SOILS AND FOUNDATIONS Vol. 47, No.

3, 559576, June 2007

Japanese Geotechnical Society

FINITE ELEMENT ANALYSIS OF LATTICE-SHAPED GROUND IMPROVEMENT

BY CEMENT-MIXING FOR LIQUEFACTION MITIGATION

TSUTOMU NAMIKAWAi), JUNICHI KOSEKIii) and YOSHIO SUZUKIiii)

ABSTRACT
The dynamic behavior of lattice-shaped ground improvements by mixing with cement was investigated using
numerical analysis. Three-dimensional eective stress nite-element analyses were conducted to examine the eects of
dimension and strength of the improved ground on the potential for liquefaction mitigation. In these analyses, both
elastic and elasto-plastic models were used for expressing the behavior of the improved ground, where the elasto-
plastic model can describe the post-peak stress-strain behavior of cement-treated soils. The numerical results suggested
that the improvement area ratio and the elastic modulus of the cement-treated soil aect the potential of the improved
ground for liquefaction mitigation. Moreover, numerical analysis using the elasto-plastic model showed that partial
failure of the improved ground causes no considerable reduction in the potential for liquefaction mitigation. Since
such partial failure of improved ground can be taken into account appropriately, the analysis using the elasto-plastic
model provides an adequate solution for the performance-based design of a lattice-shaped ground improvement.

Key words: cement-mixing, nite element method, ground improvement, liquefaction, mitigation, seismic response
analysis (IGC: E8)

INTRODUCTION
Lattice-shaped ground improvements using soil-
cement mixtures as typically shown in Fig. 1 have been
developed as a method for remediation of liquefaction.
The performance of this method was conrmed at the
time of the Hyogo-ken Nanbu earthquake in 1995
(Suzuki et al., 1996a; Tokimatsu et al., 1996). Applica-
tions of this liquefaction mitigation method have in-
creased since then.
The specication of the lattice-shaped ground improve-
ment depends mainly on the potential of liquefaction
prevention. Moreover, the internal and external stabili-
ties are examined in the current design procedure
(PWRC, 1999; PWRI, 1999). The external stability of the
improved ground is evaluated under various potential
modes of failure, and the internal stability is examined by Fig. 1. Case history of application of lattice-shaped ground improve-
ment
ensuring that the stresses induced inside the improved
ground remain within the allowable ranges.
The lattice-shaped ground improvement prevents the located on the outside of the improvement zone ( see
unimproved sand deposits that are left in-between the Fig. 2). These external forces can cause large tensile and
improved ground grids from liquefaction by impeding shear stresses inside the cement-treated soil walls. There-
their shear deformation during an earthquake. On the fore, the internal stability often dominates the specica-
other hand, the cement-treated soil walls should resist tion of the improved ground at high seismic load levels in
against the inertia force of the unimproved soil mass the current design procedure.
surrounded by the improved soil grids and the dynamic In the current allowable stress design method, partial
earth pressure exerted from the original soils that are damage to the improved soil grids is not allowed.
i)
Associate Professor, Kobe City College of Technology, Japan (namikawakobe-kosen.ac.jp).
ii)
Professor, Institute of Industrial Science, University of Tokyo, Japan (kosekiiis.u-tokyo.ac.jp).
iii)
Takenaka Corp., Japan (suzuki.yoshiotakenaka.co.jp).
The manuscript for this paper was received for review on August 22, 2006; approved on January 15, 2007.
Written discussions on this paper should be submitted before January 1, 2008 to the Japanese Geotechnical Society, 4-38-2, Sengoku,
Bunkyo-ku, Tokyo 112-0011, Japan. Upon request the closing date may be extended one month.

559
560 NAMIKAWA ET AL.

Fig. 2. External forces applied to improved soil grids during earth-

quake
Fig. 3. Site location of superstructure at Meriken Wharf

However, the improved ground might be eective against

liquefaction of the surrounded sand deposits even after it
has partially failed during an earthquake. In such a case,
the improved soil grids can be designed more rationally
by the performance-based design method in which a
partial failure of the improved soil grids is accepted.
The performance-based design requires an accurate
evaluation of the eects of partial damage on the liq-
uefaction mitigation potential of the improved ground.
Therefore, to develop an economical and rational design
procedure, it is important to evaluate the behavior of the
improved soil grids in which the induced stress reaches
the strength of cement-treated soils.
In this paper, we present simulation analyses that were
conducted to investigate the behavior of lattice-shaped
ground improvements by cement-mixing under large
earthquake loads. Three-dimensional (3D) eective stress
nite element analyses (FE-analyses) were conducted for
three dierent models of the improved ground to examine
the eects of its dimension and strength on the potential
for liquefaction mitigation. In these analyses, both elastic
and elasto-plastic models were used for expressing the
behavior of the improved ground, where the elasto- Fig. 4. Soil prole, Bor. No. 7 (Suzuki et al., 1996a)

plastic model can describe the post-peak stress-strain

behavior of cement-treated soils. The results from the University during the 1995 Hyogo-ken Nanbu earthquake
numerical analysis using the latter model provided good that was evaluated following the procedure specied by
understanding of the performance of the improved soil recommendations for design of building foundations
that was partially damaged during an earthquake. (AIJ, 1988) is less than 1 in the reclaimed layer (Suzuki
et al., 1996b).
The 14-story building is supported by cast-in-place
CASE HISTORY OF LATTICE-SHAPED GROUND reinforced concrete piles with a diameter of 2.5 m extend-
IMPROVEMENT FOR LIQUEFACTION ing to dense diluvial sand and gravel at a depth of 33 m.
MITIGATION Its section and plan diagrams are shown in Fig. 5. The
The performance of the lattice-shaped ground im- building is 134 m59 m in plan and 60 m high. The
provement for liquefaction mitigation was conrmed at lattice-shaped ground improvement was applied because
the time of the Hyogo-ken Nanbu earthquake in 1995 of the potential for liquefaction in the upper loose ll
(Suzuki et al., 1996a; Tokimatsu et al., 1996). This case with a thickness of 1012 m. The ground improvement
history is herein described briey. that surrounds the cast-in-place piles extends to a depth
A 14-story building ( see Fig. 1) located on Meriken of about 16 m.
Wharf in Kobe city, Japan ( see Fig. 3) experienced the The improved ground was constructed using a Deep
Hyogo-ken Nanbu earthquake. The soil prole at this Cement Mixing method. The cement-treated soil walls
location consists of 1012 m of soft reclaimed sand and were produced by mixing the original soil with cement
gravel layers over the seabed. The seabed soil consists of slurry in overlapping columns with a diameter of 1 m.
alternating layers of clay, sand and gravel ( see Fig. 4). The treatment area ratio as dened by the ratio of the
The factor of safety against liquefaction for the east-west improved soil area to the whole site area was approxi-
component of the recorded earthquake motion at Kobe mately 20z. The unconned compressive strength of the
 FE-ANALYSIS OF GROUND IMPROVEMENT 561

Fig. 5. Section and plane views of superstructure and improved soil

grids

improved soil samples after curing of about six weeks was

46 MPa (Suzuki et al., 1996b).
A photograph of the surface of unimproved soil sur-
rounded by the improved soil grids was taken after the
Hyogo-ken Nanbu earthquake as shown in Fig. 6(a).
Vertical temporary timber supports were found to stand
still, suggesting no occurrence of soil-strength loss. No
settlement of the unimproved ground was reported
either. Figure 6(b) shows the top of the improved soil grid
at point P5 in Fig. 5. There was no crack at the surface of
the improved soil. The head of the cast-in-place pile sup-
porting the building was found to be intact as shown in
Fig. 6(c). Moreover, no dierential settlement was ob-
served on the rst oor of the building.
As shown in Fig. 6(d), the quay walls that formed the
wharf had slid considerably into the sea by the earth-
quake. The concrete caisson type quay walls on the west,
south and east moved horizontally by 1 m, 2 m, and
0.6 m respectively and settled by 0.5 m, 0.6 m and 0.3 m.
Sand boils and ground ssures were observed at the
surface of the original soil layer near the building. There-
fore, these large displacements of the quay walls were due
mainly to the liquefaction of the original soil layer.
These inspection results indicate that ground improve-
ment by cement-mixing could mitigate the damage to the
pile foundation and the superstructure, although the
liquefaction of the original soil layer had caused severe
damage to the quay walls.
Fig. 6. Foundation of superstructure and quay wall after the Hyogo-
ken Nambu earthquake: (a) Surface of unimproved soil surround-
NUMERICAL CONDITIONS ed by improved soil grid, (b) Improved soil grid head, (c) Cast-in-
place pile head and (d) South side quay wall
Finite Element analyses were performed using the
dynamic eective stress analysis code (Shiomi et al.,
1993) to survey the behavior of the lattice-shaped ground
improvement during a large earthquake. Two-phase
562 NAMIKAWA ET AL.

jf Table 1. Calculated cases

Grid space qu of improved soil Analysis Model for improved soil

Case 1 4 m4 m 2 MPa 1a Elastic model

2a Elastic model
Case 2 2 MPa
2b Elasto-plastic model
8 m8 m
3a Elastic model
Case 3 5 MPa
3b Elasto-plastic model

Free eld 4, 5, 6

qu: Unconned compression strength

in the other cases. The treatment area ratios are 39z and
23z, respectively, for the 4 m4 m grid and 8 m8 m
grid. The unconned compression strengths qu were
assumed as 2 MPa (case 1, case 2) and 5 MPa (case 3).
The former value corresponds to the typical in-situ
strength of the improved sandy soils (Matsuo, 2002), and
the latter value was chosen by referring to the strength of
improved soil sample in the case history described in the
previous section (Suzuki et al., 1996b). For analyses 1a,
2a and 3a, the improved soil was modeled using the
elastic model. For analyses 2b and 3b, the improved soil
Fig. 7. Geometry of analyzed ground with grid-improved soil areas in was modeled using the elasto-plastic model. In addition,
a line the free eld response without ground improvement was
computed for comparison with the response of the
unimproved soil surrounded by the improved soil (anal-
formulation (u-U formulation) was used to compute the yses 4, 5 and 6).
generation and mitigation of excess pore water pressures
in liqueable sand deposits. Three-dimensional FE- Mesh
analyses were conducted using a mesh consisting of eight The nite element mesh with the denition of coor-
node isoparametric elements. Innitesimal strain was dinate system is shown in Fig. 8. Half of one section was
assumed in the present study. The iterative initial stress modeled in the analyses. The improved soil consists of
method was used incrementally for solving the equilibri- four grids in a line for the 4 m4 m grid (case 1) and two
um equation. grids in a line for the 8 m8 m grid (cases 2 and 3).
Two models, an elastic model and an elasto-plastic Boundary conditions for the 3D analyses are shown in
model, were used for assessing the behavior of the soil Fig. 9. The side boundary was modeled using a cyclic
improved by cement-mixing. The elasto-plastic model boundary and the bottom boundary was modeled using a
can describe the behavior of cement-treated soil under a viscous boundary. The inuence of the cyclic boundary
general 3D stress state (Namikawa and Mihira, 2007). will be discussed later. In addition, symmetry planes were
It models tensile and shear strain-softening responses in assumed at both sides that are perpendicular to the z
localization zones after peak stress states. direction. Since the pore uid behavior was not consid-
A hypothetical improved ground consisting of several ered in the layers except for the liqueable sand layer, the
grids in a line was chosen to study the basic behavior of undrained boundary was assumed to exist below the
the lattice-shaped improved ground ( see Fig. 7). It should liqueable sand layer.
be noted that the case history described in the previous Figure 10 shows the mesh used for the simulation of the
section was not simulated exactly in the FE-analyses. It free eld. Two-dimensional plane strain elements were
was assumed that the depth to groundwater level from used here and the side boundary was modeled by a cyclic
the ground surface is 2 m and a liqueable sand layer boundary to represent the innite lateral extent.
exists at depths between 2 m and 8 m. The improved soil
grids are embedded to the clay layer below the liqueable Material Properties
sand layer by 0.6 m. The improved soil wall is 0.8 m Soil proles and models used in the analyses are shown
thick. in Fig. 11. The saturated sand layer consists of two-phase
The calculated cases are listed in Table 1. Two elements and the other layers consist of one-phase
parameters of the improved soil were varied in the elements. These soil properties are listed in Table 2.
analyses: grid space and strength. The improved soil The behavior of the saturated sand layer was described
walls were spaced at 4 m4 m in case 1, and at 8 m8 m using the densication model based on the Mohr-
 FE-ANALYSIS OF GROUND IMPROVEMENT 563

Fig. 8. 3D nite element mesh: (a) 4 m4 m grid model and (b) 8 m 8 m grid model

Coulomb's yield criterion (Shiomi et al., 1993). In this

model, excess pore pressure is estimated using an
endochronic dilatancy model that is based on the
empirical relationship between the accumulated strain
and the built-up excess pore pressure. In Table 2, A, B
and g indicate the densication model parameters. These
parameters were determined based on the prescribed
liquefaction resistance (td W p?0 )20, in which td is the
amplitude of cyclic shear stress and p?0 is the initial mean
p?0 )20 denotes the cyclic stress ratio that
eective stress. (td W
is necessary to cause liquefaction in 20 cycles. Figure 12
shows a simulation of an undrained cyclic loading triaxial
test using the densication model. The densication
Fig. 9. Boundary condition for 3D analyses model can describe appropriately the accumulation of
excess pore pressure during cyclic loading and the drastic
564 NAMIKAWA ET AL.

reduction of soil stiness after the stress reaches the general 3D stress states. The gravel layer was assumed to
failure state. However, this model can not simulate cyclic be linear. The shear-wave velocity of the base was
mobility after the initial liquefaction. assumed as 400 m W s.
For the surface layer and the clay layer, an elastic The initial elastic shear modulus G0 was calculated by
perfectly plastic model based on the Mohr-Coulomb substituting a void ratio e0.65 into the empirical
criterion was employed because, to the best knowledge of relation (Kokusyo, 1980).
the authors, any simple non-linear model has not been
established for expressing a cyclic behavior of soils under

Fig. 11. Soil proles and models used in analyses

Fig. 10. Finite element mesh for free eld response analysis

Table 2. Material parameters for unimproved soil

Damping factor h
Shear Friction
Density Poisson's Cohesion
No. Soil type modulus angle q
m3 )
r (t W ratio n c (kPa) Anal. 1a, 2a,
G (kPa) (deg) Anal. 5 Anal. 6
2b, 3a, 3b, 4

1 Dry sand 1.80 9000 0.33 0.01 36 0.05 0.025 0.075

2 30
3 30

4 Saturated 1.95 Eq. (1) 0.33 0.0 33 0.02 0.01 0.03

5 sand 35

6 40

7 Clay 1.60 26000 0.33 49.0 20 0.10 0.05 0.15

8 Gravel 2.00 48400 0.33 0.05 0.025 0.075

Liquefaction Densication
No. parameters Permeability
resistance k (m W
s)
(td W
p?0 )20 g A B

2 0.18 0.035

3 0.20 0.030

4 0.23 0.01 0.025 10 110 5

5 0.27 0.020

6 0.33 0.015
 FE-ANALYSIS OF GROUND IMPROVEMENT 565

matrices; a, b are the damping parameters. The damping

matrix C functions dierently depending on the frequen-
cies. Therefore, it is important to know the model damp-
ing factors at each mode. The modal damping factor of
the Rayleigh damper, hr, can be given as the following:

hr
1
2 a
vr
bvr (3)

In this equation, the subscript r is the order of the mode

and v is the natural frequency. The Rayleigh damping
parameters a and b were chosen to satisfy the condition
that the amount of damping corresponds to the assumed
damping factor h in Table 2 at the rst and second
dominant frequencies of the soil column.
The soils except the saturated sand are considered to
behave in the elastic region for most of the shaking time
because the elastic model and the elastic perfectly plastic
model are used for these soils. Thus the values of h in
Table 2 were determined in consideration of the relation-
ships between the damping ratio and the shear strain
(Kokusho, 1980; Kokusho et al., 1982; Rollins et al.,
1998). Here, the developing shear strains were assumed to
range from 1104 to 1103 in the dry sand and gravel
layers, and from 1103 to 1102 in the clay layer.
Conversely, in the saturated sand for which the densica-
tion model is used, the plastic strain could occur after the
initial liquefaction accompanying the failure and the
hysteresis damping could be mobilized. Therefore, the
damping without any consideration of the strain depend-
ence was used for this layer. In the analysis for the free
eld, three damping parameters were varied to examine
the inuence of the material damping. The inuence of
the damping will be discussed later.
The soil properties that were improved by mixing with
cement are listed in Table 3. The elastic model was used
for the improved soil in analyses 1a, 2a and 3a. These
Fig. 12. Element simulation results of densication model for sand
analyses require two parameters, elastic modulus E and
(No. 3): (a) Stress path, (b) Stress-strain relationship and (c)
Liquefaction resistance Poisson's ratio n. The values of E and n for the cement-
treated soil with the unconned compression strength
qu2 MPa (analyses 1a and 2a) were determined from
laboratory test results (Mihira et al., 2003; Namikawa
(2.17e)2 0.5
G08400 p? (1) and Mihira, 2007). Assuming that E is proportional to qu,
1 e the value of E for the cement-treated soil with qu5 MPa
In the above equation, p? denotes the mean eective (case 3a) was evaluated from that for the cement-treated
stress. The amount of strain is underestimated on the soil with qu2 MPa. The value of n was assumed not to
strain-hardening process if G0, which is calculated using change with the strength.
Eq. (1), is used directly for the elastic perfectly plastic The elasto-plastic model was used for the improved soil
model. Therefore, considering the strain-hardening in the analyses 2b and 3b. Parameters for the cement-
behavior, the elastic shear modulus G was assumed to be treated soil with the unconned compression strength
one-half of G0 in the present analysis. Initial stresses were qu2 MPa (analysis 2b) were determined using laborato-
estimated from densities listed in Table 2 and the ry test results (Mihira et al., 2003; Namikawa and Koseki,
coecient of earth pressure at rest, K00.5. 2006; Namikawa and Mihira, 2007). Assuming that the
Since these models do not express hysteresis damping elastic modulus E, the cohesion c, the tensile strength Tf,
mechanisms for the pre-peak region of the stress path, and the fracture energy Gf were proportional to qu, these
the viscous damping was used as the material damping. values for the cement-treated soil with qu5 MPa were
Rayleigh damping was employed for the solid phase as: evaluated from those for the cement-treated soil with
qu2 MPa. Other parameters were assumed not to
C aM b K (2)
change with the strength.
where M, C and K are mass, damping and stiness The stress-strain relationship of the improved soil
566 NAMIKAWA ET AL.

Table 3. Material parameters for improved soil

Parameters for elastic model

Density Elastic modulus Poisson's Permeability

Analysis
m3 )
r (t W E (MPa) ratio n k (m W
s)

1a, 2a 1.95 3500 0.167 1107

3a 1.95 8750 0.167 1107

Parameters for elasto-plastic model

Analysis Density Elastic modulus Poisson's Friction angle Cohesion

m3 )
r (t W E (MPa) ratio n q (degree) c (kPa)

2b 1.95 3500 0.167 30 577

3b 1.95 8750 0.167 30 1440

Analysis Tensile strength Hardening Hardening Fracture energy Softening

Tf (kPa) parameter a parameter ey Gf (N Wm) parameter er

2b 400 1.05 0.0002 15.0 0.40

3b 1000 1.05 0.0002 37.5 0.40

Analysis Dilatancy coecient Localization size Characteristics length Permeability

Dc ts0 (mm) lc (mm) k (m W
s)

2b 0.4 0.6 300 1107

3b 0.4 0.6 300 1107

applicability of this elasto-plastic model for the cement-

treated sand was conrmed by simulating the triaxial
compression, triaxial tension, plane strain compression
and bending tests on cement-treated sands. These simula-
tion results are available elsewhere (Namikawa and
Mihira, 2007).

Input Motion
The N-S component of the recorded earthquake at
Kobe JMA (Japan Meteorological Association) station
1995 with the maximum acceleration scaled to 300 cm W s2
was used as the base input acceleration ( see Fig. 14). The
response acceleration is about 850 cm W s2 at the rst
natural period of the free eld (0.41 s). The input motion
was applied laterally in the x direction. The duration of
the input motion was 20 s.

NUMERICAL SIMULATION RESULTS

Fig. 13. Unconned stress-strain relationship of elasto-plastic model Free Field
for improved soil Time histories of the acceleration and excess pore
pressure obtained from one-dimensional numerical anal-
ysis are shown in Fig. 15. The maximum acceleration at
employing the elasto-plastic model with parameters for the ground surface is about 400 cm W s2. This value cor-
analysis 2b is shown in Fig. 13. This stress-strain relation- responds approximately to the maximum acceleration
ship shows the behavior subjected to a cyclic loading recorded at the ground surface in Port Island during the
under an unconned condition. As shown in Fig. 13, the 1995 Hyogo-ken Nanbu earthquake (CEORKA, 1995).
tensile softening behavior is modeled by a bi-linear The steadily increasing excess pore pressure at a depth of
relationship with the elasto-plastic model used here. The 3.95 m reaches the level of the eective overburden
 FE-ANALYSIS OF GROUND IMPROVEMENT 567

Fig. 14. Input motion, Kobe JMA N-S component (1995) with maxi- Fig. 16. Inuence of damping variation on free eld response: (a)
s2: (a) Time history and (b)
mum acceleration scaled to 300 cm W Proles of maximum acceleration and (b) Proles of maximum
Acceleration response spectrum shear strain

layer. These results suggest that a typical example of

liquefaction phenomenon can be simulated under the
assumed conditions.
The inuence of damping is investigated based on the
free eld analyses (analyses 4, 5 and 6) in which the
damping factors h were varied as listed in Table 2.
Figure 16 compares the maximum acceleration and
maximum shear strain distributions. No signicant dier-
ence was observed among the results from the three
analyses, indicating that the inuence of h under the
conditions employed in the present study is not apprecia-
ble. Moreover, the maximum shear strains are in the
assumed range for determining the values of h in the dry
sand, clay and gravel layers.

Analysis 1a
The lattice-shaped ground improvement specied in
case 1 (4 m4 m grid, qu2 MPa) is analyzed using the
elastic model for the improved soil. The distribution of
excess pore pressures in the liqueable sand layer at t20
s is shown in Fig. 17. The excess pore pressures in the
unimproved sand deposits surrounded by the improved
soil grids are considerably lower than those in the outside
sand deposits. Figure 18 shows the excess pore pressure
time histories in the unimproved sand deposit at point A
(refer to Fig. 8 for its location). The time history of the
Fig. 15. Acceleration and excess pore pressure (E.P.P.) in free eld excess pore pressure in the free eld at the same depth is
(Analysis 4): (a) Acceleration and (b) Excess pore pressure also shown for comparison. The excess pore pressure in
the free eld reaches the initial eective overburden
pressure at about 7.5 s, indicating occurrence of liquefac- pressure, whereas that at point A increased slightly.
tion. Subsequently, the amplitude of the acceleration at These results show that the ground improvement in case 1
the upper layer decreases because the shear wave could (4 m4 m grid, qu2 MPa) can prevent the unimproved
not be transmitted completely through the liqueable sand layer from liquefying.
568 NAMIKAWA ET AL.

Fig. 17. Distribution of excess pore pressure at t20 s (analysis 1a)

Fig. 20. s3 in improved soil at point B in Fig. 8 (analysis 1a)

Fig. 18. Excess pore pressures (E.P.P.) in unimproved sand deposit
surrounded by improved soil grid at point A in Fig. 8 and free eld
at the same depth

Fig. 21. Inuence of cyclic boundary: (a) Proles of maximum

Fig. 19. Distribution of sz in improved soil at t5.6 s (analysis 1a) displacement and (b) Proles of excess pore pressure ratio at t
20 s

Fig. 22. Distribution of excess pore pressure at t20 s (analysis 2a)

 FE-ANALYSIS OF GROUND IMPROVEMENT 569

Figure 19 shows the distribution of stress in the z two ground displacement proles are similar to each
direction sz in the improved soil at t5.6 s, when the other in shape. There is little dierence in the distribution
input acceleration becomes nearly the maximum value of the excess pore pressure. These results suggest that the
(Fig. 14). Tensile stress caused by the bending moments ground response at the line G in the 3D analysis is
occurs at the corner of the improved soil grids. During approximately identical to the free eld response and the
the earthquake, because the improved soil grids restrict inuence of the cyclic boundary is not signicant.
the deformation of the unimproved sand deposit, exter-
nal forces, i.e., the inertial force of the soil deposit and Analysis 2a
the dynamic earth pressure, are applied to the improved The lattice-shaped ground improvement specied in
soil walls that are normal to the excitation. case 2 (8 m8 m grid, qu2 MPa) is analyzed using the
The time histories of the minor principal stress s3 at elastic model for the improved soil. The distribution of
point B (refer to Fig. 8 for its location), where the large excess pore pressures in the liqueable sand layer at t20
tensile stress is caused in the improved soil, are shown in s is shown in Fig. 22. The excess pore pressures in the
Fig. 20. For most of the shaking period, the induced unimproved sand deposits surrounded by the improved
tensile stress is lower than the tensile strength, indicating soil grids are lower than those in the outside sand
that no damage would occur in the improved soil in case 1 deposits. They suggest that the lattice-shaped ground
(4 m4 m grid, qu2 MPa), where the internal stability improvement in case 2 (8 m8 m grid, qu2 MPa)
of cement-treated soil wall is satised. reduces the excess pore pressures in the liqueable sand
The ground behavior along a line G (refer to Fig. 8 for deposit.
its location) was compared to the free eld behavior to Figure 23 shows the distribution of sz in the improved
examine the inuence of the cyclic boundary. Figure 21 soil grids at t5.6 s. The tensile stress in the improved
shows the comparisons of the proles of the maximum soil spaced at 8 m8 m is much larger than that in the
ground displacement and the excess pore pressure ratio improved soil spaced at 4 m4 m (case 1). Time histories
(calculated by dividing the excess pore pressure by the of s3 at point E (refer to Fig. 8 for its location) are shown
initial vertical eective stress) at t20 s when the shaking in Fig. 24. The tensile stress is much greater than the
ended. Although the ground surface displacement along tensile strength (400 kPa) for the main shaking period,
the line G is 20z smaller than that at the free eld, the indicating that the improved soil in case 2 (8 m8 m
grid, qu2 MPa) would be partially damaged during the
large earthquake.
These numerical results show that the improved soil in
case 2 partially failed during the earthquake. Although
the partial failure might induce a reduction in the
liquefaction mitigation potential of the improved
ground, the analysis with the elastic model does not

Fig. 23. Distribution of sz in improved soil at t5.6 s (analysis 2a) Fig. 24. s3 in improved soil at point E in Fig. 8 (analysis 2a)

Fig. 25. Distribution of excess pore pressure at t20 s (analysis 2b)

570 NAMIKAWA ET AL.

Fig. 27. Excess pore pressures (E.P.P.) in unimproved sand deposit

surrounded by improved soil grid at point D in Fig. 8

Analysis 2b
The lattice-shaped ground improvement specied in
case 2 (8 m8 m grid, qu2 MPa) is analyzed using the
elasto-plastic model for the improved soil. The distribu-
tion of excess pore pressures in the liqueable sand layer
at t20 s is shown in Fig. 25. The excess pore pressure in
the unimproved sand deposit surrounded by the im-
proved soil grids is reduced by the existence of the im-
proved soil.
The tensile failure zones in the improved soil during the
earthquake are shown in Fig. 26. At t6 s, failure occurs
at the bottom and upper corners of the improved soil
grid. Moreover, the horizontal failure zone exists in the
improved soil wall normal to the excitation. However, at
this moment, the failure zones do not penetrate through
the improved soil wall. The damage volume ratio as
dened by the ratio of the failure improved soil volume to
the whole improved soil volume is about 4.2z at t6 s.
Thereafter, at t20 s, when the shaking nishes, the
failure zones extend further and penetrate through the
walls at the corners. The damage volume ratio becomes
about 8.7z nally. These results suggest that, although
the failure of the improved soil initiates at the corner of
the grid, the improved soil grid does not collapse sudden-
ly; instead, it fails gradually during the earthquake.
Figure 27 shows the excess pore pressure time histories
in the unimproved sand deposit surrounded by the
improved soil grids in analyses 2a and 2b at point D (refer
to Fig. 8 for its location). The excess pore pressure in
analysis 2b is nearly identical to that in analysis 2a during
the early stage of earthquake shaking up to t5 s.
Thereafter, the excess pore pressure in analysis 2b
Fig. 26. Failure zones in improved soil (analysis 2b): (a) t6 s and (b)
t20 s
becomes larger than that in analysis 2a. This is because
the constraint of deformation of the liqueable sand
deposit by the improved soil decreases due to its partial
incorporate such eect of the failure of the improved soil. failure in analysis 2b. Nevertheless, the dierence
Consequently, analysis 2a might underestimate the excess between analyses 2a and 2b is of limited extent. These
pore pressure generated in the unimproved sand deposit results suggest that, in case 2 in which the nal value of
surrounded by the grids. Therefore, based on numerical the damage volume ratio is limited to about 8.7z, the
results obtained in analysis 2a, it is dicult to judge partial failure occurring in the improved soil during the
whether the ground improvement in case 2 can prevent earthquake does not greatly aect the potential of the
the sand deposit from liquefaction. liquefaction mitigation.
The failure process of the improved soil grid depends
on the characteristics of the input wave. Therefore, the
 FE-ANALYSIS OF GROUND IMPROVEMENT 571

Fig. 28. Distribution of sz in improved soil at t5.6 s (analysis 3a)

Fig. 29. s3 in improved soil at point E in Fig. 8 (analysis 3a)

characteristics of the input wave might aect the liquefac-

tion mitigation potential of the improved soil grid in
which the partial failure occurs during the earthquake.
Further studies are required to investigate the inuence of
the input wave characteristics on the performance of the
lattice-shaped improved ground.

Analyses 3a and 3b
The lattice-shaped ground improvement specied in
case 3 (8 m8 m grid, qu5 MPa) is analyzed using the
elastic and elasto-plastic models for the improved
ground, respectively, in the analyses 3a and 3b.
Figure 28 shows the distribution of sz in the improved
soil at t5.6 s and Fig. 29 shows the time histories of s3
at point E (refer to Fig. 8 for its location) in analysis 3a. Fig. 30. Failure zones in improved soil (analysis 3b): (a) t6 s and (b)
Although the improved soil is assumed to have high t20 s
strength (qu5 MPa) in case 3, the stress becomes greater
than the tensile strength (1 MPa) in a few instances dur-
ing the earthquake. These results indicate that the im- at the corners of the grid. The damage volume ratio
proved soil specied in case 3 partially fails during a large becomes about 2.5z nally. The failure region in case 3
earthquake. (qu5 MPa) is much smaller than that in case 2 (qu2
Figure 30 shows the tensile failure zones in the im- MPa) ( see Fig. 26), suggesting that the failure region can
proved soil during the earthquake in analysis 3b. At t6 be reduced by increasing the strength of the improved
s, failure occurs at the bottom and upper corners of the soil.
improved soil grid. At this moment, the damage volume Figure 31 shows the excess pore pressure time histories
ratio is about 1.5z. Thereafter, at t20 s, when the in the unimproved sand deposit in analyses 3a and 3b at
shaking nishes, the failure zones extend to some extent point D (refer to Fig. 8 for its location). The excess pore
572 NAMIKAWA ET AL.

Fig. 31. Excess pore pressures (E.P.P.) in unimproved sand deposit

surrounded by improved soil grid at point D in Fig. 8

Fig. 33. Maximum displacement proles along lines C and F in Fig. 8

improved ground spaced at 4 m4 m (case 1) has the

highest potential for the liquefaction mitigation. The
excess pore pressure in the sand deposit decreases with the
increase in the strength of the improved soil when the
spacing of the improved soil wall is the same (cases 2 and
3). Figures 27 and 31 show that partial damage of the
improved soil has insignicant inuence on the increase
in the excess pore pressure in the unimproved sand
deposit surrounded by the grid. Therefore, the dierence
in excess pore pressure between cases 2 and 3 is mainly
attributable to the dierence of the elastic modulus of the
improved soil.
Ground displacement is important for design of piles
Fig. 32. Proles of maximum excess pore pressure ratios along lines C that support a structure. The proles of the maximum
and F in Fig. 8
ground displacements at the center of the improved soil
grid along lines C and F (refer to Fig. 8 for their location)
pressure in analysis 3b is almost identical to that in are shown in Fig. 33. At the free eld, large displacement
analysis 3a during the earthquake. In case 3, because the is induced at the layer in which liquefaction occurred
failure region does not extend considerably in the during the earthquake. In the unimproved sand deposit
improved soil during the earthquake, its partial damage surrounded by the grid, on the other hand, no marked
only slightly aects the potential of the ground improve- change is apparent in the displacement proles. The
ment for the remedy against liquefaction. displacement with the improved soil grid spaced at 4.0 m
4.0 m (case 1) is the smallest in all cases. The ground
displacement decreases with the increase in the strength
COMPARISON OF RESULTS OF ANALYSES of the improved soil when the spacing of the improved
Three numerical results (analyses 1a, 2b and 3b) are soil is the same (cases 2 and 3).
compared to elucidate the eects of specications of the The depth at which the larger displacement occurred at
improved ground on the potential for liquefaction miti- the free eld corresponds to the depth at which the
gation. Numerical results indicate that partial failure is improved soil wall normal to the direction of excitation
induced in the improved soil specied in cases 2 and 3. failed horizontally in analysis 2b ( see Fig. 26). The
However, because eects of such failures are considered bending moment around the z axis induced by the large
directly in analyses 2b and 3b, we can compare the soil displacement of the surrounding unimproved soil may
layer responses obtained from these numerical analyses. have damaged the improved soil wall.
The proles of the maximum excess pore pressure These comparisons suggest that the treatment area
ratios along lines C and F (refer to Fig. 8 for their loca- ratio is the most eective parameter for increasing the
tion) are shown in Fig. 32. The excess pore pressure ratio potential of liquefaction mitigation. The elastic modulus
at the free eld reaches 1.0 at the depth of 4 m. The of the improved soil also aects the potential of liquefac-
 FE-ANALYSIS OF GROUND IMPROVEMENT 573

tion mitigation. Moreover, the above results show that, grids does not increase remarkably.
with the partial failure of the improved soil, the displace-
ment of the unimproved sand deposit surrounded by the
DISCUSSIONS
Residual Potential of Damaged Ground Improvement
for Liquefaction Mitigation
Numerical results for the improved soil spaced at 8 m
8 m showed that, although the improved soil partially
fails during a large earthquake, it can prevent the unim-
proved sand deposit from liquefaction. A question then
arises as to whether or not such damaged improved soil
maintains its liquefaction mitigation capability after such
an earthquake. Therefore, additional analyses were
conducted to examine the residual potential of the
damaged, improved soil for liquefaction mitigation.
In the additional analyses, it was assumed that the
second earthquake occurs long after the rst one in cases
2 and 3. Taking account of the partial failure of the
improved soil during the rst earthquake, the internal
variable denoting the damage (Namikawa and Mihira,
2007) calculated in analyses 2b and 3b were used as the
initial value in the analyses for the second earthquake.
Other analytical conditions for the second earthquake,
including the input motion, are identical to those in
analyses 2b and 3b. The residual stress and strain
Fig. 34. Proles of maximum excess pore pressure ratios along line F accumulated during the rst earthquake were not consid-
in Fig. 8 ered in the analyses for the second earthquake.

Fig. 35. Distribution of strain in improved soil at t20 s: (a) Analysis 2b and (b) Analysis 2b second
574 NAMIKAWA ET AL.

broadly similar to those in the case history in the Hyogo-

ken Nanbu earthquake in 1995 (Fig. 1). Therefore, the
results in analysis 3b provide a good understanding of the
inspection results from this case history. In analysis 3b,
the excess pore pressure ratio in the unimproved sand
deposits surrounded by the improved soil grids does not
reach 1.0 ( see Fig. 32), suggesting no occurrence of
liquefaction. This numerical results support the inspec-
tion results that no evidence of liquefaction was found on
the surface of the unimproved soil surrounded by the
improved soil grids.
The numerical results show that the improved soil grids
in case 3 partially fail during the earthquake ( see Fig. 30).
Conversely, no crack at the surface of the improved soil
grid was observed in the case history. This dierence may
be attributed to the diculty in detecting cracks at the
damage regions by visual observation.
In the laboratory, the authors observed progressive
Fig. 36. e3 in improved soil at point E in Fig. 8: (a) Case 2 and (b)
Case 3 formation of a tensile crack with a microscope during
three point bending tests with a notch (Namikawa and
Koseki, 2005). This microscopic observation suggested
Proles of the maximum excess pore pressure ratios that, during the softening process, the tensile crack
along lines C and F (refer to Fig. 8 for their location) are can not be observed visually just after the peak stress, but
shown in Fig. 34. In case 2, the excess pore pressure at the after the development of the large strain. In the bending
second earthquake (denoted as analysis 2b second) is tests, the visible tensile crack has not appeared until the
larger than that at the rst earthquake (analysis 2b). load dropped to as little as 50z of the peak load, when
This behavior indicates that the restraint eect of the the opening crack displacement was about 0.05 mm
improved soil is reduced by the partial failure induced (Namikawa and Koseki, 2006).
during the rst earthquake. Conversely, in case 3, the In the FE-analyses, the element size of the improved
excess pore pressure at the second earthquake (analysis 3b soil part is 200300400 mm. Thus the tensile strain
second) is only slightly larger than that at the rst earth- corresponding to the above visible tensile crack can be
quake (analysis 3b). The reduction in the restraint eect roughly evaluated as 0.05 mm W 400 mm0.0125z. The
of the improved soil is negligible because the failure absolute value of the minimum residual strain calculated
region does not extend markedly during the rst earth- in analysis 3b is smaller than this value ( see Fig. 36(b)),
quake. indicating that it might be dicult to nd cracks visually
Figure 35 shows distributions of the resultant tensile under such a failure condition.
strain in the improved soil after each earthquake in case These laboratory and numerical results imply that, in
2. The resultant tensile strain at the second earthquake is the above case history, although the stress at the corner of
larger than that at the rst one. The time histories of the the improved soil grids might reach the tensile strength of
minor principal strain e3 at point E (refer to Fig. 8 for its the improved soil during the earthquake, the amount of
location) are shown in Fig. 36. In case 2, the resultant residual tensile strain was so small that no crack was
tensile strain at the second earthquake is approximately observed at the surface of the improved soil after the
four times as large as that at the rst one. On the other earthquake. Therefore, it would be reasonable to assume
hand, the resultant tensile strains in case 3 are much that the partial failure of the improved soil occurred in
smaller than those in case 2; the resultant tensile strain at the case history. However, this damage might be so small
the second earthquake in case 3 is only slightly larger than that the liquefaction mitigation potential of the improved
that at the rst one. ground deteriorated hardly, as was the case with case 3 in
These additional numerical results indicate that the the present study.
liquefaction mitigation potential of the improved soil is It should be noted that, in the case history, the
not reduced remarkably by partial damage caused by a improved ground experienced the Hyogo-ken Nanbu
preceding earthquake. In particular, if the induced earthquake twenty months after it was constructed. On
failure region is small (such as the case with case 3 in the other hand, in case 3, the unconned compression
which the damage volume ratio is about 2.5z at the rst strength qu5 MPa was chosen by referring to the sam-
earthquake), the damaged improved soil can maintain ple strengths after curing for about six weeks. At the time
practically the same potential of liquefaction mitigation. of the Hyogo-ken Nanbu earthquake, the strength of the
improved soil might be lager than the value of 5 MPa
Interpretation of Inspection Results in Case History owing to the eect of ageing. Thus, it is also possible that
Based on Numerical Analysis Results the improved soil grid was scarcely damaged in the case
The ground improvement specications in case 3 are history.
 FE-ANALYSIS OF GROUND IMPROVEMENT 575

Table 4. Summary of results of numerical analyses

Liquefaction potential Internal stability of Deterioration of

Grid space qu of unimproved sand improved soil performance after
deposit large earthquake

Case 1 4 m4 m 2 MPa Acceptable Acceptable

Case 2 8 m8 m 2 MPa Acceptable Unacceptable Appreciable

Case 3 8 m8 m 5 MPa Acceptable Unacceptable Negligible

qu: Unconned compression strength

Signicance of Results of Numerical Analyses in Design improvement by cement-mixing for liquefaction mitiga-
Procedures tion. Finite element analyses were conducted for three
The present numerical analysis results are summarized cases to examine the eect of dimensions and strength of
in Table 4. They suggest that, although the improved soil the improved ground on the potential for liquefaction
grids can prevent the unimproved soil from liquefaction mitigation. In these analyses, the behavior of the im-
in all the cases, the internal stability is not satised in proved soil was modeled using both elastic and elasto-
cases 2 and 3. plastic models, where the elasto-plastic model can
In the allowable stress-based design procedure, we describe the post-peak stress-strain behavior of cement-
need to ensure both the liquefaction potential of the treated soils.
unimproved soil and the internal stability of the Comparisons of the results of the analysis suggest that
improved soil grids. Then, only the improved soil spaced the increase in the improvement area ratio is particularly
at 4 m4 m (case 1) is acceptable as a countermeasure to eective in increasing the potential of the improved
liquefaction during a large earthquake. ground for liquefaction mitigation. The elastic modulus
Conversely, in the performance-based design, the of the cement-treated soil also aects its potential for
specication of the improved ground is determined from liquefaction mitigation.
the unimproved soil behavior surrounded by the im- Numerical analyses using the elasto-plastic model for
proved soil walls. In analyses 2b and 3b, the eects of the the improved soil show that, although the improved soil
partial failure in the improved soil are simulated directly spaced at 8 m8 m fails partially during a large earth-
using the elasto-plastic model. Therefore, the proles of quake, it can prevent liquefaction in the unimproved
excess pore pressures (Fig. 32) and ground displacements sand deposits surrounded by the improved soil grids.
(Fig. 33) obtained from these numerical analyses can be Since the eect of such partial damage of the improved
used for the design of structure foundations. Since the soil is simulated appropriately in these analyses, the
potential of liquefaction mitigation are ensured in all the obtained excess pore pressures and ground displacements
cases, the improved soil spaced at 8 m8 m (cases 2 and can be used directly in the design procedure. By using this
3), in which the partial failure occurs, may be accepted in method, required performances, such as ensuring the
the performance-based design. In addition, if the per- potential of liquefaction mitigation at single large
formance in terms of the liquefaction mitigation poten- earthquake or multiple large earthquakes, can be veried,
tial shall be maintained even after a large earthquake, and more rational design of lattice-shaped ground im-
limited extent of the induced failure region such as the provement by cement-mixing can be implemented.
case with case 3 needs to be ensured.
In a practical design, we should consider the inuence
of superstructures built on the improved ground, which REFERENCES
was not considered in this study. When there are piles 1) Architectural Institute of Japan (1988): Recommendations for
supporting the superstructure in the soil deposit sur- Design of Building Foundations (in Japanese).
2) Committee of Earthquake Observation and Research in the Kansai
rounded by the improved soil grid, the inertia force of the Area (CEORKA) (1995): Digitized Strong Motion Records in the
superstructure is exerted on the soil deposit from the Aected Area during the 1995 Hyogoken-Nambu Earthquake.
piles. In such a condition, the excess pore pressure 3) Kokusho, T. (1980): Cyclic triaxial test of dynamic soil properties
induced in the sand deposit surrounding the piles and the for wide strain range, Soils and Foundations, 20(2), 4560.
stress occurring in the improved soil grids might be 4) Kokusho, T., Yoshida, Y. and Esashi, Y. (1982): Dynamic
properties of soft clay for wide strain range, Soils and Foundations,
dierent from those in the condition without the super- 22(4), 118.
structure. An analysis involving the superstructure would 5) Matsuo, O. (2002): Determination of design parameters for deep
be required to evaluate precisely the potential of liquefac- mixing, Tokyo Workshop 2002 on Deep Mixing, 7579.
tion mitigation of the improved ground. 6) Mihira, S., Koseki, J., Sato, T., Namikawa, T. and Yoshizawa, M.
(2003): Triaxial compression and tension tests on deformation and
strength properties of cement-treated sand, Proc. 38th Annu. Conf.
CONCLUSIONS Jpn. Geotech. Soc., 875876 (in Japanese).
7) Namikawa, T. and Koseki, J. (2005): Post peak strain softening
This study presents results of the numerical investiga- properties of cement-treated sand, Proc. Symp. Cement-treated
tion into the potential of the lattice-shaped ground Soils, JGS, 357364 (in Japanese).
576 NAMIKAWA ET AL.

8) Namikawa, T. and Koseki, J. (2006): Experimental determination prediction for model No. 1. Verication of Numerical Procedures
of softening relations for cement-treated sand, Soils and Founda- for the Analysis of Soil Liquefaction Problems (eds. by Arulanan-
tions, 46(4), 491504. dan & Scott), Balkema, 213219.
9) Namikawa, T. and Mihira, S. (2007): Elasto-plastic model for 14) Suzuki, Y., Saito, S., Onimaru, S., Kimura, T., Uchida, A. and
cement-treated sand, Int. J. Num. Anal. Meth. Geomech., 31(1), Okumura, R. (1996a): Grid-shaped stabilized ground improved by
71107. deep cement mixing method against liquefaction for a building
10) Public Works Research Center (1999): Technical Manual on Deep foundation, Tsuchi-to-Kiso, JGS, 44(3), 4648 (in Japanese).
Mixing Method for Constructions on Land (in Japanese). 15) Suzuki, Y., Onimaru, S. and Uchida, A. (1996b): A study on eect
11) Public Works Research Institute (1999): Design and Construction of grid-shaped stabilized ground during 1995 Hyogo-ken Nanbu
Manual of Countermeasures against Liquefaction (in Japanese). Earthquake, Summaries of Technical Papers of Annual Meeting,
12) Rollins, K. M., Evans, M. D., Diehl, N. B. and Daily III, W. D. Architectural Institute of Japan, 629630 (in Japanese).
(1998): Shear modulus and damping relationships for gravels, J. 16) Tokimatsu, K., Mizuno, H. and Kakurai, M. (1996): Building
Geotech. Geoenviron. Engrg., 124(5), 396405. damage associated with geotechnical problems, Special Issue of
13) Shiomi, T., Shigeno, Y. and Zienkiewicz, O. C. (1993): Numerical Soils and Foundations, 219234.