International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50

Contents lists available at ScienceDirect

International Journal of
Rock Mechanics and Mining Sciences
journal homepage: www.elsevier.com/locate/ijrmms

Effect of impact velocity, block mass and hardness on the coefficients of T

restitution for rockfall analysis
⁎
Pavlos Asteriou , George Tsiambaos
National Technical University of Athens, Greece

A R T I C LE I N FO A B S T R A C T

Keywords: One of the most significant input parameters in rockfall trajectory modelling is the Coefficient of Restitution,
Coefficients of restitution which controls block rebound. In particular, in the design of rockfall barriers, it can significantly affect their
Impact velocity height. In practice, the coefficients of restitution are acquired by field or laboratory tests, back analyses of known
Block mass events, experience or, most commonly, suggested values connected to the slope material. Furthermore, to ac-
Schmidt hardness
count for the effect of impact velocity or block mass, scaling factors are available. However, the suggested values
Rockfall modelling
are based on a qualitative description of the slope material, and the scaling factors are not well documented, nor
do they account for the effects of impact velocity and block mass simultaneously. In this paper, a semi-empirical
correlation is proposed that takes into account the Schmidt hammer rebound value of both the slope and block
material as well as the impact velocity and block mass. This was derived by an extensive laboratory experimental
study (445 impact tests) of a one-dimensional drop of spherical blocks onto planar surfaces. The semi-empirical
correlation proposed adequately describes the responses observed under the circumstances imposed in the la-
boratory.

1. The coefficient of restitution 1.1. Definitions for the coefficients of restitution

The coefficient of restitution (COR ), or the rebound coefficient (R),
is defined as the decimal fractional value that represents the ratio of
velocities before and after an impact of two colliding entities.
Theoretically, a COR that equals one corresponds to a perfect elastic
collision, a COR less than one defines an inelastic collision and COR of
zero describes the instantaneous stopping of the block at the surface
area, without rebound, i.e., a perfectly plastic impact.1 However, in
practice, COR will be less than one, even in elastic normal impacts, as
argued by Imre et al.2
After an impact, the block velocity changes according to the COR
value. Hence, in rockfall engineering practice, COR is assumed to be an
overall value that takes into account all the characteristics of an impact,
including deformation, sliding at the contact area, and transformation
of rotational moments into translational ones, and vice versa.3 More-
over, COR is influenced by a wide range of parameters, as summarized
in Table 1.
Various definitions on the COR s have been proposed, such as the
kinematic, kinetic and energy COR . However, there is no consensus on
which of them is more appropriate for modelling the impact during a
rockfall.5
The kinematic approach, derived from the inelastic collision of
particles in Newtonian mechanics, is the first and simplest definition.
For an object (block) impacting on a steadfast surface (slope), kinematic
COR (vCOR ) is defined as
vr
vCOR =
vi (1)
where vi is the incident velocity magnitude and vr is the rebound ve-
locity magnitude (see Fig. 1)
In engineering practice, the most commonly used COR definition is
derived from the normal and tangential projections of the kinematic
COR to the impact surface. Normal to the slope surface, energy dis-
sipation is associated with the deformation of the colliding entities and
the propagation of elastic waves. Normal COR (nCOR also found as Rn ,
kn ) is determined by
vn, r
nCOR = −
vn, i (2)
In the tangential direction, energy losses are attributed to the fric-
tion between the colliding bodies and to the transformation of rota-
tional moments into translational ones, and vice versa. Therefore,

⁎
Corresponding author.
E-mail address: paster@central.ntua.gr (P. Asteriou).

https://doi.org/10.1016/j.ijrmms.2018.04.001
Received 12 September 2017; Received in revised form 2 April 2018; Accepted 5 April 2018
Available online 13 April 2018
1365-1609/ © 2018 Elsevier Ltd. All rights reserved.
P. Asteriou, G. Tsiambaos International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50

Table 1 Table 2
Parameters that influence the rebound response.4 Characteristic normal and tangential COR values.8
Slope characteristics Block characteristics Kinematics Slope Properties normal COR tangential COR

Strength Strength Translational velocity Clean hard rock 0.53 0.99

Stiffness Stiffness Rotational velocity Asphalt road 0.40 0.90
Roughness Weight - Size Collision angle Bedrock outcrops with boulders 0.35 0.85
Inclination Shape Configuration of block Talus cover 0.32 0.82
Soft soil with vegetation 0.30 0.80

documentation of the COR values obtained from experimental studies

Fig. 1. Quantities used in COR definitions.
tangential COR (tCOR , also known as Rt , kt ) is defined as
vt , r
tCOR =
vt , i
where the subscripts n and t in Eqs. (2) and (3) denote the normal and
tangential velocity components with respect to the slope surface (see
Fig. 1).
Another definition of COR found in the literature is given in energy
terms and is known as Energy COR (ECOR) :
0.5(mvr2 + Iωr2)
ECOR =
0.5(mvi2 + Iωi2)
where m is the mass, I is the moment of inertia and ω is the angular
velocity magnitude of the block.
However, in relevant studies, Energy COR (Eq. (4)) is mostly found
without considerating the rotational motion, according to Eq. (5). The
latter is hereafter denoted eCOR to distinguish it from Eq. (4):
can be found in Heidenreich10, Turner and Schuster11, Asteriou12 and
other studies.
The aforementioned values are suggestive. Software developers
encourage designers to perform field tests and back-analyse the tra-
jectories to acquire site-specific values. Unfortunately, in most cases,
this procedure is omitted because it is a difficult, expensive and time-
consuming task. Therefore, a design is often based on the suggested
values.
However, the COR values originating from laboratory and field tests
or back analyses of rockfall events are significantly scattered and do not
show any dominant or systematic pattern. This is depicted in Fig. 2,
where the aforementioned values are compared to those suggested in
the literature for the case of a hard rock. Suggested COR is plotted by
the blue circle (the centre is the suggested value (nCOR , tCOR ), and the
radius is the standard deviation), and the data originating from the
relevant literature13,8,7,14,3,15–20,5,21,10,22,2,23–25,11,26–28,12 are plotted by
red points.
The rebound height is very sensitive to the selected value, as ob-
served in Fig. 3 through a simple parametric analysis altering nCOR . By
contrast, the kinetic energy is not as strongly affected; in a practical
application, the proposed barrier would have the same capacity.
Moreover, selecting a COR based solely on the material type erro-
(3) neously omits the effects of the parameters given in Table 1.
1.3. Effect of rock type on COR values
As a general rule, COR increases as the slope material becomes
“harder”. This is qualitatively verified by the suggested values pre-
sented in Table 2 but becomes questionable when using values de-
termined experimentally or in back-analysis studies, as shown in Fig. 2.
(4) This difference is attributed to the highly volatile nature of the

0.5mvr2
eCOR =
0.5mvi2 (5)

It is apparent that eCOR is equal to the square root of vCOR . Therefore, this
definition does not provide any benefit compared to the simpler defi-
nition of vCOR .
Additionally, more complex definitions have been proposed, such as
the impulse COR s defined by Descoeudres et al.6 and Pfeiffer and
Bowen.7 However, they are rarely used due to their overall complexity.

1.2. Characteristic COR values

In practical applications, the COR values are selected based on the

material that constitutes the slope surface. For that purpose, some in-
dicative values are provided as guidelines. Table 2 summarizes the
values suggested by Hoek8, which have been adopted in the RocFall
simulation software9 with a standard deviation of 0.04. Similar values Fig. 2. Comparison of suggested COR (blue circle) with the values obtained
were initially proposed by Pfeiffer and Bowen7. A detailed from experimental studies for hard rock slopes (red points).

42
P. Asteriou, G. Tsiambaos
Fig. 3. Parametric analysis altering the nCOR value; Top: Trajectories in xy
space, Middle: Rebound height, hR , measured from the slope surface, and
Bottom: Kinetic Energy, Ekin .
parameters that affect the rebound (see Table 1) and is more pro-
nounced when the impact is oblique and the block has a random shape.
Rayudu29 attempted to quantify this qualitative rule by performing
laboratory tests using 14 different impact surfaces and a metal sphere
with a 4-cm diameter. A significant positive correlation arose with the
Schmidt hardness (Schmidt Hammer Rebound Value). However, the
author stated that the applicability of the derived correlation was
questionable because the elastic properties of the metal sphere differed
significantly from those of rock materials.
Extending the aforementioned study, Richards et al.20 proposed a
correlation (Eq. (6)) between nCOR and the Schmidt hardness (R), based
on free fall drop tests on different rock types (basalt, diorite, gneiss,
granite, limestone, marble, rhyolite, sandstone and schist) and on
debris and soil materials.
−110 + 9Rs + 4Rb + 4θ
nCOR =
1000 (6)
where θ is the inclination of the impact surface (in degrees) and Rb and
Rs are the Schmidt hardness of the block and the impact surface, re-
spectively.
More recently, a similar correlation (Eq. (7)) was proposed by An-
sari et al.30, who performed a laboratory study with 7 different mate-
rials in which nearly spherical blocks, with diameters between 4 and
43
International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50
5 cm, were vertically dropped to 5-cm-thick impact surfaces.
nCOR = −0.215 + 0.0183Rs − 0.007Rb (7)
1.4. Scaling methods for the COR
Among other factors, COR depends on the incident velocity and
mass (see Table 1). To account for these parameters when analysing
rockfalls, various scaling factors have been proposed. The first scaling
method was introduced in the Colorado Rockfall Simulation Program
(CRSP)7, in which both normal and tangential COR values are scaled
according to the magnitude of the normal incident velocity. The scaled
nCOR is calculated by multiplying the characteristic value nCOR, ref (as
obtained from the suggested values) by the scaling factor B v , defined as
1
Bv =
1 + (vi, n/ vref )2 (8)
where vref is the reference velocity and was proposed to be 9.14 ms−1.
However, little information is available concerning its calibration.31
This scaling factor has been adopted in many simulation codes (e.g.,
RocFall,9 RockyFor3D.32
Peng19 performed a small-scale laboratory investigation with
spherical blocks formed from various rock types. The blocks were
vertically released onto horizontal and inclined surfaces from drop
heights (hD ) that varied between 0.6 and 3 m. Based on that study,
Richards et al.20 found that a reference velocity of vref = 5 ms−1 de-
scribed their laboratory findings better.
In the CRSP software, the tangential COR is scaled according to Eq.
(9).
1
B vT =
1 + (vn, i/ vT , ref )2 (9)
where vT , ref is a reference velocity that was suggested to equal
15. 2 ms−1. The use of the normal component of impacting velocity is
based on the Coulomb friction principle. Friction is analogous to the
applied normal force, which is connected to the normal incident velo-
city. However, the calibration of the reference velocity is not well
documented. In contrast to the normal scaling factor, Eq. (9) is not used
in other simulation software. RocFall9 performs scaling according to the
slope's friction angle (ϕ ), and RockyFor3d32 does so according to Mean
Obstacle Height Principle (MOH).
Based on parabolic release laboratory tests of irregularly shaped
blocks performed by Ushiro et al.33 it was observed that nCOR decreases
as the incident velocity increases. Eqs. (10) and (11) were proposed for
surfaces of natural rock and concrete, respectively.
nCOR = 2.506e−0.1053(−vn, i) (10)
nCOR = 0.847e−0.0842(−vn, i) (11)
The aforementioned equations are both statistically not significant
because the coefficient of determination, R2 , is less than 0.5 when fitted
to the data presented by Ushiro et al.33 However, they do describe the
overall trends found in the relevant literature. Since different equations
are proposed for the two materials used, it is concluded that the ma-
terial type affects nCOR when all other parameters are similar. Ad-
ditionally, based on the experiments, it is stated that nCOR decreases
with an increase in block mass, regardless of the incident velocity;
further, tCOR was found to be insensitive to incident velocity.
An alternate velocity scaling factor (Eq. (12)) was introduced by
Rammer et al.34 in which scaling is applied for incident velocities vn, i
greater than 10 ms−1. Since the majority of COR values found in the
literature originate from testing in lower incident velocities, it is rea-
sonably assumed that scaling is included in the suggested COR values.
1 vi ≤ 10 ms−1
k = ⎧ 0.25 −0.25
⎨10 vn, i vi > 10 ms−1 (12)
⎩
P. Asteriou, G. Tsiambaos International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50

This scaling factor originates from an equation proposed by

Johnson35, according to which COR is proportional to vi−0.25 . Its cali-
bration was based on the experimental tests performed by Goldsmith1,
who conducted free fall tests of metal spheres on horizontal impact
surfaces.
A scaling factor dependent solely on mass is also available in the
RocFall software.9 This factor, Bm , presents a similar trend to that of the
scaling factor B v .
1
Bm =
1 + (m / mref )2 (13)
where mref is the reference mass required for the scaling factor Bm to
equal 0.5.
However, no information concerning the extrapolation of Eq. (13)
or the value for mref is available. Moreover, according to the software
developers, it is not advisable to use mass and velocity scaling si-
multaneously because doing so generates unrealistic rock paths. Fig. 4. Experimental apparatus.
Recently, a momentum-based COR definition was proposed by
Bourrier and Hungr.31 COR is represented by a hyperbolic function (Eq.
Impact surfaces were created for all materials used. Each impact
(14)) that yields unity in ‘soft’ impacts and zero in severe impacts, ac-
surface was the top side of a 5-cm-thick rectangular plate with 15-cm
cording to the fundamental principles of impact mechanics.
sides that was fixed to a massive dead weight base to ensure the pre-
M0.5 servation of momentum during the impact.
k′ =
Mn + M0.5 (14) The release mechanism held the block in the desired height by
where Mn is the normal incident momentum calculated as Mn = m ·vn ; suction produced by a vacuum pump. The block was released into a free
M0.5 is the reference momentum; and k′ is the scaled normal COR . fall drop by switching the vacuum pump off. Consequently, the block
Based on the preceding literature review, it becomes apparent that a had only vertical translational velocity prior to impact.
better method for selecting COR values is necessary. This method The trajectory of the block was recorded with a digital high-speed
should incorporate the material types of both the slope surface and the video camera at a frame rate of 500 fps and a resolution of 440 × 330
block as well as the block mass and its incident velocity. In this paper, a pixels. The camera was installed 1 m from the impact surface. A high-
correlation among the aforementioned parameters is proposed that is contrast background was installed to make the block visually distin-
derived by the extensive experimental campaign presented hereafter. guishable.

2. Experimental set-up and data acquisition technique 2.2. Materials

Block shape has been found to have a significant effect on the re- The physical and mechanical properties of the materials used were
bound trajectory and, therefore, on the COR obtained from experi- determined according to the ISRM suggested methods37 (see Table 3).
mental testing.36 Additionally, when performing free fall tests with These materials have also been examined to investigate the effect of
randomly shaped blocks, the rebound trajectory presents a wide de- various material properties along with other natural rock materials
viation from the fall direction27 due to the configuration of the block commonly found in Greece.27
during impact.
To eliminate these effects as well as those attributed to the para-
2.3. Data acquisition
meters shown in Table 1, which enhance randomness, tests were per-
formed with spherical blocks, flat impact surfaces and free fall drops.
A video file is a sequence of still images (frames) displayed at
Theoretically, under such conditions, the rebound angle (αr ) for a
constant time intervals (fps). A gray-scale digital image forms an array
normal and frictionless impact between a sphere and a flat steadfast
of pixels the size of which is described by the image resolution. Each
surface should be equal to 90o , and consequently, nCOR should equal
element in the array has a numerical value from 0 (black) to 255
vCOR . However, due to the micro-roughness of the colliding entities,
(white) that represents the intensity of the pixel. Thus, each image
there is a slight divergence from the vertical direction during the ex-
forms a matrix, which can easily be manipulated. An in-house MatLab
perimental tests.2
code was created to perform a fast, accurate and robust trajectory ac-
quisition.
2.1. Experimental set-up
Table 3
The equipment required for conducting impact tests in the labora-
Material properties.
tory includes a block, an impact surface, a release mechanism and a
recording system (Fig. 4). Material property cement epoxy marble sandstone
grout resin
The experimental campaign was performed in two stages: first with
artificial materials and then with natural rock materials. In the first Density, ρ (Mg/m3 ) 2.19 2.04 2.52 2.52
stage, artificial blocks were cast in silicon moulds with diameters of 3, Compressive strength, σci (MPa) 37.4 86.5 66.1 107.5
4, 5 and 6 cm. A high-strength cement grout and an epoxy resin were Young's modulus, Et (GPa) 15.7 11.0 40.2 35.4
used as mortars and were stable and cohesive, exhibited zero segrega- Poisson's ratio, ν 0.11 0.38 0.26 0.22
Tensile strength, σt (MPa) 7.5 15.6 6.2 11.4
tion and were shrinkage-compensating, ultimately resulting in uniform
P-wave velocity, vp (ms−1) 4335 4125 4954 4606
samples when hardened.
S-wave velocity, vs (ms−1) 2575 2430 2861 2981
In the second stage, natural spherical blocks were formed with three
Schmidt Hardness, R 33.4 42.3 38.2 44.7
different diameters. The specimens originated from intact blocks of a
fine-graded marble and a quartizic sandstone.

44
P. Asteriou, G. Tsiambaos International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50

Fig. 5. Tracking algorithm stages: a. recorded image; b. converted into a binary image discretizing the block; c. boundary recognition, acquisition of x i , yi coordinates
and calculation of block's geometrical properties.

Initially, each frame was converted into a binary image (black and and angular velocity. To omit their effects, only the initial impacts were
white), discretizing the block from the background (Fig. 5-b). The near taken into account. This deviation was also observed by Imre et al.2 and
end projection of the block in the capture plan resembles a circular was attributed to the grain boundaries and the micro-fissures of the
shape. Therefore, an edge recognition algorithm38 picks out the colliding entities because their surfaces can never be perfectly smooth.
boundary between the block and the background (Fig. 5-c). The co-
ordinates of each pixel comprising the boundary were used to calculate 3. Experimental programme - presentation of results
the centre point and the radius of the spherical block (Fig. 5-c), using
Umbach and Jones39 least squares method (Eq. (15)): 3.1. Effect of incident velocity and mass
n
2
SS (X , Y , r ) = ∑ (r − (x i − X )2 + (yi − Y )2 ) To assess the effect of incident velocity (vi ) on the COR , tests with
i=1 (15) various release heights (hD ) were performed for all blocks. More pre-
cisely, the heights ranged from 0.1 to 4.5 m, resulting in incident ve-
where x i , yi are the coordinates of boundary points; X , Y are the co-
locities (vi ) ranging from 1.3 to 9.3 ms−1. In total, 73 different config-
ordinates of the centre point; and r is the circle radius.
urations were examined (see Table 4), each consisting of 5 repetitions,
Repeating this process for every captured image in a test gives us a
for a total of 365 tests.
reconstruction of the trajectory.
During the free fall of a spherical block onto a flat surface, the re-
Later in the process, the trajectory was separated in the pre- and
sponse of the block to the impact was repeatable. This was seen by the
post-impact stages. As the acceleration is constant in the Y direction,
low standard deviation on the determined COR values, which reached
due to gravity, a 2nd-degree regression analysis over time was per-
0.02 within each test set. In contrast, the tests performed with the same
formed for each stage.
methodology but with blocks that were cubically shaped with rounded
Due to the finite pixel dimensions, the points on the circular
edges (see Asteriou et al.26) presented a standard deviation of up to
boundary include a positioning error that equals the pixel size.
0.11. This behaviour is attributed to the random orientation of the
However, this has a minor effect on the determination of the centre
cubical blocks at impact, in contrast to the repeatable impact config-
point because Eq. (15) is used with a vast number of points, which
uration that occurs between a sphere and a flat surface.
causes the error to be less than the pixel size.
In all tests, the rebound angle was also repeatable and had values
The positioning error between frames due to the velocity change
close to 90 o . The global minimum of the 365 tests was αr = 78. 1o ,
caused by the acceleration was diminished by using a high frame rate
which indicates that the differences between nCOR and vCOR were less
and a 2nd-degree regression. However, using a high frame rate in-
than 2.5%. Therefore, the results presented hereafter can be assessed in
creases the velocity error caused by the positioning uncertainty dras-
terms of nCOR or vCOR .
tically40. When velocity is calculated by the position measurements, an
Fig. 6 depicts vCOR as a function of vi for the tests performed with the
uncertainty of δy is included. Since the time interval between the two
sandstone spheres. It becomes apparent that vCOR reduces with an
positions is Δt , the error in velocity is 2(δy /Δt )2 . It is apparent that as Δt
decreases (while the frame rate increases) the error in velocity in-
creases. To increase the time interval and reduce the error, the incident
velocity was calculated by Eq. (16) as the average of the velocities
calculated from the impact point and each measured point. A more
analytical presentation of the error analysis is beyond the scope of this
paper. However, after implementing this methodology, the relative
error in the calculated COR value was less than 3%.
n
1 ΔYi ⎤
vi =
n
∑ ⎡⎢0.5 g ΔTi + ΔTi ⎥
i=1 ⎣ ⎦ (16)

where n is number of frames (block positions); ΔYi is the vertical dis-

tance between the centre point at frame i and the impact point; ΔTi is
the flight time from frame i until impact; and g is the gravitational
constant.
Several bounces on the impact surface occurred for each test.
However, a progressively increasing offset from the vertical direction Fig. 6. vCOR as a function of vi for the tests performed with sandstone spheres
happened after each impact, resulting in the generation of horizontal (data are grouped according to block diameter).

45
P. Asteriou, G. Tsiambaos International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50

Table 4 Table 5
Experimental programme for the investigation of the effect on COR posed by Experimental programme for the investigation of the effect posed by the dif-
block's incident velocity and mass. ference in material properties between the colliding entities (All tests are per-
formed from a drop height of hD = 0.5 m ).
Material Schmidt Block diameter, Drop Height, hD [cm]
Hardness, R [-] d [cm] Surface Material Block Material Block diameter d [cm]

Concrete 33.4 3, 4, 5 0.1, 0.25, 0.5, 1.0, 2.0, 3.0 Concrete Marble 2.7, 3.0, 3.4
Marble 38.2 2.7, 3.0, 3.4 0.125, 0.25, 0.5, 1.0 Sandstone 2.9, 4.0, 4.3
Epoxy 42.3 3, 4, 5, 6 0.1, 0.25, 0.5, 1.0, 2.0, 3.0, 4.5 Marble Concrete 3.0, 4.0
Sandstone 44.7 2.9, 4.0, 4.3 0.125, 0.25, 0.5, 1.0, 1.8 Sandstone 2.9, 4.0, 4.3
Sandstone Concrete 3.0, 4.0
Marble 2.7, 3.0, 3.4

increase in vi and that for a given vi , an increase in mass decreases vCOR .

Therefore, both parameters affect the rebound behaviour of the block.
The same response was shown from the other three materials, but the
magnitude of vCOR differed (the results for the concrete and epoxy
blocks have been presented in41). This result indicates that the rebound
is also affected by the material type.

3.2. Effect of the impact surface

To examine the effects of the difference in the material consisting

the colliding bodies, further tests were conducted by letting the spheres
impact on a surface of a different material. Sixteen more test sets were
performed by releasing spheres composed of sandstone, marble and
concrete from a release height of 0.5 m, acquiring an incident velocity
(vi ) of approximately 3.1 ms−1. The combinations tested are summarized
in Table 5.
In Fig. 7, vCOR determined from the tests with sandstone spheres is
presented as a function of the Schmidt hardness of the impact surface. It
is seen that vCOR is affected by the material type of the impact surface,
since it increases as the surface becomes harder. Additionally, an in-
crease in block mass for a given incident velocity results in a lower vCOR ,
in accordance with the observations presented in the previous section.
Similar conclusions are drawn from the tests performed with the marble
and concrete blocks.
3.3. Block fracture
Rock fragmentation is frequently observed during rock fall events.23
However, this phenomenon is not accounted for when analysing the
response of a block to an impact. A major limitation comes from the
assumptions used to define COR s.
Marble and concrete blocks were fractured during the tests when
they were released from a height (hD ) of 2.0 and 4.5 m respectively.
Fracture initiated by splitting along the loading direction (Fig. 8),
Fig. 8. Image sequence of a fracture. Top: 3 cm diameter concrete block re-
leased from a height of hD = 4.5 m; a. ta = 0.000 s; b. tb = 0.008 s; and c. tc =
0.024 s. Bottom: real block fracture documented during an earthquake-induced
rockfall in Lefkada, Greece 2015.
similar to the failure mode of an indirect Brazilian tensile test. This
implies that fracture is controlled by the tensile strength.
That kind of fracture is similar to what is documented in real
rockfall impacts. Fig. 8 (bottom) depicts the impact of a block following
the 2015 earthquake (Mw = 6.5) in Lefkada island, Greece, that was
incidentally recorded by a car dashboard video camera.
In these test sets, the rebounds of the blocks were significantly lower
after each repetition. This was also reflected in the COR values, which
were gradually reduced. Fracture occurred after 3–4 drops, indicating
the initiation of internal cracking and its propagation during the re-
petitions. However, no visible macroscopic signs of cracking appeared
until fracture.
In general, the fracture of a spherical block at impact depends on the
impact velocity and the block's diameter.42 According to Salman et al.43
for aluminium oxide particles, the first observable damage typically
occurred at approximately 15 ms−1 for a diameter of 5 mm and at
100 ms−1 for a diameter 0.4 mm . These results indicate that the de-
formation during impact is connected to the momentum of the colliding
block. Therefore energy dissipation, or the COR that describes it, must
be a function of both impact velocity and mass.

4. A new semi-empirical methodology for the estimation of COR

The momentum describes the intensity of the impact and is defined

as the product of mass and absolute translational velocity. Therefore, it
combines the effects of those parameters on the COR in the simplest
manner. Fig. 9 depicts vCOR as a function of the incident momentum (Mi )
Fig. 7. vCOR as a function of impact surface Schmidt hardness (Rs ) for the
for the four materials tested, where the block impacts on a surface of
sandstone spheres.
the same material. Thus, the Schmidt Hardness (R) for the block (Rb ) is

46
P. Asteriou, G. Tsiambaos International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50

Fig. 9. vCOR as a function of incident momentum from all tests performed with
blocks and impact surfaces of the same material.
Fig. 10. One-to-one comparison of the estimated COR values from Eq. (18)
⋆
(vCOR ) against those obtained from the laboratory tests (vCOR ).
the same as that of the impact surface (Rs ).
A significant correlation becomes apparent: for a given momentum
magnitude, the values of vCOR , originating from the same material, co-
incide independently of the mass or the incident velocity. For each
material, this trend (dashed lines in Fig. 9) can be described by a hy-
perbolic function of the form:

vCOR = aMib + 1 (17)

where a and b are the fitting parameters summarized in Table 6.

It is seen that the coefficient of determination, R2 , has high values
(>0.9), except for the tests performed with marble. This is attributed to
the limited range of the incident momentum in those tests, as releasing
the marble blocks from a higher altitude resulted in fragmentation.
However, the trend and the values obtained for the parameters a, b are
in accordance with those calculated from the other materials.
Moreover, it is seen that the parameter a is inversely related to R but
that the parameter b presents a negligible variation and therefore can
be considered constant. The function used to determine parameter a
was obtained with a least squares method over R. Hence, the following
semi-empirical correlation is proposed that accounts for the material ⋆
Fig. 11. One-to-one comparison of the estimated vCOR (calculated with the
type and the intensity of the impact.
impact surface hardness, Rs ) against the experimental vCOR obtained from the
⋆
vCOR = (0.027R − 1.536) Mi0.244 + 1 tests between different surface and block material (Data are grouped according
(18)
to the hardness ratio, Rs / Rb ).
⋆
Fig. 10 compares, in a one-to-one sense, the estimated value (vCOR )
from Eq. (18) with those obtained from the experimental tests (vCOR ). because the data points deviate significantly from the 45° line. How-
Since all values are close to the 45° line, Eq. (18) describes the re- ever, a systematic behaviour is seen: when the ratio of Rs / Rb is below
bounding of the blocks adequately. Moreover, the relative error is ⋆
unity, vCOR is underestimated, and vice versa. Therefore, a correction
bounded within a ± 10% range (dashed lines in Fig. 10). factor, chr , that considers the hardness ratio can be included.
As previously shown (Fig. 7), when the impact surface material is ⋆
Setting chr as the ratio of vCOR to vCOR and examining its relation to
different from the material of the block (Rs ≠ Rb ), vCOR is affected, and the hardness ratio (Fig. 12), we get a linear expression of chr as a
the applicability of Eq. (18) becomes questionable. In Fig. 11, a one-to- function of Rs / Rb , which is described by Eq. (19) (dashed line in
⋆
one comparison is performed between vCOR (estimated by the Schmidt Fig. 12).
hardness of the impact surface, Rs ) and the vCOR obtained from the tests
1 R
performed with different block and surface materials. The values are = 0.491 s + 0.489
chr Rb (19)
grouped according to the hardness ratio between the surface and the
block (Rs / Rb ). Moreover, this expression can be fine-tuned to obtain a result of
It is observed (Fig. 11) that Eq. (18) fails to predict vCOR in this case unity when the material of the impact surface and the block coincide:
2Rb
Table 6 crh =
Rs + Rb (20)
Fitting parameters a and b of Eq. (17), where R is the Schmidts hardness, R^2
the coefficient of determination By introducing this correction factor into the empirical model (Eq.
Material R a b R2 (21)), vCOR can be estimated as a function of the colliding entities’
Schmidt hardnesses and the incident momentum.
Concrete 33.4 − 0.602 0.243 0.924
⋆
Marble 38.2 − 0.407 0.23 0.673 vCOR = crh vCOR (21)
Epoxy 42.3 − 0.387 0.247 0.967 ⋆
Sandstone 44.7 − 0.262 0.233 0.926 A one-to-one comparison between the corrected COR , and the vCOR ,
experimentally determined vCOR is presented in Fig. 13, in a manner

47
P. Asteriou, G. Tsiambaos
⋆
Fig. 12. Ratio of experimental vCOR to the estimated vCOR as a function of col-
liding entities hardness ratio Rs / Rb .
⋆
Fig. 13. One-to-one comparison between the corrected COR , vCOR , and the ex-
perimentally determined vCOR for the tests with different materials.
similar to Fig. 11, where no correction is done. It becomes evident that
the estimation is drastically improved, as the data points approach the
45° line and the relative error is bounded within ± 10% (dashed lines in
Fig. 13).
5. Comparison of the proposed methodology with the existing
scaling approaches
The data obtained from this experimental study are used hereafter
to evaluate the scaling models that are currently used in practice. The
model used in most of the relevant software, presented by Eq. (8), in-
corporates incident velocity, a reference COR value and a reference
velocity. However, observing Fig. 6, it becomes evident that the re-
sponse differs; Eq. (8) results in a sigmoid curve, while the experimental
data are better described by a linear function. Additionally, COR de-
pends strongly on the block's mass (Fig. 6), which is not considered by
this model. Another problematic issue is selecting the reference COR ;
the values suggested in literature are based on a qualitative char-
acterization of the material that constitutes the slope surface, without
any consideration of the impact properties. Thus, this scaling factor
lacks the ability to effectively describe the effects of the impact's in-
tensity or those of the properties of the colliding entities.
The model proposed by Rammer et al.34 omits scaling for incident
velocities less than 10 ms−1 because the reference COR is usually ac-
quired from suggested values found in the literature. Since these values
are determined by experimental studies performed within that velocity
48
International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50
Fig. 14. Calibration of the model proposed by Bourrier and Hungr31 a. Best fit
for nCOR, ref = 0.785 and M0.5 = 3.5 kgms−1 where R2 = 0.88; and b. Relative error
on the predicted COR .
range, it is reasonable to assume that they already include the velocity
effect. However, based on the presented data, the velocity effect is also
significant for that range. This could explain the scatter that COR pre-
sents in the relevant literature, even for the same rock type, because
those values are rarely given with the impact properties. Additionally,
this model omits the mass dependency for the whole velocity range.
Hence, this model also cannot be used to describe the trends produced
by the presented experimental data.
The momentum-based COR definition suggested by Bourrier and
Hungr31 is, in principle, compatible with the experimental data ac-
quired in this study. The attempt to calibrate the term M0.5 of Eq. (14)
with the data originating from the epoxy block tests, is depicted in
Fig. 14. However, the relative error of the predicted COR presents non-
linearity, implying either that the proposed function is not appropriate
or that some extra parameters affect the response but are not considered
in Eq. (14). Furthermore, a model that is defined in terms of incident
momentum seems to be appropriate for the simultaneous scaling of
COR values according to incident velocity and mass.
6. Assessment of the proposed empirical model with experimental
data from the literature
The authors have conducted and published similar experimental
studies that can be used to assess the proposed empirical model.
First, in Fig. 15, the correlation of COR with R presented in Asteriou
et al.27 is plotted as a red line. In that study, the impact surface and
block materials were the same (i.e., R = Rs = Rb ), and the incident
momentum was approximately Mn, i = 0.085 kgms−1. Using the proposed
empirical model (Eq. (21)) with the same value of Mn, i results in the
black line shown in Fig. 15. These two lines have been derived from
different experimental studies conducted on similar scales. Since they
fit each other very well, it is concluded that the proposed empirical
model adequately predicts the COR .
P. Asteriou, G. Tsiambaos
Fig. 15. Assessment of the proposed empirical model compared with others in
the literature.
Additionally, the correlations presented by Richards et al.20 and
Ansari et al.30 are plotted in Fig. 15, under the assumption that the
impact surface and block materials are the same, that is, Rs = Rb = R . It
is seen that for all correlations, COR has a similar rate of change over R
but that there is a significant difference in the COR magnitude. This is
attributed to the incident momentum level. In the tests presented by
Asteriou et al.27 the incident momentum was the smallest, whereas in
Ansari et al.30 it was the highest amongst those three experimental
campaigns. This is reflected in Fig. 15 by the order in which those
correlations are plotted and is in accordance with main observation of
this study: independently of the material, COR reduces as the incident
velocity and/or mass increase.
Moreover, Giokari et al.44 performed free fall tests of spherical
blocks made from the cement grout that were released onto horizontal
surfaces of Schist with three different weathering degrees. It was found
that COR decreased as the weathering of the impact surface increased.
The degree of weathering was quantified with both the Schmidt hard-
ness and the Point load strength index, which both decrease as
weathering progresses. These tests were used to evaluate the relative
hardness correction factor crh because the blocks and the impacting
surfaces were of different materials. Using the proposed empirical
model, with the momentum imposed in44, the COR s estimated present
a relative error of less than 6% for all three test sets performed.
7. Discussion and conclusions
The Coefficient of Restitution (COR ) is a significant parameter in
rockfall modelling because it controls the rebound height of the block.
Its acquisition requires a difficult, time-consuming, and expensive field
test procedure. Therefore, it is common in practice to retrieve the COR
from suggested values that are connected to a qualitative description of
the material that forms the slope. Then, scaling methods can be im-
plemented to account for the effects of incident velocity and block mass.
However, the COR s derived from experimental studies and back
analyses of rockfalls differ significantly from the COR values found in
literature; the qualitative description of the material constituting the
slope can be vague in some cases; and the scaling methods are not well
documented and cannot be combined. Therefore, a better procedure for
acquiring COR would be useful for design practice, and this need forms
the aim of this experimental study.
To exclude some of the parameters that affect the response of the
block to the impact, a one-dimensional approach was taken. Free fall
drops of spherical blocks impacting onto planar surfaces were per-
formed to eliminate the effects of the impact angle; block shape and
configuration at impact; slope inclination and roughness; and angular
velocity. Small-scale tests were performed with four different material
types, varying incident velocities and block masses. The physical and
mechanical properties of materials tested were determined to correlate
49
International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50
them with the COR and to propose a semi-empirical model for its es-
timation. In total, 445 impact tests were performed and processed with
an in-house machine vision application that was developed for this
purpose, which provided fast, accurate and robust data.
Two major conclusions are drawn from this study. First, COR is
significantly affected by both incident velocity and block mass and can
be better interpreted using incident momentum. Second, COR is af-
fected by the Schmidt hardness of both the block and the impact sur-
face.
Based on the data acquired from this study, a semi-empirical model
is proposed that takes the aforementioned parameters into account.
This model adequately describes the response during impact for the
scale and configuration used. Moreover, it addresses the differences in
the COR magnitudes found in relevant laboratory studies. Extending
this semi-empirical model to account for more impact characteristics,
such as the impact angle and block shape, and for scale effects is part of
ongoing research.
Acknowledgements
This post-doctoral research was carried out with a scholarship
granted by the State Scholarships Foundation of Greece (IKY) under the
act “Supporting Post-Doctoral Researchers” of the Operational Program
“Human Resources Development, Education and Life Lifelong
Learning” (Thematic Priority Axes 6, 8, 9) and is co-funded by the
European Social Fund (ESF) and the Greek Government.
References
1. Goldsmith W. Impact: The theory and physical behavior of colliding solids. Arnold; 1960.
2. Imre B, Räbsamen S, Springman S. A coefficient of restitution of rock materials.
Comput Geosci. 2008;34(4):339–350.
3. Giani G. Rock Slope Stability Analysis. CRC Press; 1992.
4. Labiouse V, Heidenreich B. Half-scale experimental study of rockfall impacts on
sandy slopes. Nat Hazards Earth Syst Sci. 2009;9(6):1981–1993.
5. Chau K, Wong R, Wu J. Coefficient of restitution and rotational motions of rockfall
impacts. Int J Rock Mech Min Sci. 2002;39(1):69–77.
6. Descoeudres F, Zimmermann T, et al. Three-dimensional dynamic calculation of
rockfalls. In: 6th ISRM Congress. International Society for Rock Mechanics; 1987.
7. Pfeiffer T, Bowen T. Computer simulation of rockfalls. Bull Assoc Eng Geol.
1989;26(1):135–146.
8. Hoek E. Rockfall - a program in Basic for the analysis of rockfalls from slopes.
Unpublished notes - University of Toronto/Golder Associates; 1987.
9. RocScience. Advanced Tutorial: Determining Input Parameters for A RocFall
Analysis. www.rocscience.com; 2003.
10. Heidenreich B. Small-and Half-scale Experimental Studies of Rockfall Impacts on Sandy
Slopes [Ph.D. Thesis]. EPFL; 2004.
11. Turner A, Schuster R. Rockfall: Characterization and Control. Transport Research
Board of the National Academies; 2012.
12. Asteriou P. Investigation of the Geotechnical Parameters Which Control Rockfalls in Greek
[Ph.D. Thesis]. NTUA; 2016.
13. Piteau D, Clayton R. Discussion of paper computerized design of rock slopes using
interactive graphics for the input and output of geometrical databy cundall, p. In:
Proceedings of the 16th Symposium on Rock Mechanics, Minneapolis, USA. 1977:
62-63.
14. Chau K, Wong R, Lee C. Rockfall problems in hong kong and some new experimental
results for coefficients of restitution. Int J Rock Mech Min Sci. 1998;35(4):662–663.
15. Azzoni A, Rossi P, Drigo E, Giani G, Zaninetti A. In situ observation of rockfall
analysis parameters. Istituto sperimentale modelli e strutture; 1992.
16. Evans S, Hungr O. The assessment of rockfall hazard at the base of talus slopes. Can
Geotech J. 1993;30(4):620–636.
17. Budetta P, Santo A. Morphostructural evolution and related kinematics of rockfalls in
campania (Southern Italy): a case study. Eng Geol. 1994;36(3):197–210.
18. Robotham M, Wang H, Walton G. Assessment of risk from rockfall from active and
abandoned quarry slopes. Int J Rock Mech Min Sci Geomech Abstr. 1995;5:237A.
19. Peng B. Rockfall Trajectory Analysis: Parameter Determination and Application [Master’s
thesis]. University of Canterbury. Geological Science; 2000.
20. Richards L, Peng B, Bell D. Laboratory and field evaluation of the normal coefficient
of restitution for rocks. In: Proceedings of Eurock; 2001: 149-56.
21. Giani G, Giacomini A, Migliazza M, Segalini A. Experimental and theoretical studies
to improve rock fall analysis and protection work design. Rock Mech Rock Eng.
2004;37(5):369–389.
22. Topal T, Akin M, Ozden A. Analysis and evaluation of rockfall hazard around afyon
castle, Turkey. In: Proceedings of the 10th international congress IAEG; 2006.
23. Giacomini A, Buzzi O, Renard B, Giani G. Experimental studies on fragmentation of
rock falls on impact with rock surfaces. Int J Rock Mech Min Sci. 2009;46(4):708–715.
24. Giacomini A, Thoeni K, Lambert C, Booth S, Sloan S. Experimental study on rockfall
P. Asteriou, G. Tsiambaos
drapery systems for open pit highwalls. Int J Rock Mech Min Sci. 2012;56:171–181.
25. Saroglou H, Marinos V, Marinos P, Tsiambaos G. Rockfall hazard and risk assess-
ment: an example from a high promontory at the historical site of monemvasia,
greece. Nat Hazards Earth Syst Sci. 2012;12(6):1823–1836.
26. Asteriou P, Saroglou H, Tsiambaos G. Geotechnical and kinematic parameters af-
fecting the coefficients of restitution for rock fall analysis. Int J Rock Mech Min Sci.
2012;54:103–113.
27. Asteriou P, Saroglou H, Tsiambaos G. Rockfalls: influence of rock hardness on the
trajectory of falling rock blocks. Bulletin of the Geological Society of Greece, Greece,
XLVII 2013.
28. Sabatakakis N, Depountis N, Vagenas N. Evaluation of rockfall restitution coeffi-
cients. In: Lollino G, Giordan D, Thuro K, Carranza-Torres C, Wu F, Marinos P,
Delgado C, eds. Engineering Geology for Society and Territory-Volume 2. Springer;
2015:2023–2026.
29. Rayudu D. Computer Simulation of Rockfalls-application to Rockfalls at Fox Glacier, West
Coast, New Zealand [Ph.D. Thesis]. Lincoln University; 1997.
30. Ansari M, Ahmad M, Singh R, Singh T. Correlation between schmidt hardness and
coefficient of restitution of rocks. J Afr Earth Sci. 2015;104:1–5.
31. Bourrier F, Hungr O. Rockfall dynamics: a critical review of collision and rebound
models. Wiley Online Library; 2011.
32. Dorren L. Rockyfor3d (v5.2) revealed - transparent description of the complete 3d
rockfall model. Tech. Rep.; www.ecorisq.org; 2015.
33. Ushiro T, Shinohara S, Tanida K, Yagi N. A study on the motion of rockfalls on
slopes. In: Proceedings of the 5th symposium on impact problems in civil
50
International Journal of Rock Mechanics and Mining Sciences 106 (2018) 41–50
engineering; 2000: 91-96.
34. Rammer W, Brauner M, Dorren L, Berger F, Lexer M. Evaluation of a 3-d rockfall
module within a forest patch model. Nat Hazards Earth Syst Sci. 2010;10(4):699.
35. Johnson K. Contact Mechanics. Cambridge, UK: Cambridge University Press; 1985.
36. Asteriou P, Tsiambaos G. Empirical model for predicting rockfall trajectory direction.
Rock Mech Rock Eng. 2016;49(3):927–941.
37. Ulusay R, Hudson J. The complete ISRM suggested methods for rock characteriza-
tion, testing and monitoring: 1974–2006. International Society for Rock Mechanics,
Commission on Testing Methods; 2007.
38. Canny J. A computational approach to edge detection. IEEE Trans Pattern Anal Mach
Intell. 1986;6:679–698.
39. Umbach D, Jones K. A few methods for fitting circles to data. IEEE Trans Instrum
Meas. 2003;52(6):1881–1885.
40. Feng Y, Goree J, Liu B. Errors in particle tracking velocimetry with high-speed
cameras. Rev Sci Instrum. 2011;82(5):053707.
41. Asteriou P, Saroglou H, Tsiambaos G. Rockfall: scaling factors for the coefficient of
restitution. In: Kwaśniewski M, Łydżba D, eds. Rock Mechanics for Resources, Energy
and Envi-ronment. London: Taylor & Francis Group; 2013:195–200.
42. Higa M, Arakawa M, Maeno N. Size dependence of restitution coefficients of ice in
relation to collision strength. Icarus. 1998;133(2):310–320.
43. Salman A, Reynolds G, Fu J, et al. Descriptive classification of the impact failure
modes of spherical particles. Powder Technol. 2004;143:19–30.
44. Giokari S, Asteriou P, Saroglou C, Tsiambaos G. Rockfalls: Effect of Slope Surface
Weathering on the Coefficients of Restitution. Springer International Publishing; 2015.