Geotechnical Engineeringfor JransportationInfrastructure, Barends et al.

(eds)

(8 1999 Balkema, Rotierdam, ISBN 90 5809 047 7

Rediction on the long-term behavior of subsoils under high-speed

railways
Pgdiction du comportement h long terme de sols SOUS des chemins
de fers h grande vitesse
H.-G. Kempfert & Y. Hu
Institute of Geotechnique, Universi~,
of Kassel, Germany

Keywords: dynamic stress, granular soils, quasi-static model, permanent settlement, FEM
ABSTRACT: In this Paper, some in-situ measurements related to the dynamic stress of the substructure and the subsoil of railway traffic are illustrated. By analyzing the test results from literature, some empirical stress-straincycle number relationships of ballast and sands are formulated.
Based on these, a quasi-static model predicting the long-term deformation and the stability of the
railway foundation is proposed and implemented by using the finite element method. To illustrate
the applicability of this model, a caiculation example is given.

&sUMI?: Dans ce papier, sont illustrks quelues mesures in-situ en relation avec la sollicitation
dynamique des sous-structures et sols du trafic de chemins de fers. Par l'analyse des resultats de
tests, recherches dans la littkrature, sont formules quelques nombres empiriques de cycles de
soilicitations en relation avec du lest (bailast) et sable. Base sur cela, un modele quasie-statique
predisant la deformation i long terme et la stabilite de la fondation du chemin de fer est propose et
accomplit par la methode des elements fuiis. Pour illustrer l'application de ce modele, un exemple
de calcul est donne.
1 INTRODUCTION
Railway tracks must be laid on the ground in some way and anchored to it. The dynamic loading
from railway traffic is carried by the superstructure, substructure and subsoil. The resulting settlement may lead to the deterioration of track systems. This problem is serious on high-speed lines
and especially serious on lines with mixed traffic. Here the resulting dynamic loadings are large,
but the allowable total settlement, as well as the differential settlement of the tracks is less than on
normal lines. Therefore, the prediction of settlement under such conditions is of practical significance in evaluating the long-term behavior of track system.
In literature two different procedures exist describing the constitutive relationships of soils under cyclic loading. The fust is built on the classical theory of anisotropic elasto-plasticity and
seems to be theoretically well-founded. This is represented by the nested yield surface model of
Mroz et al (1978) as well as by the bounding surface model of Dafalias & Herrmann (1982). It is
well lmown from the test results, that plastic deformation of soils occurs, even if the cyclic loading
lies within the static yielding surface. This is put down to the translation of the yielding surface
during cyclic loading, similar as the experimentally observed Bauschinger effect of meta1 on cyclic
loading. The formulations mentioned above combine the isotropic and kinernatic hardening together and rnake the consideration of this phenomenon possible. The comparison between theoretical prediction and experimental results under triaxial conditions shows that some aspects of soil
behavior under cyclic loading can be predicted qualitatively by using these models, for example the
movement of effective stress path toward critical state line during undrained cyclic loading and the
existente of critical stress ratio under undrained condition. However, some important aspects espe-

1824 H. -G. Kempfert & Y. Hu

cially the quantitative description of them, such as the plastic volume change as well as the hysteresis phenomenon, may not be properly simulated using these models. Furthermore, using these formulations the cyclic course must be followed step by step and formidable data be stored during the
calculation. For practical problems, e.g. railway track, the cyclic number to be analyzed can be
more than 105.This prevents them from being used in practice. Although some improvements on
the simplification were rnade, the practical application of such models has been rarely reported.
Another procedure describes the soil behavior directly according to the test results of cyclic
loading, and may be more reliable, e.g. Andersen (1991), as well as Song (1990). Here, the computational procedure does not trace each cyclic course step by step. The accumulated plastic deformation dunng cyclic loading is taken into consideration using the empiric relationships obtained
from the tests on representative soil samples. As a result, the calculation procedure is significantly
simplified.
In this study, the dynamic stresses of the railway underground are evaluated by analyzing the
data from some measuring projects. Based on the analysis of the cyclic triaxial test results of ballast
and sands in literature, some empiric relationships are formulated describing the mechanical behavior of such two soils under cyclic loading, especially the long-term behavior. Upon these and
some assumptions, a quasi-static model is proposed and then numerically implemented by using the
finite element method. A computational example demonstrates the capacity of the model as well as
its applicability in practice.

2 MEASURED DYNAMIC STRESS

The dynamic stresses resulting from railway traffic depend on train speed, type of superstructure,
depth as well as type of soil. In Figure la, the measured dependency of the rnaximum vertical dynarnic stress on train speed is presented for the measuring Cross sections 1 and 2 on the HannoverWrzburg railway line. A clear increase of the resulting dynamic stress in the substructure and the
subsoil can be observed within the range of the train speed between 150 and 300 kmlh.

measured (gauge 102)

measured (gauge 202)

0 ,

e--*concipte slab t m k

A-

40

--r asphdtic slab t m k

.-a

0.86 m below rall top

m--4

F
h

bdlastisted h c k

1.48 m below rail top

n
m

E
20

50

I
1
I
100 150 200

250

Train speed [ k m h ]

a)

I
300

---3----

350

20

400

b)

Figure 1. Measured dynamic stress: a) maximum vertical stress versus train speed (project HannoverWrzburg); b) influence of superstructure (project Kutzenhausen).

The influence of the superstructure on the resulting dynamic stress in the substructure and the
subsoil is revealed by analyzing the measured results of the project Kutzenhausen, See Figure lb.

Prediction on the long-term behavior of subsoils under high-speed railways

1825

From ballasted to asphaltic as well as concrete slab tracks, the resulting maximum dynamic vertical
Stresses become smaller.
Based on the results illustrated above and other analyses, the dynamic loading in the substmcture and the subsoil arising from the railway traffic could be evaluated as follows: With reference
to the static stress 0, deterrnined from the wheel Set loads, the dynamic increasing factor @ = odo,
is 1 up to a train speed of 150 km/h.Beyond that point, the factor increases linearly with train
speed until it i ~ ~ c h ae srnaximum at 300 km/h;then, @ becom~sindependent of the train speed
again. Based on our analysis of the existing in-situ measurements, the maximum factor @ may be
about 1.3 for the substructure and the subsoil under slab tracks, and 1.7 under ballasted tracks.

3 STRESS-STRAIN-CYCLIC NUMBER RELATIONSHIP OF GRANULAR SOLS

Extensive experimental studies on ballast and sands have been made under cyclic triaxial condition,
e.g. Raymond & Williams (1978), Diyaljee & Raymond (1982), Hettler (1987). The most tests
have been carried out with dry materials and low frequency. The influence of Pore water pressure
as well as fiequency are generally not included. Some in-situ measurements indicate that during
train passing the resulting excess Pore water pressure in granular soils is very low and decreases
within very short time down to Zero after train passing. Upon this, the factor of Pore water pressure
rnay be excluded.
One essential result is that there exists some limit of cyclic deviatoric stress, defined as a ratio to
the static limit value K = (ol - C T ~/(cl
) ~ -, cr3),,f.
~
1s the cyclic stress smaller than the limit, the resulting cyclic and permanent deformation of the soil specimens will gradually converge to a corresponding stable value. In this state the soil specirnen behaves as a quasielastic material. In contrast
to this, the cyclic and plastic deformation increases nearly linearly from cycle to cycle leading to
failure within short time, if the limit is exceeded.
For the cyclic stable case, a lot of empiric relationships describing permanent vertical strain caCP
under triaxial condition have been proposed in literature. According to the reanalysis by the
authors, the semi-logarithmic and double-logarithmic functions rnay be applied for bailast and
sands, respectively:
ballast:
sand:

log&: = l o g a + a . l o g N

The analysis shows that the coefficient a is nearly independent of cyclic deviatoric stress (ol 03)~as well as static hydrostatic stress 03, and therefore can be Seen as constant with a given relative density. For the coefficient a, the following empiric relationships have been proposed:
ballast: a = . ( a , / ~ , ) X - q 2 (Pa=100kN/m2)

Here, the cyclic stress level is defined as q = (ol - 03)~/(ol - o3),,f. and X are two curve fitting coefficients.
Another important relationship for modeling the long-term deformation is the permanent radial
strain bcPunder triaxial condition. Unfortunately, there exist only few available results from literature. The reanalysis of the test results of the ballast materials from Raymond & Williarns (1978)
indicates that the ratio bCP may be expressed as a function f(N) as follows:

h and o are two curve fitting coefficients. They are dependent on q and 03 and may be assessed by
using the following empiric relationships:

1826 H.-G. Kempfert & Y. Hu

h = Al-( 1 - O J O ~ , +
~ )h2-q and

o = oo.(ofl,)ln

Xi, h2, %,o and oo are deterrnined from test results.

4 CALCULATION MODEL
For simulating the stress-strain-relationship of ballast and sands observed in cyclic triaxial tests, a
quasi-static model is proposed, See Figure 2a. Here, the rnaximum of cyclic dynarnic loading is applied as a quasi-static stress o in the system. The model consists of 4 elements: spring E, nonlinear
viscous dashpot ql, linear viscous dashpot q2 and limiting value o k .

rll>>r12

q, + constant

.
permanent strain

a)
Figure 2. a) One-dimensional quasi-static model;

---

totale strain

b)
b) Schematic strain - cycle number - relationship.

According to the results from literature, the cyclic strain part E ~ is

" generally much smaller than
the permanent strain cCP.
Therefore, it is rational to assume that the quasi-elastic modul E is independent of the cycle number. The viscous dashpot 111 is introduced to simulate permanent strain
for the cyclic stable case and is dependent on the cyclic number N. This dependency can be determined using the empiric relationship proposed in section 3. For the failure case, the conventional
visco-plastic formulation is used, See Figure 2. The limiting value o k can be calculated using the
parameter K and the static strength parameter q.
The onedimensional conceptual model has been generalized to the three-dimensional case assuming that the principle of superposition can be applied to compute the permanent strain resulting
from cyclic stress components in the three primary directions. This generalized quasi-static model
has been implemented in the FE-Program "GEOCYCL" by using the initial strain algorithm.
5 A CALCULATION EXAMPLE

In Figure 3a, a conventional ballasted track on the uniform fine-grained sand is illustrated as a calculation example. It is assumed that the wheel load of a train can be idealized as a line load with a
value of 30 kNlm The design speed is assumed to be 300 kmlh and the cycle nurnber up to 10'.
Under consideration of the dynarnic increasing factor, described in section 2, the quasi-static line
load should be 1 . 7 ~ 3 kN/m.
0
The chosen cross-section and the twodimensional FE-mesh is shown

Prediction on the long-term behavior of subsoils under high-speed railways

1827

in Figure 3b. The essential parameters used in the calculation are given in Figure 3. The numerical
calculation was carried out on PC (Pentium II) and the total calculation time was about 30 minutes.

p = 1.7~30
kN/m

ballast
I

1;
i

iI
I

Figure 3.

bound surface
uniform fine-grained sand

V
-

ballast: y = 17 kN/m3,E = 50 h4N/mz,

V = 0.33, a = 0.22, = 7.3, X = 0.22,
hl = 0.1, h2= 0.5,03,0 = 80 W/&,
% = 0.01, <p = 38", K = 0.8.
@:,y = 19 kN/m3,E = 30 MN/&,
v=0.35 a=O.l,=0.24,~=0.33,
h = 0.1, cp = 35O, K = 0.8.

a) Details of problem;

b) Computation section, FE-mesh.

The calculated total settlement of sleeper and ground surface depending on the cycle number is
illustrated in Figure 4a. A total settlement of Ca. 13.3mm is predicted at a cycle number of 10'. The
difference between the two curves shows the deformation part resulting from the ballast. The deformation rate decreases quickly in the initial20.000 cycles and then becomes slowly. In Figure 4b,
the distributions of the settlement on the ground surface are shown for some chosen cycle numbers.
The plastic Zone in ballast and underground depending on the cycle number is illustrated in Figure
5. No further extension of the plastic Zone can be observed after the cycle number of 500.

6 CONCLUSIONS
The in-situ measurements indicate that for train speed between 150 and about 300 km/h the resulting rnaximum dynamic stress in the substructure and subsoil exhibits large dependency on the train
speed. In addition, the type of superstructure has large influence on the resulting dynamic stress.
The measured stress under the conventional ballasted tracks is much larger than those under the
slab tracks.
Based on the reanalysis of the existing results of triaxial tests from literature, some empiric
stress-strain-cycle number-relationships have been formulated for ballast and sands. For simplifying the calculation procedure, a quasi-static formulation has been proposed. The corresponding
threedimensional formulation has been implemented in a finite element model.
A twodimensional example of the railway foundation on uniform fine-grained sand has been
calculated by using the developed program "GEOCYCL". The predicted settlement and differential
Settlement as well as the plastic Zone are illustrated as a function of cycle number. The calculation
results seem to be feasible and the proposed model applicable for practice.

1828 H.-G. Kempfert & Y. Hu

Cyclenumber N 1-1

a>

Distance f b m raii axis [m]

b)

Figure 4. a) Settlement depending on cycle number; b) Settlement distributions on the ground surface.

Figure 5. Development of plastic Zone.

Andersen, K. H. 1991. Foundation design of offshore gravity structures. In O'Reilly et al (eds.), Cyclic loaing
of soils,
Dafalias, Y. F. & Herrmann, L. R. 1982. Bounding surface formulation of soil plasticity. In Pande & Zienkiewicz (eds.), Soil Mechanics-Transient and cyclic loads. Wiley, London.
Diyaljee, V. A. & Raymond, G. P. 1982. Repetitive load deformation of cohesionless soil. J.Geotech. Engrg.,
ASCE, 108(10).
Hettler, A. 1987. Schottertriaxialversuchemit statischem und zyklischem Belastungsverlauf. ETR 36.
Mroz, Z., Norris, V. A., & Zienkiewicz, 0. C. 1978. An anisotropic harding model for soils and its application to cyclic loading. Int. J. Num. Ana. Methods in Geomech. Vol. 2.
Raymond, G. P. & Williams, D. R. 1978. Triaxial tests on dolornite railroad ballast. J. Geotech. Engrg.,
ASCE, 104(6).
Song, E. X. 1990. Elasto-plastic consolidation under steady and cyclic loads. Research report of the Technical University Delft.