e)). Then, a binary labelling operation is performed to change the pixel inten-
sities of islands/regions of white pixels (i.e. aggregates) to unique integers (Figure 3(f)). This is
accomplished by using a connected components algorithm. The filtering process is not a sub-
jective process since the best filtering values for an image are unique and based on the volume
fraction/area fraction and laboratory gradation/image-based gradation comparison.
The software then calculates the coarse aggregate fraction (Psv im
) and gradation (PRimi ) from
the image. The geometric properties of each labelled region are then calculated. These include
the area, equivalent diameter and centroid, with calculations shown in Equations (A6), (A7) and
(A8), respectively. The software then eliminates the labelled regions with an equivalent diameter
eq
less than the user-defined minimum aggregate size (Dj < Dmin ). The total area of the coarse
aggregates from the image (Ac ) is then calculated using Equation (A9), where Ac corresponds to
the total area of aggregates larger than Dmin . The total area of all aggregates (At ) is calculated
according to Equation (A10). The percentage of aggregates retained in each sieve size (e.g.
Di = 2.36, 4.75, 9.5 mm, etc.) PRim i is determined using Equation (A11). The percent coarse
im
aggregate (Psv ) from the image is then determined using Equation (A12).
The graphical representation of comparing the Psv im
value with Psv as well as PRim
i with PRi which
is comparing the known/entered aggregate gradation with the virtual gradation identified in the
438 A.R. Coenen et al.

processed image helps in the selection of best filtering values. The filter selection process may be
required to be repeated until these gradations and fractions agree. Once they are in agreement, the
user proceeds to the analysis for the determination of contact points, orientation and segregation;
the significant outputs of the software developed for this study.

3.2.1. Software output


3.2.1.1. Contact points. Since the total number of aggregates is a relative term, the connectivity
of aggregates within a matrix must be defined. In this study, the index representing the connectivity
is defined as an aggregate-to-aggregate contact and the total number of occurrences within the
matrix is termed the number of contact points. Since the precise point of contact within a 3D
matrix is rarely captured in a 2D slice of the specimen, a contact distance must be defined. The
2D image will not show the perimeter of aggregates to be within contact unless the specimen
was sliced immediately adjacent to the 3D contact point. For this reason, any distance larger
than immediately adjacent is seen as a small distance between aggregate perimeters in a planar
image. This distance is defined in this study as the surface distance threshold (SDT )value and
is a user-defined value and is adjustable based on the mixture type, aggregate size, method of
cutting specimen, etc. By defining the contact distance and the minimum aggregate size to be
considered for the calculation of the total number of contact points, the software searches the
entire image for locations that meet the criteria and reports the total number of contact points and
also produces an auxiliary file which assigns a reference number to each contact point, identifies
individual aggregate ID of the two in contact and the equivalent diameter of a circular (spherical)
aggregate with the same cross-sectional area as each irregular aggregate identified in the image.

3.2.1.2 Aggregate orientation. The major axis of each aggregate identified in the image is located
and measured. The centre of each major axis and the centroid are connected by a radial arm. The
orientation of each aggregate can now be defined in two ways: with respect to the horizontal (based
on a Cartesian coordinate system assigned as vertical and horizontal on the scanned image) and
the radial orientation which is defined as the magnitude of the angle between the radial arm to
each aggregate and the major axis of that aggregate.

3.2.1.3 Spatial distribution (segregation). In order to maintain aggregate characteristics as


defined by previous studies, the software of this study determines segregation of aggregate within
the image. The image is divided into regions and the total number of aggregates of each sieve size
within each region is calculated, tabulated and displayed as a histogram. Each of these outputs is
determined as follows.

3.3. Procedure for determining contact points


Contact points are calculated using the surface pixels of each aggregate. The minimum distance
between the surfaces of neighbouring aggregates is calculated. If this distance is less than the
SDT value, the aggregates are assumed to be in contact. Figure 4 shows an example of contact
points calculated between aggregates.
It should be noted that the user-defined SDT value is not arbitrarily chosen. A sample plot
of the number of contact points versus the SDT is shown in Figure 5. The relationship shows a
plateau in the middle region of the plot, where the contact distance used to define a contact point
does not significantly affect the total number of contact points found within the image. Above
this plateau region is a linearly increasing range where the number of contact points continues to
rise as the radius of search from the perimeter of each aggregate grows. This relationship would
Road Materials and Pavement Design 439

Figure 4. Illustration of contact points.

Figure 5. Plot of number of contact points versus contact distance for selection of SDT.

continue until a large enough SDT is selected to allow each aggregate in the image to be identified
as in contact with every individual aggregate, which is clearly not an accurate representation. For
this reason, this linearly increasing region should be avoided. The lower end of the plateau is
interrupted by an abrupt decrease to zero contact points identified in the image. This is caused by
the selected SDT approaching the resolution of the image, beyond which contact points are no
longer clearly identified. This check for an appropriate contact distance should be performed to
logically select an SDT value for analysis.

3.4. Procedure for determining aggregate orientation spectrum


The first step in computation of aggregate orientation is the determination of the major principal
axis (Dmax ) of an aggregate. The Dmax is determined using Equation (A13). The orientation
feature calculates the angles from the horizontal axis (α) and from the radial axis (θ), using
Equations (A14) and (A15), respectively. The angles θ and α are identified in Figure 6.
A sample histogram of the radial and horizontal distributions is shown in Figure 7.

3.4.1. Aggregate orientation analysis parameters


Images processed and analysed as part of this study were captured at a resolution of 600 dpi
(dots per inch, the number of pixels that represent one US inch, approximately 24 pixels/mm),
which is the equivalent of approximately 0.04 mm/pixel. The distinction between sand and gravel
is generally taken to be 4.75 mm diameter. At the resolution of 0.04 mm/pixel, the minimum
aggregate or gravel identified by a diameter of 4.75 mm is seen in the image as approximately
120 pixels, or a radius of roughly 60 pixels. Using the default median filter size of 5 pixels, the
440 A.R. Coenen et al.

Figure 6. Illustration of orientation angles.

25
Angle from horizontal
Percentage of aggregates (%)

20 Angle from the radial axis

15

10

0
0 20 40 60 80 100 120 140 160 180
Angle (º)

Figure 7. Example histogram of radial and horizontal angles computed for all aggregate identified in an
asphalt specimen.

software identifies the intensity of each pixel and the surrounding radius of 5 pixels, replacing
each pixel with the median value of the 5-pixel radius region. This applies a “smoothing” effect
to the image, which results in a minute loss of information in this processing step. The loss has
the potential to influence a zone of up to 5 pixels. Based on a 5-pixel shift of a diametrical line,
rotated about the centre, the minimum aggregate size (radius of 60 pixels) can shift ±5◦ , or a total
swing/rotation of 10◦ . This is illustrated in Figure 8.
This tolerance is critical information when determining the orientation of individual aggregates
relative to the major axis of the aggregate. The major axis has this tolerance of nearly 10◦ . For
this reason, the analysis performed in this study determines the frequency of angular occurrence
and also fits a harmonic function to aggregate orientation presented in the histogram form divided
by increments of 10◦ (as finer increments would exceed the tolerance).
Though the software developed in this study is capable of measuring both horizontal and radial
orientation, when comparing horizontal slices of laboratory compacted specimens from various
methods, it is not reasonable to look at the horizontal orientation. The orientation relative to
horizontal requires a fixed point of reference and is more appropriate for field cores when the
standard reference direction is assigned as either the longitudinal or transverse direction of the
roadway. For laboratory compacted specimens, this standard reference no longer applies due to
the varying geometries of compaction equipment and the ease of freely rotating the specimen after
compaction. For this reason, the effect of the compaction method on aggregate orientation was
Road Materials and Pavement Design 441

Figure 8. Tolerance of orientation calculation.

Figure 9. Conversion from (a) histogram representation to (b) harmonic fit.

determined by comparing radial orientation (angle θ measured between aggregate major axis and
the radial arm drawn from the centre of the image to the centre of the aggregate), and is presented
in Figure 9.
442 A.R. Coenen et al.

Figure 10. Additional parameters defined for harmonic fit.

Figure 10 shows a standard histogram produced by dividing the individual aggregate orientation
into 10◦ intervals. Figure 9(b) shows the same frequency of occurrence per angle represented as
line graph rather than a bar chart with the addition of a harmonic fit. Tashman, Masad, Peterson,
and Saleh (2001) presented the idea of assigning a harmonic function to be used as a fit for the data.
This fit clearly identifies (by the maximum of the sinusoidal wave) the predominant aggregate
orientation within the image. This is accomplished by using the following equations:
freqharm = freqave (1 + a∗ cos 2θ + 2b∗ sin θ cos θ − a∗ sin 2θ), (1)
where
N
[2 cos(2θk )]
a= k=1
, (2)
N
N
[2 sin(2θk )]
b= k=1
(3)
N
and N is the total number of aggregates accounted for in the image.
Since the fit used is a harmonic function, the localised maxima (180◦ apart) are identified.
Based on these maxima, additional parameters of the harmonic fit can be defined. The additional
parameters of interest are δ and A, the angle corresponding to the first maxima and the amplitude
of the harmonic, respectively. The δ identifies the predominant orientation as determined by the
fit, whereas the amplitude of the harmonic is a measure of the severity or variance from a uniform
distribution. A uniform distribution of aggregate orientation would result in an amplitude of zero –
a horizontal line. Thus, the lower the amplitude, the more uniform is the orientation.

3.5. Procedure for determining spatial distribution (segregation)


The third output is the quantification of aggregate segregation. This is accomplished by dividing
the image into three radial groups as shown in Figure 11. These three regions (radially) are:

Group 1 (aggregates within Rim /3 circle where Rim = min(Sx , Sy )/2);


Group 2 (aggregates in a ring-shaped area between 2/3 Rim and 1/3 Rim ) and
Group 3 (aggregates in a ring-shaped area between Rim and 2/3 Rim ),

where Sx and Sy are sizes of image in x and y directions, respectively.


The quantification is achieved by determining the aggregates whose centroids are within each
group. The software then plots (in the form of a histogram) the percentage of aggregates in
Road Materials and Pavement Design 443

Figure 11. Illustration of radial segregation groups.

60
Group-1
Percentage of aggregates in

50 Group-2
Group-3
each group (%)

40

30

20

10

0
4.75 9.5 12.5 19 25
Sieve sizes (mm)

Figure 12. Example distribution of different aggregate sizes in each group.

each group for each sieve size. An example of such a histogram is shown in Figure 12. Each
analysis/output function generates a results file where the data is tabulated for additional analysis.
To process and analyse the images of asphalt concrete specimens, it is essential to discuss how
the capturing of digital images was controlled.

3.6. Protocol for capturing images


Digital images were obtained using an economical and widely available technique of scanning
the specimen on a flatbed scanner at a resolution of no less than 600 dpi. Prior to scanning the
specimen, the specimen must be cut horizontally with a masonry saw at two locations dividing
the specimen into three equally dimensioned sections and allowing for multiple images of each
mix to be obtained. Locale of the slices is illustrated in Figure 13.

4. Results and analysis


In comparing the effects of various compaction methods on the aggregate structure using the
technique described, the harmonic fit was used as shown in Figure 14 and Table 1. The results
(i.e. for analysis performed on two replicate per compaction method) indicate that there are
significant effects of the compaction method on the calculated parameters (A and δ). For the
444 A.R. Coenen et al.

Figure 13. Location of slices/cuts of gyratory specimen for imaging.

Figure 14. Effect of compaction method.

Table 1. Predominant angle (δ) and severity


(A) of various compaction methods.

Compaction method δ A

Hveem 106 5.14


Marshall 122 2.78
German steel sector 138 6.38
Superpave Gyratory 83 7.3

analysed mixture, the SGC had the greatest value of A (7.3), which implies that it has the highest
tendency to orient the aggregate to a particular angle, 83◦ from the radial arm in this case, while the
Marshall had the lowest severity (value of A) or closest to the uniform value of the four methods,
with an amplitude of 2.78.
The relative ranking of the compaction method according to the severity of aggregate orientation
places Marshall as the least influential and Superpave Gyratory as the most influential, with the
Hveem and German steel sector falling in-between. The Superpave Gyratory also reported a
predominant angle closest to 90◦ , which is indicative of aligning the aggregates concentric to
the perimeter. This is expected since the gyratory is a circular axisymmetric compaction method.
To further study the effect of compaction conditions (temperature and pressure), the standard
Superpave Gyratory compaction technique was used.

4.1. Effects of compaction temperature and pressure (gyratory)


Several combinations of compaction temperatures and pressures were compared to determine the
contribution of each to the severity of aggregate orientation as well as determining the number
Road Materials and Pavement Design 445

Figure 15. Comparison of normalised contact points with percent air voids for 19 mm mixes.

Figure 16. Comparison of normalised number of contact points with percent air voids for 12.5 mm mixes.

of contact points between adjacent aggregates in each image, as shown in Figures 15 and 16.
It is seen in Figures 15 and 16 for the variety of gyratory specimens chosen (which include 19 mm
and 12.5 mm NMAS, limestone and gravel aggregates) that the number of contact points found in
the image follows a similar trend to that of percent air voids found in the mix, using conventional
laboratory techniques. To avoid the false influence of gradation of each aggregate type and NMAS,
the number of contact points identified in each image was normalised by the gradation (10% of
the percent retained on the 4.75 mm sieve). For example, the total number of contact points
identified as 160 for a gradation with 40% of the aggregate retained on the 4.75 mm sieve, would
be normalised to 40 contact points. The collective results, although showing variations, follow a
trend indicating that a higher void content usually results in less contact points. In other words,
lower temperature and lower pressure lead to less contact points. These small data sets confirm
the current practice that density is a critical indicator of asphalt mixture stability, in the sense that
it leads to higher contact points. Obviously changes in gradation and material properties for a
given density could result in higher or lower number of contact points. The argument is then if
we have a simple tool for a direct estimate of contact points, it could be a critical complement to
the density for reducing the risk of mixture instability.

4.2. Analysis of orientation results


The orientation of the aggregates within each gyratory specimen was evaluated on the same
bases as those used for comparison of different compaction methods (using the parameters of the
predominant angle of aggregate orientation (δ), and the level of severity (A) estimated from the
harmonic fit of the orientation spectrum). This evaluation is tabulated in Tables 2 and 3. No clear
trend is identified in either the predominant orientation angle, δ, or the severity, A, in either of the
446 A.R. Coenen et al.

Table 2. Orientation parameters of 19 mm gyratory specimens.

Compaction temperature and pressure δ A

12.5 mm gravel
600 kPa and 120◦ C 83 21.34
600 kPa and 60◦ C 67 15.78
300 kPa and 120◦ C 104 7.22
300 kPa and 60◦ C 102 15.17
12.5 mm limestone
600 kPa and 120◦ C 73 6.42
600 kPa and 60◦ C 91 5.42
300 kPa and 120◦ C 71 27.03
300 kPa and 60◦ C 104 13.88

Table 3. Orientation parameters of 12.5 mm gyratory


specimens.

Compaction temperature and pressure δ A

12.5 mm gravel
600 kPa and 120◦ C 83 21.34
600 kPa and 60◦ C 67 15.78
300 kPa and 120◦ C 104 7.22
300 kPa and 60◦ C 102 15.17
12.5 mm limestone
600 kPa and 120◦ C 73 6.42
600 kPa and 60◦ C 91 5.42
300 kPa and 120◦ C 71 27.03
300 kPa and 60◦ C 104 13.88

tables. This indicates that although the SGC appeared to the most heavily influenced aggregate
orientation when compared with other methods of compaction, altering additional compaction
parameters such as temperature and pressure neither exaggerates nor hides the level of influence
on orientation caused by the gyratory.

4.3. Effect of NMAS and gradation


This study also investigated the effect of NMAS and aggregate gradation on the number of contact
points. The initial results of various NMAS repeatedly show that a larger NMAS results in a greater
number of contact points identified in the image. To better account for the effect of the varying
aggregate size on the number of contact points identified, the individual gradation of each mix
must be more closely examined. To do so, the Weibull cumulative distribution function was used
to fit each mix gradation of this study as follows:

F(x, α, β) = 1 − e−(x/β) ,
α

where β indicates fineness/coarseness and α indicates how dense the mix is (Mahmoud & Bahia,
2010).
The corresponding α and β values can be seen in Table 4. With this information, the β param-
eter can be used to determine the effect of fineness/coarseness on the number of contact points.
Figure 17 shows a strong correlation between β and the number of contact points. This allows
Road Materials and Pavement Design 447

Table 4. α and β values determined by Weibull


function fit for each gradation.

Gradation α β
Mix 1 0.8827 4.791
Mix 2 0.8023 4.563
Mix 3 0.9928 8.059
Mix 4 0.7124 3.353
Mix 5 0.7596 3.422
Mix 6 0.7623 3.269

Figure 17. Beta value versus number of contact points for all gradations of study.

for the separation of data points based on the specific gradation, as seen in Figure 17, where it
is clear that for each combination of temperature and pressure, the data points for an individual
gradation are grouped near one another. This suggests that the gradation, as defined by β, has a
significant influence on the number of contact points identified within an image.

4.4. Segregation
The third type of aggregate structure characteristic analysis performed in this study is determining
the segregation of aggregate within a mix. Segregation provides further insight as to whether or
not the gyratory action has any influence on aggregate migration within the mix relative to the
particle size. To investigate this, the percent passing each sieve size from one standard sieve above
NMAS to one sieve size below 4.75 mm (the particle size used for contact points and orientation
evaluation) was determined for each mix. An example of the results of these partial gradations is
illustrated in Figure 18 for the 19 mm limestone. The results show no evidence of segregation. In
fact, none of the mix images analysed in this study displayed extreme segregation of aggregates
caused by using the gyratory compactor.

4.5. Comparison to field compaction


Due to variation in geometry as well as shipping from around the globe, all laboratory compacted
specimens were cored to 100 mm diameter to be consistent with geometry. When measuring
aggregate orientation relative to horizontal axis on the screen/image, the horizontal or vertical
direction must have a specific meaning or coincide with a known direction in the lab or the field.
Since this was not clearly identified as compaction direction, longitudinal or transverse for the
laboratory specimens, the horizontal reference becomes arbitrary and is dependent on the user
scanning the digital images of the physical specimens. To avoid this arbitrary assignment, the
448 A.R. Coenen et al.

Figure 18. Grain-size distribution of radial groups for 19 mm limestone mix used to determine the level
of segregation.

Table 5. Predominant angle (δ)


and severity (A) from field samples.

Location δr4 mm A
F10− left 94 3.46
F10− centre 97 4.14
F10− right 69 4.5
F20− left 71 1.06
F20− centre 88 0.6
F20− right 102 2.54
Average 86.83 2.72

radial angle (angle from a radial arm drawn from the centre of the image to the centre of the
aggregate to the major axis of the aggregate) was used. Any rotation of sample in shipping or
scanning will not result in a change to this parameter.
With this in mind, though longitudinal versus transverse direction can be clearly identified
when analysing images from the field (in this case, a test track), to remain consistent with the
analysis of laboratory methods and thus determine which laboratory compaction method most
closely reproduces the internal aggregate structure observed in the field, the radial angle must
again be used. Table 1 showed the predominant angle (δ) and severity (A) identified by harmonic
fits applied to the distribution of aggregate orientation found in each of the studied laboratory
compaction methods. This can be compared to the obtained field values as follows:
F10 and F20 in Table 5 represent two locations on the test track. F10 is at the central line of the
lane while F20 is under the wheel path. The corresponding left, centre and right with each location
represent each of the three images that were captured for that specific location. Comparing the
average predominant angle of approximately 87◦ with the various laboratory compaction methods,
it can be seen that this aggregate orientation is most closely reproduced by the gyratory compactor,
though an average severity of 2.72 from the field relative to 7.3 produced by the gyratory suggests
that the orientation of the aggregate in the field is not as heavily influenced as by the gyratory
and that level of severity is best represented by the minimally influencing laboratory method of
Road Materials and Pavement Design 449

Figure 19. Aggregate orientation histograms as line graph (based on bins of 10◦ as determined by
resolution of images).

Table 6. Combined predomi-


nant angle (δ) and severity (A)
from field samples.

Location δr4 mm A
F10 96 7.60
F20 93 3.74

the Marshall Hammer (a severity of 2.78). To better understand which of these two methods best
simulated aggregate orientation in the field, we must revert back to the harmonic fits derived
from the individual data. For this reason, Figure 19 shows the aggregate orientation histograms
represented by the corresponding harmonic fit for the gyratory, Marshall and field samples.
Figure 19 shows the two locations, F10 and F20, individually represented after determining
any outliers of the three based on data presented in Table 5. This additional analysis results in
data as presented in Table 6.
It can be seen from the data presented in Table 6 that the predominant aggregate orientation
continues to hover near that of the gyratory even with separating the F10 samples from F20
samples in the analysis. It can also be seen that the repeatability of orientation of the aggregate
in each image at a given field location allows for a histogram representation to begin to show
an amplified severity as peaks and valleys of individual harmonics combine for an exaggerated
harmonic fit. This increased severity, specifically at location F10, also begins to approach that of
the gyratory compactor, while location F20 still exhibits a slightly lower severity with a similar
predominant aggregate orientation.

5. Summary of findings and conclusions


In this study, a software for 2D image analysis and aggregate structure characterisation of asphalt
mixtures was developed. This all-inclusive software allows for processing of the image prior
to analysis to ensure accuracy of results. For the analysis of the image aggregate orientation,
spatial distribution, and contact zones indices are proposed as critical characteristics to describe
the aggregate structure and study effects of compaction conditions. This software was used on
images of laboratory compacted asphalt concrete specimens that were captured directly from
a flatbed scanner. The processing steps along with the calculation of contact points between
aggregates were verified by determining an appropriate SDT that maximises the accuracy of
calculations while not exceeding the resolution of the image.
450 A.R. Coenen et al.

The aggregate structural parameters obtained were used to compare samples of a single mixture
design produced using various compaction methods at laboratories in several countries. A method
for representing aggregate orientation distribution by fitting a harmonic function indicated that
there are distinct differences in the aggregate orientation from various compaction methods. The
Superpave Gyratory showed the highest influence on the aggregate orientation in a radial fashion.
Additional investigation will be required to determine which of these laboratory methods most
accurately reproduces the internal structure parameters seen in field-compacted specimens.
The purpose of comparing laboratory compaction methods is to determine the best which
reproduces the asphalt pavement as compacted in the field. By comparing the internal aggregate
structure, namely aggregate orientation for various laboratory compaction methods with the same
mix design compacted in a field-setting on a test track, it appears that the gyratory compactor best
duplicates the predominant aggregate orientation seen in the field. Data shows, however, that at
various locations in the field, the level of influence (severity) was not quite as strong as that seen
from the gyratory and was similar to alternate laboratory methods such as the Marshall or the
Hveem. It should be noted that the lower the severity, the less meaningful the specific angle of
the aggregate orientation.
This study also included the investigation of effects of varying compaction temperatures and
pressures used in the SGC on aggregate structure indicators. The specific temperature and pressure
of the gyratory compaction did not significantly exaggerate or minimise the aggregate orientation,
though a clear difference in the number of contact points between aggregates was identified. The
number of contact points within the gyratory specimens directly correlates with the percent air
voids in the mix, or density, which is commonly used to characterise flexible paving materials. A
larger NMAS also increases the number of contact points within a given mix. All this is occurring
without significant segregation of aggregates within the asphalt concrete. A distinction was seen
in the number of contact points relative to the aggregate type (gravel versus limestone) though
no real conclusions can be drawn as to whether this is image-related or angularity-related until
further investigation of angularity is conducted for materials of this study.

Acknowledgements
The authors thank the many volunteer hours of the RILEM Task Group 2 members, specifically those from the
University of California Pavement Research Center, Asian Institute of Technology, Technische Universität
Braunschweig, Michigan Technological University and University of Wisconsin-Madison. A special thanks
to Mr Carl Johnson for his contributions and evaluation of the early iterations of software development.

References
Emery, J.J., Schenk, W., Carrick, J.J., Davidson, J.K., MacInnis, W.K., & Kennepohl, J.A. (1993). Stone
mastic asphalt trials in Ontario. Transportation Research Record 1427, TRB. Washington, D.C.: Nation
Research Council, pp. 47–53.
Hunter, A.E., Airey, G.D., & Collop, A.C. (2004). Aggregate orientation and segregation in laboratory
compacted asphalt samples. Proceedings of the 83rd Annual Meeting, Washington D.C.: Transportation
Research Board.
Kose, S., Guler, M., Bahia, H.U., & Massad, E. (2000). Distribution of strains within asphalt binders in HMA
using imaging and finite element techniques. Journal of the Transportation Research Board, TRR No.
1728, 21–27, Washington, D.C.
Lakes, R.S., Kose, S., & Bahia, H.U. (2002). Analysis of high volume fraction particulates with applications
to asphalt and high damping composites. Transactions of the ASME, 124(April), 174–178.
Mahmoud, E., & Bahia, H.U. (2010, January 10–14). Methods for measuring surface characteristics of
gyratory-compacted asphalt mixes and relationships to volumetric design. Poster session presented at
Transportation Research Board 89th Annual Meeting, Washington, D.C.
Road Materials and Pavement Design 451

Masad, E., Al-Rousan, T., Bathina, M., McGahan, J., & Spiegelman, C. (2007). Analysis of aggregate
shape characteristics and its relationship to hot mix asphalt performance. International Journal of Road
Materials and Pavement Design, 8(2), 317–350.
Masad, E., Little, D., & Sukhwani, R. (2004). Sensitivity of HMA performance to aggregate shape measured
using conventional and image analysis methods. International Journal of Road Materials and Pavement
Design, 5(4), 477–498.
Masad, E., Muhunthan, B., Shashidhar, N., & Harman, T. (1998). Aggregate orientation and segregation in
asphalt concrete. ASCE Getotechnical Special Publication, 85, 69–80.
Masad, E., Muhunthan, B., Shashidhar, N., & Harman, T. (1999). Internal structure characterization of
asphalt concrete using image analysis. ASCE Journal of Computing in Civil Engineering (Special Issue
on Image Processing), 13(2), 88–95.
Masad, E., Somadevan, N., Bahia, H., & Kose, S. (2001). Modeling and experimental measurements of
strain distribution in asphalt mixes. Journal of Transportation Engineering, ASCE, 127(6), 471–477.
Meyer, F. (1994). Topographic distance and watershed lines. Signal Processing, 38(July), 113–125.
Olard, F., & Perraton, D. (2010). On the optimization of the aggregate packing characteristics for the design
of high-performance asphalt concretes. Road Materials and Pavements Design, 11(Suppl. 1), 145–169.
doi:10.3166/rmpd.11hs.145-169.
Rothenburg, L., Bogobowicz, A., & Haas, R. (1991). Micromechanical modeling of asphalt concrete in
connection with pavement rutting problems. Proceedings of the 7th International Conference on Asphalt
Pavements, Nottingham, UK, Vol. 1, pp. 230–245.
Soille, P. (1999). Morphological image analysis: Principles and applications. Berlin: Springer, pp. 170–171.
Tashman, L., Masad, E., Peterson, B., & Saleh, H. (2001). Internal structure analysis of asphalt mixes to
improve the simulation of superpave gyratory compaction to field conditions. Journal of the Association
of Asphalt Paving Technologists, 70, 605–645.
Tashman, L., Wang, L.B., & Thyagarajan, S. (2007). Microstructure characterization for modeling HMA
behavior using imaging technology. Journal of Road Materials and Pavement Design, 8(2), 207–238.
Yue, Z.Q., Bekking, W., & Morin, I. (1995). Application of digital image processing to quantitatively study
of asphalt concrete microstructure. Transportation Research Record 1492, TRB, Washington, D.C.:
National Research Council, pp. 53–60.
Zhang, L., Wang, Z., Xu, G., & Ding, Y. (2008). Characterization of HMA internal structure using image
analysis. Proceedings of the Eighth International Conference of Chinese Logistics and Transportation
Professionals, ASCE Conf. Proc., 330, 389.

Appendix
A.1 Governing calculations
Image resolution:

D(mm)
x = y = , (A1)
Dim (pixels)

where D is the known distance in physical units (e.g. mm) and Dim the distance measured in the image in
pixels.
Volumetric percentage of aggregates with respect to the total volume (Psv ):

Vs (1 − VTM)
Psv = = , (A2)
V (Pb (Gs /Gb ) + 1)

where V is the total volume of the asphalt mixture (mix), Vs the volume of aggregates in the mix, Gb the
specific gravity of binder, Gs the specific gravity of aggregates in the mixture (assumed to be constant for
all aggregates in the mixture), VTM the voids in total mix (a.k.a. Va ) and Pb the binder content (by weight).
The derivation of Equation (A2) is given in the following section.
Volumetric gradation: percent retained (PR): It is noted that volumetric gradation of the aggregates is
identical to weight-based gradation measured in the laboratory, provided that the specific gravity is constant
452 A.R. Coenen et al.

for all aggregates in the mixture:

Wi G s γw V i Vi
Percent retained = PR = = = , (A3)
Wt G s γw V t Vt

where Wt is the total weight of the aggregates, Wi the weight of aggregates retained on ith sieve, Vt the total
volume of the aggregates and Vi the volume of aggregates retained on ith sieve.
Coarse aggregate fraction (CF): CF is the fraction of the aggregates larger than Dmin calculated using the
mix volumetrics (not from the image). Coarse fraction (CF) is defined as the volume of the coarse aggregates
(Vc ) divided by the total volume of the aggregates (Vt ) as follows:

i Vi Vc
CF = = for every i ∈ Di > Dmin . (A4)
Vt Vt

Percent coarse aggregate (Psv c ): P c is defined as the volume of the coarse aggregates (V ) divided by
sv c
the total volume of the entire asphalt mixture sample (V ) as follows:

c Vc
Psv = , (A5)
V

where V is the total volume of the mix including the aggregates, binder and air voids.
eq
Area (Aj ), equivalent diameter (Dj ) and centroid (xjc , yjc ) of the aggregates in the image:

Area: Aj = Nj ∗ x2 , (A6)

where Aj is the area of the labelled region (i.e. aggregate) j, Nj is the number of pixels in the labelled region
(i.e. aggregate) j and x the resolution of the image.

eq 4Aj
Equivalent diameter: Dj = , (A7)
π
Nj Nj
1  1 
Centroid: xjc = xk , yjc = yk , (A8)
Nj Nj
k=1 k=1

where xjc and yjc are the x- and y-coordinates of the centroid of the labelled region j, respectively, and xk and
yk are the individual coordinates of each pixel within labelled region (aggregate) j.
Total area of the coarse aggregates (Ac ) in the image: This corresponds to the total area of the coarse
aggregates (i.e. aggregates larger than Dmin ):

Nag

Ac = Aj , (A9)
j=1

where Nag is the total number of aggregates (labelled regions).


Total area of all aggregates (At ): At includes the aggregates that are not visible in the image and calculated
using the CF, which was determined from the known mix volumetrics. Equation (A9) can be rewritten for
a 2D image (i.e. in terms of area rather than volume), assuming that the property in the third dimension is
homogeneous:
Ac
At = , (A10)
CF
where At is the total area of the aggregates (including the aggregates not visible in the image).
Road Materials and Pavement Design 453

i ) retained on each sieve size (e.g. Di = 2.36, 4.75, 9.5 mm, etc.) calculated
Percentage of aggregates (PRim
from the image:
Ai Ai
Percent retained = PRimi = = CF , (A11)
At Ac
where
Ni
eq
Ai = Aj ⇔ j ∈ Di < Dj < Di+1 .
j=1

Percent coarse aggregate im )


(Psv from the image:

im Vc ∼ Ac
Psv = = , (A12)
V A
where A is the total area of the specimen including the aggregates, binder and air voids in the image.
Major principle axis (Dmax ): The first step in computation of the orientation is the determination of the
major principal axis (Dmax ) of an aggregate. The Dmax is determined using the formula:
 
Dmax = max (xi − x−i )2 + (yi − y−i )2 , (A13)

where xi , and yi are x- and y-coordinates of a surface pixel, x−i , and y−i are x- and y-coordinates of a surface
pixel at the opposite side of a line going through the centroid of the aggregate.
The angles from the horizontal axis (α) and from the radial axis (θ)
 
xi − x−i
α = cos−1 , (A14)
Dmax
 c
−1
xj
θ = cos , (A15)
Dmax

where θ and α are shown in Figure 6 and xjc as defined in Equation (A8).

A.2 Calculation of volumetric percentage of aggregates

v
wb vb

ws vs

V t = Vv + V b + V s , (A16)
Wb Wb
Gb = → Vb = , (A17)
V b γw Gb γw
Ws Ws
Gs = → Vs = , (A18)
V s γw G s γw
Vv = VTM Vt , (A19)
Wb
Pb = , (A20)
Ws
where Vt is the total volume of the asphalt mixture (mix), Vv the volume of voids in the mix, Vb the volume
of binder in the mix, Vs the volume of aggregates in the mix, Gb the specific gravity of the binder, Gs the
454 A.R. Coenen et al.

specific gravity of aggregates in the mixture (assumed to be constant for all size aggregates), γw the unit
weight of water, VTM the voids in total mix (a.k.a. Va ) and Pb the binder content (by weight).
Dividing Equations (A17) by (A18) reveals

Vb W b Gs Gs Gs
= = Pb → V b = V s Pb . (A21)
Vs Ws Gb Gb Gb

Plugging Equations (A19) and (A21) into Equation (A16) reveals

Gs
Vt = VTM Vt + Vs Pb + Vs . (A22)
Gb

Rearranging Equation (A22) yields


 
Gs
Vt (1 − VTM) = Vs Pb +1 , (A23)
Gb
Vs (1 − VTM)
Psv = = , (A24)
Vt (Pb (Gs /Gb ) + 1)

where Psv in Equation (A24) represents the volumetric percentage of aggregates with respect to the total
volume.

You might also like