Transportation Geotechnics 24 (2020) 100371

Contents lists available at ScienceDirect

Transportation Geotechnics
journal homepage: www.elsevier.com/locate/trgeo

Resilient modulus prediction of RAP using the Contact Dynamics Method T

a,⁎ a a b
Laura Gaillard , Juan Carlos Quezada , Cyrille Chazallon , Pierre Hornych
a
Laboratoire des sciences de l’ingénieur, de l’informatique et de l’imagerie, ICUBE, UMR 7357, CNRS, INSA de Strasbourg, Université de Strasbourg, 24 Boulevard de la
Victoire 67084 Strasbourg Cedex, France
b
MAST-LAMES, Université Gustave Eiffel, IFSTTAR, F-44344 Bouguenais, France

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

Keywords: The Optimal Recycling of Reclaimed Asphalts in low-traffic Pavements (ORRAP) project concerns a cold re-
Reclaimed asphalt pavement cycling of 100% Reclaimed Asphalt Pavement (RAP) without binder addition, in base and subbase layers of low-
Resilient modulus traffic pavements. The mastic coating of aggregates induces a viscoelastic behaviour and changes the resilient
Discrete Element approach modulus with frequency and temperature. This modulus is obtained with repeated load triaxial tests and is a key
Burgers’ model
element for pavement design. In this context, this paper presents a numerical alternative to predict this influence
with a discrete approach. Contact Dynamics (CD) simulations are used to reproduce the resilient modulus test on
a set of rigid spherical particles with a viscoelastic contact, based on the Burgers’ model. The tests were carried
out with cylindrical samples at several frequencies for two temperatures (20 °C and 40 °C). The proposed model
is calibrated in a given range of temperatures and frequencies regarding the experimental data, and then used to
predict resilient modulus values at unreachable frequencies in the laboratory, with one set of parameters for each
temperature. Finally, using the Williams-Landel-Ferry (WLF) equation, a master curve of resilient modulus at
20 °C as a function of the frequency was established. The Time-Temperature Superposition Principle (TTSP)
reveals that the values at high frequencies and 40 °C correspond to the data at low frequencies and 20 °C.

Introduction of RAP under repeated cyclic loadings. However, this experimental

process is not able to identify the micro-scale behaviour. Therefore, a
Reclaimed Asphalt Pavement (RAP) comes from the demolition of numerical model with a discrete approach was developed to complete
asphalt road layers. Usually this material is recycled in hot the experimental campaign. The main modelling input is the capability
[7,20,22,23,32,33] or cold recycling [3,15,21,31], but RAP is seldom of the model to show that the RAP material, even when it is used in an
used without any addition. Based on a Swedish process [18], the Op- unbound state, follows the time temperature principle of full asphalt
timal Recycling of Reclaimed Asphalts in low-traffic Pavements concrete. To this end, RLTT modelling has been carried out at high
(ORRAP) project develops a cold recycling of 100% RAP, polluted with frequencies. Indeed, the Contact Dynamics (CD) method enables the
PAH (Polycyclic Aromatic Hydrocarbons), without any binder addition, modelling of aggregates and their viscoelastic contact.
in base and subbase layers of low-traffic pavements. Indeed, for PAH The following section describes the tested material and the de-
content between 50 and 500 mg/kg, in France, only cold recycling can termination of the resilient modulus from laboratory tests. Then, the
be performed or landfilling. Thus, for a better reuse of this material numerical procedures, in particular the contact model, and the resilient
than landfilling, it is necessary to evaluate the mechanical perfor- test modelling are detailed. Finally, a time-temperature equivalency is
mances of RAP with large PAH content. established using numerical simulations.
RAP is a complex material with aggregates coated with mastic of
bitumen and filler. Due to this binder phase, the material is sensitive to Laboratory work
the loading frequency and temperature. Indeed, this viscoelastic phase
has a key role in the macroscopic behaviour. Tested material
The elastic or resilient modulus is an important element to char-
acterise the mechanical behaviour of bituminous materials and for The tested material is a RAP, which comes from the demolition of
pavement design [1]. In laboratory, the Repeated Load Triaxial Test wearing courses from Strasbourg and its suburbs, and is defined by 25
(RLTT) has been used to quantify the macro-scale mechanical response RAP 0/14 [2]. Thus, the maximal size of aggregates is 25 mm and the

⁎
Corresponding author.
E-mail address: laura.gaillard@insa-strasbourg.fr (L. Gaillard).

https://doi.org/10.1016/j.trgeo.2020.100371
Received 15 December 2019; Received in revised form 4 May 2020; Accepted 11 May 2020
Available online 12 May 2020
2214-3912/ © 2020 Elsevier Ltd. All rights reserved.
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

Fig. 1. Particle size distributions.

Table 1 cycles), the permanent strains reach stable values, and the resilient
Characteristics of RAP. behaviour of the material can be observed. The RLTT have been per-
Bulk density [Mg/m3] 2.48
formed with cylindrical specimens (320 mm height –160 mm diameter)
Proctor maximum dry density [Mg/m3] 2.02 at a water content of 3.9% (Proctor optimum water content minus 2%),
Proctor optimum water content [%] 5.86 at several frequencies and for two temperatures, 20 °C and 40 °C. We
Binder content [%] 4.36 present hereafter the triaxial cell used, which enables to control and
Needle penetrability [1/10 mm] 13
measure the temperature inside the cell, as well as the axial and radial
Ring & Ball [°C] 69.4
PAH content [mg/kg] 298 strains. Before the test, the sample is kept at the target temperature in
the cell for 10 h. This step ensures a homogeneous temperature inside
the sample. The temperature during the RLTT remains constant (see
grading of particles after binder extraction is 0/14, as shown in Fig. 1. Fig. 3).
The main characteristics of this material are summarised in Table 1 and The strain measurement is based on three Hall Effect sensors fixed
Fig. 2 displays the modified Proctor compaction curve. on the sample after compaction with inner supports: two of them
measure the vertical strains and are located on the lateral surface, one
of them measures the radial strains. For more details, please refer to the
Determination of resilient modulus
paper of Gaillard et al. [14].
For each couple temperature-frequency, a constant confining pres-
The resilient modulus Mr is determined by means of Repeated Load
sure σ3 and a sinusoidal deviator stress q is applied to the sample
Triaxial Tests (RLTT) according to the European Standard EN 13286-7
(Fig. 4). Table 2 details the test parameters.
[1]. After application of a sufficient number of load cycles (here 30,000

Fig. 2. Modified Proctor compaction curve.

2
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

Fig. 3. Temperature controlled triaxial cell and instrumented specimen [14].

The resilient modulus Mr (Fig. 5) is defined as: Table 2

Test parameters.
Mr = Δq/ ε1r = (qmax − qmin )/ ε1r (1)
Temperature [°C] Frequency [Hz] Confining pressure Deviator stress
where ε1r is the resilient axial strain. σ3 [kPa] [kPa]

qmin qmax
Experimental results
20 0.1 – 0.5 – 1 – 2 – 5 70 10 350
– 10
Fig. 6 presents the evolution of the deviator stress as a function of 40 0.1 – 0.5 – 1 – 2 70 10 210
the resilient axial strains over a cycle of loading-unloading. Using Eq.
(1), the resilient modulus Mr is determined for each couple tempera-
ture-frequency (Fig. 7). The maximal experimental frequency is 10 Hz Contact Dynamics (CD) method
at 20 °C and 2 Hz at 40 °C, and the minimal is 0.1 Hz for both tem-
peratures. Significant variations of modulus with frequency and tem- The CD method is a discrete element approach for the simulation of
perature are observed. non-smooth granular dynamics. The non-smoothness refers to various
degrees of discontinuity in the velocities and contact forces arising in a
system composed of rigid particles. In this method, the equations of
Numerical procedures
motion for each particle are formulated as differential inclusions in
which velocity jumps replace accelerations [19,25,26,30]. The prin-
This section presents briefly the Contacts Dynamics method and the
cipal difference between this method and classic DEM approaches lies
viscoelastic contact model used to simulate numerically the resilient
in the expression of the contact laws as complementary relations
modulus test.

Fig. 4. Principle of RLTT [17].

3
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

known, and iteratively updating the forces and velocities until a con-
vergence criterion is fulfilled.
For the numerical simulations, the LMGC90 software was used [12].
This software is able to model rigid or deformable particles with regular
or irregular shapes.

Contact model

The viscoelastic contact model employed in the simulations was

developed by Quezada and Chazallon [28], which models a granular
material with a viscoelastic behaviour. This material is represented by
rigid particles, which are coated with a viscoelastic phase (Fig. 8). Thus,
the proposed contact model mixes the original CD formulation for rigid
particles, to ensure the non-interpenetration criterion, with a regular-
Fig. 5. Definition of resilient modulus after the conditioning phase. ized viscoelastic law [6,29], based on Burgers’ model, acting on distant
contacts.
between the contact forces and velocities. This approach ensures basi- The Burgers’ model simulates the mastic phase of a RAP. This vis-
cally the mutual exclusions between particles without introducing coelastic model presents few parameters and is able to reproduce the
regularized laws often used in explicit methods such as the distinct dynamic properties, creep and relaxation, of asphalt mixtures
element method [9,11,10] or molecular dynamics [4,16,27]. The uni- [5,8,13,24,35]. This viscoelastic model is the association in series of a
lateral contact interactions and Coulomb friction law are treated as Kelvin-Voigt model and a Maxwell model. At the micro-scale, Km and
complementarity relations or set-valued contact laws. Using the CD Cm are the stiffness and the viscosity of Maxwell, while Kk and Ck are
formulation, the equation of dynamics can be written as: the stiffness and the viscosity of the Kelvin-Voigt section. Fig. 9 shows
the schema of the Burgers’ model in the local frame for the normal and
U+ = U− + W R+dt tangential components. In this frame, the normal vector n identifies the
_ _ _ _
_ (2) direction where the minimal distance between two particles at contact
With: δn is located. The normal components of the microscopic parameters
can be determined by:
– U+ and U− are the relative velocities at times t and t + dt, where dt 2
Kmn = πrmin Emn/ l 0 (3)
_ _
is the time step,
2
– W is the Delassus matrix, which contains the values of the inverse Cmn = πrmin ηmn / l 0 (4)
_
_ 2
reduced inertia, K kn = πrmin Ekn/ l 0 (5)
– R+ is the vector of contact forces at time t + dt. 2
_ Ckn = πrmin ηk n / l 0 (6)

In this method, the time integration algorithm is implicit. All ki- where rmin is the minimum radius of two particles in contact and l0 the
nematic constraints implied by frictional contacts between particles are initial distance between the centres of particles, Em and ηm are the
simultaneously taken into account, together with the equations of dy- macroscopic Maxwell stiffness and viscosity, while Ek and ηk are the
namics, in order to determine all the velocities and contact forces in the macroscopic Kevin-Voigt stiffness and viscosity.
system. Thus, this kinematic problem is solved by an iterative proce- In this contact frame, it is possible to add two orthogonal vectors to
dure based on the nonlinear Gauss-Seidel method. It consists of solving define the tangential directions t and s. The tangential properties of the
a single contact problem with other contact forces being treated as model following both tangential directions t and s are set as equal,

Fig. 6. Stress - strain curves at (a) 20 °C (b) 40 °C.

4
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

Fig. 7. Experimental resilient moduli.

where:

– Uk+ is the complementary relative velocity generated by the Kelvin-

_
Voigt model,
– Um+K and Um+C are the complementary relative velocities generated
_ _
by the stiffness and viscosity parts of the Maxwell model

Fig. 8. Numerical representation of a granular material with a viscoelastic The contact forces generated within the Burgers’ model can be de-
behaviour. termined for each section as a function of the complementary relative
velocities, as follows:
considering an isotropic behaviour.
R+ = Rk+ = Rm+K = Rm+C
Thus, the tangential components of the model parameters are esti- _ _ _ _ (12)
mated as a function of the normal components and the Poisson’s ratio υ,
equal to 0.35, as: Rk+ = −Kk δ +
− Ck Uk+
_ _ k _ (13)
Kmt = Kms = Kmn/(2(1 + υ)) (7)
Rm+K = −Km δ +
_ m (14)
Cmt = Cms = Cmn/(2(1 + υ)) (8) _ K

K kt = K k s = K kn/(2(1 + υ)) (9) Rm+C = −Cm Um+C

_ _ (15)
Ckt = Ck s = Ckn/(2(1 + υ)) (10) Then, by replacing the complementary velocity terms obtained from
Eqs. (13) to (15) into Eq. (11), leads to a modified equation of motion in
By introducing an initial normal gap δ0 and a critical local reaction
the CD formulation, obtained as a function of the parameters of the
Rcrit, Fig. 10 describes the relations between local reactions and relative
Burgers’ model (Eq. (16)) as:
velocities for the viscoelastic model. The Rcrit value is computed as the
product between the yield stress value and the projected cross-section +
Kk δ R−
_ k _ 1 1 1⎞ +
of particles at contact. Thus, for the developed contact model, the vis- U+ = U − − + + (W − ⎛ ⎜ + + ) R dt
⎟

coelastic Burgers’ model acts together with the frictional contact law. _ _ Ck Km dt _ ⎝ K m dt C m Ck⎠ _
_ (16)
For neighboring contacts (δn > 0) the local reaction values R are
calculated from the Burgers’ model. For the case of frictional contact where R− is the initial contact force values during a time step. For more
(δn = 0) the produced local reactions are appended to the total reaction details about the contact model implementation, we invite the reader to
for each inter-particle contact. If the local reaction in one direction is refer to the article of Quezada and Chazallon [28].
larger than Rcrit, the link is lost and the frictional conditions are applied
instead. Resilient test modelling
Adding the complementary velocities generated into the Burgers’
model to the general CD formulation for frictional contacts (Eq. (2)), The main goal of these numerical simulations is to reproduce the
leads to the following equation of motion: resilient modulus values identified experimentally to perform numer-
ical tests at frequencies beyond the capacities of the experimental de-
U+ = U− + Uk+ + Um+K + Um+C + W R+dt vices. The influence of the micro-structure on the macro-scale response
_ _ _ _ _
_ _
_ (11) is not treated in detail in this work, but further analysis could focus on

5
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

Fig. 9. Burgers' model (a) Normal components (b) Tangential components [28].

this topic. compared with the laboratory specimen, while keeping the aspect ratio
at two.
Sample preparation

The cylindrical samples are composed of 25,150 rigid spherical Contact model
particles. The grading curve was cut at 2 mm, to avoid modelling fines. The viscoelastic contact models the coating of particles. To take in
The fines are considered as a part of the mastic, nevertheless the account the binder content of 4.36%, the particle size is decreased. The
amount of particles passing the different sieves has been kept. Fig. 11 radius is multiplied by a correction factor fcr (Eq. (17)).
displays the numerical and experimental particle size distributions and
Table 3 details the size and number of the generated particles. The fcr =(1−(TL/100). (ρs / ρL )1/3 (17)
generation of the sample requires several steps, which are described
below. where TL is the binder content (4.36%), ρs the initial density of particles
(2.48 Mg/m3) and ρL the density of binder (1.03 Mg/m3).
Deposit by gravity In order to keep the stone skeleton, the density of particles is in-
The particles are arranged randomly in a cubic lattice, into a cy- creased to 3.39 Mg/m3. Then, the contact model is activated between
linder confined between two plates. Then, a deposition by gravity is particles with a gap lower than 1 mm, which is the minimal radius of
carried out, with a drop height of five meters. The Coulomb friction law particles. For a particle-particle contact, the frictional contact law is
is applied for every contact, with a low coefficient of friction, equal to applied, with a coefficient of friction of 0.92, which is equivalent to the
0.2. This procedure generates a very dense sample, where a dynamic tangent of the internal friction angle (42.58°). Due to the small strains
compaction is not necessary. The cylindrical specimen is 0.05 m in during the resilient test, the interactions between particles are purely
diameter and 0.1 m in height. These dimensions have been reduced viscoelastic.

6
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

Table 3
Size and number of particles.
Particle size [mm] 2 4 6 8 10 12.5 16

Number of particles 22,432 2221 262 132 63 30 10

Fig. 12. Numerical specimen.

Stabilisation
To achieve the equilibrium state, an isotropic confining pressure of
70 kPa is applied on the lateral and top surfaces of the cylinder. Fig. 12
displays a snapshot of the numerical specimen.

Calibration of model parameters

This section discusses the procedure to determine the contact model

parameters to reproduce numerically the experimental resilient mod-
ulus values.

Fig. 10. Relations between velocities and contact forces for the Burgers’ model: Set of parameters for the temperature of 20 °C
(a) Normal components (b) Tangential components [28]. In order to calibrate the numerical model using the experimental
values, we first need an analytical model to describe the evolution of

Fig. 11. Numerical and experimental particle size distributions.

7
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

Fig. 13. Numerical calibration based on experimental data.

Table 4 Thus, the best fit macroscopic parameters was determined from Eq.
Macroscopic model parameters. (21) regarding only experimental values of the resilient modulus at 2, 5
Temperature [°C] Macroscopic model parameters
and 10 Hz. Here, the low frequencies were not taken into account for
the fit, due to significant computational time (over one month for
Em [Pa] ηm [Pa·s] Ek [Pa] ηk [Pa·s] 0.1 Hz). The curve fitting was performed by means of an iterative op-
timization algorithm based on the least squares method, implemented
20 7.92 × 107 3.41 × 108 2.96 × 108 1.39 × 107
40 3.97 × 107 1.71 × 108 1.48 × 108 6.96 × 106
in Python programming language, for a temperature of 20 °C. For each
iteration, the model parameters were adjusted in order to calibrate the
numerical results regarding the experimental data.
the resilient modulus as a function of frequency.
The Burgers model allows us to describe the evolution of the com- Set of parameters for the temperature of 40 °C
plex modulus norm as a function of frequency. In this way, we can The parameters were established from the set for the temperature of
associate the norm of the complex modulus with the values of the re- 20 °C. The ratio between the experimental resilient modulus values at
silient modulus. Both properties are the ratio between the axial stress 20 °C and 40 °C being constant, the parameters at 20 °C were divided by
and the axial strain. The complex modulus E* is defined by its norm this ratio. Indeed, the experimental device allows a maximal frequency
|E ∗|and the phase angle φ (Eq. (18)). of 2 Hz. At very low frequencies, the computational time is quite sig-
nificant.
E ∗ = |E ∗|. exp(iφ) = (σ0/ ε0). exp(iφ) (18)
With the axial stress σ and the axial strain ε defined as: Numerical simulations
σ (t ) = σ0 sin(ωt ) (19)
The tests were conducted by applying the same stresses as in the
ε (t ) = ε0 sin(ωt − φ) (20) experimental tests (Table 2) over one cycle. The numerical resilient
modulus was calculated by means of Eq. (1). The main objectives of the
For the RAP, the viscosity of bitumen does not involve a phase shift numerical work are to reproduce the experimental data and to check if
between the stress and the strain. Under cyclic loading, the norm of the this unbound material sensitive to temperature and loading frequencies
complex modulus |E ∗| and the phase angle φ can be written as a func- fulfils the TTSP.
tion of the macroscopic parameters (Eq. (21) and (22)). Fig. 13 presents the numerical results with the calibration based on
the laboratory tests and the analytical forecasted values. Table 4 sum-
|E ∗| = 1/|D∗| = 1/ D' (ω)2 + D'' (ω)2 (21)
marises the macroscopic model parameters for each temperature. One
φ = arctan(D''(ω)/ D'(ω)) (22) can note that the analytical model is able to fit the experimental values,
but fails in forecasting values outside the experimental range with a
where |D∗|
is the complex compliance, D’ and D’’ are the real and the flattened curve at higher frequencies. On the other hand, the numerical
imaginary parts of the complex compliance, and ω the pulsation, ob- calibration is quite satisfactory, but it deviates for the lower fre-
tained from Eqs. (23) and (24): quencies. Then, we have modelled the evolution of the resilient mod-
D' (ω) = 1/Em + Ek /(Ek2 + ω2ηk2) ulus at large frequencies, for that the numerical simulations were per-
(23)
formed until 14,910 Hz at 20 °C and 2600 Hz at 40 °C (Fig. 14). Beyond
D''(ω) = 1/ ωηm + ωηk /(Ek2 + ω2ηk2) these values, some difficulties with sample stability appear. Two si-
(24)
mulations for each couple temperature-frequency were performed

8
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

Fig. 14. Numerical simulations at large frequencies.

Fig. 15. Master curve at 20 °C.

(excepted at 746 Hz, 3728 Hz and 14,910 Hz). – The higher the temperature, the lower is the resilient modulus. The
Several observations can be made from these numerical results: cause is the mastic coating of aggregates, which is much more
sensitive at high temperatures.
– The resilient modulus values increase with frequency. Indeed, lower – The viscoelastic model is able to describe the evolution of the re-
frequency implies higher strains. silient modulus Mr values as a function of frequency.

9
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

−C1 (T − TR )
log(aT ) =
(T − TR) + C2 (25)

where TR is the reference temperature, C1 and C2 two constants and T

the temperature.
The RAP with only 4.36% is sensitive to frequency and temperature.
The TTSP, initially developed for viscoelastic materials, was applied to
the resilient modulus of RAP. Fig. 15 presents the master curve estab-
lished with the WLF law with a reference temperature of 20 °C, with
C1 = 45 and C2 = 185 °C. Fig. 16 shows the evolution of the coefficient
aT. The data at high frequencies and 40 °C correspond to the values at
low frequencies and 20 °C. The master curve defines the resilient
modulus values at frequencies unreachable in the laboratory, and gives
resilient moduli for low or high frequencies at 20 °C. This last in-
formation is of large interest for the design of RAP layers used as un-
bound material.
From the master curve at 20 °C, the behaviour at other temperatures
can be derived. Each frequency range is equivalent to the behaviour of
the RAP at one temperature (Fig. 17). By means of the translation
coefficient aT, the evolution of the resilient modulus at 10 °C, 20 °C,
Fig. 16. Translation coefficient. 30 °C and 40 °C is predicted in the 0.1 to 40 Hz frequency range. Fig. 18
shows the different isotherms.
The isotherms give the numerical values of the resilient modulus at
– The predictions at 0.1 Hz are less accurate. To improve these results,
each temperature between 0.1 Hz and 40 Hz. Consequently, the model
the parameters should be calibrated at this very low frequency.
parameters at 10 °C and 30 °C are estimated. Table 5 details the model
Nevertheless, this procedure requires very long computational
parameters for each temperature.
times.
– The rate of increase of the resilient modulus is similar at both
temperatures. Conclusions
– Regarding model parameters, higher temperature implies lower
stiffness and viscosity values. This paper reports the numerical modelling of the mechanical be-
haviour of a Reclaimed Asphalt Pavement (RAP). CD simulations of the
Time-Temperature Superposition Principle (TTSP) resilient modulus test were performed with the LMGC90 software to
validate and extrapolate experimental data. The tests were carried out
The TTSP identifies the temperature dependence of viscoelastic with cylindrical samples at several frequencies and two temperatures
material properties. This principle establishes the equivalence between (20 °C and 40 °C). RAP is an aggregate that is coated with bitumen.
frequency and temperature, by introducing a reduced variable. Thus, it Thus, this material is modelled by rigid spherical particles with a vis-
is possible to build a single curve of the norm of the complex modulus coelastic contact. The contact model is based on the Burgers’ model.
for a chosen reference temperature. This master curve is obtained by a Despite a low binder content of 4.36%, the RAP presents still a
translation of isotherms, with the translation coefficient log(aT). The viscoelastic behaviour. The proposed model validates the experimental
Williams-Landel-Ferry (WLF) equation [34] was used to define log(aT): values of resilient modulus Mr, but also allows to extrapolate the results

Fig. 17. Decomposition of the master curve at 20 °C.

10
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

Fig. 18. Isotherms.

Table 5 CRediT authorship contribution statement

Macroscopic model parameters at four temperatures.
Temperature [°C] Macroscopic model parameters
Laura Gaillard: Conceptualization, Software, Investigation, Writing
- original draft, Visualization. Juan Carlos Quezada:
Em [Pa] ηm [Pa·s] Ek [Pa] ηk [Pa·s] Conceptualization, Software, Writing - review & editing. Cyrille
Chazallon: Conceptualization, Writing - review & editing, Funding
10 8.84 × 107 7.76 × 1013 1.26 × 109 3.17 × 107
20 7.92 × 107 3.41 × 108 2.96 × 108 1.39 × 107
acquisition, Supervision. Pierre Hornych: Resources, Writing - review
30 6.02 × 107 2.13 × 108 2.84 × 108 6.25 × 106 & editing.
40 3.97 × 107 1.71 × 108 1.48 × 108 6.96 × 106
Declaration of Competing Interest

to frequencies outside the experimental range contrary to the analytical

The authors declare that they have no known competing financial
model, which presents a flattened curve at higher frequencies. Indeed,
interests or personal relationships that could have appeared to influ-
the viscoelastic modelling is able to describe the evolution of Mr, as a
ence the work reported in this paper.
function of the frequency, with one set of parameters for each tem-
perature. Using the Williams-Landel-Ferry (WLF) equation, a master
References
curve was established at 20 °C. The Time-Temperature Superposition
Principle (TTSP) reveals that the values of Mr at high frequencies and
[1] AFNOR. Mélanges avec ou sans liant hydraulique – Partie 7: Essai triaxial sous
40 °C correspond to the data at low frequencies and 20 °C. From the charge cyclique pour mélanges sans liant hydraulique. European Standard NF EN
master curve at 20 °C, the behaviour at 10 °C and 30 °C has been de- 13286-7; July 2004.
rived. [2] AFNOR. Mélanges bitumineux – Spécifications pour le matériau – Partie 8 : Agrégats
d’enrobés. European Standard NF EN 13108-8; October 2016.
These results are very encouraging and need to be extended to a [3] Betti G, Airey G, Jenkins K, Marradi A, Tebaldi G. Active filler’s effect on in situ
wider temperature range, in order to validate totally the TTSP. This performances of foam bitumen recycled mixtures. Road Mater Pavement Des
paper presents the first results, which provide a basis for designing RAP 2016;18(2):281–96.
[4] Brilliantov NV, Spahn F, Hertzsch JM, Pöschel T. Model for collisions in granular
layers in a pavement knowing the loading and its frequency. The nu- gases. Phys Rev E 1996;53(5):5382.
merical determination of the master curve obtained at 20 °C also needs [5] Cai W, McDowell GR, Airey GD. Discrete element visco-elastic modelling of a rea-
to be validated at other water content. It leads to numerous perspec- listic graded asphalt mixture. Soils Found 2014;54(1):12–22.
[6] Cambou B, Jean M, Radjaï F, editors. Micromechanics of granular materials. John
tives. Thus, the further numerical work will focus on the effects of water Wiley & Sons; 2013.
content and particle size distribution on Mr, as well as the rutting re- [7] Colbert B, You Z. The determination of mechanical performance of laboratory
sistance of RAP through the modelling of laboratory and field tests. produced hot mix asphalt mixtures using controlled RAP and virgin aggregate size
fractions. Construct Build Mater 2011;26(1):655–62.
[8] Collop A, McDowell GR, Lee Y. Modelling dilation in an idealized asphalt mixture
using discrete element modelling. Granular Matter 2006;8:175–84.
Funding [9] Cundall PA. A computer model for simulating progressive, large scale movement in
blocky rock systems. In: symp ISRM nancy, France, Proc.; 1971.
[10] Cundall PA. Formulation of a three-dimensional distinct element model-Part I. A
This work is part of the ORRAP (Optimal Recycling of Reclaimed scheme to detect and represent contacts in a system composed of many polyhedral
Asphalts in low-traffic Pavements) project funded by ERDF - INTERREG blocks. Int J Rock Mech Min Sci Geomech Abst 1988.
V (3.1 ORRAP), which studies the cold recycling of reclaimed asphalts [11] Cundall PA, Strack OD. A discrete numerical model for granular assemblies.
Géotechnique 1979;29(1):47–65.
aggregates without binder addition. [12] Dubois F, Jean M. LMGC90 une plateforme de développement dédiée à la
modélisation des problèmes d’interaction. In: Proceedings of Sixième Colloque
National en Calcul des Structures; 2003 May 20–23; Giens, France. 2003. 1. p.
111–8.

11
L. Gaillard, et al. Transportation Geotechnics 24 (2020) 100371

[13] Feng H, Pettinari M, Hofko B, Stang H. Study of the internal mechanical response of velocity in DEM of constant strain rate tests on asphalt. Geomech Geoeng: Int J
an asphalt mixture by 3-D discrete element modeling. Constr Build Mater 2009;4(2):171–4.
2015;77:187–96. [25] Moreau JJ. Unilateral contact and dry friction in finite freedom dynamics. In:
[14] Gaillard Laura, Chazallon Cyrille, Hornych Pierre, Quezada Juan Carlos, Raab Moreau JJ, Panagiotopoulos PD, editors. Nonsmooth mechanics and applications.
Christiane. Thermo-hydro-mechanical behaviour of cold reclaimed asphalt ag- Springer; 1988. p. 1–82.
gregates without binder addition. Road Mater Pavement Des [26] Moreau JJ. Some numerical methods in multibody dynamics: application to gran-
2019;20(sup1):S49–63. https://doi.org/10.1080/14680629.2019.1587490. ular materials. Eur J Mech A/Solids 1994;13(4):93–114.
[15] Guatimosim FV, Vasconcelos KL, Bernucci LLB, Jenkins KJ. Laboratory and field [27] Pöschel T, Buchholtz V. Molecular dynamics of arbitrarily shaped granular parti-
evaluation of cold recycling mixture with foamed asphalt. Road Mater Pavement cles. J Phys I France 1995;5(11):1431–55.
Des 2016;19(2):385–99. [28] Quezada JC, Chazallon C. Complex modulus modeling of asphalt concrete mixes
[16] Herrmann HJ, Luding S. Modeling granular media on the computer. Continuum using the Non-Smooth Contact Dynamics method. Comput Geotech
Mech Thermodyn 1998;10(4):189–231. 2020;117(103255).
[17] Hornych P, El Abd A. Work package 5 Performance-based specifications - Selection [29] Radjaï F, Dubois F. Discrete-element modeling of granular materials. Wiley-Iste;
and evaluation of models for prediction of permanent deformations of unbound 2011.
granular materials in road pavements. Extract of SAMARIS project. Document [30] Radjaï F, Richefeu V. Contact dynamics as a nonsmooth discrete element method.
n°SAM-05-DE-10; 2004. Mech Mater 2009;41(6):715–28.
[18] Jacobson T, Waldemarson A. Återvinning av asfaltgranulat vid motorvägsbygget på [31] Sangiorgi C, Tataranni P, Simone A, Vignali V, Lantieri C, Dondi G. A laboratory and
E4, Markaryd- Uppföljningar åren 2005–2010. VTI notat 25-2010; 2011. field evaluation of Cold Recycled Mixture for base layer entirely made with
[19] Jean M. The non smooth contact dynamics method. Comput Methods Appl Mech Reclaimed Asphalt Pavement. Constr Build Mater 2017;138:232–9.
Eng 1999;177(3–4):235–57. [32] Tran N, West R, Taylor A, Willis R. Evaluation of moderate and high RAP mixtures
[20] Kalman, B. Deliverable 7.3 – project final report. RE-ROAD – end of life strategies of at laboratory and pavement scales. Int J Pavement Eng 2015;18(9):851–8.
asphalt pavements; 2013. Available from: http://re-road.fehrl.org/. [33] Valdés G, Pérez-Jiménez F, Mirό R, Martίnez A, Botella R. Experimental study of
[21] Kuna K, Airey G, Thom N. Mix design considerations of foamed bitumen mixtures recycled asphalt mixtures with high percentages of reclaimed asphalt pavement
with reclaimed asphalt pavement material. Int J Pavement Eng (RAP). Constr Build Mater 2010;25(3):1289–97.
2016;18(10):902–15. [34] Williams ML, Landel RF, Ferry JD. The temperature dependence of relaxation
[22] Li X, Marasteanu MO, Williams RC, Clyne T. Effect of reclaimed asphalt pavement mechanisms in amorphous polymers and other glass-forming liquids. J Am Chem
(proportion and type) and binder grade on asphalt mixtures. Transp Res Rec Soc 1995;77(14):3701–7.
2008;2051(1):90–7. [35] Yu H, Shen S. A micromechanical based three-dimensional DEM approach to
[23] McDaniel RS, Shah A. Use of reclaimed asphalt pavement (RAP) under superpave characterize the complex modulus of asphalt mixtures. Constr Build Mater
specifications (with discussion). J Assoc Asphalt Paving Technol 2003;72:226–52. 2013;38:1089–96.
[24] McDowell GR, Collop AC, Wu JW. A dimensional analysis of scaling viscosity and

12