Computers and Geotechnics 93 (2018) 123–132

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Research Paper

Numerical prediction and back-calculation of time-dependent behaviour

of Ballina test embankment
Hans P. Jostad a,⇑, Francesca Palmieri a, Lars Andresen a, Noel Boylan b
a
NGI (Norwegian Geotechnical Institute), Sognsveien 72, 0855 Oslo, Norway
b
NGI (Norwegian Geotechnical Institute), Level 7, 40 St. Georges Tce, Perth, WA 6000, Australia

a r t i c l e i n f o a b s t r a c t

Keywords: The paper describes the Class A prediction and Class C back-calculations of the Ballina test embankment
Embankment using the finite element program Plaxis and the Soft Soil Creep Model (SSCM). The prediction underesti-
Soft soil
mated the measured settlement 3 years after construction by about 20%. This was mainly due to too high
Consolidation
stiffness in the transition zone beneath the clay and that SSCM underestimated the shear deformation of
Settlements
Drains
the clay. Furthermore, the horizontal permeability of the clay was overestimated. In the back-calculation,
FEM it was possible to obtain a excellent match with the measured settlements by reasonable modifications of
the input parameters.
Ó 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://
creativecommons.org/licenses/by/4.0/).

1. Introduction NGI delivered two different predictions, one based on hand-

calculation and on based on advanced numerical analyses using
Settlements of foundations and embankments on soft ground in the finite element program Plaxis (www.plaxis.nl). This paper
geotechnical engineering are often calculated using idealized 1D describes the numerical Class A prediction together with a Class
methods with simplified assumptions or elastic analytical solu- C back-calculation.
tions of load spread distribution with depths, pure vertical pore
pressure dissipation, and permeability and compressibility param-
eters from oedometer tests. Time dependent creep deformations 2. Background information
are added by a simple secondary consolidation term, e.g. Mesri
[1]. However, in some projects more accurate settlement predic- 2.1. Test site and embankment
tions are required. In these cases, 2D or 3D analyses using a fully
coupled displacement and pore water flow (consolidation) finite Two test embankments were constructed at the NFTF. Several
element (FE) program with a proper material model may be used. sampling, laboratory and in situ testing campaigns have been per-
In order to improve the accuracy and reliability of more formed to characterize the soil [2,3]. Based on geophysics, cone
advanced numerical analyses, the FE calculation models and the penetration (CPTU) and shear vane tests, it has been demonstrated
process of determining parameters need to be validated against that the stratigraphy is rather uniform across the site.
results from well defined and instrumented field cases. This was Seasonal groundwater variations of about ±1 m cause the in situ
the purpose of the test embankments constructed at the National pore pressure to vary with time. The average ground water level is
Soft Soil Testing Facility (NFTF) in northern New South Wales, about 0.5 m below the ground. Data obtained from vibrating wire
Australia. piezometers (VWP) installed within the Ballina clay below the
The Australian Research Council (ARC) Centre of Excellence for footprint of the western embankment (i.e. the one with vertical
Geotechnical Science and Engineering invited practising engineers drains) show that the groundwater is almost hydrostatic with
and academics to make predictions of the time dependent settle- depth.
ment, pore pressure dissipations and lateral displacements of the The depositional history suggests that the ground is likely to be
test embankment. geologically normally consolidated as substantial erosion is unli-
kely to have occurred. However, some overconsolidation through
⇑ Corresponding author. the seasonal changes in groundwater levels and creep have
E-mail address: [email protected] (H.P. Jostad). occurred.

http://dx.doi.org/10.1016/j.compgeo.2017.05.026
0266-352X/Ó 2017 The Authors. Published by Elsevier Ltd.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
124 H.P. Jostad et al. / Computers and Geotechnics 93 (2018) 123–132

The prediction is based on the stratigraphy deduced from CPT according to the calculated incremental displacements. The main
soundings and boreholes Inclo1, Mex1 and Inclo2, and aims at purpose of the updated mesh analyses is to account for that the
reproducing the settlement of the cross section 2 of the western excess weight of the embankment is gradually reduced as the
embankment. The soil layering of this cross section comprises of material settle below the ground water table. This is accounted
about 1.4 m thick alluvial clayey sandy silt, underlain by a 9.4 m for by using the ‘‘Updated water pressures” option in Plaxis.
thick estuarine clay layer, a 3.3 m thick transition zone, a 5 m thick The 15-node triangular element and the medium mesh option
sand layer and then a stiff Pleistocene clay layer. are selected in the calculation, leading to a total of 2151 elements.
In order to build the embankment, a working platform approx- This model is found to be fine enough to not be affected by any dis-
imately 95 m long by 25 m wide and 0.6 m thick was initially con- cretization errors.
structed. On top of this a 0.4 m thick sand layer was placed, before
the wick drains were installed. Lastly, a 2 m thick top earth fill 3.2. Soil models and properties
comprised of highly plastic clay was constructed on top of the sand
layer. The final crest of the embankment was 80 m long by 16 m The compressibility and shear deformation of the estuarine clay
wide. The slope of the sides was 3:2 (H:V). are modelled with the Soft Soil Creep Model (SSCM) [4]. This model
accounts for the stress dependent stiffness of the soil within the
framework of hardening plasticity. In addition, the model takes
3. Finite element analyses into account the time-dependent behaviour of the deformation,
i.e. creep. The hardening law of SSCM does not include directly
3.1. Finite element model the strain-induced destructuration such as for instance in Creep-
SCLAY1S [5,6]. Instead, the parameters are selected in order to
The numerical analyses are carried out by using the finite ele- model the significant stiffness reduction seen for this clay beyond
ment (FE) program Plaxis 2D version 2016.01 (www.plaxis.nl). the yield (pre-consolidation) stress in the stress range of interest.
Fig. 1 shows the finite element model used in the Class A predic- Thus, a strain independent value of the modified compression
tion. The model consists of 8 soil layers, the 0.6 m thick working index k⁄ is assumed to be appropriate to describe the material
platform, 0.4 m thick sand drain and the 2 m thick top embank- compressibility. The SSCM uses an associated plastic flow rule
ment. The model covers a total horizontal distance of 140 m. This based on an isotropic CamClay type cap surface as shown in
model is found to be sufficiently large enough such that end effects Fig. 2 (left). The hardening law is controlled only by the plastic vol-
do not affect the settlement beneath the embankment and the hor- umetric strain. This means that the additional shear deformation
izontal displacement at the edge of the embankment. The bottom due to slightly higher shear mobilisation than the KNC o -state may
boundary is taken at the top of the stiff Pleistocene clay. The be different than predicted by the isotropic SSCM model. However,
ground water table is in the Class A prediction taken at 1.2 m in order to control the shear deformation one need to include one
below the original terrain in order to fit the effective stress profile additional parameter that change the shape of the yield surface
given in [2]. between the KNC o -line and the failure line M. In the paper by Siva-
The effect of the wick drains is modelled by the vertical drain sithamparam et al. [7], one such model is proposed.
elements available in Plaxis, starting from the sand layer 1.0 m The input parameters to the SSCM that controls the compress-
above the ground continuing down to 14.9 m below the ground, ibility are the modified compression index, k⁄, the modified swel-
with a selected center distance of 3.2 m. When activated, the ling index, j⁄, the unloading/reloading Poisson ratio, mur, the
drains force the nodes with pore pressure degree of freedom along modified secondary compression index, l⁄, and the vertical effec-
the geometrical line to have a head equal to a specified value. In tive yield stress, rvc0 . The yield stress is defined by the over-
the Class A prediction the head is set equal to 0 m, i.e. the nodes consolidation ratio, OCR = rvc0 /rvo0 , or pre-overburden pressure,
are forced to have a hydrostatic pore pressure starting from the POP = rvc0  rvo0 . In SSCM, it is assumed that all plastic strain is
original ground level. The corrected horizontal permeabilities used time dependent. This means that the yield stress given by the
for the soil between the drains are calculated in Section 3.3. intersection between the elastic compression line and the elasto-
The ground is assumed to be horizontal even though the borings plastic virgin compression line is rate dependent, see Fig. 2 (right).
shows some small variations. Displacements along the bottom of The creep rate along the virgin compression line is de/dt = l⁄/teqv,
the model is fully fixed while the vertical boundaries are free to where teqv is given by a vertical strain increment (distance) from
move in the vertical direction and fixed in the horizontal direction. a reference line corresponding to to = 24 h, i.e. Dev,creep = l⁄ ln(teqv/
Pore water flow is prevented through the bottom and the vertical to). Therefore, when interpreting the input parameters to SSCM
boundaries of the estuarine layer. The other soil layers are consid- from a constant rate of strain (CRS) oedometer test, one need to
ered to be drained and thus pore water flow through their vertical account for the actual strain rate used in the CRS-test. Fig. 3 shows
boundaries are allowed. back-calculation of the CRS-test of the specimen from depth
In the analyses an updated mesh formulation is used. This 5.49 m in boring Inclo2 using SSCM with k⁄ = 0.263, j⁄ = 0.042,
means that after each calculation step, the nodal points are moved l⁄/k⁄ = 0.03, rvo0 = 40.5 kPa and rvc0 = 64 kPa (at the reference

Fig. 1. Finite Element Model of cross section 2 used in the Class A prediction.
 H.P. Jostad et al. / Computers and Geotechnics 93 (2018) 123–132 125

Fig. 2. The Soft Soil Creep Model, yield surface (left) and compression curves (right).

The back-calculated input parameters from the CRS tests are

shown in Fig. 4. The estuarine soft clay is divided into 4 layers, with
constant properties within each sub-layer. The parameters are
therefore average values within each sublayer.
The Soft Soil Model (SSM) is applied for the alluvial clayey
sandy silt between 0.75 and 1.4 m depth and for the transition
zone (10.8–14.0 m). SSM is similar to SSCM, however, without
the effect of time dependent creep. For the alluvial soil and transi-
tion zone, compressibility parameters are defined based on the
available oedometer tests. For the alluvial silt, there is only one
oedometer test at a depth of 0.79 m from Inclo2. From this test it
is seen that the yield stress is more diffuse and that the stiffness
at large stresses is higher than for the clay. The best fit of j⁄, k⁄
and POP for the actual stress range to this material is also given
in Table 1. Since the stress-dependent stiffness in SSM may lead
Fig. 3. Back-calculation of CRS test INCLO2_5.49m using the Soft Soil Creep Model
to unrealistic deformations in the most superficial part of the
with three different sets of input parameters.
deposit, the upper 0.75 m of the alluvial soil is modelled assuming
a constant oedometer modulus Eoed = 6900 kPa, adopting the
Mohr-Coulomb model. This value is uncertain and can be both
strain rate of 1 day) or a POP of 23.5 kPa, applying a strain rate of higher and lower, however the contribution to the total settlement
de/dt = 28%/day. This is called material set SSCM1 in Fig. 3. This is in any cases rather small.
automatically accounts for the rate dependent yield stress found A similar approach is used for the transition zone. The oedome-
in the CRS-tests. However, a good fit to the CRS-curve could also ter test (sample from depth 11.46 m) shows no clear yield stress
be obtained by increasing the creep ratio l⁄/k⁄ from 0.03 to 0.04 within the stress range considered, i.e. up to a stress of about
and increasing rvc0 from 64 to 66 kPa (SSCM2), or reducing k⁄ from 700 kPa. However, the observed nonlinear behaviour is fitted by
0.263 to 0.24 and decreasing rvc0 from 66 to 62 kPa (SSCM3). In the a ‘‘bi-linear” curve with k⁄ = 0.015 and j⁄ = 0.009, and POP = 21 kPa.
calculation of the settlement of the embankment, the effect of Due to the significantly stiffer response of this material, the depth
changing l⁄/k⁄ has largest effect since it increases the contribution where this zone is included affects the calculated settlement.
from creep when extrapolating from the CRS strain rate to the sig- The Hardening Soil Model (HSM) is used for the sand layer
nificantly slower strain rate in the field. In addition, the creep rate below the transition zone. Compressibility parameters are for the
as function of OCR is governed by the exponent, n = l⁄/(k⁄-j⁄), sand layer defined by empirical correlations in Lunne et al. [9],
which affect the creep rate to the side of the embankment not based on the measured cone resistance, qt. The main properties
experiences increased effective stresses. The same type of back- used in the analyses are the reference Young’s modulus at 50%
calculation is performed for the other CRS-tests on specimens from shear mobilisation, Eref ref
50 = 66 MPa, oedometer modulus, Eoed = 82 -
ref
Inclo2. Since the simulations of the CRS-tests start at a low isotro- MPa, unloading/reloading Young’s modulus, Eur = 161 MPa, a refer-
pic effective stress, the results are also dependent on mur, which ence stress, pref = 100 kPa, moduli exponent, m = 0.5 and friction
controls the effective stress ratio, Drh0 /Drv0 = mur/(1-mur), before angle, / = 42.5o. However, the contribution from this layer to the
o = rh /rv at the vertical
0 0
yielding and the effective stress ratio KNC total settlement is found to be negligible.
yield stress rvc0 . The Poisson’s ratio, mur, is assumed for all layers The vertical consolidation coefficient, cv is assumed based on
equal to 0.2. In particular, for soft clays it represents an average Pineda et al. [2]. In addition, these values are modified for taking
value, while this value represents a lower limit for silty sand and into account the smear effects due to the installation of prefabri-
sand as shown by Das [8]. The initial lateral earth pressure coeffi- cated vertical drains and the influence of modelling a radial flow
cients Ko are defined according to Pineda et al. [2]. pattern of drainage with a 2D model as described in Section 3.3.
Based on Fig. 12b in [2], a constant l⁄/k⁄ = Ca/Cc = 0.03 was used The most relevant parameters for the clay and silt layers used in
for the estuarine soft clay in the Class A prediction. the prediction are listed in Table 1.
126 H.P. Jostad et al. / Computers and Geotechnics 93 (2018) 123–132

Fig. 4. Input parameters used in back-calculation of CRS tests on samples from Inclo2 using the Soft Soil Creep Model and values (given by the vertical solid lines) used in the
class A prediction.

Table 1
SSCM/SSM input parameters for the clay/silt layers used in the Class A prediction.

Layer Depth [m] c [kN/m3] k* [–] j* [–] l*/k* [–] KNC

o [–] kv [m/day] kh [m/day] POP [kPa] Kinitial
o [–]
Alluvial silt 0.75–1.4 17.3 0.04 0.01  0.47 6.4e3 16.0e3 44.5 0.90
Estuarine 1 1.4–2 14.4 0.10 0.018 0.03 0.4 0.4E3 1.0E3 24 0.55
Estuarine 2 2–4.5 14.4 0.181 0.026 0.03 0.4 0.4E3 1.0E3 24 0.55
Estuarine 3 4.5–9.1 14.4 0.263 0.039 0.03 0.4 0.4E3 1.0E3 24 0.55
Estuarine 4 9.1–10.8 14.4 0.20 0.033 0.03 0.4 0.4E3 1.0E3 24 0.55
Transition zone 10.8–14.0 19.1 0.016 0.009 – 0.38 0.5e3 1.3e3 21 0.55

3.3. Wick drains Ch  t

Tr ¼ 2
ðde Þ
Preinstalled vertical drains (PVDs) installed at a square grid
spacing of 1.2 m cover the 24 m width of the main embankment
area. The PVDs are assumed to be draining at the sand layer above      
n2 n kh
the working platform and at the base level at 14.9 m depth below Fn ¼  ln þ lnðsÞ  0; 75
n2 1 s ks
the ground where they are penetrated into the sand layer. The con-
solidation of the estuarine clay layer beneath the embankment is
therefore assumed to be dominated by horizontal drainage where ch is the horizontal coefficient of consolidation, t is the con-
towards the PVDs. The vertical drains have thus been modelled solidation time, n = de/dw and s = ds/dw. With this approach the aver-
considering a system of unit cells as shown in Fig. 5 with each age degree of consolidation U(t) at any consolidation time can be
drain at the centre of a unit cell with equivalent diameter calculated. It is noted that this approach assumes at any time a con-
de = 1.05  1200 mm = 1260 mm. stant degree of drainage and thus a constant value of excess pore
The equivalent diameter of the wick drain was calculated as pressures within the entire zone of PVDs for a given soil layer.
dw = 2(w + t)/p = 2  (100 mm + 3 mm)/p = 66 mm. The dimen- In the 2D model, the same time to 90% degree of drainage, t90 is
sions (t and w) are taken from the specifications of the CeTeau obtained by using B = 3.2 m as the distance between the vertical
drain. The drains were installed using a rectangular shaped man- drains and an equivalent horizontal permeability kekv equal to
drel with dimensions 120 mm  60 mm giving an equivalent the intact horizontal permeability kh.
diameter of the mandrel dm = 115 mm. It is assumed a smeared In the Class A prediction, it was assumed a vertical coefficient of
(remoulded) zone of diameter ds and reduced permeability ks consolidation at the vertical yield stress of about 2 m2/year. Based
around each drain caused by installation. Based on recommenda- on experience from Sweden [10], the horizontal permeability was
tions in Hansbo [9] ds = 2.0  dm = 230 mm and the reduced perme- taken as 2.5 times the vertical permeability. The calculated time
ability ks is assumed as ks = kh/3. for 90% degree of drainage was approximately 0.4 year = 150 days.
The average degree of drainage U within each cell of PVDs is cal- In Fig. 10a in [2], it is shown that cv may be even smaller than
culated for each layer based on Hansbo [9] accounting for installa- 1 m2/yr at an effective vertical stress between 100 and 200 kPa.
tion smear effects but neglecting any flow resistance of the drains: The lowest value is about 0.3 m2/yr at a depth of 7.70 and
  9.73 m. With ch = cv = 0.3 m2/yr, the time to reach 90% degree of
8T r
UðtÞ ¼ 1  exp ð1Þ drainage increases to about 9 years, demonstrating the large
Fn uncertainties in the predicted drainage time.
 H.P. Jostad et al. / Computers and Geotechnics 93 (2018) 123–132 127

Fig. 5. Representation of system of PDVs by the unit cell approach.

3.4. Calculation phases 3.5. Class A prediction and comparison with measurements

For cross section 2, settlement recording starts before the work- The calculated settlement of the original terrain at the center of
ing platform construction. An initial consolidation phase of 8 days cross section 2 versus time is compared with the measured settle-
is considered leading to have comparable results. Drains are acti- ments in Fig. 6. The most representative settlement plates are SP2
vated in four different phases, thus, taking into account the instal- and SP3. However, as shown in Fig. 6, there are very small differ-
lation time of the vertical drains. The equivalent permeabilities ences between the four settlement plates. The calculated settle-
were introduced when half of the drains were installed. Informa- ment agrees very well with the measured settlement up to the
tion about the calculation phases are listed in Table 2. end of filling. After that, the calculation first overestimates the rate
In the consolidation phases the automatic time stepping proce- of settlement, before both the rate and total settlement become too
dure in Plaxis was used. That the applied time steps were not too small. In July 2016, the calculated settlement is about 1.18 m while
large was manually checked by inspection of the calculated history the measured settlements are between 1.46 and 1.52 m. The corre-
curves. sponding rate of calculated and measured settlement in July 2016

Table 2
Construction phases applied in the numerical analyses.

Phase Time
Initial phase –
Consolidation 8 days
Working platform construction 5 days
Consolidation 7 days
Drainage sand construction 8 days
Consolidation 8 days
Activate drains 1 1 day
Activate drains 2 1 day
Consolidation with eqv. permeabilities 5 days
Activate drains 3 1 day
Activate drains 4 1 day
Consolidation 5 days
Main construction 13 days
End of consolidation and creep 1770 days Fig. 6. Comparison between measured and calculated (Class A) time settlements
curves.
128 H.P. Jostad et al. / Computers and Geotechnics 93 (2018) 123–132

is about 40 mm/year and 1000 mm/year. The main reason for the 3.6. Uncertainties in Class A prediction
larger calculated settlement rate immediately after the construc-
tion of the fill is due to overestimation of the rate of pore pressure The following main uncertainties affected the predicted time-
dissipation as will be shown later, while the most likely reason for settlement curve:
under-predicting the settlement and rate of settlement in July
2016 will be discussed in the following.  The measured permeability is varying significantly within the
Fig. 7 shows the calculated and measured settlement profiles at estuarine clay (from 61010 m/s at 9.73 m depth to 6108 m/
Mex1 in July 2016. From this figure, it is seen that one reason for s at 3.65 m depth, both at the yield stress, based on Table 2 in
the under predicted settlement is that the calculated compaction Pineda et al. [2]. In addition, based on experiences from Sweden
of the transition zone at 10.8–14 m depth is too small. At 11 m [10], the intact horizontal permeability was taken as 2.5 times
depth the measured settlement is about 6 cm larger than calcu- the vertical. However, the permeability anisotropy might be dif-
lated. In addition, between 2 and 5 m depth the measured com- ferent for the estuarine clay.
paction is about 14 cm larger than calculated. Finally, in the top  Limited information was available about the stiffness of the soil
2 m of the soil profile the measured compaction is about 13 cm above and below the estuarine clay, i.e. only CRS-tests at 0.79
while the calculated compaction is about 6 cm smaller. and 11.46 m in Inclo2.
Fig. 8 shows the calculated and measured average vertical strain  The ground water table is varying with the seasons and it may
in Mex1. This shows the same results as in Fig. 7, that the largest increase up to the sand drainage layer during the consolidation
contributions to the under prediction come from the upper part process.
of the estuarine clay, and beneath and above the estuarine clay.  Spatial variation in the soil properties and layer thicknesses. The
Fig. 9 shows the calculated and measured horizontal displace- soil properties were based on CRS-tests on samples taken from
ment profiles with depth in Inclo1 at the end of construction and Inclo2.
in July 2016. The calculated and measured horizontal displacement  The effect of the disturbance of the soil during installation of the
at the end of construction agree rather well, although, the mea- drains. The permeability ks in the remoulded (smear) zone
sured displacements is larger below 8 m. The measured maximum around the drains was assumed to have a horizontal permeabil-
horizontal displacement in July 2016 is 22 cm, while the calculated ity being 1/3 of the intact horizontal permeability. However,
is only 10 cm. This means that the model was not able to predict this permeability can be both higher and lower.
the development of horizontal displacement with time, which  The stress path in the CRS-tests before yielding is different from
more or less follows the development of settlement with time. This in situ since it starts from significantly lower effective stresses
additional shear deformation is therefore one reason for the under in the CRS-tests. This may affect the predicted yield stress.
estimated settlement between 2 and 5 m depth.  SSCM may predict wrong shear strains for a shear mobilisation
Fig. 10 shows the calculated and measured total pore pressure larger than the KNC o -line.
at 2, 6 and 10 m depths. The calculated pore pressures are initially  The creep parameter is uncertain, l⁄/k⁄ = Cc/Ca is varying
too small and also decreases faster with time compared with the between 0.025 and 0.07 in the IL creep tests presented in Pineda
measured values. The calculation automatically accounts for the et al. [2].
effect of piezometers (nodes) being moved downwards with the  The idealization of the varying properties with depth within the
settlement. As the position of the ground water table is assumed estuarine soft clay into four sub-layers introduce some
to be fixed, the hydrostatic pore pressure is then increased due uncertainties.
to the increased depth below the ground water table.
A systematic evaluation of the uncertainties in the calculated
settlement is presented in a companion paper by Liu et al. [11].

4. Class C back-calculation

4.1. Required modifications

Based on comparison between predicted and measured results,

the following modifications are included in the Class C back-
calculation:

 Based on the measured pore pressure before construction, the

ground water level was increased from 1.2 m to 0.9 m below
the ground surface. In order to keep the same vertical yield
stress, POP was correspondingly increased by 3 kPa.
 The depth to the top of the transition zone is increased from
10.8 m to 11.0 m. Furthermore, in order to account for the grad-
ual increase in sand content and thus increase in stiffness, a
lower stiffness is given to the top 1 m of this layer using the Soft
Soil Creep Model. By including the creep the POP needed also to
be changed.
 The stiffness of the alluvial soil is reduced.
 The equivalent permeabilities of the estuarine clay are reduced.
This is justified by the significantly lower cv found in Fig. 10a in
[2] compared to the values at the yield stress in Table 2 in the
same paper. Furthermore, there are limited justifications for
Fig. 7. Comparison between measured and calculated (Class A) settlements profiles using a higher horizontal permeability than the vertical.
in July 2016 at Mex1.
 H.P. Jostad et al. / Computers and Geotechnics 93 (2018) 123–132 129

Fig. 8. Comparison between measured and calculated (Class A) average strain curves at Mex1.

Instead, the small differences between the measured ch and cv

as well as kh and kv suggest a low permeability anisotropy for
the estuarine clay (Kelly et al. [3])
 The stiffness of the top estuarine clay between 2.0 and 4.5 m is
reduced. This is partly done in order to compensate for that the
SSCM underestimates the shear deformation.
 The creep index is increased to l⁄/k⁄ = 0.05 for the estuarine
clay between 4.5 and 11 m.
 In order to increase the shear deformation and corresponding
horizontal displacements at the periphery of the embankment,
the KNC
o is increased (i.e. the M-value defining the cap-surface,
shown in Fig. 3 (left), is reduced). In some cases it was even nec-
essary to reduce the friction angle, since the M-value cannot be
less than the corresponding Mohr Coulomb failure line, M = 6
sin//(3-sin/) in triaxial compression.

4.2. Modified soil properties

The stiffness of the Alluvial soil and the Transition zone were
first reduced in order to fit the measured settlement profile after
3 years shown in Fig. 7, i.e. the average vertical strain in these
Fig. 9. Comparison between measured and calculated (Class A) horizontal dis-
placement profiles in Inclo1in July 2016.
zones. Then, the KNCo was increased in the Estuarine clay layers in
order to increase the shear deformations in the soil beneath the

Fig. 10. Comparison between measured and calculated (Class A) total pore pressure histories in VWP6.
130 H.P. Jostad et al. / Computers and Geotechnics 93 (2018) 123–132

embankment and thus better fit the measured horizontal displace- ity kekv by 50%, k⁄ by 20% and l⁄ by 20% compared to the parame-
ment profile at the edge of the embankment shown in Fig. 9. ters that gave the best fit to the average vertical strain between
Finally, the SSCM input parameters k⁄, l⁄/k⁄ and kekv were adjusted Magnets 2 and 3. The final parameters used in the Class C predic-
until a good fit with the measured average vertical strain curves, tion are presented in Table 3.
shown in Fig. 8, were obtained. It was chosen to keep the vertical By simulating CRS tests, however by starting from the initial
yield stresses the same as for the Class A prediction. The adjust- effective stresses, it is demonstrated that the modified parameters
ments were performed by first changing kekv, to fit the initial part given in Table 3 still agree with the CRS-tests from Inclo2 as shown
of the strain curves (e.g. typically the first 6 to 12 months). The Fig. 12.
k⁄-values were then adjusted in order to fit the magnitude of the
measured average strains after 3 years. Finally, the slope of the 4.3. Comparison with measurements
measured average vertical strain curves in July 2016 were fitted
by changing the contribution from the creep, i.e. the l⁄/k⁄-ratios. Fig. 13 shows that the Class C back-calculation gives time-
This process was repeated until a good fit was obtained. The effect settlement curves that perfectly fits the measured settlement
of changing the different parameters is demonstrated in Fig. 11. curves at Mex1. From Fig. 6, it is seen that in order to improve
The figure shows the effect of increasing the equivalent permeabil- the results compared to the Class A prediction, the total settlement

Fig. 11. Effect of changing the SSCM input parameters compared to the best fit parameters for the average vertical strain between Magnets 2 and 3.

Table 3
SSCM/SSC input parameters for the clay/silt layers in the Class C back-calculations.

Layer Depth [m] c [kN/m3] k* [–] j* [–] l*/k* [–] KNC

o [–] kv [m/day] kvekv, h [m/day] POP [kPa] Kintial
o [–]
Alluvial 2 0.75–1.5 17.3 0.05 0.010 – 0.73 0.01 0.01 47.5 0.58
Estuarine 1 1.5–2.0 14.4 0.15 0.018 0.04 0.73 1.50E3 1.50E3 27 0.58
Estuarine 2 2.0–4.5 14.4 0.20 0.020 0.04 1.94 0.80E3 0.80E3 27 0.62
Estuarine 3 4.5–9.1 14.4 0.24 0.039 0.05 0.73 0.08E3 0.04E3 27 0.58
Estuarine 4 9.1–11.0 14.4 0.18 0.033 0.05 0.73 0.05E3 0.02E3 27 0.58
Transition 1 11.0–12.0 19.1 0.08 0.015 0.04 0.73 0.32E3 0.32E3 27 0.58
Transition 2 12–14.25 19.1 0.03 0.010 0.03 0.73 0.60E3 0.60E3 50 0.58

Fig. 12. Simulated CRS tests at characteristic depths within each of the Estuarine clay layers using modified parameters for the Class C calculation together with some CRS
tests from Inclo2.
 H.P. Jostad et al. / Computers and Geotechnics 93 (2018) 123–132 131

Fig. 13. Comparison between measured and calculated (Class C) time settlements curves.

after 3 years needed to be increased, and the rate of settlement

needed to initially be reduced and after about 1 year to be
increased. Based on Fig. 8, it is seen that the strain in the upper part
of the estuarine clay (between 1 and 2 m depths) first of all needed
to be increased. From Tables 1 and 3, it is seen that k⁄ was
increased from 0.1 to 0.15. This correction could be justified by
too few tests within the actual depths and correspondingly large
uncertainty in the actual parameter. In addition, it was necessary
to compensate for that the material model, SSM, is underestimat-
ing the shear deformation in this layer. The increased settlement
rate after 1 year, is obtained by a reasonable increase in the creep
contribution, i.e. l⁄/k⁄-is increased from 0.03 to 0.04 in the upper
part of the estuarine clay and 0.05 between 4.5 and 11 m. This
means that the creep strain after 3 years may be as large as ecreep = -
l⁄ln(t/tref) = 0.050.2ln(3365 days/1 day) = 7%. This means that the
contribution due to creep for this clay is significant. In addition, the
equivalent permeability between 4.5 and 11 m depths is reduced
from 1E3 m/day to 0.02–0.04E3 m/day. The parameters used
in the Class C prediction can still be justified by the values in [2].
Fig. 14 shows that the horizontal displacement in July 2016 is
Fig. 14. Comparison between measured and calculated (Class C) horizontal
displacement profiles in Inclo1.
also increased at the periphery of the embankment and agrees sig-
nificantly better with the measurements at Inclo1 compared to the

Fig. 15. Comparison between measured and calculated (Class C) total pore pressure histories in VWP6.
132 H.P. Jostad et al. / Computers and Geotechnics 93 (2018) 123–132

Class A prediction. In order to obtained better results another friction angle below the measured value from the triaxial tests.
material model than the SSC model is recommended to be used Alternatively, a model [7] that accounts for this effect by an input
as discussed in Section 3.2. parameter that control the curvature of the yield surface could be
Fig. 15 shows that the calculated pore pressure histories also used.
fits the measured histories better. The main reason for this is the
lower equivalent permeablies used, as discussed above. Further-
Acknowledgements
more, the plane-strain idealization of the drains may overestimate
the pore pressure dissipation in the middle between the drains.
The authors gratefully acknowledge the support through an
internal research project (GBV) which is partially funded by the
5. Conclusions Norwegian Research Council. In addition fruitful discussions with
the authors (Suzanne Lacasse, Zhongqiang Liu and Jung Chan Choi)
The paper describes the Class A prediction and Class C back- of the companion paper regarding uncertainties in the calculated
calculations of the test embankment at the NFTF near Ballina in results.
Australia using the FE program Plaxis 2D. The Soft Soil Creep Model This research did not receive any specific grant from funding
(SSCM) was used for the deformation calculation of the soft estuar- agencies in the public, commercial, or not-for-profit sectors.
ine clay. The Class A prediction underestimated the measured set-
tlement 3 years after construction by about 20%. This was due to
uncertainties in the creep index of the soft estuarine clay and the References
stiffness of the soil above and below the soft clay. SSCM was also
[1] Mesri G. Coefficient of secondary compression. J Soil Mech Found Div ASCE
underestimating the shear deformation of the soft estuarine clay.
1973;99(1):123–37.
In addition, the horizontal permeability was overestimated based [2] Pineda JA, Suwal LP, Kelly RB, Bates L, Sloan SW. Characterisation of the Ballina
on wrong assumptions regarding the anisotropy ratio and neglect- clay. Géotechnique 2016;66(7):556–77.
[3] Kelly RB, Pineda JA, Bates L, Suwal LP, Fitzallen A. Site characterisation for the
ing the effect of reduction due to void ratio decrease below the
ballina field testing facility. Geotechnique 2016. 15-P-211.
yield stress. This effect could have been accounted for by using a [4] Vermeer PA, Stolle DFE, Bonnier PG. From the classical theory of secondary
void dependent permeability formulation available in Plaxis. How- compression to modern creep analysis. In: Proc. 9th int. conf. comp. meth. and
ever, since equivalent horizontal permeabilities are used that adv. geomech., Wuhan, China, vol. 4; 1998. p. 2469–78.
[5] Karstunen M, Sivasithamparam N, Brinkgreve RBJ, Bonnier PG. Modelling rate-
account for remoulding during installation of the drains and the dependent behaviour of structured clays. In: International. conference on
idealization of the 3D flow pattern by a 2D model, made it difficult installation effects in geotechnical engineering, 24–27 March 2013,
to use this feature. But, it is checked that the uncorrected vertical Rotterdam; 2013. p. 43–50.
[6] Sivasithamparam N, Karstunen M, Bonnier P. Modelling creep behaviour of
permeabilities used in the analysis agree with the void ratio depen- anisotropic soft soils. Comput Geotech 2015;69:46–57.
dent permeabilities in Fig. 11a in Pineda et al. [2]. [7] Sivasithamparam N, Castro J. An anisotropic elastoplastic model for soft clays
In the Class C back-calculation, it was possible to obtain perfect based on logarithmic contractancy. Int J Numer Anal Meth Geomech
2016;40:596–621.
match with the measured settlements by reasonable modifications [8] Das BM. Soil mechanics laboratory manual. New York, USA: Oxford University
of the input parameters. SSCM is thus generally well suited for Press; 2002. p. 99–108.
modelling settlements of embankments on soft clay including the [9] Lunne T, Robertson PK, Powel JJM. Cone penetration testing in geotechnical
practice. London: Blackie Academic & Professional; 1997.
important contribution from creep. However, SSCM may underes-
[10] Hansbo S. Consolidation of fine-grained soils by prefabricated drains. In: Proc.
timate the shear deformations for shear stress ratios above the KNCo - 10th ICSMFE, 1981, vol. 3; 1981. p. 677–82.
line. This can be mitigated by lowering the top point of the Cam- [11] Liu ZQ, Choi JC, Lacasse S, Nadim F. Uncertainty analyses of time-dependent
behaviour of Ballina test embankment. Comput Geotech 2018;93:133–49.
Clay cap surface given by M (i.e. increasing KNCo ) and reducing the