evaluation of aggregate imaging techniques for the ...docs.trb.org/prp/13-2955.pdf · 1 evaluation...
TRANSCRIPT
Evaluation of Aggregate Imaging Techniques for the Quantification of Morphological 1
Characteristics 2 3
Linbing Wang1, Ph.D., P.E., 4
Professor, Charles E. Via, Jr. Department of Civil and Environmental Engineering, Virginia 5
Polytechnic Institute and State University; Director, Center for Smart Infrastructure and Sensing 6
Technology, Virginia Tech Transportation Institute (VTTI) 7
Virginia Polytechnic Institute and State University, Blacksburg, VA 24061 8
Phone: (540) 231-5262, Fax: (540) 231-7532, Email: [email protected] 9
10
Wenjuan Sun 11 Graduate Research Assistant, Charles E. Via, Jr. Department of Civil and Environmental 12
Engineering, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061 13
Email: [email protected] 14
15
Erol Tutumluer, Ph.D. 16
Professor, Department of Civil and Environmental Engineering, 17
University of Illinois at Urbana Champaign 18
205 North Mathews, Urbana, IL 61801 19
E-mail: [email protected] 20
21 Cristian Druta, Ph.D. 22
Research Scientist, Virginia Tech Transportation Institute (VTTI), Blacksburg, VA 24060 23
Email: [email protected] 24
25
Submission Date: August 1, 2012 26
Submitted for Presentation at the 2013 TRB Annual Meeting and Publication in the 27
Transportation Research Record: Journal of the Transportation Research Board 28
29
Word Count: Abstract: 214
Text: 3780
Figures: 8 × 250 = 2000
Tables: 6 × 250 = 1500
TOTAL: 7494
30
31
32
33
34
35
36 1 Corresponding Author 37
38
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 1
TRB 2013 Annual Meeting DRAFT 11/12/2012
ABSTRACT 1
Aggregate morphological characteristics, including shape, angularity and surface texture, have a 2
significant impact on the engineering properties of construction materials, such as hot-mix 3
asphalt, and hydraulic cement concrete, etc. Consequently, quantification of morphological 4
characteristics of aggregates is essential for quality control of both aggregate production and 5
pavement construction. Imaging techniques provide a cost-effective means to conveniently 6
measure the aggregate morphological characteristics without laborious work. However, these 7
imaging techniques adopt various mathematical methods with different instrument setups, 8
resulting in different definitions of morphological descriptors that are usually incomparable with 9
each other. This paper evaluates prevalent image techniques used for aggregate morphological 10
characteristics analysis, including equipment cost, repeatability and reliability, etc. Three 11
imaging techniques, i.e., the second generation aggregate imaging measurement system (AIMS 12
II), the first generation University of Illinois Aggregate Image Analyzer (UIAIA), and the 13
Fourier Transform Interferometer (FTI) system, are further evaluated by comparing the analysis 14
results of seven types of aggregates of passing 3/4’’ sieve and retained on 1/2’’ sieve in size 15
(hereinafter referred to as 1/2’’ aggregates) with manual measurements and visual rankings. 16
Analysis of variance between measurements using different methods is also conducted to 17
evaluate the accuracy of each aggregate imaging system. Based on the data analysis, 18
recommendations are made for the selection of appropriate imaging analysis techniques 19
depending on which morphological characteristics engineers are most interested in. 20
INTRODUCTION 21
Aggregates are an important component in asphalt concrete, cement concrete, granular base and 22
treated base. The morphological characteristics of aggregates, including shape, angularity, 23
surface texture and surface area, significantly affect the workability, durability, fatigue response, 24
and friction properties of both hydraulic Portland cement concrete (PCC) and hot-mixed asphalt 25
(HMA) pavements (1-3). Consequently, a great deal of research has been conducted to better 26
understand aggregate morphological characteristics of different aggregate and to establish the 27
relationship between aggregate morphological characteristics and field performances of 28
pavements (4, 5). Motivated by the advancements in image techniques and the availability of low 29
cost and rapid image-processing software, digital imaging techniques provide a reliable means to 30
rapidly quantify morphological characteristics of aggregates. 31
Historically, tremendous efforts have been made to quantify the aggregate morphological 32
characteristics using imaging techniques and correlate these characteristics to pavement 33
performance at both state level and national level. Some national efforts in the past few years 34
focused on the evaluation of direct measurement methods using two-dimensional (2D) image 35
analysis and semi-3D methods (i.e., 2.5D). However, 2D or 2.5D image analysis may not be 36
accurate enough to represent three-dimensional (3D) morphological characteristics of aggregates. 37
Besides, the current imaging techniques use different image acquisition methods and different 38
definitions of shape, angularity and texture, sometimes making the morphological 39
characterization results incomparable to each other. Consequently, it is vital to determine which 40
quantification methods of aggregate morphological characteristics are valid and accurate. 41
This paper evaluates the most widely used image analysis techniques for aggregate 42
morphological analysis. The comparison between features of different aggregate imaging 43
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 2
techniques is presented, followed by definitions of the morphological descriptors in the second 1
generation Aggregate Imaging Measurement System (AIMS II), the first generation University 2
of Illinois Aggregate Image Analyzer (UIAIA), and the Fourier Transform Interferometry (FTI) 3
system. Then the three imaging techniques, i.e., AIMS II, UIAIA, FTI systems, are adopted to 4
quantify the morphological characteristics of seven types of 1/2’’ aggregates (i.e., aggregates of 5
passing 3/4’’ sieve and retained on 1/2’’ sieve in size). The analysis results are compared to 6
manual measurements and visual rankings for further evaluations of the three aggregate imaging 7
systems. Manual measurements were conducted by using a caliper to measure three dimensions 8
of each aggregate particle by three different operators for three times, and the average value of 9
the measurements of each particle was used in this paper as manually measured results. 10
COMPARISON BETWEEN AGGREGATE IMAGING TECHNIQUES 11
Based on whether aggregates are moving or not during the image capture process, imaging 12
techniques for the quantification of morphological characteristics of aggregates can be generally 13
divided into two categories: dynamic digital image method and static digital image method. 14
Table 1 tabulates the experimental setups, estimated cost, and aggregate size ranges that each 15
system can measure, and other features of the aggregate imaging techniques. The estimated cost 16
of each aggregate imaging system is generally higher than $20,000, and the total number of 17
cameras installed in these systems ranges from one to three. 18
19
TABLE 1 Comparison between different Aggregate Imaging Techniques 20
Test
Method
Aggregate Imaging
System
Estimated
Equipment
Cost ($)
Analysis
Speed Accuracy
Ease of
Use Repeatability
Dynamic
Digital
Image
Method
VDG-40
Videograder 45,000 H M M L
Computerized
Particle Analyzer 25,000 H L M M
PSDA 50,000 L M H L
VIS 60,000 H M L L
Camsizer 45,000 M M M L
Winshape 35,000 H M M M
UIAIA 35,000 H H M H
PIAS 25,000 H H H M
Statistic
Digital
Image
Method
AIMS II 35,000 H H H H
FTI 20,000 M H M H
Note: H = high; M = medium; L = low. (Based on (6, 7)) 21
PSDA = Micrometric Optimizer Particle Size Distribution Analyzer, VIS = Video Image 22
System, UIAIA = University of Illinois Aggregate Image Analyzer, PIAS= Portable Image 23
Analysis System, AIMS II= the second generation of Aggregate Imaging Measurement System, 24
FTI = Fourier Transform Interferometry. 25
26
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 3
1
2
TABLE 1 (Continued) 3
Aggregate Imaging
System
Aggregate
Size Range
No. of
cameras
Dimensions of
imaged data
Measured
Characteristics
Dynamic
Digital
Image
Analyzer
VDG-40
Videograder #16 - 1.5'' 1 2 Shape
Computerized
Particle Analyzer #140 - 1.5'' 1 2 Shape
PSDA #200 - 1.5'' 1 2 Shape
VIS #16 - 1.5'' 1 2 Shape
Camsizer #50 - 3/4'' 2 2 Shape
Angularity
Winshape #4 - 1'' 2 3 Shape
Angularity
UIAIA #8 - 3'' 3 2
Shape
Angularity
Texture
Surface Area
Volume
PIAS #200 – 1'' 1 2
Shape
Angularity
Texture
Statistic
Digital
Image
Method
AIMS II #200 - 1'' 1 3
Shape
Angularity
Texture
FTI #50 - 3/4'' 1 3
Shape
Angularity
Texture
4
Among the ten aggregate imaging techniques, the three most widely used aggregate 5
imaging techniques are further evaluated in this paper, i.e., AIMS II, UIAIA, and FTI system. 6
AIMS II can image and analyze aggregates of a wide size range, from #200 to 1’’. Aggregates 7
are dispersed on a round tray over the tray trough in a manner which provides separations 8
between aggregates; the tray is placed on a turntable that can be operated by the AIMS software 9
to rotate automatically. There are seven trays with different trough sizes, and the tray is selected 10
depending on the aggregate size. A tray of aggregate is rotated while the camera captures an 11
image of each aggregate. The resolution of the captured image is size dependent, ranging from 12
44.78 µm/pixel for coarse aggregates, 17.91 µm/pixel for #8 (2.36 mm), 8.96 µm/pixel for #16 13
(1.18 mm), 4.48 µm/pixel for #30 (0.60 mm), to 3.36 µm/pixel for aggregates fine than #50 (0.3 14
mm). In UIAIA, all the aggregates are individually placed on the conveyor belt, with three 15
orthogonally positioned cameras to capture projections of aggregates. UIAIA can distinguish flat 16
and elongated aggregates and automatically calculate angularity and surface texture as well as 17
surface area and volume for a wide range of coarse and fine aggregates (76.2 mm to 2.38 mm). 18
As opposed to the black and white cameras and black background, the second generation 19
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 4
enhanced version of the UIAIA (UIAIA II) is now equipped with three progressive scan color 1
cameras (1292 × 964 pixel resolution) and a blue background to capture high resolution (0.056 2
mm/pixel) color images of aggregate. Moreover, an advanced color thresholding scheme is 3
utilized in the enhanced UIAIA software control. Therefore, different types of mineral 4
aggregates with various colors can be scanned with this system also now using four LED 5
illumination lights with dimmer control to achieve the sharpest possible aggregate images with 6
optimizing light intensity as well as minimizing shadows. Since UIAIA II is now available, the 7
results from UIAIA will only be used in this paper. The FTI system is able to analyze both 8
coarse and fine aggregates (3/4’’ to 0.3 mm) from different origins of various colors. Aggregates 9
are dispersed on a particle tray, and a charge coupled device (CCD) camera captures images of 10
the aggregate top surfaces from an angled-mirror. The resolution of FTI images is 35.4 µm/pixel 11
in both x and y directions. 12
IMAGE ANALYSIS METHODS IN AIMS II, UIAIA, AND FTI SYSTEMS 13
AIMS II uses a digital camera with an autofocus microscope to automatically capture images 14
with different resolutions depending on aggregate sizes. This system measures three-dimensions 15
of aggregates to calculate sphericity and aspect ratios from the images of aggregate top surface. 16
It can also calculate the angularity of aggregate of all sizes using a gradient method and quantify 17
texture of coarse aggregates using the wavelet method (7). The influence of shape on angularity 18
is normalized via the division of measurements by the equivalent ellipse dimensions. 19
In AIMS II, the shape properties of coarse aggregates are defined by sphericity, flatness 20
ratio, elongation ratio, and flat & elongated (FE) ratio, whereas the shape properties of fine 21
aggregates are defined by the Form2D parameter instead of sphericity. 22
32
Sphericity s m
l
D D
D (1) 23
Flatness ratio /s mD D (2) 24
Elongation ratio /m lD D (3) 25
FE ratio /m sD D (1) 26
360
0
Form2DR R
R
(5) 27
where Ds is the shortest dimension of the aggregate particle; Dl is the longest dimension of the 28
aggregate particle; Dm is the dimension of the aggregate particle perpendicular to both Ds and Dl; 29
R is the radius of the particle at angle ; and is the incremental difference in the angle. 30
Angularity is defined by Eq.(6). Texture is defined by Eq.(7) at a given level using a 31
wavelet method. 32 360
0
AngularityP EE
EE
R R
R
(6) 33
23
,
1 1
1Texture ,
3
N
i j
i j
D x yN
(7) 34
where RP is the radius of the particle at a directional angle of ; REE is the radius of an 35
equivalent ellipse at the same ; N denotes the level of decomposition; Di,j is a decomposition 36
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 5
function; i takes values of 1, 2, and 3, for the three detailed images of texture, and j is the wavelet 1
coefficient index; (x,y) is the location of the coefficients in transformed domain. 2
In the UIAIA system, shape is described using sphericity, flatness ratio and elongation 3
ratio; angularity is described by the angularity index (AI) method as illustrated in Figure 1. The 4
AI method traces the changes of slope of the 2D profile outline of the particle and uses a 5
weighted average value of its angularity determined from three views (front, top, side images). 6
Eq.(8) shows the calculation of angularity for each image; and Eq.(9) is used to calculate AI. 7
n=1n=2
n=3n=4
Aggregate
8 FIGURE 1 Illustration of an n-sided polygon approximating the outline of a particle (8). 9
10 170
0
Angularity ( )e
e P e
(8) 11
Angularity Area Angularity Area Angularity AreaAI
Area Area Area
front top side
front top side
(9) 12
where e is an angle starting for each interval of 10°; P(e) is the probability that changes in angle 13
from the starting angle e to the next angle of e+10°; Areafront, Areatop, and Areaside represent 14
areas of profiles of the front view, top view and side view, respectively. 15
Surface texture is analyzed using an erosion and dilation technique (9), shown in Figure 2, 16
in which Figure 2(a) is the original aggregate image. Surface irregularity is gradually lost during 17
the erosion-dilation process with the corresponding area lost as a percentage of the area in the 18
original image. The percentage of lost area is defined as texture, defined by Eq.(10), and surface 19
texture (ST) is defined by Eq.(11). 20
1 2
1
Texture= 100%A A
A
(10) 21
Texture Area Texture Area Texture AreaST
Area Area Area
front top side
front top side
(11) 22
where A1 and A2 are areas of aggregates before and after erosion-dilation cycles, respectively. 23
24
25
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 6
(a) Area = A1 (c) Area = A2
Erosion Dilation
(b)
1 FIGURE 2 Illustration of erosion and dilation technique (9). 2
3
In the FTI system, shape properties of aggregates are described by sphericity, flatness ratio, 4
elongation ratio, and FE ratio; angularity and texture are defined using two-dimensional Fourier 5
transform method. 6 2 21 1
0 0
( , ) ( , )N N j xp yq
N N
x y
Z p q z x y e
(4) 7
2 21 1
0 0
( , ) ( , )N N j xp yq
N N
p q
z x y Z p q e
(5) 8
2 2
0 0
1 1
Angularity Factor (AF) , / , /A AN N
p q
a p q a b p q a
(8) 9
2 2
0 0
1 1
Texture Factor (TF) , / , / AFN N
p q
a p q a b p q a
(9) 10
where z(x,y) is the three-dimensional coordinate in xth
row and yth
column on an aggregate 11
surface; Z(p,q) is a coefficient in pth
row and qth
coloumn with the DFT matrix of z(x,y) in 12
frequency domain; j is the imaginary root; a0 is the average value of z(x,y); a and b are the real 13
and imaginary parts of the FFT2 coefficients; N is the size of the z(x,y) matrix; NA is defined as a 14
threshold value, with which a 3D surface is reconstructed using the inverse of FFT2 coefficients 15
that have their frequencies smaller than 2πNA/N in either x-direction or y-direction. These FFT2 16
coefficients are considered to only contribute to angularity, if the average value of the absolute 17
differences between the original surface and reconstructed surface using the inverse is greater 18
than 0.2 mm, which is the spatial spacing discernible by unaided human eyes. Other Fourier 19
coefficients are considered to contribute to texture only. 20
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 7
0.0
5.0x10-4
1.0x10-3
1.5x10-3
2.0x10-3
0 5 10 15 20 25 300.0
2.0x10-5
4.0x10-5
6.0x10-5
8.0x10-5
Blast Furnace Slag
Copper Ore
Dolomite
Glacial Gravel Crushed
Glacial Gravel Rounded
Iron Ore
Limestone
Angula
rity
Fac
tor
Slope is defined as Angularity
Slope is defined as Texture
Tex
ture
Fac
tor
Roughness matrix size (mm2) 1
FIGURE 3 Illustrations of angularity and texture of some 1/2’’ aggregates in the FTI 2
system (10). 3
4 Roughness matrix is the z(x,y) matrix of a rectangular region on aggregate surface with an 5
area ranging from 1.0 mm2 to the maximum value as 25% of an aggregate surface area, based on 6
the sieve size of aggregate being analyzed (10). Figure 3 plots the relationship between AF (or 7
TF) and roughness matrix area for all the seven types of aggregates (hereinafter referred as AF 8
plot or TF plot) for some 1/2’’ aggregates. As shown in Figure 3, the AF and TF values follow a 9
linear relationship in both AF and TF plots as the roughness matrix area increases. Furthermore, 10
aggregates with more angular surfaces tend to have steeper slopes in the AF plot, and aggregates 11
with rougher surfaces tend to show steeper slopes in the TF plot. Therefore, it is reasonable to 12
define the linear relationship (slope) in the AF and TF plots as angularity and texture of an 13
aggregate, respectively. 14
AGGREGATES 15
Table 2 contains some physical properties of the seven types of aggregates: Blast Furnace Slag, 16
Copper Ore, Dolomite, Glacial Gravel Crushed, Glacial Gravel Rounded, Iron Ore, and 17
Limestone. For each type of aggregate, there are thirty 1/2’’ aggregates in total to ensure 18
statistical stability of aggregate samples. Further statistical analysis suggests that a thirty-19
aggregate sample is large enough to provide a normal distribution and to achieve stable statistics 20
for each type. All the seven types of aggregate are manual measured with a vernier caliper first 21
and then analyzed using the FTI system, AIMS II and the first generation UIAIA systems 22
respectively for comparisons. Figure 4 shows the photographs of the 1/2’’ aggregates. As 23
illustrated by the photos, it can be observed that Blast Furnace Slag has very angular and the 24
roughest surfaces while the Glacial Gravel Rounded has the least angular and very smooth 25
surfaces. 26
27
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 8
1
TABLE 2 Aggregate Types and the Physical Properties (10) 2
Aggregate
type Origin Description
LAA
loss
(%)
Bulk dry
SpGr
(g/cm3)
24 hr soak
absorption
(%)
BFS Wayne, MI
Color: 5R 5/0 to 5R 1/0
poorly developed crystalline
structure
43 2.27 3.18
CO
Keweenaw,
MI Mixture of colors, including 5R
9/0, 5BG 8/2, 5R 8/6, and 5R 2/8
fine to coarse grains
19 2.64 0.95
Houghton,
MI 16 2.76 2.12
DLT
Mackinac,
MI
Color: 5R 9/4 to 5R 6/8
medium to coarse crystals 27 2.78 0.52
Monroe,
MI
Color: 5R 6/7
well defined small euhedral
dolomite crystals
45 2.45 4.16
GGC Kent, MI Various colors
With crushed surfaces 17 2.73 0.71
GGR Kent, MI Various colors
Very smooth texture 19 2.68 1.10
IO Marquette,
MI
Color: 10PB 2/2 with 5R 2/1
very fine grained and hard
metamorphic rock
11 -- --
LST
Schoolcraft,
MI
Color: 5R 9/4
very fine subcrystalline texture 25 2.65 0.64
Arenac, MI
Color: 10PB 7/0
very fine crystalline with
abundant frosted quartz sand
grains
42 2.56 2.13
Note: LAA=Los Angeles Abrasion; Bulk dry SpGr=Bulk dry specific gravity; BFS = Blast 3
Furnace Slag; CO = Copper Ore; DLT = Dolomite; GGC = Glacial Gravel Crushed; GGR = 4
Glacial Gravel Rounded; IO = Iron Ore; LST = Limestone. 5
Color description in the Description column is named according to Munsell color system. The 6
Munsell color system is a color system that defines color based on three dimensions: hue, value 7
(lightness), and chroma (color purity). A color can be specified by listing three numbers for hue, 8
value, and chroma. For example, 5R 5/0 means a red of medium lightness and barely saturated, 9
with 5R meaning the color in the middle of the red hue band, 5/ meaning medium value 10
(lightness), and a chroma of 0. 11
12
13
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 9
1
2 (a) Blast Furnace Slag (b) Copper Ore (c) Dolomite 3
4
5 (d) Glacial Gravel Crushed (e) Glacial Gravel Rounded (f) Iron Ore 6
7
8 (g) Limestone 9
FIGURE 4 Photos depicting the 1/2” aggregates (10). 10 11
ANALYSIS RESULTS 12
Figure 5 plots the sphericity distributions of the seven types of 1/2’’ aggregates using manual 13
measurement, AIMS II and FTI systems, respectively. Compared to the sphericity distribution of 14
manual measurement, AIMS II sphericity is distributed in a wider range, whereas FTI sphericity 15
is distributed in a narrower range. According to both manual measurements and FTI results, 16
Blast Furnace Slag, Glacial Gravel Rounded and Glacial Gravel Crushed have aggregates with 17
relatively great values of sphericity; conversely Iron Ore and Limestone have aggregates with 18
smaller values of sphericity than the other types of aggregate. 19
20
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 10
0.4 0.6 0.8 1.00
25
50
75
1000
25
50
75
1000
25
50
75
100
Per
cen
tag
e (%
) FTI
Per
cen
tag
e (%
)
BFS 1/2''
CO 1/2''
DLT 1/2''
GGC 1/2''
GGR 1/2''
IO 1/2''
LST 1/2''AIMS II
Sphericity
Per
cen
tag
e (%
) Manual
1 FIGURE 5 Sphericity distributions of the 1/2’’ aggregates. 2
3
Figure 6 plots the FE ratio distributions of all the 1/2’’ aggregates analyzed using manual 4
measurement, UIAIA, AIMS II, and FTI systems. Compared to manual measurements, Iron Ore 5
is considered with have the greatest values of FE ratio according to both AIMS II and FTI 6
systems, which is consistent to manual measurements. Since Iron Ore was not evaluated with the 7
UIAIA system, Dolomite gives the next highest values of FE ratio by the UIAIA system similar 8
to AIMS II and manual measurements. 9
10
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 11
1 2 3 4 5 6 70
25
50
75
1000
25
50
75
1000
25
50
75
1000
25
50
75
100
FTI
Per
cen
tag
e (%
)
FE ratio
Per
cen
tag
e (%
)
BFS1/2''
CO1/2''
DLT1/2''
GGC1/2''
GGR1/2''
IO1/2''
LST1/2''
AIMS II
UIAIA
Per
cen
tag
e (%
)
Per
cen
tag
e (%
)Manual
1 FIGURE 6 FE ratio distributions of the 1/2’’ aggregates. 2
3
Table 3 tabulates mean values and standard deviations for the seven types of the 4
1/2’’aggregates. According to the FTI ranking, Blast Furnace Slag aggregates are the most 5
angular aggregates with the roughest surfaces, followed by Limestone as the second most 6
angular aggregates with the second smoothest surfaces. Dolomite is the least angular aggregate 7
with the smoothest surface texture. According to AIMS II ranking, Copper Ore is the most 8
angular aggregates with the second roughest surfaces, followed by Blast Furnace Slag as the 9
second most angular aggregates with the roughest surfaces. Glacial Gravel Rounded has the least 10
angular aggregates with the smoothest surfaces. In the UIAIA ranking, Iron Ore aggregates could 11
not be imaged, because of the dark color of Iron Ore aggregates although this would not be a 12
limitation in the current enhanced UIAIA utilizing progressive color cameras and a blue 13
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 12
background to capture digital color images of aggregates. Glacial Gravel Crushed has the most 1
angular aggregates with the roughest surfaces, and Glacial Gravel Rounded aggregates are the 2
least angular ones with the third smoothest surfaces. 3
Since it is not convenient to directly compare roughness rankings of the seven types of 4
aggregates using different morphological descriptors, the angularity and texture rankings for 1/2’’ 5
aggregates are plotted in Figure 7. All the data in this figure are generated from Table 3, simply 6
dividing the mean value of each type of aggregate by the smallest mean value for either 7
angularity or texture among the FTI, AIMS II, and UIAIA angularity and texture, respectively. 8
As shown in Figure 7, both AIMS II, and UIAIA roughness rankings reach an agreement that 9
Glacial Gravel Rounded aggregates are considered as the least angular aggregates with very 10
smooth texture, whereas FTI ranks dolomite as the most angular aggregates with very smooth 11
texture. And Blast Furnace Slag aggregates are very angular aggregates with the roughest texture 12
according to angularity and texture rankings from all three systems. 13
14
TABLE 3 Angularity and Texture using FTI, AIMS II, and UIAIA for 1/2’’ Aggregates 15
1/2'' Aggregates BFS CO DLT GGC GGR IO LST
Angularity
FTI Mean 2.44×10
-4 0.98×10
-4 0.60×10
-4 1.46×10
-4 0.97×10
-4 1.00×10
-4 1.90×10
-4
Standard
deviation 5.18×10
-4 1.14×10
-4 1.13×10
-4 2.51×10
-4 1.11×10
-4 1.04×10
-4 4.54×10
-4
AIMS II Mean 3040.72 3126.44 2864.42 2650.92 1362.78 2932.80 2859.35
Standard
deviation 833.09 604.35 697.83 612.77 624.87 515.95 631.80
UIAIA Mean 367.51 394.23 333.59 458.09 239.89 NA 357.07
Standard
deviation 80.71 77 55.15 108.54 83.19 NA 51.69
Texture
FTI Mean 9.57×10
-6 4.69×10
-6 2.86×10
-6 2.70×10
-6 3.54×10
-6 5.17×10
-6 4.60×10
-6
Standard
deviation 13.70×10
-6 5.68×10
-6 5.18×10
-6 3.37×10
-6 4.32×10
-6 6.47×10
-6 8.38×10
-6
AIMS II Mean 574.74 365.75 259.11 322.72 217.55 243.40 270.21
Standard
deviation 155.81 120.04 113.33 188.26 145.06 112.90 122.26
UIAIA Mean 1.12 1.19 0.92 2.15 1.00 NA 0.91
Standard
deviation 0.39 0.44 0.39 0.98 0.76 NA 0.27
Note: NA = not available. 16
17
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 13
0
1
2
3
4
5
1=Smooth, 5=Rough FTI
AIMSII
UIAIA
An
gu
lari
ty r
ank
ing
1=Least angular, 5=Most angular
0
1
2
3
4
5
BFS CO DLT GGC GGR IO LST
Tex
ture
ran
kin
g
Note: BFS=Blast furnace slag; CO=Copper ore; DLT=Dolomite; GGC=Glacial glacial crushed;
GGR=Glacial gravel rounded; IO=Iron ore; LST=Limestone
1 FIGURE 7 Angularity and texture rankings of 1/2’’ aggregates. 2
3
EVALUATION AND DISCUSSION 4
To further evaluate the three imaging techniques, i.e., AIMS II, UIAIA, and FTI, the following 5
procedure is adopted. (i) comparison with manual measurements for shape characteristics; (ii) 6
Analysis of variance between measurements using different methods (ANOVA) test for shape, 7
angularity and texture characteristics; and (iii) comparison to visual rankings for both angularity 8
and texture. 9
Comparison with Manual Measurements 10
The relationship between manual measurements and shape characteristics quantified using 11
aggregate imaging techniques could be regressed by linear relationships. As shown in Figure 8, 12
the linear relationships for sphericity and FE ratio indicate that: (i) both AIMS II and FTI could 13
accurately quantify the sphericity of Limestone aggregates with consistent values with those of 14
manual measurements; (ii) both UIAIA and FTI quantify the FE ratio of Glacial Gravel Rounded 15
aggregates with more consistent values than AIMS II. 16
17
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 14
0.4 0.5 0.6 0.7 0.8 0.9 1.00.4
0.5
0.6
0.7
0.8
0.9
1.0
LST Sphericity
AIMS II
FTI
Sp
her
icit
y u
sin
g a
gg
reg
ate
imag
ing
tec
hn
iqu
es (
y)
Manual Sphericity (x)
y = 0.8956x + 0.1629
R2 = 0.8162
y = 0.8139x + 0.1124
R2 = 0.9623
(a)
1
1.0 1.5 2.0 2.5 3.0 3.51.0
1.5
2.0
2.5
3.0
GGR FE ratio
UIAIA
AIMS II
FTI
FE
rat
io u
sin
g a
gg
reg
ate
imag
ing
tec
hn
iqu
es (
y)
Manual FE ratio (x)
y = 1.1002x - 0.1935
R2 = 0.9711
y = 1.0327x + 0.0337
R2 = 0.9789
y = 1.1403x - 0.6315
R2
= 0.9475
(b)
2 FIGURE 8 Relationship between manual measurements and shape properties acquired 3
from image analysis techniques with regard to sphericity and FE ratio for Limestone 1/2’’ 4
aggregates. 5 6
ANOVA Test 7
ANOVA is used to compare several groups of observations, which are independent and possibly 8
having a different mean for each group. A test of importance is whether or not all the means are 9
statistically equal. The basic assumptions of the ANOVA test are as follows: (i) sample data are 10
collected using a simple random sampling method; (ii) sample groups are independent from each 11
other; (iii) equal variance across groups; (iv) each group follows a normal distribution. 12
The ANOVA tests performed in this section were used to compare sphericity and FE ratio of all 13
the 1/2’’ aggregates using the statistical software JMP (version 9), in order to determine whether 14
the differences among the three methods were significant or not for shape characteristics. 15
Angularity and texture characteristics cannot be directly compared to each other using ANOVA 16
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 15
in JMP, because the descriptors for angularity and texture are defined in different terms. The 1
ANOVA test can only detect whether there are significant differences between results using 2
different methods (i.e., FTI, AIMS II, UIAIA, and manual measurements) or not. Once there is a 3
difference between means of these groups, Tukey’s HSD test is conducted via JMP to determine 4
which mean is significantly different from the others. 5
Table 4 presents the method groups for ANOVA test, and Table 5 tabulates ANOVA test 6
results. The ANOVA test for sphericity is as follows: H0: 1s=2s=3s, vs Ha: at least one mean 7
differs; where 1s, 2s, and 3s are the means of sphericity using the FTI system, AIMS II system, 8
and manual measurements, respectively. Set =0.05, reject H0 if pvalue < , and then Tukey’s 9
method will be performed to conduct multiple comparison test to determine the mean 10
significantly different from the other means. And the ANOVA test for FE ratio is as follows: H0: 11
1fe=2fe=3fe=4fe, vs Ha: at least one mean differs; where 1fe, 2fe, 3fe, and 4fe are the means 12
of FE ratio using the FTI, AIMS II, UIAIA, and manual measurements, respectively. Set =0.05, 13
reject H0 if pvalue < , and then Tukey’s method will be performed to conduct multiple 14
comparison test to determine the mean significantly different from the other means. 15
16
TABLE 4 Analysis Method Groups for ANOVA Test 17
Sphericity FE ratio
Method Group Method Group
FTI 1 FTI 1
AIMS II 2 AIMS II 2
Manual measurement 3 UIAIA 3
Manual measurement 4
18
TABLE 5 ANOVA Test Summary for Sphericity and FE ratio of 1/2'' Aggregates 19
1/2''
Aggregates Sphericity FE ratio
BFS No significant difference among three
means
No significant difference among the
four means
CO No significant difference among three
means
No significant difference among the
four means
DLT No significant difference between
three means
The UIAIA mean is significantly
smaller than the other three means
GGC The AIMS II mean is significantly
greater than the other two means.
The AIMS II mean is significantly
smaller than the other three means.
GGR The FTI mean is the greatest one, and
the AIMS II mean is the smallest one
The AIMS II mean is significantly
smaller than the other three means.
IO No significant difference among three
means
No significant difference among the
four means
LST The FTI mean is very close to the
manual measurement mean
The FTI mean is very close to the
manual measurement mean and the
UIAIA mean
20
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 16
Comparison with Visual Ranking of Angularity and Texture 1
Angularity and texture of all the 1/2’’ aggregates were visually ranked by three operators. Mean 2
values of the same aggregate were calculated for visually evaluated rankings from the three 3
operators to get the final visual rankings of angularity and texture respectively. To compare the 4
angularity and texture rankings acquired from three imaging techniques with visual rankings, the 5
Spearman correlation coefficient was calculated. The Spearman correlation coefficient ρ is 6
defined by the following equation: 7
1
2 2
1 1
n
i i
i
n n
i i
i i
x x y y
x x y y
(10) 8
where n is the sample size; x and y represent the visual ranking results and rankings of image 9
analysis results, respectively. The Spearman correlation coefficient has a value ranging from -1 10
to 1. A Spearman correlation of zero indicates that there is no relevance between X and Y (x and 11
y are ranks of X and Y, respectively), whereas a Spearman correlation close to 1 shows a 12
monotonically related relationship between X and Y, even if the relationship is not linear. 13
Table 6 presents the Spearman correlation coefficients of angularity and texture by 14
comparing the angularity and texture ranking results using the three imaging techniques to visual 15
ranking result. Spearman correlation coefficients suggest that: (i) all three imaging techniques 16
can be used for the quantification of angularity and texture for all the 1/2’’ aggregates; 17
furthermore, FTI angularity ranking has the closer correlation to visual ranking than either 18
UIAIA or AIMS II; (ii) AIMS II ranking for texture has better agreement with visual ranking 19
than the other two. 20
21
TABLE 6 Spearman Correlation Coefficients of Angularity and Texture 22
Analysis method UIAIA AIMS II FTI
Angularity 0.257 0.771 0.860
Texture 0.200 0.657 0.371
23
CONCLUSION 24
Imaging techniques can rapidly measure aggregate shape, angularity and texture characteristics. 25
Even though that may require expensive equipment, the unit costs of incremental tests are low. 26
Further, imaging techniques can provide automatic quantifications of morphological 27
characteristics with accurate and reliable results. The following issues should be taken into 28
consideration for further improvements. 29
(i) Imaging techniques utilize various analysis methods, which may produce results 30
sometimes incomparable to each other. For instance, some imaging techniques analyze 2D 31
images while others utilize 2D projections to reconstruct 3D surfaces of aggregates. 32
(ii) Some imaging techniques could only analyze coarse aggregates; whereas the other 33
techniques could analyze aggregates of wide size ranges, including both coarse and fine 34
aggregates. Consequently, it is difficult to evaluate the merits of different imaging techniques. 35
(iii) Measurements using imaging techniques require instrument setups and well-trained 36
operators. 37
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 17
After the evaluation of three imaging techniques, i.e., UIAIA, AIMS II, and FTI, the 1
following conclusions can be drawn based on aggregate imaging procedures and quantification 2
results: 3
(i) ANOVA test results of the 1/2’’ aggregates demonstrate that the shape characteristics, 4
such as sphericity, obtained through the FTI system is in good agreement with those calculated 5
from manual measurements and AIMS II for most aggregates. The ANOVA test of FE ratio 6
show that FE ratio results obtained from the three aggregate imaging systems are generally in 7
good agreement with FE ratios calculated from manual measurements. 8
(ii) Comparison of angularity and texture results show that the three aggregate imaging 9
systems could analyze 1/2’’ aggregates with reasonable angularity and texture rankings. It is 10
found that Blast Furnace Slag has greater values of both angularity and texture than the other 11
aggregates, followed by Limestone and Iron Ore. Both Dolomite and Glacial Gravel Rounded 12
are shown by all the three aggregate imaging techniques to have very smooth surface texture. 13
However, there is some difference between the analyzed results using the three imaging 14
systems. For example, Glacial Gravel Crushed aggregates are quantified as aggregates with 15
somewhat smooth surface texture using both AIMS II and FTI, but GGC is considered to have 16
the roughest surface texture according to UIAIA results; conversely Dolomite aggregates are 17
considered the least angular ones while AIMS II and UIAIA rank them as the second least 18
angular ones. Possible reason for the diversities might lay in different definitions of angularity 19
and surface texture. When angularity and texture are significantly different, these systems may 20
be sensitive enough to differentiate them; when angularity and texture are not significantly 21
different, these systems may rank them differently. As opposed to the black and white cameras 22
and black background, the second generation enhanced version of the UIAIA is now equipped 23
with higher resolution progressive scan color cameras and a blue background to capture high 24
resolution color images (0.056 mm/pixel) and utilize an advanced color thresholding scheme for 25
improved texture quantification. 26
Based on the evaluation results, which only considered results from the first generation 27
UIAIA system, the following analysis methods are recommended: 28
(i) Both AIMS II and FTI systems are capable to reasonably quantify shape properties of 29
the seven types of 1/2’’ aggregates in terms of sphericity, flatness ratio, elongation ratio, and 30
flatness & elongation ratio. 31
(ii) It is recommended to use either AIMS II or FTI to analyze aggregates for angularity, as 32
both AIMS II and FTI system are capable to differentiate aggregates of all kinds of colors from 33
different origins with results in wide numerical ranges for both angularity and texture, 34
respectively. AIMS II is recommended to be adopted to quantify surface texture. 35
ACKNOWLEDGEMENT 36
The research reported herein was funded under the NCHRP Project 4-34. The authors would 37
like to extend their sincere gratitude to the project panel, including Ervin L. Dukatz, Jr., Edward 38
T. Harrigan, William G. Eager, Julie E. Kliewer, Jorge A. Prozzi, Alan C. Robords, William H. 39
Skerritt, G. P. Jayaprakash, and Richard C. Meininger. Great appreciations are also extended to 40
Anbo Wang, Evan M. Lally, Cris Harris, Yang Lu, Yu Zhou, Ashley Stanford, Yinning Zhang, 41
and Yue Hou for their dedicated contributions to develop the FTI system, compile the MATLAB 42
program, and analyze aggregates. The authors would also like to acknowledge the generous help 43
and strong support from Richard C. Meininger for coordinating the use of the AIMS II at FHWA 44
and the help from Yuanjie Xiao and Maziar Moaveni for analyzing aggregates using the first 45
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Sun, Tutumluer, Druta 18
generation UIAIA system. 1
REFERENCES 2
1. Chowdhury, A., J. W. Button, V. Kohale, and D. W. Jahn. Evaluation of Superpave Fine 3
Aggregate Angularity Specification. ICAP Report 201-1. International Center for Aggregate 4
Research, Texas Transportation Institute, Texas A&M University System, College Station, 5
Texas, 2001. 6
2. Masad, E. Review of imaging techniques for characterizing the shape of aggregates used in 7
asphalt mixes. Proceeding of the 9th annual symposium of the international center for 8
aggregate research (ICAR), Austin, TX, 2001. 9
3. Al-Rousan, T., E. Masad, E. Tutumluer, and T. Pan. Evaluation of image analysis 10
techniques for quantifying aggregate shape characteristics. Construction and Building 11
Materials, Vol. 21, 2006, pp. 978-990. 12
4. Masad, E., D. Olcott, T. White, and L. Tashman. Correlation of fine aggregate imaging 13
shape indices with asphalt mixture performance. In Transportation Research Record: 14
Journal of the Transportation Research Board, No. 1757, Transportation Research Board of 15
the National Academies, Washington, D.C., 2001, pp. 148-156. 16
5. Prowell, B. D., J. Zhang, and E. R. Brown. Aggregate properties and the performance of 17
Superpave-designed hot mix asphalt. NCHRP 539. Transportation Research Board, 18
National Research Council, Washington, D.C., 2005. 19
6. Wang, L., W. Sun, E. M. Lally, A. Wang, C. Druta, E. Tutumluer. Application of LADAR in 20
the analysis of aggregate characteristics. NCHRP Report 724. Transportation Research 21
Board, National Research Council, Washington, D.C., 2012. 22
7. Masad, E., T. Al-Rousan, J. Button., D. Little, and E. Tutumluer. Testing methods for 23
characterizing aggregate shape, texture, and angularity. NCHRP Report 555. 24
Transportation Research Board, National Research Council, Washington, D.C., 2007. 25
8. Rao, C., E. Tutumluer, and I. T. Kim. Quantification of coarse aggregate angularity based 26
on image analysis. In Transportation Research Record: Journal of the Transportation 27
Research Board, No. 1787, Transportation Research Board of the National Academies, 28
Washington, D.C., 2002, pp. 117-124. 29
9. Masad, E., and J. Button. Unified imaging approach for measuring aggregate angularity and 30
texture. Journal of Computer-Aided Civil and Infrastructure Engineering, Vol. 15, No.4, 31
2000, pp. 173-280. 32
10. Sun, W., L. Wang, and E. Tutumluer. Image analysis technique for aggregate morphology 33
analysis with two-dimensional Fourier transform method. In Transportation Research 34
Record: Journal of the Transportation Research Board, No. 2267, Transportation Research 35
Board of the National Academies, Washington, D.C., 2012, pp. 3-13. 36
37
TRB 2013 Annual Meeting Paper revised from original submittal.