Professional Documents
Culture Documents
Jemal Jibril Muhammed-Settlement of A Railway Embankment On PVD-improved Karakore Soft Alluvial Soil
Jemal Jibril Muhammed-Settlement of A Railway Embankment On PVD-improved Karakore Soft Alluvial Soil
a r t i c l e i n f o a b s t r a c t
Article history: This paper presents a case history of the settlement performance of a railway embankment built on a
Received 3 September 2019 Prefabricated Vertical Drain (PVD) improved soft alluvial soil of Karakore area, Wollo province, north-
Revised 23 January 2020 east of Ethiopia. The embankment was constructed on Awash - Kombolcha - Haragebaya (AKH) railway
Accepted 11 March 2020
project for the purpose of preloading so as to facilitate the consolidation settlement in a short period of
Available online 24 March 2020
time; and it was monitored for more than 750 days. Numerical predictions of Class A and Class C (based
on back-calculation procedure) were performed using Plaxis 2D. The soft soil creep (SSC) model was
Keywords:
employed for the soft soil layers, while the Mohr-Coulomb (MC) model was used for the drainage and fill
Karakore soft alluvial soil
PVD-improved
layers. The predictions were compared with the field monitoring settlement data. Quite a reasonable
Embankment agreement was achieved from the finite element modelling of Class C prediction when the prediction
Settlement was compared with the actual field settlement values. Parametric sensitivity studies were carried out
Numerical modelling to examine the influence of parameters on the rate and magnitude of settlements. The results of the
Class A prediction numerical predictions and parametric studies have been discussed in detail. The soil on the flood plain
Class C prediction area of Karakore is found to be very soft to soft alluvial deposit and formed over Tertiary Eocene stiff clay.
Consolidation parameters The variable conditions of the soil formation and deposition highly dictated the deformation and excess
pore water pressure behaviour of the soft subsoil sediment. The parametric sensitivity study showed that,
following the compression index, the initial void ratio, the hydraulic conductivity, and the OCR highly
influenced the settlement responses of PVD-improved soft alluvium soil respectively.
Ó 2020 Karabuk University. Publishing services by Elsevier B.V. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
https://doi.org/10.1016/j.jestch.2020.03.004
2215-0986/Ó 2020 Karabuk University. Publishing services by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1016 J.J. Muhammed et al. / Engineering Science and Technology, an International Journal 23 (2020) 1015–1027
consolidation of the soft subsoil accompanied by inclusion of PVDs the effect of vertical and horizontal permeability of the natural
would eventually increase the shear strength before placing the subsoil and the drainage capacity of the PVDs [5,14,16–18]. The
subsequent railway structures. The drainage capacity of PVDs is fourth method has been found to be more convenient and simple
generally affected by several factors such as time, deformation, fil- to apply in an actual engineering practice.
tration and clogging of the wick drain [4]. The vertical hydraulic
conductivity of a soft soil can be increased remarkable by inclusion 2.2. Methods in deformation prediction and constitutive soil models
of PVDs with up to one order of magnitude [5]. For the reasons of
low cost and short installation time, like stone columns [3], PVDs Several methods have been proposed for predicting the magni-
are usually installed before the embankment is constructed on tude and rate of settlement of structures supported on soft soils.
the soft subsoil. Lambe (1973) proposed that the predictions in soil engineering
The behaviour of PVD-improved soft clay under an embank- can be classified into different types. Among them Class A and
ment surcharge loading can be modelled by finite element (FE) Class C predictions are widely used for estimating settlement in
analysis to predict the anticipated consolidation settlement. Settle- geotechnical structures such as embankments and piles. Class A
ments of the fills can be monitored at selected locations on the prediction is carried out using available laboratory and field soil
field. The field monitoring data may be used to compare and adjust investigation data before the construction activity is begun. In
the magnitude and rate of the settlements estimated by the Class C prediction, the estimation of values is performed after the
numerical modelling. The comparison of the numerically- construction activity had started and when a complete set of field
predicted and field-measured deformation of the embankment data is available. Class C prediction can also be used to back-
constructed on soft soil mainly contributes to a mechanism that calculate the properties of soils by matching of curves from the
can be used to examine the relevance of the soil constitutive model numerical and field data analyses [19,20]. A Class A prediction is
employed in modelling activities. Many of the study efforts made performed without having measured field data, while a Class C pre-
to model soft grounds which are improved with PVDs to predict diction is performed after data has been obtained from field mea-
settlements are done on coastal area deposits such as marine soft surements. Both Class A and Class C methods have relative
clay [4,6–8]. However, the soft soils in the current study area are advantages to predict values in different design situations.
different in their mode of formation, characteristic, and Mohr-Coulomb (MC) and Modified Cam-Clay (MCC) are the
classification. most popular elastoplastic soil models. They are widely used to
Hence, this study aims at evaluating the field settlement perfor- model the behaviour of soils because of their simplicity and having
mance of the AKH railway embankment built on PVD-improved the capability to describe strain softening, yielding conditions, and
Karakore soft alluvial soil deposit. The numerical simulations were failure mechanisms. Particularly, MCC is used to simulate the
performed based on Class A and Class C prediction approaches and behaviour of clay soils. But, since MCC is an isotropic model, it does
field monitoring settlement data. Sensitivity analyses of stiffness not consider creep (viscous behaviour) of soft soils [14]. To repli-
and permeability parameters were carried out depending on Class cate the behaviour of soft natural soil accurately, it is essential to
C prediction. Systematic comparisons were made between develop a constitutive model that capable of capturing soil aniso-
numerically-predicted values of the Class A and Class C predictions tropy, de-structuration, and creep [21]. A constitutive law integrat-
and field measured settlement values. ing all three characteristics, however, is very complicated and
would require many constitutive parameters to be determined by
different complicated stress path tests. The method which
2. Fundamentals in PVD modelling accounts the behaviour of soft soil materials including the creep
effect, though it does not account for the de-structuration effect,
2.1. Effects of PVD is Soft Soil Creep (SSC) model. For this reason, the Soft Soil Creep
(SSC) model was employed in this study to reproduce the beha-
Engineering practitioners commonly prefer simplified analysis viour of the Karakore soft deposit. SSC is a second order model for-
methods to determine the behaviour of PVD-improved soft ground mulated based on a visco-plastic framework [22].
under an embankment surcharge loading. The two prominent
authors who developed analytical frameworks on this subject are 2.3. Parameters in PVD modelling
Barron [9] and Hansbo [10]. These authors are pioneers in studying
on the concepts, theories and applications of a unit cell condition The main parameters required for the PVD-induced consolida-
(surrounded by a soil cylinder) and a radial drainage in the PVD- tion process are the discharge capacity of PVD (qw), the smear zone
induced consolidation processes. However, in many practical situ- diameter (ds) around the PVD and the ratio of horizontal perme-
ations, the forms of ground deformations do not represent the ability of the undisturbed zone (kh) to the soil smear zone (ks),
behaviour of a unit cell condition. As a result, a number of simpli- i.e. kh/ks [11,23]. In the past decades, several efforts have been
fied approximate methods have been devised as alternatives. The made to determine these key parameters. Authors suggest that if
simplified methods are represented by an equivalent vertical per- test data are not readily available , qw can be assumed as
meability which take into account the effects of the natural soil 100 m3/year and ds = 3 dm, where dm is an equivalent diameter
on the vertical drainage and the PVD on the radial permeability of the mandrel of the machine which is used to install the PVDs
[2,11]. Four types of methods have been commonly used in mod- [11].
elling the behaviour of PVD-improved soft subsoils under embank- According to Eriksson et al. (2000) and Chu et al. (2004), kh/ks
ment [12,13]. The first method is a one dimensional (1D) drainage ranges from 2 to 6 [24]. As reported by Long et al. (2013) kh/ks cov-
element model that the PVD is represented by a line element with- ers a wide range from 1.4 to higher than 10 [6]. In the 3D model,
out a distinct cross-sectional area [14]. The second one assumes as the ratio of kh/ks can be assumed to be 20 and ds = 5 dm for Poko
a cylindrical unit cell model of the PVDs and the surrounding soil clay in Finland [25]. The vertical permeability was always half
mass. The PVD is considered as a macro-element representing the horizontal value [18]. Hansbo (1987) proposed that that kv = ks
the vertical drainage behaviour [5]. The third one represents the based on laboratory test results. But, laboratory Oedometer tests
PVDs as a solid element [15]. The fourth method assumes an usually underestimate the coefficient of permeability of the field
approximate and equivalent vertical hydraulic conductivity (kve) deposits [11]. In general, the relationship between field and labora-
of the PVD-improved subsoil. In this method, kve is computed from tory coefficient of permeability can be estimated as [11]:
J.J. Muhammed et al. / Engineering Science and Technology, an International Journal 23 (2020) 1015–1027 1017
ðkh =ks Þf ¼ C f :kh =ks l orðkh Þf ¼ C f :ðkh Þl ð1Þ where ksp is permeability of the smear zone for plane strain case; ks
is permeability of smear zone for axisymmetric case; n = R/rw - R is
where Cf = coefficient of permeability ratio of field to laboratory val- the radius of influenced zone by the PVD and for a rectangular pat-
ues. The value of Cf is greater than one and needs to be estimated tern of drain installation and computed as in [10], rw is the equiva-
from field and laboratory coefficient of permeability tests or by lent radius of the drain considering a corner effect and can be
back-analysis [5,23]. calculated as suggested by [28], s = rs/rw - rs is the radius of the
Tavenas et al. [42] studied for Canadian marine clays that kh/kv is in smear zone and related to the diameter of the mandrel [11]; l is
the range of 0.91–1.42. Similarly, Bergado et al. (1991) estimated that the length of the drain; z is the depth of the drain; qw is the dis-
the ratio for Bangkok clay is between 1.5 and 2.0. For certain Japanese charge capacity of the drain. qz is the equivalent discharge capacity
clays, Shogaki et al. (1995) reported that the range falls between 1.36 of the drain in plane strain, and dimensions B is half-width of the
and 1.57 [24]. Jamiolkowski et al. [48] and other authors reported that influence zone, bs is half-width of the smear zone, and bw is half-
the ratio of kh/kv ranges from about 2 to 15 [14,26]. equivalent width of the drain in 2D, and the schematic detail is pre-
sented as in [15].
2.4. Geometry model Ignoring the effect of well resistance, the equation becomes:
1 2 bs 2 2
The existing ground is generally gently undulating, crossed by
b¼ 2
ðbs bw Þ þ 3
ð3bw bs Þ ð2eÞ gullies, with few volcanic rock outcrops. The map of the study area
B 3B
is presented in Fig. 1. At station KM 160 + 750 (Borehole 160BH-
2 01), the groundwater table (GWT) is close to the ground surface
2khp bw
h¼ 0 1 ð2fÞ and it is located at 1.9 m from the original ground level. The
k sp Bqz B
GWT measurement was taken during the rainy season on August
8th, 2015. The bearing capacity and settlement performance of
2 these soil was not sufficient to carry the embankment and railway
qz ¼ qw ð2 gÞ
pr e traffic loads without proper ground treatment.
1018 J.J. Muhammed et al. / Engineering Science and Technology, an International Journal 23 (2020) 1015–1027
Fig. 1. Location of the project area (source: Author’s reconstructed map and from Google map).
In the study site of AKH railway project, four major groups of properties are also reported in [31,47]. The soft deposit has an
lithostratigraphic units are encountered. The units are Eocene vol- average relative resistance of SPT value of N = 2. From particle size
canic, Oligocene-Miocene volcanic, Upper Miocene volcanic, Qua- distribution (PSD) curve, both clay and silt soil are the dominant
ternary volcanic. Besides, these units are overlain by associated grain sizes. At shallower depths, the soft silty clays have water con-
Quaternary deposits of lacustrine, fluvial and colluvium origin in tents that exceed their liquid limits. It indicates that these soils in
the plains and sometimes on the slope areas. The Tertiary Eocene this region contain organic materials. The average bulk density of
volcanic of Ashangi formation (Tas) is overlain by Quaternary- the soft soil is nearly 17 kN/m3. Pre-overburden pressure (POP) is
ancient alluvium (Qal-A) and in river banks and gully areas is by preconsolidation pressure (r0 p) less vertical effective stress (r0 vo)
recent alluvium (Qal-R) [30]. Hence, the geological profile of the and has an average value of 27 kPa. The soft alluvium deposit
Karakore site corresponds to the Quaternary lacustrine and fluvial has an average OCR value of 1.6 and can be classified as lightly
deposits, particularly of alluvium sediments (Qal-A). The antici- over-consolidated clay. The soft soil has low hydraulic conductivity
pated railway tracks and embankments were laid on this ancient and lower shear strength. The stiffness parameters such as com-
alluvium deposit. The subsoil deposit is mainly composed of cohe- pression index (Cc) and recompression index (Cr) were determined
sive soils with coarse-grained interbeds in irregular patterns. The from incremental loading (IL) of Oedometer test. Cc and Cr range
top crust can be classified as a black cotton soil and characterized from 0.13 to 1.38 and 0.017 to 0.14 respectively, whereas an
by dark gray, dark brownish gray or black in colour. adopted mean value of initial void ratio (eo) is 1.28. Vertical
hydraulic conductivity (kv) was estimated from the coefficient of
3.2. Vertical soil profile and laboratory test results consolidation (cv) of consolidation test. However, for the loose
gravelly sand (GS) layer, the values were adopted from literature
Tables 1–3 present the soil properties and classification of the for the numerical analysis and its values are presented as in Table 5
soft silty clay deposit at KM 160 + 750 station. Some of the soil and under subsection 5.3.
J.J. Muhammed et al. / Engineering Science and Technology, an International Journal 23 (2020) 1015–1027 1019
Table 1
Particle size distribution and classification of Karakore soft subsoil.
Soil Depth (m) Gravel/Sand (%) Silt (%) Clay (%) Natural water content, wn (%) Liquid limit, LL (%) Plasticity index, PI (%) Classification (ASTM-D2487)
0.50–3.25 16 50 34 57 41 27 CL
3.25–6.50 27 49 24 135 58 23 MH
6.50–9.50 89 11 – 108 – – SP-SM
9.50–12.75 20 40 40 41 52 28 CH
12.75–15.75 5 35 60 42 57 31 CH
15.75–18.25 4 26 70 40 65 40 CH
18.25–21.25 6 45 49 40 59 37 CH
21.25–26.00 13 53 34 42 48 26 CL
Table 2
In-situ properties and pressure history of Karakore soft subsoil.
Soil Depth (m) SPT (N) Gs cb (kN/m3) r’vo(kPa) r’p (kPa) POP OCR
0.50–3.25 2 2.59 16.83 21 52 32 2.4
3.25–6.50 2 2.59 15.50 27 45 18 1.7
6.50–9.50 3 2.64 – – – – –
9.50–12.75 2 2.50 17.67 88 103 15 1.2
12.75–15.75 8 2.48 17.17 109 125 16 1.2
15.75–18.25 11 2.48 17.00 127 210 83 1.6
18.25–21.25 10 – – – – – –
21.25–26.00 15 – – – – – –
Table 3
Compression and consolidation parameters from IL Oedometer tests.
4. Field monitoring of embankment square pattern. The preloading fill activity was done stage by stage
with 0.5 m thickness at regular time intervals.
4.1. Description of the embankment system
4.2. Monitoring of the embankment fill
The embankment considered in this analysis is a part of the
2 km long PVD-improved soft ground and located near Karakore A field monitoring was performed to measure the vertical set-
town at KM 160 + 750. The embankment has a base width of tlement of the fill on the PVDs-improved soft ground. Several opti-
60 m and 6.5 m top width with side slope of 3H:1V. The total cal monitoring points (MP) (at selected 25 different locations) were
height of the embankment surcharge fill including extra fill is placed on the top surface of the fill (embankment + extra sur-
8 m. A typical vertical section of the embankment is presented in charge) at reasonable distances along the railway line after the fill
Fig. 2. The embankment section has three important parts. The first activity was completed. The survey readings were taken using an
part involved of the PVD-improved soft alluvial subsoil. The first accurate optical survey equipment (with 1 1011m accuracy)
layer of this part is a very soft silty clay (SC1) deposit of 6.5 m thick. from the centreline (CL) and edges of the fill. The monitored had
The next layer is an interbedded loose gravelly sand (GS2) having a been done on a regular basis (every seven days).
3.0 m thickness and is followed by a 3.0 m thick very soft silty clay This paper utilised the field settlement data from the monitor-
(SC3). The entire PVD-improved zone was 12 m thick and extended ing point of the 23rd (KM 160 + 750) location. The monitoring
to this depth. Below the soft clay soil is a firm to very stiff Eocene activity was started on December 2, 2016 and has been continued
silty clay (ESC4) having a 13. 50 m thickness and extending down for more than 600 days. The observation usually provides data for
to 25.95 m depth. The thickness of the soft silty clay deposit ongoing fill and future construction on similar soils and will also
improved with PVDs is about 12 m, while the stiff silty clay portion ensure that the rate of settlement has subsided to an acceptable
without PVDs is about 13.5 m. For the stiff clay, the N value is level prior to laying of the railroad tracks. Observation of the settle-
greater than 8. The second part consisted of a working platform, ments was designed to use settlement beacons, with a base plate
drainage blanket, and filter layers. The third part comprised of a placed under the fill and more sophisticated instrumentation (such
surcharge embankment fill constructed with selected granular as rod extensometers). Unfortunately, these instrumentations
materials. Procedurally, the working platform with a 0.5 m thick- were damaged due to different reasons. Instead, optical topograph-
ness was constructed before the installation of the PVDs. Over ical surveying (OTS) has been used reliably to measure the vertical
the top of the platform a 0.6 m crushed rock drainage and 0.2 m settlements of the fill. The description of the monitored embank-
filter layers were placed after the installation of PVDs in a 1.3 m ment section at the 23rd point and detail of PVDs are provided in
1020 J.J. Muhammed et al. / Engineering Science and Technology, an International Journal 23 (2020) 1015–1027
the Table 4. The PVDs related information was obtained from the 3D -axisymmetric condition to 2D plane strain condition, perme-
ceTeau company [30]. ability matching procedure was employed based on several inten-
sive studies [11,13,15,23,34,35]. These studies reported that both
3D and 2D methods reasonably achieved similar time-settlement
5. Finite element modelling
curves. According to Indraratna and Redana [15], Rujikiatkamjorn
et al. [35] and Muthing et al. [17] the settlement computed from
5.1. The modelling approaches
a 3D modelling and a transformed modelling from 3D in to 2D gave
excellent matching of results.
The numerical analysis was carried out by finite element (FE)
FE analysis was performed by PLAXIS 2D geotechnical computer
program using Plaxis 2D. Class A prediction followed by Class C
program. To modify some of the input soil parameters such as per-
back-calculation were employed for the prediction of settlement.
meability, so as to represent the radial behaviour of the field con-
In Class A prediction, laboratory test values were used in estimat-
dition, the axisymmetric problem (3D) was transformed in to plane
ing settlements. The parameters adopted in the Class A prediction
strain condition (2D). The permeability in the horizontal direction
are summarized as in Table 5.
was converted in to 2D model using the matching formula given in
SCC model considers time-dependent behaviour of soft soils
Eq. (5). This relationship considers smear effect, relatively simple
such as primary and secondary compressions [32]. SSC model
and most convenient in computational efficiency and provides a
was applied in the FE analysis to overcome the limitations of its
reasonable estimation for horizontal permeability in plane strain
predecessor models. Hardening soil (HS) model also used for com-
condition [18,24]. For qw greater than 150 m3/year, the impact of
parison with SSC modelling results. Mohr-Coulomb was employed
well resistance and kinking can be neglected [36]. Hence, in this
for the granular fill materials such as for embankment, drainage
study, as the PVD has qw of more than 2300 m3/year, Eq. (5) is
and working platform layers. For the stiff silty clay layer underlain
rationally applicable to use.
the PVD-improved region, as this layer was found to be nearly nor-
The equivalent drain radius (R) was calculated as 2R = 1.13.S
mally consolidated clay, Modified Cam-Clay (MCC) model was used
and R = 1.13/2.1.3 m = 0.735 m, where S is the drain spacing; radius
[33]. The input parameters for MCC were determined from labora-
of smear zone (rs) is related to the diameter of mandrel (dm) and
tory tests of strain-log pressure history curve. They were obtained
was assumed as 2rs = 3 dm [11] and hence, rs = 3/2.(120 mm) =
and calibrated based on loading–unloading of two cycle process.
180 mm = 0.18 m; 2rw = (w + t)/2 and rw = 1/2.(98 + 3)/2 mm =
25.25 mm = 0.0253 m, where w and t are width and thickness of
5.2. Permeability matching procedure the PVD respectively; the reduced permeability in the smear zone
was assumed as ks = kh/2. As the value of (kh/ks)Cf can be affected
PVD-improved grounds are usually 3D problems and can be by several uncertain field factors, Cf was assumed to be 1 for Class-
solved with axisymmetric solutions. However, several studies have A prediction, and it was back-calculated during Class C prediction.
been performed by transforming 3D into 2D finite element method
to reduce time of computation and effort. In such studies, widely
acceptance and successful transformations are made by Indraratna 5.3. Model input parameters
and Redana [15], Hird et al. [27] and Chai et al. [11]. Transformed
2D modelling captures well the settlement behavior of soft soils In Class A prediction, laboratory tests data accompanied by
[19]. In this study, so as to achieve a realistic transformation from some assumptions for the unknown parameters, were directly
Table 4
Embankment geometry and PVD description.
Embankment Name H (m) Hp (m) Hd (m) Hcomp (m) hPVD(m) Remarks on CT-D1010 PVD
E23 6.2 8 1.3 12 12 w = 98 mm, t = 3 mm, cc = 36 channels
Note: H = final embankment height; Hp = Height of embankment before removing the extra fill; Hd = thickness of drainage and working platform layer; hcomp = thickness of
compressible layer; hPVD = depth of PVD improved; w = width of PVD; t = thickness of PVD; cc = core configuration.
J.J. Muhammed et al. / Engineering Science and Technology, an International Journal 23 (2020) 1015–1027 1021
used for the analysis. In Class C prediction the influential and sen- fill’s effective weights. The analysis simulates the dependency of
sitive parameters were revised to match the field measured settle- settlements with the real thickness of the soft and compressible
ment. For the fill of the embankment, drainage and working layers by considering the deformation of the previous phases.
platform a unit weight of 20 kN/m3 and an internal angle of friction The reduced thickness of the soil is therefore logically resulted in
(/) of 40° were used. The strength parameters such as reference lesser settlements, and it makes the design reasonably economical
cohesive strength (cref) of 4 kN/m2 and an internal angle of friction [44].
of (/) 26° were used for all the soft layers. The boundary condition for displacement was modelled in such
For the loose gravelly sand layer (GS2), basic engineering prop- a way that at the bottom both vertical and horizontal displace-
erties such as unit weight and classification were determined from ments were fixed; for the left and right side boundaries the hori-
laboratory tests. Whereas, as coefficient of permeability and zontal movement was fixed while the vertical boundaries were
strength parameters were not carried out in the lab, the values free to move. The bottom, vertical and horizontal drainage bound-
for the model analyses were systematically adopted from literature aries were closed to flow and consolidation effect while the top
[37–39]. As the layer consists of gravelly size with poor gradation boundary was free to flow and consolidation.
and loose deposition, a coefficient of permeability of 0.5 m/day and
an internal angle of friction of 34° were used. 6. Modelling results and comparison with the field data
The soil parameter values and employed soil models are listed
in Tables 5–7. As literature suggest, the horizontal coefficient of 6.1. Class a prediction
permeability of the soil (kh) was assumed to be two times of the
vertical permeability (kv). The secondary consolidation coefficient Class A settlement was predicted from numerical soil model
(Ca) was reasonably supposed to be 3% of Cc [4,40]. Coefficient of using the laboratory test data from Borehole160BH-01 (station
permeability variation with void ratio was estimated as per the KM 160 + 750) and the soil parameters are presented in Table 5.
suggestion by Taylor’s solution [41]. The settlement results were considered from the top centre of
the embankment which was at the centreline of the railway line.
k ¼ ko :10ðeo eÞ=C k ð8Þ
The comparison between Class A prediction and field monitoring
where k0 is the initial permeability; e0 is the initial void ratio; k is results for the vertical settlement-time curve is presented in
the current permeability; e is the present void ratio; and Ck is the Fig. 3. The field monitoring was started 7 days after the completion
permeability change index and is a constant value and can be of the final fill placement. It was nearly 128 days from the start of
adopted from Ck = 0.5e0 [42]. the surcharge fill placement. The analysis was carried out from 128
to 438 days (from December 2, 2016 to October 17, 2017) which
5.4. Numerical analysis was for the period of 310 effective days from the beginning of
observation of the embankment fill.
In PLAXIS 2D, for both the embankment fill and the soft subsoil The numerical simulation revealed that there were uncertain-
layers, a plane strain condition with a discretization of 15-node of ties associated with the settlement estimation of the alluvial soft
triangular elements were used. The embankment fill, working plat- subsoil deposit. Overall, though the simulation underestimated
form, the drainage layer and the interbedded gravelly sand layers the settlement value, the prediction of the settlement-time profile
were modelled as ‘drained’ condition, while the other soft and stiff curve agreed with the in situ measurements. Unlike the field mea-
silty clay layers were modelled as ‘undrained’ [40]. surement, the settlement-time curve from the numerical simula-
The PVDs were modelled as vertical drains which are prescribed tion was very smooth, because it represented an ideal situation.
lines inside the geometry model where excess pore water pressure The discrepancy between the predicted and measured data was
(EPWP) are supposed to be zero in the consolidation analysis in nearly 25% on the 438th day of observation. The limitation with
PLAXIS 2D [22]. But PLAXIS has limitations that it does not explic- Class A prediction was that it overestimated the stiffness of the soft
itly account for the drainage capacity of vertical drains, smear zone subsoil. The reason for the difference in results might be attributed
effect and well resistance. to the uncertainties associated with the laboratory test results. The
When the anticipated deformation of the soft subsoil under the laboratory test values could depend on sample qualities such as
embankment is large, Large Strain (LS) analysis is a more rational sample disturbance, less representation of the entire subsoil, test
approach than small deformation (SS) one. Large strain calculation conditions and soil heterogeneity. It clearly suggested that the soil
is significant when deformation affects the geometry of the model parameters in Class A prediction need to be modified and
embankment geometry [22]. To consider the large deformation back-calculated in the subsequent analysis for Class C prediction.
effect, the mesh update option in PLAXIS 2D incorporates a LS anal-
ysis based on the updated Lagrange formulation [43]. The LS 6.2. Back-Analysis: class C prediction
approach accounts for the increase in unit weight of soil mass by
taking into consideration the change in void ratio due to surcharge 6.2.1. Class C modelling
stress increment. Furthermore, the LS analysis also accounts In this study, the authors have attempted to develop a practical
updated water pressure which results in considering the surcharge and simplified methodology for selecting appropriate soil parame-
Table 5
Input parameters of alluvial subsoil in Class A prediction.
Layer Thickness (m) cb (kN/m3) mur() eo () k* () j* () l* () kvo (m/day) Kho (m/day) Khp (m/day)
04 04
Soft silty clay (SC1) 6 16.10 0.15 1.3 0.147 0.023 0.004 1.93 10 3.86 10 2.81 1005
Loose sand (GS2) 3 18.00 – – – – – 0.5 0.5 –
Soft silty clay (SC3) 3 17.30 0.15 1.1 0.038 0.006 0.001 2.72 1004 5.43 1004 3.95 1005
Stiff Eocene silty clay (ESC4) 13.45 17.30 0.15 0.96 0.11a 0.04b 0.003c 2.84 1004 5.69 1004 4.14 1005
Note: mur = Poisson’s ratio for loading/unloading; modified compression index, k*= Cc/2.303(1 + eo), modified swelling index, j* = 2Cr/2.303(1 + eo), modified secondary
compression index, m* = Ca/2.303(1 + eo); kvo = initial vertical permeability, kvo = initial horizontal permeability; MC = Mohr-Coulomb; SSC = soft soil creep, MCC = modified
cam clay. a,b,c are values given for Cam-Clay compression, swelling and secondary compression indices, respectively.
1022 J.J. Muhammed et al. / Engineering Science and Technology, an International Journal 23 (2020) 1015–1027
Table 6
Modified model parameters of the soft subsoil in Class C prediction.
Layer Thickness (m) cb (kN/m3) mur () eo () k* () j* () l* () kvo (m/day) kho (m/day) Khp (m/day)
Soft silty clay (SC1-1) 3.0 16.10 0.15 1.30 0.189 0.038 0.006 2.00E-04 2.42E-04 3.50E-05
Soft silty clay (SC1-2) 3.0 16.10 0.15 2.50 0.236 0.049 0.007 1.00E-04 1.70E-04 2.50E-05
Loose sand (GS2) 3.0 18 – – – – – 0.5 0.5 –
Soft silty clay (SC3) 3.0 17.3 0.15 1.10 0.062 0.012 0.002 3.00E-04 4.50E-04 4.00E-05
Stiff Eocene silty clay (ESC4) 13.5 17.3 0.15 0.96 0.11a 0.04b – 2.84E-04 5.69E-04 4.14E-05
a,b
Input parameters for MCC modelling.
Table 7
Modified model parameters of the soft subsoil in Class C prediction for HS Model.
fill was removed since the 438th day, the settlement and EPWP
the embankment body itself, although the fill was assumed as a curves had been raised up. The settlement curve resilience was
rigid in the modelling, (2) the creep settlement was started during likely attributable to its swelling/elastic behaviour of the subsoil
earlier period of observation than it was considered in the simula- and elastic rebound of the embankment body; whereas, for the
tion, and (3) time dependent deformation of the interbedded gran- pore water pressure curve, it was owing to the development of
ular layer (GS2) was not modelled explicitly. EPWP when the water ingressed to the subsoil mass. The ingres-
In class C prediction, the standard error of the estimates (SEest.) sion of water resulted in the development of suction pressure
for the conformity between the prediction and field settlement at around the PVD wells and subsequently the settlement showed a
different time of observations was found to be 5.5 for SSC mod- heaving response (Fig. 5). The developed suction pressure under
elling. Similarly, the coefficient of determination (R2) from the the centre of the embankment ranged from 19 to 2 kPa, which
overlap/association of estimated and field settlement was 0.974. was due to the removal of extra surcharge loading (36 kPa). The
Both inferential statistic values showed that the prediction well curves again went down (after the 481th day) very gradually with
correlated with the field condition with a very good precision. time. It was most likely because of uncompleted primary consoli-
Fig. 4 also compares the results of SSC modelling with HS mod- dation from the embankment preloading, the loading effect from
elling. Generally, SSC better predicted the field condition than HS the newly laid track components and due to creep behaviour from
model. During early period of observations, the two models closely the surcharge fill and superstructure track layers. The simulated
predicted the settlement – time relations. But during the latter per- settlement curve has been compared with the field measured data
iod of observations the two models showed significant differences. (Fig. 5).
The differences in predictions were attributed to the effect of creep From Fig. 6, one can observe that when the extra surcharge fill
behaviour in which SSC model accounted while HS did not involve was removed, the water has got the freedom to enter the soil mass
creep coefficient. for both with PVDs and without PVDs scenarios. Subsequently, the
However, the Class C prediction was not perfectly simulated the EPWP reduced(dropped) due to the development of suction pres-
settlement rate and magnitude of the field conditions. The discrep- sure (negative pressure). For both scenarios, the same trend of
ancies might be attributed to some of the following expected rea- reduction in EPWP was clearly observed. However, the rate of drop
sons. One, the field surface settlements were performed under was different. The rate for without PVDs case was relatively rapid
variable environmental conditions such as moisture content varia- with 0.48 kPa/day, while it was slower for with PVDs case with a
tions and groundwater table fluctuations in different seasons. Sec- rate of 0.39 kPa/day.
ond, as the geotechnical investigation showed that the subsoil is The ratio of kh/kv assumed during modelling was 2 and used for
interbedded by gravelly sand and very compressible organic mate- the analysis in all layers. After the back-analysis and sensitivity
rial, the SSC model could not adequately simulate these interbed- studies were carried out, the values of kh/kv were found in the
ded layers. The third reason could be the that the high fill range of 0.95 to 1.75. The average value of coefficient of permeabil-
embankment itself might have been deformed, though the ity ratio (Cf) of the field to the laboratory was found to be 1.5 from
assumption made was that the compacted embankment fill was the back-analysis of Class C prediction.
rigid and its body deformation was negligible.
Installation of PVDs has significantly improved the subsoil
behaviour in accelerating consolidation settlements and the dissi- 7. Parametric sensitivity analysis
pation of EPWP as presented in Figs. 4 and 6. The EPWP
distribution-time curves of Fig. 6 was considered at 4.5 m depth 7.1. Model parameters of consolidation
from the bottom of the drainage layer. The EPWP was found higher
at the centre of the more compressible layer (SC1-2) which is In soft soil deformation analysis, stiffness and consolidation
nearly at 4.5 m depth from the ground surface. The curves have parameters (Cc, Cr, Ca, eo, kv) are deemed key inputs. The effect of
been demonstrated for the Class C prediction with PVDs and with- the model parameters on the results of the analysis was assessed
out PVDs. The deformation and pore water pressure development, by varying one parameter at a time and keeping others the same
distribution and dissipation have been considerably dictated by the and by comparing the output with the reference value which is
interference of highly permeable soil layer (GS2). so called the most probable value (MPV). To evaluate the sensitiv-
The settlement (Fig. 5) and excess pore water pressure (Fig. 6) ity of parameters, a lower limit (LB) and an upper limit (UB)
against time curve profiles were influenced by loading of the sur- estimation were established. The LB and UB were estimated from
charge fill and removing of the extra surcharge fill. When the extra mean values for 68% confidence interval, which means one
1024 J.J. Muhammed et al. / Engineering Science and Technology, an International Journal 23 (2020) 1015–1027
Fig. 8. Effect of (a) initial void ratio, (b) vertical coefficient of permeability (c) over
consolidation ratio on settlement responses.
behaviour with the same extent that the initial void ratio had influ-
enced. The value of eo did not affect only the magnitude of the ver-
tical settlement but also the shape of the settlement-time curve.
Recompression index has the least impact on the magnitude of
deformation of the soft subsoil. Since the field settlement-time
curve was considered after the final surcharge load (160 kPa)
was applied, the level of stress applied was much higher than the
preconsolidation pressure (yield stress (52 kPa)) of the subsoil.
Applying a load beyond the yield stress, it makes the soil to be in
a state of normally consolidated condition.
and its effect distributed and reduced as it went deep down to the
soft soil mass.
The thickness of soft soil usually determines the depth of PVDs.
Fig. 9(b) presents the effect of PVD depth on settlement perfor-
mance. Increasing the PVD depth has improved the rate and mag-
nitude of consolidation settlement. Up to 6 m depth from the
surface of the ground, the depth of installation of the PVD has a
substantial contribution on the rate of deformation. But, increasing
the depth beyond 6 m has not contributed much for the improve-
ment. This is likely due to the presence of highly permeable grav-
elly sandy layer below the depth of 6 m. The 3 m thick gravelly
sand layer has considerably affected the development, distribution
and dissipation of pore water pressure. Hence, in this particular
section, 6 m PVDs might be sufficient to improve the rate of consol-
idation. However, in consideration to the pressure applied from the
machine during installation, additional depth of PVD might be
required.
Fig. 10 compares the effect of different permeability change
index on the settlement prediction. Initially, Ck = 0.5eo was adopted
to model the predicted settlement. Later, different Ck values ranged
from 0.2eo to 0.9eo were used. 0.2eo and 0.3eo under-predicted,
0.4eo, 0.5eo and 0.9eo are closely predicted whereas 0.6eo best pre-
dicted the settlement of the soft soil deposit of the study site.
[31] Muhammed J.J., Jayawickrama P.W., Theoretical and Numerical Settlement [42] F.P.J. Tavenas, P. Leblond, S. Leroueil, The Permeability of natural soft clays.
Predictions of an Embankment on PVD-improved Soft Alluvial Soil, In: 15th PartII: Methods of Laboratory measurement, Can. Geotech. J. 20 (1983) 644–
Int. Conf. Geotech. Eng., Lahore, Pakistan, 2019, In Press. 659.
[32] R.B. Brinkgreve, W.M. Swolfs, S. Kumarswamy, PLAXIS 2D Reference Manual. [43] T. Benz, N. Steinar Numerical Methods in Geotechnical Engineering. In:
Plaxis Bv 2014; Anniversar. Proceeding 7th Eur. Conf. Numer. Methods Geotech. Eng. Trondheim Norway,
[33] PLAXIS 2D. Scientic Manual 2018:1–120. 2–4 June 2010, 2010.
[34] J.-C. Chai, N. Miura, S. Sakajo, D. Bergado, Behavior of vertical drain improved [44] P.J.V. Oliveira, L.J.L. Lemos, Numerical analysis of an embankment on soft soils
subsoil under embankment loading, Soils Found 35 (1995) 49–61. considering large displacements, Comput. Geotech. 38 (2011) 88–93, https://
[35] C. Rujikiatkamjorn, B. Indraratna, J. Chu, 2D and 3D numerical modeling of doi.org/10.1016/j.compgeo.2010.08.005.
combined surcharge and vacuum preloading with vertical drains, Int. J. [45] D. Zheng, J. Huang, D.Q. Li, R. Kelly, S.W. Sloan, Embankment prediction using
Geomech. 8 (2008) 144–156. testing data and monitored behaviour: a Bayesian updating approach, Comput.
[36] B. Indraratna, I.W. Redana, Numerical modeling of vertical drains with smear Geotech. 93 (2018) 150–162, https://doi.org/10.1016/j.compgeo.2017.05.003.
and well resistance installed in soft clay, Can. Geotech. J. 37 (2000) 132–145, [46] R.B. Kelly, S.W. Sloan, J.A. Pineda, G. Kouretzis, J. Huang, Outcomes of the
https://doi.org/10.1139/t99-115. Newcastle symposium for the prediction of embankment behaviour on soft
[37] J.K. Mitchell, K. Soga, J. Wiley, Fundamentals of Soil Behavior, Third Ed., soil, Comput. Geotech. 93 (2018) 9–41, https://doi.org/10.1016/
2005. j.compgeo.2017.08.005.
[38] Kirsch K, Bell A. Ground Improvement, 2009. https://doi.org/10.1007/978-3- [47] J.J. Muhammed, P.W. Jayawickrama, S. Ekwaro-Osire, Uncertainty Analysis in
540-89230-4. Prediction of Settlements for Spatial Prefabricated Vertical Drains Improved
[40] Y. Gong, Y.H. Chok, Predicted and measured behaviour of a test embankment Soft Soil Sites, Geosciences 10 (2/42) (2019) 1–24, https://doi.org/
on Ballina clay, Comput. Geotech. 93 (2018) 178–190, https://doi.org/10.1016/ 10.3390/geosciences10020042.
j.compgeo.2017.06.003. [48] M. Jamiolkowski, R. Lancellotta, W. Wolski, Pre-compression and speeding up
[41] D.W. Taylor, Fundamentals of Soil Mechanics, 2nd., John Wiley and Sons Inc., consolidation, General Report, Special Session 6, Proc. Eighth Europe Conf. Soil
New York, 1948. Mech. and Found. Engrg. (1983) 1026–1201.