Soil Dynamics and Earthquake Engineering 82 (2016) 6372

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

journal homepage: www.elsevier.com/locate/soildyn

Dynamic elastic analysis of 3-D exible pavements under moving

vehicles: A unied FEM treatment
Niki D. Beskou a, Stephanos V. Tsinopoulos b,n, Dimitrios D. Theodorakopoulos a
a
Department of Civil Engineering, University of Patras, 26500 Patras, Greece
b
Department of Mechanical Engineering, Technological Educational Institute of Western Greece, 26334 Patras, Greece

art ic l e i nf o a b s t r a c t

Article history: The dynamic response of exible road pavements to moving vehicles is numerically obtained by the time
Received 23 September 2015 domain nite element method under three-dimensional conditions with the aid of the commercial
Received in revised form program ANSYS. Both the uniform elastic half-space and the layered elastic half-space with or without
24 November 2015
damping are considered. The moving with constant speed loads (wheels) of the vehicle are simulated by
Accepted 26 November 2015
assigning time dependent load values at all the surface nodes along the vehicle path, which are activated
Available online 17 December 2015
at the time it takes for every load to travel the distance from the origin to every node's location. Both
Keywords: concentrated (one or a series of them) and distributed moving loads are considered. Simple supports on
Flexible pavements rollers and viscous absorbing boundaries of the pavement domain are both considered. Comparisons
Half-space
with analytical exact solutions for the half-space model under moving loads and analytical approximate
Layered system
solutions for the layered half-space model under static loads are made for validation and comparison
Elastodynamic analysis
Finite elements reasons, respectively. The cases of static versus dynamic loads, of a single load versus a series of loads and
Moving loads low versus high vehicle's velocity are also compared and discussed from the viewpoint of how they affect
the response of the pavement.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction Even though some works dealing with nonlinear pavement

During the last 35 years or so, a lot of research on the analysis
and design of exible and rigid pavements under stationary or
moving vehicles and realistic geometries and material behavior
has been successfully conducted, as it is evident in the books of
Huang [1] and Cebon [2]. This was made possible with the con-
struction of computer programs for the already existing multi-
layered elastic theories of Burmister [3,4] and Odemark [5] under
distributed static loading and mainly with the development of
numerical methods capable of considering elastic and inelastic
multi-layered systems under complex dynamic loads (stationary
or moving) [1,2]. A comprehensive review on the subject for the
case of dynamic loading has been recently reported by Beskou and
Theodorakopoulos [6]. Thus, the vehicle models can be either
concentrated or distributed loads moving with constant speed, or
more complicated ones involving springmassdamper systems
with variable speed. The pavement models can be simple beams or
plates on elastic or inelastic layered half-spaces under conditions
of plane strain or full three-dimensionality [6].
n
Corresponding author. Tel.: 30 2610 369073.
E-mail address: stsinop@teipat.gr (S.V. Tsinopoulos).
behavior have appeared in the literature, most of the works on
pavements under static or dynamic loads are associated with lin-
ear elastic material behavior including viscoelasticity and por-
oelasticity [1,6]. Due to the complexity of vehicle and pavement
modeling, even when elastic material behavior is assumed, use of
numerical methods of analysis is imperative.
In this work, the nite element method (FEM) is employed for
the numerical determination of the response of exible pavements
composed of three layers on deformable soil to constant speed
moving vehicles consisting of multiple axles under three-
dimensional conditions. The material behavior of the layers and
the supporting soil is assumed to be linear elastic with or without
internal damping. The FEM is employed and used in the time
domain with the aid of the general purpose computer program
ANSYS [7].
Flexible pavements, modeled as elastic or viscoelastic layered half-
spaces under static or dynamic stationary or moving loads, have been
analyzed by various numerical methods. One can mention here the
works of Wang et al. [8] under static loading and use of layered system
theories in matrix form, Cho et al. [9], Jooste [10] and Mulungye et al.
[11] under static load and use of the FEM; the works of Mamlouk and
Davies [12], Tu [13] and Lu et al. [14] under stationary dynamic load
and use of layered system theories in matrix form; the works of
Nazarian and Boddapati [15] and Shoukry and William [16] under

http://dx.doi.org/10.1016/j.soildyn.2015.11.013
0267-7261/& 2015 Elsevier Ltd. All rights reserved.
64 N.D. Beskou et al. / Soil Dynamics and Earthquake Engineering 82 (2016) 6372

stationary dynamic load and use of the FEM; the works of Sousa et al.
[17], Zar et al. [18], Siddharthan et al. [19], Grundmann et al. [20],
Chatti et al. [21], Lefeuve-Mesgouez et al. [22] and Lee et al. [23] under
moving loads and use of the layered system theory in matrix form; the
works of Yang and Hung [24], Elsei et al [25], Yin et al. [26], Yoo and
Al-Qadi [27], Al-Qadi et al. [28], Ju [29], Kang et al. [30], Liao and
Sargand [31] and Khavassefat et al. [32] under quasi-statically or
dynamically moving loads and use of the FEM; the works of Hirose
[33] and de Barros and Luco [34] under moving loads and use of the
boundary element method and nally the work of Cebon [2] and Chen
[35] under moving loads and use of the impulse response method.
The dynamic formulation of the above methods was done either in
the frequency or the time domain. For many more applications of
numerical methods (mainly FEM) to problems of dynamic analysis of
pavements (especially rigid ones), the interested reader can consult
the review paper [6]. For an excellent review on exible pavement
analysis and design with an emphasis on practical aspects, one is
advised to look at Monismith [36].
Looking at the aforementioned papers for exible pavements
with linear elastic or viscoelastic material behavior and the cor-
responding ones in [6], one can observe that i) not a single work
considers all the parameters affecting the accuracy and realism of Fig. 1. General geometry of half of pavement structure domain (symmetry with
the response results and ii) in many cases there is not a general respect to the XZ plane): (OR) 30.00 m, (OA) (BR) 9.00 m, (AQ) (QB) 6.00 m,
(OC) 15.00 m, (CD) 0.15 m, (DF) 0.30 m, (FG) 29.00 m.
agreement among the authors about the kind and magnitude of
the effect a parameter has on the response (e.g., effect of boundary
direction. It consists of three layers fully bonded to each other and
conditions or the nature of loading). In this work, an effort has
to the supporting soil with the top one being the asphaltic layer of
been made not only to consider most of the important parameters
thickness 0.15 m, the intermediate one, the granular base layer, of
affecting the response, but also to reach convincing conclusions
thickness 0.30 m and the bottom one, the subgrade layer, of
related to the importance of these parameters on the response of
thickness 29.00 m.
the pavement in a quantitative manner.
For this nite domain, three discretizations, A, B and C in the
More specically, in the framework of a linear elastic analysis
order of increasing mesh renement, as shown in Fig. 2, were
with the FEM, this work considers i) geometrical parameters (half-
considered and tested with respect to accuracy and efciency. It
space versus layered system, size of pavement domain, absorbing
was decided to adopt discretization B as a compromise between
boundaries versus boundaries on rollers and supporting soil ver-
accuracy and efciency. This discretization consists of 45,920 8-
sus bedrock); ii) loading parameters (concentrated versus dis-
noded 3-D solid nite elements (bricks) symbolized as SOLID185
tributed load, stationary versus moving load and single load versus
in the commercial computer program ANSYS [7] and having in
a series of loads); iii) material parameters (linear elastic material
total 24 degrees of freedom, as shown in Fig. 3.
with or without damping) and iv) discretization parameters (right
Because the three vertical faces (the fourth one is the zx plane
combination of element size and time step for acceptable error
of symmetry) and the bottom horizontal face of the pavement
with respect to analytic and/or numerical solutions). With all the
model are articial boundaries, waves generated by the motion of
above features, the authors hope this work to become a useful
the vehicle along the surface of the pavement propagating inside
source of information on the subject of the elastodynamic analysis
of exible pavements under moving vehicles with the aid of the the pavement domain, are reected at those articial boundaries
FEM and the basis for future developments in the area of the more and pollute the wave propagation inside that domain. For this
realistic inelastic material behavior. reason, one has to introduce absorbing or non-reecting devices at
those boundaries. The standard viscous boundaries of Lysmer and
Kuhlemeyer [37], as described for the 3-D case in Hatzigeorgiou
2. Finite element modeling of pavement structure and Beskos [38], are used in this work. These viscous boundaries,
even though are not perfectly absorbing, are widely used because
The three-dimensional nite element modeling of a typical ex- of their simplicity. The basic idea of this technique is the appli-
ible pavement structure consists of selecting the appropriate dimen- cation of traction boundary conditions at free articial boundaries
sions of a nite domain and the degree of its discretization in order to which forces any reected stresses to become zero. Thus one has
fairly accurately simulate the dynamic behavior of a semi-innite t n V P u_ n 0 ; t s V s u_ s 0 1
space at a reasonable computational cost. The nally selected nite
where tn and ts are the normal and shear tractions at the boundary
domain modeling of the pavement structure evolved through a trial
respectively, is the mass density, u_ n and u_ s are the normal and
and error procedure with the goal of combining realism, accuracy and
tangential velocities respectively, and are the dimensionless
computational efciency. Various combinations of length, width and
parameters depending on Poisson's ratio v and usually taken as
depth of the nite domain were tested with or without symmetry
1 and Vp and Vs are the dilatational and shear wave velocities of
considerations. It was found that the domain should resemble neither
propagation, respectively, dened as
a beam nor a plate but rather a cube-like solid with dimensions more
s s
than twice the vehicle path. E1  v E
The result was that the optimum nite domain simulating the Vp ; VS 2
1 v1  2v 21 v
pavement structure is the one shown in Fig. 1, where symmetry
considerations have been employed. This model has dimensions where E is the elasticity modulus.
29.45 m along the vertical z direction, 15.00 m along the lateral The articial vertical and bottom boundaries in this work are dis-
(transverse) y direction and 30.00 m along the longitudinal x cretized by rectangular vertical and horizontal elements, respectively,
 N.D. Beskou et al. / Soil Dynamics and Earthquake Engineering 82 (2016) 6372 65

Fig. 3. The 8-node 3-D nite element (SOLID185, with 24 d.o.f.).

Number of Nodes: 56810

Number of nodes: 155961

Fig. 4. Vertical articial boundary (external face of brick nite elements) and vis-
cous dashpots.

values
cs a  bV s ; cp a  bV p 3
as shown in Fig. 4. At the bottom face one has also to introduce elastic
springs at every node to simulate the supporting soil medium. The two
horizontal spring constants KH and the vertical spring constant KV at
every bottom node, as shown in Fig. 5, are provided by the Mulliken
and Karabalis [39] expressions
4:6ED 2:35ED
KH ; KV   4
1 v2  v 1  v2
where D is the half-length of the side of an equivalent square to the
rectangular a  b of Fig. 4, which is given by
p
D a  b=2 5
Number of nodes: 284427
The aforementioned dashpots and springs are appropriately
Fig. 2. Three discretizations of domain of Fig. 1: (a) A-coarse; (b) B-intermediate; introduced into the main ANSYS [7] program.
and (c) C-dense. In addition to the above mentioned absorbing boundaries, two
much more simple types of boundaries are used in the rst comparison
studies of this work in order to assess their effectiveness, since they
which are actually the faces of the 3-D solid elements (bricks) used to have been successfully used by some others [26,31,32] and are attrac-
discretize the domain. The dimensions of those elements/faces are tive because of their low computational cost. These are boundaries with
a  b as shown in Fig. 4. To every nodal point one assigns two tan- rollers at the three vertical faces and the bottom of the domain and
gential and one perpendicular to the element/face dashpots with boundaries free at the three vertical faces and on rollers only at the
66 N.D. Beskou et al. / Soil Dynamics and Earthquake Engineering 82 (2016) 6372

Fig. 5. Bottom articial boundary with nodal soil springs.

bottom of the domain, as shown in Fig. 6. Of course, waves reaching

these boundaries are reected and one expects the distance of the
points of interest from those boundaries to be large enough so that
those reected waves do not have the time to appreciably affect the
maximum values of the response. This distance can be selected when
deciding about the dimensions of the domain in order for the response
error, as compared to the exact solution, to be of acceptable level.

3. Modeling of moving vehicle loads

Consider a typical heavy vehicle (DAF truck model FAD CF75) as

shown in Fig. 7 with four axles of concentrated loads PA PB
PC 16 kN and PD 32 kN and axle distances LAB 1.40 m,
LBC 2.95 m and LCD 2.05 m. Thus, the total axle length L 6.40 m
and the total weight of the vehicle is 80 kN.
Consider next the top of the discretized plane y0 of the
pavement system of Fig. 1 with four successive surface nodes along
the direction of the vehicle's motion as shown in Fig. 8. It is
assumed that the distance a between two successive nodes is less
than or equal to 1.0 m so that a oLAB. It is assumed that at time
t0 wheel D is on node i. At that instant one has Pi PD,
Pi 1 Pi 2 Pi 3 0. For the vehicle to move a distance a the
required time is ta a/V, where V is the constant speed of the
vehicle. Hence wheel D will be on nodes i 1, i2 and i 3 at
times ta, 2ta and 3ta, respectively. In that case one has

Pi 1 PD at t ta ; Pi 2 PD at t 2t a ; Pi 3 PD at t 3t a
6
Without loss of generality, the subsequent analysis concentrates Fig. 6. Simple boundary conditions: (a) rollers on three vertical faces and bottom of
on node i1. At time t0 node i 1 is unloaded. After that, node i 1 domain; (b) rollers only at bottom and free vertical faces of domain.
starts being loaded since load PD leaves node i and moves towards
node i1. Using isoparametric nite element concepts, load PD is
distributed to nodes i and i 1 in a linear fashion during the time
interval 0rtrta and the same happens for nodes i1 and i2 for
the time interval ta rtr2ta, as shown in Fig. 9. The full unloading of
node i 1 occurs after load PD has reached node i2. This triangular
moving load variation with time shown in Fig. 9, is very close to that
of a haversine variation used elsewhere [1] but simpler and even-
though represents an approximation of the real situation, agrees well
with experimental results [40] in the framework of the linear theory
used herein. Taking into account that the vehicle considered has four
axles, the total loading will be the result of the superposition of loads
PA, PB, PC and PD, as shown in Fig. 10. The loading conguration of
Fig. 10 for node i1 is repeated for every other node of the pavement
surface with the only difference being from node to node the com-
mencement time of the above quadruple loading. The aforemen-
tioned procedure of load application in the time domain has been
programmed and that program has been appropriately interfaced
with the main ANSYS [7] program. Fig. 7. Typical heavy vehicle (truck) with four axles and total weight of 80 kN.
 N.D. Beskou et al. / Soil Dynamics and Earthquake Engineering 82 (2016) 6372
Fig. 8. Top of discretized plane y 0 of the pavement system of Fig. 1 with four
successive nodes along the x axis.
Fig. 9. Loading of node i 1 by load PD.
Fig. 10. Total loading of node i 1 by the whole vehicle.
In reality, the applied moving or stationary loads on pavements
are distributed rather than concentrated and refer to a single tire
(wheel) or a group of tires of the vehicle in contact with the pave-
ment. The contact area of a single tire or a group of tires can be
assumed to be of rectangular form with length l and width b along
the x and y directions, respectively [1] and this is compatible with
the fact that the faces of the nite elements at the top surface of the
domain are also of rectangular form. The distributed load is simu-
lated by an equivalent system of concentrated loads placed sym-
metrically on either side of the symmetry axis x. These loads are
treated as series of concentrated loads parallel to x as described for
the case of moving concentrated loads. Use is made of 28 axles of
loads numbered along the x axis for an almost perfect simulation.
Because of the symmetry of domain B adopted here (Fig. 2), only half
of the load and its rectangular area is taken into account.
4. Finite element solution in time domain
On the basis of the spatial modeling of the pavement structure
of Section 2, ANSYS [7] formulates the equation of its motion in
67
matrix form as [41]
Cfug
Mfug _ Kfug fF g 7
where [M], [C] and [K] are the mass, damping and stiffness
matrices, respectively, {u} and {F} are the vectors of the nodal
displacements and nodal external forces, respectively and overdots
denote differentiation with respect to time t. External forces refer
to nodes at the surface z 0 of the pavement structure coming
from the moving wheel loads of the vehicle. The moving load
starts at point A under zero initial conditions and terminates at
point B, as shown in Fig. 1.
Structural damping is assumed to be of the Rayleigh type,
meaning that
C d M d K 8
where the coefcients d and d are determined in terms of two of
the natural frequencies of the pavement system i and j and the
corresponding modal damping ratios i and j. Assuming i j ,
coefcients d and d have the form [41]
d 2i j =i j ; d 2=i j 9
Eq. (7) is solved in the time domain to directly determine the in
time pavement response evolution to the moving vehicle loads.
The step-by-step time integration algorithm of Newmark [41] of
the constant average acceleration type under zero initial condi-
tions is employed in this work. This algorithm is unconditionally
stable and provides highly accurate results provided an appro-
priate time step t has been selected compatible with the domain
discretization and the dominant frequency of the moving loads,
which is an increasing function of their speed.
The selection of the appropriate t is rst done on the basis of the
guidelines in [41]. Thus, if Lel is the minimum nite element length
and Vp the dilatational wave velocity as given by Eq.(2), one has
t r Lel =V p 10
An empirical relation provides t directly as [7]
t r 1=20 f 11
where f is the highest loading frequency in Hz.
For example, for Lel 0.15 m (elements along the motion axis in
Fig. 1) and Vp 500 m/s (a mean value for pavements) one has
from (10) that tr0.3  10  3 sec. On the other hand, on the
assumption that for a truck load, f 20 Hz, one obtains from (11)
that tr 2.5  10  3 sec. Thus, computations can commence with
tr 0.6  10  3 sec and continue with smaller and smaller values
of t until convergence of the maximum response values is
established or an acceptable error is reached at convergence in
case the exact solution for the response is available.
5. Numerical response studies
In this section validation studies of the proposed modeling asso-
ciated with geometry, boundary conditions, loading and solution
algorithms are presented. These have to do with comparisons on the
basis of a uniform half-space under point and distributed moving and
stationary constant loads between numerical and analytical solutions.
Comparisons on the basis of a layered half-space under distributed
moving and stationary constant loads between numerical and
approximate analytical methods for the cases of stationary loads on a
layered half-space are also made. Numerically obtained responses for
a single concentrated load, a single distributed load and a series of
concentrated loads are also compared. Finally, the effect of viscous
damping on the dynamic response of half-space or layered half-space
to single concentrated and distributed moving loads is also
investigated.
68 N.D. Beskou et al. / Soil Dynamics and Earthquake Engineering 82 (2016) 6372

5.1. Half-space elastic medium under moving and stationary loads displacement increases for increased velocity and occurs under the
moving load and at time, which decreases with increasing velocity.
Consider the road pavement structure of Fig. 1 with identical Consider now the case of a moving distributed load on the
properties in all three layers, i.e., with E 50  106 N/m2, v 0.25 surface of the half-space. The area over which the load P 80 kN
and 2000 kg/m3. The resulting domain becomes a bounded acts, has dimensions l and b along the x and y directions respec-
elastic half-space with wave velocities Vp 173.2 m/s and tively. In view of the fact that the dimensions of the elements used
Vs 100 m/s as obtained with the aid of Eq. (2). For this elastic on the surface of the full domain are 0.15 m  0.15 m, one can
half-space adopted from Eason [42], the Rayleigh wave velocity assume l 3  0.15 0.45 m and b 2  0.15 0.30 m resulting in
VR 92 m/s is the resonance wave velocity for a moving con- an area A l  b0.45  0.30 0.135 m2. Thus, the moving pres-
centrated load. In road pavements, realistic velocities of moving sure p P/A 0.5926 MPa, which is a reasonable value [1]. Table 3
vehicles, do not exceed V 50 m/s (180 km/h) and hence one deals provides the maximum values of the vertical displacements uz at a
with the subsonic case (VoVR) and does not worry about reso- depth of z  1.00 m below the surface point Q (Fig. 1) for two
nance. In this work, four vehicle velocities in the subsonic case are velocities V(m/s) of the above distributed load p moving along the
considered: V 10 m/s (36 km/h), V 20 m/s (72 km/h), V35 m/s x axis, three types of boundary conditions and three values of t
(126 km/h) and V50 m/s (180 km/h). for the domain of Fig. 2b, as obtained numerically by ANSYS [7]
Table 1 lists the maximum values of the vertical displacements and analytically by an approximate modication of the results of
uz at a depth of z  1.0 m below the surface point Q (15, 0, 0) in Eason [42], as explained in the Appendix. From the results of
Fig. 1 due to a single concentrated load P 80 kN moving along the Table 3 one can observe that convergence of the numerical results
x axis with V20 m/s and 50 m/s. The results correspond to three is obtained for t0.3  10  3 s for all cases of boundary condi-
types of boundary conditions and three values of t for the space tions, as in the case of concentrated load (Table 1). However, here
discretization of domain model B of Fig. 2b and are obtained the error for the case of rollers everywhere is lower and almost the
numerically by ANSYS [7] and analytically by Eason [42]. The case same (maximum e 2.65%) with that of the absorbing boundaries
indicating that for distributed loading one can employ the former
of the single concentrated load P is realized by assigning the values
type of boundary conditions at a reduced computational cost
of PD P 16 16 16 3280 kN and PA PB PC 0 in the gen-
without jeopardizing accuracy.
eral four wheel case of Fig. 7. The above table also records the
Furthermore, from Table 3 (see also the Appendix), one can
relative per cent error e dened as
observe that deections due to distributed load are consistently
  
e uapp uex =uex 100 12 slightly smaller than the ones for the concentrated load. This was
expected, as in the former case the total load is distributed over a
of the approximate numerical values uapp with respect to the large area and is not a point load as in the lateral case. Something
analytic exact ones uex of Eason [42]. From the results of Table 1 similar happens between a single point load and a series of point
one can observe that convergence of the numerical results is loads with a resultant force equal to the former point load. Indeed,
obtained for t0.3  10  3 sec for all cases of boundary condi- consider the displacement uz at a depth of 1.00 m below the sur-
tions. The errors for the cases of rollers everywhere (maximum face point Q (Fig. 1) as computed numerically for the cases of a
e3.72%) and absorbing boundaries (maximum e3.67%) are single load P 80 kN and a series of four concentrated loads
considered acceptable for displacements. Naturally one expects
the error to be higher for stresses and strains. Table 2 provides Table 2
maximum values of stresses x, y, z for the case of boundary Maximum stresses x, y, z at depth z  1.0 m due to a vertical concentrated load
conditions with rollers everywhere and t 0.15  10  3 and P 80 kN moving at speed V 20, 50 m/s on the surface of half-space for bound-
aries with rollers everywhere; values in parentheses denote % errors. Positive and
0.30  10  3 sec and compares them against the analytical ones in negative signs indicate tensile and compressive stresses, respectively.
Eason [42]. It is observed that, eventhough the maximum error for
two cases is 17.50%, for almost all the other cases, the maximum Stress (Pa) Analytic: Numerical:FEM [7] Numerical:FEM [7]
error is 8.63% which can be considered acceptable for design Eason t 0.30  10  3sec t 0.15  10  3sec
[42]
purposes. The same observations mentioned for Tables 1 and 2 for
V 20 m/s and 50 m/s hold also true for the other two velocities V 20 m/s x 3513 3488 (  0.71%) 3778 (7.54%)
V 10 m/s and 35 m/s considered here. y 3246 3708 (14.23%) 3814 (17.50%)
Fig. 11 depicts the time variation of uz at depth z  1.00 m z  39,065  36,476 (  6.63%)  36,317 (  7.03%)
V 50 m/s x 5903 6142 (4.05%) 6685 (13.25%)
below the surface point Q (Fig. 1) for the case of boundaries with y 3524 4128 (17.14%) 3828 (8.63%)
rollers everywhere, t 0.3  10  3 s and V 10, 20, 35, and 50 m/s z  45,180  43,064 (  4.68%)  42,331 (  6.30%)
as obtained by ANSYS [7]. It is observed that the maximum

Table 1
Maximum vertical displacement uz in m at depth z  1.0 m due to a vertical concentrated load P 80 kN moving at speed V 20,50 m/s on the surface of half-space
(domain model B) for three types of boundary conditions and three values of t.

Eason [42] t (sec) Rollers at bottom Rollers everywhere Absorbing boundaries

V 20 m/s uz 0.81608  10  3 m 0.6  10  3 uz 0.85186  10  3 uz 0.826032  10  3 uz 0.816635  10  3
e 4.38% e 1.22% e 0.08%
0.3  10  3 uz 0.878057  10  3 uz 0.80333  10  3 uz 0.80312  10  3
e 7.59% e  1.56% e  1.59%
0.15  10  3 uz 0.879081  10  3 uz 0.79735  10  3 uz 0.79651  10  3
e 7.72% e  2.29% e  2.40%
Eason [42] 0.6  10  3 uz 1.009815  10  3 uz 0.983899  10  3 uz 0.978023  10  3
V 50 m/s uz 0.96338  10  3 m e 4.82% e 2.13% e 1.52%
0.3  10  3 uz 1.041125  10  3 uz 0.92754  10  3 uz 0.92801  10  3
e 8.07% e  3.72% e  3.67%
0.15  10  3 uz 1.042955  10  3 uz 0.93122  10  3 uz 0.93141  10  3
e 8.26% e  3.34% e  3.32%
 N.D. Beskou et al. / Soil Dynamics and Earthquake Engineering 82 (2016) 6372
PA PB PC 16 kN and PD 32 kN, as shown in Fig. 7. It was found
that for the cases of V20 m/s and V50 m/s with employment of
absorbing boundaries and t0.3  10  3 sec, the values of max-
imum displacement uz due to the series of moving loads of Fig. 7,
are 0.352775  10  3 m and 0.4114520  10  3 m, respectively, i.e.,
smaller than the corresponding ones for the cases of the single
moving load.
In pavement design, static analytic solutions for the uniform or
the layered half-space under distributed load are widely used [1].
Thus, some comparisons of numerical static results against ana-
lytic ones are in order. ANSYS [7] can easily perform static ana-
_ {u}
lyses, i.e., solve Eq. (7) with {u} {0}. Use of the domain with
rollers everywhere at the boundaries under a static distributed
vertical load p 0.5926 MPa acting on the surface of the uniform
half-space of Eason [42] can lead to numerical results for the
vertical deection uz and stresses z, x and y under that load at a
depth of z  1.00 m below the surface point Q (Fig. 1) as shown in
Table 4. In that table, analytical results obtained from Boussinesq
[43] through the equations
 
69
are also shown for comparison. In the above, a is the radius of the
circular area equivalent to the rectangular one acted upon by the
distributed load. Thus, for the rectangular area 0.30  0.45 0.135 m2,
one has a0.20735 m. It is observed that the error for the displace-
ment is almost the same with that of the dynamic case (Table 1),
while the error for stresses is higher than that for displacements, as
expected. Furthermore, one can observe that for the numerical results
x a y, as expected, because of lack of axisymmetry in the
numerical model.
5.2. Layered half-space elastic medium under moving and stationary
loads
Consider the layered pavement structure of Fig. 1 with elastic
material properties as shown in Table 5 under the moving dis-
tributed load p 0.5926 MPa of Section 5.1. Using the domain
model B in conjunction with rollers everywhere boundary condi-
tions and t 0.3  103 sec, one can determine at z 0.00 m,
 0.15 m and  0.45 m below the surface point Q (Fig. 1) the
 maximum values of the dynamic vertical deection uz, the stresses
1 vpa   2   0:5  2 0:5  
uz 1 z= 1 2v 1 z=  z= 13 z, x, y and the strains z, x, y, for two values of the speed V
E
(20 m/s and 50 m/s), as shown in Table 6. It is observed that, as in
  
 2   1:5 the case of the half-space (Table 3), the response increases with
z p 1  1 a=z 14
velocity but not as much.
Consider now the same layered pavement structure of the pre-
  
 2   0:5   2   1:5 vious case under the same value of the distributed load applied sta-
x y p 1 2v=2 1 v 1 =z 0:5 1 =z 15
tically. The computed numerically static response values for uz, z, x,
y, z, x, y at z 0.00,  0.15,  0.45 are recorded in Table 7 together
1.0 with their approximate analytic values obtained by the method of
0.0 Odemark [5], which is used for design purposes. One can observe
from Table 7 that, while the displacement and strain results of Ode-
-1.0
mark [5] are higher than the numerical ones (displacements are even
-2.0
higher than the dynamic ones of Table 6), his stress results are lower
-3.0 than the numerical ones. Out of the displacements, strains and
uz (x 10-4) [m]

-4.0 stresses of Tables 6 and 7, in their dynamic and static versions,

respectively, uz at z 0.00 m, x/y at z  0.15 m and z at
-5.0
z  0.45 m are the most important ones for design purposes [1]. For
-6.0

-7.0 Table 4
V = 10 m/s Vertical deection and stresses at z  1.0 m due to a vertical distributed static
-8.0
V = 20 m/s load p 0.5926 MPa acting on a circular area of radius 0.20735 m at the surface
-9.0 V = 35 m/s of the uniform half-space. Positive and negative signs indicate tensile and com-
V = 50 m/s pressive stresses, respectively.
-10.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Analytic [43] Numerical [7] error %
Time [sec]
uz 0.781237  10  3m 0.762646  10  3m  2.38
Fig. 11. Time variation of vertical displacement uz at depth z  1.00 m below z  36.2610  10  3MPa  34.9024  10  3MPa  3.75
point Q of the half space for a concentrated load P 80 KN moving with speeds x 2.70265  10  3MPa 2.60308  10  3MPa  3.68
V 10, 20, 35, and 50 m/s (domain model B, rollers everywhere and y 2.70265  10  3MPa 2.79802  10  3MPa  3.53
t 0.3  10  3 s).

Table 3
Maximum vertical displacement uz in m at depth z  1.0 m below the surface point Q (Fig. 1) due to a distributed load p 0.5926 MPa moving at speed V 20,50 m/s on the
surface of half-space for three types of boundary conditions and three values of t.

Modied Eason [42] Rollers at bottom Rollers everywhere Absorbing boundaries t (sec)
V 20 m/s uz 0.800765  10  3 m uz 0.785836  10  3 uz 0.752302  10  3 uz 0.748779  10  3 0.6  10  3
e  1.86% e  6.05% e  6.49%
uz 0.819808  10  3 uz 0.785013  10  3 uz 0.780815  10  3 0.3  10  3
e1.90% e  1.96% e  2.49%
uz 0.820634  10  3 uz 0.786049  10  3 uz 0.782750  10  3 0.15  10  3
e2.48% e  1.84% e  2.25%
Modied Eason [42] uz 0.921687  10  3 uz 0.888979  10  3 uz 0.881889  10  3 0.6  10  3
V 50 m/s uz 0.9453199  10  3 m e  2.50% e  5.96% e  6.71%
uz 0.965644  10  3 uz 0.924995  10  3 uz 0.921498  10  3 0.3  10  3
e2.15% e  2.15% e  2.52%
uz 0.972167  10  3 uz 0.924523  10  3 uz 0.920269  10  3 0.15  10  3
e2.84% e  2.20% e  2.65%
70 N.D. Beskou et al. / Soil Dynamics and Earthquake Engineering 82 (2016) 6372

these displacements, strains and stresses their time histories for
V20 m/s and 50 m/s are depicted in Figs. 1214.
6. Effect of viscous damping
and hence the damping matrix [C] can be obtained in ANSYS [7] by
Eq. (8). Using this damping matrix one can nally obtain the
maximum vertical deections uz for V 20 m/s, 50 m/s at point Q
and at a depth of z  1.00 m using rollers everywhere and
t 0.30  10  3 sec as

For the two cases of the uniform and the layered elastic half- uz 0:802285  10  3 m ; uz 0:896853  10  3 m 18
space, as modeled by domain model B in conjunction with rollers
everywhere boundaries, the effect of viscous damping on the respectively, which are lower than the corresponding values of
pavement response to moving concentrated and distributed load, 0.80333  10  3 and 0.92754  10  3 m computed numerically for
respectively, will be studied in this section by using Eqs. (8) and the case of zero damping and shown in Table 1. These results of
(9). (18) can be compared with the numerical ones in Hung and Yang
Consider rst the case of the uniform half-space with 2% [44], which for the present data read uz 0.803930  10  3 m and
under the moving load P with V 20 m/s and 50 m/s. The rst ve uz 0.935440  10  3 m for V20 m/s and V 50 m/s, respectively.
natural frequencies of the domain model B with elastic properties The agreement is satisfactory.
as in Section 5.1 are determined by ANSYS [7] and found to be Consider now the case of the layered half-space with 2%
1 9:069904 ; 2 10:559835 ; 3 17:446960 under the distributed load p 0.5926 MPa moving with V 20 m/s
4 19:381113 ; 5 19:384820 rad=sec 16 and V 50 m/s. Following exactly the same procedure as before,
one obtains natural frequencies
With 0.02 for all modes and following Bathe [41], one can
determine 1 12.358899 and 2 19.382967 rad/sec as the 1 13:322238; 2 15:304206; 3 22:113608
averages of the rst three and last two frequencies in (16),
respectively and then the coefcients d and d from (9) assuming
4 24:830269; 5 24:8335107 rad=sec 19
i1, j2. The result is
damping coefcients
d 0:301875 ; d 0:002061 17
d 0:301875 ; d 0:002061 20
Table 5
Elastic material properties of layered domain of Fig. 1. (d1, d2, d3 are layer
and maximum vertical deections at z=0.00 m
thicknesses).
uz 0:862177  10  3 m ; uz 0:879694  10  3 m 21
Layers E (Pa) v (kg/m3)
for V20 m/s and 50 m/s, respectively, which are lower than the
Asphalt d1 0.15 m 1000  106 0.35 2500
corresponding values of uz 0.865126  10  3 m and 0.90190 
Base d2 0.30 m 400  106 0.35 2100
Subgrade d3 29.00 m 80  106 0.40 2000 10  3 m computed numerically for the case of zero damping and
shown in Table 6.

Table 6
Maximum vertical displacements, stresses and strains in elastic layered pavement structure under a moving distributed load p 0.5926 MPa for V 20 m/s and 50 m/s.
Stresses and strains with positive signs are tensile and those with negative signs are compressive.

V (m/s) z (m) uz (  10  3m) z (MPa) x (MPa) y (MPa) z (  10  3) x (  10  3) y (  10  3)

20 0.00  0.865126  0.35404  0.34933  0.36591  0.15411  0.10577  0.18898

0.00904 0.02593 0.00619 0.11791 0.05907 0.00567
 0.15  0.817536  0.44709  0.12554  0.06734  0.60676  0.11818  0.00724
0.01065 0.00823 0.00431 0.03258 0.16272 0.22423
 0.45  0.636677  0.0825  0.0166  0.00344  0.569  0.06467  0.00678
0.00462 0.05262 0.05477 0.03065 0.26645 0.27862
50 0.00  0.90190  0.35383  0.36685  0.37198  0.14274  0.12183  0.19358
0.01031 0.05871 0.00712 0.12115 0.06472 0.00593
 0.15  0.854722  0.4531  0.12958  0.07142  0.61553  0.12171  0.00727
0.012 0.01295 0.00713 0.03581 0.16891 0.22323
 0.45  0.670753  0.08437  0.02015  0.00371  0.58365  0.08253  0.01219
0.00483 0.05598 0.05656 0.03263 0.28031 0.28242

Table 7
Displacements, stresses and strains in elastic layered pavement structure under static distributed load (Approximate analytic values are in parentheses). Positive and
negative signs indicate tension and compression, respectively.

z uz (  10  3m) z (MPa) x (MPa) y (MPa) z (  10  3 MPa) x (  10  3 MPa) y (  10  3 MPa)

0.00  0.8403709  0.3316569  0.3468572  0.3636657  0.08297390  0.1034943  0.1261858

(  0.91830)
 0.15  0.7926878  0.4464080  0.09659979  0.06623167  0.6064473 0.1627205 0.2212876
(  0.86250) (  0.3890) (  0.04512) (  0.04512) (  0.88217) (0.26272) (0.26272)
 0.45  0.6121315  0.08216961 0.05164941 0.05436905  0.56583259 0.2620921 0.2774817
(  0.69230) (  0.0742) (0.00092) (0.00092) (  0.93669) (0.37789) (0.37789)
 N.D. Beskou et al. / Soil Dynamics and Earthquake Engineering 82 (2016) 6372 71

1.0 0.01

0.0 0.00
-1.0 -0.01
-2.0
-0.02
-3.0
-0.03
uz (x 10-4) [m]

-4.0

z [MPa]
-0.04
-5.0 V = 20 m/s
-0.05 V = 50 m/s
-6.0

-7.0 -0.06

-8.0 V = 20 m/s -0.07

V = 50 m/s
-9.0 -0.08

-10.0 -0.09
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
Time [sec] Time [sec]
Fig. 12. Displacement uz versus time at depth z 0.00 m under the distributed load Fig. 14. Stress z versus time at depth z  0.45 m under the distributed load
p 0.5926 MPa moving at speed V 20, 50 m/s on the layered half-space (domain p 0.5926 MPa moving at speed V 20, 50 m/s on the layered half-space (domain
model B, rollers everywhere and t 0.3  10  3 s). model B, rollers everywhere and t 0.3  10  3 s).

0.20

0.15
of the layered elastic half-space under a static distributed load, it
v = 20 m/s was found that Odemark's approximate analytic method used in
0.10 v = 50 m/s
design provides higher values for displacements and strains and
lower ones for stresses. Finally, it was found, as expected, that
0.05
viscous damping reduces the response.
(x 10 )

4) This work has been restricted to linear elastic material behavior and
0.00
its validation studies have been done on the basis of available ana-
lytical solutions. Extension of this work to the more realistic case of
-0.05
inelastic material behavior and its validation studies on the basis of
experimental results will be treated in a forthcoming publication.
-0.10

-0.15
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 Acknowledgments
Time [sec]
The rst author (Niki D. Beskou) acknowledges with thanks the
Fig. 13. Strain x versus time at depth z -0.15 m under the distributed load
p 0.5926 MPa moving at speed V 20, 50 m/s on the layered half-space (domain
support provided to her by the IKY Fellowships of Excellence for
model B, rollers everywhere and t 0.3  10  3 s). postgraduate studies in Greece-Siemens program.

7. Conclusions
Appendix
On the basis of the preceding studies one can state the fol-
lowing conclusions: Determination of approximate modied Eason's displacement
results for the case of distributed load. Consider the static cases of
1) An effort has been made to construct in the framework of a concentrated load P and a distributed load p P/2 over a cir-
ANSYS computer program, a general 3-D model for the simula- cular area of radius , both acting vertically on an elastic half-
tion in the time domain of the dynamic response of a exible space. The vertical deection under the load at a depth z is given
road pavement to a vehicle moving with constant speed under by [43]
linear elastic material behavior. 1 v3  2vP
ucz A:1
2) Comparison and convergence studies for the cases of the 2 zE
homogeneous and the layered half-space on the basis of  

available analytical solutions, helped to determine the domain 1 vpa   2   0:5  2 0:5  
udz 1 z= 1  2v 1 z=a  z=
geometry and its degree of discretization, the type of bound- E
aries to be employed and the time step to be used in order to :2
obtain acceptable accuracy at a reasonable computational cost.
for the cases of concentrated load P and distributed load p,
3) It was found, that the dynamic response is always higher than the
respectively. For the particular case of v 0.25, z1.0 m and
corresponding static one and that increasing values of vehicle
(0.45  0.30/)0.5 0.20735 m one can obtain from the above
speed increase the dynamic pavement response (more in the case
equations
of the half-space than in the case of the layered system). Further-
more, it was also found that a series of moving point loads or a udz =ucz 0:98123 A:3
distributed moving load result in a lower response than a single
Thus, on the assumption that relation (A.3) holds also for the
point moving load of intensity equal to the sum of those of the load
dynamic case (which is true for Vo 50 m/s and z/ 4 1 [42]), one
series or to the resultant force of the distributed load. For the case
can obtain maximum dynamic deections for the distributed load
72 N.D. Beskou et al. / Soil Dynamics and Earthquake Engineering 82 (2016) 6372
by simply multiplying those of Eason [42] for the concentrated
load by the factor 0.98123 shown at the right hand side of (A.3).
References
[1] Huang YH. Pavement analysis and design. Upper Saddle River, N. Jersey, USA:
Pearson Prentice-Hall; 2004.
[2] Cebon D. Handbook of vehicleroad interaction. London: Taylor & Francis;
1999.
[3] Burmister DM. The theory of stresses and displacements in layered systems
and applications to the design of airport runways. With discussion by Palmer
LA, Barber ES, Casagrande A, Middlebrooks TA. Proceedings of the Highway
Research Board. 23; 1943. p. 126148.
[4] Burmister DM. The general theory of stresses and displacements in layered
systems, I, II & III. J Appl Phys 1945;16:8994 126127 and 296302.
[5] Odemark N. Investigations as to the elastic properties of soils and design of
pavements according to the theory of elasticity, Meddelanden 77 fran Statens
Vaginstitut, Stockholm, Sweden, 1949. (In Swedish: Translation in English by
Hibbs MA, Silfwerbrand J, Ioannides AM, editors, available through TRID of
Transportation Research Board, USA; 1990).
[6] Beskou ND, Theodorakopoulos DD. Dynamic effects of moving loads on road
pavements: a review. Soil Dyn Earthq Eng 2011;31:54767.
[7] ANSYS. ANSYS LS-DYNA Users Guide, Release 13. Canonsburg, PA, USA.:
ANSYS Inc.Southpointe; 2010.
[8] Wang D, Roesler JR, Guo DZ. Innovative algorithm to solve axisymmetric
displacement and stress elds in multilayered pavement systems. J Transp Eng
ASCE 2011;137(4):28795.
[9] Cho YH, McCullough BF, Weissmann J. Considerations on nite element
method application in pavement structural analysis. Transp Res Rec
1996;1539:96101.
[10] Jooste FJ. Flexible pavement response evaluation using the semi-analytical
nite element method. Road Mater Pavement Des 2002;3:21125.
[11] Mulungye RM, Owende PMO, Mellon K. Finite element modelling of exible
pavements on soft soil subgrades. Mater Des 2007;28:73956.
[12] Mamlouk MS, Davies TG. Elasto-dynamic analysis of pavement deections. J
Transp Eng ASCE 1984;110(6):53650.
[13] Tu W. Response modelling of pavement subjected to dynamic surface loading
based on stress-based multi-layered plate. Department of civil engineering.
Columbus, OH, U.S.A: Ohio State University; 2007 Ph. D. Thesis.
[14] Lu Z, Yao HL, Wu WP, Cheng P. Dynamic stress and deformation of a layered
road structure under vehicle trafc loads: experimental measurements and
numerical calculations. Soil Dyn Earthq Eng 2012;39:10012.
[15] Nazarian S, Boddapati KM. Pavement-falling weight deectometer interaction
using dynamic nite-element analysis. Transp Res Rec 1995;1482:3343.
[16] Shoukry SN, William GW. Performance evaluation of backcalculation algo-
rithms through three-dimensional nite-element modelling of pavement
structures. Transp Res Rec 1999;1655:15260.
[17] Sousa JB, Lysmer J, Chen SS, Monismith CL. Effects of dynamic loads on per-
formance of asphalt concrete pavements. Transp Res Rec 1988;1207:14568.
[18] Zar Z, Siddharthan R, Sebaaly PE. Dynamic pavement strain histories from
moving trafc load. J Transp Eng ASCE 1994;120(5):82142.
[19] Siddharthan RV, Yao J, Sebaaly PE. Pavement strain from moving dynamic 3D
load distribution. J Transp Eng ASCE 1998;124(6):55766.
[20] Grundmann H, Lieb M, Trommer E. The response of a layered half-space to
trafc loads moving along its surface. Arch Appl Mech 1999;69:5567.
[21] Chatti K, Kim T, Yun K. Verication of the SAPSI-M program using full-scale
instrumented pavements and moving truck tests. In: Proceedings of the
international conference on accelerated pavement testing. Paper No CS13-4,
Reno, Nevada; 1999.
[22] Lefeuve-Mesgouez G, Peplow AT, LeHouedec D. Surface vibration due to a
sequence of high speed moving harmonic rectangular loads. Soil Dyn Earhq
Eng 2002;22:45973.
[23] Lee JH, Kim JK, Tassoulas JL. Dynamic analysis of a layered half-space subjected
to moving line loads. Soil Dyn Earthq Eng 2013;47:1631.
[24] Yang YB, Hung HH. A 2.5D nite/innite element approach for modelling
visco-elastic bodies subjected to moving loads. Int J Numer Methods Eng
2001;51:131736.
[25] Elsei MA, Al-Qadi IL, Yoo PJ. Viscoelastic modelling and eld validation of
exible pavements. J Eng Mech ASCE 2006;132(2):1728.
[26] Yin H, Salaimanian M, Kumar T, Stoffels S. The effect of loading time on
exible pavement dynamic response: a nite element analysis. Mech Time-
Depend Mater 2007;11:26588.
[27] Yoo PJ, Al-Qadi IL. Effect of transient dynamic loading on exible Pavements.
Transp Res Rec 2007;1990:12940.
[28] Al-Qadi IL, Wang H, Yoo PJ, Dessouky SH. Dynamic analysis and in situ vali-
dation of perpetual pavement response to vehicular loading. Transp Res Rec
2008;2087:2939.
[29] Ju SH. Finite element investigation of trafc induced vibrations. J Sound Vib
2009;321:83753.
[30] Kang Z, Liu D, Jiang R, Yang Y. Effect of vehicle speed and vehicle load on
damage of pavement. In: Proceedings of the international conference on
computational intelligence and vehicular system (CIVS 2010). IEEE; 2010. p.
428431.
[31] Liao J, Sargand S. Viscoelastic FE modelling and verication of a U.S.30 per-
petual pavement test section. Road Mater Pavement Des 2010;11:9931008.
[32] Khavassefat P, Jelagin D, Birgisson B. A computational framework for viscoe-
lastic analysis of exible pavements under moving loads. Mater Struct
2012;45:165571.
[33] Hirose S. Boundary element modelling and analysis of moving loads in time
and frequency domain. In: Chouw N, Schmid G, editors. Wave, 2000. A.A.
Balkema Rotterdam; 2000. p. 7182.
[34] de Barros FCP, Luco JE. Response of a layered viscoelastic half-space to a
moving point load. Wave Motion 1994;19:189210.
[35] Chen Y. Viscoelastic Modelling of Flexible Pavement. Akron, Ohio, USA: Uni-
versity of Akron; 2009 Ph.D.Thesis.
[36] Monismith CL. Flexible pavement analysis and design-a half century of
achievement. In: Rollins K, Zekkos D, editors. Geotechnical engineering state
of the art and practice, GSP 226. ASCE; 2012. p. 187220.
[37] Lysmer J, Kuhlemeyer RL. Finite dynamic model for innite media. J Eng Mech
Div ASCE 1969;95(4):85977.
[38] Hatzigeorgiou GD, Beskos DE. Soil-structure interaction effects on seismic
inelastic analysis of 3-D tunnels. Soil Dyn Earthq Eng 2010;30:85161.
[39] Mulliken JS, Karabalis DL. Discrete model for dynamic through-the-soil cou-
pling of 3-D foundations and structures. Earthq Eng Struct Dyn 1998;27:687
710.
[40] Mamlouk MS. General outlook of pavement and vehicle dynamics. J Transp
Eng ASCE 1997;123(6):5157.
[41] Bathe KJ. Finite element procedures. Englewood Cliffs, New Jersey: Prentice
Hall; 1996.
[42] Eason G. The stresses produced in a semi-innite solid by a moving surface
force. Int J Eng Sci 1965;2:581609.
[43] Boussinesq JV. Applications des Potentiels a LEtude de lEquilibre et du
Movement des Solides Elastiques. Paris: Gauthier Villars; 1885.
[44] Hung HH, Yang YB. Elastic waves in visco-elastic half-space generated by
various vehicle loads. Soil Dyn Earthq Eng 2001;21:117.