INTERNATIONAL CONFERENCE

ON GEOTECHNICAL ENGINEERING

Soil-Structure
Interaction.
Underground
Structures and
Retaining Walls.

16-18 June 2014 SAINT PETERSBURG

SIMSG ISSMGE
PSEUDO-STATIC ANALYSIS OF FLEXIBLE RETAINING
STRUCTURES INCLUDING SEISMIC THRUST DEPENDENCY
ON WALL DEFORMATION

Bruno Becci & Marco Carni

Ce.A.S. s.r.l. Centro di Analisi Strutturale- Milano, Italy

ABSTRACT: The design of flexible retaining structures is very frequently conducted using simplified soil-
structure interaction schemes such as the well known non-linear spring method. With this method, several
distinguishing aspects of flexible wall behaviour can be usually captured with reasonable accuracy with regards
to construction stages and static conditions. Seismic design, however, still represents a challenging aspect. In this
paper some common issues of seismic analysis with the spring method are addressed and a proposal to include
seismic effects within a pseudo-static framework is presented. Some examples analyzed with the proposed
approach are finally discussed.

1. INTRODUCTION within selected lower bound thrusts (usually the

Mononobe-Okabe (M-O in the following),
The design of flexible retaining walls is usually Mononobe & Matsuo (1929), Okabe (1926))
based on numerical analyses, aiming at model- and an upper bound distribution (to be applied
ling real construction sequence, in particular to non yielding walls). Some examples taken
when several supports are included. One of the from available experimental or numerical
most used numerical approaches in this field is benchmarks are reproduced by this approach.
the non-linear spring method, a natural exten- The results and limitations of this proposal will
sion of the Winkler method traditionally used be finally discussed.
by practicing engineers in the design of founda-
tion beams and slabs. Even today, in spite of the 2. THE NON-LINEAR SPRING METHOD
wide availability of advanced modeling tools,
such as 2D and 3D finite element or finite Within a structural or finite element program,
difference codes, the non-linear spring method the wall is modeled by an array of elastic or
is very frequently adopted, at least in the early inelastic BEAM elements and just lateral
stages of a wall design, because, in authors' behaviour is considered. The interaction with
opinion, the merits of such simple engineering soil is modelled by special non-linear elements
method still outweigh its intrinsic shortcomings. at each face, which may be either one-
On the other hand, in seismic design, the non- dimensional simple lateral springs applied at
linear spring approach highlights severe limita- each wall node, or more refined interface
tions that may, at first glance, discourage its elements whose stiffness is calibrated by means
usage. However, once its assumptions and of continuum finite element models (Pappin et
limitations have been clearly addressed, this al. (1985)).
method may still be helpful in seismic analysis. In the constitutive model of such elements,
In this paper, the non-linear spring method as lateral stress 'h in each spring is initially set
implemented by the authors (Becci & Nova, equal to the assumed at rest pressure K0'v.
(1987)) is briefly presented. Then the main Hence the wall is initially subjected to self-
assumptions of the pseudo-static methods for balanced distribution of pressures, provided
seismic design and the so-call performance same geostatic conditions are prescribed at each
design approach are reviewed, in the light of its side. Subsequent stages are then studied in
implementation by the non-liner spring ap- which an excavation or the insertion of a prop is
proach. A simple modification to the non-linear modelled, as well as any other modification to
spring constitutive law is proposed, to automati- previous conditions. Based on lateral wall
cally include seismic actions on the wall, falling

277
deformations , soil elements react as shown in In essence such geometrical parameters are
Figure 1. set equal to the average mobilized soil wedge
'h widths in active conditions (driving side) or
KP'v passive conditions (resisting side).
The (secant) elastic modulus E' can be
determined by means of usual correlations
adopted for other geotechnical problems.
For sandy soils, a reasonable estimate can
K0'v obtained by the one of the following equations:
KA'v E' (2 3) N SPT (MPa) , (3a)

or
Figure 1. Constitutive law for a non-linear spring E' (0.8 1) G 0 , (3b)
(cohesionless soil)
For granular soils, yield limits linearly de- where G0 is the small strain shear modulus.
pend on the effective overburden stress 'v via In Becci & Nova (1987) a simplified cap
the usual active and passive thrust coefficients logic is also included in order to differentiate
KA and KP, usually determined according to loading modulus Evc from unloading one EUR.
classical Coulomb equations for KA and values Including such model in a general nonlinear
by Caquot et al. (1973) for KP, accounting for finite element code, several modeling features
wall-soil friction and dredge line slope. can be easily considered, such as stage analysis
The spring stiffness depends on the elastic by adding and/or removing parts of the soil as
properties of the soil and on the excavation well as ground anchor installation and also the
geometry. According to Becci & Nova (1985), interaction with water. At each analysis stage,
at each step, the spring stiffness K is computed balanced conditions are computed by means of
by the following equation: Newton-Raphson iterations. The straight-
forward model definition as well as the
E' t
K= , (1) simplicity in soil parameters selection makes
L such kind of procedure a quite attractive option,
where E' is the soil Young modulus, t is the at least in preliminary design phases, albeit
spring spacing and L is a geometrical parameter some important design aspects, such as
accounting for actual wall geometry. This value settlements of nearby foundations or global
is different between up- (LM) and down-hill stability checks are not directly available.
(LV) side:
3. PSEUDO-STATIC SEISMIC ANALYSIS
2 '
LM = H' tan Pseudo-static seismic analysis is a widely
3 4 2 , (2) used method used in the design of retaining
(H'D ) tan + '
2 walls. Stemming from the pioneering proposal
LV = by Mononobe & Okabe, several extensions have
3 4 2 been published during last decades, in order to
D is the excavation depth, H is the overall wall encompass most of the aspects that may arise in
height and H =min (2D,H) (Figure 2). the practice. In essence a uniform seismic
acceleration with horizontal and vertical
components is statically applied to the wall and
to the soil mass interacting with the wall. The
M-O method and its extensions offer a closed
form solution of the total soil thrust on the wall,
assuming that the soil mass behind yields
during seismic event, thus considering a wall
subjected to relevant deformations. Following
the approximation proposed by Seed &
Figure 2. Definition of geometrical scaling factor for Whitman (1970), the M-O increment to static
non-linear spring stiffness assessment

278
active pressures can be applied as a constant Eq. 5 or any other elastic method. In this re-
pressure distribution given by spect, it is clear that, for many kinds of walls,
such as multi-propped bulkheads, the perfor-
a
p E ,M O 0.375 H , (4) mance (or capacity) seismic analysis with
g pseudo-static approach requires particular care
and further research.
where H is the wall height, the average back-
Finally, within the performance design
fill unit weight and a/g the normalized seismic
framework:
acceleration. Eq. 4 strictly holds for a dry
granular fill with a friction angle '=35 and a p E , Wood a
soil-wall angle = '. 2.67 max , (6)
Other methods, including the widely used
p E , M O ac
Wood method (Wood, 1973)) provide the Therefore active seismic thrusts may drop down
seismic increment of the soil thrust on a wall to a quite low percentage of elastic thrusts: for
which is essentially rigidly restrained so as to example, taking ac/amax 0.70, according to Eq.
prohibit remarkable wall deformations. Such 6, active thrust is just 26% of elastic thrust.
increment is represented by a constant pressure
distribution given by: 4. PSEUDO-STATIC SEISMIC ANALYSIS
a BY THE NON-LINEAR SPRING
p E ,Wood = H, (5) METHOD
g
Comparing the formulations above, it can be
realized that the seismic thrust increment for 4.1. General
rigid walls (Eq. 5) may exceed the M-O incre- Once the construction process has been
ment (Eq. 4) by a factor greater than 2.5. There- completed, seismic conditions are usually
fore it is important to select the appropriate modelled according to one the two following
approach, in the light of wall behaviour under alternatives:
consideration.
Beyond the method adopted in computing A. the active and passive coefficients are
seismic pressures, a most important issue arose modified to account for seismic effects or;
even since early '70s: i.e. the selection of an
appropriate design acceleration ac with respect B. just passive coefficients are modified
to the prescribed maximum seismic acceleration whereas, on the driving side, an appropriate
amax. Thanks to the fundamental works by distribution of external pressures is applied,
Richard & Elms (1979), Whitman & Liao whose resultant corresponds with the ex-
(1985) and later works, the concept of design pected seismic thrust increment: for example
acceleration ac<amax related to selected wall Eqs. 4 or 5 may be used, depending on the
performance (acceptable deformation) rather expected wall behaviour.
than to maximum site intensity was clearly
established. Since then, most design standards Approach A is usually not recommended since
such as Eurocode comply with this approach. it is just applicable to very flexible walls, in
Research efforts have been devoted in which active conditions are reached at the end
extending this method that was originally of excavation process and in general may not
proposed for gravity walls, to other kinds of sufficiently reproduce expected seismic actions.
retaining structures, such as cantilevered or Approach B, on the contrary, allows the design-
propped bulkheads. Since the performance er to keep under his/her control the seismic
design method requires the calculation of a actions which, however, do not depend on wall
desired ultimate capacity, just rigid-plastic deformations.
methods, such as the Blum method or similar
extensions (e.g. Conti & Viggiani, (2013), 4.2. The proposed procedure
Callisto (2014)), are essentially applicable. In
other words, reduced design accelerations ac As an alternative to such conventional
given by a performance design approach must approaches, the following procedure is
be applied only if seismic increments are proposed aiming at automatically include
determined by a method like Eq. 4, but not by intermediate seismic conditions, based on actual

279
wall behavior. Consider a soil region in the A-B-C-D path corresponds with an elastic
uphill side on the retaining wall. In a very short uphill soil region in which the lateral stress was
time lagging in between seismic wave arrival released, in a previous static stage, from at rest
and the actual wall response, incremental wall conditions, yet not sufficiently to reach active
deformations may be assumed to be negligible: limit state (pt A). Initial rigid seismic stress
i.e. the wall may be considered as very rigid, for increment is represented by A-B segment.
a while. During such very transient conditions, Further stress evolution toward seismic active
the seismic thrust increment may be estimated conditions is represented by B-C-D path, in
by a rigid approach, such as the Wood method which B-C is the elastic release part, whereas
(Eq. 5). Due to such incremental pressure spike, C-D represents the development of seismic
previous equilibrium conditions are violated. plastic strains.
Therefore additional deflections must develop,
in order to reestablish equilibrium conditions, E-F-G-H path represents an uphill soil region
which are pursued by means of the usual that reached active static conditions (pt. E) prior
iterative process, exactly as in any static stage. to earthquake. Initial rigid seismic stress incre-
The proposed algorithm requires the ment (E-F) corresponds with a temporary elastic
following two modifications to the standard reloading path. Subsequent elastic strain release
calculation scheme: is represented by F-G segment, whilst seismic
plastic deformation development is associated
1. In the first iteration of the seismic to segment G-H
stage, in any uphill soil element only, the
effective lateral pressure is increased by the I-J-K path: an uphill soil element is represented,
rigid (Wood) pressure increment (Eq. 5): note which was actually pushed inward rather than
that such stress increment is assigned with no being released (>0, point I): initial rigid
strain increment, thus it can be revised as an seismic stress increment is I-J segment, whereas
inelastic increment. subsequent unloading is J-K.

2. During further iterations, strain incre- Active and passive limit seismic conditions
ments are allowed and, at the same time, active should be computed according to general
and passive limits are updated to the seismic criteria (for example using M-O equations). KP
values as discussed later. should be determined so as to account for wall-
soil friction, based on safe formulations (e.g.
Through the iterative process, the initial seismic Lancellotta (2007), Soubra (2000)).
stress increment may diminish or even vanish at
all, but the appropriate seismic yield conditions 4.3. Discussion
are ensured, at least.
According to this procedure, the computed
In figure 3 some particular stress paths for up- seismic pressures on the driving side fall be-
hill soil elements are outlined, during such tween a minimum (active) and a maximum
automatic seismic procedure stage. (Wood) distribution. This feature may be
considered the most relevant advantage of the
Passive limits
procedure because all wall typologies may be
h J analyzed without preliminary (and arbitrary)
pd assumptions on their mechanical behavior. Of
K
I course, it also suffers from some limitation,
Seismic (M-O) active limit
pd = initial (Wood) increment at
1st iteration
including the following:
B - the decrease rate from Wood to active condi-

G
F pd tions is governed by the same stiffness con-
D C
H
A h sidered in static calculation;
E - the position of the overall active seismic thrust

Static active limit
= soil element deformation (+ve for compression) is essentially the same as the static active
thrust.
In the light of the performance design meth-
Figure 3. Automatic seismic algorithm in non-linear od, on authors' opinion, this approach should be
spring model used just when the allowed unrecoverable

280
deformation are very small or even negligible, forces have been recorded at several model
i.e. when the wall is designed to behave (al- positions.
most) elastically during seismic conditions. The following sequence is reproduced:
- stage 0: at rest conditions are established;
- stage 1:a small excavation is performed down
5. EXAMPLES to strut position;
- stage 2: the strut is inserted and the excavation
is lowered down to final level (results at this
5.1. Propped bulkhead - centrifuge model stage are compared with static results in Conti
The propped wall PW2 studied by Conti et al et al (2012));
(2012) by means of a centrifuge reduced scale - stage 3: EQ1 seismic conditions are applied.
model is considered. Prototype dimensions are
summarized in Figure 4. It should be noted that a reduced value of peak
acceleration of the input motion must be usually
considered in a pseudo-static analysis, all the
more because, in this case, numerical results are
compared with the available results at the end of
seismic excitation. In this case, a pseudo-static
acceleration equal to 0.15 g is considered,
representing ~70% of maximum peak EQ1
excitation.

In Figure 5, numerical and experimental

results are compared. Bending moments and
prop forces are quite well matched by the
numerical model, whereas the predicted wall
movements are different from the measured
Figure 4. PW2 model (dimensions in [m])
ones, in both static and seismic conditions.

The following soil parameters have been con-

sidered:

D r = 42% e 0 = 0.84
d = 14.37 kN / m 3 cv = 32
0.5
'h
E vc (MPa ) = 50 E UR E vc = 2
98kPa
38 uphill side
= = 0.30
35 downhill side
Figure 5. PW2 model results comparison

In order to account for a likely reduced soil However, it should be noted that measured
density inside excavation (see discussion deflections in centrifuge tests, in initial static
below), a reduced friction angle (35) and stage, most likely suffer from a insufficient
reduced elastic moduli (50% of reference local sand densification, thus displaying some
values) are assigned to soil elements at the wall uneven initial tilting not reproduced by our
toe. numerical analysis. In this respect, in Conti
In the experimental study, the model was (2010), such issue is thoroughly addressed.
excited by a set of five contiguous wave As for seismic incremental deformations,
motions, modelling five seismic excitations it should be observed that unrecoverable defor-
(EQ1 to EQ5), with increasing intensities from mations are accumulated throughout the real
0.21 g up to 0.41 g seismic event, when applied acceleration
Thanks to a set of complex instrumentations, exceeds the threshold acceleration ac related to
wall displacements, bending moments and strut wall capacity; of course, such contribution

281
could not be directly reproduced by a pseudo- E vc (kPa ) 10000 z[m] z 20m
static analysis, even if much more advanced
numerical tools were used. Some a posteriori E vc (kPa ) 50000 z[m] z > 20m
correction to pseudo-static deformed shape may
be possibly added, depending on maximum wall E UR E vc = 1.5 = 35 = 0.7
capacity as compared with maximum seismic
acceleration, but such topic is beyond the scope Dewatering operations are reproduced by
of this paper. assigning a lower permeability coefficient to the
Nonetheless, in spite of its simplicity, the 2 m thick hydraulic plug inside excavation (Fig.
non-linear spring algorithm can predict the most 6) and computing the pore pressures by means
important results for design purposes reasonably of a simplified mono-dimensional seepage
well. However it must be noted that since in this scheme. Ground anchors are modelled as linear
case active limit conditions are reached along springs whose stiffness is given by the axial
all the wall, seismic results are poorly affected stiffness (EA) of the steel stem divided by the
by the amount of Wood pressure applied at first deformable length, set equal to the free length
seismic iteration. plus 50% of the grouted length.

5.2. A numerical exercise - seismic analysis of a

real wall
A flexible retaining wall benchmark is consid-
ered in the following, as a starting point to
perform some parametric seismic analyses with
the proposed algorithm.
The multi-anchored retaining wall described
by Schweiger (2002) is shown, in Figure 6.
Such problem, which is selected as a benchmark
by some of the most popular numerical
modelling tools, such as FLAC (Itasca (2011)
and PLAXIS (Brinkgreve et al. (2010)), is
limited to static conditions.

Figure 7. Real wall benchmark (adapted after

Brinkgreve et al. (2010))

In figure 7, computed and measured de-

formed shapes are compared, showing that the
numerical model reproduces quite well the
observed behaviour, unless in the top part where
some reduced anchor stiffness is highlighted by
the measures. This effect is likely related to a
limited upper anchor length whose foundation is
somehow disturbed by wall movements: such
aspect cannot be easily reproduced by a non-
linear spring method, whereas may be high-
lighted by a continuum model (Itasca (2011)).
Bending moments are also compared with
Figure 6. Real wall benchmark (adapted after the values reported in the cited reference. Both
Schweiger (2002)) absolute values and overall pattern are fairly in
agreement, on the safe side. Thus such non-
A non-linear spring analysis of the linear spring model can be considered a realistic
constructions stages is performed first, and the model of the observed wall behaviour during
obtained results are compared with the reported construction stages, and a valuable starting
measures. The following soil properties are point for numerical experiments including
assumed (see Schweiger (2002)): seismic conditions.

282
 In this respect, seismic experiments outlined Table 2. Rigid support reactions (kN/m)
in Table 1 are discussed.
EL EXP3 EXP4
Table 1. Numerical experiments on real wall 0.00 96 6
Initial -5.50 823 520
experiment ac/g Lateral restraints
Wood p. -11.00 1065 792
EXP1 Yes -16.80 636 558
Soft
EXP2 No
0.15
EXP3 Yes On authors' experience on similar examples,
Rigid this observation can be considered quite gen-
EXP4 No
eral, in agreement with similar findings by other
Soft lateral restraints correspond with ground authors (e.g. Psarropoulos et al (2005)).
anchors only, whereas rigid restraints are
represented by very stiff lateral supports applied This statement holds for moderate seismic
at elevations shown in Figure 8, along with conditions and for walls designed according to
ground anchor removal. Bending moments for usual criteria, with regards to the most relevant
various assumptions are also shown in Figure 8. aspects such as the length of ground anchors,
which should be normally designed to ensure
expected resistance and stiffness in both static
and seismic conditions.

6. CONCLUSIONS

A simple algorithm to allow pseudo-static

seismic analysis of flexible retaining walls with
the non-linear spring method is presented.
As most pseudo-static methods, this ap-
proach has a reasonable ability to capture
internal forces in soil, as well as in wall and
props, but fails in computing physical defor-
mations when relevant plastic deformations are
progressively accumulated during seismic
events: this circumstance is particularly evident
in case of cantilevered or singly propped walls
and must be clearly taken into account in final
design.
However, on authors' best knowledge, this
Figure 8. Numerical experiment results limitation is also present in any kind of pseudo-
static analysis, even using more advanced
With soft supports, EXP1 and EXP2 results numerical models.
are almost the same since active seismic Based on the examples studied with this ap-
conditions are easily restored so as to make the proach, the following concluding remarks can
initial Wood pressure effect negligible. be proposed:
With rigid supports, initial Wood pressures
(EXP3) significantly increase bending moments - Multi-propped walls with deformable inter-
and support forces, by ~30% in the upper part mediate supports such as ground anchors are
and ~9% in the middle part, with respect to an normally so flexible to reach active limit state
ordinary M-O analysis (EXP4). Support at driving side at the end of excavation as well
reaction differences are even larger (Table 2). as during seismic events. In this case both the
In practical problems, similar to this proposed approach and the classical M-O
example, representing a quite typical pressure distribution can be used in practical
configuration for deep excavations in urban designs.
areas, the proposed algorithm produces safer
predictions with respect to traditional simplified - For multi-propped walls with rigid supports,
analysis limited to M-O seismic pressures. active seismic pressures may be inappropriate,

283
 whereas fully elastic solution such as the one walls. Proc., Numerical Methods in Engineer-
by Wood (1973) may be too conservative. The ing Theory and Applications (NUMETA85),
proposed algorithm can provide a more rea- Balkema, Rotterdam, pp 789802.
sonable estimate of seismic conditions falling Psarropoulos P.N., Klonaris G., Gazetas G. 2005.
Seismic earth pressures on rigid and flexible re-
between lower and upper bound solutions.
taining walls, Soil Dynamics and Earthquake
Engineering, vol. 25, pp. 795809
7. ACKNOWLEDGMENT Richards, R. & Elms, D. 1979. Seismic behaviour of
gravity retaining walls, Journal of geotechnical
The authors are grateful to Prof. Giulia Viggiani Engineering Division, ASCE, Vol. 105, No.
and Dr. Riccardo Conti of Universit Tor GT4, pp. 449-464.
Vergata (Roma, Italy) for their kind assistance Schweiger, H. F. 2002. Results from numerical
in providing detailed information on their benchmark exercises in geotechnics, Proc. 5th
researches, as well as for their valuable sugges- European Conf. Numerical Methods in Ge-
tions. otechnical Engineering, Presses Ponts et
chaussees, Paris, 2002, pp 305314.
Seed, H. B. & Whitman, R. V. 1970. Design of earth
8. REFERENCES retaining structures for dynamic loads. Proceed-
ings of ASCE Special Conference on Lateral
Becci, B. & Nova, R. 1987. Un metodo di calcolo Stresses, Ground Displacement and Earth Re-
automatico per il progetto di paratie, Rivista taining Structures, Ithaca, N.Y., pp. 103147.
Italiana di Geotecnica,1,3347 (in Italian). Soubra, A. H. 2000. Static and seismic passive earth
Brinkgreve, R.B.J., Engin E., Engin H.K., 2010. pressure coefficients on rigid retaining structures
Validation of empirical formulas to derive mod- - Can. Geotech. Journal. Vol. 37, pp. 463478.
el parameters for sands. Proc. Numerical Meth- Whitman, R. V. & Liao, S. 1985. Seismic design of
ods in Geotechnical Engineering NUMGE2010, retaining walls, Miscellaneous Paper GL-85-1,
Trondheim, Norway, June 2-4 , pp. 137142 U.S. Army Engineer Waterways Experiment
Callisto, L. 2014. Capacity design of embedded Station, Vicksburg, Mississippi.
retaining structures, Gotechnique, Vol. 64, No. Wood, J. H. 1973. Earthquake Induced Soil Pressures
3, pp 204214. on Structures, Doctoral Dissertation, EERL 73-
Caquot A., Kerisel J., Absi E. 1973. Tables de bute 05, California Institute of Technology, Pasade-
et de pousse, Gautiers-Villars, Paris. na, CA.
Conti, R. 2010. Modellazione fisica e numerica del
comportamento di opere di sostegno flessibili in
condizioni sismiche. PhD thesis, Universit di
Roma Tor Vergata, Rome, Italy (in Italian).
Conti R., Madabhushi G. S. P, Viggiani G. M. B.
2012. On the behaviour of flexible
retainingwalls under seismic actions.
Gotechnique, Vol. 62, No. 12, pp 108194.
Conti, R. & Viggiani, G. 2013. A new limit
equilibrium method for the pseudostatic design
of embedded cantilevered retaining walls, Soil
Dynamics and Earthquake Engineering, Vol.50,
pp. 143150.
Itasca Consulting Group, 2011. Installation of a
Triple-Anchored Excavation Wall in Sand,
FLAC version 7.0, Example Applications
Manual, section 17.
Lancellotta, R. 2007. Lower-bound approach for
seismic passive earth resistance, Gotechnique,
Vol. 57, No. 3, pp. 319321.
Mononobe, N., and Matsuo, H. 1929. On the deter-
mination of earth pressure during earthquakes.
Proceedings of World Engineering Conference,
Vol 9.
Okabe, S. 1926. General theory of earth pressure.
Journal, Japanese Society of Civil Engineers,
Vol. 12, No. 1.
Pappin, J. W., Simpson, B., Felton, P. J., and Raison,
C. 1985. Numerical analysis of flexible retaining

284
Published by Georeconstruction Engneering Co.
St. Petersburg, Russia, 2014

1618 June 2014

Saint Petersburg, Russia
tc207ssi.org