Roundness and Sphericity of Soil Particles in Assemblies

by Computational Geometry
Junxing Zheng, S.M.ASCE 1; and Roman D. Hryciw, M.ASCE 2

Abstract: The use of computational geometry methods for determining soil roundness (R) and sphericity (S) were evaluated and extended to
particles segmented from images of three-dimensional particle assemblies. Two Adobe Photoshop lasso tools were used to delineate particles
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

with full projections from the assemblies. Results were in excellent agreement with values published in traditional roundness and sphericity
charts, thus confirming that the computational method can replace the much slower and less objective chart methods. Complete volume-based
distributions of particle roundness and sphericity were presented for three soils with vastly different particle shapes. Values of R and S
obtained from images of three-dimensional assemblies were almost indistinguishable from values obtained using images of detached particles
showing their largest projected areas. Mean R values were also computed for 10 different soils of various geologic origins. As expected,
crushed sands exhibited the smallest mean values of R whereas alluvial and glacio-fluvial soils showed the largest values. DOI: 10.1061/
(ASCE)CP.1943-5487.0000578. © 2016 American Society of Civil Engineers.

Introduction By contrast to the ease with which S is computed, the most popu-
lar and commonly used measure of roundness (R) developed by
Soil particle form (shape) and roundness are intrinsic grain-level Wadell (1932, 1933, 1935), does not lend itself to simple image
characteristics that, along with state parameters, control assembly analysis. Wadell defined R as the ratio of the average radius of
level soil properties such as minimum and maximum void ratio, curvature of a particle’s corners to the radius of the largest possible
strength, compressibility, and shear wave velocity (Santamarina inscribed circle. In Wadell’s original procedure, the particle outline
and Cho 2004; Cho et al. 2006; Bareither et al. 2008; Chapuis was traced and enlarged to a size of 7.0 cm (2.7 in.). Each corner
2012; Shin and Santamarina 2013; Cabalar et al. 2013). However, along the outline was then compared to a set of circle templates
the difficulties involved in determining particle form and roundness in search of the largest circle that would fit it. Obviously, consider-
have impeded their application in practice. On the computational able effort was required to evaluate even a small number of particles;
front, discrete element methods (DEMs) have shown great potential a statistically valid sampling was simply unachievable. Nevertheless,
to simulate soil assembly behavior. However, in DEMs, soil Wadell’s manual procedure is still occasionally used today when rel-
particles are generally idealized as perfect spheres, and thus form atively accurate R values are sought for a small number of particles
and roundness effects must be accounted for in roundabout ways, (Moroto and Ishii 1990; Rouse et al. 2008; Yang and Wei 2012).
for example, by artificially restricting particle rotations. Recently, To accelerate the estimation of R, charts containing particle
DEM researchers have been developing the computational tools to silhouettes of known Wadell R were developed by Krumbein
simulate more realistic and complex particles. This will require (1941), Krumbein and Sloss (1951), and Power (1953). Users could
more accurate and statistically valid quantification of the form
quickly estimate R by observing their particles under a microscope
and roundness of real soil particles.
and comparing them to the particles in the standard charts.
Particle form is commonly quantified by a sphericity (S) value.
Although more rapid, this approach is subjective and yields incon-
Various definitions have been proposed for this parameter.
Almost all of the definitions are based on particle dimensions sistent results between evaluators. Nevertheless, the chart methods
measured in images of two-dimensional (2D) projections of the are widely used in a variety of disciplines including geotechnical
particles (Tickell 1931; Wadell 1935; Krumbein and Sloss 1951; engineering, soil science, agriculture, powder engineering, pave-
Santamarina and Cho 2004; Altuhafi et al. 2013). Computation ment engineering, and geology (Eisma 1965; Youd 1973; Frossard
of S by the various definitions can be readily automated by digital 1979; Sladen et al. 1985; Vepraskas and Cassel 1987; Sagga et al.
image analysis techniques (Kuo et al. 1996; Kuo and Freeman 1993; Vallejo and Zhou 1995; Santamarina and Cho 2004; Mitchell
2000; Rao and Tutumluer 2000; Fletcher et al. 2003; Kumara et al. and Soga 2005; Masad et al. 2007; Cho et al. 2006; Mehring and
2012; Altuhafi et al. 2013 and others as surveyed by Hryciw McBride 2007; Bareither et al. 2008; Altuhafi et al. 2013; Chapuis
et al. 2014). 2012; Shin and Santamarina 2013; Cabalar et al. 2013; Kandasami
and Murthy 2014; Oh et al. 2014).
1
Graduate Student and Research Assistant, Dept. of Civil and Environ- Some researchers have developed alternative roundness descrip-
mental Engineering, Univ. of Michigan, 2340 GG Brown, Ann Arbor, MI tors whose parameters could be obtained through digital image
48109-2125. E-mail: junxing@umich.edu analysis. These have included Fourier coefficients (Bowman et al.
2
Professor, Dept. of Civil and Environmental Engineering, Univ. of Mi- 2001; Wettimuny and Penumadu 2004; Wang et al. 2005; Mollon
chigan, 2366 GG Brown, Ann Arbor, MI 48109-2125 (corresponding
and Zhao 2012a, b, 2014), an angularity index (Sukumaran and
author). E-mail: romanh@umich.edu
Note. This manuscript was submitted on March 4, 2015; approved on
Ashmawy 2001; Tutumluer and Pan 2008), and a fractal dimension
December 22, 2015; published online on March 24, 2016. Discussion per- (Arasan et al. 2011; Vallejo 1995; Vallejo and Zhou 1995). However,
iod open until August 24, 2016; separate discussions must be submitted for these alternate definitions of roundness have not prevailed over
individual papers. This paper is part of the Journal of Computing in Civil Wadell’s because of the latter’s long history and deserved
Engineering, © ASCE, ISSN 0887-3801. popularity.

© ASCE 04016021-1 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 In light of the continuing usage of Wadell’s R in research and validation. The corners on the soil particles are identified through
practice, Zheng and Hryciw (2015) proposed a computational computational geometry. Four main steps are required to compute
geometry algorithm to automate the computation of Wadell’s R. Wadell’s R from a binary soil particle image.
Their method was tested on two particles from Wadell’s original In Step 1 the maximum inscribed circle is found. The particle
1935 paper and an additional 20 reference particles from the perimeter in Fig. 1(b) is obtained by tracing the outline of the
roundness chart of Krumbein and Sloss (1951). Excellent agree- binary particle silhouette in Fig. 1(a). For each point inside the
ment was found between the computed R values and the chart soil particle, the minimum distance D to the particle perimeter
values. The present paper further tests the roundness algorithm is computed, which produces a Euclidean Distance Map shown
on another 93 reference particles from published standard charts in Fig. 1(c). The point C having maximum D is the center of
[81 from Krumbein (1941) and 12 from Powers (1953)]. The paper the maximum inscribed circle and D is its radius as shown in
then extends the algorithm to particles selected and delineated Fig. 1(d).
from 13 images of three-dimensional (3D) soil assemblies having In Step 2 the roughness on the particle outline is removed. A
various geologic origins. locally weighted scatter plot smoothing (LOESS) technique
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

(Cleveland and Devlin 1988) and K-fold cross-validation technique

(Efron and Tibshirani 1993) are adopted to estimate a mean surface.
Computational Geometry Technique for Wadell The particle perimeter is discretized by polar coordinates (θ, ρ) in
Roundness Fig. 2(a) and plotted in Fig. 2(b). The LOESS and K-fold cross
validation are used to filter out the roughness resulting in a
As previously described, the Wadell procedure requires identifica-
tion of the particle corners and fitting of circles to them. The pro- smoothed mean surface, also shown in Fig. 2(b). The two lines
cedure is difficult to automate because the number of corners and are indistinguishable unless the plot is highly magnified as shown
their curvature vary considerably from particle to particle. Human by expansion of areas (1) and (2). Then, the mean surface is replot-
evaluators readily identify the corners and easily fit corresponding ted to generate the new smoothed soil particle outline in Fig. 2(c). It
corner circles to them using intuition, experience, and judgment. is noted that the LOESS and K-fold cross validation could also be
The other previously mentioned roundness descriptors (Fourier used to quantify the roughness of soil particles (Zheng and Hryciw
coefficients, angularity index, and a fractal dimension) avoid the 2015). In Step 3 the corners on the particle perimeter are identified.
complex corner identification process by quantifying the curvature An arbitrary curved segment on the perimeter, such as AB in
over the entire particle perimeter without placing emphasis on Fig. 3(a), could be approximated by the chord AB in Fig. 3(b).
corners. Secondly, digital noise and actual roughness along the The chord must be short enough so that the divergence of the curve
particle perimeter must be filtered out prior to computing round- from the chord approximating it does not exceed a predetermined
ness. The noise and roughness could be easily filtered out by a hu- critical value given by δ 0. Figs. 3(b and c) show the chord approx-
man operator but is challenging for a computer. Some researchers imations of the curve AB using a larger and a smaller value of δ 0
including Sukumaran and Ashmawy (2001) and Tutumluer and Pan respectively. Clearly, a large δ 0 will result in missed small corners.
(2008) sought to remove the roughness by discretizing the outline The physical meaning and determination of the appropriate δ0 value
of a particle into an N-sided polygon. The selected N value de facto will be discussed shortly. Assuming an appropriate δ 0 is used, the
served as the cutoff between what would be defined as angularity soil particle in Fig. 3(a) is approximated by line segments connect-
and what would be roughness. Sukumaran and Ashmawy (2001) ing key points in Fig. 3(d).
suggested that N ¼ 40 would be appropriate while Tutumluer and The key points could be either on corners or not, as shown in
Pan (2008) believed N ¼ 24 to be adequate. In reality, a single N Fig. 3(d). Only the key points that lie on the corners, hereafter
value cannot be applied to all particles; N should vary depending on termed corner points, will be needed to compute roundness.
each particle’s individual angularity and roughness. Another The straightforward method to identify whether a key point [such
limitation of this N-approach is that small sharp corners can be as point E in Fig. 3(d)] is a corner point is as follows. The centroid
missed. As such, this approach is problematic for very angular O of the soil particle is located as shown in Fig. 3(d). The two
particles. nearest key points on each side of E (points C and D) are connected
The issues described previously were addressed and overcome and a line from O through E is constructed. The point F lies at the
by Zheng and Hryciw (2015). The noise and roughness are re- intersection of CD and the line from O through E. Then, if
moved from particle perimeters using two statistical techniques: OE > OF, point E is a corner point, whereas if OE < OF it is
locally weighted scatter plot smoothing (LOESS), and K-fold cross not. This process is repeated for all key points. Only the points

D (pixels)

rin =642

C Cin

(a) (b) (c) (d)

Fig. 1. Finding the maximum inscribed circle: (a) the input binary particle image; (b) extracted particle outline; (c) Euclidean distance map; (d) com-
puted maximum inscribed circle

© ASCE 04016021-2 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 550
(1)
500
(2)
Original 450 (1)
particle

400
(a)
350 (2)

(1)
Smoothed 300
particle
Original Smoothed
250
0 45 90 135 180 225 270 315 360
(c) (b)
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

(degrees)

Fig. 2. Removing roughness from a particle perimeter: (a) discretization of particle outline using polar coordinates; (b) estimation of particle mean
surface; (c) smoothed particle outline

B
C
F Corner 1
E
A
B (b) D Corner 2
O Corner 4
B
A Corner 3

(c)
Key points A Corner points
(a) (d) (e)

Fig. 3. Identifying the corners on the soil particle perimeter: (a) two key point on a particle outline; (b) chord between key points using a large δ 0 ;
(c) chord between key points using a small δ 0 ; (d) soil particle approximated by line segments connecting key points and the geometry for determining
if points are corner points; (e) identified corner points

identified as corner points, 43 in all, are plotted in Fig. 3(e). They consideration. In the next loop, shown in Fig. 4(f), points 9 to 43 are
reveal four particle corners. used in the search for the circle that will fit the second corner. The
In Step 4 circles are fitted to the corner points as follows. second proper circle C2 for which d ≈ r is finally found when the
Initially, a large circle is fitted to all 43 corner points as shown first point is 10 and the last point moves to 19 in Fig. 4(g).
in Fig. 4(a). The center of the circle is C and its radius is r. After all the loops are completed and the corner circles are
The minimum distance from C to any point on the particle found, a special situation must be checked. The first fitted circle
boundary is d. If d is smaller than r, the circle cannot be tangent and the last fitted circle may coincide. This happens when
to the particle boundary. As such, it is not an acceptable circle. If point 1 is on a corner, in which case one of the circles containing
this happens, as it does in Fig. 4(a), the end point 43 is eliminated point 1 is redundant and should be ignored. The final corner circle
and point 42 becomes the new last point. Points 1 to 42 are now fitting results are shown in Fig. 4(h). As expected, four circles were
fitted with a new circle. The recomputed d and r values are com- found for the four corners. The radii of these circles are then used to
pared. If d is still smaller than r, point 42 is eliminated and 41 be- compute the average corner radius.
comes the new last point. The process continues until a circle is In the previous procedure, the satisfactory circles are found by
found satisfying d ≈ r or until only three points are left. d ≈ r. However, it is rare that d is exactly the same as r due to
Fig. 4(b) shows the last point having moved down to point 22 computational round-off. In fact, Zheng and Hryciw (2015) found
but still no satisfactory circle was found. Fig. 4(c) shows that that 0.98 ≤ d=r ≤ 1.00 is accurate enough for determining particle
the first satisfactory circle was found when the last point moved roundness.
all the way to point 7. The circle C1 thus defines the curvature The last issue is how to determine an appropriate δ 0 . The value
of the first corner. Because points 1 to 7 have now been used to of δ 0 effectively establishes the threshold between corners and non-
define a corner, in the next loop, the first point moves to point corners as shown in Figs. 3(b and c). Therefore, δ 0 should be set as
8 and the last point returns to point 43 as shown in Fig. 4(d). small as possible to capture all the corners. However, once δ 0 is
The previously described process continues with the last point smaller than a threshold value by which all the corners are captured,
successively moving downward in search of a circle whose d ≈ r. a further decrease in δ 0 will not improve the results but merely
In this loop the last point moved all the way to point 10 without increase computational effort. Zheng and Hryciw (2015) found that
finding an acceptable circle. Thus, the loop ends without an accept- a δ 0 equal to 0.03% of the diameter of the smallest circle that will
able circle having been found as shown in Fig. 4(e). Because point circumscribe the particle will adequately define the corners of even
8 was not part of any acceptable circle it is eliminated from further the most angular particles. To facilitate digital analysis they

© ASCE 04016021-3 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 43 43

d d C
r 1 1 r 1
C d d1 r1
r
22 d C r C1
C
7
8
(a) (b) (c) (d) 36
40
r4 d Corner
43 C4
4
circle 4
30
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

d C Corner d3
r3
circle 3 C3 r1 1
r d1 r 1 d1 r1 d1 r1
C1 r 24 C1 d Corner
d C1 C1 1
circle 1
C r2
d2 r2 C2 7
C2 d2
8 9
19 19
10 10 Corner 10
(e) (f) (g) (h) circle 2

Fig. 4. (a–h) Corner circle fitting process as described in the paper

quantified the diameter by the number of pixels per circumscribing even more comprehensively by comparing its results to roundness
circle diameter (PCD). For the particle shown in Figs. 1–4, PCD ¼ charts proposed by Krumbein (1941) and Powers (1953). The
998 pixels as shown in Fig. 5. Therefore, δ 0 must be at least 0.3 Krumbein chart contains 81 reference particles redrawn from peb-
pixels. Another important finding is that the spatial resolution must bles and manually assessed by Wadell’s method. As shown in
be set to capture a particle at a PCD of 200 or more. Otherwise, the Fig. 6, the 81 reference particles were binned by Krumbein into
resolution will not be sufficient to delineate the small corners of nine groups having R increments of 0.1. The authors determined
very angular particles. Using these criteria, the roundness of the the R values of all 81 particles using computational geometry.
particle shown in Fig. 5 was found to be 0.40. The spatial resolution was set to have a PCD of approximately
1,200 pixels for all of the particles and δ 0 was set to 0.3 for all
computations. The resulting corner circles and maximum inscribed
Comparison of the Computational Geometry circles are plotted in Fig. 6. The computed R values are shown with
Technique Results with Chart Values of Roundness two significant figures over each of the particles in Fig. 6. They are
The described computational geometry algorithm was previously in excellent agreement with the one significant figure values pro-
implemented on 22 particles: 2 particles from Wadell’s original vided by Krumbein (1941) at the bottom of each of the nine groups.
1935 paper and 20 from the Krumbein-Sloss (1951) chart by Zheng When the computational results are rounded to one significant fig-
and Hryciw (2015). They found excellent agreement between the ure, perfect agreement is found for 79 of the 81 particles. The only
computed and chart values. In this paper, the algorithm is evaluated exceptions are two very well rounded particles that had R ¼ 0.97,
which rounds to 1.0.
It is clear that the computational method furnishes a precise
computation of R whereas the chart methods provide only estimates
of it. Indeed, the successful development of the former eliminates
r1 = 188 the need for the latter.
Another widely used chart for estimating particle sphericity and
roundness was provided by Powers (1953). He separated particles
rin = 642 having Wadell R values from 0.12 to 1.00 into six roundness
r2 = 199 classes as shown in Fig. 7. The ratio of the upper limit to the lower
limit of R in every class is 0.7. Each roundness range is illustrated
with two particles: one having high S and one with low S. The R
r4 = 336 values of the 12 soil particles were determined using the computa-
r3 = 304
tional methods described in this paper with δ0 set to 0.3. The results
are shown in Fig. 7. The computed R values of the two particles in
each class are very close to the upper and lower limit in each
class. The particles having high S displayed the upper R value
PCD = 998 = 0.3 whereas the particle having low S displayed the lower R value
in each range. The computed values agree remarkably well with
the values reported by Powers (1953). As such, the authors again
Fig. 5. Results of the computational geometry analysis to compute conclude that the computational geometry method proposed in this
roundness
paper can replace the imprecise and subjective chart method.

© ASCE 04016021-4 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 PCD = 1218 PCD = 1259 PCD = 1243 PCD = 1157 PCD = 1050
PCD = 1072
R = 0.12 R = 0.10 R = 0.09 R = 0.24 PCD = 1218 R = 0.27
R = 0.18 PCD = 1153 PCD = 1162
R = 0.17
R = 0.28 R = 0.31

PCD = 1186
PCD = 1153 PCD = 1236 PCD = 1201
PCD = 1242 R = 0.33
R = 0.12 R = 0.10 R = 0.20 PCD = 1162 PCD = 1200
R = 0.12 PCD = 1233 R = 0.23 PCD = 1040
R = 0.32
R = 0.22 R = 0.34
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

PCD = 1100 PCD = 1137

PCD = 1020 PCD = 1245 PCD = 1029 R = 0.30
PCD = 1231 PCD = 1222 PCD = 1231 R = 0.29
R = 0.10 R = 0.16 R = 0.30
R = 0.08 R = 0.12 R = 0.22
PCD = 1171
R = 0.22

0.1 0.2 0.3

PCD = 1117
PCD = 1230 PCD = 1185 PCD = 1178
PCD = R = 0.51 PCD = 1167
PCD = 1123 R = 0.40 R = 0.51 PCD = 1190 R = 0.59
PCD = 1281 1127 R = 0.58
R = 0.37 R = 0.39 R = 0.53 R = 0.61

PCD = 1302
PCD = PCD = 1261 PCD = PCD = 990 PCD = 1221
PCD = 1237 PCD = 1146 PCD = 1160 R = 0.62
1275 R = 0.47 1270 R = 0.60 R = 0.63
R = 0.39 R = 0.42 R = 0.41
R = 0.49 R = 0.51

PCD = 1116 PCD = 1212 PCD = 1126

PCD = 1178 PCD = 1161 PCD = 1136 PCD = 1186 PCD = 1320 R = 0.61 R = 0.61 R = 0.60
R = 0.38 R = 0.41 R = 0.37 R = 0.49 R = 0.51
PCD = 1116
R = 0.49

0.4 0.5 0.6

PCD = 1145
PCD = 1041 PCD = 1136 PCD = 1187 PCD = 1188
PCD = 1244 PCD = 1200 R = 0.76 PCD = 1078
PCD = 1245 R = 0.83 R = 0.90 R = 0.88 R = 0.93
R = 0.72 R = 0.73 R = 0.79
R = 0.70

PCD = 1183 PCD = 1225 PCD = 1134

PCD = 1137 PCD = 1196 PCD = 1137
PCD = 1214 PCD = 1256 PCD = 1284 R = 0.97 R = 0.92 R = 0.92
R = 0.80 R = 0.83 R = 0.79
R = 0.72 R = 0.73 R = 0.67

PCD = 1069 PCD = 1327

PCD = 1202 PCD = 1108 PCD = 1248 PCD = 1178 PCD = 1265 R = 0.92 R = 0.86 PCD = 1107
R = 0.69 R = 0.73 PCD = 1242 R = 0.81 R = 0.84 R = 0.78 R = 0.97
R = 0.68

0.7 0.8 0.9

Fig. 6. Comparison to results reported by Krumbein (1941) (δ0 ¼ 0.3)

© ASCE 04016021-5 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 High Sphericity
PCD = 1076.7 PCD = 932.1
PCD = 1097.7 R = 0.25 R = 0.33
R = 0.16 S = 0.69, 0.83 S = 0.75, 0.87
S = 0.81, 0.90 0.72, 0.86 0.77, 0.87
0.81, 0.86 0.82 0.86
0.98

PCD = PCD =
Low Sphericity 1182.0 PCD =
1363.1
R = 0.12 R = 0.18 1210.4
S= S = 0.41 R = 0.26
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

0.58, 0.76 0.63 S=

0.55, 0.80 0.41 0.58, 0.76
0.59 0.76 0.54, 0.79
0.47
0.54

Very angular Angular Subangular

0.12 – 0.17 0.17 – 0.25 0.25 – 0.35
High Sphericity

PCD = 969.7 PCD = 943.7 PCD = 1038.4

R = 0.47 R = 0.96
R = 0.66
S = 0.82, 0.90 S = 0.91, 0.96
S = 0.90, 0.95
0.79, 0.92 0.92, 0.94
0.89, 0.94
0.87 0.93
0.94

PCD =
Low Sphericity

PCD = 1285.3
1331.3 PCD = 1154.6
R = 0.36 R = 0.52 R = 0.72
S= S= S=
0.49, 0.70 0.61, 0.78 0.57, 0.75
0.47, 0.83 0.60, 0.89 0.54, 0.89
0.50 0.50 0.54

Subrounded Rounded Well rounded

0.35 – 0.49 0.49 – 0.70 0.70 – 1.00

Fig. 7. Comparison of computational method to estimtes of angularity by Powers (1953)

Application of the Algorithm to Particle Assemblies three-dimensional contacting assemblies. The Vision Cone Pen-
etrometer (VisCPT) developed by Raschke and Hryciw (1997),
Provided binary particle images, the computational algorithm Ghalib et al. (2000), Hryciw and Ohm (2013), and Zheng and
can directly extract a particle’s outline and compute Wadell’s R. Hryciw (2015) captures images in situ without taking soil
Therefore, it could be readily integrated into existing optical soil specimens from the ground. Obviously, the soil particles in VisCPT
characterization systems capturing binary soil images. Some of images are also in three-dimensional assemblies. Finally, some par-
the systems include: the University of Illinois Aggregate Image ticles such as fine sands are so small that it is unrealistic to separate
Analyzer (UIAIA) (Rao and Tutumluer 2000; Tutumluer and them prior to image capture, even in a laboratory. Therefore, a
Pan 2008); the Aggregate Imaging System (AIMS) (Fletcher et al. procedure was sought that could computationally extract particles
2003; Chandan et al. 2004; Mahmoud and Masad 2007); the Qicpic from images of three-dimensional assemblies so that the new com-
system (Altuhafi et al. 2013); and the Translucent Segregation putational algorithm for Wadell’s R could be used on them.
Table (TST) system (Ohm and Hryciw 2013). In all of these In 3D assemblies, particles are not only in contact with each
systems the particles are prepared to lie detached from one another other, they also block and are blocked from view by other particles.
thereby facilitating image collection for simple analysis. Some soil particles may have a full projection of their area in view
By contrast to the systems listed previously, in other image- while others will be occluded by foreground particles. Secondly,
based soil characterization systems the soil particles are not or the voids between soil particles are hard to distinguish from actual
cannot be detached. For example, in the Sedimaging system particles. Naturally, only particles exhibiting full projections are
(Ohm and Hryciw 2013) a 213 cm (7 ft) tall sedimentation column useful for characterization of form and roundness.
is used to rapidly sort soil particles by size prior to image Therefore, the challenge is to distinguish particles with full
capture. In these images, the sedimented soil particles are in projections from voids and occluded particles. Ideally, a computer

© ASCE 04016021-6 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 algorithm would make these distinctions. However, this is a ticles have a variety of colors, shapes, and roundnesses. In the Ot-
daunting task as soils have various colors, size distributions, inter- tawa sand image 129 particles showed full projections, whereas
nal textures, particle forms, and roundnesses. The authors’ future 203 particles were found in the 2NS image. The computational re-
research efforts will aim to teach computers to pick out the particles sults for the two soils are superimposed on the original images.
with full projections through machine learning and pattern recog- Once again, the maximum inscribed circles are dashed while the
nition techniques. However, to date, only human judgment is corner circles are solid. They again appear to nicely fit the particles
capable of making the selections. Therefore, a semiautomated ap- and their corners.
proach is used in this paper; it combines human judgment with a
computer’s rapid computational abilities. In this hybrid approach,
operators first pick out the particles with full projections. Wadell Roundness and Sphericity Distributions
Then, binary particle images are automatically generated. Finally,
roundness and other descriptors of particle geometry are deter- After identifying the maximum inscribed circle and corner circles,
mined using the computational methods described in this paper. the Wadell R of each particle is easily computed. Other geometric
Fig. 8(a) shows a natural soil aggregate image called brown descriptors such as length (d1 ), width (d2 ), sphericity, aspect ratio,
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

fused alumina oxide sand (BFAO). The particles are brown and orientation can also be readily obtained. Assuming the soil particles
have a complex texture. The particles whose projections are fully are ellipses, a relative volume of each particle can be computed as
visible can be manually picked out using the image processing d1 × d2 × d2 . The distribution of Wadell R by volume for the three
software Adobe Photoshop. To begin, the operator can trace the soils was computed and is shown in Fig. 11. The use of volume-
particle boundaries using the Photoshop tools polygonal lasso or based distributions for civil engineering (geomechanics) applications
magnetic lasso. When using the polygonal lasso, users must is more logical and appropriate than simple distributions based on
manually specify the perimeter points and Photoshop will connect particle counts. Nevertheless, it’s recognized that for relatively
the points to generate particle boundaries. The magnetic lasso uniform sized soil particles the distributions will be similar.
automatically detects the particle boundaries. Although it is a very Powers’ classification of R (according to Fig. 7) is also shown in
powerful tool that does not require much human interaction, the Fig. 11. As shown, the Ottawa sand is 75% by volume well-
magnetic lasso is ineffective for particles that exhibit complex rounded with about 25% rounded; 2NS contained approximately
textures because of roughness or mineral variability. 45% rounded and 35% subrounded with smaller volumes of
As such, the polygonal lasso was used for the BFAO. A total of well-rounded (10%) and subangular (10%); the BFAO was 50%
89 fully projected particles were identified. After delineating the subangular, 20% angular, and 30% subrounded by volume.
boundaries, the regions within them are filled with a distinct color By convention, R should be evaluated in the view showing the
as shown in Fig. 8(b). Fig. 8(b) is the input into the R computation largest projected area of a particle (Sneed and Folk 1958). There-
program previously described. The program easily extracts the fore, approximately 200 random particles from each of the three
newly colored particles and computes the PCD of each one. Each sands were laid out on a flat surface exposing their largest area.
particle is then upscaled or downscaled so that PCD equals 1,000 Images were captured and analyzed using the computational
pixels and δ0 is set to 0.3. The results have been superimposed on method. The results, shown by dashed lines in Fig. 11 are in very
the original image in Fig. 8(c) and a magnification of the shown good agreement with those obtained from the images of three-
rectangular region in Fig. 8(d). The dashed circles are the maxi- dimensional assemblies. As summarized in Table 1, the difference
mum inscribed circles and the solid circles are corner circles. between the average R determined from images of assemblies and
Images of two other soils were collected using different optical the average R from images of detached particles was 0.01 for all
systems. Fig. 9 shows standard Ottawa #20–#30 sand captured us- three soils.
ing the VisCPT. The particles are light brown, rounded, and spheri- Having identified particles with full projections, particle
cal. The soil in Fig. 10 is called 2NS by the Michigan Department sphericities can also be computed and their distributions can also
of Transportation; an image of its 1.4–2.0 mm size increment was be developed. Sphericity quantifies the similarity of a particles
captured in the Sedimaging device. The detailed Sedimaging test overall shape (form) to a perfect circle or sphere. Over the years,
procedures are provided by Ohm and Hryciw (2013). The 2NS par- at least five different definitions have been used to quantify S

Area of soil particle

Area Sphericity∶ SA ¼ ð1Þ
Area of minimum circumscribed circle

Diamter of circle with same area of soil particle

Diameter Sphericity∶ SD ¼ ð2Þ
Diameter of minimum circumscribed circle

Diameter of maximum inscribed circle

Circle Ratio Sphericity∶ SC ¼ ð3Þ
Diameter of minimum circum scribed circle

Perimeter of circle with same area of soil particle

Perimeter Sphericity∶ SP ¼ ð4Þ
Perimeter of soil particle

Width of soil particle

Width to Length Ratio Sphericity∶ SWL ¼ ð5Þ
Length of soil particle

© ASCE 04016021-7 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

Fig. 8. Circle fitting results for BFAO sand: (a) image of the three-dimensional assembly of BFAO; (b) delineated particles using Photoshop
polygonal lasso tool; (c and d) fitted corner circles (solid) and maximum inscribed circles (dashed)

Eqs. (1)–(5) were either proposed or have been adapted from for BFAO, Ottawa #20–#30, and 2NS, respectively. The results
Tickell (1931), Wadell (1935), Santamarina and Cho (2004), confirm that SWL indeed gives a widest numerical range for S.
Altuhafi et al. (2013), Krumbein and Sloss (1951, 1963), respec- The dashed lines in the three figures are the S distributions
tively. Sphericity values by all five definitions are shown for com- obtained using the largest projected areas of the same particles as
parison for the particles in Fig. 7. Zheng and Hryciw (2015) were used for computing roundness. The average values for all three
reviewed the origins of the five equations, presented their computa- soils by all five definitions of S are listed in Table 1. As observed, the
tional methods, and extensively compared their effectiveness at quan- differences in average values were at most 0.02.
tifying sphericity by two criteria: (1) their abilities to discriminate
sphericity (displaying a significant numerical range for particles
for angular to rounded particles); and (2) being unaffected by the Required Sample Size for Determining Particle
roundness of soil particles. The study showed that SWL is the ideal Roundness of a Soil
definition for S as it best satisfies both criteria.
Nevertheless, for comparison, all five sphericities were In most earlier studies only a mean R value for a soil was deter-
computed for the three soils. Figs. 12(a–c) are the S distributions mined. The ability to rapidly determine R for numerous particles

© ASCE 04016021-8 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

Fig. 9. Circle fitting results for Ottawa #20–#30: (a) image of the three-dimensional assembly of Ottawa #20–#30 by VisCPT; (b) fitted corner circles
(solid) and maximum inscribed circles (dashed)

Fig. 10. Circle fitting results for 2NS sand: (a) image of a three-dimensional assembly of 2NS sand by sedimaging; (b and c) fitted corner circles
(solid) and maximum inscribed circles (dashed)

in a specimen raises the question of how many are needed to obtain 50 particles. Cavarretta et al. (2010) believed that examining 30–40
statistically valid value. To obtain an average R for a soil specimen, particles would be adequate. Yang and Wei (2012) reported using
Youd (1973) reported that at least 50 particles are needed. Edil et al. 40 particles. Others who computed and reported average R values
(1975) estimated R for sand particles by Krumbein’s chart and include Eisma (1965), Frossard (1979), Sladen et al. (1985),
reported that viewing at least 25 particles were needed to yield Vepraskas and Casselkan (1987), Sagga et al. (1993), Frossard
a reliable mean. Cho et al. (2006) visually compared 30 particles (1979), Mehring and McBride (2007), Bareither et al. (2008),
to obtain the mean. Rouse et al. (2008) concluded that at least Chapuis (2012), and Cabalar et al. (2013). In summary, the typical
30 particles are needed to compute the mean R based on the prob- sample size used to compute an average R value has been in the
ability theorem law of large numbers. Bareither et al. (2008) used range of 30–50 particles.

© ASCE 04016021-9 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 100
WR R SR SA A VA SA

Percentage Smaller by Volume (%)

100
SD
Percentage smaller by Volume (%)

Solid lines: from 80

Assembly
particle assemblies.
image SC
80 BFAO Dashed lines: from
maximum projected
(89 particles) SP
60
areas. SWL
60
VA: Very angular
2NS Maximum
A: Angular 40 Dashed
40 SA: Subangular projections
lines
Ottawa SR: Subrounded (219 particles)
R: Rounded
20
20 WR: Well-rounded

0 0
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 0.4 0.2 0

Roundness, R (a) Sphericity, S

Fig. 11. Wadell roundness cumulative distributions 100

SA

Percentage Smaller by Volume (%)

80 SD
Assembly
From statistics, for normal distributions the minimum sample image SC
size, nmin necessary for reliably estimating the mean value of a (129 particles) SP
60
population is computed by
SWL
 
zα=2 σ 2 40
Maximum
Dashed
nmin ¼ ð6Þ projections
lines
E (170 particles)
20
where E = allowed error between the estimated population mean
and the actual population mean; σ = population standard deviation;
and Zα=2 = value related to the confidence level of 100ð1 − αÞ%, 0
which is obtained from a Z-table. 1 0.8 0.6 0.4 0.2 0
(b) Sphericity, S
If a maximum E of 0.05 is established for the mean roundness
with a confidence level of 98%, then α ¼ 0.02 and Zα=2 ¼ Z0.01 ¼ 100
2.33 from the Z-table. From study of over 20 different sands, SA
Percentage Smaller by Volume (%)

the authors have observed standard deviations of no more than

SD
0.17 for roundness, with most values below 0.15. Using the worst 80
Assembly
case σ ¼ 0.17 with E ¼ 0.05 and a 98% desired confidence image SC
(α ¼ 0.02), the computed nmin is 63. Because this paper presents (203 particles) SP
60
a more rapid and precise method for computing roundness than by
SWL
chart methods it is also logical that more particles could and should
be used than the previously used 30–50. Maximum
40 Dashed
projections
Eight natural and two crushed sands from various locations lines
(204 particles)
in Michigan, Texas, Missouri New Mexico, and California having
various particle shapes were evaluated using the computational 20
geometry methods for R and S. In each case, the average value
of 64 particles was computed. The mean R values with their
0
standard deviations and the mean SWL with their standard devia- 1 0.8 0.6 0.4 0.2 0
tions are shown in Fig. 13. As expected, the two crushed sands, (c) Sphericity, S
30A (R ¼ 0.15) and a crushed gabbro rock (R ¼ 0.23) were the
most angular, as expected. Fort Davis, Texas (R ¼ 0.41) is very Fig. 12. Sphericity distributions for (a) BFAO; (b) Ottawa #20-#30;
recent colluvium. Scotts Valley, California (R ¼ 0.40) is a residual (c) 2NS
sand from a mildly cemented sandstone. Rincon, New Mexico

Table 1. Mean R and S from Images of Assemblies and Images of Maximum Area Projections
R SA SD SC SP SWL
Soil IA MP IA MP IA MP IA MP IA MP IA MP
BFAO 0.31 0.32 0.62 0.62 0.78 0.78 0.62 0.62 0.88 0.89 0.69 0.70
Ottawa 0.75 0.76 0.75 0.77 0.87 0.88 0.74 0.76 0.96 0.96 0.79 0.80
2NS 0.53 0.52 0.62 0.64 0.78 0.79 0.62 0.63 0.91 0.92 0.67 0.67
Note: IA = from images of assemblies; MP = from images of maximum area projections.

© ASCE 04016021-10 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 (R ¼ 0.55) is a dessert sand, probably windblown. Capitola,
California (R ¼ 0.48) is an alluvial river bed sand. The most
rounded soils are a Mississippi River alluvium from New Madrid,
Missouri (R ¼ 0.57); a glaciofluvial sand from Oakland County,
Michigan (R ¼ 0.65); and a Lake Michigan Dune sand (R ¼ 0.62).
Visual observation of the images appears to confirm the
reasonableness of the R values and classifications. Just as impor-
tantly, the inscribed circles and circles fitted to corners are correctly
(a) (b) constructed. The largest observed standard deviation was 0.17
confirming the reasonableness of using σ ¼ 0.17 in Eq. (1) to
R = 0.15, = 0.08 R = 0.23, = 0.08 compute nmin .
SWL = 0.69, = 0.15 SWL = 0.56, = 0.14

Discussion and Future Studies

Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

Mean particle roundness has been correlated to various soil proper-

ties including minimum and maximum packing densities, the angle
of internal friction, critical state soil parameters, and compressibil-
ity. No attempts have been made to relate such properties to
distributions of particle R and S such as shown in Figs. 11
(c) (d) and 12. Although such research is yet to be performed, it is worth
considering what number of particles is needed to yield a statisti-
R = 0.40, = 0.09 R = 0.41, = 0.09 cally valid distribution of R. If the R values exhibit a normal
SWL = 0.73, = 0.13 SWL = 0.68, = 0.14
distribution, an empirical rule of thumb used by many statisticians
suggests that only 30 samples are sufficient to produce the
distribution. Intuitively, soils with larger standard deviations and
longer tails on the ends of the distribution curves should require
more samples. Of course, non-normal distributions would require
a larger number of samples. It is also noted that sampling bias could
outweigh the considerations of adequate sample size.
Finally, the soil specimens used in this study consisted of
relatively uniform-sized particles. Real soils, particularly if
(e) (f) engineered for foundations, pavement subbases and other fills
R = 0.48, = 0.12 R = 0.51, = 0.17
are typically better graded. Additional study is needed to evaluate
SWL = 0.72 , = 0.11 SWL = 0.69, = 0.14 how R and S values could vary with the size of particles in
such sands.

Conclusions

A computational geometry method was recently developed to de-

termine soil particle roundness and sphericity by their traditional
definitions developed in the 1930s by Hakon Wadell. In this paper,
the computed values were compared to visual charts of particle R
(g) (h) and S developed in the 1950s. The agreement was excellent and
R = 0.55, = 0.13 R = 0.57, = 0.14 given that the computational method is rapid and nonsubjective,
SWL = 0.70, = 0.13 SWL = 0.73, = 0.11 it should eliminate the need for usage of such charts, except
possibly as a qualitative visual guide.
In this paper, the computational geometry methods were ex-
tended to particles in three-dimensional assemblies that exhibited
full and unobscured projections. For particles exhibiting uniform
internal textures, Photoshop’s magnetic lasso tool was used to
define the perimeter, whereas for particles with complex internal
textures Photoshop’s polygonal lasso was used. Cumulative vol-
ume-based distribution of particle roundness and sphericity were
developed for uniformly textured Ottawa sand and Michigan
(i) (j)
2NS as well as for a highly textured brown fused aluminum oxide
R = 0.62, = 0.12 R = 0.65, = 0.13
SWL = 0.72, = 0.14 SWL = 0.72, = 0.12 sand (BFAO). As expected, the Ottawa sand proved to be mostly
well-rounded (75%) with about 25% by volume rounded particle;
Fig. 13. Mean R and S values and their standard deviations for sands 2NS ranged from 35% subrounded to 45% rounded with smaller
of various geologic origins: (a) Michigan 30A; (b) Crushed Gabbro; volumes of well-rounded (10%) and subangular particles (10%);
(c) Scotts Valley, California; (d) Fort Davis, Texas; (e) Capitola, California; the BFAO was 50% subangular, 20% angular, and 30% sub-
(f) Upper Peninsula, Michigan; (g) Rincon, New Mexico; (h) New Madrid, rounded. Sphericity distributions for these three soils confirmed
Missouri; (i) Lake Michigan Dunes; (j) Oakland County, Michigan that the ratio of particle width to length [as sphericity was defined
by Krumbein and Sloss (1963)] provides a broader range of values

© ASCE 04016021-11 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 than four other occasionally used definitions of sphericity. Average particles.” J. Comput. Civ. Eng., 10.1061/(ASCE)0887-3801(2004)18:1
values of R and S obtained from the images of three-dimensional (75), 75–82.
assemblies were also compared to average values obtained from Chapuis, R. P. (2012). “Estimating the in situ porosity of sandy soils
images of detached particles of the same sands laid out to expose sampled in boreholes.” Eng. Geol., 141–142(19), 57–64.
their largest projected areas. The differences in computed R and S Cho, G. C., Dodds, J., and Santamarina, C. (2006). “Particle shape effects
were insignificant. on packing density, stiffness, and strength: Natural and crushed sands.”
J. Geotech. Geoenviron. Eng., 10.1061/(ASCE)1090-0241(2006)132:
Finally, mean R was computed for 10 different sands of various
5(591), 591–602.
geologic origins. Their values ranged from 0.15 and 0.25 for
Cleveland, W. S., and Devlin, S. J. (1988). “Locally weighted regression:
crushed sands to 0.40–0.65 for alluvial and glacio-fluvial sands.
An approach to regression analysis by local fitting.” J. Am. Stat. Assoc.,
It was shown that 64 particles were adequate to compute a mean 83(403), 596–610.
roundness with 0.05 accuracy and 98% confidence. Edil, T. B., Krizek, R. J., and Zelasko, J. S. (1975). “Effect of grain
characteristics on packing of sands.” Proc., of Istanbul Conf. on Soil
Mechanics and Foundation Engineering, Istanbul Technical Univ.,
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

Acknowledgments Istanbul, Turkey, 46–54.

Efron, B., and Tibshirani, R. J. (1993). An introduction to the bootstrap,
This material is based upon work supported by the U.S. National Chapman and Hall/CRC, New York.
Science Foundation under Grant No. CMMI 1300010. ConeTec Eisma, D. (1965). “Eolian sorting and roundness of beach and dune sands.”
Investigations Ltd. and the ConeTec Education Foundation are Neth. J. Sea Res., 2(4), 541–555.
acknowledged for their support to the Geotechnical Engineering Fletcher, T., Chandan, C., Masad, E., and Sivakumar, K. (2003). “Aggre-
Laboratories at the University of Michigan. gate imaging system for characterizing the shape of fine and coarse
aggregates.” Transp. Res. Rec., 1832, 67–77.
Frossard, E. (1979). “Effect of sand grain shape on the interparticle friction;
indirect measurements by Rowe’s stress dilatancy theory.” Géotechni-
Notation
que, 29(3), 341–350.
The following symbols are used in this paper: Ghalib, A. M., Hryciw, R. D., and Susila, E. (2000). “Soil stratigraphy
delineation by VisCPT.” Innovations and Applications in Geotechnical
d1 = length of a particle;
Site Characterization, ASCE, Reston, VA, 65–79.
d2 = width of a particle;
Hryciw, D. R., Zheng, J., Ohm, H., and Li, J. (2014). “Innovations in
E = the allowed error between the estimated population optical geocharacterization.” Geo-Congress 2014 Keynote Lectures:
mean and the actual population mean; Geo-Characterization and Modeling for Sustainability, Vol. 235,
nmin = the minimum sampling number for estimating ASCE, Reston, VA, 97–116.
population mean; Hryciw, R. D. and Ohm, H.-S. (2013). “Soil migration and piping suscep-
PCD = pixels per circumscribing circle diameter; tibility by the VisCPT.” Geo-Congress 2013: Stability and Performance
R = roundness; of Slopes and Embankments III, ASCE, Reston, VA, 192–195.
S = sphericity; Kandasami, R. K., and Murthy, T. G. (2014). “Effect of particle shape on
Zα=2 = value related to the confidence level of 100ð1 − αÞ%, the mechanical response of a granular ensemble.” Proc., Int. Symp. on
which is obtained from a Z-table; Geomechanics from Micro to Macro, Taylor & Francis Group, Boca
δ 0 = maximum allowed divergence of a curve from the Raton, FL, 1093–1098.
straight line segment approximating it; Krumbein, W. C. (1941). “Measurement and geological significance of
θ and ρ = polar coordinates; and shape and roundness of sedimentary particles.” J. Sediment. Petrol.,
σ = the population standard deviation. 11(2), 64–72.
Krumbein, W. C., and Sloss, L. L. (1951). Stratigraphy and sedimentation,
W. H. Freeman and Company, San Francisco.
Krumbein, W. C., and Sloss, L. L. (1963). Stratigraphy and sedimentation,
References 2nd Ed., W. H. Freeman and Company, San Francisco.
Kumara, G. J. J., Hayano, K., and Ogiwara, K. (2012). “Image analysis
Altuhafi, F., O’Sullivan, C., and Cavarretta, I. (2013). “Analysis of an im- techniques on evaluation of particle size distribution of gravel.” Int.
age-based method to quantify the size and shape of sand particles.” J. J. Geomate., 3(1), 290–297.
Geotech. Geoenviron. Eng., 10.1061/(ASCE)GT.1943-5606.0000855, Kuo, C. Y., and Freeman, R. B. (2000). “Imaging indices for quantification
1290–1307. of shape, angularity and surface texture of aggregates.” Transp. Res.
Arasan, S., Akbulut, S., and Hasiloglu, A. S. (2011). “The relationship Rec., 1721, 57–65.
between the fractal dimension and shape properties of particles.” KSCE Kuo, C. Y., Frost, J. D., Lai, J. S., and Wang, L. B. (1996). “Three-
J. Civ. Eng., 15(7), 1219–1225. dimensional image analysis of aggregate particles from orthogonal
Bareither, C. A., Edil, T. B., Benson, C. H., and Mickelson, D. M. (2008).
projections.” Transp. Res. Rec., 1526, 98–103.
“Geological and physical factors affecting the friction angle of
Mahmoud, E., and Masad, E. (2007). “Experimental methods for the
compacted sands.” J. Geotech. Geoenviron. Eng., 10.1061/(ASCE)
evaluation of aggregate resistance to polishing, abrasion, and breakage.”
1090-0241(2008)134:10(1476), 1476–1489.
Bowman, E. T., Soga, K., and Drummond, W. (2001). “Particle shape J. Mater. Civ. Eng., 10.1061/(ASCE)0899-1561(2007)19:11(977),
characterization using Fourier descriptor analysis.” Géotechnique, 977–985.
51(6), 545–554. Masad, E., Al-Rousan, T., Button, J., Little, D., and Tutumluer, E. (2007).
Cabalar, A. F., Dulundu, K., and Tuncay, K. (2013). “Strength of various “Test methods for characterizing aggregate shape, texture, and angular-
sands in triaxial and cyclic direct shear tests.” Eng. Geol., 156(1), ity.” National Cooperative Highway Research Program Rep. No. 555,
92–102. Transportation Research Board, Washington, DC.
Cavarretta, I., O’Sullivan, C., and Coop, M. (2010). “The influence of Mehring, J. L., and McBride, E. F. (2007). “Origin of modern quartzarenite
particle characteristics on the behaviour of coarse grained soils.” beach sands in a temperate climate, Florida and Alabama, USA.” Sed.
Géotechnique, 60(6), 413–423. Geol., 201(3–4), 432–445.
Chandan, C., Sivakumar, K., Masad, E., and Fletcher, T. (2004). Mitchell, J. K., and Soga, K. (2005). Fundamentals of soil behavior, 3rd
“Application of imaging techniques to geometry analysis of aggregate Ed., Wiley, New York.

© ASCE 04016021-12 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021

 Mollon, G., and Zhao, J. D. (2012a). “Fourier-Voronoi-based generation of Sukumaran, B., and Ashmawy, A. K. (2001). “Quantitative characterization
realistic samples for discrete modelling of granular materials.” Granular of the geometry of discrete particles.” Géotechnique, 51(7), 619–627.
Matter, 14(5), 621–638. Tickell, F. G. (1931). The examination of fragmental rocks, Stanford
Mollon, G., and Zhao, J. D. (2012b). “Generating realistic 3D sand particles University Press, Stanford, CA.
using Fourier descriptors.” Granular Matter, 15(1), 95–108. Tutumluer, E., and Pan, T. (2008). “Aggregate morphology affecting
Mollon, G., and Zhao, J. D. (2014). “3D generation of realistic granular strength and permanent deformation behavior of unbound aggregate
samples based on random fields theory and Fourier shape descriptors.” materials.” J. Mater. Civ. Eng., 10.1061/(ASCE)0899-1561(2008)
Comput. Meth. Appl. Mech. Eng., 279, 46–65. 20:9(617), 617–627.
Moroto, N., and Ishii, T. (1990). “Shear strength of uni-sized gravels under Vallejo, L. E. (1995). “Fractal analysis of granular materials.” Géotechni-
triaxial compression.” Soils Found., 30(2), 23–32.
que, 45(1), 159–163.
Oh, T. M., Joo, G. W., Yun, K. J., and Cho, G. C. (2014). “Crush character-
Vallejo, L. E., and Zhou, Y. (1995). “The relationship between the fractal
istics of abrasive particles during abrasive waterjet rock cutting.” Proc.,
dimension and Krumbein’s roundness number.” Soils Found., 35(1),
Int. Symp. on Geomechanics from Micro to Macro, Taylor & Francis
Group, Boca Raton, FL, 1577–1581. 163–167.
Ohm, H. S., and Hryciw, R. D. (2013). “Translucent segregation table test Vepraskas, M. J., and Cassel, D. K. (1987). “Sphericity and roundness of
Downloaded from ascelibrary.org by University of Michigan on 11/02/16. Copyright ASCE. For personal use only; all rights reserved.

for sand and gravel particle size distribution.” Geotech. Test. J., 36(4), sand in coastal plain soils and relationships with soil physical
592–605. properties.” Soil Sci. Soc. Am. J., 51(5), 1108–1112.
Powers, M. C. (1953). “A new roundness scale for sedimentary particles.” J. Wadell, H. (1932). “Volume, shape and roundness of rock particles.” J.
Sediment. Petrol., 23(2), 117–119. Geol., 40(5), 443–451.
Rao, C., and Tutumluer, E. (2000). “Determination of volume of Wadell, H. (1933). “Sphericity and roundness of rock particles.” J. Geol.,
aggregates.” Transp. Res. Rec., 1721, 73–80. 41(3), 310–331.
Raschke, S., and Hryciw, R. (1997). “Vision cone penetrometer for direct Wadell, H. (1935). “Volume, shape, and roundness of quartz particles.”
subsurface soil observation.” J. Geotech. Geoenviron. Eng., 10.1061/ J. Geol., 43(3), 250–280.
(ASCE)1090-0241(1997)123:11(1074), 1074–1076. Wang, L., Wang, X., Mohammad, L., and Abadie, C. (2005). “A unified
Rouse, P. C., Fannin, R. J., and Shuttle, D. A. (2008). “Influence of method to quantify aggregate shape angularity and texture using Fourier
roundness on the void ratio and strength of uniform sand.” Géotechni- analysis.” J. Mater. Civ. Eng., 10.1061/(ASCE)0899-1561(2005)17:5
que, 58(3), 227–231. (498), 498–504.
Sagga, M. A. (1993). “Roundness of sand grains of longitudinal dunes in Wettimuny, R., and Penumadu, D. (2004). “Application of Fourier analysis
Saudi Arabia.” Sediment. Geol., 87(1–2), 63–68. to digital imaging for particle shape analysis.” J. Comput. Civ. Eng., 10
Santamarina, J. C., and Cho, G. C. (2004). “Soil behaviour: The role of .1061/(ASCE)0887-3801(2004)18:1(2), 2–9.
particle shape.” Advances in Geotechnical Engineering: The Skempton
Yang, J., and Wei, L. M. (2012). “Collapse of loose sand with the addition
Conf., Thomas Telford, London, 604–617.
of fines: The role of particle shape.” Géotechnique, 62(12), 1111–1125.
Shin, H., and Santamarina, J. C. (2013). “The role of particle angularity on
Youd, T. L. (1973). “Factors controlling maximum and minimum densities
the mechanical behavior of granular mixtures.” J. Geotech. Geoenviron.
Eng., 10.1061/(ASCE)GT.1943-5606.0000768, 353–355. of sands.” Evaluation of relative density and its role in geotechnical
Sladen, J. A., D’Hollander, R. D., and Krahn, J. (1985). “The liquefaction projects involving cohesionless soils, STP 523, ASTM, Philadelphia,
of sands, a collapse surface approach.” Can. Geotech. J., 22(4), 98–112.
564–578. Zheng, J., and Hryciw, R. D. (2015). “Traditional soil particle sphericity,
Sneed, E. D. and Folk, R. L. (1958). “Pebbles in the lower Colorado River, roundness and surface roughness by computational geometry.”
Texas: A study of particle morphogenesis.” J. Geol., 66(2), 114–150. Géotechnique, 65(6), 494–506.

© ASCE 04016021-13 J. Comput. Civ. Eng.

J. Comput. Civ. Eng., 2016, 30(6): 04016021