Manual FLAC PDF
Manual FLAC PDF
Manual FLAC PDF
1 DYNAMIC ANALYSIS
1.1 Overview
* The data files in this volume are stored in the directory “ITASCA\FLAC600\Dynamic” with the
extension “.DAT.” A project file is also provided for each example. In order to run an example
and compare the results to plots in this volume, open a project file in the GIIC by clicking on the
File / Open Project menu item and selecting the project file name (with extension “.PRJ”). Click
on the Project Options icon at the top of the Project Tree Record, select Rebuild unsaved states,
and the example data file will be run and plots created.
The “equivalent-linear” method is common in earthquake engineering for modeling wave transmis-
sion in layered sites and dynamic soil-structure interaction. Since this method is widely used, and
the fully nonlinear method embodied in FLAC is not, it is worth pointing out some of the differences
between the two methods.
In the equivalent-linear method (Seed and Idriss 1969), a linear analysis is performed, with some
initial values assumed for damping ratio and shear modulus in the various regions of the model.
The maximum cyclic shear strain is recorded for each element and used to determine new values for
damping and modulus, by reference to laboratory-derived curves that relate damping ratio and secant
modulus to amplitude of cycling shear strain. Some empirical scaling factor is usually used when
relating laboratory strains to model strains. The new values of damping ratio and shear modulus are
then used in a new numerical analysis of the model. The whole process is repeated several times,
until there is no further change in properties. At this point, it is said that “strain-compatible” values
of damping and modulus have been found, and the simulation using these values is representative
of the response of the real site.
In contrast, only one run is done with a fully nonlinear method (apart from parameter studies, which
are done with both methods), because nonlinearity in the stress-strain law is followed directly by
each element as the solution marches on in time. Provided that an appropriate nonlinear law is
used, the dependence of damping and apparent modulus on strain level are automatically modeled.
Both methods have their strengths and weaknesses. The equivalent-linear method takes drastic
liberties with physics but is user-friendly and accepts laboratory results from cyclic tests directly.
The fully nonlinear method correctly represents the physics but demands more user involvement and
needs a comprehensive stress-strain model in order to reproduce some of the more subtle dynamic
phenomena. Important characteristics of the two methods are examined in Sections 1.2.1 and 1.2.2.
FLAC contains an optional form of damping, hysteretic damping, that incorporates strain-dependent
damping ratio and secant modulus functions, allowing direct comparisons between the equivalent-
linear method and the fully nonlinear method. This form of damping is described in Section 1.4.3.4.
There is a comparison between FLAC and SHAKE (a one-dimensional equivalent-linear program –
Schnabel, Lysmer and Seed 1972) in Section 1.7.2 for the case of a linear elastic, layered system,
and in Section 1.7.3 for the case of a nonlinear elastic, layered system.
2. The interference and mixing phenomena that occur between different fre-
quency components in a nonlinear material are missing from an equivalent-
linear analysis.
3. The method does not directly provide information on irreversible displace-
ments and the permanent changes that accompany liquefaction, because only
oscillatory motion is modeled. These effects may be estimated empirically,
however.
4. It is commonly accepted that, during plastic flow, the strain-increment tensor
is related to some function of the stress tensor, giving rise to the “flow rule” in
plasticity theory. However, elasticity theory (as used by the equivalent-linear
method) relates the strain tensor (not increments) to the stress tensor. Plastic
yielding, therefore, is modeled somewhat inappropriately.
5. The material constitutive model is built into the method: it consists of a stress-
strain curve in the shape of an ellipse (see Cundall 1976). Although this
pre-choice relieves the user of the need to make any decisions, the flexibility
to substitute alternative shapes is removed. However, the effects of a different
shape to the curve are partially allowed for by the iteration procedure used in the
method. It should be pointed out that a frequency-independent hysteresis curve
in the form of an ellipse is physically impossible, since the continuous change
in slope prior to reversal implies preknowledge (and rate information is not
available to the model because the model is defined as being rate-independent).
6. In the case where both shear and compressional waves are propagated through a
site, the equivalent-linear method typically treats these motions independently.
Therefore, no interaction is allowed between the two components of motion.
7. Equivalent linear methods cannot be formulated in terms of effective stresses
to allow the generation and dissipation of pore pressures during and following
earthquake shaking.
The following characteristics of the fully nonlinear method should be compared to the corresponding
points listed in Section 1.2.1.
1. The method follows any prescribed nonlinear constitutive relation. If a hyst-
eretic-type model is used and no extra damping is specified, then the damping
and tangent moduli are appropriate to the level of excitation at each point in
time and space, since these parameters are embodied in the constitutive model.
If Rayleigh or local damping is used, the associated damping coefficients
remain constant throughout shaking. Consult Section 1.4.3 for more details
on damping.
The standard practice for dynamic analysis of earth structures, and especially analyses dealing with
liquefaction, is based primarily upon the equivalent-linear method. The nonlinear numerical method
has not been applied as often in practical design. However, as more emphasis is placed on making a
reliable prediction of permanent deformations and liquefaction-induced damage of earth structures,
practical applications with nonlinear numerical codes have increased. Byrne et al. (2006) provide
an overview of the different methods used for liquefaction assessment, and discuss the benefit of the
nonlinear numerical method over the equivalent-linear method for different practical applications.
There are several publications describing applications of nonlinear numerical models for analysis
and design of earth structures subjected to seismic loading.* Many of the publications describing
nonlinear numerical models pertain to back-analyses of geotechnical case histories which recorded
large permanent ground deformations and failures of earth dams. These studies revisit analyses
previously performed with equivalent-linear models. The response of the Upper and Lower San
Fernando dams to the 1971 San Fernando earthquake is one of the most commonly sited case
histories. See Beaty and Byrne (2001) for a review of the observed response of both dams, and
an assessment of the key parameters affecting the response. An important observation from this
case history is that although the characteristics of the dams were similar, the earthquake-induced
responses were quite different. While the Upper San Fernando dam experienced large lateral
displacements of approximately 2 meters, a flow slide occurred at the upstream face and crest of
the Lower San Fernando dam some 20 to 30 seconds after the earthquake, and nearly resulted in a
catastrophic failure. Beaty and Byrne (2000) describe nonlinear numerical analyses of both dams
using FLAC, incorporating a liquefaction constitutive model based upon a total stress procedure.
The analyses directly consider the triggering of liquefaction and post-liquefaction response of
the dam material. Beaty and Byrne (2000) conclude that the total stress approach is a logical
extension of the equivalent-linear method because it incorporates both liquefaction-triggering and
residual strength charts in the approach. The approach calculates progressive liquefaction-induced
ground deformations that compare reasonably well with observed response, especially for the
Upper San Fernando dam. However, excess pore pressures are not computed directly in the total
stress approach, and Beaty and Byrne (2000) state that an effective stress analysis is warranted to
investigate the response of the Lower San Fernando dam properly.
Dawson et al. (2001) present a back-analysis of the Lower San Fernando dam based upon an
effective stress analysis with FLAC and a semi-empirical constitutive model. The constitutive
model is described as a “decoupled” effective stress model, because it generates pore pressure
directly in response to the number of shear stress cycles required to trigger liquefaction. Pore
pressures are generated incrementally in relation to the cyclic strength of the material as defined
by a cyclic strength curve. The same modeling approach is also applied to a back-analysis of the
Upper San Fernando dam as described in Inel et al. (1993).
* It is interesting to note that the proceedings of the “Geotechnical Earthquake Engineering and
Soil Dynamics IV” conference, held May 18-22, 2008 in Sacramento, California, contains more
than 20 publications that describe nonlinear numerical analysis related to geotechnical earthquake
engineering. (See Zeng et al. 2008.)
The delayed failure of the Upper San Fernando dam was also observed in the Mochikoshi tailings
dam failure, which occurred in 1978 in Izu-Ohshim-Kinkai, Japan as a result of a magnitude M7
earthquake followed by a magnitude M5.8 aftershock. Two dams failed: Dam No. 1 failed during
the main shaking, and Dam No. 2 failed approximately 24 hours after the main shock. Byrne
and Seid-Karbasi (2004) suggest that the delayed failure of Dam No. 2 may be related to the
low permeability silt layers contained within the sands of the tailings dam. These layers could
impede vertical drainage of excess pore pressures and greatly reduce stability because they cause a
water bubble to develop beneath the layers. Byrne and Seid-Karbasi performed coupled, nonlinear
effective stress analyses to evaluate the excess pore pressures and deformations that develop during
the earthquake and help assess the suggested failure mode.
Back-analyses of full-scale case histories are subject to many uncertainties with respect to material
behavior and input motions, which make it difficult to verify nonlinear numerical analyses. Con-
fidence in the accuracy of the nonlinear seismic deformation analysis is primarily subject to the
uncertainty related to the understanding of liquefaction. Mitchell (2008) lists four difficulties that
contribute to this uncertainty:
Difficulties in the constitutive modeling of liquefiable soils, in estimating the extent of
liquefaction, in determining the time at which liquefaction is triggered during shaking
and in estimating the post-liquefaction residual strength...
Centrifuge model tests are commonly used to attempt to address these difficulties, and permit
verification of nonlinear numerical models. The VELACS (Verification of Liquefaction Analysis
by Centrifuge Studies) project (Arulmoli et al. 1992) is one example that has provided experimental
data for use in the verification of nonlinear liquefaction analysis. Comparisons are typically made
in terms of excess pore pressure, acceleration and displacement time histories. Publications by
Inel et al. (1993), Byrne et al. (2003), Andrianopoulos et al. (2006) and Kutter et al. (2008)
describe different constitutive models that have been tested in FLAC by comparison to results from
centrifuge tests.
Nonlinear numerical analyses are presently being applied to provide seismic vulnerability assess-
ments and evaluate remedial measures for dam rehabilitation projects. The application of the
decoupled effective stress model to assess liquefaction potential of the Pleasant Valley dam in Cal-
ifornia is described by Roth et al. (1991). Deformation analyses using this constitutive model
helped determine a safe operating level for the reservoir, and supported the renewal of Pleasant
Valley dam’s operating license for the lower pool level. Seismic retrofitting of the Success Dam
in Southern California is being guided by a combination of deformation analysis methods, ranging
from simplified procedures based on the equivalent-linear method and limit equilibrium analyses,
to decoupled and fully coupled effective-stress analyses with FLAC. Perlea et al. (2008) provide
an overview of the analyses and remediation design. Salah-Mars et al. (2008) report the use of
nonlinear deformation analyses with FLAC as part of a probabilistic seismic-hazard analysis to
estimate the seismic hazard of the Sacramento-San Joaquin Delta levees in California.
In addition to seismic analyses for earthfill dams and levees, nonlinear numerical models have been
used to assess the seismic stability of concrete gravity dams (e.g., Bureau et al. 2005), concrete
water reservoirs (e.g., Roth et al. 2008), mechanically stabilized earth (MSE) walls (e.g., Lindquist
2008) and bridge foundations (e.g., Yegian et al. 2008). Several other applications of the fully
nonlinear method can also be found in the proceedings edited by Zeng et al. (2008).
The finite difference formulation is similar to that described in Section 1.3 in Theory and Back-
ground except that “real” masses are used at gridpoints rather than the fictitious masses used to
improve convergence speed when a static solution is required. Each triangular subzone contributes
one-third of its mass (computed from zone density and area) to each of the three associated grid-
points. The final gridpoint mass is then divided by two in the case of a quadrilateral zone that
contains two overlays. In finite-element terminology, FLAC uses lumped masses and a diagonal
mass matrix.
The calculation of critical timestep involves contributions of stiffness and mass at each degree
of freedom, so that the effects of nonuniform grids, structural members, interfaces and fluid can
be accommodated. For each triangular subzone, the following stiffness contribution (in units of
force/distance) is made from each of the three gridpoints of the subzone:
4 (Lmax )2
k= K+ G T (1.1)
3 6A
where Lmax is the maximum edge-length of the triangle, A is the area of the triangle and T is the
out-of-plane dimension, equal to 1.0 for a plane-strain analysis. Thus for the full quadrilateral zone,
the total contribution to each of the four gridpoints is the summation of those for the three triangles
meeting at the gridpoint. For example, for the northwest gridpoint (assuming two overlays, with
notation as illustrated in Figure 1.3 in Theory and Background),
2
K + 43 G (Lmax
a )
2 (Lmax
c )
2 (Lmax
d )
knw = + + T (1.2)
6 Aa Ac Ad
2
4 L
kz = K + G d T (1.3)
3 Az
where Az is the area of the rectangular zone, and Ld the length of its diagonal. Note that Eq. (1.3)
only applies in the specific case of a rectangular full-zone, and is provided for interest only; the
general form of the stiffness contribution is given by expressions similar to Eq. (1.2).
Masses are also accumulated at zone gridpoints from each triangular subzone. As an example, for
the northwest gridpoint (assuming two overlays),
ma + mc + md
Mnw = (1.4)
6
where ma , mc and md are the masses of triangles a, c and d, respectively. For the case of a rectangular
full-zone (containing four triangular subzones), the mass contributed to each gridpoint is
Mgp = mz /4 (1.5)
M
tcrit = 2 (1.6)
k
For the case of a rectangular zone, we can substitute stiffness and mass values from Eqs. (1.3) and
(1.5):
mz Az
tcrit = 2 (1.7)
4 K + 43 G L2d T
Substituting mz = Az ρT ,
Az ρ Az
tcrit = = (1.8)
Ld K + 43 G Ld Cp
where Cp is the speed of longitudinal waves. This expression is identical to that given in Section 1.3.5
in Theory and Background. However, the more general form (based on Eq. (1.6)) is used in
deriving the dynamic timestep, td , using a safety factor of 0.5 (to allow for the fact that the
calculation of timestep is an estimate only). Thus,
M 1
td = min · (1.9)
k 2
where the min() function is taken over all gridpoints and structural degrees of freedom, and is a
summation over all contributions to the gridpoint or structural degree-of-freedom. For a simple grid
consisting of only rectangular zones, the computed timestep may be verified using Eq. (1.8), noting
that td = tcrit /2. However, a more complicated model will contain unequal zones, different
materials connected to common gridpoints, structural elements, interfaces and the added stiffness
of coupled fluid. Each of these objects or conditions will contribute to the summations of Eq. (1.9),
so that the final timestep will be a combined function of all items. Note that stiff or small zones
may control the timestep chosen by FLAC, due to the min() function and the division by stiffness.
The above derivation is for plane strain; related expressions are obtained for axisymmetric analysis,
accounting for the effects of the varying “out-of-plane” thickness on masses and stiffnesses. For
zones containing only one overlay, the contribution from two subzones (instead of four) is summed
as above, but a divisor of 3 instead of 6 is used in Eqs. (1.1), (1.2) and (1.4).
If stiffness-proportional damping is used (see Section 1.4.3.1), the timestep must be reduced for
stability. Belytschko (1983) provides a formula for critical timestep, tβ , that includes the effect
of stiffness-proportional damping:
2
tβ = 1 + λ2 − λ (1.10)
ωmax
where ωmax is the highest eigenfrequency of the system, and λ is the fraction of critical damping
at this frequency. Both ωmax and λ are estimated in FLAC, since an eigenvalue solution is not
performed. The estimates are
2
ωmax = (1.11)
td
0.4 β
λ= (1.12)
td
given
where ξmin and ωmin are the damping fraction and angular frequency specified for Rayleigh damping
(see Section 1.4.3.1). The resulting value of tβ is used as the dynamic timestep if stiffness-
proportional damping is in operation.
The maximum stable timestep for dynamic analysis is determined by the largest material stiffness
and smallest zone in the model (see Eq. (1.1)). Often, the stiffness and zone size can vary widely
in a model (e.g., in the case of a finely zoned concrete structure located in a soft soil). A few zones
will then determine the critical timestep for a dynamic analysis, even though the major portion of
the model can be run at a significantly larger timestep.
A procedure known as dynamic multi-stepping is available in FLAC to reduce the computation time
required for a dynamic calculation. In this procedure, zones and gridpoints in a model are ordered
into classes of similar maximum timesteps. Each class is then run at its timestep, and information
is transferred between zones at the appropriate time.
Dynamic multi-stepping uses a local timestep for each individual gridpoint and zone. At the start of
an analysis, the grid is scanned and the local stable timestep for each gridpoint, tgp , is determined
and stored. The value of tgp depends on the size, stiffness and mass of the neighboring subzones
(as shown in Eq. (1.1)), attached structural elements and interfaces. The global timestep, tG , is
determined as the minimum of all tgp , as in the standard formulation.
Integer multipliers, Mgp , to the global timestep are then determined for each gridpoint according
to the algorithm illustrated by the flow chart in Figure 1.1. This algorithm ensures that multipliers
are powers of 2. In the current implementation, Mgp is set to 1 for nodes that are assigned a null
material model, connected to structural elements, attached to other gridpoints, or part of a quiet
boundary. All zones are then scanned, and an integer multiplier, Mz , is calculated for each zone as
the minimum of the multipliers for the four surrounding gridpoints.
Null,
attached, structure, n
n=0
quiet boundary
y y
n
2n<=Dtgp/DtG<2n+1 n=n+1 n<5
n
y
Mgp=1 Mgp=2n Mgp=32
Calculations for a zone (i.e., derivation of new stresses from surrounding gridpoint velocities and ac-
cumulation of gridpoint force sums from stress components) are only performed every Mz timesteps.
In all expressions involving a timestep, the global timestep is replaced by tG Mz .
Calculations for a gridpoint (i.e., derivation of new velocities and displacements from gridpoint
force sums) are only performed every Mgp timesteps; otherwise, the force sums are reset to zero,
which is normally done after every motion calculation. In all expressions involving a timestep, the
global timestep is replaced by tG Mgp .
The effect of the prescriptions described above is to skip calculation of selected gridpoints and
zones, thereby speeding up the overall calculation. The use of gridpoint and zone multipliers (Mgp
and Mz , respectively) ensures the following characteristics:
1. The force sum at each gridpoint is composed of component forces from each
connected zone that exist at the same point in time. The simultaneous nature
of the component forces is guaranteed by the fact that multipliers are powers
of two. Arbitrary integral multipliers would not have this characteristic.
2. Velocities seen by a zone (at the four surrounding gridpoints) are not updated
between zone updates. This is guaranteed by the fact that the zone multiplier is
the minimum of the surrounding gridpoint multipliers. Since stress increments
are derived from strain and displacement increments, the displacement contri-
bution of a gridpoint is felt by a zone at each update, even though the gridpoint
is updated less frequently than the zone. In essence, the total displacement
increment of the gridpoint is divided into Mgp /Mz equal parts.
This scheme is accurate for dynamic simulations that represent waves with frequencies well below
the natural frequencies of individual elements. The condition is usually guaranteed by the wave-
length criterion described by Eq. (1.29). For higher frequencies, it is believed that inaccuracies
arise from the fact that velocities used in computing strain increments are not defined (in time) at
the center of the time interval, t, for the case of a zone multiplier being unequal to the gridpoint
multiplier. This represents a departure from the second-order accuracy of the central difference
scheme used in FLAC. However, it is always possible to assess the accuracy of the scheme for
any part of the simulation by running a short period of the simulation with and without dynamic
multi-stepping. The results may be directly compared.
Dynamic multi-stepping is invoked with the command SET multi on. The effect of dynamic multi-
stepping on calculation speed is model-dependent (i.e., the more zones that have a high multiplier,
the greater the increase in speed). Although multi-stepping is not implemented within structural
elements, substantial savings can still be obtained by using multi-stepping for a system in which stiff
structures are connected to soft continuum elements. In a typical system, only a small proportion
of computer time is spent in structural calculations, so there is only a small penalty for performing
these calculations at every timestep, compared to the savings obtained by performing infrequent
grid calculations.
Example 1.1 illustrates the effect of dynamic multi-stepping. The model consists of a “wall” of
material with a modulus 20 times greater than the surrounding soil material. A shear wave is
applied at the base of the model for a 1 second time period. With SET multi on, the wall zones have
a multiplier of 1 and the soil zones have a multiplier of 4. (The gridpoint and zone multipliers are
stored in separate FISH extra variables for monitoring.) The calculation is 2.25 times faster with
dynamic multi-stepping. Velocity histories monitored at the base of the model and top of the wall
are identical with and without multi-stepping. Figure 1.2 plots the histories for the multi-stepping
run.
There is no direct printout of the multi-stepping multipliers, but FISH intrinsics zmsmul and gmsmul
(see Section 2.5.3 in the FISH volume) may be used to determine the multipliers used during cycling.
Dynamic multi-stepping can be used with structural elements. The grid timestep multipliers are
set to 1 for all gridpoints connected to structural nodes. Multipliers are not used in structures; their
natural timestep is used. This timestep may be small, but if the grid not attached to the structure does
have a large natural timestep, these gridpoints will have large multipliers, thus saving execution
time.
A user-defined integer multiplier can be specified with the optional max keyword.
For additional information and example applications of dynamic multi-stepping, see Unterberger,
Cundall and Zettler (1997). The application of dynamic multi-stepping in numerical predictions
of vibrations caused by rail traffic in tunnels is presented in Unterberger, Hochgatterer and Poisel
(1996) and Daller, Unterberger and Hochgatterer (1996).
Example 1.1 Shear wave applied to a stiff wall in a soft soil – with dynamic multi-stepping
;--- Test multistepping option ---
; ... model has a stiff retaining wall
conf dyn ext=5
grid 40 20
mod elas
prop dens 2000 bulk 2e8 shea 1e8
model null i=1,10 j=11,20
prop bulk 4e9 shear 2e9 i=11,12 j=11,20 ; 20 times stiffness
fix y i=1
fix y i=41
def setup
freq = 1.0
omega = 2.0 * pi * freq
end
setup
def wave
wave = sin(omega*dytime)
end
apply xvel=1 hist=wave j=1
apply yvel=0 j=1
hist xvel i 11 j 21
hist yvel i 11 j 21
hist xvel i 11 j 1
hist dytime
set ncw=50
set multi=on ; Comment out this line, and compare times & histories
def tim
tim = 0.01 * (clock - old_time)
end
cyc 1
def qqq ; Save multipliers in ex_1 and ex_2 - for interest
loop i (1,izones)
loop j (1,jzones)
ex_1(i,j) = zmsmul(i,j)
endLoop
endLoop
loop i (1,igp)
loop j (1,jgp)
ex_2(i,j) = gmsmul(i,j)
endLoop
endLoop
old_time = clock
end
qqq
solve dytime 1.0
; pri ex_1 zon ; (look at multipliers)
; pri ex_2
save dyn_ms.sav
JOB TITLE : .
LEGEND
-1.000
-1.500
1 2 3 4 5 6 7 8 9
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.2 Velocities at model base (i = 11, j = 1), and top of wall (i = 11, j =
21)
There are three aspects that the user should consider when preparing a FLAC model for a dynamic
analysis: (1) dynamic loading and boundary conditions; (2) mechanical damping; and (3) wave
transmission through the model. This section provides guidance on addressing each aspect when
preparing a FLAC data file for dynamic analysis. Sections 1.5 and 1.6 illustrate the use of most of
the features discussed here.
FLAC models a region of material subjected to external and/or internal dynamic loading by applying
a dynamic input boundary condition at either the model boundary or at internal gridpoints. Wave
reflections at model boundaries are minimized by specifying either quiet (viscous), free-field or
three-dimensional radiation-damping boundary conditions. The types of dynamic loading and
boundary conditions are shown schematically in Figure 1.3; each condition is discussed in the
following sections.
1.4.1.1 Application of Dynamic Input
In FLAC, the dynamic input can be applied in one of the following ways:
(a) an acceleration history;
(b) a velocity history;
(c) a stress (or pressure) history; or
(d) a force history.
Dynamic input is usually applied to the model boundaries with the APPLY command. Accelerations,
velocities and forces can also be applied to interior gridpoints by using the INTERIOR command.
Note that the free-field boundary, shown in Figure 1.3, is not required if the only dynamic source
is within the model (see Section 1.4.1.4).
The history function for the input is treated as a multiplier on the value specified with the APPLY or
INTERIOR command. The history multiplier is assigned with the hist keyword and can be in one of
three forms:
(1) a table defined by the TABLE command;
(2) a history defined by the HISTORY command; or
(3) a FISH function.
With TABLE input, the multiplier values and corresponding time values are entered as individual
pairs of numbers in the specified table; the first number of each pair is assumed to be a value of
dynamic time. The time intervals between successive table entries need not be the same for all
entries. Note that the use of tables to provide dynamic multipliers can be quite inefficient compared
to the other two options. When using the HISTORY command to derive the history multiplier, the
values stored in the specified history are assumed to be spaced at constant intervals of dynamic
time. The interval is contained in the data file that is input with the HISTORY read command and
associated with a particular history number. If a FISH function is used to provide the multiplier,
the function must access dynamic time within the function, using the FLAC scalar variable dytime,
and compute a multiplier value that corresponds to this time. Example 1.9 provides an example of
dynamic loading derived from a FISH function.
Dynamic input can be applied either in the x- or y-direction corresponding to the xy-axes for the
model, or in the normal and shear directions to the model boundary. Certain boundary conditions
cannot be mixed at the same boundary segment (see Table 1.3 in the Command Reference for a
summary of the compatibility of boundary conditions).
One restriction when applying velocity or acceleration input to model boundaries is that these
boundary conditions cannot be applied along the same boundary as a quiet (viscous) boundary
condition (compare Figure 1.3(a) to Figure 1.3(b)), because the effect of the quiet boundary would
be nullified. See Section 1.4.1.3 for a description of quiet boundaries. To input seismic motion at
a quiet boundary, a stress boundary condition is used (i.e., a velocity record is transformed into a
stress record and applied to a quiet boundary).
A velocity wave may be converted to a stress wave using the formula
σn = 2(ρ Cp ) vn (1.14)
or
σs = 2(ρ Cs ) vs (1.15)
3-D
damping
structure
quiet boundary
quiet boundary
internal
free field
free field
dynamic
input
quiet boundary
3-D
damping
structure
quiet boundary
quiet boundary
internal
free field
free field
dynamic
input
Cp is given by
K + 4G/3
Cp = (1.16)
ρ
and Cs is given by
Cs = G/ρ (1.17)
The formulae assume plane-wave conditions. The factor of two in Eqs. (1.14) and (1.15) accounts
for the fact that the applied stress must be double that observed in an infinite medium, since half the
input energy is absorbed by the viscous boundary. The formulation is similar to that of Joyner and
Chen (1975). To illustrate wave input at a quiet boundary, consider Example 1.2, in which a pulse is
applied as a stress history to the bottom of a vertical, 50 m high column. The bottom of the column
is declared “quiet” in both horizontal directions, and the top is free. The properties are chosen such
that the shear wave speed is 100 m/sec, and the product, ρCs , is 105 . The amplitude of the stress
pulse is set, therefore, to 2 × 105 , according to Eq. (1.14), in order to generate a velocity amplitude
of 1 m/sec in the column. Figure 1.4 shows time histories of x-velocity at the base, middle and top
of the column; the amplitude of the outgoing wave is seen to be 1 m/sec, as expected. The first three
pulses in Figure 1.4 correspond, in order, to the outgoing waves at the base, middle and top. The
final two pulses correspond to waves reflected from the free surface, measured at the middle and
base, respectively. The velocity-doubling effect of a free surface, as well as the lack of waves after
a time of about 1.3 seconds, can be seen, which confirms that the quiet base is working correctly.
The doubling effect associated with a free surface is described in texts on elastodynamics (e.g.,
Graff 1991).
hist xvel i 1 j 1
hist xvel i 1 j 26
hist xvel i 1 j 51
hist dytime
solve dytime 1.8
save dyn_02.sav
JOB TITLE : .
LEGEND
4-Jun-08 16:34
step 930 1.600
Dynamic Time 1.8009E+00
HISTORY PLOT
Y-axis : 1.200
1 X velocity ( 1, 1)
2 X velocity ( 1, 26)
3 X velocity ( 1, 51) 0.800
X-axis :
5 Dynamic time
0.400
0.000
2 4 6 8 10 12 14 16 18
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.4 Primary and reflected waves in a bar: stress input through a quiet
boundary
If a “raw” acceleration or velocity record from a site is used as a time history, the FLAC model may
exhibit continuing velocity or residual displacements after the motion has finished. This arises from
the fact that the integral of the complete time history may not be zero. For example, the idealized
velocity wave form in Figure 1.5(a) may produce the displacement wave form in Figure 1.5(b) when
integrated. The process of “baseline correction” should be performed, although the physics of the
FLAC simulation usually will not be affected if it is not done. It is possible to determine a low
frequency wave (for example, Figure 1.5(c)) which, when added to the original history, produces a
final displacement which is zero (Figure 1.5(d)). The low frequency wave in Figure 1.5(c) can be a
polynomial or periodic function, with free parameters that are adjusted to give the desired results.
An example baseline correction function of this type is given as a FISH function in Example 1.25
(see Section 1.6.1.4).
Baseline correction usually applies only to complex wave forms derived, for example, from field
measurements. When using a simple, synthetic wave form, it is easy to arrange the process of
generating the synthetic wave form to ensure that the final displacement is zero. Normally, in
seismic analysis, the input wave is an acceleration record. A baseline-correction procedure can
be used to force both the final velocity and displacement to be zero. Earthquake engineering texts
should be consulted for standard baseline correction procedures.
velocity
time
(a) velocity history
displacement
time
(b) displacement history
velocity
time
(c) low frequency velocity wave
displacement
time
(d) resultant displacement history
An alternative to baseline correction of the input record is to apply a displacement shift at the end of
the calculation if there is a residual displacement of the entire model. This can be done by applying
a fixed velocity to the mesh to reduce the residual displacement to zero. This action will not affect
the mechanics of the deformation of the model. Computer codes to perform baseline corrections
are available from several Internet site (e.g., http://nsmp.wr.usgs.gov/processing.html).
The modeling of geomechanics problems involves media which, at the scale of the analysis, are
better represented as unbounded. Deep underground excavations are normally assumed to be
surrounded by an infinite medium, while surface and near-surface structures are assumed to lie on
a half-space. Numerical methods relying on the discretization of a finite region of space require
that appropriate conditions be enforced at the artificial numerical boundaries. In static analyses,
fixed or elastic boundaries (e.g., represented by boundary-element techniques) can be realistically
placed at some distance from the region of interest. In dynamic problems, however, such boundary
conditions cause the reflection of outward propagating waves back into the model and do not allow
the necessary energy radiation. The use of a larger model can minimize the problem, since material
damping will absorb most of the energy in the waves reflected from distant boundaries. However,
this solution leads to a large computational burden. The alternative is to use quiet (or absorbing)
boundaries. Several formulations have been proposed. The viscous boundary developed by Lysmer
and Kuhlemeyer (1969) is used in FLAC. It is based on the use of independent dashpots in the
normal and shear directions at the model boundaries. The method is almost completely effective
at absorbing body waves approaching the boundary at angles of incidence greater than 30◦ . For
lower angles of incidence, or for surface waves, there is still energy absorption, but it is not perfect.
However, the scheme has the advantage that it operates in the time domain. Its effectiveness has
been demonstrated in both finite-element and finite-difference models (Kunar et al. 1977). A
variation of the technique proposed by White et al. (1977) is also widely used.
More efficient energy absorption (particularly in the case of Rayleigh waves) requires the use of
frequency-dependent elements, which can only be used in frequency-domain analyses (e.g., Lysmer
and Waas 1972). These are usually termed “consistent boundaries,” and involve the calculation of
dynamic stiffness matrices coupling all of the boundary degrees-of-freedom. Boundary element
methods may be used to derive these matrices (e.g., Wolf 1985). A comparative study of the
performance of different types of elementary, viscous and consistent boundaries was documented
by Roesset and Ettouney (1977).
The quiet-boundary scheme proposed by Lysmer and Kuhlemeyer (1969) involves dashpots attached
independently to the boundary in the normal and shear directions. The dashpots provide viscous
normal and shear tractions given by
tn = −ρ Cp vn (1.18)
ts = −ρ Cs vs
where:vn and vs are the normal and shear components of the velocity at the boundary;
ρ is the mass density; and
Cp and Cs are the p- and s-wave velocities.
These viscous terms can be introduced directly into the equations of motion of the gridpoints lying
on the boundary. A different approach, however, was implemented in FLAC, whereby the tractions
tn and ts are calculated and applied at every timestep in the same way that boundary loads are applied.
This is more convenient than the former approach, and tests have shown that the implementation
is equally effective. The only potential problem concerns numerical stability, because the viscous
forces are calculated from velocities lagging by half a timestep. In practical analyses to date, no
reduction of timestep has been required by the use of the non-reflecting boundaries. Timestep
restrictions demanded by small zones are usually more important.
Dynamic analysis starts from some in-situ condition. If a velocity boundary is used to provide
the static stress state, this boundary condition can be replaced by a quiet boundary; the boundary
reaction forces will be automatically calculated and maintained throughout the dynamic loading
phase. Note that the boundaries must not be freed before applying the quiet boundary condition;
otherwise, the reaction forces will be lost.
Care should be taken to avoid changes in static loading during the dynamic phase. For example, if
a tunnel is excavated after quiet boundaries have been specified on the bottom boundary, the whole
model will start to move upward. This is because the total gravity force no longer balances the
total reaction force at the bottom that was calculated when the boundary was changed to a quiet
one. If a stress boundary condition is applied for the static solution, a stress boundary condition of
opposite sign must also be applied over the same boundary when the quiet boundary is applied for
the dynamic phase. This will allow the correct reaction forces to be in place at the boundary for the
dynamic calculation.
Quiet boundary conditions can be applied in the x- and y-directions, or along inclined boundaries,
in the normal and shear directions, using the APPLY command with appropriate keywords (xquiet,
yquiet, nquiet or squiet). When applying quiet boundary conditions in the normal and shear di-
rections, nquiet and squiet should always be specified together. These conditions, individually, do
not account for the coupling between x- and y-directions for inclined boundaries. When using
the APPLY command to install a quiet boundary condition, it must be appreciated that the material
properties used in Eq. (1.18) are obtained from the zones immediately adjacent to the boundary.
Thus, appropriate material properties for boundary zones must be in place at the time the APPLY
command is given, in order for the correct properties of the quiet boundary to be stored.
Quiet boundaries are best-suited when the dynamic source is within a grid. Quiet boundaries should
not be used alongside boundaries of a grid when the dynamic source is applied as a boundary
condition at the top or base, because the wave energy will “leak out” of the sides. In this situation,
free-field boundaries (described below) should be applied to the sides.
1.4.1.4 Free-Field Boundaries
Numerical analysis of the seismic response of surface structures such as dams requires the dis-
cretization of a region of the material adjacent to the foundation. The seismic input is normally
represented by plane waves propagating upward through the underlying material. The boundary
conditions at the sides of the model must account for the free-field motion which would exist in
the absence of the structure. In some cases, elementary lateral boundaries may be sufficient. For
example, if only a shear wave were applied on the horizontal boundary AC, shown in Figure 1.6, it
would be possible to fix the boundary along AB and CD in the vertical direction only (see the exam-
ple in Section 1.7.4). These boundaries should be placed at distances sufficient to minimize wave
reflections and achieve free-field conditions. For soils with high material damping, this condition
can be obtained with a relatively small distance (Seed et al. 1975). However, when the material
damping is low, the required distance may lead to an impractical model. An alternative procedure is
to “enforce” the free-field motion in such a way that boundaries retain their non-reflecting properties
(i.e., outward waves originating from the structure are properly absorbed). This approach was used
in the continuum finite-difference code NESSI (Cundall et al. 1980). A technique of this type was
developed for FLAC, involving the execution of a one-dimensional free-field calculation in parallel
with the main-grid analysis.
B D
free field
free field
A C
seismic wave
Figure 1.6 Model for seismic analysis of surface structures and free-field
mesh
The lateral boundaries of the main grid are coupled to the free-field grid by viscous dashpots to
simulate a quiet boundary (see Figure 1.6), and the unbalanced forces from the free-field grid are
applied to the main-grid boundary. Both conditions are expressed in Eqs. (1.19) and (1.20), which
apply to the left-hand boundary. Similar expressions may be written for the right-hand boundary:
In this way, plane waves propagating upward suffer no distortion at the boundary because the free-
field grid supplies conditions that are identical to those in an infinite model. If the main grid is
uniform, and there is no surface structure, the lateral dashpots are not exercised because the free-
field grid executes the same motion as the main grid. However, if the main-grid motion differs
from that of the free field (due, say, to a surface structure that radiates secondary waves), then the
dashpots act to absorb energy in a manner similar to the action of quiet boundaries.
The free-field model consists of a one-dimensional “column” of unit width, simulating the behavior
of the extended medium. An explicit finite-difference method was selected for the model. The
height of the free field equals the length of the lateral boundaries. It is discretized into n elements
corresponding to the zones along the lateral boundaries of the FLAC mesh. Element masses are
lumped at the n +1 gridpoints. A linear variation of the displacement field is assumed within each
element; the elements are, therefore, in a state of uniform strain (and stress).
The following conditions are required in order to apply the free-field boundary condition:
1. The lateral boundaries of the grid must be vertical and straight.
2. The free-field boundaries may be applied to the whole grid or to a sub-grid,
starting at (1,1), with the left-hand boundary being i = 1. The right-hand
boundary corresponds to the last-encountered non-null zone, scanning along
j = 1 with increasing i numbers. Any other disconnected sub-grids are not
considered when the free-field boundaries are created. Therefore, if sub-grids
are used in a simulation that requires free-field boundaries to the main grid,
this grid must be the “first” one (i.e., its left and bottom sides must be lines i =
1 and j = 1, respectively). The optional keyword ilimits forces the free field to
be applied on the outer i limits of the grid (as specified in the GRID command).
This keyword should be used if null zones exist on the j = 1 row of zones. It is
advisable to perform PLOT apply to verify that the free field is applied to the
correct boundary before starting a dynamic simulation.
The application of the free-field boundary is illustrated in Example 1.3. A shear-stress wave is
applied to the base of the model. Figure 1.7 shows the resulting x-velocity at the top of the model
at different locations in the free field and the main grid.
JOB TITLE : .
21-Jul-08 10:19
1.000
step 1005
Dynamic Time 2.0016E-02
0.200
0.000
2 4 6 8 10 12 14 16 18
-03
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
A vibrating structure located on the surface of the modeled region creates a disturbance both in
the plane of analysis and in the out-of-plane direction. The energy radiated in-plane is reasonably
absorbed by the quiet boundary condition. However, in a three-dimensional system, energy would be
radiated in the out-of-plane direction. To represent this effect approximately, dashpots are connected
from all gridpoints in the main grid to corresponding gridpoints in the free field (although the force is
not applied to the free-field grid). This mechanism is termed three-dimensional radiation damping
and is invoked by the SET 3d damp command. The 3D damper acts on the difference between the
actual particle velocity under the structure and the free-field velocity around the model region. The
scheme is identical to that described by Lysmer et al. (1975). The dashpot constant, c, has the value
2 ρ Csff
c= (1.21)
W
where:c = coefficient of 3D damping;
Csff = free-field shear wave velocity; and
W = out-of-plane width of structure.
The free-field boundaries (i.e., APPLY ff) must be specified when using 3D damping. The dashpot
can be connected to either the left-hand side or the right-hand side of the free field (see Figure 1.3).
Design earthquake ground motions developed for seismic analyses are usually provided as outcrop
motions, often rock outcrop motions.* However, for FLAC analyses, seismic input must be applied
at the base of the model rather than at the ground surface as illustrated in Figure 1.8. The question
then arises, “What input motion should be applied at the base of a FLAC model in order to properly
simulate the design motion?”
The appropriate input motion at depth can be computed through a “deconvolution” analysis using
a 1D wave propagation code such as the equivalent-linear program SHAKE. This seemingly simple
analysis is often the subject of considerable confusion resulting in improper ground motion input
for FLAC models. The application of SHAKE for adapting design earthquake motions for FLAC
input is described. Two typical cases are:
1 A rigid base, where an acceleration-time history is specified at the base of the
FLAC mesh.
2 A compliant base, where a quiet (absorbing) boundary is used at the base of
the FLAC mesh.
Input of an earthquake motion into FLAC is typically done using either a “rigid base” or a “compliant
base.” For a rigid base, a time history of acceleration (or velocity or displacement) is specified for
gridpoints along the base of the mesh. While simple to use, a potential drawback of a rigid base
is that the motion at the base of the model is completely prescribed. Hence, the base acts as if it
were a fixed displacement boundary reflecting downward-propagating waves back into the model.
Thus, a rigid base is not an appropriate boundary for general application unless a large dynamic
* This section is abstracted with permission from the publication by Mejia and Dawson (2006).
impedance contrast is meant to be simulated at the base (e.g. low velocity sediments over high
velocity bedrock).
For a compliant base simulation, a quiet boundary is specified along the base of the FLAC mesh.
See Section 1.4.1.3 for a description of quiet boundaries. Note that if a history of acceleration is
recorded at a gridpoint on the quiet base, it will not necessarily match the input history. The input
stress-time history specifies the upward-propagating wave motion into the FLAC model, but the
actual motion at the base will be the superposition of the upward motion and the downward motion
reflected back from the FLAC model.
SHAKE (Schnabel et al. 1972) is a widely used 1D wave propagation code for site response
analysis. SHAKE computes the vertical propagation of shear waves through a profile of horizontal
visco-elastic layers. Within each layer, the solution to the wave equation can be expressed as the
sum of an upward-propagating wave train and a downward-propagating wave train. The SHAKE
solution is formulated in terms of these upward- and downward-propagating motions within each
layer as illustrated in Figure 1.9:
Figure 1.9 Layered system analyzed by SHAKE (layer properties are shear
modulus, G, density, ρ and damping fraction, ζ )
The relation between waves in one layer and waves in an adjacent layer can be solved by enforcing
the continuity of stresses and displacements at the interface between the layers. These well-known
relations for reflected and transmitted waves at the interface between two elastic materials (Kolsky
1963) can be expressed in terms of recursion formulas. In this way, the upward- and downward-
propagating motions in one layer can be computed from the upward and downward motions in a
neighboring layer.
To satisfy the zero shear stress condition at the free surface, the upward- and downward-propagating
motions in the top layer must be equal. Starting at the top layer, repeated use of the recursion
formulas allows the determination of a transfer function between the motions in any two layers of
the system. Thus, if the motion is specified at one layer in the system, the motion at any other layer
can be computed.
SHAKE input and output is not in terms of the upward-and downward-propagating wave trains, but
in terms of the motions at: a) the boundary between two layers, referred to as a “within” motion; or
b) at a free surface, referred to as an “outcrop” motion. The within motion is the superposition of the
upward- and downward-propagating wave trains. The outcrop motion is the motion that would occur
at a free surface at that location. Hence the outcrop motion is simply twice the upward-propagating
wave-train motion. If needed, the upward-propagating motion can be computed by taking half
the outcrop motion. At any point, the downward-propagating motion can then be computed by
subtracting the upward-propagating motion from the within motion.
The SHAKE solution is in the frequency domain, with conversion to and from the time-domain
performed with a Fourier transform. The deconvolution analysis discussed below illustrates the
application of SHAKE for a linear, elastic case. Section 1.7.2 describes a comparison between
FLAC and SHAKE for a layered, linear-elastic soil deposit. SHAKE can also address nonlinear
soil behavior approximately, through the equivalent linear approach. Analyses are run iteratively
to obtain shear modulus and damping values for each layer that are compatible with the computed
effective strain for the layer. See Section 1.7.3 for a comparison of FLAC to SHAKE for a layered,
nonlinear elastic soil.
Deconvolution for a Rigid Base – The deconvolution procedure for a rigid base is illustrated in
Figure 1.10. The goal is to determine the appropriate base input motion to FLAC such that the
target design motion is recovered at the top surface of the FLAC model. The profile modeled
consists of three 20-m thick elastic layers with shear wave velocities and densities as shown in the
figure. The SHAKE model includes the three elastic layers and an elastic half-space with the same
properties as the bottom layer. The FLAC model consists of a column of linear elastic elements. The
target earthquake is input at the top of the SHAKE column as an outcrop motion. Then, the motion at
the top of the half-space is extracted as a within motion and is applied as an acceleration-time history
to the base of the FLAC model. Mejia and Dawson (2006) show that the resulting acceleration at the
surface of the FLAC model is virtually identical to the target motion. The SHAKE within motion is
appropriate for rigid-base input because, as described above, the within motion is the actual motion
at that location, the superposition of the upward- and downward-propagating waves.
Deconvolution for a Compliant Base – The deconvolution procedure for a compliant base is il-
lustrated in Figure 1.11. The SHAKE and FLAC models are identical to those for the rigid body
exercise, except that a quiet boundary is applied to the base of the FLAC mesh. For application
through a quiet base, the upward-propagating wave motion (1/2 the outcrop motion) is extracted
from SHAKE at the top of the half-space. This acceleration-time history is integrated to obtain a ve-
locity, which is then converted to a stress history using Eq. (1.15). Again, the resulting acceleration
at the surface of the FLAC model is shown by Mejia and Dawson (2006) to be virtually identical
to the target motion. As an additional check of the computed accelerations, they also show that
the response spectra for both the compliant-base and rigid-base cases closely match the response
spectra of the target motion.
Although useful for illustrating the basic ideas behind deconvolution, the previous example is not
the typical case encountered in practice. The situation shown in Figure 1.12, where one or more
soil layers (expected to behave nonlinearly) overlay bedrock (assumed to behave linearly), is more
common. A FLAC model for this case will usually include the soil layers and an elastic base of
bedrock. To compute the correct FLAC compliant base input, a SHAKE model is constructed as
shown in the figure. The SHAKE model includes a bedrock layer equal in thickness to the elastic
base of the FLAC mesh, and an underlying elastic half-space with bedrock properties. The target
motion is input to the SHAKE model as an outcrop motion at the top of the bedrock (point A).
Designating this motion as outcrop means that the upward-propagating wave motion in the layer
directly below point A will be set equal to 1/2 the target motion. The upward-propagating motion
for input to FLAC is extracted at Point B as 1/2 the outcrop motion.
For the compliant-base case there is actually no need to include the soil layers in the SHAKE model,
as these will have no effect on the upward-propagating wave train between points A and B. In fact,
for this simple case, it is not really necessary to perform a formal deconvolution analysis, as the
upward-propagating motion at point B will be almost identical to that at point A. Apart from an
offset in time, the only differences will be due to material damping between the two points, which
will generally be small for bedrock. Thus, for this very common situation, the correct input motion
for FLAC is simply 1/2 of the target motion. (Note that the upward-propagating wave motion must
be converted to a stress-time history using Eq. (1.15), which includes a factor of 2 to account for
the stress absorbed by the viscous dashpots.)
For a rigid-base analysis, the within motion at point B is required. Since this within motion
incorporates downward-propagating waves reflected off the ground surface, the nonlinear soil layers
must be included in the SHAKE model. However, soil nonlinearity will be modeled quite differently
in FLAC and SHAKE. Thus, it is difficult to compute the appropriate FLAC input motion for a rigid-
base analysis.
Another typical case encountered in practice is illustrated in Figure 1.13. Here, the soil profile is
deep, and rather than extending the FLAC mesh all the way down to bedrock, the base of the model
ends within the soil profile. Note that the mesh must be extended to a depth below which the soil
response is essentially linear. Again, the design motion is input at the top of the bedrock (point A)
as an outcrop motion, and the upward-propagating motion for input to FLAC is extracted at point
B. As in the previous example, for a compliant-base analysis there is no need to include the soil
layers above point B in the SHAKE model. These layers have no effect on the upward-propagating
motion between points A and B. Unlike the previous case, the upward-propagating motion can be
quite different at points A and B, depending on the impedance contrast between the bedrock and
linear soil layer. Thus, it is not appropriate to skip the deconvolution analysis and use the target
motion directly.
A rigid base is only appropriate for cases with a large impedance contrast at the base of the model.
However, the use of SHAKE to compute the required input motion for a rigid base of a FLAC model
leads to a good match between the target surface motion and the surface motion computed by FLAC,
only for a model that exhibits a low level of nonlinearity. The input motion already contains the
effect of all layers above the base, because it contains the downward-propagating wave.
A different approach must be taken if a FLAC model with a rigid base is used to simulate more
realistic systems (e.g., sites that exhibit strong nonlinearity, or the effect of a surface or embedded
structure). In the first case, the real nonlinear response is not accounted for by SHAKE in its estimate
of base motion. In the second case, secondary waves from the structure will be reflected from the
rigid base, causing artificial resonance effects.
A compliant base is almost always the preferred option because downward-propagating waves are
absorbed. In this case, the quiet-base condition is selected, and only the upward-propagating wave
from SHAKE is used to compute the input stress history. By using the upward-propagating wave
only at a quiet FLAC base, no assumptions need to be made about secondary waves generated by
internal nonlinearities or structures within the grid, because the incoming wave is unaffected by
these; the outgoing wave is absorbed by the compliant base.
Although the presence of reflections from a rigid base is not always obvious in complex nonlinear
FLAC analyses, they can have a major impact on analysis results, especially when cyclic-degradation
or liquefaction-soil models are employed. Mejia and Dawson (2006) present examples that illustrate
the nonphysical wave reflections calculated in models with a rigid base. One example, shown in
Figure 1.14, demonstrates the difficulty with a rigid boundary. The nonphysical oscillations that
result from a rigid base are shown by comparison to results for a compliant base in Figure 1.15. The
inputs in both cases (rigid and compliant) were derived by deconvoluting the same surface motion.
The dynamic interaction between water in a reservoir and a concrete dam can have a significant
influence on the performance of the dam during an earthquake. Westergaard (1933) established
a mathematical basis for procedures to represent this interaction, and this approach is commonly
used in engineering practice. Although the advent of computers has enabled numerical solution
of coupled differential equations of fluid-structure systems, the formula proposed by Westergaard
is widely used for stability analysis of smaller dams, and preliminary calculations in the design of
large dams.
The hydrodynamic pressure acting on a rigid concrete dam over a reservoir height, H , is depicted
in Figure 1.16. The pressure can be derived from the equation of motion for a fluid. The equation
of motion for a fluid with small Reynold’s number can be written as
∂ 2 ∂ 2 ∂ 2
c2 + = (1.22)
∂x12 ∂x22 ∂t 2
where c is the speed of sound in water, and is the velocity potential. The water pressure can be
written as a function of the velocity potential:
∂
p = ρw (1.23)
∂t
∂
2. The free surface of the reservoir is assumed to be at rest. Thus, ∂t = 0 at x2 = H .
3. The reservoir is assumed to be infinitely long. Therefore, as x1 → ∞, → 0.
∂
4. Hydrodynamic motion is assumed to be horizontal only: ∂x2 = 0 at x2 = 0.
∂
5. The upstream face of the dam is vertical and the dam is rigid: ∂x1 = f (t) at x1 = 0.
The solution of Eq. (1.22) with the above assumptions can be obtained for an arbitrary acceleration,
ẍ0 (t), in the form of an infinite Fourier series:
∞
(−1)n+1 −(2n−1)x1 (2n − 1)x2
p(0, x2 , t) = 8ẍ0 (t)ρw H e 4H cos (1.24)
((2n − 1)π)2 4H
n=1
Cm
x22 x22
p(0, x2 , t) = ρw ẍ0 (t)H 1− 2 + 1− 2 (1.25)
2 H H
Ag
p̄(0, x2 , t) = ρec ẍ0 (t) (1.26)
x2
where Ag is the area associated with the gridpoint, and x2 is the contact length on the upstream
face of the dam through which the water load is applied for the gridpoint.
where
H x2 Cm
x2 x22
ρsc = ρw 1 − 22 + 1− (1.28)
Ag 2 H H2
ρc is the density of concrete such that the gridpoint mass is given by mg = Ag ρc . The scaled
gridpoint mass msg = Ag ρec is used only for the motion calculation in the horizontal direction; the
effect of the increased mass does not influence the vertical forces.
The gridpoint mass is adjusted by adding the term (as determined from Eq. (1.28)) to account for the
hydrodynamic pressure. The FISH gridpoint variable gmscl is available to store the gridpoint mass
adjustment. The FISH function westergaard is provided to apply the hydrodynamic pressures
to a vertical dam face. The FISH function requires the following input:
dx x1 component of the unit vector pointing in the direction
of the water
dy x2 component of the unit vector pointing in the direction
of the water
height height of the water in the reservoir
yb x2 coordinate of the base of the reservoir
den w density of water
A simple example is presented to illustrate the effect of hydrodynamic pressures on a concrete dam.
The dynamic loading is applied in three different ways. First, the dam is subjected to a dynamic
loading without taking into account the hydrodynamic pressure. Second, the hydrodynamic pressure
is applied as a boundary condition by means of the Westergaard scaling of the gridpoint mass, as
described above. Third, the hydrodynamic pressure is simulated by modeling the water directly
as zones adjacent to the dam zones. Figure 1.17 shows the model for the first two loading cases,
and Figure 1.18 shows the model for the third case. The dynamic loading is a velocity sine wave
applied to the base of the model. The models are first brought to a static equilibrium state with
the reservoir loading applied along the upstream vertical face of the dam. The dynamic loading is
then applied for a period of 10 seconds. The horizontal displacement at the top of the dam at the
upstream face is monitored for all three cases. The results are plotted for comparison in Figure 1.19.
These results illustrate the effect on displacement of the hydrodynamic pressures. The case using
the Westergaard adjustment is in good agreement with the case modeling the water explicitly.
LEGEND
5.750
21-Jul-08 9:22
step 2130
2.733E+01 <x< 5.267E+01
3.683E+01 <y< 6.217E+01
0 5E 0
4.750
4.250
7.500
FLAC (Version 6.00)
LEGEND
User-defined Groups
water 5.500
dam
Grid plot
0 1E 1
4.500
3.500
2.500
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
0.500 1.500 2.500 3.500 4.500
(*10^1)
JOB TITLE : .
LEGEND
19-Nov-07 18:01
1.500
step 0
x-displacement histories
with hydrodynamic pressure 1.000
with water zones
without hydrodynamic pressure
0.500
0.000
-0.500
-1.000
1 2 3 4 5 6 7 8 9
set tabin 2
tabtofile
save small_NO_wester_result.sav
;
new
config dyn ex 10
grid 80,60
model elastic
prop dens 2000 bulk 1e8 shear 3e7
model null j = 40
model null j = 41 60 i = 37 80
gen 0,40 0 59 30 59 30 40 i 1 31 j 41 61
gen 30.0,40.0 30.0,59.0 35.0,59.0 50.0,40.0 i 31 37 j 41 61
group ’dam’ i 31 37 j 41 61
model null i 1 30 j 41 60
set grav 10
model null j 1 40
fix x y j 41
set dyn off
solve
save water_grid.sav
initial xdisp 0 ydisp 0
initial xvel 0 yvel 0
group ’water’ i 1 29 j 41 60
model elastic group ’water’
prop density=1000.0 bulk=2E9 shear=0 group ’water’
fix x i 1
ini x add 1.0 y add 0.0 nmregion 29 41
interface 1 aside from 30,41 to 30,61 bside from 31,41 to 31,61
interface 1 unglued kn=1.4E9 ks=1.4E9 cohesion=0.0 dilation=0.0 &
friction=0.0 tbond=1e10 bslip=Off
history 999 unbalanced
solve
save water_pressure.sav
ini xdisp 0 ydisp 0 xvel 0 yvel 0
set dyn on
set step 100000000
def sine_wave
sine_wave = sin(2*pi*freq*dytime)
end
hist sine_wave
hist xdisp i 31 j 61
hist xdisp i 31 j 54
hist xdisp i 31 j 50
hist xdisp i 31 j 41
set dytime 0.0 freq 1
Numerical distortion of the propagating wave can occur in a dynamic analysis as a function of
the modeling conditions. Both the frequency content of the input wave and the wave-speed char-
acteristics of the system will affect the numerical accuracy of wave transmission. Kuhlemeyer
and Lysmer (1973) show that for accurate representation of wave transmission through a model,
the spatial element size, l, must be smaller than approximately one-tenth to one-eighth of the
wavelength associated with the highest frequency component of the input wave – i.e.,
λ
l ≤ (1.29)
10
where λ is the wavelength associated with the highest frequency component that contains appreciable
energy.
1.4.2.2 Filtering
For dynamic input with a high peak velocity and short rise-time, the Kuhlemeyer and Lysmer
requirement may necessitate a very fine spatial mesh and a corresponding small timestep. The
consequence is that reasonable analyses may be prohibitively time- and memory-consuming. In
such cases, it may be possible to adjust the input by recognizing that most of the power for the input
history is contained in lower-frequency components (e.g., use “FFT.FIS” in Section 3 in the FISH
volume). By filtering the history and removing high-frequency components, a coarser mesh may
be used without significantly affecting the results.
The filtering procedure can be accomplished with a low-pass filter routine such as the Fast Fourier
Transform technique (e.g., “FILTER.FIS” in Section 3 in the FISH volume). The unfiltered velocity
record shown in Figure 1.20 represents a typical wave form containing a very high frequency spike.
The highest frequency of this input exceeds 50 Hz but, as shown by the power spectral density plot
of Fourier amplitude versus frequency (Figure 1.21), most of the power (approximately 99%) is
made up of components of frequency 15 Hz or lower. It can be inferred, therefore, that by filtering
this velocity history with a 15 Hz low-pass filter, less than 1% of the power is lost. The input
filtered at 15 Hz is shown in Figure 1.22, and the Fourier amplitudes are plotted in Figure 1.23. The
difference in power between unfiltered and filtered input is less than 1%, while the peak velocity is
reduced 38%, and the rise time is shifted from 0.035 to 0.09 second. Analyses should be performed
with input at different levels of filtering to evaluate the influence of the filter on model results.
If a simulation is run with an input history that violates Eq. (1.29), the output will contain spurious
“ringing” (superimposed oscillations) that is nonphysical. The input spectrum must be filtered
before being applied to a FLAC grid. This limitation applies to all numerical models in which a
continuum is discretized; it is not just a characteristic of FLAC. Any discretized medium has an
upper limit to the frequencies that it can transmit, and this limit must be respected if the results
are to be meaningful. Users of FLAC commonly apply sharp pulses or step wave forms to a FLAC
grid; this is not acceptable under most circumstances, because these wave forms have spectra that
extend to infinity. It is a simple matter to apply, instead, a smooth pulse that has a limited spectrum,
as discussed above. Alternatively, artificial viscosity may be used to spread sharp wave fronts
over several zones (see Section 1.4.3.10), but this method strictly only applies to isotropic strain
components.
4
Velocity (cm/sec)
3
(Thousands)
-1
0 0.2 0.4
Time (sec)
130
120
110
100
Fourier Amplitude
90
(Times 10E9)
80
70
60
50
40
30
20
10
0
0 2 4 6 8 10 12 14 16 18 20
Frequency
3
2.8
2.6
2.4
2.2
2
Velocity (cm/sec)
1.8
(Thousands)
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
-0.2
-0.4
0 0.2 0.4
Time (sec)
130
120
110
100
90
Fourier Amplitude
80
(Times 10E9)
70
60
50
40
30
20
10
0
0 2 4 6 8 10 12 14 16 18 20
Frequency
Natural dynamic systems contain some degree of damping of the vibration energy within the system;
otherwise, the system would oscillate indefinitely when subjected to driving forces. Damping is
due, in part, to energy loss as a result of internal friction in the intact material and slippage along
interfaces, if these are present.
FLAC uses a dynamic algorithm for solution of two general classes of mechanical problems: quasi-
static and dynamic. Damping is used in the solution of both classes of problems, but quasi-static
problems require more damping for rapid convergence to equilibrium. The damping for static
solutions is discussed in Section 1.3.4 in Theory and Background.
For a dynamic analysis, the damping in the numerical simulation should reproduce in magnitude
and form the energy losses in the natural system when subjected to a dynamic loading. In soil
and rock, natural damping is mainly hysteretic (i.e., independent of frequency – see Gemant and
Jackson 1937, and Wegel and Walther 1935). It is difficult to reproduce this type of damping
numerically because of at least two problems (see Cundall 1976, and comments in Section 1.2.2).
First, many simple hysteretic functions do not damp all components equally when several wave
forms are superimposed. Second, hysteretic functions lead to path-dependence, which makes results
difficult to interpret. However, if a constitutive model that contains an adequate representation of
the hysteresis that occurs in a real material is found, then no additional damping would be necessary.
This comment is addressed to users who program their own constitutive models in the FISH language
or in C++; the built-in models are not considered to model dynamic hysteresis well enough to omit
additional damping completely.
For several reasons, it is impractical to use the “real” stress/strain response of the material in
numerical simulations. For example: (a) there are no laws that describe the complete material
response; and (b) existing laws that capture many important aspects have many material parameters,
requiring extensive calibration.
In time-domain programs, Rayleigh damping is commonly used to provide damping that is approx-
imately frequency-independent over a restricted range of frequencies. Although Rayleigh damping
embodies two viscous elements (in which the absorbed energy is dependent on frequency), the
frequency-dependent effects are arranged to cancel out at the frequencies of interest. Rayleigh
damping is described in Sections 1.4.3.1 through 1.4.3.3.
An alternative damping algorithm, hysteretic damping, is described in Sections 1.4.3.4 through
1.4.3.6. This form of damping allows strain-dependent modulus and damping functions to be
incorporated directly into the FLAC simulation. This makes it possible to make direct comparisons
between calculations made with the equivalent-linear method and a fully nonlinear method, without
making any compromises in the choice of constitutive model.
For routine engineering design, we must use an approximate representation of cyclic energy dissipa-
tion. In FLAC, when using simple plasticity models such as Mohr-Coulomb, the choice is between
Rayleigh damping and hysteretic damping. Here, we make some general comparisons between the
two approaches, to enable a choice to be made. In general, hysteretic damping is the more realistic
of the two, and it entails no reduction in timestep. For further discussion, see Section 1.4.3.11.
For low levels of cyclic strain and fairly uniform conditions, Rayleigh damping and hysteretic
damping give similar results, provided that the levels of damping set for both are consistent with
the levels of cyclic strain experienced. The results will differ in the following two circumstances:
1. When the system is nonuniform (e.g., layers of quite different properties), then cyclic
strain levels may be different in different locations and at different times. Using hysteretic
damping, these different strain levels produce realistically different damping levels in time
and space, while constant and uniform Rayleigh damping parameters can only reproduce
the average response. It would be possible to adjust the Rayleigh damping parameters to
account for spatial variations in damping using an iterative (strain-compatible) scheme,
as used in the equivalent linear method (see Section 1.2.1). It may also be possible to
adjust the Rayleigh damping parameters in time, although some practical difficulties may
be encountered.
2. As yield is approached, neither Rayleigh damping nor hysteretic damping account for
the energy dissipation of extensive yielding. Thus, irreversible strain occurs externally to
both schemes, and dissipation is represented by the yield model (e.g., Mohr-Coulomb).
Under this condition, the mass-proportional term of Rayleigh damping may inhibit yield-
ing because rigid-body motions that occur during failure modes are erroneously resisted.
Hysteretic damping may give rise to larger permanent strains in such a situation, but
this condition is usually believed to be more realistic compared to one using Rayleigh
damping.
We note that hysteretic damping provides almost no energy dissipation at very low cyclic strain
levels, which may be unrealistic. To avoid low-level oscillation, it is recommended that a small
amount (e.g., 0.2%) of stiffness-proportional Rayleigh damping be added when hysteretic damping
is used in a dynamic simulation.
Another form of damping in FLAC, the local damping embodied in FLAC ’s static solution scheme,
may be used dynamically, but with a damping coefficient appropriate to wave propagation. Local
damping in dynamic problems is useful as an approximate way to include hysteretic damping.
However, it becomes increasingly unrealistic as the complexity of the wave forms increases (i.e.,
as the number of frequency components increases). Local damping cannot properly capture the
energy loss of multiple frequency cyclic loading. Local damping is described in more detail in
Section 1.4.3.7.
A fourth form of damping, artificial viscosity, is also provided in FLAC. This damping may be
used for analyses involving sharp dynamic fronts – it is described in Section 1.4.3.10.
Rayleigh damping was originally used in the analysis of structures and elastic continua, to damp
the natural oscillation modes of the system. The equations, therefore, are expressed in matrix form.
A damping matrix, C, is used, with components proportional to the mass (M) and stiffness (K)
matrices:
C =α M +β K (1.30)
where:α = the mass-proportional damping constant; and
β = the stiffness-proportional damping constant.
The mass-proportional term is analogous to a dashpot connecting each FLAC gridpoint to “ground.”
The stiffness-proportional term is analogous to a dashpot connected across each FLAC zone (re-
sponding to the strain rate). Although both terms are frequency-dependent, an approximately
frequency-independent response can be obtained over a limited frequency range, with the appro-
priate choice of parameters, as discussed below.
For a multiple degree-of-freedom system, the critical damping ratio, ξi , at any angular frequency
of the system, ωi , can be found from (Bathe and Wilson 1976)
α + β ωi2 = 2 ωi ξi (1.31)
or
1α
ξi = + β ωi (1.32)
2 ωi
The critical damping ratio, ξi , is also known as the fraction of critical damping for mode i with
angular frequency ωi .
Figure 1.24 shows the variation of the normalized critical damping ratio with angular frequency
ωi . Three curves are given: mass components only; stiffness components only; and the sum of
both components. As shown, mass-proportional damping is dominant at lower angular-frequency
ranges, while stiffness-proportional damping dominates at higher angular frequencies. The curve
representing the sum of both components reaches a minimum at
ξmin = (α β)1/2
(1.33)
ωmin = (α / β)1/2
or
α = ξmin ωmin
(1.34)
β = ξmin / ωmin
Note that at frequency ωmin (or fmin ) (and only at that frequency), mass damping and stiffness
damping each supply half of the total damping force.
β=0
5
α= 0
4
total
ξ i / ξ min
0
0 5 10 15 20 25 30
ωi
Figure 1.24 Variation of normalized critical damping ratio with angular fre-
quency
Rayleigh damping is specified in FLAC with the parameters fmin in Hertz (cycles per second) and
ξmin , both specified with the command SET dy damp rayleigh or INITIAL dy damp rayleigh.
Stiffness-proportional damping causes a reduction in the critical timestep for the explicit solution
scheme (see Belytschko 1983). As the damping ratio corresponding to the highest natural frequency
is increased, the timestep is reduced (see Eq. (1.10)). This can result in a substantial increase in
runtimes for dynamic simulations.
In FLAC, the internal timestep calculation takes account of stiffness-proportional damping, but it is
still possible for instability to occur if the large-strain calculation is in effect (SET large) and very
large mesh deformation occurs. If this happens, it is necessary to reduce the timestep manually (via
the SET dydt command).
For the case shown in Figure 1.24, ωmin = 10 radians per second. It is evident that the damping
ratio is almost constant over at least a 3:1 frequency range (e.g., from 5 to 15). Since damping
in geologic media is commonly independent of frequency (as discussed in Section 1.4.3), ωmin is
usually chosen to lie in the center of the range of frequencies present in the numerical simulation
– either natural frequencies of the model or predominant input frequencies. Hysteretic damping is
thereby simulated in an approximate fashion.
Viscous stress increments representing the stiffness-proportional component of Rayleigh damping
are added to the total stress increments during a timestep in order to compute gridpoint forces.
Thus,
β
σv = σn + σn − σo (1.36)
t
where σv is a component of the combined stress tensor used to derive gridpoint forces, σn is the
corresponding component returned from the constitutive
law, and σo is the value of the component
prior to invoking the constitutive law. The notation denotes the vector of all components.
Viscous stresses are not included with the accumulated total stresses (σn in Eq. (1.36)). However,
the total stresses including viscous stresses (σv in Eq. (1.36)) can be printed, plotted and recorded
as histories by using the vsxx, vsyy, vszz and vsxy keywords.
A stiffness matrix is not needed in this formulation. Rayleigh damping operates directly on the
tangent modulus for the constitutive model, whether it is linear or nonlinear.
Stiffness-proportional damping is turned off when plastic failure occurs within a FLAC zone. Mass-
proportional damping, however, remains active. If excessive failure occurs in a model, the mass-
proportional term may inhibit yielding. In this case, it may be advisable to exclude Rayleigh
damping from regions of strong plastic flow (by using the INITIAL dy damp command to set Rayleigh
damping in selected regions, as described in Section 1.4.3.8).
1.4.3.2 Example Application of Rayleigh Damping
In order to demonstrate how Rayleigh damping works in FLAC, the results of the following four
damping cases can be compared; the example consists of a square grid in which gravity is suddenly
applied. The conditions are:
(a) undamped;
(b) Rayleigh damping (both mass and stiffness damping);
(c) mass damping only; and
(d) stiffness damping only.
Eq. (1.5) provides data corresponding to each case in turn. The Rayleigh parameters are adjusted
to give critical damping in cases (b), (c) and (d).
Example 1.5 Block under gravity – undamped and 3 critically damped cases
conf dy
gr 3 3
m e
prop den 1000 bu 1e8 sh .3e8
fix y j=1
set grav 10.0
hist n 1
hist ydisp i=3 j=4
hist dytime
save damp.sav
restore damp.sav
step 200
save damp1.sav
restore damp.sav
set dy_damp=rayl 1 25.0
step 445
save damp2.sav
restore damp.sav
set dy_damp=rayl 2 25.0 mass
step 80
save damp3.sav
restore damp.sav
set dy_damp=rayl 2 25.0 stiff
step 870
save damp4.sav
In the first case, with no damping, a natural frequency of oscillation of approximately 25 Hertz is
observed (see Figure 1.25). The problem should be critically damped if: (1) a fraction of critical
damping, ξmin , of 1 is specified; (2) the natural frequency of oscillation, fmin , of 25 Hertz is
specified; and (3) both mass and stiffness damping are used.
The results in Figure 1.26 show that the problem is critically damped. If only mass or stiffness
damping is used, then ξmin must be doubled to obtain critical damping (since each component
contributes one-half to the overall damping). Figures 1.27 and 1.28 again show that the system is
critically damped.
Note that the timestep is different for the three damped simulations. This is a result of the influence
of stiffness-proportional damping, as discussed above.
JOB TITLE : .
26-Sep-07 9:15
0.000
step 200
Dynamic Time 1.8898E-01
-0.600
-0.800
-1.000
2 4 6 8 10 12 14 16 18
-02
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.25 Plot of vertical displacement versus time, for gravity suddenly
applied to a square grid (no damping)
JOB TITLE : .
-0.500
26-Sep-07 9:16
step 445
Dynamic Time 7.5499E-02 -1.000
-3.000
-3.500
-4.000
-4.500
10 20 30 40 50 60 70
-03
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.26 Plot of vertical displacement versus time, for gravity suddenly
applied to a square grid (mass and stiffness damping)
JOB TITLE : .
-0.500
26-Sep-07 9:17
step 80
Dynamic Time 7.5593E-02 -1.000
-3.000
-3.500
-4.000
-4.500
10 20 30 40 50 60 70
-03
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.27 Plot of vertical displacement versus time, for gravity suddenly
applied to a square grid (mass damping only)
JOB TITLE : .
-0.500
26-Sep-07 9:17
step 870
Dynamic Time 7.5616E-02 -1.000
-3.000
-3.500
-4.000
-4.500
10 20 30 40 50 60 70
-03
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.28 Plot of vertical displacement versus time, for gravity suddenly
applied to a square grid (stiffness damping only)
Range of Predominant
Frequencies
Velocity
Spectrum
Frequency
* A spectral analysis based on a Fast Fourier Transform is supplied as a FISH function in the FISH
library in Section 3 in the FISH volume (see “FFT.FIS”). Application of “FFT.FIS” is shown in
Section 1.6.1.
If the highest predominant frequency is three times greater than the lowest predominant frequency,
then there is a 3:1 span or range that contains most of the dynamic energy in the spectrum. The idea
in dynamic analysis is to adjust fmin of the Rayleigh damping so that its 3:1 range coincides with
the range of predominant frequencies in the problem. The predominant frequencies are neither the
input frequencies nor the natural modes of the system, but a combination of both. The idea is to try
to get the right damping for the important frequencies in the problem.
For many problems, the important frequencies are related to the natural mode of oscillation of the
system. Examples of this type of problem include seismic analysis of surface structures such as
dams, or dynamic analysis of underground excavations. The fundamental frequency, f , associated
with the natural mode of oscillation of a system is
C
f = (1.37)
λ
For a continuous, elastic system (e.g., a one-dimensional elastic bar), the speed of propagation, Cp ,
for p-waves is given by Eq. (1.16), and for s-waves by Eq. (1.17). If shear motion of the bar gives
rise to the lowest natural mode, then Cs is used in the above equation; otherwise, Cp is used if
motion parallel to the axis of the bar gives rise to the lowest natural mode.
The longest wavelength (or characteristic length or fundamental wavelength) depends on boundary
conditions. Consider a solid bar of unit length with boundary conditions as shown in Figure 1.30(a).
The fundamental mode shapes for cases (1), (2) and (3) are as shown in Figure 1.30(b). If a wave-
length for the fundamental mode of a particular system cannot be estimated in this way, then a
preliminary run may be made with zero damping (for example, see Figure 1.25). A representative
natural period may be estimated from time histories of velocity or displacement. Section 1.7.1
contains a simple example in which natural periods are estimated by undamped simulations. Sec-
tion 1.6.1 describes seismic analyses in which predominant frequencies are estimated based on
both the input frequencies and the natural modes of the system.
The equivalent-linear method (see Section 1.2) has been in use for many years to calculate the
wave propagation (and response spectra) in soil and rock at sites subjected to seismic excitation.
The method does not capture directly any nonlinear effects because it assumes linearity during
the solution process; strain-dependent modulus and damping functions are only taken into account
in an average sense, in order to approximate some effects of nonlinearity (damping and material
softening). Although fully nonlinear codes such as FLAC are capable – in principle – of modeling
the correct physics, it has been difficult to convince designers and licensing authorities to accept
fully nonlinear simulations. One reason is that the constitutive models available to FLAC are
either too simple (e.g., an elastic/plastic model, which does not reproduce the continuous yielding
seen in soils), or too complicated (e.g., the Wang model [Wang et al. 2001], which needs many
parameters and a lengthy calibration process). Further, there is a need to directly accept the same
degradation curves used by equivalent-linear methods (see Figures 1.31 and 1.32 for examples), to
allow engineers to move easily from using these methods to using fully nonlinear methods.
A further motivation for incorporating such cyclic data into a hysteretic damping model for FLAC
and FLAC 3D is that the need for additional damping such as Rayleigh damping would be eliminated.
Rayleigh damping is unpopular with code users because it often involves a drastic reduction in
timestep, and a consequent increase in solution time.
Optional hysteretic damping is described here; it may be used on its own, or in conjunction with the
other damping schemes, such as Rayleigh damping or local damping. (It may also be used with any
of the built-in constitutive models, except for the transversely isotropic-elastic, modified Cam-clay
and creep material models.)
The hysteretic damping formulation is not intended to be a complete constitutive model: it should
be used as a supplement to one of the built-in nonlinear models, and not as a primary way to
simulate yielding. If models that embody both plastic yield and an adequate hysteretic response
are used, then there is no reason to add further damping; the hysteretic damping option in FLAC is
only intended to provide damping for those models lacking intrinsic damping when not yielding.
Further, Rayleigh damping should be unnecessary when hysteretic damping is in operation, apart
from its possible use (at low levels: e.g., 0.2%) to remove high frequency noise.
Figure 1.31 Modulus reduction curve for sand (Seed & Idriss 1970 – “upper
range”). The data set is from the file supplied with the SHAKE-91
code download. (http://nisee.berkeley.edu/software/ )
Figure 1.32 Modulus reduction curve for clay (Seed & Sun 1989 – “upper
range”). The data set is from the file supplied with the SHAKE-
91 code download. (http://nisee.berkeley.edu/software/ )
Formulation
Modulus degradation curves, as illustrated in Figures 1.31 and 1.32, imply a nonlinear stress/strain
curve. If we assume an ideal soil in which the stress depends only on the strain (not on the number
of cycles, or time), we can derive an incremental constitutive relation from the degradation curve,
described by τ̄ /γ = Ms , where τ̄ is the normalized shear stress (= τ/Go ) (where Go is the small-
strain shear modulus of the material), γ is the shear strain and Ms is the strain-dependent normalized
secant modulus.
τ̄ = Ms γ (1.38)
d τ̄ dMs
Mt = = Ms + γ (1.39)
dγ dγ
where Mt is the normalized tangent modulus. The incremental shear modulus in a nonlinear
simulation is then given by Go Mt . This is used in place of the given shear modulus, Go (i.e., the
property assigned with the name shear).
In order to handle two- and three-dimensional strain paths, an approach similar to that described for
the “Finn model” (e.g., see Section 1.4.4.2) is used, whereby the shear strain is decomposed into
components in strain space, and strain reversals are detected by changes in signs of the dot product
of the current increment and the previous mean path. Following the formulation of the Finn model
(replacing ε with γ , but otherwise using the same notation):
γ2 := γ2 + 2e12 (1.41)
√
z= υi υi (1.43)
υi
noi = (1.44)
z
A reversal is detected when |d| passes through a maximum, and the previous-reversal strain values
are updated as given by Eqs. (1.46) and (1.47). Note that there is no “latency” period, as used in the
Finn model (see Section 1.4.4.2); there is no minimum number of timesteps that must occur before
a reversal is detected.
γio = γi (1.47)
Between reversals, the shear modulus is multiplied by Mt , using γ = |d| in Eq. (1.39). The
multiplier is applied to the shear modulus used in all built-in constitutive models, except for the
transversely isotropic-elastic, modified Cam-clay and creep material models.
Implementation
The formulation described above is implemented in FLAC by modifying the strain-rate calculation so
that the mean strain-rate tensor (averaged over all subzones) is calculated before any calls are made
to constitutive model functions. At this stage the hysteretic logic is invoked, returning a modulus
multiplier which is passed to any called constitutive model. The model then uses the multiplier Mt
to adjust the apparent value of tangent shear modulus of the full zone being processed.
In addition to the “backbone curve,” provided by applying Eq. (1.39) to a modulus degradation
curve (described below), two Masing (1926) rules are used to specify the behavior at reversal
points. Essentially, these state that: (1) a new (but inverted) function is started upon reversal,
implying that the initial unload modulus is G; and (2) the first quarter-cycle of loading is scaled by
one-half relative to all other cycles. Although Pyke (2004) concludes that “neither of the Masing
rules is valid,” some simplifying assumption is necessary to ensure repeatable, closed loops. The
fact that real soil departs from this ideal behavior is not believed to be too important in this context
because the formulation is not intended to be a complete constitutive model, but simply to provide
hysteretic damping as an alternative to Rayleigh damping.
An additional rule deals with sub-cycles: the hysteretic logic contains push-down FILO* stacks
that record all state information (e.g., stress, strain, tangent modulus and previous reversal point)
at the point of reversal, for both positive- and negative-going strain directions. If the strain level
returns to – and exceeds – a previous value recorded in the stack (of the appropriate sign), the state
information is popped from the stack, so that the behavior (and, hence, tangent multiplier) reverts
back to that which applied at the time before the reversal.
The operation of the various rules is illustrated by the example shown in Figure 1.33. The initial
loading is interrupted by a small unload/load cycle; after this there is a complete unloading (extend-
ing to negative strain) followed by a loading part that continues to a higher positive strain level than
before. In particular, note: the half-scale initial loading curve; the slope of Go at each reversal;
and the restoration of the original loading path after execution of the small loop. In this case, both
the positive and negative stacks are popped upon closure of the small loop (i.e., the entire loop is
forgotten), but only the information from the positive stack is used to restore state information; the
negative-stack information is discarded.
JOB TITLE : .
7-Sep-07 13:38
step 5600 2.000
Dynamic Time 5.6000E-01
HISTORY PLOT
Y-axis : 1.000
1 Ave. SXY ( 1, 1)
X-axis :
2 X displacement( 1, 2)
0.000
-1.000
-2.000
-10 -5 0 5 10
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.33 Various load/unload cycles, to illustrate rules used in the hys-
teretic damping formulation (Example 1.6)
his xdis i 1 j 2
his nstep 1
ini xvel 1e-2 j=2
cyc 700
ini xvel mul -1
cyc 200
ini xvel mul -1
cyc 500
ini xvel mul -1
cyc 2000
ini xvel mul -1
cyc 2200
save revloop.sav
The degradation curves used in earthquake engineering are usually given as tables of values, with
cyclic strains spaced logarithmically. Since the derivative of the modulus-reduction curve is required
here (i.e., for Eq. (1.39)), the coarse spacing (e.g., 11 points in the curve shown in Figure 1.31) leads
to unacceptable errors if numerical derivatives are calculated. Thus, the implemented hysteretic
model uses only continuous functions to represent the modulus-reduction curve, so that analytical
derivatives may be calculated. The various implemented functions are described below. If degra-
dation curves are available only in table form, they must be fitted to one of the built-in functional
forms before simulations can be performed.
The hysteretic damping feature is invoked with the command
initial dy damp hyst name <v1 v2 v3 …> <range>
Where name is the name of the fitting function (chosen from the list: default, sig3, sig4 and hardin
– see below), and v1, v2, v3 . . . are numerical values for function parameters. The optional range
may be any acceptable range phrase for zones. Hysteretic damping may be removed from any range
of zones with the command
initial dy damp hyst off <range>
Note that the INITIAL dy damp hyst command only applies where the CONFIG dyn mode of operation
has been selected, and when SET dyn=on applies. Hysteretic damping operates independent of
all other forms of damping, which may be also specified to operate “in parallel” with hysteretic
damping.
Tangent-Modulus Functions
Various built-in functions are available to represent the variation of the shear modulus reduction
factor, G/Gmax , with cyclic strain (given in percent), according to the keyword specified on the
INITIAL dy damp hyst command.
Default model – default
The default hysteresis model is developed by noting that the S-shaped curve of modulus versus
logarithm of cyclic strain can be represented by a cubic equation, with zero slope at both low strain
and high strain. Thus, the secant modulus, Ms , is
Ms = s 2 (3 − 2s) (1.48)
where
L2 − L
s= (1.49)
L2 − L1
L = log10 (γ ) (1.50)
The parameters L1 and L2 are the extreme values of logarithmic strain (i.e., the values at which the
tangent slope becomes zero). Thus, giving L1 = −3 and L2 = 1 means that the S-shaped curve
will extend from a lower cyclic strain of 0.001% (10−3 ) to an upper cyclic strain of 10% (101 ).
Since the slopes are zero at these limits, it is not meaningful to operate the damping model with
strains outside the limits. (Note that Eq. (1.48) is only assumed to apply for 0 ≤ s ≤ 1, and that
the tangent modulus will be set to zero otherwise.) The tangent modulus is given by
dMs
Mt = Ms + γ (1.51)
dγ
dMs dMs ds dL
= · · (1.52)
dγ ds dL dγ
we obtain
6s(1 − s)
Mt = s 2 (3 − 2s) − log10 e (1.53)
L2 − L1
There is a further limit, s > smin , such that the tangent modulus is always positive (no strain
softening). Thus,
6smin (1 − smin )
2
smin (3 − 2smin ) = log10 e (1.54)
L2 − L1
or
2
2smin − smin (A + 3) + A = 0 (1.55)
A+3− (A + 3)2 − 8A
smin = (1.56)
4
a
Ms = (1.57)
1 + exp(−(L − xo )/b)
sig4 model
a
Ms = yo + (1.58)
1 + exp(−(L − xo )/b)
The command line for invoking these models requires that 3 symbols (a, b and xo ) are defined
by the parameters v1, v2, and v3, respectively, for model sig3 (Eq. (1.57)). For model sig4, the 4
symbols, a, b, xo and yo , are entered by means of the parameters v1, v2, v3 and v4, respectively.
Numerical fits for the two models to the curve of Figures 1.31 and 1.32 are provided in Tables 1.1
and 1.2, respectively.
Hardin/Drnevich model – hardin
The following function was suggested by Hardin and Drnevich (1972):
1
Ms = (1.59)
1 + γ /γref
It has the useful property that the modulus reduction factor is 0.5 when γ = γref , so that the sole
parameter, γref , may be determined (by inspection) from the strain at which the modulus-reduction
curve crosses the G/Gmax = 0.5 line. Choosing a value of γref = 0.06 produces a match to the
curve of Figure 1.31, and a value of 0.234 produces a match to the curve of Figure 1.32.
Table 1.1 Numerical fits to Seed & Idriss data for sand
Data set Default Sig3 Sig4 Hardin
Sand – L1 = −3.325 a = 1.014 a = 0.9762 γref = 0.06
upper range L2 = 0.823 b = −0.4792 b = −0.4393
(Seed & xo = −1.249 xo = −1.285
Idriss 1970) yo = 0.03154
Table 1.2 Numerical fits to Seed & Sun data for clay
Data set Default Sig3 Sig4 Hardin
Clay – L1 = −3.156 a = 1.017 a = 0.922 γref = 0.234
upper range L2 = 1.904 b = −0.587 b = −0.481
(Seed & xo = −0.633 xo = −0.745
Sun 1989) yo = 0.0823
The results are summarized in Figures 1.34 and 1.35, which present the tangent modulus results and
damping ratio results from FLAC together with the Seed & Idriss results. Although the modulus
results match the target data well over five orders of magnitude (Figure 1.34), the measured damping
does not conform well with the published damping curves for the same material over the same range
(Figure 1.35).
Example 1.7 One-zone sample exercised at several cyclic strain levels (using sig3 model)
conf dy
def setup
givenShear = 1e8
CycStrain = 0.1 ; (percent cyclic strain)
;---- derived ..
setVel = 0.01 * min(1.0,CycStrain/0.1)
givenBulk = 2.0 * givenShear
timestep = min(1e-4,1e-5 / CycStrain)
nstep1 = int(0.5 + 1.0 / (timestep * 10.0))
nstep2 = nstep1 * 2
nstep3 = nstep1 + nstep2
nstep5 = nstep1 + 2 * nstep2
end
setup
;
gri 1 1
m e
prop den 1000 sh givenShear bu givenBulk
fix x y
ini xvel setVel j=2
set dydt 1e-4
ini dy_damp hyst sig3 1.014 -0.4792 -1.249
* Examples 1.7 and 1.8 illustrate the calculations for one cyclic shear-strain value. The data file to
develop the complete curve is given in “MODRED.DAT” in the “\Dynamic” directory. For further
explanation see “MODRED.FIS” in the FISH library in Section 3 in the FISH volume.
his sxy i 1 j 1
his xdis i 1 j 2
his nstep 1
cyc nstep1
ini xv mul -1
cyc nstep2
ini xv mul -1
cyc nstep2
his write 1 vs 2 tab 1
def HLoop
emax = 0.0
emin = 0.0
tmax = 0.0
tmin = 0.0
loop n (1,nstep5)
emax = max(xtable(1,n),emax)
emin = min(xtable(1,n),emin)
tmax = max(ytable(1,n),tmax)
tmin = min(ytable(1,n),tmin)
endLoop
slope = ((tmax - tmin) / (emax - emin)) / givenShear
oo = out(’ strain = ’+string(emax*100.0)+’% G/Gmax = ’+string(slope))
Tbase = ytable(1,nstep3)
Lsum = 0.0
loop n (nstep1,nstep3-1)
meanT = (ytable(1,n) + ytable(1,n+1)) / 2.0
Lsum = Lsum + (xtable(1,n)-xtable(1,n+1)) * (meanT - Tbase)
endLoop
Usum = 0.0
loop n (nstep3,nstep5-1)
meanT = (ytable(1,n) + ytable(1,n+1)) / 2.0
Usum = Usum + (xtable(1,n+1)-xtable(1,n)) * (meanT - Tbase)
endLoop
Wdiff = Usum - Lsum
Senergy = 0.5 * xtable(1,nstep1) * yTable(1,nstep1)
Drat = Wdiff / (Senergy * 4.0 * pi)
oo = out(’ damping ratio = ’+string(Drat*100.0)+’%’)
end
HLoop
save cyclic.sav
Figure 1.34 Results of several cyclic FLAC simulations for sig3 model – secant
modulus values versus cyclic shear strain in %. Seed & Idriss
data also shown.
Figure 1.35 Results of several cyclic FLAC simulations for sig3 model – damp-
ing values versus cyclic shear strain in %. Seed & Idriss data also
shown.
Clearly, the published data for modulus and damping is inconsistent with a conceptual model of
strain- and time-independent material response. It is unclear whether the two sets of published data
came from different tests, or if the nature of the test led to the inconsistencies. For example, the
steady-state response (after many cycles of applied strain) may be different from the initial – single
cycle – response. If this is true, then it is not evident that the steady-state response (presumably
encompassed by the published results) is a better representation in typical earthquake simulations
than the single-cycle response, because many earthquakes contain only one or two large-amplitude
cycles. Thus, the single-cycle response may more correctly represent material behavior under
earthquake loading. In this case, the damping and modulus curves are consistent.
Note that the numerical results are self-consistent because the two curves (normalized shear modulus
and damping ratio) derive from the same basic stress-strain relation. The derivation of incremental
stress/strain relations from modulus-reduction curves assumes that the hysteretic mechanism is
stationary (i.e., that stress depends only on strain, and not on the number of cycles executed or on
past history). In real soils there are often hardening or softening processes that cause successive
cycles to be different, which may also explain why the modulus reduction is apparently inconsistent
with the corresponding damping curve in a typical lab test.
Although the implemented hysteretic damping ignores these non-stationary effects, it is still possible
to match both modulus and damping curves to a reasonable accuracy. In the absence of consistent
laboratory data, it is suggested that a compromise approach be taken, in which both the damping
and modulus curves are fitted over a reasonable range of strains (corresponding to the strains being
modeled). As an example of this strategy, the default model is used (with data file Example 1.8),
giving the FLAC results shown in Figures 1.36 and 1.37. The hysteretic damping in this case was
invoked with the following command (see default parameters in Table 1.1):
ini dy damp hyst default -3.325 0.823
The results show that, over a middle range of strain (say, 0.001% to 0.3% strain), there is an
approximate fit to both the modulus and damping curves of Seed & Idriss.
Example 1.8 One-zone sample exercised at several cyclic strain levels (using default
model) with approximate fit over selected strain range
conf dy
def setup
givenShear = 1e8
end
setup
gri 1 1
m e
prop den 1000 sh givenShear bu 2e8
fix x y
ini xvel 1e-2 j=2
set dydt 1e-4
ini dy_damp hyst default -3.5 1.3
his sxy i 1 j 1
his xdis i 1 j 2
his nstep 1
cyc 1000
ini xv mul -1
cyc 2000
ini xv mul -1
cyc 2000
his write 1 vs 2 tab 1
def HLoop
emax = 0.0
emin = 0.0
tmax = 0.0
tmin = 0.0
loop n (1,5000)
emax = max(xtable(1,n),emax)
emin = min(xtable(1,n),emin)
tmax = max(ytable(1,n),tmax)
tmin = min(ytable(1,n),tmin)
endLoop
slope = ((tmax - tmin) / (emax - emin)) / givenShear
oo = out(’ strain = ’+string(emax*100.0)+’% G/Gmax = ’+string(slope))
Tbase = ytable(1,3000)
Lsum = 0.0
loop n (1000,2999)
meanT = (ytable(1,n) + ytable(1,n+1)) / 2.0
Lsum = Lsum + (xtable(1,n)-xtable(1,n+1)) * (meanT - Tbase)
endLoop
Usum = 0.0
loop n (3000,4999)
meanT = (ytable(1,n) + ytable(1,n+1)) / 2.0
Usum = Usum + (xtable(1,n+1)-xtable(1,n)) * (meanT - Tbase)
endLoop
Wdiff = Usum - Lsum
Senergy = 0.5 * xtable(1,1000) * yTable(1,1000)
Drat = Wdiff / (Senergy * 4.0 * pi)
oo = out(’ damping ratio = ’+string(Drat*100.0)+’%’)
end
HLoop
save cyclefit.sav
Figure 1.36 Results of several cyclic FLAC simulations for default model –
secant modulus values versus cyclic shear strain in %. Seed &
Idriss data also shown.
Figure 1.37 Results of several cyclic FLAC simulations for default model –
damping values versus cyclic shear strain in %. Seed & Idriss
data also shown.
In most attempts to match laboratory and numerical damping curves, it is noted that the damping
provided by the hysteretic formulation at low cyclic strain levels is lower than that observed in
the laboratory. This may lead to low-level noise, particularly at high frequencies. Although such
noise hardly affects the essential response of the systems, for cosmetic reasons it may be removed
by adding a small amount of Rayleigh damping. It is found that 0.2% Rayleigh damping (at an
appropriate center frequency) is usually sufficient to remove residual oscillations without affecting
the solution timestep.
Simple Application
An example of a 20 m layer excited by a digitized earthquake is provided to show that plausible
behavior occurs for a case involving wave propagation, multiple and nested loops, and reasonably
large cyclic strain. The data file Example 1.9 is listed:
The digitized earthquake record is described as “LOMA PRIETA GILROY.” The stress/strain loops
for the bottom and middle of the layer are shown in Figures 1.38 and 1.39, respectively, and the
acceleration histories for 3 positions are shown in Figure 1.40. The simulation is in one dimension,
for excitation in the shear direction only. Note that for this example, the initial stresses are zero. If
a non-zero initial vertical and horizontal stress state is specified, then the left and right boundaries
should be attached to produce a one-dimensional simulation.
The hysteretic model seems to handle multiple nested loops in a reasonable manner. There is clearly
more energy dissipation at the base of the model than at the middle. The maximum cyclic strain is
about 0.15%. The magnitude of timestep is unaffected by the hysteretic damping.
Observations
A method has been developed to use cyclic modulus-degradation data directly in a FLAC simula-
tion. The resulting model is able to reproduce the results of constant-amplitude cyclic tests, but it is
also able to accommodate strain paths that are arbitrary in strain space and time. Thus, it should be
possible to make direct comparisons between calculations made with an equivalent-linear method
and a fully nonlinear method, without making any compromises in the choice of constitutive model.
The developed method is not designed to be a plausible soil model; rather, its purpose is to allow
current users of equivalent-linear methods a painless way to upgrade to a fully nonlinear method.
Further, the hysteretic damping of the new formulation will enable users to avoid the use of Rayleigh
damping and its unpopular timestep penalties. A comparison of a layered model, assuming non-
linear elastic material using SHAKE, to one using FLAC with hysteretic damping is provided in
Section 1.7.3.
JOB TITLE : .
11-Sep-07 9:21
3.000
step 91288
Dynamic Time 2.5000E+01
0.000
-1.000
-2.000
-10 -5 0 5 10 15
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.38 Shear stress vs shear strain for base of the layer; default FLAC
hysteretic model
JOB TITLE : .
-0.500
-1.000
-1.500
-05
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.39 Shear stress vs shear strain for middle of the layer; default FLAC
hysteretic model
JOB TITLE : .
0.800
26-Sep-07 9:31
step 91288
Dynamic Time 2.5000E+01 0.600
-0.200
-0.400
-0.600
-0.800
4 8 12 16 20 24
JOB TITLE : .
26-Sep-07 9:31
1.000
step 91288
Dynamic Time 2.5000E+01
-0.500
-1.000
-1.500
4 8 12 16 20 24
JOB TITLE : .
-0.500
-1.000
-1.500
4 8 12 16 20 24
The following conditions should be checked when hysteretic damping is applied in a dynamic
simulation:
Compare hysteretic damping response to laboratory tests – It is generally not useful to compare
apparent damping response from the hysteretic damping model used with a yield constitutive model
to damping curves from laboratory results, because laboratory tests are likely to be unstable when
shear failure occurs, and therefore unable to provide meaningful results. It is recommended that
FLAC ’s hysteretic response be matched to cyclic laboratory results obtained for a strain range that
excludes failure, and that plasticity parameters (e.g., cohesion and friction angle) be matched to
static laboratory strength tests. FLAC is able to combine the response in both regimes during a
typical simulation of seismic response.
Check cyclic strain level – Hysteretic damping not only adds energy loss to dynamic straining, it
also causes the mean shear modulus to decrease for large cyclic strains. This may lead to unexpected
results (e.g., an increased response amplitude, due to the shift in resonance frequency closer to the
dominant frequency of input waves).
Before running a dynamic model with hysteretic damping, an elastic simulation should be made
without damping, to observe the maximum levels of cyclic strain that occur. If the cyclic strains are
large enough to cause excessive reductions in shear modulus for a given modulus reduction curve,
then the use of hysteretic damping is questionable – it will be performing outside of its intended
range of application. The model properties and input should be checked. If the properties and input
are reasonable, and the large cyclic strains are limited to small regions, then consider the possibility
of excluding hysteretic damping from these regions and using only a yield model in these regions
(because the large strains imply that yielding should occur).
Check modulus reduction factor – Even if cyclic strains under elastic conditions are small, the
use of a yield model may increase the strains. The hysteretic damping formulation is not intended
to be a substitute for a yielding constitutive model. It may be used in conjunction with a yield
model (such as Mohr-Coulomb), but conflicts in the domain of application should be avoided, if
meaningful results are to be expected.
Hysteretic damping is “switched off” for each zone while plastic flow is occurring (including the
associated strain accumulation for the hysteretic damping calculation). Note that the stiffness-
proportional component of Rayleigh damping is also switched off during plastic flow.
Even so, the apparent modulus used by the hysteretic damping logic may drop to low values, which
can cause unrealistic response. Low (or zero) values of the modulus can be checked by monitoring
modfac, which is the factor by which the small-strain modulus, Go , is multiplied for hysteretic
damping. (For example, use HISTORY hyst modfac for selected zones, or plot contours of modfac
at critical times.)
In some simulations, large modulus reductions have been noticed in areas remote from regions
of plastic flow (e.g., near the base of the model). It appears that the use of a single modulus-
reduction curve is unrealistic in these cases. There is evidence (e.g., see Darendeli 2001) that
degradation curves depend on the mean stress level: for example, at depth (high mean stress) there
is less damping and modulus reduction. By making the hysteretic damping depth-dependent, the
simulation should be more realistic.
Check initial shear stress state – In laboratory tests, the initial shear stress is assumed to be zero,
leading to hysteresis loops that are generally symmetrical. In practical applications, the initial shear
stress is unlikely to be zero (for example, within soil elements near the surface of an embankment).
The user of hysteretic damping must decide on the best estimate of the initial state of the material,
because the hysteretic formulation in FLAC depends on the past history of shear strain. Two cases
can be identified:
1. If hysteretic damping is activated after a set of equilibrium stresses has been installed,
then the initial shear strain will be zero, and cyclic excursions of shear stress will tend to
be symmetrical about the starting point.
2. If the initial stresses are built up by straining the model while hysteretic damping is active,
then subsequent cyclic excursions of shear stress will tend to be asymmetrical because
the initial bias in strain causes the slope of the stress/strain curve to be flatter on the side
with higher stress.
These cases are illustrated by modifying the simple application given in Example 1.9. With no
initial shear stress, the cyclic response of the model is nearly symmetrical, as shown in Figures 1.38
and 1.39.
The simulation is repeated with uniform shear stresses of 0.1 MPa initialized in the 20 m layer,
and with an equal static shear stress applied at the boundary to maintain initial equilibrium. The
following commands are added to Example 1.9 after the FIX y command:
initial sxy 1e5
apply sxy 1e5 j 21
apply sxy 1e5 j 1
solve ;; to check equilibrium
The resulting dynamic response is identical to that of the original simulation, but the set of loops
is shifted upward by 0.1 MPa. Figure 1.43 shows the result at the base of the model; compare to
Figure 1.38. This corresponds to Case 1 above.
In order to make the initial strains compatible with the initial stresses, hysteretic damping is arranged
to be active during the establishment of the initial stress state. For this case, the following commands
are added to Example 1.9 after the FIX y command:
apply sxy 1e5 j 21
apply sxy 1e5 j 1
ini dy damp hyst default -3.325 0.823
ini dy damp local 0.7
solve
initial xvel 0 yvel 0
set dytime 0
ini dy damp local 0.0
Shear stresses are applied to the boundaries of the grid, but not initialized within the grid; thus, a
static solution (using additional local damping with the hysteretic damping to speed convergence)
is used to build up internal stresses. Because hysteretic damping is active during the static solution,
strains will be compatible with stresses at the start of the dynamic simulation. The dynamic response
of the system is indicated by the stress/strain loops plotted in Figure 1.44. In this case there is a
marked asymmetry. The soil is already partially yielded (as a result of the initial stress/strain state).
Further straining in the same direction of loading produces more yielding, while straining in the
opposite direction initially acts to reduce the yielding. This corresponds to Case 2 above.
The approach of using a full static solution while hysteretic damping is active, in order to obtain
a compatible starting state for both stress and strain, may be used for more complicated models,
such as embankments in which the slope area contains initial shear stresses. One drawback of the
approach is that the static solution may involve many diminishing cycles of oscillation as the state
of equilibrium is approached. Although these cycles tend to be quite small (and hardly affect the
desired stress state), they cause many states to be stored on the memory stack (see Section 1.4.3.3).
These stored states are deleted from the stack early in dynamic loading, but they occupy memory,
and entail some initial overhead in computer time. It may be possible (in a future version of FLAC)
to include logic to allow users to flush the stacks at the end of the static initialization, while retaining
the latest state of stress-compatible strains for a dynamic simulation with hysteretic damping.
JOB TITLE : .
0.000
-1.000
-2.000
-15 -10 -5 0 5 10 15 20
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.43 Shear stress vs shear strain for base of the layer; with shear stress
simply initialized to 0.1 MPa
JOB TITLE : .
0.000
-1.000
-2.000
-5 0 5 10 15 20 25 30 35
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.44 Shear stress vs shear strain for base of the layer. The shear stress
is 0.1 MPa and the initial strain is 0.041%, following the static
solution.
Local damping (see Section 1.3.4 in Theory and Background) was originally designed as a means
to equilibrate static simulations. However, it has some characteristics that make it attractive for
dynamic simulations. It operates by adding or subtracting mass from a gridpoint or structural
node at certain times during a cycle of oscillation; there is overall conservation of mass, because
the amount added is equal to the amount subtracted. Mass is added when the velocity changes
sign, and subtracted when it passes a maximum or minimum point. Hence, increments of kinetic
energy are removed twice per oscillation cycle (at the velocity extremes). The amount of energy
removed, W , is proportional to the maximum, transient strain energy, W , and the ratio W/W
is independent of rate and frequency. Since W/W may be related to fraction of critical damping,
D (Kolsky 1963), we obtain the expression
αL = π D (1.60)
where αL is the local damping coefficient. Thus, the use of local damping is simpler than Rayleigh
damping, because we do not need to specify a frequency. To compare the two types of damping,
we repeat Example 1.5 with 5% damping, which is a typical value used for dynamic analyses.
Example 1.10 provides the data file; we also set fmin to 24.1, which is a more accurate estimate of
the natural frequency of the block. A similar run is done with local damping, with the coefficient set
to 0.1571 (= 0.05π) – see Example 1.11. In both runs, we specify the timestep at 5×10−4 , so that
we can execute the same number of steps in each to obtain the same elapsed time. Displacement
histories from the two runs are given in Figures 1.45 and 1.46, respectively. The results are quite
similar.
A modified form of local damping – combined damping – may also be used in dynamic mode, but
its performance is unknown. The formulation for combined damping is given in Section 1.3.4 in
Theory and Background, and the command to invoke it is SET dy damp combined value.
JOB TITLE : .
-1.000
26-Sep-07 10:08
step 1000
Dynamic Time 5.0000E-01 -2.000
-6.000
-7.000
-8.000
-9.000
5 10 15 20 25 30 35 40 45 50
-02
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
JOB TITLE : .
-1.000
26-Sep-07 10:09
step 1000
Dynamic Time 5.0000E-01 -2.000
-6.000
-7.000
-8.000
-9.000
5 10 15 20 25 30 35 40 45 50
-02
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
CAUTION: Local damping appears to give good results for a simple case because it is frequency-
independent and needs no estimate of the natural frequency of the system being modeled. However,
this type of damping should be treated with caution, and the results compared to those with Rayleigh
damping for each application. There is some evidence to suggest that, for complicated wave forms,
local damping underdamps the high frequency components, and may introduce high frequency
“noise.”
Local damping is not recommended for seismic simulations, because this type of damping cannot
completely represent the energy loss of multiple cyclic loading properly.
1.4.3.8 Spatial Variation in Damping
Rayleigh damping and local damping are both assigned as global parameters by the SET command
in FLAC. A spatial variation in the damping parameters (and the damping type) for Rayleigh,
local and hysteretic damping can be prescribed via the INITIAL dy damp command. For example, if
different materials are known to have different fractions of critical damping, a different value for
ξmin can be assigned to each material. This can be demonstrated by modifying the example of a
wave propagating in a column (Example 1.2). In Example 1.12, two separate identical grids are
constructed, to enable a direct comparison to be made. Both grids contain two layers: a stiff layer in
the lower half, and a soft layer in the upper half. The left-hand grid has uniform Rayleigh stiffness
damping, while the right-hand grid has two values for the damping coefficient, corresponding to the
two materials, although the average damping coefficient is the same as that of the left-hand grid.
The velocity histories at the free surface are plotted in Figure 1.47 for both grids. Differences in
response can be observed particularly in the second pulse (reflected from the material discontinuity).
JOB TITLE : .
LEGEND
HISTORY PLOT
Y-axis : 0.800
2 X velocity ( 1, 51)
4 X velocity ( 3, 51) 0.600
X-axis :
5 Dynamic time 0.400
0.200
0.000
-0.200
5 10 15 20 25 30
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.47 Velocity histories at a free surface for spatial variation in damping
The specification of nonuniform damping with the INITIAL command follows the syntax of both
the SET dy damp command and the INITIAL command. For example, variations, additions and
multipliers can be prescribed for all parameters. In its simplest form, the INITIAL dy damp command
resembles the SET dy damp command (e.g., the following two commands produce identical results):
set dy damp rayl 0.05 25.0
ini dy damp rayl 0.05 25.0
Note that a SET dy damp command implicitly sets damping for all grid elements (and overrides
any previous INITIAL dy damp specifications). By using range parameters, several INITIAL dy damp
commands can be used to install different damping values (and even different damping types) in
various locations. The var keyword can also be used. For example, we can modify the previous
example of uniform damping:
ini dy damp rayl 0.05 var 0.1,0.2 25.0 var -5,0
In this case, there are spatial variations in both the damping coefficient and the center frequency.
The syntax follows the general rule for the INITIAL command, in that any parameter value may be
followed by the keyword var, add or mul, with appropriate parameters for those keywords. Note that
all damping parameters pertain to gridpoints. In particular, the Rayleigh stiffness-proportional term,
which acts on zone strain rates, is derived by averaging, from values specified at the neighboring
gridpoints.
The command PRINT dy damp produces a normal grid printout, consisting of two or more blocks
of data: the first block denotes the damping type (L, C or R, for local, combined or Rayleigh,
respectively, with modifiers m and s for mass and stiffness); and the second block records the
damping coefficient. In the case of Rayleigh damping, there is a third block of output that records
the center frequency.
There is no direct plot of damping information, but the FISH grid intrinsic damp can be used to
transfer appropriate data to the extra arrays for plotting. See Section 2 in the FISH volume for a
description of the intrinsic damp.
If damping parameters are modified with the FISH intrinsic damp, the change will not necessarily
take effect immediately, because the code uses derived coefficients. In small-strain mode, derived
coefficients are computed from user-given parameters when a CYCLE or STEP command is given;
in large-strain mode, the derivation is done every 10 steps. A user-written FISH function may force
the derived coefficients to be computed by executing the intrinsic do update. Note that the timestep
may change as a result (if the Rayleigh stiffness term is changed).
1.4.3.9 Structural Element Damping for Dynamic Simulations
Rayleigh or local damping can also be specified independently for structural elements by giving
the struct keyword immediately following SET dy damp. Damping is then applied specifically for
all structural elements in the model. See, for example, Example 1.17.
Structural damping operates in a way similar to damping in the grid. However, if a structural node
is rigidly attached to a gridpoint, the gridpoint damping value is used rather than the structural node
damping value. For the special case of a structural node attached to a null gridpoint (one surrounded
by null zones), the damping for that gridpoint/node is zero.
Note that stiffness damping is included by default for pile coupling springs. This damping can be
turned off by using the SET dy damp pile sd off command.
1.4.3.10 Artificial Viscosity
Von Neumann and Landshoff artificial viscosity terms are implemented in FLAC to control damping
involving sharp fronts in dynamic analysis. These viscous damping terms are a generalization of the
one-dimensional equations (1) and (3) in Wilkins (1980), and correspond to the original viscosity
formulation of von Neumann and Richtmyer (see Wilkins 1980).
The artificial viscosity method was initially developed for numerical calculation of shock propa-
gation in fluid dynamics. The method may not apply to elastic or plastic waves when shear stress
components are significant when compared to mean pressure, because shear waves are not damped
by the method. The purpose of the quadratic von Neumann term q1 is to spread the shock over
a number of grid spacings, and damp the oscillations behind the front. The effect of the linear
Landshoff term q2 is to diffuse the shock front over an increased number of zones as the shock
progresses.
In the FLAC implementation, a linear combination, q, of the scalar viscosity terms q1 and q2 is
used on a zone basis:
q = an q1 + al q2 (1.61)
where an and al are two constants. The viscous terms have the form
q1 = b c0 2 ρ L2 ˙ 2 (1.62)
q2 = b c1 ρ L a ˙ (1.63)
where:L is a characteristic zone dimension (square root of the zone area);
˙ is the zone volumetric rate;
ρ is the zone density;
(K+ 43 G)
a is the material p-wave speed: a = ρ
where K and G are bulk and shear moduli for the zone;
c0 is a constant set = 2; and
c1 is a constant set = 1.
and, to accommodate both compressive and dilatant shocks, we specify
b = −sgn(˙ )
The isotropic viscous stress contribution is added to the out-of-balance force for the nodes before
resolution of the equations of motion.
The following command is provided to activate artificial damping for a FLAC model:
SET dy damp avisc an al
where an and al are the two constants defined above, which should, in most instances, be assigned
the value of 1.
Note that the presence of damping terms results in a slightly more stringent stability condition that
has not been taken into consideration in the implementation. Hence, in some cases, it may be
necessary to reduce the timestep to achieve satisfactory stability.
The data file in Example 1.13 corresponds to a model with a sharp velocity wave (of the form shown
in Figure 1.48) applied to the left boundary. The data file is run in both plane-strain and axisymmetry
mode using the artificial viscosity model. (Replace the CONFIG dyn command with CONFIG dyn
axi for the axisymmetry analysis.) The effect on wave transmission through the grid is illustrated
by the x-velocity plots in Figure 1.49 for the plane-strain model without artificial viscosity (SET
dy damp avisc command removed), compared to Figure 1.50 for the model with artificial viscosity.
Figure 1.51 shows the results for the axisymmetry model with artificial viscosity.
Note that an alternative form of damping for shock waves is described in Section 1.7.7. In this case,
the scheme is coded in FISH (function leak in Example 1.45).
JOB TITLE : .
LEGEND
26-Sep-07 10:42
step 259 1.000
Dynamic Time 1.0029E-03
HISTORY PLOT
Y-axis : 0.800
2 wave (FISH)
X-axis :
1 Dynamic time
0.600
0.400
0.200
2 4 6 8 10
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
JOB TITLE : .
LEGEND
HISTORY PLOT
Y-axis : 0.800
4 X velocity ( 10, 150)
5 X velocity ( 20, 150) 0.600
6 X velocity ( 30, 150)
X-axis : 0.400
1 Dynamic time
0.200
0.000
-0.200
2 4 6 8 10
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.49 x-velocity histories for plane-strain model without artificial vis-
cosity
JOB TITLE : .
LEGEND
26-Sep-07 10:41
1.000
step 259
Dynamic Time 1.0029E-03
0.200
0.000
2 4 6 8 10
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.50 x-velocity histories for plane-strain model with artificial viscosity
JOB TITLE : .
HISTORY PLOT
Y-axis : 3.000
4 X velocity ( 10, 150)
5 X velocity ( 20, 150) 2.000
6 X velocity ( 30, 150)
X-axis : 1.000
1 Dynamic time
0.000
-1.000
-2.000
2 4 6 8 10
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.51 x-velocity histories for axisymmetry model with artificial viscosity
1.4.3.11 Integration of Damping Schemes and Nonlinear Material Models for Geo-materials
Energy dissipation in soil and rock is largely hysteretic in nature; the specific loss for each
load/unload cycle of shear strain is independent of the rate at which the cycle is executed, but
dependent on the amplitude of the cycle. Ideally, this behavior would be reproduced by an appro-
priate constitutive model, but adequate nonlinear models tend to be complicated, embodying many
material parameters. Simpler models, such as Mohr Coulomb, are often used instead, in order
to reproduce irreversible strain accumulation (e.g., slumping or slip on shear surfaces) that may
occur during seismic loading. In such models, additional damping must be included to account for
cyclic dissipation during the elastic part of the response, and during wave propagation through the
site. Rayleigh damping is commonly used, but, as discussed in Section 1.4.3.1, it only provides
(approximately) rate-independent damping over a limited frequency range, and it entails a large
reduction in critical timestep, and consequent long runtimes.
The following discussion illustrates the ways damping can be integrated with a simple material
model. We consider the use of hysteretic damping plus damping arising from plastic flow. As
discussed in Section 1.4.3.4, hysteretic damping is based on a secant modulus-reduction curve for
primary loading (the backbone curve) and a Masing rule assumption for unloading/reloading to pro-
vide energy dissipation. Hysteretic damping is applied in the elastic range only, and natural damping
provided by the constitutive model operates in the plastic range. Three simple cases are presented.
First, the energy dissipation provided by a standard elastic/plastic Mohr-Coulomb is shown. Sec-
ond, damping is incorporated into a linearly elastic model using hysteretic (Hardin/Drnevich model)
damping. And third, the Hardin/Drnevich model is combined with the Mohr-Coulomb model. The
energy dissipation is compared for all three cases by evaluating the change in shear modulus and
damping ratio for each case.
Natural damping with the Mohr-Coulomb model
Standard elastic/plastic models such as Mohr-Coulomb can produce shear-modulus reduction and
damping ratio curves. Consider an elastic/plastic model with a constant shear modulus, Gmax , and
a constant yield stress, τm , subject to cyclic shear strain of amplitude |γ |. Below yield, the secant
shear modulus, G, is simply equal to Gmax . For cyclic excitation that involves yield, the secant
modulus is
G = τm /|γ | (1.64)
The modulus-reduction curve relates the ratio G/Gmax to the amplitude of shear strain, |γ |; it is
simply obtained by dividing Eq. (1.64) by Gmax , and using γm = τm /Gmax , we obtain, for |γ | > γm ,
G γm
= (1.65)
Gmax |γ |
The maximum stored energy, W , during the cycle (assuming G represents an elastic modulus) is
1
W = τm γc (1.66)
2
and the dissipated energy (corresponding to the area enclosed by the loop) is
Hence,
W 8(γc − γm )
= (1.68)
W γc
Denoting the damping ratio by D, and noting that for small D (Kolsky 1963),
1 W
D≈ (1.69)
4π W
2 (γc − γm )
D= (1.70)
π γc
We plot normalized modulus (G/Gmax ) from Eq. (1.65), and damping, D, from Eq. (1.70) against
normalized cyclic strain, γ /γm , in Figure 1.53. It can be seen that even a simple model (where
“simple” is taken in the context of dynamics) exhibits an evolution of modulus and damping that
can be matched to experimental results over limited ranges of cyclic strain.
Figure 1.53 Modulus and damping ratio versus cyclic strain for
elastic/plastic Mohr-Coulomb model
τ γ
= γ (1.71)
Gmax 1 + γref
where γref is the constant for the Hardin/Drnevich function. γref is the ultimate value of τ/Gmax :
τm
γref = (1.72)
Gmax
This curve is followed for primary loading. For unloading/reloading, the Masing rule holds. For
the case of cyclic shear loading at constant amplitude, γc , the Masing rule gives, for unloading,
γc − γ
τdown = −Gmax γc −γ + τc (1.73)
1+ 2γref
γc + γ
τup = Gmax γc +γ − τc (1.74)
1+ 2γref
where
γc
τc = Gmax (1.75)
1 + γγrefc
The initial loading curve and loop traced in one cycle of unloading/reloading is sketched in Fig-
ure 1.54:
The energy dissipated in one full unloading-reloading cycle is given by the integral
γc
W = (τup − τdown )dγ (1.76)
−γc
After introduction of Eqs. (1.73) and (1.74) in Eq. (1.76), and performing the integration, we obtain
2
γc
γc γc γref
W = 2
4Gmax γref 2 − ln(1 + ) − γc (1.77)
γref γref 1+ γref
1
W = τc γc (1.78)
2
γc
2 1 + γref γc γc
D= 2 2 − ln(1 + ) −1 (1.79)
π γc γref γref
γref
lim
γc
D=0 (1.80)
γref →∞
It is interesting to note that for the Hardin/Drnevich hysteretic damping and the elastic model, the
damping ratio does not depend on Gmax . Also, D is larger for smaller values of γref . For an elastic,
cyclic shear test of constant amplitude at constant volume, the use of hysteretic damping produces
a response that is independent of the number of cycles performed.
Hysteretic damping with the Mohr-Coulomb model
When hysteretic damping is used with an elastic/plastic model in FLAC, the modulus-reduction
technique is applied in the elastic range, and natural damping applies in the plastic range. In this
case, we combine the Hardin/Drnevich hysteretic damping with a Mohr-Coulomb model. The
Mohr-Coulomb model has a constant, tangent, elastic shear modulus, Gmax , and a constant yield
stress, τm . The Hardin/Drnevich model is used to provide energy dissipation in the elastic range
(but not to simulate yielding by means of a hyperbolic plasticity model). Accordingly, the yield
level from the hyperbolic law must be higher than the Mohr-Coulomb yield stress. This will be the
case provided that the following requirement is met:
where
τm
γm = (1.82)
Gmax
An initial loading curve involving Mohr-Coulomb yielding, and a loop traced in one cycle of
unloading/reloading, are sketched in Figure 1.55:
The elastic range is defined by γc < γm , where the shear strain, γm , is found from the following
relation (see Eq. (1.71)):
τm γm
= (1.83)
Gmax 1 + γγref
m
In the elastic range, γc < γm , the modulus reduction factor is given by Eq. (1.71), or
G 1
= (1.84)
Gmax 1 + γ|γref|
G 1
= (1.85)
Gmax γm |γ |
1+ γref γm
Also, in the plastic range, the energy dissipated in one cycle is the area enclosed by the loop in
Figure 1.55. This energy may be expressed as the sum of two contributions:
2
γm
γm γm γref
WH = 2
4Gmax γref 2 − ln(1 + ) − γm (1.87)
γref γref 1+ γref
G
max 2 γc
WMC = 4 γm γm −1 (1.88)
1 + γref γm
1
W = τm γc (1.89)
2
1 WH + WMC
D= (1.90)
4π W
After substituting Eqs. (1.87) and (1.88) in Eq. (1.90), we obtain with some manipulation,
γm
2 1 + γref γm γm 1 2 (γc − γm )
D= 2 2 − ln(1 + ) − 1 γc + (1.91)
π γm γref γref γref π γc
γref
The energy dissipation for the three damping cases is compared by exercising the equations for
G/Gmax and D over a cyclic strain range. A FISH function, listed in Example 1.14, performs this
exercise over a cyclic shear strain, γc , from 0.0001 to 4.0. The value for γm is set to 0.01 and the
value for γref is set to 0.02. The results for G/Gmax versus log γc /γm , based upon Eqs. (1.65),
(1.71), (1.84) and (1.85), are plotted in Figure 1.56. The results for D versus log γc /γm , based
upon Eqs. (1.70), (1.79) and (1.91), are plotted in Figure 1.57.
The inclusion of hysteretic damping is shown to reduce the shear modulus from the initial value of
Gmax , and increase the damping ratio (compared to the elastic-only response). The damping ratio
increases monotonically with shear strain amplitude, and approaches the asymptotic value of 2/π
for all three cases.
JOB TITLE : .
LEGEND
1.000
20-Jun-08 11:42
step 0
0.900
Table Plot
0.800
G/Gmax - MC only
G/Gmax - hyst only 0.700
G/Gmax - MC+hyst
0.600
0.500
0.400
0.300
0.200
0.100
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.56 Normalized shear modulus vs log normalized shear stain for three
damping cases
JOB TITLE : .
3.000
2.000
1.000
0.000
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.57 Damping ratio vs log normalized shear stain for three damping
cases
Liquefaction is defined as the loss of shear strength of soil under monotonic or cyclic loading,
arising from a tendency for loose soil to compact under shear loading. The term “liquefaction”
was originally coined by Mogami and Kudo (1953). Note that this definition covers both static and
dynamic liquefaction; the effective stress does not necessarily have to be zero for a soil to liquefy.
In particular, when a saturated cohesionless soil is submitted to rapid static or cyclic loading, the
tendency for the soil to densify causes the effective stress to decrease, and the process leads to soil
liquefaction.
It is known that pore pressures may build up considerably in some sands during cyclic shear
loading. Eventually, this process may lead to liquefaction when the effective stress decreases.
Although excess pore pressure is generally associated with liquefaction, it is not the direct cause
of liquefaction. In constant volume tests with no applied load, it is the decrease in contact forces
between particles that is responsible for the decrease in effective stress. The process is documented
by Dineash et al. (2004), who modeled similar tests (in which no change in pore pressure occurs)
using the distinct element method. Alternatively, in undrained, simple shear tests under normal
pressure, it is the irrecoverable reduction of porosity during cyclic compaction that generates pore
pressure and, consequently, a decrease in effective stress.
Dilation plays an important role in the liquefaction process. As soil densifies under repeated shear
cycles, grain rearrangement may be inhibited. Soil grains may then be forced to move up against
adjacent soil particles, causing dilation to occur, the effective stress to increase and the pore pressure
to decrease. Thus, densification is a self-limiting process.
The standard built-in constitutive models in FLAC do not model the liquefaction process directly.
Coupled dynamic-groundwater flow calculations can be performed with FLAC. However, by de-
fault, the pore fluid simply responds to changes in pore volume caused by the mechanical dynamic
loading; the average pore pressure remains essentially constant in the analysis.
There are many different models that attempt to account for pore pressure build-up, but they often do
it in an ill-defined manner because they refer to specific laboratory tests. In a computer simulation,
there will be arbitrary stress and strain paths. Consequently, an adequate model must be robust
and general, with a formulation that is not couched in terms that apply only to specific tests. The
following section describes a model that is simple, but that accounts for the basic physical process.
In Section 1.4.4.3, a review is presented of more comprehensive models that have been developed,
and in Section 1.4.4.4 we compare the simple and comprehensive models.
As mentioned in the previous section, in reality, pore pressure build-up is a secondary effect,
although many people seem to think it is the primary response to cyclic loading. The primary effect
is the irrecoverable volume contraction of the matrix of grains when a material is taken through a
complete strain cycle with the confining stress held constant. Since it is grain rearrangement rather
than grain volume change that takes place, the volume of the void space decreases under constant
confining stress. If the voids are filled with fluid, then the pressure of the fluid increases and the
effective stress acting on the grain matrix decreases. Note that pore pressures would not increase if
the test were done at constant volume; it is the transfer of externally applied pressure from grains
to fluid that accounts for the fluid-pressure increase.
Finn Model – Martin et al. Formulation
This mechanism is well-described by Martin et al. (1975), who also note that the relation between
irrecoverable volume-strain and cyclic shear-strain amplitude is independent of confining stress.
They supply the following empirical equation that relates the increment of volume decrease, vd ,
to the cyclic shear-strain amplitude, γ , where γ is presumed to be the “engineering” shear strain:
2
C3 vd
vd = C1 (γ − C2 vd ) + (1.92)
γ + C4 vd
vd vd
= C1 exp(−C2 ( )) (1.93)
γ γ
where C1 and C2 are constants with different interpretations from those of Eq. (1.92). In many
cases, C2 = 0.4 C1 , so Eq. (1.93) involves only one independent constant; however, both C1 and C2
have been retained for generality. In addition, a third parameter, C3 , sets the threshold shear strain
(i.e., the limiting shear-strain amplitude below which volumetric strain is not produced).
The shear induced volumetric strain for constant amplitude of cyclic shear strain predicted by this
formula is plotted versus number of cycles in Figure 1.58. The formula predicts an increase in
shear-induced (compactive) volumetric strain with the level of cyclic shear-strain. Also, for a given
strain amplitude, γ , the rate of accumulation decreases with the number of cycles.
The incremental volumetric behavior of the Finn/Byrne model (at the end of a cycle) may be
expressed as
where σm = σii /3 is the mean stress, p is pore pressure, α is Biot coefficient (= 1 for soil), K is the
drained bulk modulus of the soil and is the volumetric strain. Note that is positive in extension,
while vd is positive in compression. For undrained conditions, the change in pore pressure is
proportional to the change in volumetric strain:
p = −αM (1.95)
where M is Biot modulus. After substitution of Eq. (1.95) into (1.94), and solving for , we
obtain
σm − Kvd
= (1.96)
K + α2 M
If the fluid is very stiff compared to the solid matrix (M >>> K), Eq. (1.96) predicts no change
in volume. Further, using = 0 in Eq. (1.94) gives
Eq. (1.97) predicts a decrease in magnitude of effective stress with cyclic shear strain (produced
by an increase of shear induced compaction). Under conditions of constant stress, σm = 0, there
will be an increase in pore pressure:
p = Kvd (1.98)
that is proportional to the drained bulk modulus of the soil. While under free stress conditions, the
pore pressure will remain unchanged (p = 0), and the magnitude of the total stress will decrease
according to
Note that in both situations, the drained (tangent) bulk modulus, K, plays an important role in
determining the magnitude of the cyclic loading impact on effective stress. The Finn/Byrne model,
therefore, captures the main physics of liquefaction.
Finn Model Implementation in FLAC
FLAC contains a built-in constitutive model (named the Finn model)* that incorporates both
Eq. (1.92) and Eq. (1.93) into the standard Mohr-Coulomb plasticity model – it can be modi-
fied by the user as required. The use of Eq. (1.92) or Eq. (1.93) can be selected by setting parameter
ff switch = 0 or 1, respectively. As it stands, the model captures the basic mechanisms that can
lead to liquefaction in sand. In addition to the usual parameters (friction, moduli, etc.), the model
needs the four constants for Eq. (1.92), or three constants for Eq. (1.93). For Eq. (1.92), Martin et
al. (1975) describe how these may be determined from a drained cyclic test. Alternatively, one may
imagine using some trial values to model an undrained test with FLAC, and compare the results
with a corresponding laboratory test. The constants could then be adjusted to obtain a better match.
(See Example 1.15 for an example.) For Eq. (1.93), Byrne (1991) notes that the constant, C1 , can
be derived from relative densities, Dr , as follows:
* A FISH constitutive model is also provided for the Finn model (see “FINN.FIS” in Section 3 in the
FISH volume).
Further, using an empirical relation between Dr and normalized standard penetration test values,
(N1 )60 ,
1
Dr = 15(N1 )60
2
(1.101)
then,
C1 = 8.7(N1 )−1.25
60 (1.102)
C2 is then calculated from C2 = 0.4C1 in this case. Note that, as expected, the volumetric strain is
larger for smaller values of the blow count (compare Eq. (1.102) to Eq. (1.93)). Refer to Byrne
(1991) for more details.
In the Finn model there is logic to detect a strain reversal in the general case. In Martin et al.
(1975) (and most other papers on this topic), the notion of a strain reversal is clear, because they
consider one-dimensional measures of strain. In a two-dimensional analysis, however, there are
at least three components of the strain-rate tensor. By eliminating the volumetric strain, we have
a 2D “strain space.” In the general case of earthquake loading (where there is vertical as well as
horizontal motion), the trajectory of each element in this strain space is very complicated.
For example, Figure 1.59 shows the locus of strain states for a few seconds of typical earthquake
shaking, where e11 − e22 is plotted on the x-axis and 2e12 is plotted on the y-axis. What is a strain
cycle in this case? We adopt a formulation that degenerates to the conventional notion of strain
cycle when the amplitude on one axis is zero, or if there is a constant offset in strain. Note that
a simple magnitude measure (e.g., distance from the center point) is not good enough. Denoting
the two orthogonal strain measures as 1 and 2 , we accumulate strain, as follows, from FLAC ’s
“input” strain increments:
2 := 2 + 2e12 (1.104)
We use the following scheme to locate extreme points in strain space. Denoting the previous point
by superscript (◦ ), and the one before that with (◦◦ ), the previous unit vector, n◦i , in strain space is
computed:
√
z= vi vi (1.106)
vi
n◦i = (1.107)
z
where subscript i takes the values 1,2, and repeated indices imply summation.
.000
-1.000
-2.000
-3.000
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
The perpendicular “distance,” d, from the old point to a new point is given by the dot product of
the new vector with the previous unit vector:
We use the rule that d must be negative (so that the new strain segment corresponds to a reversal
compared to the previous segment). We then monitor the absolute value of d and do the following
calculation when it passes through a maximum, dmax , provided that a minimum number of timesteps
has elapsed (to prevent the reversal logic being triggered again on transients that immediately follow
a reversal). This threshold number of timesteps is controlled by the property named ff latency,
which is set to 50.0 in the runs reported here.
γ = dmax (1.109)
i◦ = i (1.111)
Note that there are two factors of 2 implied in Eq. (1.109) that cancel out: the shear strain is half
the excursion dmax , but γ is the engineering strain, which is twice FLAC ’s strain. Having obtained
γ , we insert it into Eq. (1.92) and obtain vd . We then update vd , as follows, and save it for use
in Eq. (1.92):
We also save one-third of vd , and revise the direct strain increments input to the model at the
next cycle:
vd
e11 := e11 + (1.113)
3
vd
e22 := e22 + (1.114)
3
vd
e33 := e33 + (1.115)
3
Note that FLAC ’s compressive strain increments are negative, and vd is positive. Hence, the
mean effective stress decreases.
The logic described above is certainly not perfect, but it seems to work in simple cases. However, the
user must verify that the algorithm is appropriate before applying it to real cases. In particular, the
number of “cycles” detected depends strongly on the relative magnitude of horizontal and vertical
motion. Hence, the rate of build-up of pore pressure will also be sensitive to this ratio. It may be
more practical to consider just the e12 component of strain for something like a dam, which is wide
compared to its height. Ultimately, we need better experimental data for volume changes during
complicated loading paths; the model should then be revised accordingly. One effect that has been
shown to be very important (see, for example, Arthur et al. 1980) is the effect of rotation of principal
axes: volume compaction may occur even though the magnitude of deviatoric strain (or stress) is
kept constant. Such rotations of axes occur frequently in earthquake situations. Another effect that
is not incorporated into the Finn model is that of modulus increase induced by compaction – it is
known that sand becomes stiffer elastically when compaction occurs by cyclic loading. It would
be easy for the user to add this modification to the “FINN.FIS” model.
The Finn model is implemented in FLAC with the MODEL command (i.e., MODEL finn). The code
must be configured for dynamic analysis (CONFIG dynamic) to apply the model. As with the other
built-in models, the property names are assigned with the PROPERTY command. The following
keywords are used to assign properties for the Finn model:
bulk bulk modulus
cohesion cohesion
dilation dilation angle in degrees
ff c1 Eqs. (1.92) and (1.93) constant C1
ff c2 Eqs. (1.92) and (1.93) constant C2
ff c3 Eq. (1.92) constant C3 , and threshold shear strain for Eq. (1.93)
ff c4 Eq. (1.92) constant C4
ff latency minimum number of timesteps between reversals
ff switch = 0 for Eq. (1.92), and 1 for Eq. (1.93)
friction friction angle in degrees
shear shear modulus
tension tension cutoff
JOB TITLE : .
0.000
-0.200
-0.400
1 2 3 4 5 6 7 8 9
Figure 1.60 Pore pressure (top) and effective stress (bottom) for shaking table,
using Eq. (1.92)
JOB TITLE : .
0.000
-0.200
-0.400
1 2 3 4 5 6 7 8 9
Figure 1.61 Pore pressure (top) and effective stress (bottom) for shaking table,
using Eq. (1.93)
settlement = (ydisp(1,jgp)+ydisp(2,jgp))/2.0
end
set dy_damp=rayl 0.05 20.0
his dytime
his pp i 1 j 2
his eff_stress
his settlement
his nstep 20
set ampl=0.005 freq=5.0
solve dyt=10.0
save Byrne.sav
;
new
conf dyn gw
; shaking table test for liquefaction
g 1 5
m finn
gen 0 0 0 5 50 5 50 0
fix x y j=1
fix x
set grav 10, flow=off
prop dens 2000 shear 2e8 bulk 3e8
prop fric 35 poros 0.5
water dens 1000 bulk 2e9 tens 1e10
ini pp 5e4 var 0 -5e4
ini syy -1.25e5 var 0 1.25e5
ini sxx -1e5 var 0 1e5 szz -1e5 var 0 1e5
prop ff_latency=50
; parameters for Martin formula
prop ff_switch = 0
prop ff_c1=0.8 ff_c2=0.79
prop ff_c3=0.45 ff_c4=0.73
; parameters for Byrne formula
;prop ff_switch = 1
def _setCoeff_Byrne
ff_c1_ = 8.7*exp(-1.25*ln(n1_60_))
ff_c2_ = 0.4/ff_c1_
ff_c3_ = 0.0000
end
set n1_60_ = 7
;_setCoeff_Byrne
;prop ff_c1=ff_c1_ ff_c2=ff_c2_
;prop ff_c3=ff_c3_
set ncwrite=50
def sine_wave
while_stepping
The state of practice for seismic analysis involving liquefiable materials is currently experiencing a
shift from using empirical schemes (first developed in the 1970s) to simulate liquefaction, to time-
marching numerical methods incorporating liquefaction constitutive models currently at various
stages of development. There are several ways in which liquefaction behavior is included in
numerical methods, ranging from total-stress empirical schemes to estimate liquefaction conditions
(e.g., the UBCTOT model – see Beaty and Byrne 2000), to simple effective-stress shear-volume
coupling schemes (e.g., the Finn model described in Section 1.4.4.2 and the URS model – see
Dawson et al. 2001), to more comprehensive constitutive models (e.g., the UBCSAND model –
see Byrne et al. 1995; and bounding surface models such as the model described in Wang 1990
and the model described in Papadimitriou et al. 2001) that address cyclic shearing via kinematic
hardening.
To help practicing engineers choose a procedure best-suited to their needs, a selection of approaches
is outlined below, ranging from simple to elaborate in terms of complexity and model parameter
determination. This is not a comprehensive list, but is a selection that illustrates the different types
of liquefaction models that have been developed, and have been used in FLAC.* One important
factor to keep in mind is that, in engineering practice, the use of a very complicated model for
liquefaction analyses is often hardly justified, considering the many uncertainties with respect to
soil properties and earthquake motions, and the numerous approximations which must be made
(see Dawson et al. 2001). Before describing the individual models, it will be helpful to review the
current state of practice for liquefaction analysis. (See Byrne et al. 2006 for further discussion on
state-of-practice analysis.)
State of Practice – The standard practice approach for liquefaction analysis of earthquake loading is
based on a total-stress analysis in which it is assumed that the liquefiable soil remains undrained at
the in-situ void ratio (Byrne and Wijewickreme 2006). Typically, this analysis approach is divided
into three steps:
1. Triggering Evaluation: Typically, an equivalent-linear elastic, dynamic analysis (such
as SHAKE) using strain-compatible moduli and damping is conducted for the design
earthquake. The cyclic stress ratio (CSR)** is evaluated from the numerical simulation
and compared to the value of cyclic resistance that the soil has because of its density
* The liquefaction models discussed in this section have been implemented in FLAC either as FISH
or C++ user-defined constitutive models. Contact the authors of these liquefaction models in order
to receive additional information about the model, or to inquire about receiving a copy of the model.
Itasca does not provide technical support for these models.
** Cyclic Stress Ratio: ratio of maximum dynamic shear stress to the initial vertical effective stress
prior to the earthquake at a specific location.
(CRR),*** derived from empirical curves. A factor of safety against triggering liquefac-
tion is evaluated using the ratio of CRR and CSR (e.g., see Byrne and Anderson 1991,
and Youd et al. 2001).
2. Flow Slide Assessment: Post-liquefaction (undrained) strengths are assigned in zones pre-
dicted to liquefy from the triggering evaluation analysis, and a standard limit-equilibrium
analysis is carried out to evaluate the factor of safety against a flow slide. Post liquefac-
tion strengths may be derived from penetration resistance (corrected blow count (N1 )60 )
using empirical charts (e.g., see Seed and Harder 1990, and Olson and Stark 2002).
3. Seismic Displacements: Displacements are evaluated using the Newmark approach (see
Newmark 1965). In this step, the potential sliding block of soil is modeled as a rigid
mass resting on an inclined plane. The design time history of acceleration is applied at
the base, and the equation of motion is solved to obtain the displacement of the mass
caused by the shaking.
The main shortcomings of the standard practice approach are that the three aspects of liquefaction
(triggering, flow slide and deformation) are treated sequentially, when in reality they may interact
locally in various zones of the soil structure and affect the overall behavior of the soil mass. Also,
no direct account is made of excess pore-pressure redistribution and dissipation.
Total-Stress Synthesized Procedure
The synthesized procedure of Beaty and Byrne (2000) uses FLAC and the UBCTOT constitutive
model to combine the three steps of the standard practice approach (triggering, flow slide and
estimate of liquefaction-induced displacements) into one single analysis. The procedure, which
assumes undrained behavior, uses a (two-dimensional) total-stress approach to liquefaction analysis,
and relies on adjustment of liquefied element properties (stiffness and strength) at the instant of
triggering of liquefaction. The main features of the UBCTOT model are summarized below.
A seismic analysis using the synthesized procedure starts from a static state of equilibrium for
the FLAC model. The seismic analysis is conducted in total-stress space. UBCTOT uses Mohr-
Coulomb elasto-plastic logic with zero friction and a value of cohesion equal to the undrained
shear strength, in combination with Rayleigh damping. The elastic shear modulus is assigned a
value of Gmax multiplied by a Modulus Reduction Factor (MRF). As opposed to equivalent-linear
methods, this approach is not iterative, and appropriate values of MRF and damping are selected at
the start of the seismic analysis. Triggering of liquefaction is based on changes of shear stress on
the horizontal plane, τxy . The irregular shear-stress history caused by the earthquake is interpreted
in each FLAC zone as a succession of half-cycles with the contribution to triggering determined by
the maximum value of τcyc , defined as the difference between τxy and the initial horizontal shear-
stress prior to earthquake loading (i.e., the static bias). A cumulative damage approach is used
to combine the effects of each half-cycle. The approach converts the nonuniform history into an
equivalent series of uniform stress cycles with amplitude equal to τ15 (i.e., the value of τcyc required
to cause liquefaction in 15 cycles, which is approximately the number of cycles in a magnitude 7.5
*** Cyclic Resistance Ratio: commonly taken as the value of CSR that causes liquefaction in 15 cycles
of dynamic loading.
earthquake). This is done using an empirical chart giving the cyclic stress ratio (CSR) versus cycles
to liquefaction. Several property changes are imposed when liquefaction is detected in a FLAC
zone: a residual shear strength is assigned; a reduced loading stiffness is used; and unloading uses
a stiffer modulus than loading, according to a bilinear stiffness model. Also, a hydrostatic stress
state is imposed when a zone experiences a shear-stress reversal. Finally, reduced viscous damping
is assigned in a liquefied zone. The constitutive model also has logic to account for anisotropy in
stiffness and strength. See Beaty (2001) for additional information.
The UBCTOT model removes some of the limitations associated with the sequential approach to
problem solving used in the state-of-practice procedure, while relying on similar empirical charts
for triggering of liquefaction and residual strength.
Some of the drawbacks of the model are:
• the use of equivalent modulus ratio that may not capture the pre-liquefaction phase well;
• the cyclic shear stresses are accounted for on the horizontal plane only;
• the simplified manner in which the undrained shear strength is specified;
• pore pressure is not taken into account explicitly; and
• liquefaction due to monotonic loading is not considered.
Loosely Coupled Effective-Stress Procedure
The URS model is a loosely coupled effective-stress constitutive model to generate pore pressure
from shear stress cycles using the Seed cyclic stress approach (Seed and Idriss 1971). This model
is built around the standard FLAC Mohr Coulomb model. The (two-dimensional) model counts
shear stress cycles by tracking the shear stress acting on horizontal planes (τxy ) and looking for
stress reversals. The cyclic stress ratio (CSR) of each cycle is measured, and this is used to compute
the incremental “damage” that is then translated into an increment of excess pore pressure. The
procedure is “loosely coupled” because pore pressures are only computed after each 1/2 cycle
of strain or stress as the analysis proceeds. The model incorporates residual strength (a critical
parameter for seismic stability analyses) by using a two-segment failure envelope consisting of a
residual cohesion value and zero friction angle that is extended to meet with the traditional Mohr-
Coulomb failure envelope.
The model is simple, robust and practice-oriented; it is based on the widely accepted cyclic-stress
approach with input parameters readily obtainable from routine field investigations. (Note that
liquefaction due to monotonic loading is not considered.) A disadvantage of the model is that
liquefaction-induced consolidation settlements are not captured, because the actual physical mech-
anism of liquefaction, whereby pore pressure is generated through contraction of the soil skeleton,
is bypassed. The model is applicable to problems where slope movements due to reduced shear
strength are the main concern (such as seismic stability of dams, and waterfront retaining struc-
tures), while shaking-induced consolidation settlements are of secondary importance. See Roth et
al. (1991), Inel et al. (1993), Roth et al. (1993), and Perlea et al. (2008) for some field applications.
The URS model is similar to the built-in Finn model (described in Section 1.4.4.2), which is also
considered a loosely coupled effective-stress model. The primary difference is that in the Finn
model, the volumetric strains induced by cyclic loading are evaluated based on an experimental
curve of irrecoverable volumetric strain versus number of constant amplitude cycles. Pore pressures
are then generated from these volumetric strains, as well as from contraction of the soil skeleton.
Also, the Finn model in FLAC, at present, does not include a post-liquefaction residual strength.
Further discussion and numerical testing of the Finn model is presented in Section 1.4.4.4.
Fully Coupled Effective-Stress Procedure
The UBCSAND model is a fully coupled (kinematic hardening) effective-stress constitutive model
to predict seismic response and liquefaction of cohesionless soils in plane strain problems. The
model uses an elasto-plastic formulation, based on an assumed hyperbolic relation between stress
ratio and plastic shear strain, similar to the Duncan and Chang (1970) formulation. It is applicable
for monotonic as well as cyclic loading (e.g., see Byrne et al. 2003, 2006).
The model implementation is a modified form of the built-in Mohr-Coulomb model in FLAC that
accounts for a strain-hardening frictional behavior, neglects cohesion, and applies to plane strain
conditions. The hardening law is a hyperbolic function of plastic shear strain. Unloading is assumed
to be nonlinear elastic, with bulk and shear modulus as functions of mean (in-plane) effective stress.
Stress reversal is detected by a change of sign in horizontal shear stress, τxy . Reloading is elasto-
plastic, with the yield locus reset to the value at the reversal point. Plastic flow is non-associated;
the logic is based on a variation of Rowe stress-dilatancy theory. According to this theory, there is
a constant-volume stress ratio, φcv , below which the material contracts (i.e., for mobilized friction,
φm , smaller than φcv ), while for higher stress ratios (i.e., for φm > φcv ), the material dilates. The
effect of relative density is addressed through the choice of material properties. Most properties
are calibrated to field experience as well as centrifuge tests, and are conveniently related to blow
count, (N1 )60 .
The model is able to capture the stiff pre-liquefaction stage, the onset of liquefaction at the appro-
priate number of cycles, and the very much softer post-liquefaction response observed in cyclic,
simple shear-constant volume tests.
The coupled effective-stress approach corrects many drawbacks of the previous approaches. Al-
though most parameters are related to blow count, and rely on a growing body of data and experience,
it is always good practice to check on model parameters for each layer (using numerical simulation
of a simple shear test) to verify that, if it has to liquefy in N cycles according to the field data
during dynamic loading, it will. Also, because comparison with standard procedures may not be
straightforward, it is recommended that the model be used with supervision from an experienced
practitioner. With time, and with the increase of its usage, the model should become more prominent
and be used for problems ranging from simple to complex with little effort.
The primary disadvantages are that (1) the logic for detection of stress reversal is based on hor-
izontal shear stress only, and (2) the formulation applies only to two-dimensional analysis. See
Section 1.4.4.4 for additional discussion and numerical testing of the UBCSAND model.
Comparisons with centrifuge experiments have been made (Andrianopoulos et al. 2006a), and the
ability of the model to study a practical problem of geotechnical earthquake engineering has been
demonstrated (Andrianopoulos et al. 2006b). The disadvantages are the model calibration, which
is a rather tedious procedure and requires a test database not readily available in most cases, and
the long computational time required for the solution of practical problems. Also, comparison to
standard practice is not straightforward. Application of the model to study boundary-value problems
at the field scale is not recommended at this time without the assistance of the model developer,
for model calibration, interpretation of results, and eventual support for issues related to numerical
implementation.
FLAC is able to follow the full nonlinear stress/strain behavior of soil or rock, provided that a
suitable constitutive model is provided. In particular, a comprehensive constitutive soil model for
dynamic liquefaction analysis should exhibit hysteresis loops (and, hence, damping) and progressive
volume change with continuing cyclic shearing (to produce liquefaction). One such model is the
UBCSAND model (described previously in Section 1.4.4.3). Alternatively, simple formulations
for hysteretic damping (as discussed in Section 1.4.3.4) and liquefaction modeling (as discussed
in Section 1.4.4.2) can be combined in a simple elastic/plastic model (such as Mohr-Coulomb) to
produce comparable liquefaction behavior.
In this section, the UBCSAND model is compared to a simple liquefaction scheme (Finn/Byrne
liquefaction formulation in the Mohr-Coulomb model) incorporating hysteretic (Hardin/Drnevich
model) damping. To compare the predictions of the Finn/Byrne model to the UBCSAND model,
we consider a constant amplitude, constant volume, cyclic-shear test. There is no applied load;
therefore, the pore pressure remains constant. First, we compare the volumetric behavior of the
two models, and then we compare the energy dissipation.
The FLAC grid for this test consists of one zone with unit dimensions. The initial conditions of the
test are:
σyy = −150.0
σxx = −100.0
σzz = −125.0
p = 50.0
The material properties for the UBCSAND model are:
φcv = 33
(N1 )60
φf = φcv + 10
Pa = 100.0
(N1 )60
where φf is the ultimate friction angle, Pa is the reference pressure, and 10 is the normalized
standard penetration test value.
The Finn/Byrne model properties used for the comparison are:
φ = φf
ψ =0
K = Kini
U BCSAN D
G = GU BCSAN D
ini
The Finn/Byrne elastic properties, K and G, are selected to be equal to the values of tangent bulk
and shear moduli recorded after the first step of calculation in the UBCSAND model simulation.
The simulations are conducted for four values of (N1 )60 (5, 10, 20 and 30), and f or two values of
shear strain amplitude (0.01% and 0.02%).
Volumetric behavior – The volumetric behavior of the Finn/Byrne model in a constant-volume
cyclic, shear test of constant amplitude under no applied load is calculated to be (see Eq. (1.99))
Integration of the Finn/Byrne formula (Eq. (1.93)) with respect to cycle number, for constant shear
strain amplitude gives
γ C1
vd = ln(1 + 0.4n) (1.117)
0.4
where n is the number of cycles. After substitution of Eq. (1.117) into Eq. (1.116), and some
manipulation, we obtain
σm 1
= ln(1 + 0.4n) (1.118)
Kγ C1 0.4
Thus, in the framework of the Finn/Byrne model, and for the proposed cyclic test, the dimensionless
stress measure appearing on the left side of Eq. (1.118) is predicted to be solely a function of the
number of cycles.
For comparison of model behavior, we use the following dimensionless pressure measure:
P
− (1.119)
Kγ C1
+ σ
σxx
yy
P =− (1.120)
2
and P is the difference between the current and initial value of P for the test. K is a constant
equal to the UBCSAND constant Kini U BCSAN D , and C is related to the blow count, as defined in
1
Eq. (1.102).
The Finn/Byrne model and UBCSAND model pressure predictions are compared for four different
values of blow count in Figures 1.62 through 1.65. Also, in each figure, two different values of
cyclic shear strain are considered.
The plots in Figures 1.62 through 1.64 show that for the Finn/Byrne model, the dimensionless
pressure measure is independent of shear-strain modulus. This is expected, provided the response
remains elastic as in Figures 1.62 through 1.64. This is in contrast to the UBCSAND behavior,
which shows higher pressure measure for the higher amplitude of shear strain. This difference is
attributed in part to the value of bulk modulus, which is constant for the Finn/Byrne model, and
evolving with stress level, P , and other quantities in the UBCSAND formulation (the K value
used for scaling pressure is the constant bulk-modulus value assigned in the Finn/Byrne model, and
U BCSAN D ).
equal to Kini
Note that both the Finn/Byrne and UBCSAND pressure plots for γ = 0.2% exhibit a plateau at a
value of dimensionless pressure of approximately 7 in Figure 1.65. This plateau corresponds to
yielding of the sample at the ultimate value of friction angle φf .
Figure 1.62 Pressure response in cyclic shear test for Finn/Byrne and UBC-
SAND models – (N1 )60 = 30
Figure 1.63 Pressure response in cyclic shear test for Finn/Byrne and UBC-
SAND models – (N1 )60 = 20
Figure 1.64 Pressure response in cyclic shear test for Finn/Byrne and UBC-
SAND models – (N1 )60 = 10
Figure 1.65 Pressure response in cyclic shear test for Finn/Byrne and UBC-
SAND models – (N1 )60 = 5
Damping behavior – The Finn/Byrne model, which builds on the logic of a Mohr-Coulomb model,
naturally accounts for damping arising from plastic flow (see Section 1.4.3.11). However, the model
does not account automatically for cyclic energy dissipation during the elastic response. Some
additional form of damping must be provided, in the form of Rayleigh or hysteretic damping. In
this section, we apply hysteretic damping (with the Hardin/Drnevich formulation) to the Finn/Byrne
model, and compare the damping ratio and shear modulus reduction factor from a cyclic shear test,
at constant amplitude and constant volume, to those derived from the same test using the UBCSAND
model.
We note that although the Finn/Byrne scheme in FLAC is intended for liquefaction modeling, it
may still affect the response of a constant-volume shear test because a reduction in effective stress
results in a reduction in shear strength.
The FLAC model, initial conditions and test properties are the same as those used previously for
the comparison of volumetric responses. The values of shear and bulk moduli for the Finn/Byrne
model are chosen to be equal to the values of tangent shear and bulk moduli recorded after the
first calculation step in the UBCSAND model simulation. Also, friction is set equal to the ultimate
value specified for the UBCSAND model, and dilation is zero.
The simulations are conducted for four values of blow count: (N1 )60 = 5, 10, 20 and 30. Six
values of shear strain amplitude are considered, ranging from 0.01% to 0.05% at an interval of
0.01%. The damping ratio and modulus-reduction ratio are calculated from the second loop of the
constant-amplitude shear test. A convention is adopted for definition of the ratios because the strain
loop is not “closed” for all cases investigated. Figure 1.66 shows a typical loop with the parameters
used for the definition of the ratios.
The energy dissipated in the cycle, W , is calculated as the (algebraic) area under the loading
curve (from τ i to τ m ) minus the (algebraic) area under the unloading curve (from τ m to τ f ). The
“peak energy” in the cycle of amplitude, γ , is evaluated using W = τ av γ /2. The damping ratio is
then evaluated using the small ratio formula (i.e., D = W/(4π W )). The shear modulus ratio is
estimated using the ratio of (τ av + τ m )/(2γ ) to the initial value of shear modulus, G. Also, for the
model comparison, the Hardin/Drnevich parameter, γref , is calibrated to match the damping ratio
of the UBCSAND model at 0.03% strain.
Figure 1.66 Definition of parameters used for damping and shear modulus
measures
The results for damping ratio and modulus ratio for the four different values of blow count are
shown in Figures 1.67 through 1.70.
Figure 1.67 Damping ratio (%) and G/Gmax versus shear strain (%) in cyclic
shear test – (N1 )60 = 30
Figure 1.68 Damping ratio (%) and G/Gmax versus shear strain (%) in cyclic
shear test – (N1 )60 = 20
Figure 1.69 Damping ratio (%) and G/Gmax versus shear strain (%) in cyclic
shear test – (N1 )60 = 10
Figure 1.70 Damping ratio (%) and G/Gmax versus shear strain (%) in cyclic
shear test – (N1 )60 = 5
The results of the simulations show similar damping trends for the Finn/Byrne model with hysteretic
damping and the UBCSAND model for a shear cycle of constant amplitude and constant volume.
Also, the values of damping ratio and shear-modulus reduction ratio compare rather well given the
complex elastic/plastic behavior involved. (The complexity is unavoidable because of the difficulty
in exercising both models in only the elastic range.)
The main advantages of using the Finn/Byrne model combined with hysteretic damping are: (1) the
model reproduces the main mechanisms of liquefaction; (2) it is a robust approach that works both
in two-dimensional and three-dimensional analysis (because the logic for stress reversal is valid in
both 2D and 3D); and (3) it gives a fair comparison to UBCSAND model behavior.
The inconveniences of the Finn/Byrne model with hysteretic damping are: (1) the pore-pressure
generation is not smooth (because the pore-pressure update occurs when a half-cycle is completed);
(2) the approach does not apply to monotonic loading; (3) the elastic stiffness, K, does not evolve
with effective stress level; (4) the damping ratio is rather high during plastic flow; (5) the approach
applies to over-consolidated soil (because the Mohr-Coulomb envelope does not evolve); and (6)
the response may be too conservative (because dilation is assumed constant).
The UBCSAND model is a more comprehensive approach. The advantages of this approach are:
(1) the correct physics are produced, based on laboratory test behavior; (2) the model works for
monotonic and cyclic loading; (3) it is calibrated extensively with field case histories; (4) it is flexible
and easy to use (e.g., most of the material properties are related to blow count); (5) the evolution
of physical variables, including volumetric strain, is continuous; and (6) dilation is accommodated
for stress states above a constant volume friction angle.
The disadvantages of the UBCSAND model are: (1) the logic for stress reversal is based on
horizontal shear stress only; and (2) the formulation of the model applies only to two-dimensional
analysis.
Approaches for modeling dynamic problems are described in the following three subsections. The
first subsection (Section 1.5.1) discusses procedures for dynamic, mechanical-only calculations.
The second subsection (Section 1.5.2) discusses dynamic coupled analyses, including the effect
of groundwater on the dynamic response. The first two subsections include simple examples to
illustrate the modeling approaches. In the third subsection (Section 1.5.3), recommended steps are
given for a full-scale seismic analysis of an earth structure with the soil behavior represented by a
nonlinear effective-stress material model including liquefaction behavior. Section 1.6 contains an
example that illustrates the application of these steps.
Dynamic analysis is viewed as a loading condition on the model, and as a distinct stage in a modeling
sequence, as described in Section 3.5 in the User’s Guide. A static equilibrium calculation always
precedes a dynamic analysis. There are generally four components to the dynamic analysis stage:
1. Ensure that model conditions satisfy the requirements for accurate wave trans-
mission (by adjusting zone sizes with the GENERATE command – see Sec-
tion 1.4.2). This check must be performed even before the static solution
is performed, because gridpoints must not be relocated by the user after the
calculation starts.
2. Specify appropriate mechanical damping, representative of the problem ma-
terials and input frequency range. Use the SET dy damp or INITIAL dy damp
command, as described in Section 1.4.3.
3. Apply dynamic loading and boundary conditions (by using the APPLY and
INTERNAL commands – see Section 1.4.1). A given time history may need to
be filtered in order to comply with the requirements noted in Section 1.4.2.
4. Set up facilities to monitor the dynamic response of the model (by using the
HISTORY command).
The procedure for dynamic analysis is illustrated by Example 1.16, and then in Example 1.17. The
model is greatly simplified for rapid execution, but it still illustrates the steps in a dynamic analysis.
Consider the problem of a structure built at the top of a soil slope. The slope is initially stable under
the applied structural loading. The data file for the initial static loading stage is given below. The
stress state of the model at equilibrium is shown in Figure 1.71.
LEGEND
1.400
11-Jun-08 9:51
step 1421
-1.111E+00 <x< 2.111E+01
-4.611E+00 <y< 1.761E+01
1.000
Grid plot
0 5E 0
Principal stresses
Max. Value = 6.014E+04 0.600
Min. Value = -7.361E+05
0 2E 6
Beam plot
0.200
-0.200
The slope material is simulated as a strain-softening soil; the cohesion weakens as a function of
plastic strain. This example demonstrates the development of slope failure as a consequence of loss
of material strength following dynamic loading.
The four steps identified previously are now followed to prepare for dynamic analysis:
1. Check Wave Transmission – The dynamic loading for this problem is a sinusoidal
velocity wave applied at the base of the model in the x-direction. The wave has an
amplitude of 1 m/sec and a frequency of 10 Hz.
Based upon the elastic properties for this problem, the compressional and shear wave
speeds are (from Eqs. (1.16) and (1.17))
Cp = 840 m/sec
Cs = 485 m/sec
The largest zone dimension for this model is 1 m. Based upon Eqs. (1.37) and (1.29),
the maximum frequency which can be modeled accurately is
Cs Cs
f = = ≈ 48 Hz
λ 10 l
Therefore, the zone size is small enough to allow velocity waves at the input frequency
to propagate accurately.
2. Specify Damping – The plastic flow associated with the strain-softening model can dis-
sipate most of the energy and, hence, tends to make the selection of damping parameters
less critical to the outcome of the analysis. This model was run with no damping, and
with a small amount of Rayleigh damping (5%, at the natural frequency), to evaluate the
influence of damping.
To estimate the lowest natural frequency for this model (used as a Rayleigh damping
parameter), Example 1.16 is run with SET dyn on and with no damping. A plot of velocity
history (Figure 1.72) indicates that the dominant natural frequency of the system is
approximately 25 Hz. This is unrealistically high, but the value reflects the simplifications
made for this example.
3. Apply Dynamic Loading and Boundary Conditions – The APPLY command is used
with the hist keyword to specify the dynamic input. The FISH function wave supplies the
history (a sinusoidal wave of 1 m/sec amplitude, 10 Hz frequency and 0.25 sec duration).
Free-field boundaries are invoked along the left and right boundaries to absorb energy.
4. Monitor Dynamic Response – Three velocity histories are located in the model: the
first at the position of the applied input wave; the second along the slope face; and the
third within the grid.
The data file for the dynamic stage is reproduced in Example 1.17:
hist dytime
hist xvel i=8 j=7
hist xvel i=8 j=1
hist xvel i=18 j=10
solve dytime = 0.5
save stage3.sav
The response of the slope at 0.5 sec (0.25 sec after the dynamic wave is stopped) is shown in
Figure 1.73. A rotational failure mechanism develops beneath the structure, resulting from the loss
of cohesive strength. The velocity histories in Figure 1.74 illustrate the input history (at i = 8, j = 1),
the continuous movement at the slope face (at i = 8, j = 7), and the gradual return to equilibrium
at a position remote from the slope (at i = 18, j = 10).
The response is similar for both no damping and 5% damping, although velocities are lower for
the damped case. To see this, Example 1.17 may be rerun with the SET dy damp rayl and SET
dy damp struct rayl commands enabled (i.e., with the comment characters removed). The results
are shown in Figures 1.75 and 1.76. Note, also, that the structural damping has a minor influence
on these results; if only SET dy damp rayl is applied, the results are nearly the same as those shown
in Figures 1.75 and 1.76.
JOB TITLE : .
-1.000
-2.000
-3.000
10 20 30 40 50
-02
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
LEGEND 1.400
11-Jun-08 11:55
step 10849
Dynamic Time 5.0003E-01
-1.110E+00 <x< 2.119E+01
1.000
-4.986E+00 <y< 1.731E+01
Grid plot
0 5E 0
0.600
Beam plot
Velocity vectors
max vector = 5.024E+00
0 1E 1
0.200
-0.200
JOB TITLE : .
LEGEND
1.000
11-Jun-08 11:55
step 10849
Dynamic Time 5.0003E-01 0.500
-2.000
-2.500
-3.000
5 10 15 20 25 30 35 40 45 50
-02
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.74 Velocity histories at base, slope face and remote from slope
– undamped simulation
LEGEND 1.400
11-Jun-08 11:56
step 48573
Dynamic Time 5.0000E-01
-1.123E+00 <x< 2.116E+01
-4.799E+00 <y< 1.749E+01 1.000
Grid plot
0 5E 0
Beam plot 0.600
Velocity vectors
max vector = 1.733E+00
0 5E 0
0.200
-0.200
JOB TITLE : .
LEGEND
1.000
11-Jun-08 12:57
step 48573
Dynamic Time 5.0000E-01 0.500
-2.000
-2.500
-3.000
10 20 30 40 50
-02
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.76 Velocity histories at base, slope face and remote from slope
– with Rayleigh damping for soil and structure
Prior to performing a dynamic simulation with groundwater present, an equilibrium state must be
obtained. This consists of several stages, which are illustrated by an analysis of the earthquake
response of an idealized dam resting on a foundation (Figure 1.77 shows the original shape of the
dam). Note that this dynamic example is not very realistic, as it subjects the dam to a few cycles of
very high amplitude; however, it runs quickly and illustrates some important points.
First, the foundation is set in place and brought to equilibrium. The data for this initial stage is
given in Example 1.18.
1.750
LEGEND
11-Jun-08 13:15
step 181
-1.667E+01 <x< 3.167E+02 1.250
Grid plot
0.750
0 1E 2
0.250
-0.250
-0.750
1. Gravity Compaction of the Dam – We create the dam in a single placement of material,
which is saturated (see Example 1.19 for the data file). If we are not interested in the
time of settlement, we can set the bulk modulus of water to zero for this stage, so that
numerical convergence is rapid.
3. Allow Phreatic Surface to Develop – Again, assume that we are only interested in the
final flow pattern, not in the time it takes to occur. (If consolidation time is important,
then consult Section 1.8.6 in Fluid-Mechanical Interaction.) To allow rapid adjustment
of the phreatic surface, we set the fluid modulus to a low value (1 MPa, compared with
the “real” value of 2 × 103 MPa). We also do the fluid calculation and the mechanical
adjustment separately (since the fully coupled solution takes much longer) – i.e., for this
stage, flow=on and mech=off. The tensile limit for water is set to zero so that a phreatic
surface develops. Pore pressure is applied to the upstream face, with fixed saturation of
1.0; on the other surfaces, pore pressure is fixed at its default value of zero.
4. Mechanical Adjustment to New Flow Field – Once the equilibrium flow field is estab-
lished, we need to do a final mechanical adjustment, because: (a) some of the material
is now partially saturated so the gravity loading is less; and (b) the effective stress has
changed, which may cause plastic flow to occur. During this stage, we prevent fluid flow
and pore pressure changes (setting fluid modulus temporarily to zero), since we are not
concerned with the consolidation process here.
water bulk=2e3
save step4.sav
We now have a system that is in mechanical and fluid equilibrium, ready for dynamic
excitation; the fluid modulus is at the value for pure water (no entrained air). Note
that the separation into several stages (just fluid or just mechanical) was done to reduce
calculation time. The fully coupled simulation could be done if required.
5. Apply Dynamic Excitation to Dam – The dynamic simulation may now be done. What
is being modeled is the response of the dam and its trapped groundwater. It is assumed
here that no fluid flow occurs, and that no pore-pressure generation occurs due to particle
rearrangement. However, pore pressure changes do occur because of the dynamic volume
changes induced by the seismic excitation. The excitation is by rigid sinusoidal shaking
at the base of the foundation.
Figure 1.78 shows the deformed grid, and Figure 1.79 shows the pore-pressure history
in zone (4,3). The dashed lines represent the original shape.
1.700
FLAC (Version 6.00)
LEGEND 1.500
11-Jun-08 13:22
step 44509 1.300
Flow Time 4.0566E+11
Dynamic Time 1.0000E+01
7.500E+01 <x< 2.500E+02
0.000E+00 <y< 1.750E+02 1.100
0 5E 1 0.700
0.500
0.300
0.100
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
0.900 1.100 1.300 1.500 1.700 1.900 2.100 2.300
(*10^2)
JOB TITLE : .
3.000
2.000
1.000
1 2 3 4 5 6 7 8 9
The dynamic simulation given previously may be repeated with the Finn constitutive model, which
replaces the Mohr-Coulomb model in the dam; the stresses remain, even though the model is
replaced. The constants C1 , C2 , C3 and C4 are given the values that Martin et al. (1975) provide
in their paper (although the condition C1 C2 C4 = C3 , mentioned earlier, is violated slightly).
The same quantities as before are plotted – see Figures 1.80 and 1.81. We now have considerable
pore pressure build-up, and there is much larger horizontal movement in the dam; undoubtedly,
liquefaction is occurring.
1.700
FLAC (Version 6.00)
LEGEND 1.500
11-Jun-08 13:47
step 44319 1.300
Flow Time 4.0566E+11
Dynamic Time 1.0000E+01
7.500E+01 <x< 2.500E+02
0.000E+00 <y< 1.750E+02 1.100
0 5E 1 0.700
0.500
0.300
0.100
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
0.900 1.100 1.300 1.500 1.700 1.900 2.100 2.300
(*10^2)
JOB TITLE : .
3.000
2.000
1.000
1 2 3 4 5 6 7 8 9
Although very little dissipation of pore pressure is likely during seismic excitation in most structures,
FLAC is able to carry out the groundwater flow calculation in parallel with the dynamic calculation.
When both dynamic and groundwater options are selected together (SET flow=on dyn=on), the two
timesteps (groundwater and dynamic) are forced to be equal; the overall timestep is set to whichever
is the smallest. The previous example may be repeated with the addition of the following lines (the
Finn model generates pore pressures and the flow logic dissipates them):
prop perm 0.5 i 3 6 j 3 6
set flow=on
The permeability is unrealistically high, for demonstration purposes. The resulting plot of pore
pressure in zone (4,3) is shown in Figure 1.82. There is clear evidence of pore pressures dropping
off in the later stages of the simulation, but the situation is complicated because pressures generated
in other zones appear to flow into zone (4,3) initially. Note that it is possible – in principle – for the
pore pressure in a particular zone to increase when dissipation is allowed, if the surrounding zones
contribute excess fluid.
JOB TITLE : .
3.000
2.000
1.000
2 4 6 8 10
Figure 1.82 Pore pressure history at zone (4,3) – finn model, with dissipation
A seismic analysis of an earth structure can be quite complicated, especially if there is a potential
for dynamic liquefaction of the soil materials. The following steps are recommended to perform
a seismic analysis using a nonlinear effective-stress material model. In this procedure, the non-
linear stress/strain behavior is represented by a “simple” elastic/plastic (Mohr-Coulomb) model,
with formulations added to simulate liquefaction and hysteretic damping of the soils during cyclic
loading. In general, these steps for performing a seismic analysis can also be applied for a more
comprehensive liquefaction model.
The steps are listed and described below. Section 1.6 provides an example application of this
procedure for a seismic analysis of an embankment dam.
1. Determine representative static and dynamic material characteristics – The
effective-stress analysis approach in FLAC is performed as a fully coupled mechanical-
fluid flow analysis, which requires the selection of CONFIG gw, and assignment of drained
stiffness and strength material properties, and the unsaturated (“dry”) density (or unit
weight). In some situations, only undrained properties may be available. However,
it is possible, with certain assumptions, to estimate drained properties from undrained
properties. See Section 1.9.4 in Fluid-Mechanical Interaction for guidelines on relating
undrained properties to drained properties.
In addition, the fully coupled effective-stress calculation requires that the water bulk
modulus be specified explicitly. The water bulk modulus must be selected carefully for
this type of analysis. The behavior of the model depends on the stiffness ratio, Rk , as
defined by Eq. (1.67) in Section 1.8.1 in Fluid-Mechanical Interaction, and repeated
here:
Kw /n
Rk = (1.121)
K + 4G/3
where Kw is the water bulk modulus, n is the porosity, and K and G are the bulk and
shear moduli of the unsaturated soil. There is a temptation to decrease the water bulk
modulus arbitrarily in order to increase the timestep (and reduce the simulation time).
However, there are two cases to consider when selecting Kw :
1. If the calculated value of Rk based upon the given soil properties and an
assumed value of Kw = 4.18×107 psf, or 2 GPa (pure water), is greater than
20, then the water bulk modulus can be reduced such that Rk = 20 without
affecting the results significantly (and reducing the simulation time).
2. If the calculated value for Rk , using the actual water bulk modulus, is less than
20, then that value of water bulk modulus should be used.
The water bulk modulus can be made to depend on the local elastic moduli, and can be
specified differently for different soil units, provided that the two conditions listed above
are satisfied.
The material response to dynamic cyclic loading is typically quantified by a shear modulus
degradation curve and a damping ratio curve. Representative curves should be selected
for each material in order to provide an accurate representation of wave attenuation and
energy dissipation during dynamic loading. In addition, it may be necessary to make the
curves depth-dependent to make the energy dissipation more realistic (e.g., see Darendeli
2001).
As discussed in Section 1.4.4.3, a comprehensive material model to represent liquefaction
behavior can be quite complex. The simplest forms of liquefaction models, as discussed
in Section 1.4.4.2, are based on a volume-change modification to the Mohr-Coulomb
model. In this case, the only additional property is the blow count associated with the
material. For example, see the Finn/Byrne formula defined by Eq. (1.93) and Eq. (1.102).
2. Evaluate seismic motion characteristics – A design earthquake ground motion is often
provided as an acceleration record. It is typically an outcrop motion that is often recorded
at a rock outcrop. It is important to know the location of the recorded motion because
the motion may need to be modified for input to the FLAC model. The deconvolution
analysis that is performed to obtain the appropriate input motion is discussed below in
Step 6.
The characteristics of the seismic motion should always be checked because these charac-
teristics can influence the model conditions. As discussed in Section 1.4.2, the frequency
content of the input motion affects the selection of mesh size for accurate wave prop-
agation. Eq. (1.29) should be used to choose the appropriate maximum zone size for
an accurate representation. If the highest frequencies associated with the input motion
necessitate an extremely fine mesh (and a correspondingly small timestep), it may be
possible to remove the high frequency components to permit a coarser mesh. If most
of the power for the input is contained in the lower frequency components (say 80% to
90%) then the history can be filtered to remove the higher frequencies without signifi-
cantly affecting the results. The FISH function ‘FFT.FIS” (see Section 3 in the FISH
volume) can be used to evaluate the frequency content of the wave. The FISH function
“FILTER.FIS” which is also described in Section 3 in the FISH volume can be used to
perform the filtering.
For assessing frequency content, it is suggested that the input motion be evaluated in
the form of a particle velocity (not acceleration) history. (An acceleration record can be
converted into a velocity record using “INT.FIS,” described in Section 3 in the FISH
volume.) In a plane wave propagating through a continuous medium, it is easy to show
(e.g., see textbooks on wave propagation) that stress σ = Cρv (where C is the wave
speed, ρ is the density and v is the particle velocity). Therefore, when considering the
potential for yield or failure (which is determined by the level of induced stress), it is the
particle velocity that is most relevant when evaluating the dominant frequencies, because
the induced stress is directly proportional to velocity, not acceleration.
The input record should also be checked for baseline drift (i.e., continuing residual
displacement after the motion has finished, see Section 1.4.1.2). The FISH function
“INT.FIS,” is used to integrate the velocity record to produce the displacement wave
form related to the input acceleration. If needed, a baseline correction can be performed
by adding a low frequency sine wave to the velocity record; the sine wave parameters
are adjusted so that the final displacement is zero. An example is given in Section 1.6.1;
see Example 1.25.
3. Estimate material damping parameters to represent inelastic cyclic behavior – When
using a simple elastic-perfectly plastic material model (such as the Mohr-Coulomb model)
in a seismic analysis, it is necessary to incorporate additional material damping to account
for cyclic energy dissipation during the elastic part of the response, as discussed in
Section 1.4.3.11. A difficult aspect is how to determine the appropriate material damping
input. Generally, two different schemes are used: either Rayleigh damping (as described
in Section 1.4.3.1), or hysteretic damping (as described in Section 1.4.3.4).
The equivalent-linear method, as applied, for example, in program SHAKE, is one way
to estimate material damping input for the Rayleigh damping scheme and the hysteretic
damping scheme. An equivalent-linear analysis is performed on a soil column, represen-
tative of the site conditions, using the shear wave speeds and densities for the different
soil layers in the column, the modulus reduction and damping ratio curves selected as rep-
resentative of the materials, and the target earthquake design motion for the site. Elastic
strain-compatible values are then determined for the shear-modulus reduction factors and
damping ratios. Average modulus-reduction factors and damping ratios can be estimated
for each soil layer; these are the input parameters for the model with Rayleigh damping.
The anticipated range of cyclic shear-strain magnitudes for the given site conditions is
needed to specify a best-fit range for the modulus reduction and damping ratio curves used
with hysteretic damping (as discussed in Section 1.4.3.5). This range can be estimated
from the range of equivalent uniform cyclic strains provided from a SHAKE simulation.
Note that these estimates are derived from the equivalent-linear analysis that assumes a
low level of nonlinearity. Some adjustment in parameters may be required, especially if
the actual model exhibits strong nonlinearity. This is discussed further in Step 7.
4. Create appropriate model grid for accurate wave propagation – The characteristics of
the input motion are used to help select the appropriate mesh size and adjust the input wave
record in order to provide an accurate solution in the seismic analysis. (See Section 1.4.2
for further information on the relation between wave propagation characteristics and
mesh size.)
In most seismic simulations, especially if excessive motion and strains are anticipated,
a large-strain simulation should be performed in order to provide a more accurate defor-
mation solution. When significant deformation and distortion of the grid is anticipated,
it is important to minimize the number of triangular-shaped zones in the mesh and, in
particular, those along slope faces. Triangular zones along slope faces are prone to be-
come badly distorted during large-strain calculations, because triangular zones do not
contain overlaid sets of subzones. Quadrilateral-shaped zones contain two overlaid sets
of subzones, which provide a more accurate calculation for materials undergoing plastic
yield. (See Sections 1.3.2 and 1.3.3.2 in Theory and Background.)*
5. Calculate static equilibrium state for site – It is important to model the construction
sequence of the earth structure as closely as possible in order to provide a reasonable
representation of the initial, static shear stresses in the structure. This is important,
particularly in a liquefaction analysis, because the initial static shear stresses can affect
the triggering of liquefaction.
Simple analyses typically assume that the initial shear modulus (Gmax ) is uniform
throughout each material unit. It may be more appropriate to vary Gmax for soils as
a function of the in-situ effective stress (e.g., see Kramer 1996). For example, it may be
considered that the maximum shear modulus varies as a function of effective stress as
defined by the Seed et al. (1986) expression,
σm 0.5
Gmax = 21.7 × Pa × K2,max (1.122)
Pa
where Pa is the atmospheric pressure, K2,max is a constant determined from the rela-
tive density and σm is the effective mean stress. This initial modulus variation can be
implemented via FISH during the static loading stage.
It is important, if hysteretic damping is applied in the seismic simulation, to check the
initial shear stress at the static equilibrium state. The hysteretic damping formulation
is assumed to initiate hysteresis from an initial value of zero shear strain. If the initial
shear stresses are high, then the shear stress and shear strain state for hysteretic damping
may not be compatible. If shear stresses are high at the static equilibrium state of the
model, then in order to ensure that shear stresses and strains are consistent during the
dynamic phase, hysteretic damping should be invoked before the model is brought to the
initial equilibrium state. See Section 1.4.3.6 for a recommended procedure to incorporate
hysteretic damping during the static solution stage.
6. Apply deconvoluted dynamic loading derived from target seismic record for site –
The input seismic record to the FLAC model should produce a calculated motion that can
be matched to the target design earthquake motion. If the target motion is at a different
location than the model input motion, then a deconvolution analysis should be performed
to apply an input motion that will produce a motion that is comparable to the target motion
at the target’s location.
* If a badly distorted zone causes a calculation to stop prematurely in a mesh containing a few
triangular zones, it may be possible to prevent this by increasing the strength of the individual zone.
This should not significantly affect the model results, provided that the strengths of only a few zones
are changed. Alternatively, the automatic rezoning logic in FLAC can be used to correct the mesh
automatically during cycling when zones become badly distorted. See Section 6 in Theory and
Background.
It is important to note that a different form of the deconvoluted motion is applied for
a rigid-base model than for a compliant-base model, in order to reproduce the target
motion. (See Section 1.4.1.6 for further explanation.) For the rigid-base boundary, the
input motion to FLAC is the within motion calculated by SHAKE-91 at the specified
depth. For the compliant-base boundary, the input motion is the upward-propagating
motion, which is half of the outcrop motion output by SHAKE at the specified depth.
Figure 1.83 illustrates the different base input motions:
The dynamic boundary and loading conditions should be specified in a specific order for
a seismic analysis. This is because the free-field boundaries that are used to represent
the lateral extent of the far-field are assigned properties and initial conditions that are
transferred from the main-grid zones adjacent to the free-field boundaries. Any changes
to these zones or gridpoints after the free field is assigned are not seen by the free field.
σs = −2(ρ Cs ) vs (1.123)
Note that the factor of two in Eq. (1.123) accounts for the input energy dividing
into downward- and upward-propagating waves.
2. Monitor the x-acceleration at the foundation surface during the dynamic run to
compare this acceleration to the target acceleration. If material failure occurs
within the model, this will affect the attenuation of the wave. Also, if the
model is shallow, the free surface can cause an increase in the velocity and
acceleration that could extend to the base.* Some adjustment to the input stress
wave may be required in order to produce an acceleration that is comparable
to the target acceleration.
7. Perform undamped, elastic-material simulation – Before running a dynamic model
with actual material strength and damping properties, an elastic simulation should be
made without damping, to estimate the maximum levels of cyclic strain and natural
frequency ranges of the model system. If the cyclic strains are large enough to cause
excessive reductions in shear modulus, then the use of additional damping is questionable.
In such a case, the damping will be performing outside of its intended range of application.
The model properties and input amplitude should be checked if excessive strains are
calculated throughout the model.
Velocity histories should be monitored throughout the model to provide an estimate of
the dominant natural frequencies of the model system. Also, shear-strain histories should
be recorded to estimate maximum cyclic shear strain levels when no material damping
is provided.
8. Perform simulations with damping and actual strength properties – The results of
the undamped run can be used to help select appropriate damping characteristics for the
materials in the model. Additional damping may be prescribed for the model in order to
damp the natural oscillation modes identified from the undamped simulation.
Acceleration histories should be checked first in the initial run with damping and actual
strength properties, and compared to the target motion for the site. Some adjustment to
the model input motion may be required to improve the comparison.
It is important to monitor several different variables during the seismic shaking phase.
For example, shear stress/shear strain plots can illustrate the level of hysteretic damping
that occurs throughout the model. Excess pore-pressure histories can help quantify
the potential for liquefaction, and contours of cyclic pore-pressure ratio can delineate
regions of liquefaction in a model. Example variables are shown in the practical exercise
in Section 1.6.1.
* This is a result of the velocity-doubling effect of the free surface. Note that the effect of a free
surface on a simple sinusoidal velocity wave extends beneath the surface to, approximately, a depth
of one-fourth of the wavelength of the wave transmitted through the medium. (The extent of velocity
doubling can be shown simply by applying a wave to the base of a column of zones with a free
surface, and monitoring the maximum amplitude experienced by each zone as the wave travels
through the column.)
An example application of a nonlinear seismic analysis is presented in this section. The example
illustrates several of the topics discussed in this volume. The data files for this example are contained
in the “ITASCA\FLAC600\Dynamic” directory.
1.300
LEGEND
User-defined Groups
’foundation:soil 2’
0.700
’foundation:soil 1’
’embankment:soil 1’
’embankment:soil 2’
0.500
0.300
0.100
-0.100
-0.300
The earth dam is subjected to seismic loading representative of the 1987 Loma Prieta earthquake
in California. The earthquake target motion for this model is taken from that recorded at the left
abutment of the Lexington Dam during the Loma Prieta earthquake and, for this analysis, the input is
magnified somewhat and assumed to correspond to the acceleration at the surface of the foundation
soils at elevation 550 ft. The target record is provided in the file named “ACC TARGET.HIS.” The
estimated peak acceleration is approximately 12 ft/sec2 (or 0.37 g), and the duration is approximately
40 sec. The record is shown in Figure 1.85:
JOB TITLE : .
5-Jun-08 13:32
step 0 0.800
Target Acceleration
(ft/sec2 vs sec) 0.400
0.000
-0.400
-0.800
5 10 15 20 25 30 35
The foundation and embankment soils are modeled as elastic-perfectly plastic Mohr-Coulomb
materials. Drained properties are required because this is an effective-stress analysis. The properties
for the different soil types are listed in Table 1.3:
The dynamic characteristics of all of the soils in this model are assumed to be governed by the
modulus reduction factor (G/Gmax ) and damping ratio (λ) curves, as shown in Figures 1.86 and
1.87, and denoted by the “SHAKE91” legend. These curves are considered to be representative
of clayey soils with an average unit weight of 125 pcf and an average shear modulus of 6270
ksf; the data is derived from the input file supplied with SHAKE-91 (for more information see
http://nisee.berkeley.edu/software/).
The dynamic characteristics of the soils are simulated in the FLAC model in two different ways,
for comparison. Simulations are made with either Rayleigh damping (Section 1.4.3.1) or hysteretic
damping (Section 1.4.3.4) included with the Mohr-Coulomb model and with the Finn/Byrne model,
to evaluate and compare their representation of the inelastic cyclic response of the soils during
dynamic loading.
The equivalent linear program SHAKE-91 is run to estimate material damping parameters to rep-
resent the inelastic cyclic behavior of the soils in the FLAC model, based upon the curves in
Figures 1.86 and 1.87. A SHAKE-91 free-field column model is created for the foundation soils.
The SHAKE-91 analysis is performed using the shear wave speeds, densities, the modulus-reduction
and damping-ratio curves for the two foundation soil layers, and the target earthquake motion spec-
ified for the site. Strain-compatible values for the shear-modulus reduction factors and damping
ratios throughout the soil column are determined from the SHAKE analysis.
Average modulus-reduction factors and damping ratios can then be estimated for the foundation
soils based upon the values calculated by SHAKE-91. The selected damping ratio and modulus-
reduction parameters correspond to the equivalent uniform strain (which is taken as 50% of the
maximum strain) for each layer. In this exercise, one value is selected as representative for all
materials. The maximum equivalent uniform strain for the soils is calculated to be 0.08%, the
average damping ratio is 0.063 and the average modulus reduction factor is 0.8. The damping ratio
and modulus reduction factor values are the input parameters when Rayleigh damping is applied
in the FLAC model. For a more comprehensive analysis, different values may be selected for the
different soil units.
The maximum equivalent uniform shear strain calculated from the SHAKE-91 simulation is used
to specify the range over which the modulus reduction and damping ratio curves developed for
hysteretic damping should best-fit the representative curves for the soils. The default hysteretic
damping function (Eq. (1.48)) is used to best-fit the modulus-reduction factor and damping-ratio
curves. The parameter values L1 = −3.156 and L2 = 1.904 for the default model provide a reasonable
fit to both curves over the range of 0.08% strain, as shown in Figures 1.86 and 1.87. Note that the
parameters for the default model may need to be adjusted after the maximum shear strains are
calculated from an undamped elastic FLAC model of the entire site (see Section 1.6.1.9).
Figure 1.86 Modulus reduction curve for clayey soils (from SHAKE-91 data)
FLAC default hysteretic damping with L1 = −3.156 and
L2 = 1.904
Figure 1.87 Damping ratio curve for clayey soils (from SHAKE-91 data)
FLAC default hysteretic damping with L1 = −3.156 and
L2 = 1.904
The dynamic simulations in this example are fully coupled effective-stress calculations, which
require that the water bulk modulus be specified explicitly. For the properties listed in Table 1.3,
and assuming the actual modulus Kw = 4.18×106 psf for the site, the value of Rk (see Eq. (1.121))
for the foundation soils is approximately 0.8, and for the embankment soils it is approximately 1.5.
In this example, a uniform value of Kw = 4.18×106 psf is selected as representative of the actual
condition.
The liquefaction condition is estimated for embankment soils in terms of Standard Penetration Test
results. A normalized Standard Penetration Test value, (N1 )60 , of 10 is selected as representative
for these soils. This value is used to determine the parameters C1 and C2 in the liquefaction model
in FLAC (selected by setting the property ff switch = 1 for the Finn model – Byrne formulation).
For a normalized SPT blow count of 10, the Finn/Byrne model parameters are C1 = 0.4904 and
C2 = 0.8156. See Section 1.4.4.2 for a description of the formulation, and see Byrne (1991) for a
discussion on the derivation of these parameters.
1.6.1.4 Perform Deconvolution Analysis and Estimate Seismic Motion Characteristics
The target motion provided for this example, Figure 1.85, is assumed to correspond to the motion
at the ground surface of the foundation soils near the site.* It is necessary to modify this motion to
apply the appropriate seismic input at the base of the model (in this case at elevation 400 ft). The
appropriate input motion at depth is computed by performing a deconvolution analysis using the
equivalent-linear program SHAKE. This approach is reasonable, provided the model exhibits a low
level of nonlinearity. A check on the approach is made in Section 1.6.1.9.
SHAKE-91 is used in this example to estimate the appropriate motion at depth corresponding to
the target (surface) motion. This deconvoluted motion should then produce the target motion at
the surface. In this example, the upward-propagation motion calculated from SHAKE-91 is used
with a compliant-base boundary. (See Figure 1.83.) Note that SHAKE-91 accelerations are in g’s
versus seconds, and need to be converted into ft/sec2 versus seconds when applied in FLAC. The
input record (i.e., the upward-propagating motion from the deconvolution analysis and converted
to ft/sec2 ) is in the file named “ACC DECONV.HIS,” and is shown in Figure 1.88.
A Fast Fourier Transform (FFT) analysis of the input acceleration record (using “FFT.FIS” in
Section 3 in the FISH volume) results in a power spectrum as shown in Figure 1.89. This figure
indicates that the dominant frequency is approximately 1 Hz, the highest frequency component is
less than 10 Hz, and the majority of the frequencies are less than 5 Hz. The dominant frequencies of
the input velocity are also checked by first converting the acceleration record into a velocity record
(using the FISH function “INT.FIS,” described in Section 3 in the FISH volume, to integrate
the acceleration record), and then performing an FFT analysis to determine the power spectrum.
Figure 1.90 shows the resulting power spectrum for the velocity record. The dominant frequency
is seen to be less than 1 Hz.
* This target motion is used for illustrating the principles of deconvolution analysis and may not be the
typical case encountered in many practical situations. It is more common that the outcrop motion,
recorded at the top of bedrock underlying the soil layers, is the target motion. Deconvolution in
this case is described in Mejia and Dawson (2006), and is summarized in Section 1.4.1.6.
The input record is also checked for baseline drift. The FISH function “INT.FIS” is used to integrate
the velocity record again to produce the displacement wave form related to the input acceleration.
The resulting residual displacement is found to be approximately 0.3 ft. A baseline correction is
performed by adding a low frequency sine wave to the velocity record; the sine wave parameters
are adjusted so that the final displacement is zero. (See “BASELINE.FIS” in Example 1.25.) The
uncorrected and corrected resultant displacement histories are shown in Figure 1.91.
JOB TITLE : .
LEGEND
5-Jun-08 15:00
step 0 4.000
0.000
-2.000
-4.000
5 10 15 20 25 30 35 40
JOB TITLE : .
2.500
2.000
1.500
1.000
0.500
5 10 15 20 25
JOB TITLE : .
2.500
2.000
1.500
1.000
0.500
2 4 6 8 10 12
JOB TITLE : .
26-Nov-07 14:48
3.000
step 0
displacement histories
corrected disp. 2.000
uncorrected disp.
1.000
0.000
-1.000
-2.000
5 10 15 20 25 30 35 40
1.6.1.5 Adjust Input Motion and Mesh Size for Accurate Wave Propagation
The mesh size for the FLAC model is selected to ensure accurate wave transmission (see Sec-
tion 1.4.2). Based upon the elastic properties listed in Table 1.3, embankment soil 2 has the lowest
shear wave speed (840 ft/sec). If the largest zone size in the FLAC model is selected to be 10 ft
in order to provide reasonable runtimes for this example, then the maximum frequency that can be
modeled accurately is
Cs
f = ≈ 8.4 Hz (1.124)
10 l
Before applying the acceleration input record, it is filtered to remove frequencies above 5 Hz (by
using the FISH function “FILTER.FIS” described in Section 3 in the FISH volume). This filtering
value is selected to account for the reduction in shear wave speed that may occur in some of the
materials during the dynamic loading stage, as indicated in Figure 1.86. The acceleration history
filtered at 5 Hz is shown in Figure 1.92, the power spectrum for the filtered acceleration wave is
shown in Figure 1.93, and the power spectrum for the corresponding velocity wave is shown in
Figure 1.94. Note that the difference between the frequency content of the unfiltered and filtered
acceleration and velocity waves is minor (compare Figures 1.89 and 1.90 to Figures 1.93 and 1.94).
The data file “INPUT.DAT,” listed in Example 1.26, includes the different steps performed in
filtering the input acceleration record, integrating this record to produce velocity and displacement
histories, and correcting for baseline drift. The resultant, corrected velocity record, stored in table
104 in this data file, is the input motion for the embankment dam analysis.
JOB TITLE : .
LEGEND
5-Jun-08 15:35
step 0 4.000
Target Acc.
Table 101
2.000
0.000
-2.000
-4.000
5 10 15 20 25 30 35 40
JOB TITLE : .
2.500
2.000
1.500
1.000
0.500
5 10 15 20 25
JOB TITLE : .
2.500
2.000
1.500
1.000
0.500
2 4 6 8 10 12
The input motion can be generated for this example by using the GIIC. The procedure to create the
filtered and baseline-corrected input motion and save it as table 104 is as follows. The acceleration
history (“ACC DECONV.HIS”) is read into FLAC via the Read button in the Utility/History tool. The
Execute
button should be pressed to execute the command. The history is then converted into a table
by pressing the History -> Table button in the Utility/History tool. The dialog shown in Figure 1.95
appears, and the acceleration history (previously assigned ID number 100) is converted into a table
(designated by ID number 100).
The power spectrum is calculated in a two-step procedure. First, the FISH function named
“FFT TABLES.FIS” (accessed from the Tables menu item in the Utility/FishLib tool) is used to
assign tables for inputting the acceleration history and storing the calculated power spectrum. And
then “FFT.FIS” is executed from the same location to create the power spectrum.
The acceleration record in table 100 is converted into a velocity record, using FISH function
“INT.FIS” accessed from the Tables menu item in the Utility/FishLib tool, and stored in table 201. A
power spectrum is calculated for the velocity record using the same procedure as for the acceleration
record.
The FISH function “FILTER.FIS” is accessed from the Tables menu item in the Utility/FishLib tool
to filter the acceleration at 5 Hz. The filtered table is given the ID number 101, and the cutoff
frequency is set to 5 Hz, as shown in the dialog in Figure 1.96. OK is pressed to execute this FISH
function and create the filtered record.
In order to perform the baseline correction, the filtered acceleration in table 101 is integrated (with
“INT.FIS”) to produce a velocity record and stored in table 102. Then, a low frequency sine wave
is added to this velocity record to produce a final displacement of zero. The sine wave is given in
“BASELINE.FIS” in Example 1.25. This FISH function is accessed from the FISH editor resource
pane. The input dialog for this FISH function is shown in Figure 1.97. The corrected velocity wave
is written to table 104. Table 104 is written to a file named “TABLE104.DAT” by selecting SAVE in
the Utility/Table tool.
The FLAC model options selected for this analysis are shown activated in the Model options dialog
displayed in Figure 1.98. The dynamic analysis and groundwater flow options are selected. Ad-
vanced constitutive models are also included in order to access the Finn model, which will be used
for the liquefaction calculation phase. The Imperial system of units is specified for this analysis.
Figure 1.98 Model options selected for the embankment dam example
The embankment dam model can now be created. (The commands that will be generated for this
model are listed in “EARTHDAM.DAT” in Example 1.30.)
The dynamic calculation phase is performed using the large-strain mode in FLAC. When significant
deformation and distortion of the grid is anticipated, as in this example, it is important to minimize
the number of triangular-shaped zones in the mesh and, in particular, those along slope faces, as
discussed in Section 1.5.3, Step 4. A special FISH tool is provided in the Utility/FishLib library to
assist with the creation of a mesh for this model. The gentabletop tool, shown in the Fish
Library dialog in Figure 1.99, is used to transform a grid to fit an irregular upper surface defined by
a table. The grid beneath the table is adjusted to provide a uniform spacing of quadrilateral zones
below the table surface.
The grid zoning is defined first using the Generate/Simple tool. The grid dimensions selected for this
model are shown in the Range dialog of the Simple tool, in Figure 1.100. The number of zones
selected for this model (180 × 28) corresponds to a zone size of 10 ft.
After this grid is created, the Utility/Table tool is used to define the slope of the embankment dam.
Figure 1.101 shows the pairs of x- and y-values that are entered in the Edit Table points dialog to
define the surface.
Figure 1.101 Edit Table points dialog showing x- and y-values defining em-
bankment dam surface
After this table is defined, the gentabletop tool is executed from the Utility/FishLib library to
create the grid. The resulting mesh is shown in Figure 1.102. Note that the grid is totally composed
of quadrilateral-shaped zones.
1.300
LEGEND
Grid plot
0.700
0 5E 2
0.500
0.300
0.100
-0.100
-0.300
The foundation soil layers and the embankment core and shell regions are delineated by lines
generated using the Alter/Shape tool. Then the different materials and properties, listed in Table 1.3,
are specified, corresponding to Mohr-Coulomb materials, and assigned using the Material/Assign and
Material/GwProp
tools. The resulting model is shown in Figure 1.103. Note that some triangular
zones are created within the mesh when the different soil regions are defined (see Figure 1.104).
Triangular zones are also created at the slope toe and crest. It is difficult to eliminate triangular
zones completely in this model. However, there are only a small number of these zones along the
slope face, and the strengths of these zones can be readily adjusted if there is a distortion problem.
The model state, after the geometry shaping is complete, is saved in the GIIC Project Tree with
the name “EDAM1.SAV”; and after the materials have been added, with the name “EDAM2.SAV.”
The model is now ready to begin the analysis stage.
1.300
LEGEND
User-defined Groups
’foundation:soil 2’
0.700
’foundation:soil 1’
’embankment:soil 1’
’embankment:soil 2’
0.500
Grid plot
0 5E 2
0.300
0.100
-0.100
-0.300
Figure 1.103 Embankment dam model with foundation and embankment soils
assigned
LEGEND
7.000
26-Nov-07 15:29
step 0
5.509E+02 <x< 1.000E+03 6.500
Grid plot
0 1E 2 5.000
4.500
4.000
3.500
The water density of 1.94 slugs/ft3 and gravitational magnitude of 32.2 ft/sec2 are assigned, and
fluid-flow and dynamic-analysis modes are turned off in the global Settings tools.
The most efficient way to achieve an equilibrium stress state in a saturated, horizontally layered
soil is to use the special FISH tool ininv, provided in the Utility/FishLib library. This function
calculates the pore pressures and stresses automatically for a model containing a phreatic surface.
The function requires the phreatic surface height (wth = 550 in this example) and the ratios of
horizontal to vertical effective stresses (assumed to be k0x = k0z = 0.5 in this example). The pore
pressure, total stress and effective stress distributions are then calculated automatically, accounting
for the different soil unit weights, and the position of the water table. The equilibrium state is checked
(using the SOLVE elastic option in the Run/Solve tool). Figure 1.105 shows the initial pore-pressure
distribution in the foundation soils. This state is saved in the Project Tree as “EDAM3.SAV.”
The embankment materials can be added to the model in stages, to simulate the construction
process, by using the Material/Cut&Fill tool. In this example, both embankment soils 1 and 2 are added
simultaneously, and pore pressures are assumed not to change. The displacements resulting from
adding the embankment in one step are shown in Figure 1.106. This “construction step” is done to
simplify the example; a more rigorous analysis should follow the construction sequence as closely
as possible, in order to produce a more realistic displacement pattern and initial stress state. The
saved state at this stage in named “EDAM4.SAV.”
1.400
FLAC (Version 6.00)
LEGEND 1.200
5-Jun-08 10:51
step 1795 1.000
-0.200
0 5E 2
-0.400
1.300
LEGEND
Y-displacement contours
-3.00E-01
0.700
-2.50E-01
-2.00E-01
-1.50E-01
0.500
-1.00E-01
-5.00E-02
0.00E+00
0.300
0.100
-0.100
-0.300
The model is run in small-strain mode up to this stage and, consequently, the gridpoint positions
are not changed. This is done so that the embankment crest elevation (680 ft) does not change. If
significant deformation occurs during embankment construction, making it necessary to perform
this stage in large-strain mode, then the initial embankment crest elevation for the embankment
zones (prior to construction) would need to be raised in order to obtain a specified elevation after
construction.
State of Stress with the Reservoir Level Raised
The earthquake motion is considered to occur when the reservoir level is at full pool (i.e., at its
full height at elevation 670 ft). For Stage 2 of the analysis, the pore pressure distribution through
the embankment and foundation soils is calculated for the reservoir raised to this height. The
In Situ/Apply
tool is used to set the pore pressure distribution on the upstream side of the embankment,
corresponding to the reservoir elevation at 670 ft. The mouse is dragged in this tool along the
upstream boundary starting from the 670 elevation (at gridpoint i = 70, j = 29) and ending at the
400 elevation at the bottom-left corner of the model (at gridpoint i = 1, j = 1). The distribution
parameters, shown in the Apply value dialog in Figure 1.107, produce a pore-pressure distribution
along this boundary that ranges from zero at elevation 670 ft to 16,866.36 psf at elevation 400 ft.
This calculation is first run in flow-only mode. The groundwater-flow calculation is turned on,
and a water bulk modulus of 4.1 × 106 psf is assigned, in the Settings/GW tool. The water modulus
corresponds to water with entrained air. The fast unsaturated-flow calculation funsat and water bulk
scaling with permeability and porosity fastwb are set in order to speed the calculation to steady-state
flow. The mechanical calculation mode is turned off in the Settings/Mech tool. In the In Situ/Fix tool,
the pore pressures are fixed at gridpoints along the downstream slope to allow flow across this
surface. The porosity and permeability values are also specified for the embankment materials, in
the Material/GwProp tool.
Figure 1.108 plots pore-pressure histories at different locations in the model, indicating that con-
stant values are reached for the equilibrium ratio limit. Figure 1.109 displays the pore-pressure
distribution through the embankment and foundation at steady state. The saved state at steady-state
flow is named “EDAM5.SAV.”
The static equilibrium state is now calculated for the new pore-pressure distribution. A pressure
distribution is applied along the upstream slope to represent the weight of the reservoir water. This
time a mechanical pressure is assigned in the In Situ/Apply tool. The pressure ranges from zero at
elevation 670 ft (at gridpoint i = 70, j = 29) to 8120.8 psf at elevation 550 ft at the toe of the
slope (at gridpoint i = 1, j = 29); the dialog is displayed in Figure 1.110. The groundwater-flow
calculation is turned off, and the water bulk modulus is set to zero (in the Settings/GW tool). The
mechanical calculation is turned on (in the Settings/Mech tool). The model is now solved for this
applied condition, and the resulting total vertical-stress contour plot for the model at this stage is
shown in Figure 1.111. We also note that the shear stresses at this stage are quite low (less than
10% of the total vertical stresses throughout most of the model) and should not adversely affect the
application of hysteretic damping during the dynamic loading phase. (See Step 5 in Section 1.5.3.)
The saved state at this stage is named “EDAM6.SAV.”
JOB TITLE : .
5-Jun-08 10:55
6.000
step 13523
Flow Time 1.5168E+09
2.000
1.000
2 4 6 8 10 12 14
08
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
1.300
LEGEND
0 5E 2
-0.100
-0.300
Figure 1.109 Pore-pressure distribution at steady state flow for reservoir raised
to 670 ft
1.300
LEGEND
YY-stress contours
0.700
-3.50E+04
-3.00E+04
-2.50E+04
0.500
-2.00E+04
-1.50E+04
-1.00E+04
-5.00E+03 0.300
0.00E+00
Extrap. by averaging
Boundary plot
-0.100
0 5E 2
-0.300
Figure 1.111 Total vertical-stress distribution at steady state flow for reservoir
raised to 670 ft
For the dynamic loading stage, pore pressures can change in the materials due to dynamic volume
changes induced by the seismic excitation. However, pore-pressure generation due to particle
rearrangement does not occur. In order for pore pressures to change as a result of volume change,
the actual value of water bulk modulus must be prescribed. The value of 4.18 × 106 psf is re-
specified for the water bulk modulus. Note that the groundwater-flow mode is not active, because
it is assumed that the dynamic excitation occurs over a much smaller time frame than required for
pore pressures to dissipate. (FLAC can carry out the groundwater flow calculation in parallel with
the dynamic calculation if dissipation is considered important. See Section 1.5.2.3.)
The following conditions are set in this first dynamic simulation, following the steps listed in Step
6 of Section 1.5.3. The dynamic calculation mode is turned on, using the Settings/Dnya tool, and the
large-strain mode is selected in the Settings/Mech tool.
The filtered and baseline-corrected input velocity created previously and stored in table 104 is called
into FLAC using the Run/Call tool and selecting the “TABLE104.DAT” file.
The displacements and velocities in the model are initialized by pressing the Displmt & Velocity button
in the In Situ/Initial tool. In this way, only seismic induced motions and deformations are shown
in the model results. Damping is not prescribed for this initial dynamic simulation. Acceleration
histories are recorded at several gridpoints throughout the model via the Utility/History tool. Also,
special FISH functions are implemented to monitor the shear strain and excess pore pressure at
selected locations, and relative displacements along the upstream slope near the crest.* Examples
of these functions are listed in Example 1.27. The velocity and shear strain histories are used to
evaluate the dominant natural frequencies and maximum cyclic shear strains in the model, when no
additional damping is prescribed.
The model state is saved at this point, with these dynamic conditions set, and named “EDAM7.SAV.”
This will provide a convenient starting state for the dynamic analysis when damping is added.
The dynamic boundary conditions are now applied in the In Situ/Apply tool. First, the free-field
boundary is set for the side boundaries by selecting the Free-Field button.
Next, the dynamic input is assigned to the bottom boundary. In this model, a compliant boundary
condition is assumed for the base (i.e., the foundation materials are assumed to extend to a significant
depth beneath the dam). Therefore, it is necessary to apply a quiet (viscous) boundary along the
bottom of the model to minimize the effect of reflected waves at the bottom.
Quiet boundary conditions are assigned in both the x- and y-directions by first selecting the xquiet
button and dragging the mouse along the bottom boundary, and then selecting the yquiet button and
repeating the procedure.
* Note that relative displacement is referenced to the base of the model. See “RELDISPX.FIS” in
Example 1.27.
The dynamic wave is applied as a shear-stress boundary condition along the base in the following
manner. The Stress/sxy boundary-condition type is selected in the In Situ/Apply tool, and the mouse is
dragged from the bottom-left corner of the model (gridpoint i = 1, j = 1) to the bottom-right corner
(i = 180, j = 1). The Assign button is pressed, which opens the Apply value dialog. The velocity
record, in table 104, is considered a multiplier, vs , for the applied value. The velocity record is
applied by checking the Table radio button, and selecting table number 104 as the multiplier.
The applied value for sxy in the Apply value dialog is set to −2ρ Cs (from Eq. (1.123)), in which ρ
and Cs correspond to the properties for foundation soil 2. The input selections for the Apply value
dialog are shown in Figure 1.112.
The model state is saved again at this point and named “EDAM8.SAV.”
Figure 1.112 Apply shear stress boundary condition in Apply value dialog
Before running a dynamic model with actual material strength and damping properties, an elastic
simulation is made without damping, to estimate the maximum levels of cyclic strain and natural
frequency ranges of the model system. Acceleration and shear stress/strain histories are recorded
at several locations in the FLAC model for this elastic simulation. Figure 1.113 plots shear stress
versus shear strain in zone (77,20) located in embankment soil 2. Maximum shear strains of
approximately 0.12% were observed in this region of the model. These strains are not considered to
be sufficient to cause excessive reductions in shear modulus. The shear-modulus reduction factor
is roughly 0.6 for this strain level, as shown in Figure 1.86.
The frequency range for the natural response of the system is observed to be relatively uniform
throughout the model. Figure 1.114 displays a typical power spectrum for this simulation; the
predominant frequency was found to be approximately 1.0 Hz.
JOB TITLE : .
-1.000
-2.000
-3.000
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.113 Shear stress versus shear strain in embankment soil 2 at zone
(77,20) – elastic material and undamped
JOB TITLE : .
4.000
3.000
2.000
1.000
2 4 6 8 10 12
In this simulation, the dynamic input produces x-accelerations at the surface of the foundation
material that are very similar to the target motion (“ACC TARGET.HIS”), as shown in Figure 1.115.
The peak acceleration is approximately 11 ft/sec2 .
The undamped, elastic calculation is saved as “EDAM9.SAV.” The Fast Fourier analysis results are
saved in “EDAM10.SAV.”
JOB TITLE : .
5-Jun-08 17:06
1.200
step 93839
Flow Time 1.5168E+09
Dynamic Time 4.0000E+01
0.800
HISTORY PLOT
Y-axis :
104 X acceleration( 14, 29) 0.400
106 target acceleration at el. 550
X-axis :
5 Dynamic time 0.000
Input Time
-0.400
-0.800
5 10 15 20 25 30 35
The damping is applied at the model state “EDAM7.SAV” by double-clicking on this state name
in the Project Tree. Hysteretic damping is assigned in the In Situ/Initial tool. The dialog shown in
Figure 1.116 is opened by selecting the Zones type, checking the Hysteretic Damping menu item, and
then Assign , to assign the same values for all zones in the model.
Hysteretic damping does not completely damp high-frequency components, so a small amount of
stiffness-proportional Rayleigh damping is also applied. A value of 0.2% at the dominant frequency
(1.0 Hz) is assigned in the Dynamic damping parameters dialog shown in Figure 1.117. Note that
Rayleigh damping is applied by selecting the GPs type, and then Dynamic Damping in the In Situ/Initial
tool.
After Execute is pressed in the In Situ/Initial tool, a new branch is created in the project tree to
perform a calculation including the additional hysteretic damping (and the small amount of stiffness-
proportional Rayleigh damping). The dynamic boundary conditions must be applied again from the
In Situ/Apply
tool. The free-field is applied on the side boundaries, and sxy-stress history and quiet
boundaries are applied at the base, in the same way as for the undamped simulation. The model
state is saved at this stage as “EDAM11.SAV.”
A new simulation is now made for a dynamic time of 40 seconds. Note that the dynamic timestep
used for this calculation is approximately 3.97 × 10−4 seconds.
The amplitude of the x-acceleration at the foundation surface is somewhat smaller for this simulation
than that for the target motion; in this simulation, the peak acceleration is approximately 9 ft/sec2 .
Compare Figure 1.118 to Figure 1.85. Some adjustment to the input motion is required to provide
a better comparison. For this exercise, the input shear stress is increased by a factor of 1.33, and the
hysteretic damping simulation is repeated. The peak acceleration is now nearly 12 ft/sec2 and there
is a closer agreement with the target motion, as shown by comparing Figure 1.119 to Figure 1.85.
This factor is applied for the following simulations.
Movement of the embankment after 40 seconds is primarily concentrated along the upstream slope.
This is shown in the x-displacement contour plot (in Figure 1.120) and the shear-strain increment
contour plot (in Figure 1.121). The movement of gridpoint (62,29) along the upstream slope is
shown in Figure 1.122. The effect of material failure and hysteresis damping is evident in the cyclic
shear strain response; compare Figure 1.123 to Figure 1.113.
The pore pressure and effective vertical stress histories in Figure 1.124, recorded at (i = 49, j =
23) near the upstream face, illustrate the minor pore-pressure change in the embankment materials
during the seismic loading.
The model state is saved at this stage as “EDAM12.SAV.”
JOB TITLE : .
6-Jun-08 10:56
step 115897 0.800
Flow Time 1.5168E+09
Dynamic Time 4.0000E+01
-0.400
-0.800
5 10 15 20 25 30 35
JOB TITLE : .
6-Jun-08 10:55
step 115770 0.800
Flow Time 1.5168E+09
Dynamic Time 4.0000E+01
-0.400
-0.800
5 10 15 20 25 30 35
8.500
LEGEND
6-Jun-08 10:35
step 115770 7.500
X-displacement contours
-1.25E+00
5.500
-1.00E+00
-7.50E-01
-5.00E-01
-2.50E-01 4.500
0.00E+00
2.50E-01
5.00E-01
3.500
Contour interval= 2.50E-01
Boundary plot
0 2E 2 2.500
1.500
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
2.000 3.000 4.000 5.000 6.000 7.000 8.000 9.000
(*10^2)
LEGEND
6-Jun-08 10:36
step 115770
Flow Time 1.5168E+09 1.000
0 5E 2
JOB TITLE : .
LEGEND
HISTORY PLOT
Y-axis : -1.000
9 reldispx (FISH)
10 reldispy (FISH) -1.500
X-axis :
5 Dynamic time
-2.000
-2.500
-3.000
5 10 15 20 25 30 35
JOB TITLE : .
HISTORY PLOT
Y-axis : 1.000
7 Viscous SXY ( 77, 20)
X-axis :
8 str_77_20 (FISH) 0.000
-1.000
-2.000
-3.000
-40 -20 0 20 40 60
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.123 Shear stress versus shear strain in embankment soil 2 at zone
(77,20) – Mohr-Coulomb material and hysteretic damping
JOB TITLE : .
HISTORY PLOT
Y-axis : 0.400
12 Pore pressure ( 49, 23)
13 Effective SYY ( 49, 23) 0.200
X-axis :
5 Dynamic time
0.000
-0.200
-0.400
5 10 15 20 25 30 35 40
Figure 1.124 Pore-pressure and effective vertical stress near upstream slope
– Mohr-Coulomb material and hysteretic damping
Also note that, for this case, with mass- and stiffness-proportional Rayleigh damping of 6.3%
at the natural frequency of 1.0 Hz, the limiting timestep is approximately 3.6 × 10−5 seconds.
This timestep is roughly eleven times smaller than that for hysteretic damping. The model state
is saved at this stage as “EDAM13.SAV.” The Rayleigh damping run at 40 seconds is saved as
“EDAM14.SAV.”
If Rayleigh damping alone is used, the results are comparable to those with hysteretic damping.
Figure 1.126 plots the x-displacement contours at 40 seconds for Rayleigh damping. Figure 1.127
shows the shear-strain increment contours at this time. Both plots compare reasonably well with
those using hysteretic damping (compare to Figures 1.120 and 1.121). Figure 1.128 plots the
relative movement at gridpoint (62,29). The displacements are slightly less than those for hysteretic
damping (compare to Figure 1.122).
The effect of Rayleigh damping is also evident in the cyclic shear strain response; compare Fig-
ure 1.129 to Figures 1.113 and 1.123.
Pore pressure and effective vertical stress histories for the Rayleigh damping run are also similar to
those for the hysteretic damping run (compare Figure 1.130 to Figure 1.124).
As material yield is approached, neither Rayleigh damping nor hysteretic damping account for
energy dissipation of extensive yielding. Irreversible strain occurs external to both schemes, and
dissipation is represented by the Mohr-Coulomb model. The mass-proportional term of Rayleigh
damping may inhibit yielding because rigid-body motions that occur during failure are erroneously
resisted. Consequently, hysteretic damping may be expected to give larger permanent deformations
in this situation, but this condition is believed to be more realistic than one using Rayleigh damping.
This comparison demonstrates the substantial benefit of hysteretic damping. The results are com-
parable to those using Rayleigh damping for similar damping levels, and the runtime with hysteretic
damping is greatly reduced.
8.500
LEGEND
8-Jun-08 13:18
step 1138558 7.500
X-displacement contours
-7.50E-01
5.500
-5.00E-01
-2.50E-01
0.00E+00
2.50E-01 4.500
2.500
1.500
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
2.000 3.000 4.000 5.000 6.000 7.000 8.000 9.000
(*10^2)
LEGEND
8-Jun-08 13:18
step 1138558
Flow Time 1.5168E+09 1.000
-0.200
0 5E 2
JOB TITLE : .
LEGEND
HISTORY PLOT
Y-axis : -1.000
9 reldispx (FISH)
10 reldispy (FISH) -1.500
X-axis :
5 Dynamic time
-2.000
-2.500
-3.000
5 10 15 20 25 30 35
JOB TITLE : .
HISTORY PLOT
Y-axis : 1.000
7 Viscous SXY ( 77, 20)
X-axis :
8 str_77_20 (FISH) 0.000
-1.000
-2.000
-3.000
-04
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.129 Shear stress versus shear strain in embankment soil 2 at zone
(77,20) – Mohr-Coulomb material and Rayleigh damping
JOB TITLE : .
8-Jun-08 13:22
0.800
step 1138558
Flow Time 1.5168E+09
Dynamic Time 4.0000E+01
0.600
HISTORY PLOT
Y-axis :
12 Pore pressure ( 49, 23) 0.400
13 Effective SYY ( 49, 23)
X-axis :
5 Dynamic time 0.200
0.000
-0.200
5 10 15 20 25 30 35
Figure 1.130 Pore-pressure and effective vertical stress near upstream slope
– Mohr-Coulomb material and Rayleigh damping
The embankment soils are now changed to liquefiable materials. The Finn/Byrne liquefaction model
is prescribed for embankment soils 1 and 2, with parameters set to correspond to SPT measurements.
For a normalized SPT blow count of 10, the Byrne model parameters are C1 = 0.4904 and C2 = 0.8156
(see Section 1.6.1.3).
The liquefaction simulation starts at the saved state “EDAM6.SAV.” The embankment soils are
changed at this state by using the Material/Model tool. (Note that this tool is activated when the
Include Advanced Constitutive Models?
box is checked in the Model options dialog.) The Region range is
selected and the Dynamic
models box is checked in this tool. The Finn model is then assigned to each
region of the embankment soils. When the mouse is clicked within one of the embankment soil
regions, a dialog opens to prescribe the model properties. Figure 1.131 shows the dialog with the
properties selected for embankment soil 1. Note that the Byrne radio button is checked in order to
prescribe the appropriate parameters for the Byrne formulation. Also, the value for Latency is set to
1,000,000 at this stage. This is done to prevent the liquefaction calculation from being activated
initially. The model is first checked to make sure that it is still at an equilibrium state when switching
materials to the Byrne model, before commencing the dynamic simulation.
Figure 1.131 Model finn properties dialog with properties for embankment
soil 1
The model is now ready for the dynamic analysis. The water bulk modulus is assigned as
4.1 × 106 psf using the Settings/GW tool. The value for latency of the embankment soils is reduced to
50 in the Material/Property tool. The dynamic conditions are now set again in the same manner as in
Stage 3. The model state is “EDAM15.SAV” before damping and dynamic boundaries are applied;
it is “EDAM16.SAV” after they are applied.
The model is now run for a dynamic time of 40 seconds. The results in Figures 1.132 through 1.137
show the effect of pore-pressure generation in the embankment soils. There is now a substantial
movement along the upstream face, as shown by figures Figures 1.132 through 1.135. The relative
vertical settlement at gridpoint (62,29) is now approximately 14 in, and the relative shift upstream
is approximately 3.6 ft, as shown in Figure 1.135.
A significant increase in pore pressure (and decrease in effective stress) is calculated in the upstream
region, as indicated in Figure 1.136. The location of the pore pressure/effective stress measurement
is at gridpoint (49,23), which is at a depth of approximately 45 ft below the upstream slope face,
and 135 ft from the toe of the upstream slope.
The normalized excess pore-pressure ratio (or cyclic pore-pressure ratio), ue / σc ,* can be used to
identify the region of liquefaction in the model. Contours of the cyclic pore-pressure ratio greater
than 0.99 are plotted in Figure 1.137. These contours show the extent of the liquefied embankment
soils, primarily in the upstream region. (The excess pore-pressure ratio is calculated in FISH
function “GETEXCESSPP.FIS,” and stored in FISH extra array ex 5, see Example 1.29.)
The final state is “EDAM17.SAV.”
* where ue is the excess pore-pressure and σc is the initial effective confining stress. Note that a
liquefaction state is reached when ue / σc = 1.
The liquefaction run is repeated using only Rayleigh damping with the same damping and modulus-
reduction parameters as the simulation with Mohr-Coulomb material (damping ratio of 0.063,
center frequency of 1.0 Hz, and modulus reduction factor of 0.8). The same plots created for the
liquefaction run with hysteretic damping are re-created for the run with Rayleigh damping. See
Figures 1.138 through 1.143. As the figures show, the results are very similar. The final state in
this case is “EDAM19.SAV.”
Comments
This simple example assumes that the initial shear modulus, Gmax (before the seismic loading),
is uniform throughout each material unit. It may be more appropriate to vary Gmax for soils
as a function of the in-situ effective stress (e.g., see Kramer 1996). An initial variation can be
implemented during the static loading stage.
This example also assumes that the shear strength parameters of the liquefiable soils do not change.
It has been shown (e.g., Olson et al. 2000) that if the effective stress goes to zero, the shear strength
reduces to a strain-mobilized (liquefied) shear strength, which implies a residual cohesion. There
are several ways to incorporate a change of strength envelope in the FLAC model, such that residual
cohesion is developed as the material liquefies. For example, a FISH function can be used to adjust
the strength parameters as a function of change in the effective confining stress. A more rigorous
approach is to modify a bilinear strength model (such as the strain-softening bilinear model, MODEL
subi) to include the liquefaction behavior (e.g., the Byrne model). The existing MODEL finn in FLAC
incorporates the pore-pressure generation effect into the Mohr-Coulomb model. This can also be
done with other models, using either the FISH constitutive model facility (see Section 2.8 in the
FISH volume) or the C++ DLL model facility (see Section 3 in Theory and Background) to
create a user-defined model.
If there is a potential for flow slides to occur either at liquefaction or during post-liquefaction,
the automatic rezoning logic in FLAC can be used to simulate the development of the deformed
flow-slide state. See Section 6 in Theory and Background for further information.
Acknowledgment
This example is derived from data provided by Dr. Nason McCullough of CH2MHill. His assistance
and critical review of this exercise are gratefully acknowledged.
8.500
LEGEND
14-Jul-08 15:41
step 115795 7.500
X-displacement contours
-3.00E+00
5.500
-2.00E+00
-1.00E+00
0.00E+00
1.00E+00 4.500
2.00E+00
3.00E+00
4.00E+00
5.00E+00 3.500
0 2E 2
1.500
LEGEND
6.750
14-Jul-08 15:43
step 115795
Flow Time 1.5168E+09
Dynamic Time 4.0000E+01 6.250
3.391E+02 <x< 8.089E+02
3.014E+02 <y< 7.712E+02
5.750
Exaggerated Grid Distortion
Magnification = 1.000E+00
Max Disp = 5.051E+00
Exaggerated Boundary Disp. 5.250
Magnification = 0.000E+00
Max Disp = 5.051E+00
4.750
4.250
3.750
3.250
LEGEND
14-Jul-08 15:44
step 115795
Flow Time 1.5168E+09 1.000
0 5E 2
-0.200
JOB TITLE : .
LEGEND
-2.500
-3.000
-3.500
5 10 15 20 25 30 35 40
JOB TITLE : .
HISTORY PLOT
Y-axis : 0.600
12 Pore pressure ( 49, 23)
13 Effective SYY ( 49, 23) 0.400
X-axis :
5 Dynamic time
0.200
0.000
-0.200
0 5 10 15 20 25 30 35 40
Figure 1.136 Pore-pressure and effective vertical stress near upstream slope –
Byrne (liquefaction) material and hysteretic damping
LEGEND
14-Jul-08 15:46
step 115795
Flow Time 1.5168E+09 1.000
EX_ 5 Contours
9.00E-01 0.600
1.00E+00
1.10E+00
1.20E+00
1.30E+00
1.40E+00
1.50E+00 0.200
1.60E+00
1.70E+00
1.80E+00
0 5E 2
Figure 1.137 Excess pore-pressure ratio contours (values greater than 0.99) –
Byrne (liquefaction) material and hysteretic damping
8.500
LEGEND
8-Jun-08 14:02
step 1179516 7.500
X-displacement contours
-3.00E+00
5.500
-2.00E+00
-1.00E+00
0.00E+00
1.00E+00 4.500
2.00E+00
3.00E+00
4.00E+00
3.500
Contour interval= 1.00E+00
Boundary plot
0 2E 2 2.500
1.500
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
2.000 3.000 4.000 5.000 6.000 7.000 8.000 9.000
(*10^2)
LEGEND
6.750
8-Jun-08 14:01
step 1179516
Flow Time 1.5168E+09
Dynamic Time 4.0000E+01 6.250
3.391E+02 <x< 8.089E+02
3.014E+02 <y< 7.712E+02
5.750
Exaggerated Grid Distortion
Magnification = 1.000E+00
Max Disp = 4.713E+00
Exaggerated Boundary Disp. 5.250
Magnification = 0.000E+00
Max Disp = 4.713E+00
4.750
4.250
3.750
3.250
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
3.750 4.250 4.750 5.250 5.750 6.250 6.750 7.250 7.750
(*10^2)
LEGEND
8-Jun-08 14:02
step 1179516
Flow Time 1.5168E+09 1.000
0 5E 2
-0.200
JOB TITLE : .
LEGEND
-2.500
-3.000
-3.500
5 10 15 20 25 30 35
JOB TITLE : .
8-Jun-08 14:04
0.800
step 1179516
Flow Time 1.5168E+09
Dynamic Time 4.0000E+01
0.600
HISTORY PLOT
Y-axis :
12 Pore pressure ( 49, 23) 0.400
13 Effective SYY ( 49, 23)
X-axis :
5 Dynamic time 0.200
0.000
-0.200
5 10 15 20 25 30 35
Figure 1.142 Pore-pressure and effective vertical stress near upstream slope –
Byrne (liquefaction) material and Rayleigh damping
LEGEND
8-Jun-08 14:04
step 1179516
Flow Time 1.5168E+09 1.000
EX_ 5 Contours
9.00E-01 0.600
1.00E+00
1.10E+00
1.20E+00
1.30E+00
1.40E+00
1.50E+00 0.200
1.60E+00
-0.200
0 5E 2
Figure 1.143 Excess pore-pressure ratio contours (values greater than 0.99) –
Byrne (liquefaction) material and Rayleigh damping
set funsat=on
set fastwb=on
solve
save edam5.sav
strain_ini_hist
set echo off
call strain_hist.fis
strain_hist
set echo off
call reldispx.fis
reldispx
set echo off
call inipp.fis
inipp
set echo off
call excpp.fis
excpp
history 5 dytime
history 7 vsxy i=77, j=20
history 8 str_77_20
history 9 reldispx
history 10 reldispy
history 11 excpp
history 12 pp i=49, j=23
history 13 esyy i=49, j=23
history 14 xaccel i=77, j=1
history 15 xaccel i=77, j=29
history 16 xaccel i=172, j=29
history 17 xaccel i=9, j=29
set step=10000000
prop ff_latency 50 region 78 19
prop ff_latency 50 region 60 22
prop ff_latency 50 region 103 26
save edam15.sav
save edam16.sav
;*** BRANCH: RAY 0.063 GMAX 0.8 WITH EXCESSPP AMP FACTOR=1.33 ****
restore edam15.sav
Several examples to validate and demonstrate the dynamic option in FLAC are presented. The data
files for these examples are contained in the “ITASCA\FLAC600\Dynamic” directory.
A column of elastic material resting on a rigid base has natural periods of vibration, depending on
the mode of oscillation and the confining conditions. Three cases are examined: an unconfined
column; a confined column in compression; and a column in shear.
The column is loaded by applying gravity either in the x- or y-direction and observing the oscillations
with zero damping. The case of confined compression is modeled by inhibiting lateral displacement
along the vertical boundaries, which prevents lateral deformation of the mesh. For unconfined
compression, lateral displacement is not inhibited. For the column in shear, vertical motion is
inhibited to eliminate bending modes; the loading is applied laterally.
The theoretical value for natural period of oscillation, T , is given by Eq. (1.125),
ρ
T = 4L (1.125)
E∗
FLAC data files for the three cases are given in Examples 1.31, 1.32 and 1.33. Material properties
are given below.
The theoretical periods and calculated (FLAC) natural periods of oscillation averaged over several
periods by the FISH function crossings are compared in Table 1.6 (see Example 1.34):
The program SHAKE is widely used in the field of earthquake engineering for computing the
seismic response of horizontally layered soil deposits. Here, we compare FLAC with SHAKE for
the case of a one-dimensional layered, linear-elastic soil deposit, driven at its base by the horizontal
acceleration given by Eq. (1.126)
ü(t) = βe−αt t γ sin(2πf t) (1.126)
where:α = 2.2;
β = 0.375;
γ = 8.0; and
f = 3 Hz.
This input acceleration wave, plotted in Figure 1.144, shows a maximum horizontal acceleration of
0.2 g reached after 3.75 seconds. The wave form selected for this comparison test does not require
a baseline correction (see Section 1.4.1.2); the final velocity and displacement are both zero. Also,
this form does not contain high-frequency components that could cause numerical distortion of the
wave (see Section 1.4.2).
0.20
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.20
0 2 4 6 8 10 12 14
FLAC SHAKE
The version of SHAKE used in the comparison is SHAKE-91 (Idriss and Sun 1992). SHAKE-91 is a
modified version of SHAKE (originally published in 1971 by the Earthquake Engineering Research
Center at the University of California in Berkeley, California, SHAKE-91 can be downloaded at
http://nisee.berkeley.edu/software/). SHAKE-91 computes the response of a semi-infinite horizon-
tally layered soil deposit overlying a uniform half-space subjected to vertically propagating shear
waves. The program performs a linear analysis in the frequency domain; an iterative procedure
accounts for some of the nonlinear effects in the soil. In this verification problem, we assume that
the soil is linear. (A comparison assuming nonlinear elastic material is given in Section 1.7.3).
FLAC is compared to SHAKE-91 for the following problem conditions. A layered soil deposit is
160 feet thick and contains two materials, as shown in Figure 1.145. The stiffer layer (material 2)
is 40 feet thick, starts at a depth of 40 feet, and is sandwiched between the softer layers (material
1).
11-Jun-08 14:25
step 29649
4.000
Dynamic Time 2.0000E+01 3
5
-3.099E+01 <x< 3.404E+01
-8.128E+00 <y< 5.690E+01 4
6
shear_mod 3.000
1.500E+08
3.000E+08
History Locations
2.000
1.000
8 0.000
The soil is treated as a linear elastic material, with the following properties:
material 1 2
shear modulus (MPa) 150 300
density (kg/m3 ) 1800 2000
fraction of critical damping 10% 10%
By assuming that shear modulus and damping are strain-independent, the same properties are used
in FLAC and SHAKE-91. The data file for the analysis with SHAKE-91 is shown in Example 1.35.
The file for the FLAC analysis is given in Example 1.36.
The FLAC model consists of 16 square zones, each with a length of 10 feet (3.05 m); the zone length
is well within 1/10 of the longest wavelength, to provide accurate wave transmission. Vertical
movement is prevented at the sides of the model. Rayleigh damping is specified at 10%, operating
at a center frequency of 3 Hz.
Figure 1.146 shows the horizontal acceleration at the top of the model (gridpoint 17 in FLAC, and
sub-layer 1 in SHAKE-91) as a function of time. Both records are very similar: the maximum
acceleration calculated by FLAC is 0.160 g; the maximum acceleration calculated by SHAKE-91 is
0.156 g (a 2.6% difference).
Figures 1.147 and 1.148 show the evolution of shear stress and shear strain at a depth of 35 feet
(within material 1). The FLAC results have been obtained through a history of σxy at zone 13 and
the FISH function shrstr13. The SHAKE-91 results have been obtained at the top of sub-layer
5, using analysis option 7 in the code. The results from both codes are again very similar, with
a difference of less than 4%. Note that the stress histories do not contain the viscous component
contributed by Rayleigh damping.
Figure 1.149 shows two shear stress vs strain curves calculated by SHAKE-91: one at a depth of
35 feet (in material 1), and the other at a depth of 45 feet (in material 2). They have been obtained
through option 7 at the top of sub-layers 5 and 7. In both cases, the relation between stress and
strain is linear, with a slope equal to the shear modulus.
Figure 1.150 shows a viscous shear stress (σxy v ) vs strain plot, calculated by FLAC at the same
locations as in the previous figure. While the average slope is again equal to the shear modulus,
this plot shows hysteresis loops due to the viscous damping. If we were to plot FLAC shear stress
(σxy ) versus shear strain, we would obtain a linear relation.
0.20
0.15
0.10
0.05
0.00
-0.05
-0.10
-0.15
-0.20
0 2 4 6 8 10 12 14
FLAC SHAKE
2.0E-02
1.5E-02
1.0E-02
5.0E-03
0.0E+00
-5.0E-03
-1.0E-02
-1.5E-02
-2.0E-02
0 2 4 6 8 10 12 14
FLAC SHAKE
40
30
20
10
0
-10
-20
-30
-40
0 2 4 6 8 10 12 14
FLAC SHAKE
Figure 1.149 Shear stress versus shear strain in material 1 and material 2
(SHAKE-91 results)
Figure 1.150 Viscous shear stress versus shear strain in material 1 and material
2 (FLAC results)
1. 3. 10.
10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00
10.00 10.00 10.00
8 material #3 modulus
.0001 0.0003 0.001 0.003 0.01 0.03 0.1 1.0
1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
5 damping for material #3
.0001 0.001 0.01 0.1 1.
10.00 10.00 10.00 10.00 10.00
3 1 2 3
Option 2 -- Soil Profile
2
1 19 Example -- 160-ft layer with 2 materials
1 2 10.00 3130.086 .100 .11225
2 2 10.00 3130.086 .100 .11225
3 2 10.00 3130.086 .100 .11225
4 2 5.00 3130.086 .100 .11225
5 2 5.00 3130.086 .100 .11225
6 1 5.00 6260.173 .100 .12473
7 1 5.00 6260.173 .100 .12473
8 1 10.00 6260.173 .100 .12473
9 1 10.00 6260.173 .100 .12473
10 1 10.00 6260.173 .100 .12473
11 2 10.00 3130.086 .100 .11225
12 2 10.00 3130.086 .100 .11225
13 2 10.00 3130.086 .100 .11225
14 2 10.00 3130.086 .100 .11225
15 2 10.00 3130.086 .100 .11225
16 2 10.00 3130.086 .100 .11225
17 2 10.00 3130.086 .100 .11225
18 2 10.00 3130.086 .100 .11225
19 3 .100 .140 4000.
Option 3 -- input motion:
3
2048 2048 .01 inp3.acc (8f10.6)
1.0 100. 0 8
Option 4 -- sublayer for input motion (1):
4
19 1
Option 5 -- number of iterations & ratio of avg strain to max strain
5
1 2 1.00
Option 6 -- sublayers for which accn time histories are computed and saved:
6
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
0 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 0 0 0 0 1 0 0 0 0 1 0 0 0 0
Option 6 -- sublayers for which accn time histories are computed and saved:
6
16 17 17 19 19
1 1 1 1 0
0 0 0 0 1
option 7 -- sublayer shear stress or strain are computed and saved:
7
5 1 1 0 2048 -- stress in level 5
5 0 1 0 2048 -- strain in level 5
option 7 -- sublayer shear stress or strain are computed and saved:
7
7 1 1 0 2048 -- stress in level 7
7 0 1 0 2048 -- strain in level 7
execution will stop when program encounters 0
0
1.7.3 Comparison of FLAC to SHAKE for a Layered, Nonlinear, Elastic Soil Deposit
In this section, we compare a simulation of wave propagation through a geological profile using
FLAC, first with hysteretic damping (described in Section 1.4.3.4) and then with Rayleigh damping
(described in Section 1.4.3.1) to that using SHAKE. The FLAC and SHAKE models simulate the
following problem: a horizontally layered soil deposit overlying a rigid bedrock is subjected to
a horizontal acceleration base motion. The soil deposit is 150 feet (45.72 m) deep and contains
10 different soil types. The soils are treated as nonlinear elastic materials, by assuming that shear
moduli and damping are strain-dependent. Table 1.7 summarizes the properties and locations of
the soil layers in the deposit. The dynamic characteristics of these soils are governed by two sets
of modulus reduction factor (G/Gmax ) and damping ratio (λ) versus shear strain curves: the first
set for sand, and the second set for clay.* These curves are shown in Figures 1.151 through 1.154,
and are denoted by the “SHAKE91” legend.
This example is derived from the data file provided with the downloaded SHAKE-91 code (see
http://nisee.berkeley.edu/software/). The SHAKE-91 data file is listed in Example 1.37; it is modified
such that the deepest layer is given a large wave speed, to correspond to rigid bedrock. It should
be noted that, to simulate a rigid base (Figure 1.3), we increase the stiffness (shear modulus) of the
bedrock from 3.33 GPa to 2080 GPa in the SHAKE-91 data file, for the purpose of comparison.
The base acceleration input is a set of seismic data recorded in the Loma Prieta Earthquake,
which is also downloadable (see http://nisee.berkeley.edu/software/) with the SHAKE-91 code (i.e.,
“DIAM.ACC”). The input accelerogram is shown in Figure 1.155. The record has a peak acceler-
ation of approximately 0.11 g, duration of 40 seconds, and dominant frequency of 2.47 Hz.
* More variations of G/Gmax and λ for soils are available in the literature mentioned in the SHAKE-91
manual (Idriss and Sun, 1992) and duplicated here for easy reference (e.g., Hardin and Drnevich
1970, Seed and Idriss 1970, Seed et al. 1986, Sun et al. 1988, and Vucetic and Dobry 1991).
Figure 1.151 Modulus reduction curve for dynamic property set 1 – sand
Figure 1.152 Damping ratio curve for dynamic property set 1 – sand
Figure 1.153 Modulus reduction curve for dynamic property set 2 – clay
Figure 1.154 Damping ratio curve for dynamic property set 2 – clay
SHAKE model – The free-field response of the layered soil deposit to the base-motion acceleration
is calculated using SHAKE-91. An iterative solution is performed until the secant shear moduli
and the damping ratios throughout the column converge to within a given tolerance of the values
corresponding to 50% of the maximum shear strain. Depth profiles of the maximum cyclic stress
ratio, CSR (ratio of maximum cyclic shear stress to initial vertical effective stress) and acceleration
amplification ratio (ratio of maximum acceleration to the maximum acceleration at the base) are
computed from the solution. Calculations are made for four different input accelerations, with the
record scaled to a maximum base acceleration of 0.001 g, 0.01 g, 0.10 g and 1.0 g.* Figures 1.156
and 1.157 present profiles of CSR with depth for the input record scaled to 0.001g and 0.1g,
respectively. Figure 1.158 plots the acceleration amplification ratios versus the four maximum
base accelerations.
* The scaling is accomplished through the built-in functionality in SHAKE-91, using either of two
scaling parameters in OPTION 3.
FLAC model with hysteretic damping – A 1D column FLAC model composed of 30 5-feet (1.524 m)
square elastic zones is assigned the soil properties listed in Table 1.7. Model conditions are pre-
scribed to simulate the free-field motion of a layered soil deposit with a rigid base. The FLAC data
file is listed in Example 1.38. The horizontal acceleration record is applied as an input boundary
condition at the base of the model. FLAC takes the scaled base motion from SHAKE-91. For
example, note that the history record “DIAM-FLAC-0001.ACC,” used in Example 1.38, is scaled
with a maximum value of 0.0001 g. Vertical movement is prevented in the model.
Hysteretic damping is added to this model by fitting a tangent-modulus function curve to the
(G/Gmax ) versus strain data, as shown in Figures 1.151 and 1.153. Different fitting functions
are available in FLAC (see Section 1.4.3.5); in this example, the default (two-parameter) and sig3
(three-parameter sigmoidal) functions are used. The parameters of the tangent-modulus functions
are determined by best-fitting (e.g., least squares regression analysis) the laboratory modulus-
reduction curves, assuming that the modulus reduction factor versus shear strain follows either the
default or sig3 relations. The best-fit values for dynamic property set 1 – sand are listed in Table 1.1
and for dynamic property set 2 – clay in Table 1.2. A comparison between the tangent-modulus
function curves and the laboratory curves is shown in Figures 1.151 and 1.153.
Damping ratio curves that correspond to the tangent-modulus function curves are then calculated.*
Both shear modulus reduction factor (G/Gmax ) versus strain and damping ratio (λ) versus strain
should be monitored, and parameters adjusted, if necessary, to provide a reasonable fit to both
curves over a specified strain range. As shown in Figures 1.151 though 1.154, fairly reasonable fits
for both modulus-reduction factor and damping-ratio curves are obtained with the default and sig3
hysteretic damping functions over a strain range of 0.0001% to 0.1% for both the sand and the clay.
The fitting curves of the FLAC default model (Figures 1.151 through 1.154) show that, for small
strain (corresponding to small input acceleration), the behavior is approximately linear (i.e., both
shear modulus and damping ratio are constants for both FLAC and SHAKE-91). Both codes are
thus expected to give similar results in this circumstance. Here we compare accelerograms and
response spectra at the top of the model for very low input acceleration. Figure 1.159 shows the
horizontal acceleration at the top of the model (gridpoint 31 in FLAC and sub-layer 1 in SHAKE-91)
as a function of time with maximum input acceleration amplitude of 0.0001 g. Both records are
very similar; the maximum acceleration calculated by FLAC is 0.000592 g, while the maximum
acceleration calculated by SHAKE-91 is 0.000590 g (0.4% difference).
Figure 1.159 Accelerograms at the top of the model with small input
* The data file “MODRED.DAT,” provided in the “\Dynamic” directory, can be used to perform a
series of simple shear tests on a single-zone FLAC model in order to develop damping ratio curves
that correspond to the tangent-modulus functions.
Figures 1.160 through 1.163 provide pseudo-acceleration and pseudo-velocity spectra calibrated
in SHAKE-91 and FLAC when the maximum input acceleration amplitude is small (0.0001 g).
In SHAKE-91, the response spectra are calculated using OPTION 9. In FLAC, response spectra
are computed using a FISH function, “SPEC.FIS.” Here the damping ratio, minimum period and
maximum period of interest are 5%, 0.01 and 10, respectively. From these plots it can be seen
clearly that the FLAC and SHAKE-91 results correspond quite closely for small-amplitude input
motion.
Depth profiles of cyclic stress ratio (CSR) and acceleration amplification ratio are computed from
the FLAC simulations for comparison to SHAKE-91 results (see Figures 1.156 through 1.158). It was
observed that at very low cyclic strain levels (corresponding to 0.001 g base acceleration), hysteretic
damping provides almost no energy dissipation, so a small amount of stiffness-proportional Rayleigh
damping (0.2%) was included to avoid low-level oscillations. The dynamic timestep for these FLAC
runs is 5.0×10−4 sec.
FLAC model with Rayleigh damping – The 1D column model is also run using Rayleigh damping
in place of hysteretic damping. In this example, one set of average Rayleigh damping coefficients
is assigned for each input acceleration. The parameters for the Rayleigh damping model are
selected to be compatible with the maximum strain levels calculated from the SHAKE-91 simulation.
Maximum shear strains at every layer in the SHAKE-91 model are averaged over the entire model,
and an equivalent uniform strain (the equivalent uniform strain is taken as 50% of the maximum
strain) is then used to find appropriate shear modulus-reduction factors and damping ratios for
the Rayleigh damping runs, calculated as average values from the clay and sand curves shown in
Figures 1.151 through 1.154. The values are summarized in Table 1.8. The table also lists the
resulting dynamic timestep for the selected Rayleigh damping parameters. The center frequency
for Rayleigh damping is set to the dominant frequency of the input record, 2.47 Hz.
JOB TITLE : .
4.500
11-Jun-08 14:51
step 68602
Dynamic Time 4.0480E+01 4.000
2.500
2.000
1.500
1.000
0.500
1 2 3 4 5 6 7 8 9 10
Figure 1.160 Pseudo-acceleration spectrum at the top of the model (FLAC de-
fault) – m/s2
JOB TITLE : .
11-Jun-08 14:51
3.000
step 68602
Dynamic Time 4.0480E+01
2.000
1.500
1.000
0.500
1 2 3 4 5 6 7 8 9 10
Figure 1.162 Pseudo-velocity spectrum at the top of the model (FLAC default)
– m/s
Depth profiles of CSR and acceleration amplification are also created for the Rayleigh damping
runs and are plotted in Figures 1.156 through 1.158 for comparison. The CSR profiles and the
acceleration amplifications are similar for all simulations. The results indicate that at low input
acceleration magnitude (0.001 g) the amplification is slightly lower for the SHAKE runs than the
FLAC runs with either hysteretic damping or Rayleigh damping. This corresponds to maximum
strains of less than 0.0003%. At higher acceleration magnitudes, SHAKE predicts somewhat higher
amplification than FLAC with hysteretic damping, but lower amplification than FLAC with Rayleigh
damping. This suggests that at higher amplitudes, hysteretic damping absorbs somewhat more
energy during wave transmission than both Rayleigh damping and the equivalent linear method.
The highest equivalent strain level in these models is approximately 0.3% (at 1.0 g). For this layered
system, cyclic strain levels will be different in different locations at different times. Using hysteretic
damping, these different strain levels produce realistically different damping levels in time and
space. Constant and uniform Rayleigh damping parameters can only produce an average response.
It is possible to adjust the Rayleigh damping parameters throughout the model to account for spatial
variations, using an iterative, strain-compatible scheme, as in the equivalent linear method, in order
to produce more realistic damping levels. The 1D wave propagation simulation was repeated for
selected cases, with spatial variation in Rayleigh damping corresponding to the damping levels
derived from SHAKE. These results were essentially the same as those from the Rayleigh damping
simulations using average values.
The FLAC simulations with Rayleigh damping are slower than those with hysteretic damping
because the stiffness-proportional component of Rayleigh damping causes a reduction in the critical
timestep in an explicit solution scheme. The speed of the hysteretic damping runs range from 9%
faster (at 0.001 g) to 7 times faster (at 1.0 g) than the Rayleigh damping runs.
his write 231 vs 2 table 231 ; top accn hist to table, to gen. res spectra
call spec.fis
def compuspec
; setup values for fish function spectra
dmp=0.05 ; damping ratio
pmin=0.01 ; minimum period
pmax=10.0 ; maximum period
acc_in=231 ; input acc table
sd_out = 501 ; relative displacement table
sv_out = 502 ; pseudo-velocity table
sa_out = 503 ; pseudo-acc table
n_point = 500 ; # of computation points
spectra
end
compuspec
set hisfile inp-flac-0001.his
his write 231, 224, 201 vs 2 ;accn
save inp-flac-0001.sav
This problem concerns the effects of a planar discontinuity on the propagation of an incident shear
wave. Two homogeneous, isotropic, semi-infinite elastic media, separated by a planar discontinuity
with a limited shear strength, are shown in Figure 1.164. A normally incident, plane harmonic,
shear wave will cause slip at the discontinuity, resulting in frictional energy dissipation. Thus, the
energy will be reflected, transmitted and absorbed at the discontinuity. The problem is modeled
with FLAC, and the results are used to determine the coefficients of transmission, reflection and
absorption. These coefficients were compared with ones given by an analytical solution (Miller
1978).
UT
UI UR
A
The coefficients of reflection (R), transmission (T ) and absorption (A) given by Miller (1978) are
ER
R= (1.127)
EI
ET
T = (1.128)
EI
A= 1 − R2 − T 2 (1.129)
where EI , ET and ER represent the energy flux per unit area per cycle of oscillation associated
with the incident, transmitted and reflected waves, respectively. The coefficient A is a measure of
the energy absorbed at the discontinuity. The energy flux, EI , is given by
t1 +T
EI = σs vs dt (1.130)
t1
σs = ρ c vs (1.131)
Hence,
t1 +T
EI = ρ c vs2 dt (1.132)
t1
Boundary Conditions
• Non-reflecting viscous boundaries are located at GH and CD.
• Vertical motion is prevented along lateral boundaries GC and DH.
Loading Conditions
• Shear stresses corresponding to the incident wave are applied along CD.
• The maximum stress of the incident wave is 1 MPa and the frequency is 1 Hz.
Material Conditions
• elastic media
Note that the magnitude of the incident wave must be doubled in the numerical model to account
for the simultaneous presence of the non-reflecting boundary (see Section 1.4.1.1).
Example 1.39 provides a data file that makes four complete simulations of the problem: the first
simulation is for a fully elastic case; and the remaining simulations correspond to the various values
of cohesion. Computed values for R, T and A are written to the log file “FLAC.LOG” if the model
is run in command-driven mode. If run in GIIC mode, the computed values are displayed in the
Console resource pane.
end
def common
command
hist reset
set dytime = 0.0
ini xvel = 0.0 yvel = 0.0 xdis = 0.0 ydis = 0.0
ini sxx = 0.0 syy = 0.0 szz = 0.0 sxy = 0.0
apply remove i 1,5 j 1,16
apply remove i 1,5 j 17,32
apply xquiet yquiet j=1
apply xquiet yquiet j=32
apply sxy 2 his fsin j=1
his nstep 10
his unbal
his dytime
his xvel i=2 j=1
his xvel i=2 j=32
his xdisp i=2 j=1
his xdisp i=2 j=32
his sxy i=1 j=1
his sxy i=1 j=31
his fsin
solve dytime 5
end_command
end
def energy ;- compute energy coefficients for slipping-joint example -
;
; table 1 - x-velocity at point A for elastic joint case
; table 2 - x-velocity at point A for slipping joint case
; table 3 - x-velocity at point B for slipping joint case
; Ei - energy flux for incident wave
; Et - energy flux for transmitted wave
; Er - energy flux for reflected wave
; AAA - a measure of energy absorbed at the interface
; items - no. of items in tables
;
Cs = sqrt(mat_shear / mat_dens)
factor = mat_dens * Cs
Ei = 0.0
Et = 0.0
Er = 0.0
t_n_1 = 0.0
nac = 0
AAA = 0.0
TTT = 0.0
RRR = 0.0
loop i (1,items)
t_n = xtable(1,i)
d_t = t_n - t_n_1
t_n_1 = t_n
Vsa_e = ytable(1,i)
Vsa_s = ytable(2,i)
Vsb_s = ytable(3,i)
Ei = Ei + factor * Vsa_e * Vsa_e * d_t
Et = Et + factor * Vsb_s * Vsb_s * d_t
Er = Er + factor * (Vsa_s-Vsa_e) * (Vsa_s-Vsa_e) * d_t
if i > i_mean
nac = nac + 1
RRR = RRR + sqrt(Er/Ei)
TTT = TTT + sqrt(Et/Ei)
endif
end_loop
RRR = RRR / float(nac)
TTT = TTT / float(nac)
AAA = AAA + sqrt(1.0-RRR*RRR-TTT*TTT)
command
set log on
end_command
ii = out(’ R = ’+string(RRR))
ii = out(’ T = ’+string(TTT))
ii = out(’ A = ’+string(AAA))
command
set log off
end_command
end
setup
prop den=0.00265 bulk=16667 shear=10000
int 1 kn=10000 ks=10000 coh=2.5 fric=0.0
fix y
set clock 1000000 step 10000000
set dynamic on
common
his write 3 vs 2 tab 1 ; save elas incident wave
save dinte.sav
int 1 coh 0.5
common
his write 3 vs 2 tab 2
his write 4 vs 2 tab 3
set items 701 i_mean=600
energy
save dintp5.sav
int 1 coh 0.1
common
his write 3 vs 2 tab 2
his write 4 vs 2 tab 3
energy
save dintp1.sav
int 1 coh 0.02
common
his write 3 vs 2 tab 2
his write 4 vs 2 tab 3
energy
save dintp02.sav
viscous boundary
G H
point “B”
E F interface
y
fixed in y
x
point “A”
C D
The initial assumption of an elastic discontinuity is achieved by assigning a high cohesion (2.5 MPa,
in this case) to the interface. Figure 1.166 shows the time variation of shear stress near points A and
B. From the amplitude of the stress history at A and B, it is clear that there was perfect transmission
of the wave across the discontinuity. It is also clear from Figure 1.166 that the viscous boundary
condition provides perfect absorption of normally incident waves. Following the execution of the
elastic case, the velocity history at point A is saved in table 1, to be used later for calculating EI ,
used in the equations for energy coefficients.
The cohesion of the discontinuity is then set, successively, to 0.5, 0.1 and 0.02 MPa to permit slip to
occur. The recorded shear stresses at points A and B for the three cases are shown in Figures 1.167,
1.168 and 1.169, respectively. The peak stress at point A is the superposition of the incident wave
and the wave reflected from the slipping discontinuity. It can be seen in Figures 1.167 through
1.169 that the shear stress of point B is limited by the discontinuity strength.
After each inelastic simulation, the velocity histories at points A and B are saved in tables 2 and
3, and the energy flux and coefficients R, T and A are computed by the FISH function energy
and written to the log file. All conditions are then reset to zero and requested histories are deleted,
in preparation for the next simulation; this is done in function common. It was determined that at
least five cycles of the input wave were necessary before the computed coefficients settled down
to steady-state values. Even then, there is a periodic fluctuation in the values. In order to obtain
mean values, the coefficient values were averaged over the final 100 timesteps: the FISH variable
i mean controls the step number at which this averaging process starts. Figure 1.170 compares the
numerical results with the exact solution for the coefficients for three values of the dimensionless
parameter
ωγ U
τs
JOB TITLE : .
LEGEND
0.800
11-Jun-08 15:04
step 7016
Dynamic Time 5.0001E+00 0.600
-0.400
-0.600
-0.800
5 10 15 20 25 30 35 40 45
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.166 Time variation of shear stress at points A and B for elastic dis-
continuity (cohesion = 2.5 MPa)
JOB TITLE : .
LEGEND
0.800
11-Jun-08 15:04
step 14032
Dynamic Time 5.0001E+00 0.600
-0.400
-0.600
-0.800
5 10 15 20 25 30 35 40 45
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.167 Time variation of shear stress at points A and B for slipping
discontinuity (cohesion = 0.5 MPa)
JOB TITLE : .
LEGEND
11-Jun-08 15:05
step 21048 0.800
Dynamic Time 5.0001E+00
HISTORY PLOT
Y-axis : 0.400
7 Ave. SXY ( 1, 1)
8 Ave. SXY ( 1, 31)
X-axis : 0.000
2 Dynamic time
-0.400
-0.800
5 10 15 20 25 30 35 40 45
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.168 Time variation of shear stress at points A and B for slipping
discontinuity (cohesion = 0.1 MPa)
JOB TITLE : .
LEGEND
11-Jun-08 15:06
step 28064 0.800
Dynamic Time 5.0001E+00
HISTORY PLOT
Y-axis : 0.400
7 Ave. SXY ( 1, 1)
8 Ave. SXY ( 1, 31)
X-axis : 0.000
2 Dynamic time
-0.400
-0.800
5 10 15 20 25 30 35 40 45
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
Figure 1.169 Time variation of shear stress at points A and B for slipping
discontinuity (cohesion = 0.02 MPa)
1
A
R
T
0.8
0.6
Coefficients
0.4
0.2
0
0.1 1 2 10 50 100 1000
λ
ω U
τs
Figure 1.170 Comparison of transmission, reflection and absorption coeffi-
cients (analytical solution from Miller 1978)
The displacement amplitude for the incident wave (U ) was obtained by monitoring the horizontal
displacement at point A for non-slipping discontinuities. As can be seen, the FLAC results agree
well with the analytical solution.
This problem concerns the propagation of a spherical wave due to an impulsive pressure (explosion)
in a sphere. In unbounded (i.e., infinite) media, two types of waves can exist: compression and
shear waves. In this problem, the axisymmetric nature of the problem eliminates the shear wave.
Therefore, only the solution for the compression wave needs to be sought. The problem provides
a test of the dynamic capabilities of FLAC, and is applicable to impact and explosion modeling.
The analytical solution (assuming that the material is elastic and isotropic) for this problem was
derived by Blake (1952). The solution is based on the following governing equation:
∂ 2φ
= Cp2 2 φ (1.133)
∂t 2
where:Cp = compressional wave velocity;
t = time;
φ = a potential function; and
2 = Laplacian operator.
Let p(t) be an impulse which jumps from zero to p0 at t = 0, and then decays exponentially with
time constant α −1 . Thus, the pressure function can be defined by
A step function of the pressure (α = 0) will be used for this problem. For such a pressure function,
the potential function which satisfies the governing equation is
p0 a 3 K 4K −1 1
φα=0 = −1 + exp(−α0 τ ) cos ω0 τ − tan √ (1.135)
ρ Cp2 r 4K − 1 4K − 1
where:a = radius of the sphere;
1−ν
K = 2(1−2ν) ;
ν = Poisson’s ratio;
r = radial coordinate;
Cp
α0 = 2aK = radiation damping constant;
τ =t− r−a
Cp ; and
√
ω0 = c
2aK 4K − 1 = natural frequency.
The radial displacement can be found by differentiating the potential function with respect to radial
distance:
∂φ p0 a 3 K √ −1 1
ur = =− −1 + 2 − 2ν exp(−α0 τ ) cos ω0 τ − tan √
∂r ρ Cp2 r 2 4K − 1
p0 a 3 K α0 √ −1 1
+ 2 − 2ν exp(−α0 τ ) cos ω0 τ − tan √
ρ Cp2 r Cp 4K − 1
ω0 √ −1 1
+ 2 − 2ν exp(−α0 τ ) sin ω0 τ − tan √ (1.136)
Cp 4K − 1
A pressure equal to 1000 Pa is applied at the inner boundary to simulate the blasting.
1.000
LEGEND
11-Jun-08 15:14
step 1773
Dynamic Time 1.0004E-01
-8.333E+01 <x< 1.833E+02 0.500
-1.333E+02 <y< 1.333E+02
Grid plot
0 5E 1
0.000
-0.500
-1.000
1.000
LEGEND
11-Jun-08 15:15
step 1773
Dynamic Time 1.0001E-01
-8.333E+01 <x< 1.833E+02 0.500
-1.333E+02 <y< 1.333E+02
Grid plot
0 5E 1
0.000
-0.500
-1.000
The radial displacement histories recorded up to 0.1 second at r = 2.051a, 3.424a and 4.867a are
given in Figures 1.173 and 1.174 for circular and rectangular outer boundaries, respectively. The
delay of the response at locations far from the sphere can be noted in both cases. In both cases,
FLAC is able to capture the response at peak and steady states. The fluctuation at late time is due
to the fact that the radiated wave is not absorbed completely by the viscous boundary.
JOB TITLE : .
11-Jun-08 15:14
step 1773 1.000
Dynamic Time 1.0004E-01
HISTORY PLOT
Y-axis : 0.800
3 X displacement( 5, 31)
4 X displacement( 10, 31)
5 X displacement( 15, 31) 0.600
13 ur5 (FISH)
14 ur10 (FISH)
15 ur15 (FISH) 0.400
X-axis :
2 Dynamic time
0.200
10 20 30 40 50 60 70 80 90
-03
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
JOB TITLE : .
11-Jun-08 15:15
step 1773 1.000
Dynamic Time 1.0001E-01
HISTORY PLOT
Y-axis : 0.800
3 X displacement( 5, 31)
4 X displacement( 10, 31)
5 X displacement( 15, 31) 0.600
13 ur5 (FISH)
14 ur10 (FISH)
15 ur15 (FISH) 0.400
X-axis :
2 Dynamic time
0.200
10 20 30 40 50 60 70 80 90
-03
(10 )
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
; Boundary Condition
; ---------------------
apply pressure 1000 i=1
apply xquiet yquiet i=31
fix x j=1
fix x j=61
; -----------
; Histories
; -----------
his nstep 10
his unbal
his dytime
his xdisp i=5 j=31
his xdisp i=10 j=31
his xdisp i=15 j=31
his xvel i=5 j=31
his xvel i=10 j=31
his xvel i=15 j=31
his sig1 i=5 j=31
his sig1 i=10 j=31
his sig1 i=15 j=31
; ---------------------
; Analytical Solutions
; ---------------------
def Anal
jp=31
ip=5
ur
ur5 = ur
ip=10
ur
ur10 = ur
ip=15
ur
ur15 = ur
end
def const
p0=1000.
mu=0.25
zou = density(1,1)
Cp=sqrt(bulk_mod(1,1)+4.0*shear_mod(1,1)/3.0)/sqrt(zou)
W0=(Cp/rmin)*sqrt(1.0-2.0*mu)/(1.0-mu)
alpha0=(Cp/rmin)*(1.0-2.0*mu)/(1.0-mu)
cap_k = (1.0-mu)/(2.*(1.-2.*mu))
c1 = p0*rmin*rmin*rmin*cap_k/(zou*Cp*Cp)
c2 = atan(1./sqrt(4.0*cap_k-1.0))
c3 = sqrt(2.0-2.0*mu)
c4 = alpha0/Cp
c5 = W0/Cp
end
const
def ur
r = sqrt(x(ip,jp)*x(ip,jp)+y(ip,jp)*y(ip,jp))
tau = dytime - (r-rmin)/Cp
if tau >= 0.0 then
temp1 = -(c1/(r*r))*( -1.0+c3*exp(-alpha0*tau)*cos(W0*tau-c2) )
temp2 = (c1/r) *( c4*c3*exp(-alpha0*tau)*cos(W0*tau-c2) )
temp3 = (c1/r) *( c5*c3*exp(-alpha0*tau)*sin(W0*tau-c2) )
ur = temp1 + temp2 + temp3
else
ur = 0.0
end_if
end
his Anal
his ur5
his ur10
his ur15
set step 1000000 clock 100000000
set large
set dy_damp=rayl 0.01 50 stiff
solve dytime 0.1
save sphere1.sav
new
config dyn axi
grid 30 60
model elastic
; --------------------
; Poisson’s ratio 0.25
; --------------------
prop shear=1e10 bulk=1.665e10 dens=1675
; create cavity
; ------------------
; circular boundary
; ------------------
;call hdonut.fis
; ---------------------
; rectangular boundary
; ---------------------
call hhole.fis
set rmin 10.0 rmul 10 gratio 1.01
;hdonut
hhole
; ---------------------
; Boundary Condition
; ---------------------
apply pressure 1000 i=1
apply xquiet yquiet i=31
fix x j=1
fix x j=61
; -----------
; Histories
; -----------
his nstep 10
his unbal
his dytime
his xdisp i=5 j=31
his xdisp i=10 j=31
his xdisp i=15 j=31
his xvel i=5 j=31
his xvel i=10 j=31
his xvel i=15 j=31
his sig1 i=5 j=31
his sig1 i=10 j=31
his sig1 i=15 j=31
; ---------------------
; Analytical Solutions
; ---------------------
def Anal
jp=31
ip=5
ur
ur5 = ur
ip=10
ur
ur10 = ur
ip=15
ur
ur15 = ur
end
def const
p0=1000.
mu=0.25
zou = density(1,1)
Cp=sqrt(bulk_mod(1,1)+4.0*shear_mod(1,1)/3.0)/sqrt(zou)
W0=(Cp/rmin)*sqrt(1.0-2.0*mu)/(1.0-mu)
alpha0=(Cp/rmin)*(1.0-2.0*mu)/(1.0-mu)
cap_k = (1.0-mu)/(2.*(1.-2.*mu))
c1 = p0*rmin*rmin*rmin*cap_k/(zou*Cp*Cp)
c2 = atan(1./sqrt(4.0*cap_k-1.0))
c3 = sqrt(2.0-2.0*mu)
c4 = alpha0/Cp
c5 = W0/Cp
end
const
def ur
r = sqrt(x(ip,jp)*x(ip,jp)+y(ip,jp)*y(ip,jp))
tau = dytime - (r-rmin)/Cp
if tau >= 0.0 then
temp1 = -(c1/(r*r))*( -1.0+c3*exp(-alpha0*tau)*cos(W0*tau-c2) )
temp2 = (c1/r) *( c4*c3*exp(-alpha0*tau)*cos(W0*tau-c2) )
temp3 = (c1/r) *( c5*c3*exp(-alpha0*tau)*sin(W0*tau-c2) )
ur = temp1 + temp2 + temp3
else
ur = 0.0
end_if
end
his Anal
his ur5
his ur10
his ur15
set step 1000000 clock 100000000
set large
set dy_damp=rayl 0.01 50 stiff
solve dytime 0.1
save sphere2.sav
def hdonut
figp=igp
fjgp=jgp
loop j (1,jgp)
alfa= -pi/2.0 + (j-1)*pi/(jgp-1)
rmaxit=rmin*rmul
loop i (1,igp)
ro=rmin+(rmaxit-rmin)*(gratioˆ(i-1)-1)/(gratioˆ(igp-1)-1)
x(i,j)=ro*cos(alfa)
y(i,j)=ro*sin(alfa)
end_loop
end_loop
end
The design of a machine foundation includes the estimation of the anticipated translational and
rotational motions of the machine-foundation-soil system. This example demonstrates the calcula-
tion of the vertical response of a machine foundation consisting of a rigid, massive, strip foundation
resting on a soil and excited by an oscillating machine force.
A foundation vibrating under the action of a time-varying force, P , transmits to the soil a force, R,
causing a vertical uniform displacement, δ, of the soil beneath the foundation (see Figure 1.175).
The dynamic equilibrium of the mass is expressed as
R + M δ̈ = P (1.137)
The analytical solution to this problem, assuming an elastic soil behavior, is provided by Gazetas
and Roesset (1979). Their solution is for the case of a sinusoidal vertical loading of the form
P = Po sin ωt (1.138)
in which Po is the force amplitude and ω is the operational frequency of the machine in radians per
second.
For a harmonic exciting force, P , a dimensionless displacement (compliance) function, Fv , is
defined, to relate the soil reaction force, R, to the soil displacement, δ:
R
δ= Fv sin(ωt + ψ) (1.139)
G
G is the shear modulus of the soil, and Fv is the compliance function of the operational frequency,
ω, and phase angle, ψ.
Fv is a complex number and can be written as
in which f1,v is the real part representing the recoverable component of deformation, and f2,v is
the imaginary part expressing the energy dissipated by the propagating waves and soil hysteresis.
The amplitude of motion, δo , can be expressed in terms of the amplitude of the machine force, Po .
The expression is given by Gazetas and Roesset in dimensionless form to be
2 + f2 1/2
δo G f1,v 2,v
δ̄o = (1.141)
Po (1 − bao2 f1,v
2 )2 + (ba 2 f 2 )2
o 2,v
M ωB
b= 2
; ao = (1.142)
ρB Vs
where ρ = density, Vs = s-wave velocity of the soil, B = half-width of the strip foundation, and M =
total foundation mass per unit length.
Gazetas and Roesset use a semi-analytical approach to obtain the compliance function, Fv , for a
homogeneous half-space as a function of the frequency ratio, ao . The result is presented in Fig-
ure 1.176. The response of the foundation can then be evaluated for a set of operational frequencies
from Eq. (1.141).
The results presented in Figure 1.176 are for an elastic material with a Poisson’s ratio, ν, of 0.4 and
a critical damping ratio, β, of 0.05. For this example, we assume that the s-wave velocity of the
material is 1000 ft/sec, and the unit weight is 128.8 pcf (mass density is 4 slugs/ft3 ). For ν = 0.4,
the shear modulus G is 4.0 × 106 psf. The half-width of the footing foundation is 10 ft.
0.5
FLAC Gazetas
& Roesset
0.4 f1,v
f2,v
0.3
0.2
fv
0.1
0.0
-0.1
ao
The FLAC model consists of an 80 × 40 zone grid with the foundation represented by structural-
beam elements. A vertical symmetry plane is assumed through the center of the foundation. Viscous
boundaries are located on the bottom and right side of the mesh. Figure 1.177 shows the model
grid, beam elements and boundary conditions.
LEGEND
-0.900
-1.100
The foundation is made rigid by slaving the structural nodes together. For the case of an oscillating
rigid footing, we fix the x and rotational degrees-of-freedom, and we slave all y degrees-of-freedom
together. Hence, the final structural model consists of only one degree of freedom. The density is
set to a low value so that the structural mass is insignificant, and the Young’s modulus is set low
because it is now irrelevant: this allows the timestep to be that of the soil. The machine forcing
function is applied as a stress because this is more convenient.
Calculations are performed over a range of frequency ratios: ao is varied from 0.5 to 2.5. Several
constants are computed prior to each calculation. These constants and the displacement results
are stored for comparison to the analytical solution. The data file for this example is contained in
Example 1.43.
The values for f1,v and f2,v at each ao are computed in the FISH function “COMPL.FIS” (see
Example 1.44). The average phase angle and displacement amplitude are also calculated. The
FLAC results for f1,v and f2,v are plotted in Figure 1.176 for comparison with the Gazetas and
Roesset solution.
The FLAC results compare well with the analytical results. Note that the agreement is better at the
higher values of ao . For lower frequencies, a larger grid is required. For example, for ao = 0.5,
the wavelength is 125 ft. The distance to the boundaries should be several wavelengths from the
foundation, particularly for the top surface, because Rayleigh waves are not damped very efficiently
by the quiet boundaries.
The foundation could also be simulated without structural elements. An oscillating velocity could
be applied to a line of fixed gridpoints, and the cyclic reaction force measured. The same information
(compliance) could be determined from the ratio between displacement and force, and their phase
angle.
end
ggg
def s_wave
s_wave = sin(omega * dytime)
end
def dummy ; ... count number of history points
count = count + 1
end
set dy_damp rayl 0.05 frq_cent
fix x i=1
app xquiet yquiet i=81
app xquiet yquiet j=1
app syy=0.05 hist=s_wave i=1,6 j=41 ; pressure of amplitude 0.05 (F=1.0)
hist dytime
hist syy i 1 j 40
hist syy i 5 j 40
hist ydis i 1 j 41
hist dummy
set ncw=50 clock=10000000 step=10000000
solve dytime=final_t
save mach.sav
call compl.fis
save mach2.sav
The analysis of the jump conditions across a shock wave is given in many textbooks (e.g. Zukas
and Walters 1997). The main points of such analyses are reproduced here.
Inert shock wave – Consider a frictionless tube containing a polytropic fluid obeying the equation
of state given by Eq. (1.143),
pV
e= (1.143)
(γ − 1)
where e is the internal energy, p is the pressure, V = 1 / ρ is the specific volume and γ is the ratio of
specific heats. The gas in the tube starts with an initial density of ρ0 (= 1 / V0 ) and pressure of ρ0 . If
one end of the tube contains a piston that moves in the positive coordinate direction with velocity u,
then a shock front, moving at velocity D, develops at some distance ahead of the piston. We wish
to determine D, the pressure, p1 , and the density, ρ1 (= 1 / V1 ), behind the shock. The conservation
laws (mass, momentum and energy, respectively) applied to the jump across the shock are
V0 (D − u) = V1 D (1.144)
Du = V0 (p1 − p0 ) (1.145)
p1 V1 p0 V0 u2
D − + = V0 p1 u (1.146)
γ −1 γ −1 2
δ
p1 = p0 1+ (1.147)
V1 = V0 (1 − ) (1.148)
u
D= (1.149)
where
2 1/2
(γ + 1)δ (γ + 1)δ δ
=− + + (1.150)
4γ 4γ γ
u2
δ= (1.151)
p0 V0
Shock smoothing scheme – The equations derived above imply that there is a step jump in pressure,
velocity and density across the shock, but a “standard” FLAC simulation shows oscillations in these
quantities superimposed on the steps. As Zel’dovich and Raizer (2002) remark:
The increase in entropy indicates that irreversible dissipative processes (which can be
traced to the presence of viscosity and heat conduction in the fluid) occur in the shock
wave. A theory that does not take into account these processes is not capable of describing
either the mechanism or the structure of the very thin layer where the gas undergoes a
transition from the initial to the final state.
Thus, we add a mechanism (Eq. (1.152)) that accounts for the necessary entropy increase across
the shock, by “leaking” heat from the hot side to the cold side. Note that energy is conserved:
∂e ∂p ē
=α (1.152)
∂t ∂x p̄
where ē is the mean energy and p̄ the mean pressure at a point. In discrete form,
pi − pi−1 t
ei = α (ei + ei−1 ) (1.153)
pi + pi−1 x
where t and x are timestep and zone size, respectively, and the subscripts refer to gridpoint or
zone indices: ei is the increment in energy flowing from zone i to zone i − 1 in one timestep.
Using the equation of state Eq. (1.143), the energy increment can be written as follows:
pi − pi−1 pi Vi + pi−1 Vi−1 t
ei = α (1.154)
pi + pi−1 γ −1 x
The energy flux is first computed for each gridpoint i. Then all zones are scanned and energy
added to, or subtracted from, each zone based on the flux already computed at its two associated
gridpoints:
pi Vi
= ei − ei + ei+1 (1.155)
γ −1
ei − ei+1
pi = pi − (γ − 1) (1.156)
Vi
It is found that a value of α between 1000 and 2000 m/s gives generally good results for the systems
reported here. The shock-smoothing scheme described above is intended to perform a function
similar to the artificial viscosity scheme described by Wilkins (1980), but this approach uses only
one parameter and is believed to be related more closely to physical mechanisms.
A user-defined constitutive model is written to implement an incremental form of Eq. (1.143) in
FLAC. The model, named EoS, is written in C++ using the methodology given in Section 3 in
Theory and Background, and the DLL (dynamic link library) compiled file is named “EoS.dll.”
The DLL is loaded in FLAC using the command Model load eos.dll. Note that FLAC must first be
configured to accept the DLL model by giving the command CONFIG cppudm.
Example of one-dimensional inert shock – We apply various velocities, u, to the base of a one-
dimensional column consisting of 10 mm zones, containing gas of density 1.2 kg/m3 , γ = 3 and
initial pressure 105 Pa. FLAC is operated in Lagrangian mode, with an update frequency of 1 (SET
update=1), with α = 1000 m/s. As an example of the shock front, Figure 1.178 shows a profile, at t
= 3.027 ms, of pressure versus distance from the original origin, for the case of u = 750 m/s. Note
that the bottom of the column has moved about 2.27 m (as evidenced by the trace not starting from
zero distance) due to the applied velocity, while the shock front has moved about 5 m. There is
almost no over-shoot or oscillation in the pressure trace, which confirms that the shock-smoothing
scheme described above is working well. Table 1.9 shows the results for five driving velocities
(denoted by “num”) compared to results from the analytical expressions Eqs. (1.147), (1.148) and
(1.149), denoted by “exact.” It can be seen that the agreement is generally better than 0.5%.
The data file for this example is listed in Example 1.45. The mechanism that accounts for entropy
increase across the shock is implemented as a FISH function in this model.
JOB TITLE : .
Pressure profile
pressure (Pa) vs distance (m) 1.200
1.000
0.800
0.600
0.400
0.200
25 30 35 40 45 50 55
-01
(10 )
Itasca Consulting Group, Inc.
Minneapolis, MN 55401
Figure 1.178 Pressure profile in column at time 3.027 ms (note displaced loca-
tion of base, at 2.27 m)
AvDens = 0.0
count = 0
loop j (NzTarg-50,NzTarg-20)
AvPres = AvPres - syy(1,j)
AvDens = AvDens + zoneMulˆ2 * dens0 / area(1,j)
count = count + 1
endLoop
AvPres = AvPres / float(count)
AvDens = AvDens / float(count)
Dnorm = AvDens / dens0
Pnorm = AvPres / pzero
end
ini y mul zoneMul x mul zoneMul
prop dens=dens0 m_gamma=mgam
ini sxx=pres0 syy=pres0 szz=pres0
set large update=1
fix x
fix y j=601
def ramp
if step > nsRamp
ramp = vfinal
else
ramp = vfinal * float(step) / float(nsRamp)
endif
end
;
set nsRamp=500
fix x y j=1
ini yvel=vfinal j=1
hist dytime
hist yvel i=1 j=1
hist yvel i=1 j=50
def leak
while_stepping
ex_1(1,1) = 0.0
ex_1(1,jgp) = 0.0
loop j (2,jzones)
pdif = (syy(1,j)-syy(1,j-1))/(syy(1,j)+syy(1,j-1))
enorm = syy(1,j)/density(1,j)+syy(1,j-1)/density(1,j-1)
ex_1(1,j) = Lfac * pdif * enorm
endLoop
loop j (1,jzones)
delEn = (ex_1(1,j+1) - ex_1(1,j)) * density(1,j)
syy(1,j) = syy(1,j) + delEn
sxx(1,j) = syy(1,j)
szz(1,j) = syy(1,j)
endLoop
end
def Dfront
jyymax = 1
loop j (1,jzones)
jj = jzones - j + 1
if abs(syy(1,jj)) > abs(pres0)*1.1
Dfront = 0.5 * (y(1,jj)+y(1,jj))
exit
endif
endLoop
end
history 6 Dfront
set dydt=dtGiven
set NzTarg=200
RunIt
getResults
set NzTarg=300
RunIt
getResults
set NzTarg=400
RunIt
getResults
set NzTarg=500
RunIt
getResults
his write 6 vs 1 tab 100
def LSfit
nmax = table_size(100)
sum1 = 0.0
sum2 = 0.0
sum3 = 0.0
sum4 = 0.0
nstart = int(3.0*float(nmax)/4.0)
count = 0
loop n (nstart,nmax)
count = count + 1
sum1 = sum1 + xtable(100,n)
sum2 = sum2 + ytable(100,n)
sum3 = sum3 + xtable(100,n)ˆ2
sum4 = sum4 + xtable(100,n) * ytable(100,n)
endLoop
nff = float(count)
fitVOD = (nff*sum4 - sum1*sum2) / (nff*sum3 - sum1ˆ2)
oo = out(’ best fit VOD = ’+string(fitVOD))
end
LSfit
def pressure_profile
loop jj(1,jzones)
xtable(11,jj) = (y(1,jj+1)+y(1,jj))/2.0
ytable(11,jj) = -(sxx(1,jj)+syy(1,jj)+szz(1,jj))/3.0
endloop
end
pressure_profile
save eos.sav
;*** plot commands ****
;plot name: pressure profile
label table 11
pressure (Pa) vs distance (m)
plot hold table 11 line alias ’Pressure profile’
1.8 References
Byrne, P. M. “A Cyclic Shear-Volume Coupling and Pore-Pressure Model for Sand,” in Pro-
ceedings: Second International Conference on Recent Advances in Geotechnical Earthquake
Engineering and Soil Dynamics (St. Louis, Missouri, March 1991), Paper No. 1.24, 47-55, 1991.
Byrne, P. M., and D. L. Anderson (co-chairs). “Earthquake Design in the Fraser Delta, Task Force
Report,” City of Richmond, B.C. publication and Soil Mech. Series No. 150, Dept. of Civil
Engineering, University of British Columbia, Vancouver, B.C., 1991.
Byrne, P. M., R. Debasis, R. G. Campanella and J. Hughes. “Predicting Liquefaction Response of
Granular Soils from Self-Boring Pressuremeter Tests,” ASCE National Convention, San Diego,
California, October 23-27, ASCE, 56(GSP), pp. 122-135 (1995).
Byrne, P. M., S. S. Park and M. Beaty. “Seismic liquefaction: centrifuge and numerical modeling,”
in FLAC and Numerical Modeling in Geomechanics – 2003 (Proceedings of the 3rd International
FLAC Symposium, Sudbury, Canada, October 2003), pp. 321-331. R. Brummer, P. Andrieux, C.
Detournay and R. Hart, eds. Rotterdam: A. A. Balkema (2003).
Byrne, P. M., E. Naesgaard and M. Seid-Karbasi. “Analysis and Design of Earth Structures to Resist
Seismic Soil Liquefaction,” in Sea to Sky Geotechnique (59th Canadian Geotechnical Conference
& 7th Joint CGS/IAH-CNC Groundwater Specialty Conference, Vancouver, Canada, October
2006), pp. 1-24. R. M. Hardy Address, 2006.
Byrne, P. M., and D. Wijewickreme. “Liquefaction Resistance and Post-Liquefaction Response
of Soils for Seismic Design of Buildings in Greater Vancouver,” in Sea to Sky Geotechnique
(59th Canadian Geotechnical Conference & 7th Joint CGS/IAH-CNC Groundwater Specialty
Conference, Vancouver, Canada, October 2006), pp. 1267-1278 (2006).
Cundall, P. A. “Explicit Finite Difference Methods in Geomechanics,” in Numerical Methods
in Engineering (Proceedings of the EF Conference on Numerical Methods in Geomechanics,
Blacksburg, Virginia, June, 1976), Vol. 1, pp. 132-150 (1976).
Cundall, P. A., H. Hansteen, S. Lacasse and P. B. Selnes. “NESSI – Soil Structure Interaction
Program for Dynamic and Static Problems,” Norwegian Geotechnical Institute, Report 51508-9,
December 1980.
Dafalias, Y. F. “Bounding surface plasticity. I: mathematical foundation and hypoplasticity,”
Journal of Engineering Mechanics, ASCE, 112(9), 966-987, 1986.
Daller, J., W. Unterberger and B. Hochgatterer. “Vibration Control in Railway Tunnels,” in Pro-
ceedings of the 3rd Symposium on Tunnel Construction and Underground Structures (Ljubljana,
Slovenia, September 1996).
Dames and Moore (San Francisco, California) and Science Applications, Incorporated (Oakland,
California). “Study of Nonlinear Effects on One-Dimensional Earthquake Response,” Final Report
NP-865 to Electric Power Research Institute (Palo Alto, California), August 1978.
Darendeli, M. B. Development of a New Family of Normalized Modulus Reduction and Material
Damping Curves. Ph.D. dissertation, presented to the Faculty of the Graduate School of The
University of Texas at Austin, August 2001.
Kutter, B. L., J. C. Chou and T. Travasarou. “Centrifuge Testing of the Seismic Performance of a
Submerged Cut-and-Cover Tunnel in Liquefiable Soil,” in Geotechnical Earthquake Engineering
and Soil Dynamics IV (Proceedings of the Conference, Sacramento, California, May 2008),
Geotechnical Special Publication No. 181. D. Zeng, M. T. Manzari and D. R. Hiltunen, eds. New
York: ASCE (2008).
Lindquist, D. D. “Seismic Modeling of a 135-Foot-Tall MSE Wall,” in Geotechnical Earthquake
Engineering and Soil Dynamics IV (Proceedings of the Conference, Sacramento, California,
May 2008), Geotechnical Special Publication No. 181. D. Zeng, M. T. Manzari and D. R. Hiltunen,
eds. New York: ASCE (2008).
Lysmer, J., and R. L. Kuhlemeyer. “Finite Dynamic Model for Infinite Media,” J. Eng. Mech.,
95(EM4), 859-877, 1969.
Lysmer, J., T. Udaka, C. F. Tsai and H. B. Seed. “FLUSH – A Computer Program for Approximate 3-
D Analysis of Soil-Structure Interaction Problems,” University of California, Berkeley, Earthquake
Engineering Research Center, Report No. EERC 75-30, 1975.
Lysmer, J., and G. Waas. “Shear Waves in Plane Infinite Structures,” ASCE J. Eng. Mech., 98(EM1),
85-105, 1972.
Martin, G. R., W. D. L. Finn and H. B. Seed. “Fundamentals of Liquefaction under Cyclic Loading,”
J. Geotech., Div. ASCE, 101(GT5), 423-438, 1975.
Mejia, L. H., and E. M. Dawson. “Earthquake Deconvolution for FLAC” in FLAC and Numerical
Modeling in Geomechanics (Proceedings of the 4th International FLAC Symposium, Madrid,
Spain, May 2006), pp. 211-219. P. Varona & R. Hart, eds. Minneapolis, Minnesota: Itasca
Consulting Group Inc. (2006).
Miller, R. K. “The Effects of Boundary Friction on the Propagation of Elastic Waves,” Bull. Seismic.
Assoc. America, 68, 987-998, 1978.
Mitchell, J. K. “Recent Developments in Ground Improvement for Mitigation of Seismic Risk
to Existing Embankment Dams,” in Geotechnical Earthquake Engineering and Soil Dynamics
IV (Proceedings of the Conference, Sacramento, California, May 2008), Geotechnical Special
Publication No. 181. D. Zeng, M. T. Manzari and D. R. Hiltunen, eds. New York: ASCE (2008).
Mogami, T., and K. Kubu. “The behavior of soils during vibration,” Proceedings, 3rd International
Conference on Soil Mechanics and Foundation Engineering, Zurich, Switzerland, Vol. 1, pp 152-
155 (1953).
Newmark, N. M. “Effects of Earthquakes on Dams and Embankments,” Geotechnique, 15, No. 2,
pp. 139-160, 1965.
Newmark, N. M., and W. J. Hall. Earthquake Spectra and Design. Earthquake Engineering
Research Institute, Berkeley, California, 1982.
Olson, S. M., T. D. Stark, W. H. Walton and G. Castro. “1907 Static Liquefaction Flow Failure of
the North Dike of Wachusett Dam,” Journal of Geotechnical and Geoenvironmental Engineering,
126(12), 1184-1193, 2000.
Olson, S. M. and T. D. Stark. “Liquefied strength ratio from liquefaction flow failure case histories,”
Canadian Geotechnical Journal, 39, 629-647, 2002.
Papadimitriou, A. G., G. D. Bouckovalas and Y. F. Dafalias. “Plasticity model for sand under
small and large cyclic strains,” Journal of Geotechnical and Geoenvironmental Engineering, ASCE
127(11), 973-983, 2001.
Papadimitriou, A. G. and G. D. Bouckovalas. “Plasticity model for sand under small and large cyclic
strains: a multiaxial formulation,” Soil Dynamics and Earthquake Engineering, 22, 191-204, 2002.
Perlea, M., D. Serafini, E. Dawson and M. Beaty. “Deformation Analysis for Seismic Retrofit of an
Embankment Dam,” in Geotechnical Earthquake Engineering and Soil Dynamics IV (Proceed-
ings of the Conference, Sacramento, California, May 2008), Geotechnical Special Publication
No. 181. D. Zeng, M. T. Manzari and D. R. Hiltunen, eds. New York: ASCE (2008).
Pyke, R. “Evolution of Soil Models Since the 1970s,” prepared for Pacific Earthquake Engineering
(PEER) Center Workshop, 2004. Available from Itasca on request.
Roesset, J. M., and M. M. Ettouney. “Transmitting Boundaries: A Comparison,” Int. J. Num. &
Analy. Methods Geomech., 1, 151-176, 1977.
Roth, W. H., G. Bureau, and G. Brodt. “Pleasant Valley Dam: An approach to quantifying the
effect of foundation liquefaction,” 17th International Congress on Large Dams, Vienna, Austria,
1991.
Roth, W. H., S. Inel, C. Davis and G. Brodt. “Upper San Fernando Dam 1971 Revisited,” 10th
Annual Conference of the Association of State Dam Safety Officials, Kansas City, Missouri, 1993.
Roth W. H., et al. “Nonlinear Dynamic SSI Analysis of a Concrete Water Reservoir,” in Geotechni-
cal Earthquake Engineering and Soil Dynamics IV (Proceedings of the Conference, Sacramento,
California, May 2008), Geotechnical Special Publication No. 181. D. Zeng, M. T. Manzari and
D. R. Hiltunen, eds. New York: ASCE (2008).
Salah-Mars, S., et al. “Seismic Vulnerability of the Sacramento-San Joaquin Delta Levees,” in
Geotechnical Earthquake Engineering and Soil Dynamics IV (Proceedings of the Conference,
Sacramento, California, May 2008), Geotechnical Special Publication No. 181. D. Zeng, M. T.
Manzari and D. R. Hiltunen, eds. New York: ASCE (2008).
Schnabel, P. B., J. Lysmer and H. Bolton Seed. “SHAKE: A Computer Program for Earthquake
Response Analysis of Horizontally Layered Sites,” University of California, Berkeley, Earthquake
Engineering Research Center, Report No. UCB/EERC-71/12, 1972.
Seed, R. B. and L. F. Harder. “SPT-based analysis of cyclic pore pressure generation and undrained
residual strength,” in Proceedings of the H. B. Seed Memorial Symposium, Vol. 2, pp. 351-376.
Bi-Tech Publishing Ltd. (1990).
Seed, H. B., and I. Idriss. “Influence of Soil Conditions on Ground Motion During Earthquakes,”
J. Soil Mech. Found., Div. ASCE, 95, 99-137, 1969.
Seed, H. B., and I. Idriss. “Simplified Procedure for Evaluating Soil Liquefaction Potential,” J. Soil
Mech. & Foundations, Div. ASCE, 97(SM9), 1249-1273, 1971.
Seed, H. Bolton, and I. M. Idriss. “Soil Moduli and Damping Factors for Dynamic Response
Analysis,” Earthquake Engineering Research Center, University of California, Berkeley, Report
No. UCB/EERC-70/10, p. 48, December 1970.
Seed, H. B., P. P. Martin and J. Lysmer. “The Generation and Dissipation of Pore Water Pressures
during Soil Liquefaction,” University of California, Berkeley, Earthquake Engineering Research
Center, NSF Report PB-252 648, August 1975.
Seed, H. Bolton, R. T. Wong, I. M. Idriss and K. Tokimatsu. “Moduli and Damping Factors
for Dynamic Analyses of Cohesionless Soils,” Journal of the Geotechnical Engineering Division,
ASCE, Vol. 112, No. GT11, pp. 1016-1032, November 1986.
Sun, J. I., R. Golesorkhi and H. Bolton Seed. “Dynamic Moduli and Damping Ratios for Cohesive
Soils,” Earthquake Engineering Research Center, University of California, Berkeley, Report No.
UCB/EERC-88/15, p. 42, 1988.
Unterberger, W., P. A. Cundall and A. H. Zettler. “Dynamic Substepping – Increasing the Power of
Explicit Finite Difference Methods,” in Proceedings of the 9th International Conference on Com-
puter Methods and Advances in Geomechanics (Wuhan, China, November 1997). Rotterdam:
A. A. Balkema (1997).
Unterberger, W., R. Poisel and C. Honeger. “Numerical Prediction of Surface Vibrations Caused
by High-Speed Rail Traffic in Tunnels,” presented at the World Tunnel Congress (ITA), Vienna,
Austria, April 1977.
Vucetic, M., and R. Dobry. “Effect of Soil Plasticity on Cyclic Response,” in J. Geotech. Eng.
Div., ASCE, Vol. 111, No. 1, pp. 89-107, January 1991.
Wang, Z.-L. Bounding Surface Hypoplasticity Model for Granular Soils and Its Applications.
Ph.D. dissertation, presented to the Faculty of the Graduate School of The University of California,
Davis. Information Service, Order No. 9110679, Ann Arbor, MI 48106, 1990.
Wang, Z.-L., J. Egan, L. Scheibel and F. I. Makdisi. “Simulation of Earthquake Performance of
a Waterfront Slope Using Fully Coupled Effective Stress Approach,” in FLAC and Numerical
Modeling in Geomechanics – 2001 (Proceedings of the 2nd International FLAC Symposium,
Lyon, France, October 2001), pp. 101-108. D. Billaux, X. Rachez, C. Detournay and R. Hart, eds.
Rotterdam: A. A. Balkema (2001).
Wegel, R. L., and H. Walther. “Internal Dissipation in Solids for Small Cyclic Strains,” Physics, 6,
141-157, 1935.
Westergaard, H. M. “Water Pressure on Dams During Earthquakes,”Proceedings of ASCE, Trans-
actions number 98, Vol. 59, No. 8, Part 3, 1933.
White, W., S. Valliappan and I. K. Lee. “Unified Boundary for Finite Dynamic Models,” J. Eng.
Mech., 103, 949-964, 1977.
Wilkins, M. L. “Use of Artificial Viscosity in Multidimensional Fluid Dynamic Calculations,” in
Journal of Computational Physics, 36, 281-303, 1980.
Wolf, J. P. Dynamic Soil-Structure Interaction. New Jersey: Prentice-Hall (1985).
Yegian, M. K., et al. “Seismic Soil-Foundation Investigation of the Brooklyn Bridge,” in Geotechni-
cal Earthquake Engineering and Soil Dynamics IV (Proceedings of the Conference, Sacramento,
California, May 2008), Geotechnical Special Publication No. 181. D. Zeng, M. T. Manzari and
D. R. Hiltunen, eds. New York: ASCE (2008).
Youd, T. L., et al. “Liquefaction Resistance of Soils, Summary Report from the 1996 NCEER
and 1998 NCEER/NSF Workshops on Evaluation of Liquefaction Resistance of Soils,” Journal of
Geotechnical and Geoenvironmental Engineering, ASCE, 127(4), 297-313, 2001.
Zel’dovich, Ya. B. and Yu. P. Raizer. Physics of Shock Waves and High-Temperature Hydrody-
namic Phenomena. New York: Dover Publications (2002).
Zeng, D., M. T. Manzari and D. R. Hiltunen, eds. “Geotechnical Earthquake Engineering and Soil
Dynamics IV,” in Geotechnical Earthquake Engineering and Soil Dynamics IV (Proceedings of
the Conference, Sacramento, California, May 2008), Geotechnical Special Publication No. 181.
D. Zeng, M. T. Manzari and D. R. Hiltunen, eds. New York: ASCE (2008).
Zukas, J. A., and W. P. Walters Explosive Effects and Applications. New York: Springer (1997).