Вы находитесь на странице: 1из 22

Road Materials and Pavement Design, 2015

Vol. 16, No. 2, 256–276, http://dx.doi.org/10.1080/14680629.2014.990402

Dynamic response of flexible pavements at vehicle–road interaction


P. Khavassefat ∗ , D. Jelagin and B. Birgisson

Highway and Railway Engineering, KTH Royal Institute of Technology, Brinellvägen 23, Stockholm 100
44, Sweden

(Received 21 January 2014; accepted 16 November 2014 )

In the present paper a robust and general computational framework that captures the dynamic
response of flexible pavements to a moving vehicle is presented. A finite element method
is relied upon in order to establish the response function for a linear viscoelastic pavement
structure with dynamic effects taken into account. In order to characterise the dynamic loads
induced on the pavement by moving traffic, a quarter car model combined with measured road
profiles is used. Once both the traffic loads and pavement response functions are known, the
stresses and strains induced in the pavement can be obtained in the frequency–wavenumber
domain through the convolution procedure. The computational procedure developed is applied
in the present study to evaluate the effect of the pavement surface roughness on the pavement
structure response to truck traffic loading. Stress field parameters governing fracture initiation
in asphalt layers are reported for two measured road roughness profiles. It is shown that the
dynamic effects at vehicle–road interaction may have a profound influence on the stresses
induced in flexible pavements; therefore, these effects need to be taken into account for the
accurate estimation of the road resistance to cracking.
Keywords: finite element; viscoelasticity; moving load; dynamic axle loads; road roughness;
flexible pavement

Introduction
Traffic loading is one of the major causes of distresses in flexible pavements. The two most com-
mon distress modes in pavements: rutting and fatigue cracking are to a great extent controlled by
the level of traffic-induced stresses and strains in the structure (cf. COST 333, 1999). An accurate
description of stress state induced by vehicles in pavements is crucial for prediction of pavement
performance in the field. In the absolute majority of current design and evaluation procedures
for flexible pavements, linear elastic analysis is used to calculate stresses and strains induced
by moving loads in the structure. The pavement structure is then represented as an assembly of
isotropic linear elastic layers infinite in horizontal directions and having finite thickness. Fur-
thermore, the loaded area is assumed to be circular, dynamic effects are neglected and only loads
normal to the pavement surface are considered.
However, the analysis based on the aforementioned assumptions may not provide accurate
qualitative and quantitative description of the pavement mechanical behaviour. The viscoelas-
tic behaviour of asphalt pavement plays a crucial role in pavement performance, especially
when it comes to the moving load. As it has been shown by several studies (cf. Arraigada,
Partl, Angelone, & Martinez, 2009; Kim, Roque, & Byron, 2009; Pouget, Sauzéat, Benedetto, &

*Corresponding author. Email: parisak@kth.se

© 2014 Taylor & Francis


Road Materials and Pavement Design 257

Olard, 2010, Saevarsdottir & Erlingsson, 2014), taking viscoelasticity into account results in a
more accurate description of pavement’s mechanical response as compared with the layered elas-
tic approach. Finite element modelling of pavement structures has been widely used by different
researchers (cf. Arraigada et al., 2009; Khavassefat, Jelagin, & Birgisson, 2012; Pouget et al.,
2010) due to the evident advantage of the method in terms of capability of taking into account
various aspects of material behaviour, arbitrary geometry of the loaded areas and different types
of loading and boundary conditions. For instance, it has been shown that taking into account
the viscoelastic response of pavement structures combined with the actual tyre–road contact area
shape leads to more accurate qualitative and quantitative predictions regarding pavement per-
formance in the field as compared with the traditional layered elastic approach (Al-Qadi, Yoo,
Elseifi, & Janajreh, 2005). Furthermore, the effect of the presence of tangential shear stresses at
the tyre–road interface on the stress state induced on the flexible pavement has been investigated
in several studies (cf. Al-Qadi, Wang, Yoo, & Dessouky, 2008; Drakos, Roque, & Birgisson,
2001; Wang & Al-Qadi, 2009). They found that the shear stresses on the pavement surface may
have a profound influence on the distress modes initiating close to the surface, that is, rutting
in the asphalt layer and top-down cracking. An accurate description of the normal and tangen-
tial contact stress distributions may not be achieved without performing full 3D analysis of the
pavement structure.
Although considering an accurate model with 3D geometry, more realistic material behaviour,
dynamics of the structure as well as accurate tyre contact stresses taken into account are essen-
tial for capturing many distress types in pavement structures, it comes at the expense of a much
longer computation time. Recently, Khavassefat et al. (2012) proposed a computationally effec-
tive incremental procedure allowing evaluation of stresses and strains induced in the viscoelastic
flexible pavement by the moving traffic. The procedure is based on the superposition princi-
ple and is computationally favourable, as it requires only reduced incremental problem to be
solved numerically. The computational procedure reported by Khavassefat et al. (2012) is how-
ever based on the quasi-static solution and thus is incapable of capturing the dynamic effects
at vehicle–road interaction. In the present study, the computational framework proposed by
Khavassefat et al. (2012) is extended further to account for dynamic aspects at vehicle–road
interaction. The ultimate goal of the present paper is to propose a comprehensive computational
framework evaluating the effects that road surface roughness, vehicle type and structural proper-
ties of the pavements have on the dynamic response and degradation of flexible pavements under
traffic loading.
The effect of dynamic vehicle–road interaction on the loads exerted on pavement is hardly
negligible, especially with the increase in traffic weight and density in recent years (cf. European
Commission, 2008). Furthermore, the vibrations resulting from dynamic loads are an impor-
tant contributing factor in issues such as noise emissions, safety and ride comfort (Bilgiri &
Way, 2014; Degrande & Lombaert, 2001). The deterioration of road surface during the road’s
service life is an important factor to be considered, as it results in higher surface roughness
and thus an increase in dynamic loads exerted on the pavement (Kropáč & Múčka, 2008). The
effect of surface roughness is initially small and increases with time as the surface becomes
rougher and consequently the development of different crack types on the road surface accelerates
considerably (Liu, 2001; Liu, McCullough, & Oey, 2000).
The interaction between the moving load and flexible pavement model has to be studied in
order to quantify the dynamic response of the pavement to moving loads. The problem of mov-
ing loads has been studied by various researchers in different fields (cf. Aubry, Clouteau, &
Bonnet, 1994; De Barros & Luco, 1994; Degrande & Lombaert, 2001; Grundmann, Lieb, &
Trommer, 1999; Lombaert, Degrande, & Clouteau, 2000). De Barros and Luco (1994) have
studied the steady-state response of a layered viscoelastic half-space due to a constant moving
258 P. Khavassefat et al.

load. Grundmann et al. (1999) applied an additional wavelet transform on the response of the
layered half-space in the frequency–wavenumber domain in order to efficiently evaluate the
inverse wavenumber domain transformations. The substructure technique was used by Aubry
et al. (1994) in order to account for the interaction between the road and its underlying soil.
Lombaert et al. (2000) used the same technique to couple a vehicle model and compute the
vibrations induced by traffic in the free field. In the pavement field, the response of the viscoelas-
tic pavement to moving loads has been studied by Hajj, Sebaaly, and Siddharthan (2006) and
Chatti and Yun (1996) utilising the finite layer method combined with surface loads. SAPSI-M
computer program (Chatti, & Yun 1996) calculates the dynamic response of the flexible pave-
ment to moving transient loads with Fourier transform techniques. The model is an axisymmetric
pavement model thus is unable of capturing the real contact stresses on the surface. On the other
hand, 3D-Move (Hajj et al., 2006) deals with three-dimensional surface stresses but ignores the
dynamic effects.
A detailed review of the progress in the field of evaluation of dynamic response of pave-
ments to moving loads was given by Beskou and Theodorakopoulos (2011) where studies were
classified with respect to different road models, material models, method of solution and load
characteristics. It has to be emphasised, however, that to the best of authors’ knowledge, no
attempt has been made to quantitatively evaluate the dynamic response of pavements with
viscoelastic behaviour of asphalt material and with three-dimensional response field. It is the
intention of the present study to close this gap.
In the present paper, a computationally efficient procedure for analysing the response of pave-
ment structures composed of layers with linear material behaviour to moving vehicle loads is
proposed. The procedure developed is capable of taking into account dynamic effects arising at
vehicle–road interaction. The frequency response function (FRF) of the linear viscoelastic pave-
ment structure is obtained using finite element analysis. The quarter car model combined with
the measured road roughness profiles is relied upon to estimate the dynamic loads induced on
the pavement by a moving vehicle. The response of the pavement structure to moving vehicles is
obtained then in a frequency–wavenumber domain through a convolution procedure. In the com-
putational framework proposed presently the FRF for a given pavement should be calculated only
once. As the FRF is obtained, the effects of the road surface condition (i.e. road roughness) and
traffic characteristics (i.e. average speed, distance between vehicles and types of vehicles) can
be evaluated in a computationally effective manner as only a convolution in the wavenumber–
frequency domain is involved, which takes only several seconds to perform. Combined with the
fatigue damage laws used in the pavement design, as presented in, for example, Gullberg, Birgis-
son, and Jelagin (2012), the present computational procedure may be thus used to relate the road
roughness measurements performed presently on many roads annually with the road damage ini-
tiation and propagation. The computational procedure established in the present paper is used
to evaluate the effect of road surface roughness on the stresses induced in pavements. Attention
is focused on stress components governing fracture initiation in pavement structure. It is shown
that the dynamic effects at vehicle–road interaction have a significant influence on the stresses
induced in asphalt layers and thus have to be taken into account to achieve accurate predictions
of the pavement structures service life.

Problem formulation and computational procedure


The major components of the vehicle–road interaction problem are illustrated in Figure 1, along
with the spatial coordinate system used in the present analysis. The components depicted in
Figure 1 interact in the following way. The vehicle moving on an uneven road surface induces
Road Materials and Pavement Design 259

Figure 1. Schematic view of the problem description.

dynamic loads on the pavement. The loads are transferred from the vehicle to the pavement
through tyres; the contact stresses on the pavement surface and the geometry of the loaded area
are thus affected by the tyre type and inflation pressure. The stress–strain state induced in a
pavement by the contact stresses on its surface is controlled by the geometry and mechanical
properties of the pavement layers. The models used in the present computational framework to
represent the vehicle–road interaction components are described in detail below.
The response of the pavement system provided that the pavement structure comprises materi-
als with linear behaviour, for example, linear elastic or linear viscoelastic layers, to mechanical
loads can be calculated from the input through a convolution integral, provided that the impulse
response function is known:
 ∞
f (t) = F(τ ) · h(t − τ ) · dτ , (1)
−∞

where F(τ ) is the input for the linear system, that is, vehicle loads and h(t) is the impulse
response function, that is, pavement response to an impulse load. The convolution integral
in Equation (1) can be transformed to frequency domain in order to reduce the convolution
operation to a simple multiplication:

f˜ (ω) = F̃(ω) · h̃(ω), (2)

where F̃(ω) is the input function in the frequency domain and h̃(ω) is the FRF. The transforma-
tion from time to frequency domain can be obtained by Fourier transform:
 ∞
F̃(ω) = F(t) · e−2π itω · dt. (3)
−∞

The load moving along the x-axis in three-dimensional space may be presented as below with
t indicating the time vector and v being the speed of the vehicle:

F(x, y, z, t) = Fx (x − vt) · Fy (y) · Fz (z) · Ft (t). (4)


260 P. Khavassefat et al.

Applying a quadruple forward Fourier transformation to Equation (4), one may write the
equation of moving load in the frequency–wavenumber domain (Lombaert, Degrande, &
Clouteau, 2004):
   +∞ 
(4)
F̃ (kx , ky , kz , ω) = Fx (x)Fy (y)Fz (z)Ft (t) e(ix(kx +ky +kz )) e(ikx vt) e(−iωt) dx dy dz dt,
−∞
(5)
where kx , ky and kz are wavenumbers (cycle/m) in x, y and z dimensions, representing the rate
of change in distance. Moreover, the superscript (4) expresses the quadruple forward Fourier
transform. Analogous to Equation (2), the response of the linear system to moving load is written
as
f˜ (4) (kx , ky , kz , ω) = F̃ (4) (kx , ky , kz , ω) · h̃(4) (kx , ky , kz , ω). (6)
Substituting Equation (5) into Equation (6), one may re-formulate the response of the system
to a moving load as
 +∞ 
f˜ (4) (kx , ky , kz , ω) = F̃x (kx ) · F̃y (ky ) · F̃z (kz ) · Ft (t) · e−it(ω−kx v) dt · h̃(4) (kx , ky , kz , ω).
−∞
(7)
In Equation (7) the bracket term is the frequency shifted Fourier transform of time-dependent
amplitude of the moving load (Lombaert & Degrande, 2001). Applying shift in the frequency
domain, Equation (7) can be elaborated further as

f˜ (4) (kx , ky , kz , ω) = F̃x (kx ) · F̃y (ky ) · F̃z (kz ) · F̃t (ω − kx v) · h̃(4) (kx , ky , kz , ω). (8)

For a special case of loads applied on the surface of the pavement and moving along the x-axis,
the problem is reduced to only two variables, kx and ω. Equation (8) may be re-written as

f̃˜ (kx , y, z, ω) = [F̃x (kx ) · F̃t (ω − kx v)] · h̃(k


˜ , y, z, ω),
x (9)

where the bracket term is the dynamic axle load induced from the vehicle on the pavement and
˜ , y, z, ω) is the pavement response function, both characterised in the frequency–wavenumber
h̃(kx
domain. The double forward fast Fourier transform for obtaining the transfer function in the
frequency–wavenumber domain from the impulse response function in the spatial–time domain
is shown as below:
 +∞
h̃(x, y, z, ω) = h(x, y, z, t) exp(−iωt)dt, (10)
−∞
 +∞
˜ , y , z , ω) =
h̃(k h̃(x, y0 , z0 , ω) exp(+ikx x)dx. (11)
x 0 0
−∞

Finally, it is possible to obtain the response of the pavement structure to arbitrary loading
history in the spatial–time domain by applying one double inverse Fourier transform:
 +∞
˜f (x, y, z, ω) = 1 f̃˜ (kx , y, z, ω) exp(−ikx x)dkx , (12)
2π −∞
 +∞
1
f (x, y, z, t) = f˜ (x, y, z, ω) exp(+iωt)dω. (13)
2π −∞

The dynamic axle load F̃(k˜ , y, z, ω) in Equation (6) is the load induced from the moving
x
vehicle on the road with a surface roughness. Thus, as long as the response function is known
Road Materials and Pavement Design 261

Figure 2. Quarter car model with two-degree-of-freedom.

˜ , y, z, ω), one may obtain the pavement response to a


along with moving load description, F̃(k x
moving load through Equation (9). In the following, the approach used in the present study to
˜ , y, z, ω), are presented.
˜ , y, z, ω), and the response function, h̃(k
find the load function, F̃(k x x
Vehicle models with discrete masses connected with springs and dashpots are often used in
studying the vehicle–road interaction. As the dynamic stiffness of the pavement structure is very
high as compared with the vehicle suspension, the pavement is usually represented as a rigid
surface with a certain roughness profile. In the present study, a linear two-degree-of-freedom
system, that is, a quarter car model, is relied upon to evaluate the effect of the road surface
roughness on the dynamic vehicle loads. Although quarter car models are missing some impor-
tant aspects of the vehicle behaviour, for example, pitching and bouncing, they have the sufficient
frequency content for predicting the dynamic loads on the pavement (Cole & Cebon, 1989; Hardy
& Cebon, 1994) and therefore has been widely used in different research works (Sun, 2001; Sun
& Kennedy, 2002). The principal sketch of the vehicle model used is shown in Figure 2. The
system is represented by a car body mass and a tyre mass being connected with a spring and
dashpot, representing the suspension system. The tyre mass is connected to the rigid ground with
a spring and dashpot (Figure 2).
The quarter car model used in this paper is based on the work of Sun and Kennedy (2002).
The equation of motion for such a system is shown below:
[M ]{Z̈} + [C]{Ż} + [K]{Z} = {F(t)}, (14)
where {Z} = [zt zs ]T is a vector containing the relative vertical displacement of sprung and
unsprung mass from the surface and the normalised forcing function matrix is {F(T)} = [ξ̈ ξ̈ ]T .
The normalised mass matrix [M ], damping matrix [C] and stiffness matrix [K] are given by Sun
and Kennedy (2002):
     
1 0 αt −αs  h βt −βs  h
[M ] = , [C] = , [K] = ,
1 1 0 αs 0 βs

where  h = ms /mt , αs = cs /ms , αt = ct /mt , βs = ks /ms , βt = kt /mt , and Ż and Z̈ are first and
second derivative of relative displacement in accordance to time, representing the velocity and
acceleration of the system.
262 P. Khavassefat et al.

Taking the Fourier transform from both sides of the equation one can obtain the transfer
function for this system of equations (Sun & Kennedy, 2002):
 2 
ω − iωαt − βt iωαs  h + βs  h  Ht (ω)  ω2 
= 2 , (15)
ω2 ω2 − iωαs − βs Hs (ω) ω

where Ht (ω) and Hs (ω) are the FRFs for the vehicle body and the tyre mass displacements.
By solving the system of equations as presented in Equation (15) and obtaining FRFs for vehicle
body and tyre mass displacements, the FRF for the load induced on the pavement can be obtained
as follows:
hf (ω) = |(kt + ict ω)Ht (ω)|. (16)
The response of the system, F̃(ω), to any arbitrary vertical displacement, ξ(ω), can be obtained
through the multiplication below:

F̃x (ω) = hf (ω)ξ(ω). (17)

The moving load in frequency–wavenumber domain was presented in Equation (5). The
equation for the special case of the moving load along the x-axis can be simplified and expressed
as
 +∞ 
˜ , y, z, ω) =
F̃(k [Fx (x − vt)Ft (t)]Fy (y(y=0) )Fz (z(z=0) )e(ixkx ) e(−iωt) dx dy dz dt, (18)
x
−∞

which can be written as


˜ , y, z, ω) = F̃ (k ) · F̃ (k ) · F̃ (k )F̃ (ω − k v),
F̃(kx x x y y=0 z z=0 t x

where F̃x (kx ) is the load description in wavenumber domain and F̃t (ω − kx v) is the frequency of
moving load.
It remains now to discuss how the transfer function is calculated for the pavement structure.
˜ , y, z, ω), requires the solution of the pave-
The calculation of the pavement transfer function, h̃(k x
ment system response to a given spatial loading distribution at a fixed position. In this paper,
the transfer function is obtained through analysing the pavement structure using the 3D finite
element. This allows obtaining the three-dimensional transfer function with an arbitrary geome-
try, for example, layer thickness, non-symmetric boundary conditions such as pavements loaded
close to the pavement shoulder, contact area shape and 3D distribution of tyre contact stresses.
In the present study, the attention is focused however on the normal load transferred from the
vehicle to the pavement. The presence of the shear tractions on the tyre–pavement interface is
thus neglected and the load from the vehicle is assumed to transfer entirely through uniform
normal pressure distributed over the tyre–pavement contact area. In order to obtain a more real-
istic 3D distribution of the tyre–pavement contact stresses, a truck tyre model coupled with the
truck model has to be incorporated into the analysis method presented here. The transfer func-
tion obtained in this paper has been carried out by solving a layered system which represents the
typical pavement structure as shown in Figure 3.
The finite element model presently used is identical to the model previously developed by the
authors (Khavassefat et al., 2012). The first layer of the model represents the bituminous layer
and is modelled as a linear viscoelastic material, while all the layers beneath are considered lin-
ear elastic. The finite element (FE) mesh used in the present analysis is shown in Figure 4. As
the mesh is symmetric with respect to the x-axis, for clarity only half of the model is shown in
Road Materials and Pavement Design 263

Figure 3. The cross-section and longitudinal view of a typical pavement structure.

Figure 4. View of half of the finite element model for obtaining the transfer function.

Figure 4. The structure is constrained from displacements in all three directions in the bottom
plane. There are rollers on ZY and ZX planes, which constrain the movement in x and y direc-
tions. The size of the model is designed to be 20 times of the width of the loading strip, which
corresponds to the tyre width, in order to minimise the boundary effect. A more detailed expla-
nation of how the distances are chosen is given in a previously published work of the authors
(Khavassefat et al., 2012). The loading strip, distributed along the x-axis, is shown in Figure 4 as a
pink rectangle which has the finest mesh pattern, in order to capture high stress gradients induced
in this area. Each element in this area has a maximum dimension of 1.25 cm, corresponding to
the minimum wavelength of the analysis, and is chosen in accordance to ISO 13473–1(1997)
corresponding to macrostructure wavelength which is essential for vehicle–pavement vibrations.
The mesh grows coarser as it gets further away from the loaded area.
The FE analysis is performed in the frequency domain for this problem. Therefore, the
response of the pavement, f˜p (x, y, z, ω), to the load input, F̃(x, y, z, ω), is in the spatial–frequency
domain. One may use Equation (11) to transform the response and the input to the frequency–
wavenumber domain. In addition, the Hamming window has been used in order to avoid spectral
leakage of the data. Then, the transfer function is obtained by dividing the response by the input
in the frequency–wavenumber domain:
264 P. Khavassefat et al.

˜
˜ , y, z, ω) = f̃ p (kx , y, z, ω) .
h̃(k (19)
x
F̃˜ (kx , y, z, ω)

It must be mentioned that the response can be any stress, strain or displacement component.
As an example, the transfer function for the horizontal stress along the x-axis can be written as

f̃˜ S (kx , y, z, ω)
h̃˜sxx (kx , y, z, ω) = xx .
˜ , y, z, ω)
F̃(kx

Validation
In order to verify the computational procedure described above, the following tests have been
performed. First, the developed procedure is compared with an analytical solution for calculating
the response of a two-dimensional beam to a concentrated load moving with a constant velocity.
In the second part, the element size and boundary condition for the finite element model are
verified against an analytical solution for a homogeneous half-space case. Finally, in the last part
the transfer function is obtained for a pavement structure as shown in Figure 4. The pavement
is loaded with the two different load distributions depicted in Figure 7, both having a uniform
distribution in the frequency domain and magnitude of one. The FE results obtained with load
1 are used to find the pavement response function. Afterwards, using the procedure described in
the previous section, the response of the pavement to load 2 is obtained and compared with the
response from the FE analysis.

Simply supported beam subjected to a moving constant force


The response of a simply supported beam with finite length of L to a moving concentrated force
with magnitude of P and constant velocity of c can be calculated through the equation below
(Frýba, 1999):

∞ 
2Pl3  j πx 1 α
u(x, t) = sin sin j ωt − sin ω(j ) ,
t (20)
π 4 EJ j =1 l j 2 (j 2 − α 2 ) j

where α is the ratio of the speed to the critical speed and ω(j ) is the circular frequency for the
j th mode of vibration. In order to validate the numerical approach, the dynamic deflection from
Equation (20) is transformed to the frequency–wavenumber domain. The load is a concentrated
load with magnitude of 1 N and α is set to 0.5. The cross-section of the beam is a square with
dimension of 10 cm with the length of 85 m. The results for deflection of a mid-span point for
one specific frequency are shown in Figure 5. One may see in Figure 5, the good match with a
maximum error of 2% between the numerical and analytical results for different wavenumbers
and frequencies.

Elastic half-space subjected to dynamic concentrated loading


In this section the radial displacement within a body in an elastic half-space is compared for
an FE model against an analytical solution. The analytical solution is given by Johnson (1987)
where the displacement is calculated when a point load is applied. Within a body at a radial
Road Materials and Pavement Design 265

Figure 5. Dynamic deflection of a point in the mid-span of the simple beam subjected to a moving
concentrated load (E = 5 GPa, l = 85 m, ρ = 2000 kg/m3 , α = 0.5).

distance R from the point which the load is applied, when the distance is large enough compared
with the wavelength, the radial displacement uR is entirely due to the pressure wave and is given
by (Lamb, 1904; Miller & Pursey, 1954):

P0 cos θ (μ2 − 2 sin2 θ )cos(ωt − k2 R)


uR =
, (21)
2πGR (2sin2 θ − μ2 )2 4 sin θ (sin2 θ − μ2 )(sin2 θ − 1)

where the harmonic point load is considered as

P(t) = P0 cos ωt. (22)

In Equation (21): θ = cos−1 (z/R), k1 = ω/c1 , k2 = ω/c2 , μ = c1 /c2 , and c1 and c2 are the
elastic wave velocities.
In Figure 6 the results from the analytical solution against the FE model are shown. The FE
model is an axisymmetric model with a maximum element size of 1.25 cm. The rollers are placed
at 7.5 m distance from the origin on the vertical edge and the bottom of the model is fixed at the
same distance from the origin. One may see in Figure 6 that after approximately 0.2 s, which
is the time for the wave to propagate to the point of interest, the results of the FE model and
analytical solution have a perfect match.

Multi-layered solid subjected to different loading conditions


The response of the pavement structure in spatial and wavenumber domain for two different
loading types of 1 and 2 shown in Figure 7 is obtained using FE analysis. The response is obtained
in the spatial–frequency domain and then by means of Fourier transformation is transformed
into the wavenumber domain. Afterwards, the transfer function for the pavement structure is
obtained from response of the pavement to load 1 by applying Equation (19). The selected range
266 P. Khavassefat et al.

Figure 6. Dynamic radial displacement of a point within a half-space subjected to a harmonic load (E = 1
MPa, ν = 0.25, ρ = 1e + 5 kg/m3, P0 = 100, ω = 2 rad/s, z = 0.6 m, R = 1.1 m).

Figure 7. Load distributions in spatial and wavenumber domain.

of wavenumber, 0–15 m−1 , corresponds to the range of road unevenness which is important for
vehicle dynamics and is characterised by wavelengths λ (λ = 2π/ky ) between 0.5 and 50 m
(Descorret & Boulet, 1996). By applying Equation (6) the response of the pavement structure to
load 2 is obtained and then is compared with the results from FE when the structure is loaded
directly by load 2. One may see in Figure 8 that the response obtained by applying Equation (6)
matches the solution obtained directly from FE.

Numerical study of the vehicle–road interaction effect on pavement degradation


In order to evaluate the practical importance of the dynamic vehicle–road interaction effects
for the degradation of the pavement structure, the following computational study has been
Road Materials and Pavement Design 267

Figure 8. Horizontal stress at frequency of 8e − 3 Hz.

performed. Two different dynamic load levels corresponding to two different roughness profiles,
induced on the pavement by a quarter car model which represents a heavy vehicle, are considered
for the purpose of investigating the effect of surface roughness on the dynamic response of the
pavement structure. The two load levels used in the present study correspond to two different
roughness profiles, that is, to roads with comparatively smooth and rough surfaces, as discussed
in detail below and are depicted in Figure 9. The roughness profiles shown in Figure 9 have
been measured on two Swedish highways. The first set of data is the longitudinal roughness of
a highway located north of Stockholm. The highway is fairly new and also fairly even, with
the International Roughness Index (IRI) value of 0.99 m/km. The second profile belongs to a
highway 600 km north of Stockholm and has a higher level of unevenness (IRI = 2.30 m/km)
compared with the first road profile. The IRI value is a statistical measure proposed by the World
Bank to evaluate pavement roughness (Gillespie, Sayers, & Segel, 1980). IRI is a measure of
the relative displacement between the axle and sprung mass of a quarter car which traverses the
pavement at a constant speed of 80 km/h (Gillespie et al., 1980; Sun, 2003). In Sweden the stan-
dard increment for longitudinal direction is 0.1 m and thereafter the data are averaged for every

Figure 9. Longitudinal profile roughness for two different roads.


268 P. Khavassefat et al.

Figure 10. DSD for road profiles 1 and 2.

Table 1. Parameter values for quarter car model (Hardy & Cebon, 1994).

ms (kg) mt (kg) cs (N s/m) ct (N s/m) ks (N/m) kt (N/m)

Sprung mass Unsprung mass Sprung damping Unsprung damping Sprung stiffness Unsprung stiffness
4450 550 15 × 103 2 × 103 1 × 106 1.75 × 106

20 m in order to obtain the IRI value. The data for these measurements are from the test vehicle
laser that follows the right rut of the road.
Power spectral density (PSD) is one of the standard measures that has been used by the vehi-
cle industry in order to characterise the pavement (Gillespie et al., 1980; Sun, 2003). PSD of
displacement is known as displacement spectral density (DSD). In Figure 10, the DSD for both
profiles is shown. The DSD is calculated by performing Fourier transform of the longitudinal
profile measurements and is plotted versus wavenumber. The blue and red lines in Figure 10 are
the DSP for a good and average road according to the ISO standard (ISO, 1996). The ISO lines
are shown in Figure 10 only for reference and are not involved in the calculations.
The parameters used for the quarter car model are summarised in Table 1. The parameters
represent a quarter car model of a heavy vehicle (Hardy & Cebon, 1994). The load transfer
function for this vehicle model is obtained according to Equation (16) and is shown in Figure 11
which agrees with the results shown by Sun and Kennedy (2002). One may notice the two main
peaks in the load transfer function at 2 and 12 Hz. The first peak is the sprung mass resonant
frequency and the latter is the wheel hop frequency.
Figure 12 shows the dynamic load induced by passage of the quarter car over road profiles 1
and 2 with a speed of 30 m/s. The main visible peak at approximately 2 Hz corresponds to the
main body bounce mode of vibration, while the resonant frequency for wheel hop is not visible in
this figure as the level of the road excitation is very low at that range frequency (Cebon, 2000).
One may see in Figure 12 that the maximum force amplitude from profile 2 is approximately
40% higher than the one from profile 1.
In order to obtain the response of the pavement to these two dynamic load descriptions, one
should calculate the transfer function for the pavement. The transfer function for an arbitrary
pavement structure can be calculated as described in the second section. The pavement structure
Road Materials and Pavement Design 269

Figure 11. Transfer function of the quarter car used in the present study.

Figure 12. Load applied on the pavement by a quarter car.

chosen for this study is a three-dimensional multilayer structure, as shown in Figure 3, corre-
sponding to a typical Swedish pavement and is analogous to the model used by Khavassefat
et al. (2012). The first layer is modelled as an isotropic linear viscoelastic material presented
with the generalised Kelvin model:
 

N
−t/τiG
G(t) = G0 1 − ḡ Pi (1 −e ) , (23)
i=1

where ḡiP and τiG are material constants, N is the number of terms in the Prony series expansion
and G0 is the instantaneous (elastic) shear modulus. The underlying layers are considered to be
homogeneous and linear elastic with different stiffness. The material parameters for each layer
are given in Tables 2 and 3.
270 P. Khavassefat et al.

Table 2. Pavement structure geometry and material properties (Khavassefat et al., 2012).

Layer type Thickness (mm) Material properties, E (MPa) υ ρ (kg/m3 )

Bituminous layer 160 See Table 3 2200


Base layer 150 450 0.35 2400
Sub-base layer 350 240 0.35 2400
Subgrade Infinite 100 0.35 2400

Table 3. Prony series coefficients for defining viscoelastic


behaviour of asphalt mixture (Khavassefat et al., 2012).

i 1 2 3 4

ḡiP 8.13 e–1 1.70 e − 1 1.47 e − 2 8.44 e–2


τiG [s] 1.75 e–1 5.03 1.14 e + 2 2.72 e + 5

Note: G0 = 4740 MPa.

The 3D steady-state dynamic analysis is performed using Abaqus/CAE 6.11 (Simulia, 2011)
in order to obtain the dynamic response of the pavement in a selected range of frequencies. The
˜ , y, z, ω), is obtained according to Equation (19) where the load applied on
transfer function, h̃(k x
the pavement is defined as below:

F(x, y, z, ω) = F(x) · F(y) · F(z) · F(ω),



1, x < 30 cm,
F(x) =
0, x > 30 cm,
700 700
F(y) = − − 1, (24)
1 + exp(−30 · (y − 4.1)) 1 + exp(−30 · (y − 3.97))2

1, z = zsurface ,
F(z) =
0, z = zsurface ,
F(ω) = 1.

Equation (24) is the analytical form of an approximate step function with a width of 30 cm
on the surface of the pavement, which has a unit magnitude over the frequency range. For this
specific analysis, the range of frequencies is selected between 0 and 15 Hz.
It remains now to discuss the numerical results concerning the influence of the dynamic loads
reported in Figure 12 on the stress state induced by traffic in flexible pavement. Below, the
attention will be focused on the dynamic component of horizontal stresses at the top and bottom
of the pavement’s asphalt layer. Maximum tensile stresses induced at the bottom and at the top
of asphalt layer are of significant practical importance as they are commonly associated with the
initiation of bottom-up and top-down cracking (Khavassefat et al., 2012). Figure 13 depicts the
horizontal stress distribution on the surface of the pavement along a line under the moving path
for different frequencies. The horizontal stress is caused by a moving quarter car model with the
speed of 30 m/s as shown in Figure 13. The tensile stress on the surface of the pavement can be
responsible for top-down crack initiation, which has been reported as one of the most common
distress modes in pavements (COST 333, 1999). These stresses are mainly a result of viscoelastic
stress redistribution. The elastic layers under the asphalt layer tend to recover instantaneously,
Road Materials and Pavement Design 271

Figure 13. Horizontal stress on the surface of the pavement.

while the viscoelastic asphalt layer recovers in time. Thus, the unbound layers push the asphalt
layer upward and result in tensile stresses on the surface (Khavassefat et al., 2012; Kim et al.,
2009; Roque et al., 2010). One may notice that the range of wavenumbers in pavement response
increases by increasing the frequency. Moreover, for the majority of the wavenumber range for
profile 2, which was the rougher pavement, has higher maximum values as compared with the
ones for profile 1.
The horizontal stress at the bottom of the asphalt layer is depicted in Figure 14. The tensile
stress at the bottom of the asphalt layer is one of the criteria for fatigue crack initiation in pave-
ments, also known as bottom-up cracking. Analogous to Figure 13, the response from profile 2
has a higher amplitude in most of the wavenumbers in comparison with profile 1. One also may
notice that for both cases of horizontal stress at top and bottom of the asphalt layer, the maximum
peak for all of the curves is at 0.4 m−1 , which is the wavenumber for maximum load from the
vehicle.
272 P. Khavassefat et al.

Figure 14. Horizontal stress at the bottom of asphalt layer.

In order to quantify the magnitude of the response in the analysis, one may use the root mean
square (RMS) value. The RMS value is an average measure of the magnitude of a varying vari-
able (i.e. for the purposes of the present study – of the dynamic stress components). The value in
the transformed domain may be obtained as below:

n
1  1  n
RMS =  xi (t) =  2
2 |Xi (f )|2 , (25)
n i=1 n i=1

in which x(t) is the variable in the normal domain (time or space), X (f ) is the transformed
variable in frequency or the wavenumber domain and n is the number of wavenumbers used in
computing the RMS value.
The RMS value for the horizontal stress at the top and bottom of the asphalt layer for the two
different road profiles at four arbitrary frequencies is listed in Table 4. It must be noted that for
calculating the RMS value at each frequency, a range of wavenumber is chosen and is reported
Road Materials and Pavement Design 273

Table 4. RMS values for horizontal stress at different frequencies.


Maximum
wavenumber for
Location % Difference of RMS2 calculating RMS
Frequency(Hz) top(T)/bottom(B) RMS1 RMS2 to RMS1 (m−1 )

4e − 3 T 79.9 122.5 53.4 2.0


B 77.9 119.5 53.5
1.0 T 124.3 188.7 51.8 4.0
B 204.0 311.5 52.7
4.9 T 142.9 217.6 52.3 5.0
B 172.8 262.4 51.8
15.1 T 62.1 80.3 29.3 10.0
B 43.4 47.9 10.3

in Table 4. This is to avoid adding extra zeroes to the equation. As it was mentioned before
and according to Figures 13 and 14, the wavenumber range expands as the frequency increases.
In general, according to the data listed in Table 4, profile 2 causes a 50–60% higher horizontal
stress on the surface of the pavement as well as at the bottom of asphalt layer in comparison with
profile 1.
It may thus be concluded that taking into account the dynamic vehicle–pavement interaction
results in significantly higher traffic loads and thus higher stresses induced in the pavement as
compared with the quasi-static analysis. Furthermore, taking viscoelastic pavement properties
and dynamic effects in the pavement structure into account results in approximately 10% addi-
tional increase in tensile stresses induced in pavements. Therefore, when the traffic load increases
for 40%, it results in 50% higher horizontal stresses as compared with the ones obtained at lin-
ear quasi-static analysis. In most of state-of-practice design methods pavement deterioration is
predicted by empirical relations correlating maximum tensile stress or strain in the asphalt layer
to the number of cycles to pavement fatigue fracture failure (cf. Swedish Transport Adminis-
tration [STA], 2011). These empirical relations have a form of power-law where the number
of cycles to failure is inversely proportional to a certain degree of the maximum tensile stress
or strain induced in the pavement. The exponential parameter used in different design methods
varies between approximately 2 (Gullberg et al., 2012) and 4 (STA, 2011). The additional 10%
increase in the horizontal tensile stresses reported in Table 4 will thus have a profound effect on
the predicted pavement fatigue life, as an example according to the design method presented by
Gullberg et al. (2012) the 10% increase in tensile stresses induced in the pavement will result in
approximately 20% reduction of the predicted service life of the pavement structure. Moreover,
the rutting damage is correlated in state-of-practice design methods to the maximum vertical
compressive strain on top of the subgrade. Similar influence as fatigue failure (i.e. shortening of
pavement service life) may be observed by including dynamic components of the vertical strain
on top of the subgrade into rutting failure calculations.

Conclusions
A numerical framework for quantifying the dynamic response of the viscoelastic flexible pave-
ment to moving loads has been developed. The dynamic nature of the vehicle–pavement
interaction problem is taken into account by bringing the road profile roughness into calcula-
tions. In order to obtain the response of the pavement to the moving load, first the transfer
function for the pavement structure using three-dimensional finite element method (FEM) is
274 P. Khavassefat et al.

calculated. Afterwards, the response is found in the frequency–wavenumber domain by multi-


plying dynamic load and the transfer functions. The computational framework is numerically
efficient as it utilises the double fast Fourier transform and the transfer function with the FEM
needed to be calculated only once. However, it has to be mentioned that the procedure is limited
to pavement structures with linear behaviour and thus is unable to account for material nonlin-
earities as well as for nonlinear boundary conditions, for example, incomplete slip between the
layers, in the pavement.
A typical flexible pavement dynamic response to a moving quarter car model is investigated
numerically with the procedure developed. In the numerical study, two different measured road
profiles are considered. The dynamic load caused by the same vehicle model moving on the
rougher profile was approximately 40% higher in comparison with the smoother profile. Employ-
ing the developed procedure for these two different cases, it was found that as the frequency
increases, a higher range of wavenumbers gets involved in the pavement response. For the spe-
cific case of horizontal stress at the surface and the bottom of the asphalt layer, the response of
the pavement to the rougher profile is approximately 50% higher in the majority of the frequency
ranges. The results of this study clearly show the importance of the dynamic vehicle–road inter-
action in stress analysis of flexible pavements; as these effects can have a considerable influence
on the expected pavement service life.
It has to be pointed out that as it has been shown by several studies (cf. Al-Qadi et al., 2005,
Drakos et al., 2001) taking into account the non-uniform tyre contact pressure results in more
realistic pavement response. In the current study, the vehicle load was transferred to the pavement
with uniform normal pressure. However, the developed procedure is based on a 3D solution thus
is capable of dealing with non-uniform normal and shear tractions. A detailed study on dynamic
effects when more realistic tyre contact pressure is taken into account is intended for future
studies.

Disclosure statement
No potential conflict of interest was reported by the authors.

ORCID
P. Khavassefat http://orcid.org/0000-0003-2434-6957

References
Al-Qadi, I. L., Yoo, P. J., Elseifi, M. A., & Janajreh, I. (2005). Effects of tire configurations on pave-
ment damage (With Discussion). Journal of the Association of Asphalt Paving Technologists, 74,
921–962. Presented at the 2005 Journal of the Association of Asphalt Paving Technologists: From
the Proceedings of the Technical Sessions.
Al-Qadi, I. L., Wang, H., Yoo, P. J., & Dessouky, S. H. (2008). Dynamic analysis and in situ validation
of perpetual pavement response to vehicular loading. Transportation Research Record: Journal of the
Transportation Research Board, 2087(1), 29–39.
Arraigada, M., Partl, M. N., Angelone, & S. M., Martinez, F. (2009). Evaluation of accelerometers to
determine pavement deflections under traffic loads. Materials and Structures, 42, 779–790.
Aubry, D., Clouteau, D. & Bonnet, G. (1994). Modelling of wave propagation due to fixed or mobile
dynamic sources. In N. Chouw & G. Schmid (Eds.), ‘Workshop Wave ’94, Wave propagation and
reduction of vibrations’, Ruhr University, Bochum, Germany, pp. 109–121.
Beskou, N. D., & Theodorakopoulos, D. D. (2011). Dynamic effects of moving loads on road pavements:
A review. Soil Dynamics and Earthquake Engineering, 31, 547–567.
Bilgiri, K. P., & Way, G. B. (2014). Noise-damping characteristics of different pavement surface wearing
courses. Road Material and Pavement Design. doi:10.1080/14680629.2014.902768
Road Materials and Pavement Design 275

Cebon, David. (2000). Handbook of vehicle-road interaction. London: Taylor & Francis.
Chatti, K., & Yun, K. K. (1996). SAPSI-M: Computer program for analyzing asphalt concrete pave-
ments under moving arbitrary loads. Transportation Research Record: Journal of the Transportation
Research Board, 1539(1), 88–95.
Cole, D. J., & Cebon, D. (1989). Simulation and measurement of dynamic tyre forces. Proceedings of 2nd
International Symposium of Heavy Vehicle Weights and Dimensions, Kelowna, Canada.
COST 333. (1999). Development of new bituminous pavement design method. Final Report., Directorate
General Transport, Office for Official Publications of the European Communities. Brussels.
De Barros, F. C. P., & Luco, J. E. (1994). Response of a Layered Viscoelastic Half-space to a Moving Point
Load. Wave Motion, 19(2), 189–210. doi:10.1016/0165-2125(94)90066-3
Degrande, G., & Lombaert, G. (2001). An efficient formulation of Krylov’s prediction model for train
induced vibrations based on the dynamic reciprocity theorem. The Journal of the Acoustical Society of
America, 110, 1379–1390.
Descorret, G., & Boulet, M. (1996). Road surface design and tyre/road surface interactions. Presented at
the TYRETECH 95. 21ST Century Technology, Turin, Italy.
Drakos, C., Roque, R., & Birgisson, B. (2001). Effects of measured tire contact stresses on near-surface
rutting. Transportation Research Record: Journal of the Transportation Research Board, 1764, 59–69.
European Commission. (2008). Effects of adapting the rules on weights and dimensions of heavy com-
mercial vehicles as established within Directive 96/53/EC. Directorate-General Energy and Transport,
European Commission, Brussels.
Frýba, L. (1999). Vibration of solids and structures under moving loads. London: Thomas Telford.
Gillespie, T. D., Sayers, M. W., & Segel, L. (1980). Calibration of response-type road roughness measuring
systems. NCHRP Rep. No. 228. Ann Arbor, MI: Transportation Research Board.
Grundmann, H., Lieb, M., & Trommer, E. (1999). The Response of a layered half-space to traffic loads
moving along its surface. Archive of Applied Mechanics, 69(1), 55–67.
Gullberg, D., Birgisson, B., & Jelagin, D. (2012). Evaluation of predictive material models used in the new
Swedish mechanistic-empirical design module. Road Materials and Pavement Design, 13, 300–311.
Hajj, E. Y., Sebaaly, P. E., & Siddharthan, R. V. 2006. Response of Asphalt Pavement Mixture under
a Slow Moving Truck. In Asphalt Concrete: Simulation, Modeling, and Experimental Characteriza-
tion, Geotechnical Special Publication, No. 146. American Society of Civil Engineers, Baton Rouge,
Louisiana, pp. 134–146.
Hardy, M. S. A., & Cebon, D. (1994). Importance of speed and frequency in flexible pavement response.
Journal of Engineering Mechanics, 120, 463–482.
ISO. (1996). Mechanical vibration. Road surface profiles. Reporting of measured data. BSI standards.
Geneva: International Organization for Standardization.
ISO 13473-1. (1997). Characterization of pavement texture by use of surface profiles – Part 1: Determina-
tion of mean profile depth. Geneva: International Organization for Standardization.
Johnson, K. L. (1987). Contact mechanics. Cambridge: Cambridge University Press.
Khavassefat, P., Jelagin, D., & Birgisson, B. (2012). A computational framework for viscoelastic analysis
of flexible pavements under moving loads. Materials and Structures, 45(11), 1655–1671.
Kim, J., Roque, R., & Byron, T. (2009). Viscoelastic analysis of flexible pavements and its effects on
top-down cracking. Journal of Materials in Civil Engineering, 21, 324–332.
Kropáč, O., & Múčka, P. (2008). Deterioration model of longitudinal road unevenness based on its
power spectral density indices. Road Materials and Pavement Design, 9, 389–420. doi:10.1080/
14680629.2008.9690125
Lamb, H. (1904).On the propagation of tremors over the surface of an elastic solid. Philosophical Trans-
actions of the Royal Society of London (Series A, containing papers of a mathematical or physical
character), 203, 1–42.
Liu, C. (2001). Pavement response to moving loads. Road Materials and Pavement Design, 2, 263–282.
Liu, C., McCullough, B.F., & Oey, H.S. (2000). Response of rigid pavements due to vehicle-road
interaction. Journal of Transportation Engineering, 126, 237–242.
Lombaert, G., & Degrande, G. (2001). Study of determining factors for traffic induced vibrations in build-
ings (Report No. BWM-1999–04). Department of Civil Engineering, Katholieke Universiteit Leuven,
July 1999. DWTC Research Programme Sustainable Mobility, Research Project MD/01/040.
Lombaert, G., Degrande, G. & Clouteau, D. (2000). Numerical modelling of free field traffic induced
vibrations. Soil Dynamics and Earthquake Engineering, 19(7), 473–88.
Lombaert, G., Degrande, G., & Clouteau, D. (2004). The non-stationary freefield response for a moving
load with a random amplitude. Journal of Sound and Vibration, 278, 611–635.
276 P. Khavassefat et al.

Miller, G. F., & Pursey, H. (1954). The field and radiation impedance of mechanical radiators on the free
surface of a semi-infinite isotropic solid. Proceedings of the Royal Society A: Mathematical, Physical
and Engineering Sciences, 223, 521–541.
Pouget, S., Sauzéat, C., Benedetto, H. D., & Olard, F. (2010). Numerical simulation of the five-point
bending test designed to study bituminous wearing courses on orthotropic steel bridge. Materials and
Structures, 43, 319–330.
Roque, R., Zou, J., Kim, Y. R., Baek, C., Thirunavukkarasu, S., Underwood, B. S., &
Guddati, M. N. (2010). Top-down cracking of hot-mix Asphalt layers: Models for initiation
and propagation. Transportation Research Board of the National Academies. Retrieved from
http://onlinepubs.trb.org/onlinepubs/nchrp/nchrp_w162.pdf
Saevarsdottir, T., & Erlingsson, S. (2014), Modelling of responses and rutting profile of a flexi-
ble pavement structure in a heavy vehicle simulator test. Road Material and Pavement Design.
doi:10.1080/14680629.2014.939698
Simulia, D. C. S. (2011). ABAQUS 6.11 CAE. Abaqus 6.11 Documentation.
Sun, L. (2001). Computer simulation and field measurement of dynamic pavement loading. Mathematics
and Computers in Simulation, 56, 297–313.
Sun, L. (2003). Simulation of pavement roughness and IRI based on power spectral density. Mathematics
and Computers in Simulation, 61, 77–88.
Sun, L., & Kennedy, T.W. (2002). Spectral analysis and parametric study of stochastic pavement loads.
Journal of Engineering Mechanics, 128, 318–327.
Swedish Transport Administration. (2011). Bitumen bound layers. Publication number: 2011:072.
Borlange.
Wang, H., & Al-Qadi, I. (2009). Combined effect of moving wheel loading and three-dimensional con-
tact stresses on perpetual pavement responses. Transportation Research Record: Journal of the
Transportation Research Board, 2095, 53–61.
Reproduced with permission of the copyright owner. Further reproduction prohibited without
permission.

Вам также может понравиться