Computers and Geotechnics 93 (2018) 163–177

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Research Paper

Prediction of embankment performance using numerical analyses –

Practitioner’s approach
Kim F. Chan ⇑, Bosco M. Poon, Darshana Perera
GHD Pty. Ltd., Level 2, 29 Christie Street, St Leonards, NSW 2065, Australia

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

Keywords: A 3 m high embankment with prefabricated vertical drains was constructed over Ballina clay. It has been
Soft soil thoroughly instrumented for monitoring over three years after construction. Based on the data available
Consolidation at the site, the authors undertook Class A predictions of embankment performance using two approaches:
Elastic viscoplastic model a simple 1D consolidation analysis and a sophisticated large strain finite element analysis (FEA) using Soft
Creep
Soil Creep (SSC) model. Class C predictions were then conducted using the SSC model in FEA, with and
Vertical drains
without large strain. It is demonstrated that the SSC model can give satisfactory results when large strain
Finite element analysis
FEA was used.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction With reference to Fig. 1, the soil stratigraphy comprises gener-

ally a thin layer of soft to firm alluvial sandy silt near the ground
This paper presents comparisons of measured settlements, soil surface, followed by very soft estuarine silt/clay, over firm sandy
layer compressions, pore pressures and horizontal displacements clay transitioning to clayey sand, and underlain by medium dense
at the Ballina trial embankment with those of Class A predictions sand and stiff Pleistocene clay. Initial pore water pressures from all
using two soil models: a simple 1D consolidation analysis with vibrating wire piezometers (VWP) at the site indicated that the
Mesri type creep model (Hypothesis A) preferred by engineering groundwater table varied between RL
0.1 m and RL +0.1 m prior
practitioners due to its simplicity; and a sophisticated large strain to embankment construction. For the present settlement predic-
finite element analysis (FEA) using elastic viscoplastic Soft Soil tion analysis, a groundwater level of RL 0 was adopted. With the
Creep model (Hypothesis B) that is mostly endorsed by academia. original ground surface level typically at about RL 0.3 to RL 0.5,
The merits and limitations of these two prediction approaches are the groundwater table was about 0.4 m below the original ground
outlined and improvements to the predictive capability are dis- surface. There were fluctuations in groundwater level since the
cussed in the subsequent Class C predictions. A reflection on completion of embankment filling. The influence of groundwater
whether the complex reality of foundation soils can or cannot be level variation on predictions is discussed in Section 6 of this paper.
captured by the complexity of the adopted soil model is provided
in the conclusions section.
3. Adopted parameters in class A prediction

2. Background 3.1. Stress history and undrained shear strength

The 3 m high embankment was constructed in 2013 over Bal- As discussed in Pineda et al. [1], Constant Rate of Strain (CRS)
lina clay treated with prefabricated vertical drains (PVD). The tests on high quality piston-sampled Ballina clay were conducted
embankment was thoroughly instrumented including vibrating at a displacement rate of 0.004 mm/min. The preconsolidation
wire piezometers, settlement plates, extensometers, horizontal stress, r0 p estimated from these tests are higher than that when
profile gauges and borehole inclinometers to monitor the perfor- tested at a slower displacement rate, but no important changes
mance of the foundation soils during progressive filling and three in the shape of the compressibility curves are expected. To correct
years of subsequent consolidation. for the strain rate effect, the approach of Watabe et al. [2] was
adopted in which a correction factor of 0.84 was applied to the
⇑ Corresponding author. r0 p values. Fig. 2a shows the r0 p profile assessed from CRS tests,
E-mail addresses: [email protected] (K.F. Chan), [email protected] (B.M. as well as from conventional oedometer tests. While the r0 p of
Poon), [email protected] (D. Perera). the CRS are higher than that of the oedometer tests, good

http://dx.doi.org/10.1016/j.compgeo.2017.07.012
0266-352X/Ó 2017 Elsevier Ltd. All rights reserved.
164 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177

~RL 0m

Fig. 1. Geotechnical section of the instrumented trial embankment.

agreement is indicated after r0 p of the CRS results are corrected for data for the derivation of r0 p and OCR as outlined above than the
the strain rate effect by applying a reduction factor of 0.84. Also inferred OCR from other testing such as the in-situ piezocone
shown in Fig. 2a is the in-situ vertical effective stress, r0 v with RL (CPTu) test results. Furthermore, by utilising the strain rate cor-
which was calculated based on the adopted groundwater level out- rected OCR values obtained from the CRS tests and the undrained
lined in Section 2, in conjunction with the bulk unit weights shear strength, Su, measured from field vane tests (FVT) and the
obtained from laboratory test results. The over-consolidation ratio, CPTu, a site specific Su
OCR correlation can be established with
OCR calculated based on r0 p/r0 v, is shown in Fig. 2b. Fig. 2c shows great confidence as follows.
the strain rate corrected OCR from all available CRS tests, along Jamiolkowski et al. [3] and Ladd [4] indicated that the variation
with the adopted OCR profile for Class A prediction. in Su/r0 v with OCR can be approximated by the SHANSEP equation
Given the high quality of the samples and the CRS tests, it is
Su =r0v ¼ S  OCRm ð1Þ
considered appropriate to place greater emphasis on the CRS test

Fig. 2. Profiles of (a) r0 p, (b) OCR, and (c) adopted OCR.

 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177 165

where m was set to 0.95 for this prediction based on past experi- 0.4
ence in the region. The strength ratio, S, was assessed to be 0.225
0.35
by matching the calculated Su from Eq. (1) (using the strain rate cor-
TC
rected OCR values from CRS tests) with the field vane shear, Su (FV), 0.3
which have been adjusted by applying the Bjerrum’s [5] correction

su / σ'yield-corrected
SS
factors for appropriate plasticity index, PI (see Fig. 3a and b). The PI 0.25 TE

profile with depth is shown in Fig. 6a. 0.2

As pointed out by Bjerrum [6], Su(FV) can be greater than the
average shear strength mobilised along the surface of sliding dur- 0.15
ing embankment slope failure. This is due to (i) the vane test is car- 0.1 Ladd (1991)
ried out at a fast strain rate, and hence overestimates the field Natural Ballina clay
TC tests
TC tests
undrained strength, and (ii) the vane shear does not accurately 0.05 TE tests
SS tests
TE tests
measure the effects of anisotropy. Kelly et al. [7] presented data
0
that confirms the anisotropic behaviour of Ballina clay. These 0 10 20 30 40 50 60 70 80 90 100
results are reproduced in Fig. 4, which show that the undrained PI (%)
shear strength measured from Triaxial Compression (TC) tests are
greater than those obtained from Triaxial Extension (TE) tests. As Fig. 4. Undrained strength anisotropy of Ballina Clay.
illustrated in Fig. 5, TC test is analogous to those parts of the slip
surface where shearing is induced by increasing vertical stress
(i.e. beneath the embankment), whereas TE test is to the area
where shearing is induced by increasing the horizontal stress (i.e.
near the toe of the slip surface). To account for the different strain
rates and the effect of anisotropy, Bjerrum [6] suggested that the
vane strength be adjusted empirically by the Bjerrum correction
factors.
There are a number of CPTu conducted in the area. Owning to
the homogeneity of the foundation estuarine soil at the trial
embankment site, all of the CPTu derived Su match reasonably well
with the corrected vane shear by adopting appropriate cone fac-
tors, Nkt, ranging between 15 and 17.5 as summarised in Table 1.
In Kelly et al. [7], the CPTu data from the test site were com-
pared with field vane shear and TC test results. However, both of
these tests may overestimate the average strength mobilised along
Fig. 5. Relevance of laboratory shear tests to shear strength in the field [6].
the surface of sliding in the field as noted by Bjerrum [6]. Kelly
et al. [7] reported that the values of Nkt obtained from assessing
shear vane (uncorrected) and TC tests are 13.2 and 12.2, respec-
tively. These Nkt values are lower than those presented in the pre- The OCR, in turn, can be inferred from the CPTu derived Su via
sent paper (see Table 1) after correlating CPTu data with Eq. (1) and using the above SHANSEP parameters of m = 0.95 and
‘‘corrected” vane shear test results. S = 0.225. These CPTu inferred OCR values compare favourably

Fig. 3. Profiles of (a) OCR inferred from CPT08, (b) Su obtained from CPT08, (c) Su obtained from T-bar TB08, and (d) Su obtained from dilatometer SDMT-8.
166 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177

Fig. 6. Plots of (a) Index properties; (b) CR, (c) RR, (d) cae(NC) and (e) cae(NC)/CR.

Table 1
Summary of CPT tests and comparisons.

CPT Associated CRS tests Associated shear vane testsb Associated SDMTb Associated T-bar
a a
CPT ID Nkt CRS ID S m SV ID SDMT ID T-bar ID N T-bar

CPT34 17 INCLO_2 0.225 0.95 SV34 SDMT-34 – –

CPT08 15 INCLO_2 0.225 0.95 SV08 SDMT-08 TB08_UWA 11
CPT07 16 INCLO_2 0.225 0.95 SV07, SV02 and SV06 – – –
CPT06 17.5 INCLO_2, MEX09 0.225 0.95 SV06 – – –
a
SHANSEP parameters from Eq. (1).
b
Bjerrum correction factor applied.

with the strain rate corrected OCR from the CRS tests. Fig. 3b shows 3.2. Compressibility
a typical Su fit for CPT08 using a calibrated Nkt value of 15. The
comparison of the corresponding OCR profiles is shown in Fig. 3a. The compressibility curves obtained from CRS tests exhibit
Two dilatometer tests and one T-bar test were also conducted at non-linear response in the normally consolidated range with the
the test site. The Su profile derived from the dilatometer tests were largest compressibility just passing r0 p and then reduces with
applied with the Bjerrum correction factors. These corrected Su increasing r0 v, which is the consequence of progressive soil
profiles are shown to compare well with the corrected field vane de-structuration. For the present trial embankment, the stress
shear Su(FV), as well as the Su profiles calculated from the OCR of range was less than 200 kPa after the application of 3 m high
CRS tests using the SHANSEP parameters of m = 0.95 and embankment load, hence justifying the use of a constant CR for this
S = 0.225. Fig. 3d shows a typical example of the comparison for stress range just passing r0 p. Using the conventional Casagrande’s
dilatometer SDMT-8. The results of T-bar TB08 also compare well graphical approach, the assessed compression ratio CR (=cc/(1
with the Su profile established from the vane shear and CRS results, + e0)) and recompression ratio RR (=cr/(1 + e0)) are typically 0.65
as shown in Fig. 3c. The adopted T-bar factor, Nt-bar, was 11, which and 0.055, respectively, at RL
4 m to RL
10 m (Fig. 6b and c).
falls within the expected range of about 10–13 from literature. The adopted normally consolidated creep strain rate, cae(NC) is up
The assessment of CRS and CPT results outlined above indicated to 0.02 per log time cycle as obtained from oedometer creep tests
good match between Su and OCR profiles using a single set of (Fig. 6d). The cae(NC)/CR ratios generally lie at about 3% (Fig. 6e).
SHANSEP parameters of m = 0.95 and S = 0.225. This is consistent This is deemed to be consistent with the published literature
with the finding by Mesri [8] which indicated that by associating (e.g. Mesri et al. [11]), which suggests that ca/cc is about
Bjerrum’s correction factor with the relationship between 0.04 ± 0.01 for inorganic clays and silts. For 2D FEA, the 1D com-
Su(FV)/r0 p and PI, the average strength mobilised along the slip pressibility parameters of CR, RR and cae(NC) can be related to the
surface Su(mob) can be expressed, independent of PI, as critical state parameters of k⁄, j⁄ and l⁄, respectively, as given in
the FEA program PLAXIS user’s manual.
SuðmobÞ ¼ 0:22r0p ð2Þ
The clayey soils of different index properties can be seen to
Note that this relationship implies implicitly that m = 1 in Eq. (1). show distinct compressibility characteristics. For example, the PI
Conversely, for the adopted m = 0.95, S was about 0.225. As men- and LL of the soils between RL
4 m and RL
10 m are generally
tioned in Wood [9] back analysis data of field failures presented about 70% and 120%, respectively. This soil layer exhibits similar
in Larsson [10] confirmed that Eq. (2) provides a reasonable compressibility values, which are distinctively higher than those
estimate of the field strength for inorganic clays. for the soil layer above (between RL
2 and RL
4 m) with lower
 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177 167

PI and LL. Furthermore, the OCR also varies within each of the soil (a) Normally consolidated C v(NC) (b) Liquid limit (%)
layers. Therefore, it is prudent to discretise the soil layer appropri- (m2/yr)
ately in the geotechnical model in order to capture the changes in 0 1 2 0 40 80 120 160
0 0
soil compressibility outlined in Fig. 6, as well as the changes in OCR
delineated in Fig. 2c. The adopted geotechnical model for FEA is -2 -2

RL (mAHD)
summarised in Table 3 (refer to Section 4.2 for discussion).

RL (mAHD)
-4 -4

-6 -6
3.3. Rate of consolidation properties for 1 D consolidation analysis
-8 -8

For 1D consolidation analysis, the dissipation of excess pore -10 -10

water pressure depends on the coefficient of consolidation, cv. It
-12 -12
is well known that cv is stress and stress history dependent and 100.0
is not an intrinsic property of the material. From the CRS test (c) Undisturbed samples:

Coefficient of Consolidation c v
cv in range of recompression lies
results provided in Pineda et al. [1] and shown in Fig. 7a, cv reduces above this line
as r0 v increases up to r0 p, beyond which cv is fairly constant.
10.0 cv in range of virgin
For the 1D consolidation analysis employed in the Class A pre- compression

(m2/year)
diction, two sets of cv values were input for each discretized soil
layer, for a soil in the normally consolidated state, cv(NC) and for a
1.0
soil in the over-consolidated state, cv(OC). Fig. 8a shows the cv(NC) 0.8
measured in the CRS tests with RL (cv at r0 v > r0 p), which are con- 0.4
Completely remoulded samples: c v
sistent with the empirical correlation proposed by Navfac [12] as lies below this line
shown in Fig. 8c for a liquid limit (LL) of up to about 120% 0.1
0 20 40 60 80 100 120 140
(Fig. 8b). Fig. 7a can be replotted as cv vs. OCR (Fig. 7b). For the soil LiquidLimit (LL)
in the lightly over-consolidated state with OCR typically about 1.1–
1.5 during fill placement, cv(OC) is about 1–4 m2/year. Fig. 8. (a) cv(NC) with RL, (b) LL with RL, and (c) NAVFAC [9] correlation of cv with LL.

3.4. Rate of consolidation properties for 2D FEA 4.1. Simple 1-D consolidation analysis method

For 2D FEA using PLAXIS, cv is not a direct input, but is a compos- The simple 1-D consolidation analysis was carried out using an
ite parameter that depends theoretically on both the coefficient of in-house finite difference program. Whilst the consolidation of
permeability, k, and the coefficient of volume compressibility of soils was under 1-D conditions, the vertical stress that was applied
the soil, mv. The mv value is calculated based on the bi-linear e- on each of the discretised soil layer had been adjusted with the
log r0 v compressibility model defined by CR and RR, whereas k is Boussinesq stress influence factor to account for 2-D load spread-
related to the void ratio, e, in accordance with e = e0 + ck log (k/k0), ing. The 1-D primary consolidation settlement was assessed based
where e0 is the initial void ratio, ck is the permeability index and on the bi-linear recompression and virgin compression model on a
k0 is the initial permeability value. Fig. 9 shows the measured e0 semi-logarithmic space. Creep settlement was assumed to com-
and k0 from CRS tests, along with the adopted design profile. In par- mence near the end of the primary consolidation when the degree
ticular, the e0 values were used in assessing ck for each discretised of consolidation (DoC) reached 90%.
soil layer based on ck = 0.5e0 as proposed by Tavenas et al. [13]. Settlement in soft soil often accompanies by lateral extrusion of
the foundation material. In the 1-D analysis, the maximum lateral
sub-soil displacement, ym, at the toe of the fill batter can be
4. Prediction approaches assessed as a proportion of the maximum settlement, s, of the
embankment (Fig. 10a). For embankments constructed over soft
Two analytical methods were carried out for the Class A predic- soils treated with PVD, Ladd [4] indicated that the deformation
tion of the embankment performance, namely, (i) a simple 1-D ratio, ym/s is about 0.2 during fill placement. For long term consol-
consolidation analysis with Hypothesis A (Mesri type) creep model idation of embankments after fill placement, ym/s = 0.16 can be
and (ii) a sophisticated large strain 2-D FEA using elastic viscoplas- adopted as suggested by Tavenas and Leroueil [14]. The shape of
tic Soft Soil Creep Model (Hypothesis B) embedded in PLAXIS 2D. the lateral soil displacement profile with depth was evaluated

(a) Over consolidated Normally (b)

consolidated
100 100
cv (m2/year)

10
cv (m2/year)

10

1 CRS_INCLO2_RL-5.09
1
CRS_INCLO2_RL-6.23
CRS_INCLO2_RL-4.49
CRS_INCLO2_RL-6.87
0.1 0.1
0. 1 1 1 1.2 1.4 1.6 1.8 2
σ'v / σ'p (insitu) OCR

Fig. 7. Plots of (a) cv versus r0 v/r0 p and (b) cv versus OCR (from CRS tests).
168 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177

(a) Initial Void Ratio, e0 (b) Permeability (m/s)

0.0 1.0 2.0 3.0 4.0 1.E-10 1.E-09 1.E-08 1.E-07 1.E-06
0 0
-2 -2

RL (mAHD)
-4 -4

RL (mAHD)
-6 -6

-8 -8

-10 -10
CRS Yield (INCLO2)
-12 CRS - INCLO2 -12 IL CREEP
Pineda et al. 2016 CRS Yield (MEX9)
-14 CRS - MEX09
-14 BAT
Adopted profile
-16 -16

Fig. 9. (a) Initial void ratio and (b) initial permeability from CRS tests.

(a) (b) 0 0.2 0.4 0.6 0.8 1.0 Y = y/ym

Z Y = 1, when y=ym
=
z/
D

Proposed shape of horizontal displacement by Tavenas

(1979) [15]: Y = 1.78Z3- 4.7Z2 + 2.21Z + 0.71

Fig. 10. (a) Settlement induced lateral soil displacement, (b) Shape of horizontal displacement.

using the empirical expression proposed by Tavenas et al. [15] available software program PLAXIS 2D. The model utilises a form
(Fig. 10b). of time-dependent consolidation and creep analysis that also
The salient features of the 1-D analysis (denoted as ‘‘Class incorporates concepts of Modified Cam-Clay and visco-plasticity.
A-01”) are summarised in Table 2. The analysis does not consider Hence, it takes into account of the simultaneous nature of primary
shear deformation associated with yielding of soil elements. Also, consolidation and creep, and can be categorized as Hypothesis B
it is a small strain analysis that does not consider the buoyancy creep model. The coupled analysis of pore pressure and deforma-
of the embankment fill (i.e. reduction of applied fill load) as it set- tion in the foundation soils was conducted using 15 node triangu-
tles below the groundwater table. Furthermore, the 1D consolida- lar elements, each with 12 Gauss points.
tion analysis method used was not a fully coupled solution and As outlined in Table 2, the 2-D FEA (denoted as ‘‘Class A-02”) was
according to Huang and Griffiths [16], this approach can lead to performed with updating mesh and pore water pressure. When
errors when settlements are estimated using the average degree embankment deformations are large, the use of a method that
of consolidation based on pore pressures with multiple soil layers can facilitate large strain analysis helps to achieve a more realistic
being considered. result than small strain analysis (PLAXIS 2D [17]). The mesh update
option incorporating large strains in PLAXIS is based on an approach
4.2. 2-D large strain FEA with soft soil creep model known as the updated Lagrangian formulation [18]), which
accounts for the increase in unit weight of soil element by consid-
More advanced than the 1-D analysis, the 2-D FEA utilised Soft- ering the change in void ratio. Furthermore, the large strain analysis
Soil-Creep (SSC) model that is embedded in the commercially also calculates the applied fill load accurately by considering the

Table 2
Salient features of the different Class A and Class C predictions performed.

Analysis ID Class A-01 Class C-02 Class C-03 Class A-02/Class C-01
Method 1-D consolidation 2D FEA with SS model without 2D FEA with SS model with 2D FEA with SSC model with
mesh update mesh update mesh update
Sophistication indexa 1 2 3 4
Small/large strain Small Small Large Large
Soil model Bi-linear e0-log r0 v defined by Soft soil (SS) model embedded Soft soil (SS) model Soft soil creep (SSC) model
CC and CR in FEA embedded in FEA embedded in FEA
Creep Hypothesis A Hypothesis A Hypothesis A Hypothesis B
Consolidation Direct inputs of cv(NC) and cv calculated implicitly from mv and k. Refer to Section 3.4
cv(OC)
Plastic deformation Not considered Considered Considered Considered
Buoyancy of settled fill below Not considered Not considered Considered Considered
groundwater
a
Sophistication index scales from 1 to 4, with 4 being the most sophisticated analysis in the present assessment.
 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177 169

Table 3
Input parameters for SSC model.

Depth to layer base (m) cbulk (kN/m3) OCR e0 k* j* l* c0 (kPa) /’ kv (m/s) Ck

Class A Class C Class A Class C Class A Class C
0.50 18.00 15.0 15.0 0.90 0.065 0.013 0.0022 15 15 34 3.3E
08 1 0.5
1.00 17.00 8.80 8.8 1.00 0.065 0.013 0.0022 15 15 34 3.3E
08 1 0.5
1.30 16.65 4.80 4.8 1.00 0.065 0.013 0.0022 10 10 34 3.3E
08 1 0.5
2.40 16.30 2.70 2.0 1.75 0.109 0.016 0.0035 2 0 34 8.0E
09 1 0.9
3.30 14.50 2.05 1.8 2.52 0.174 0.023 0.0052 2 0 32 3.1E
09 1 0.9
4.50 14.00 1.80 1.7 2.52 0.174 0.023 0.0052 2 0 32 3.1E
09 1 0.9
5.90 14.00 1.80 1.5 2.70 0.283 0.034 0.0087 2 0 32 8.7E
10 1.6 1.1
7.50 14.00 1.60 1.4 3.08 0.283 0.034 0.0087 2 0 32 8.7E
10 1.6 1.0
9.50 14.00 1.50 1.4 3.18 0.283 0.034 0.0087 2 0 32 8.7E
10 1.6 1.125
10.50 14.50 1.35 1.3 2.68 0.261 0.033 0.0087 2 0 32 8.7E
10 1.6 1.125
12.00 19.00 2.00 1.3 0.78 0.065 0.013 0.0022 5 0 34 1.3E
07 10 1.125
14.20 19.00 2.00 1.15 0.67 0.065 0.013 0.0009 0 0 39 1.3E
07 10 1.125
Lower sand layers represented by Hardening Soil Model(1) embedded in PLAXIS
Depth to layer base (m) cbulk (kN/m3) OCR Eref
ur Eref
50 Eref
oed m c0 (kPa) /’ kv (m/s) Ck
Class A Class C Class A Class C Class A Class C
18.20 19 2.0 2.0 33852 11284 12538 0.5 0 0 39 4.00E
06 1.6 1.6
26.50 19 2.0 2.0 9488 3163 3514 1.0 5 5 38 3.33E
08 1.1 1.1

buoyancy effect as the embankment fill settled below the ground-
1.6 m
RL
7.57 m); Layer 3 between magnets 3 and 4 (RL
water table. The input parameters for the SSC model are given in
7.57 m
RL
10.57 m); and Layer 4 between magnets 4 and 5
Table 3. (RL
10.57 m
RL
14.2 m). The soil stratigraphy in relation to
the different subsoil layers can be related to Fig. 1. The registered
4.3. Prefabricated vertical drains settlement in Layer 2 increased markedly during the construction
of the embankment, which is an indication of shear deformations
Prefabricated vertical drains (PVD) were installed below the associated with the yielding of soft soils at shallow depth. The 1-
trial embankment at 1.2 m spacing in a square pattern, to a depth D analysis was insufficient for prediction as it could not consider
of 15 m. The PVD installation caused remoulding of the soft clay shear deformations. The FE analysis with SSC model is able to cap-
surrounding the drains. Indraratna et al. [19] conducted research ture the shear deformations, although the predicted magnitude
of smear zone characteristics of actual drain installation at the trial could be less than what can be a better fit to the measured settle-
embankment site by studying the extracted soil samples collected ment. The shear deformations of the soil elements within the
around the drains. The study indicated that the smear zone was upper soft layer is evident from the FE results as shown in
about 5–7 times greater than the equivalent dimensions of the Fig. 12. These plastic points are considered to be caused by bearing
mandrel, or about 11 times greater than the drain well radius. In failure of the foundation soil, as opposed to slope instability. Slight
addition, the ratio of in-situ permeability to disturbed permeability increase in pore water pressure can be expected due to this local
in the smear zone was about 1.7–2. In the present Class A predic- shear failure [22]. Fig. 3b shows that the Su was as low as about
tion, the adopted smear radius ratio and permeability ratio were 11 10kPa between RL
1.5 and RL
4 m. The undrained bearing
and 2, respectively. capacity was therefore about 51kPa (=5.14  Su), which was less
The axisymmetric vertical drains installed in the ground have to than the applied embankment fill load of about 60kPa. Separately,
be correctly modelled in the 2D FEA in order to obtain an accurate slope stability analyses using the limit equilibrium method via
consolidation response. This was achieved by using the permeabil- SLOPE/W and the strength reduction method in FEA were under-
ity matching method incorporating smear zone as described in taken. Both methods indicated a global factor of safety of about 1.5.
Indraratna and Redana [20] and Sathananthan et al. [21]. CLASS C PREDICTION - The under-prediction in ‘‘Class A-02”
using the SSC model can be addressed through minor refinements
in the selection of material properties. These included:
5. Comparison of predictions and measurements
(i) Slight reduction of preconsolidation pressure and OCR – As
5.1. Settlements
shown in Fig. 2c, the adopted OCR in the Class C prediction
were biased towards conservative values by selecting a pro-
CLASS A PREDICTION - Fig.11a compares Class A predictions of
file close to the lower bound of the test points, whereas a
the total foundation settlement with measurements at settlement
median line was adopted in the Class A analysis. As noted
plate SP2 (along the centreline of the embankment). While both
in Section 3.1, r0p is dependent on the strain rate. The appli-
the 1-D (Class A-01) and the SSC (Class A-02) models underesti-
mated the overall settlement, the SSC model gave a closer predic- cation of the 0.84 factor to the r0p is a method to normalise
tion to the measurements than that of the 1-D model, in terms of the effect of strain rate adopted in the CRS tests to that of
the magnitude and the shape of the predicted settlement vs. time a slower strain rate generally used in the oedometer tests.
curve. The 1-D model grossly underestimated the construction set- Hence, the corrected r0p values from the CRS tests became
tlement during fill placement. consistent with the r0p obtained from the oedometer tests
The extensometer results could provide insight into the cause of as shown on Fig. 2a. However, actual settlement data
the under-estimation of settlement. Fig. 11b shows the soil com- demonstrated that the field strain rate could in fact be
pressions of four sub-layers measured by the extensometer slower than that of laboratory testing. For example, the mag-
MEX1, namely, Layer 1 between existing ground surface and mag- netic extensometer data from MEX1 indicated that the field
net 1 (RL 0.4 m
RL
1.6 m); Layer 2 between magnets 1 and 3 (RL strain rate of soil at depths 2 m to 8 m was between
170 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177

(ii) Deletion of the apparent cohesion of the soil effective strength –

Davis in Thorne [23] proposed a relationship to link the
effective strength parameters and undrained parameters:

c0 cos/0 þ r0v sin /0 ð1 þ 2K 0 Þ=3

Su ¼ ð3Þ
1 þ ð2Af
1Þ sin /0
where Af is the Skempton pore pressure parameter ‘A’ at failure, c0 is
the effective cohesion, /0 is the effective friction angle, and K0 is the
earth pressure coefficient at rest. With the known Su profile outlined
in Fig. 2b and an averaged friction angle of 32° obtained from the
triaxial tests given in Pineda et al. [1], the effective cohesion c0
can be estimated using Eq. (3) with appropriate K0 and Af values.
K0 was assessed based on Equation (4) in conjunction with /0
= 32° and OCR values inferred from CRS test results as outlined in
Fig. 2c. The calculated K0 values were in the range of 0.45 to 0.65,
which were consistent with the results of a push-in pressure cell
PIPC12 installed outside of the trial embankment footprint and out-
lined in Kelly et al. [7].
pffiffiffiffiffiffiffiffiffi
K 0 ¼ ð1
sin/0 Þ OCR ð4Þ
Af depends on the stress history of the soil (i.e. OCR). Despite data
scatter over a narrow OCR range of 1.3 to 2, Fig. 13 shows that Af
values for Ballina clay obtained from TC tests lied in the range of
variation reported in Bishop and Henkel [24] for Weald clay and
London clay. Using the trend of Af
OCR for, say, London clay
and the other aforementioned input parameters, the calculated c0
for the very soft estuarine silt/clay (depths 1.3 m to 14.2 m) is about
0 kPa. However, the adopted c0 in the Class A prediction was 2 kPa.
Fig. 14 presents the distribution of maximum shear stress, smax,
with depth obtained from PLAXIS analysis at the initial stage prior
to embankment construction, and compares with the in-situ Su pro-
file inferred from CPTu and corrected vane shear. Note that smax can
be considered as the simulated undrained shear strength in PLAXIS
analysis since it is the maximum value of shear stress where the
Mohr’s circle is expanded to touch the Coulomb failure envelope
while keeping the centre of Mohr’s circle constant. It can be seen
that smax matched well with the Su profile when the effective
strength parameters of c0 = 0 kPa and /0 = 32° were used in the
SSC model, but exceeded the Su profile when values of c0 = 2 kPa
Fig. 11. Comparison of Class A predictions and measurements for (a) surface
and /0 = 32° were used in the Class A-02 prediction.
settlement at settlement plate SP2 and (b) layer compressions at extensometer
MEX1.
(iii) Increase in the compressibility of the sand-clay mixtures at
depths 10.5 m to 14.25 m below ground surface – In Class A
Shear Plastic Yielding PVD prediction, the adopted compressibility parameters for the
lower soil layers (depths 10.5 – 14.25 m) were based on

1
0.9
0.8 Ballina Clay
0.7
0.6 Weald Clay
0.5 London Clay
0.4
0.3
Af

0.2
0.1
0
Fig. 12. Plastic yielding of soil elements indicated in 2D FEA model. -0.1
-0.2
-0.3
-0.4
1.3  10
8 s
1 and 1.7  10
9 s
1, which was more than an -0.5
order of magnitude slower than a typical strain rate of -0.6
1  10
7 s
1 used in the oedometer tests. The slower strain 1 10
Over-consolidaon Rao
rate in the field therefore justified the use of a lower bound
OCR profile in the Class C prediction, in preference to the Fig. 13. Effect of over-consolidation on Skempton pore pressure parameter ‘A’ at
median OCR profile adopted in the Class A prediction. failure.
 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177 171

Su (kPa)

10
15
20
25
30
35
40
45
50
0
5
1
0
-1
-2 τma x from Class A with
-3 c' = 2kPa and φ
-4
-5
-6
RL (mAHD)

-7
-8
-9
-10
-11
-12
-13 Fig. 15. Class C predictions and measurements for surface settlements.
τma x from Class C with
-14 c' = 0kPa and φ
-15
CPT-8 (Nkt =15)
-16
Vane shear VS08 (with Bjerrum correction)
-17
Fig. 14. Comparison between in-situ Su profiles with the distribution of maximum
shear stress, smax, with depth obtained from PLAXIS analyses at the initial stage
prior to embankment construction.

CRS testing on the clayey sand sample extracted from bore-

hole INCLO2 located at the embankment toe. While there is
no laboratory test data available at the same soil depths
from borehole MEX1 at the central of the embankment,
the sand-clay mixtures underneath the trial embankment
could be more compressible than those materials at the
embankment toe, based on the tactile description as being
sandy clay in MEX1, as opposed to clayey sand in INCLO2.
When the actual extensometer data at the embankment central
is available, it is evident that the soil layers between 11.0 m and
14.5 m have gained considerably more compression (about
67 mm) than predicted (37 mm) in Class A analysis. This is consis-
tent with our expectation as well as the tactile description in
MEX1. Accordingly, the compression indices of these layers were
increased in the Class C back-analysis to match the measurements.

(iv) Reduction of ck values – This is discussed in Section 5.2.

The revised model parameters are summarised in Table 3. The

revised analysis results, denoted as ‘‘Class C-01” and presented in
Figs. 15 and 16, agree well with the measured surface settlement
and layer compressions, respectively.
Owning to the limitation of the 1D analysis, an additional 2D
FEA analysis was performed (denoted as ‘‘Class C -02” analysis in
Table 2) in which a simple elasto-plastic, Soft Soil (SS) model, avail-
able in PLAXIS 2D, was employed. The SS model resembles the
Modified Cam-clay model for the assessment of the changes in Fig. 16. Class C predictions for soil compressions of (a) Layers 2 and 3; (b) Layers 1
elastic and plastic volumetric strains. A Mohr-Coulomb hexagon and 4.

yield surface in the deviatoric plane is included in the model to

account for the plastic shear strain when the yield stress is placement (see Section 5.2). The parameter inputs listed in Table 3,
exceeded. Creep is not considered during primary consolidation except for creep strain rate, l⁄, were used for ‘‘Class C-02” analysis
(Hypothesis A), but may be added manually once the DoC of the with SS model. This analysis was based on small strain without
soil elements reached a specific threshold such as 90%. For the cur- mesh update.
rent Class C prediction, however, creep was not anticipated to be The results of ‘‘Class C-02” analysis for surface displacement
significant using Hypothesis A as the measured DoC based on pore and layer compressions are in surprisingly good agreement with
pressure remained at about 70% after 3 years following the fill the measurements, as shown in Figs. 15 and 16. It should be noted
172 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177

that the current small strain analysis adopted a constant load with during the primary consolidation, leading to relatively reasonable
time after fill placement. However, the trial embankment has set- estimate of the settlement compared to the measurements. In
tled a maximum of 1.5m over the past 3 years. Part of the fill mate- other words, while the simple small strain ‘‘Class C-02” analysis
rial that was originally above the GWT has settled below the GWT. (with SS model) gave similar predicted settlements compared to
The effective unit weight of the soil mass that was above the orig- the more sophisticated large strain ‘‘Class C-01” analysis (with
inal GWT reduced due to buoyancy forces. This entailed a load SSC model), its theoretical considerations were far less rigorous
reduction of about 11 kPa  (1.5–0.4 m)  10 kN/m3. This is signif- than the latter. It is therefore important to be aware of the limita-
icant compared to the applied load of about 60 kPa, i.e. about 18% tions when a simple analysis with simple soil model, such as that
load reduction. If this effect had been taken into consideration, in the ‘‘Class C-01” analysis, is used.
then the model would have underestimated the measured settle-
ment. The buoyancy effect on embankment settlement in soft clay 5.2. Excess pore pressure
has been pointed out by Chang [25] and Degago et al. [26].
To investigate the buoyancy effect, a large strain 2D FEA with SS CLASS A PREDICTION – In the interpretation of excess pore
model was conducted, denoted as ‘‘Class C-03”. The sophistication water pressure, it is important to correct the measured total pore
level of ‘‘Class C-03” analysis is ranked between ‘‘Class C-02” and water pressure for the increase in pressure head due to the settle-
‘‘Class C-01” analyses as outlined in Table 2. As can be seen in ment of piezometer under load. The settlement of the piezometer
Figs. 15 and 16, the large strain SS model underestimated the total with time can be estimated from the registered settlement of the
settlement and soil layer compression responses, when the buoy- magnet in the nearby extensometer. Fig. 17a shows an erroneous
ancy effect was taken into account and an Updated Lagrangian interpretation of the excess pore water pressures at VWP6 (at
UL formulation was used. When compared with ‘‘Class C-01” the centre of the embankment), in which the settlements of indi-
results (large strain SSC model), it can be seen from Fig. 15 that vidual piezometers were not taken into consideration. It can be
up to about 300 mm of creep settlement had been accumulated seen that the interpreted excess pore pressure dissipations were
during primary consolidation using the SSC model (Hypothesis much slower than those correctly interpreted in Fig. 17b with
B), which was disregarded using the SS model (Hypothesis A). Sub- the consideration of settlements.
sequently, it can be recognised that the exclusion of the buoyancy There was no standpipe piezometer installed at the site and the
effect and Updated Lagrangian UL formulation in the ‘‘Class C-01” variation of static water level is inferred to be represented by the
analysis (with greater primary settlement due to the greater changes in pore water pressure measured by VWP6. As shown in
applied fill load) compensated the effect of disregarding creep Fig. 18a, the VWP at RL
1.6 m within the upper sandy layer

Time (Days) Time (Days)

(a) (b)
10 100 1000 10 100 1000
70 70
VWP6 RL-1.6m VWP6 RL-1.6m
Excess pore water pressure (kPa)

Excess pore water pressure (kPa)

Applied fill load Applied fill load

60 VWP6 RL-5.6m 60 VWP6 RL-5.6m

VWP6 RL-9.6m VWP6 RL-9.6m

50 Applied Load
50 Applied Load

40 VWP6 40
RL-5.6m
VWP6
30 30 RL-5.6m

20 VWP6 20
VWP6
RL-9.6m
RL-9.6m
10 10
VWP6 VWP6
RL-1.6m RL-1.6m
0 0

Fig. 17. Excess pore pressures (a) prior to and (b) after the correction of total pore pressures for the increase in static pore pressure due to the settlements of piezometers.

Time (Days) Time (Days)

(a) (b)
10 100 1000 10 100 1000
70 70
VWP6 RL-1.6m VWP6 RL-1.6m
Applied fill load Applied fill load
Excess pore water pressure (kPa)

Excess pore water pressure (kPa)

60 VWP6 RL-5.6m 60 VWP6 RL-5.6m

VWP6 RL-9.6m VWP6 RL-9.6m

50 50
Applied Load Applied Load

40 40
VWP6
30 RL-5.6m 30
VWP6
RL-5.6m
20 Risein hydrostatic 20
VWP6
pore pressure
RL-9.6m VWP6
10 10 RL-1.6m
VWP6
VWP6 RL-
RL-1.6m 9.6m
0 0

Fig. 18. Excess pore pressures (a) prior to and (b) after the correction of static pore water pressure.
 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177 173

indicated a rise in the hydrostatic pore water pressure after the Time (Days)
(a) 10 100 1000
completion of excess pore pressure dissipation at about Day 230.
70
This rise in pore pressure is attributed to (i) seasonal changes in VWP6 RL-1.6m

Excess pore water pressure (kPa)

groundwater level and (ii) discharged water from the PVD during VWP6 RL-5.6m Applied fill load
60
consolidation. Further discussions on the groundwater fluctuation VWP6 RL-9.6m
and the hydraulic gradient across the embankment are given in Class A-01 1D
50
Section 6. Fig.18b shows the revised excess pore pressure plot that RL-1.6m
Class A-01 1D
has been corrected for the changes in hydrostatic pore water pres- RL-5.6m
40 Class A-01 1D
sure after Day 230. The corrected excess pore pressures at the end RL-9.6m
of monitoring were 18 kPa and 4kPa at RL
5.6 m and RL
9.6 m Applied Load
30
respectively. The corresponding DoC were about 70% and 93% at
VWP6
these two levels. RL-5.6m
20
Fig. 19 shows the results of Class A predictions for the excess
pore pressure. Both the 1D and SSC models predicted faster pore
10
pressure dissipation rates than the measurements, but the SSC VWP6
VWP6 RL-9.6m
model generally gave a higher predicted values than the 1-D model RL-1.6m
because of the creep and yielding of soil elements that were not 0
captured in the 1D model. Time (Days)
CLASS C PREDICTION - There are many parameters that can (b)70 10 100 1000
affect the predicted pore pressure responses, including the adopted VWP6 RL-1.6m

Excess pore water pressure (kPa)

soil permeability, soil stiffness, as well as the smear radius ratio VWP6 RL-5.6m Applied fill load
60
and permeability ratio of the wick drains. One influential parame- VWP6 RL-9.6m
ter, however, that affects most of the post peak excess pore pres-
50 Class A-02 SSC
sure dissipation rate was identified to be the permeability index, RL-1.6m
Class A-02 SSC
ck. The smaller the ck, the slower is the rate of excess pore pressure RL-5.6m
dissipation, and the flatter is the slope of the excess pore pressure 40 Class A-02 SSC
RL-9.6m
vs. log time curve after fill placement. The adopted ck in ‘‘Class A- Applied Load

02” analysis was based on the empirical relation ck = 0.5e0 [13]. 30

This gave a ck value of about 1.6 for the soft soil layer between VWP6
RL-5.6m
RL
5 m and RL
9 m, where e0 was about 3.2 (Fig. 9a). The CRS 20
data presented in Pineda et al. [1] (Fig. 20), however, indicated a VWP6
lower measured ck values, in the range of 0.8–1.2, or 0.25e0– 10 RL-1.6m
VWP6
0.375e0. The use of these measured ck values in the prediction, as RL-9.6m

opposed to Tavenas’s empirical relation, gave rise to a better fit 0

to the measured excess pore pressure. Fig. 21 shows the predicted
Fig. 19. (a) Comparison of Class A-01 (1D) and (b) Class A-02 (SSC) analyses with
excess pore pressures from ‘‘Class C-01” analysis, which was essen- excess pore pressure measurements.
tially the same as ‘‘Class A-02” analysis but using the revised model
parameters listed in Table 3 including the revised inputs of ck. In
general, the revised analysis gave a closer match in excess pore
pressure response than that of the Class A analysis. Also shown
in Fig. 21 are the predicted excess pore pressures from ‘‘Class C-
02” analysis using the SS model (Hypothesis A) and without mesh ck = 1.2
update. For the same reasons outlined in Section 5.1 for settle-
ments, these predicted excess pore pressures are in surprisingly
good agreements with those of ‘‘Class C-01” analysis. ck = 0.8

5.3. Horizontal displacements

The comparison of the predicted and the measured horizontal

displacements at INCLO2 are presented in Fig. 22a and b for the
end of fill embankment construction and after 3 years from the
start of filling, respectively. The predicted lateral displacements
using the empirical approach outlined in Section 4.1 relating lat-
eral displacements to the corresponding settlements measured at
settlement plate SP02 at the centre of the embankment are also
shown in the figures.
Fig. 20. Variation of k during CRS loading [1].
The FE results (i.e. ‘‘Class C-01” and ‘‘Class C-02”) over-predicted
the short term lateral displacements upon the completion of fill
embankment (Fig. 22a). This is probably due to the overestimation
of undrained lateral displacements in numerical modelling at the tion of ‘‘Class C-02” analysis was also comparable with the
beginning of primary consolidation. The empirical approach gave measurements, but the predicted peak horizontal displacements
a more comparable result with the measurement, although it occurred at a position higher than that of the measurements. The
over-predicted the lateral displacements near the ground surface. predicted long term lateral displacements using the empirical
For the long term lateral displacements, there was remarkable approach were less accurate than those using the FEA approaches.
agreement between ‘‘Class C-01” prediction with the measure- They were over-predicted by about 20% and the position of the pre-
ments, in terms of the shape and magnitude (Fig. 22b). The predic- dicted peak did not correspond with that of the measurements.
174 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177

Time (Days) in groundwater level, and its effect on the prediction analysis
10 100 1000 results.
70

Applied fill load 6.1. Natural groundwater levels and discharged water from PVD
Excess pore water pressure (kPa)

60
Class C-01 (SSC Large Strain)
Class C-02 (SS Small Strain) In addition to VWP6 beneath the trial embankment, the fluc-
50 Applied Load
tuation of groundwater level is also inferred from the pore pres-
sure readings taken by the push-in pressure cell PIPC12
40
installed outside of the trial embankment footprint (see Kelly
et al. [7]). The pore pressure filter at this PIPC lies at 4.35 m
30 VWP6 depth.
RL-5.6m
As discussed in Section 5.2, the variation in static water level
20
was inferred from the pore pressure change registered at the top
VWP6 piezometer of VWP6 after Day 230 (i.e. after the completion of
10 RL-1.6m
VWP6
excess pore pressure dissipation). This inferred static water level
RL-9.6m was then compared with the fluctuation of groundwater level
0 inferred from PIPC12, as shown in Fig. 23. The following points
Fig. 21. Class C predictions for excess pore pressure responses.
can be drawn from the plot:

 The variation of the water level inferred from VWP at RL
1.6 m

6. Groundwater fluctuation underneath the embankment bore a general resemblance to
that of PIPC12 outside of the embankment footprint. This indi-
The analyses undertaken thus far have considered a constant cates that both sensors were subject to the same seasonal
groundwater level with time. This section discusses the fluctuation changes in groundwater level.

Lateral Displacement (mm) Lateral Displacement (mm)

-50 50 150 250 -50 50 150 250
0 0
(a) (b)

-5 -5

-10 -10
RL (mAHD)

RL (mAHD)

-15 -15

Measured (04/10/2013) Measured (27/07/2016)

Class C-01 (SSC-large Class C-01 (SSC-large

-20 -20 strain)
strain)
Class C-02 (SC- small Class C-02 (SC- small
strain) strain)
Empirical approach Empirical approach

-25 -25

Fig. 22. Lateral displacement profiles for (a) end of fill placement, (b) long term after 3 years from the start of filling.

Time (Days)
1.2
0.9
0.6
0.3
RL (mAHD)

0.0
-0.3
-0.6
Adopted Ground water level
-0.9
Ground Water level based on PIPC 12
-1.2 Ground Water level based on VWP 6a
-1.5
10 100 1000

Fig. 23. Static water levels inferred from PIPC12 and VWP6.
 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177 175

 The natural groundwater table could be inferred from PIPC12,

which is away from the influence of the preloading area. This
inferred variable groundwater level was used in the additional
analysis outlined in Section 6.2. The elevated water level
inferred from the top piezometer at VWP6 was affected by the
dissipation of the excess pore pressure and the hydraulic gradi-
ent required to discharge the groundwater collected from the
PVD. Therefore, the top piezometer was measuring a water level
slightly higher (about 0.6 m) than the natural groundwater
level.
 To interpret correctly the excess pore pressure from VWP6, it is
important to correct the total pore pressure readings for:
(i) The increase in static pore water pressure due to the settle-
ment of piezometer under load as outlined in Fig. 17 in
Section 5.2.
(ii) the changes in static pore water pressure due to fluctuation
of the natural groundwater level and the discharged water
from the PVD during consolidation.

The correction for the latter was carried out for data after Day
230, i.e., by subtracting all piezometer readings in VWP6 from
the pore pressure rises recorded in the top piezometer (assumed
to be fully consolidated at this level) after Day 230 (see Fig. 18 in
Section 5.2).

6.2. Additional Class C predictions using variable groundwater levels

with time

Fig. 24. Class C-01X prediction for surface settlements.

A further Class C analysis, denoted as ‘‘Class C-01X” was under-
taken. This analysis was an extension to the previous ‘‘Class C-01”
analysis, in which a large strain 2-D FEA using elastic viscoplastic
SSC model (Hypothesis B) was employed. The only difference is
that now a variable groundwater table with time was introduced
in the analysis. For analysis purposes, an idealised groundwater-
time function was adopted (see Fig. 23), which was inferred from
the pore pressure readings taken by PIPC12 outside of the embank-
ment footprint.
Surface settlement Fig. 24 shows the comparison between mea-
sured surface settlements at SP2 and the predictions of ‘‘Class C-
01X” analysis. This figure indicates that with the consideration of
groundwater fluctuation, further improvement in prediction accu-
racy can be achieved. Some of the localised irregularities or anoma-
lies manifested on the measured settlement curve can now be
captured by the new prediction. For example, the momentary
increase in settlement rate between Day 230 and Day 330 as
shown in Fig. 24 can now be explained by the lowering of the
groundwater level (from about RL 0 m to RL
0.8 m) during the
time period ahead of the settlement rate rise (i.e. between Day
190 and Day 250). The lowering of the groundwater caused an
increase in the mean effective stress in the soil, thus leading to
increase in soil settlements. The time-lagged response of the soil
settlement activation to the lowering of the groundwater level is
dependent on the soil permeability and the dissipation rate of
excess pore pressure (refer to discussion on excess pore pressure
below). Conversely, the reduction of settlement rate from Day
330 to Day 400 was due to the rise in groundwater level (from
about RL
0.8 m to RL +0.2 m) that took place at around Day 250.
Soil layer compressions from extensometer Fig. 25 shows the com-
parison between new predictions and the soil layer compressions Fig. 25. Class C-01X predictions for soil layer compressions.
measured by extensometer MEX1. Similar to the surface settle-
ment prediction, the analysis results compare well with the field
data. Some of the localised features of the soil layer compression by piezometer VWP6. Significant improvement to the excess pore
responses can now be captured by changing the water level in pressure prediction has been made since the adoption of variable
the analysis. groundwater level with time. Taking the piezometer at RL
Excess pore water pressures Fig. 26 shows the comparison
9.6 m as an example, the lowering of groundwater level during
between new predictions and the excess pore pressure measured Day 190 to Day 250 was accompanied by some minor reloading
176 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177

Time (Days) piezometer VWP6 was possibly installed slightly off centre, thus
10 100 1000 registering slightly lower excess pore water pressure when the
70
Field Data degree of consolidation was relatively low.
Excess pore water pressure (kPa)

Best Prediction Horizontal displacements The adoption of a variable groundwa-

60
(Class C-01X) ter level in the ‘‘Class C-01X” analysis has minimal effect on the
Applied Load
prediction of horizontal soil displacements at INCLO2 location, as
50
compared to that of ‘‘Class C-01” analysis using constant water
40
level. In general, the analysis result over-predicted the short term
lateral displacement upon the completion of fill embankment,
30 but agreed remarkably well for the long term horizontal displace-
ment after 3 years from the start of filling (see Fig. 27).
VWP6
20
VWP6
10
RL-1.6m 7. Conclusions
VWP6
RL-9.6m
0 Geotechnical engineering practitioners often use simplistic soil
1.2 models for design, possibly because of time and cost constraints.
0.8 Adopted Ground water level Simple models can act as a first step approach towards a better
RL (mAHD)

Ground Water level based on PIPC 12

0.4 understanding of the complex reality, but these basic models can
0.0 over-simplify the ‘‘real” situation. It is important to understand
-0.4
the limitations that these simplified approaches entail before
-0.8
-1.2
adopting simplistic models for design. For example, this paper
-1.6 has demonstrated that the 1D analysis under-estimated the
10 100 1000 embankment settlement because it could not capture shear defor-
Time (Days)
mation of the soil elements at shallow depth.
Fig. 26. Class C-01X predictions for excess pore pressures. In choosing a soil model, geotechnical engineers must decide
which factors are likely to be sufficiently important to be included
in the model. The more factors there are to be built into the model,
of the foundation soil due to the increase in mean effective stress of the more complex that model becomes. It is in the notion that the
the soil. This reloading retarded the excess pore pressure dissipa- degree of analytical sophistication should reflect the quantity and
tion during this period. When the groundwater level rose during quality of geotechnical data available, and should correspond to
Day 250 to Day 400, the foundation soil experienced some equiva- the complexity of the soil model adopted. Simplified soil models
lent unloading due to a decrease in mean effective stress of the soil. can be applied more efficiently and are more consistent with sim-
This accelerated the excess pore pressure dissipation rate during ple numerical methods, whereas complex models should be
this period. The pattern of cyclic loading of soil due to groundwater accompanied with more sophisticated computational approaches.
fluctuation was repeated after Day 400. Mixing complex soil models with simple analytical methods, or
Fig. 26 also shows over-prediction of the excess pore water vice versa, may not give desirable outcomes. For example, the com-
pressures during the early stages of dissipation after embankment plex elastic viscoplastic SSC model will overestimate the founda-
construction (up to about Day 350, especially for the piezometer at tion settlement of the present trial embankment if it was
RL
5.6 m). This is likely due to the fact that the analysis predicted employed in a small strain consolidation analysis. Conversely, the
the maximum excess pore water pressures in-between PVDs, while simpler elastoplastic SS model will underestimate the settlement

Lateral Displacement (mm) Lateral Displacement (mm)

-50 50 150 250 -50 50 150 250
0 0
(a) (b)

-5 -5
RL (mAHD)

RL (mAHD)

-10 -10

-15 -15

-20 -20

Measured (04/10/2013) Measured (27/07/2016)

Best prediction (Class C-01X)

Best Prediction (Class C-01X)
-25 -25

Fig. 27. Class C-01X predictions on lateral displacements for (a) end of fill placement, (b) long term after 3 years from the start of filling.
 K.F. Chan et al. / Computers and Geotechnics 93 (2018) 163–177 177

if large strain (updating mesh and pore pressures) numerical the groundwater table are known or for back-analysis purposes
method was applied to the consolidation modelling of soft soil such as the subject Ballina embankment prediction exercise.
without creep.
The number of model parameters increases with the complexity References
of soil model. The more parameters that are required, the more
sophisticated and complex laboratory testing becomes in order to [1] Pineda JA, Suwal LP, Kelly RB, Bates L, Sloan SW. Characterisation of Ballina
clay. Géotechnique 2016;66:556–77.
determine their values. There is therefore a balance between the [2] Watabe Y, Udaka K, Nakatani Y, Leroueil S. Long-term consolidation behaviour
costs of testing large numbers of samples to evaluate advanced soil interpreted with isotache concept for worldwide clays. Soils Found 2012;52
parameters for a heterogeneous and layered soil profile and the (3):449–64.
[3] Jamiolkowski M, Ladd CC, Germaine JT, Lancellotta R. New developments in
possible benefit of using a complex rather than a simple model. field and laboratory testing of soils. In: Proc 11th int conf. on soil mechanics
The design risks involved in a project, the type of loading, the and foundation eng, San Francisco, 1; 1985. p. 57–154.
impacts on the surrounding structures, and the quality of site [4] Ladd CC. Stability evaluation during staged construction, the 22nd terzaghi
lecture. ASCE J Geotech Eng 1991;117(4):540–615.
investigation or laboratory data available all may or may not justify [5] Bjerrum, L., 1972. Embankment on soft ground. In: Proc. Speciality conference
the complexity of the model. on Performance on earth and earth retaining structures, Purdue, vol. 2. New
In order to provide a good prediction of consolidation responses York: ASCE. p. 1–54.
[6] Bjerrum L, Problems of soil mechanics and construction on soft clays and
of an embankment built over a soft soil deposit, collection of high
structurally unstable soils. In: Proc. VIIth international conference on soil
quality samples with minimal disturbance, together with soil test- mechanics and foundation engineering, Moscow, vol 3; 1973. p. 111–59 .
ing conducted with care and analysis using a soil model which is [7] Kelly RB, Pineda JA, Bates L, Suwal LP, Fitzallen A. Site characterisation for the
Ballina field testing facility. Géotechnique 2017;67(4):279–300.
capable of simulating realistic soil behaviour is extremely impor-
[8] Mesri G. Discussion on ‘‘New design procedure for stability of soft clays” by
tant. For the Ballina prediction exercise, high quality samples were Ladd and Foott. ASCE J Geotech Eng Div 1975;101(GT4):409–12.
extracted below the proposed embankment. In addition, the [9] Muir Wood D. Soil behaviour and critical state soil mechanics. Cambridge
amount of in-situ and laboratory test data available was typically University Press; 1990.
[10] Larsson R. Undrained shear strength in stability calculation of embankments
more than what practising geotechnical engineers would usually and foundations on soft clays. Can Geotech J 1980;17(4):591–602.
have from their typical investigation and design projects. Adopting [11] Mesri G, Lo DOK, Feng TW. Settlements of embankments on soft clays. In:
a Hypothesis B type of model (SSC) and considering the permeabil- Proceedings of settlement ’94: vertical and horizontal deformations of
foundations and embankments, College Station, TX 1; 1994. p. 8–56.
ity variation with void ratio change, groundwater fluctuations and [12] Navfac, 1971 Soil mechanics, foundations and earth structures. Design Manual
the strain rate effects to the pre-consolidation pressure, the DM-7, Naval Facilities Engineering Command.
authors were able to accurately predict the consolidation [13] Tavenas F, Leblond P, Jean P, Leroueil S. The permeability of natural soft clays,
Part 2: Permeability characteristics. Can Geotech J 1983;20:645–60.
responses in their Class C prediction by simply applying minor [14] Tavenas F, Leroueil S. The behaviour of embankments on clay foundations. Can
adjustments to the originally adopted Class A prediction Geotech J 1980;17(2):236–60.
parameters. [15] Tavenas F, Mieussens C, Bourges F. Lateral displacements in clay foundations
under embankments. Can Geotech J 1979;16(3):532–50.
Whilst numerical analyses using constant groundwater levels
[16] Huang J, Griffiths DV. One-dimensional consolidation theories for layered soil
are able to predict the soil settlement/excess pore pressure with and coupled and uncoupled solutions by the finite-element method.
time well, the adoption of variable groundwater levels can further Géotechnique 2010;60(9):709–13.
[17] Plaxis BV. 2D reference manual; 2016.
improve the accuracy of predictions on the localised features of the
[18] Bathe KJ. Finite element analysis in engineering analysis. New Jersey: Prentice-
measured settlement/excess pore pressure vs. time curves. These Hall; 1982.
localised features were associated with the transient nature of [19] Indraratna B, Perera D, Rujikiatkamjorn C, Kelly R. Soil disturbance analysis
groundwater fluctuations. due to vertical drain installation. Proceedings of the Institution of Civil
Engineers. Geotech Eng 2015;168(3):236–46.
By way of an example, the lowering of groundwater led to a [20] Indraratna B, Redana IW. Plane strain modelling of smear effects associated
momentary ‘‘dive” in the settlement-time curve and the apparent with vertical drains. ASCE J Geotech Eng 1997;123(5):474–8.
retardation of excess pore pressure dissipation due to reloading. [21] Sathananthan I, Indraratna B, Rujikiatkamjorn C. Evaluation of smear zone
extent surrounding mandrel driven vertical drains using the cavity expansion
However, these were merely the manifestation of short term tran- theory. Int J Geomech 2008;8(6):355–65.
sient effects. As the groundwater returned to the higher normal [22] Sassa K, Fokuda H, Wang F, Wang G. Progress in landslide science. Springer;
level, the settlement rate reduced and the dissipation rate of excess 2007.
[23] Thorne CP. Strength assessment and stability analysis for fissured clays.
pore water pressure increased. Therefore, the overall predictive Géotechnique 1984;34(3).
trends based on variable groundwater level did not deviate from [24] Bishop AW, Henkel DJ. The measurement of soil properties in the triaxial test.
those using constant water level for the current prediction of Bal- Edward Arnold; 1962.
[25] Chang YCE. Discussion: settlement analysis of embankments on soft clays.
lina trial embankment.
ASCE J Geotech Eng 1987;113(9):1063–7.
For general design and analysis purposes, the use of constant [26] Degago SA, Nordal S, Grimstad G, Jostad HP. Analysis of Vasby test fill
groundwater level could be assumed. Variable groundwater level according to creep hypothesis A and B. In: Proceedings 13th international
conference of the international association for computer methods and
may be adopted only if the fluctuation and recharge patterns of
advances in geomechanics, Melbourne, 1; 2011. p. 307-312.