Professional Documents
Culture Documents
Flac and Numerical Modeling in Geomechanics
Flac and Numerical Modeling in Geomechanics
Edited by
Cover picture:
The “Big Nickel” is a famous landmark and symbol of Sudbury, the “Nickel Mining
Capital of the World”.
All rights reserved. No part of this publication or the information contained herein may be reproduced,
stored in a retrieval system,or transmitted in any form or by any means, electronic, mechanical,
by photocopying, recording or otherwise, without written prior permission from the publisher.
Although all care is taken to ensure the integrity and quality of this publication and the information
herein, no responsibility is assumed by the publishers nor the author for any damage to property or
persons as a result of operation or use of this publication and/or the information contained herein.
Table of contents
Preface IX
Organisation XI
Constitutive models
Compensation grouting analysis with FLAC3D 3
X. Borrás, B. Celada, P. Varona & M. Senís
An automated procedure for 3-dimensional mesh generation 9
A.K. Chugh & T.D. Stark
A new constitutive model based on the Hoek-Brown criterion 17
P. Cundall, C. Carranza-Torres & R. Hart
A study of compaction band formation with the Double-Yield model 27
C. Detournay, P. Cundall & J. Parra
A new viscoplastic model for rocks: application to the Mine-by-Test of AECL-URL 35
F. Laigle
Prediction of deformations induced by tunneling using a time-dependent model 45
A. Purwodihardjo & B. Cambou
Modeling of anhydrite swelling with FLAC 55
J.M. Rodríguez-Ortiz, P. Varona & P. Velasco
Scenario testing of fluid-flow and deformation during mineralization: from simple to
complex geometries 63
P.M. Schaubs, A. Ord & G.H. German
Constitutive models for rock mass: numerical implementation, verification and validation 71
M. Souley, K. Su, M. Ghoreychi & G. Armand
Slope stability
A parametric study of slope stability under circular failure condition by a numerical method 83
M. Aksoy & G. Once
Numerical modeling of seepage-induced liquefaction and slope failure 91
S.A. Bastani & B.L. Kutter
Complex geology slope stability analysis by shear strength reduction 99
M. Cala & J. Flisiak
Analysis of hydraulic fracture risk in a zoned dam with FLAC3D 103
C. Peybernes
Mesh geometry effects on slope stability calculation by FLAC strength reduction method –
linear and non-linear failure criteria 109
R. Shukha & R. Baker
3D slope stability analysis at Boinás East gold mine 117
A. Varela Suárez & L.I. Alonso González
Mining applications
Sill pillar design at the Niobec mine using FLAC3D 181
P. Frenette & R. Corthésy
Stability analyses of undermined sill mats for base metal mining 189
R.K. Brummer, P.P. Andrieux & C.P. O’Connor
FLAC numerical simulations of tunneling through paste backfill at Brunswick Mine 197
P. Andrieux, R. Brummer, A. Mortazavi, B. Simser & P. George
FLAC3D numerical simulations of ore pillars at Laronde Mine 205
R.K. Brummer, C.P. O’Connor, J. Bastien, L. Bourguignon & A. Cossette
Modeling arching effects in narrow backfilled stopes with FLAC 211
L. Li, M. Aubertin, R. Simon, B. Bussière & T. Belem
FLAC3D numerical simulations of deep mining at Laronde Mine 221
C.P. O’Connor, R.K. Brummer, P.P. Andrieux, R. Emond & B. McLaughlin
Three-dimensional strain softening modeling of deep longwall coal mine layouts 233
S. Badr, U. Ozbay, S. Kieffer & M. Salamon
FISH functions for FLAC3D analyses of irregular narrow vein mining 241
H. Zhu & P.P. Andrieux
VI
The usability analyses of HDPE leachate collection pipes in a solid waste landfill 287
F. Ma
FLAC numerical simulations of the behavior of a spray-on liner for rock support 295
C.P. O’Connor, R.K. Brummer, G. Swan & G. Doyle
A numerical study of the influence of piles in the passive zone of embedded retaining walls 301
T.Y. Yap & C. Pound
VII
Preface
The first two International FLAC Symposia were held in Minneapolis (USA) in September 1999, and in Lyon
(France) in October 2001. In 2003, the third International Symposium on FLAC and Numerical Modeling in
Geomechanics returned to North America and was held in Sudbury (Ontario, Canada) from October 21 to
October 24, 2003, with two days of short courses before the symposium.
Technical contributions to the conference were received from a wide range of different disciplines, repre-
senting virtually the entire globe. A volunteer Technical Committee reviewed the papers, and where necessary
clarifications were suggested to the authors prior to finalization of their manuscripts.
The contributions in this volume cover seven main topics:
• Constitutive Models
• Slope Stability
• Underground Cavity Design
• Mining Applications
• Soil Structure Interaction
• Dynamic Analyses
• Thermal Analyses
The FLAC conferences provide all FLAC and FLAC3D users with an opportunity to meet and learn from each
other and from the people who develop the code. Conversely, they also allow Itasca staff members to learn from
the practical experiences of code users “out there in the real world”. These interactions improve our collective
knowledge and allow us to improve the performance of these numerical models in simulating the behavior of
geomaterials. These proceedings contain a comprehensive collection of FLAC & FLAC3D applications – case
studies as well as research presentations. We believe that this publication will help users by documenting a valu-
able resource for the solution of geomechanical problems.
The compilation presented here would not have been possible without the efforts of our authors and our Technical
Committee, and we thank them. We particularly thank and recognize the efforts of Michele Nelson, who served as
an extremely capable and efficient Technical Editor.
IX
Organisation
XI
Constitutive models
X. Borrás
Gestió D’Infrastructures SA (GISA), Barcelona, Spain
B. Celada
Geocontrol SA, Madrid, Spain
ABSTRACT: The Barcelona Metro Line 3 extension was excavated crossing 6.5 meters below a main water
supply pipe. Compensation grouting was used to minimize the deformations in the gallery. A FLAC3D model
was developed to investigate the efficiency of this process. The model was first calibrated to reproduce the
extensometer measurements and was later re-run without the compensation grouting in order to assess the
effectiveness of such treatment.
E c
(MPa) v (°) (t/m2) (t/m3)
4.0E+04
3.5E+04
3.0E+04
y = 6644.4Ln(x) + 11076
R2 = 0.9262
2.5E+04
E (MPa)
2.0E+04
1.5E+04
1.0E+04
5.0E+03
0.0E+00
0 5 10 15 20 25 30 35
Distancia al frente (m)
03
20
The methodology used for the simulation of the
21
injection process is based in the bulb expansion
model proposed by Buchet et al. (1999). According to
02 them the injection effect can be modeled by increas-
ing the volume of the elements in which the injection
01 is made. This volume increment is carried out applying
some “fictitious” hydrostatic stresses in the element,
which makes it expand.
These stresses are applied instantaneously, initial-
izing an hydrostatic stress increment of the element as
pulses and then reaching a mechanical equilibrium.
Figure 8. Drills and sleeves location. This process is repeated until the volumetric strain
induced in the element is the one corresponding to a
fraction of the volume injected. The volumetric strain
2.4 Simulation of the compensation grouting increment due to the injection is defined by:
Compensation grouting injections consist of a mix-
ture of cement and bentonite. To simulate these injec-
tions a methodology which reproduces the injection (1)
process carried out during the real construction has
been developed. Figure 8 shows a plan view of the
treatment area with the location of the drills and where Vi is the injected volume, V0 is the initial vol-
sleeves used. ume of the element and is the efficiency of the
This real geometry has been reproduced in the injection. The process followed during the injection
FLAC3D model. Figure 9 shows the location of all modeling is shown in Figure 10.
the drills considered and the sleeves used in one of Figure 11 shows, as an example of the process, the
the injection cycles. increments applied to the vertical stress and their later
2.5
2.0
T no
v v 1.5
1.0
yes
0.5
End of injection Itasca Consulting Group, Inc. 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0
Minneapolis, MN USA NⴗPasos ⴛ10^3
Figure 11. Vertical stresses during the injection process. Figure 13. Vertical displacement increments in one of the
injection cycles, at extensometer depth.
sleeves that are injected every cycle are reproduced in Plane Origin:
X: 0.000e+000
Plane Orientation:
Dip: 0.000
10/10
11/10
12/10
13/10
14/10
15/10
16/10
17/10
18/10
19/10
20/10
21/10
22/10
23/10
E13 E6
10/9
11/9
12/9
13/9
14/9
15/9
16/9
17/9
18/9
19/9
20/9
21/9
22/9
23/9
24/9
25/9
26/9
27/9
28/9
29/9
30/9
1/10
2/10
3/10
4/10
5/10
6/10
7/10
8/10
9/10
Geología
Plane: on E7 0.002
E9
1.111000e+007 E11 0.000
Vertical displacement (m)
1.880000e+007
2.884500e+007 -0.002
Taladros E10 E8 -0.004
Extensometros -0.006
E12 -0.008
E1B
Tunel -0.010
GaleriaATLL E2B -0.012
E14
-0.014
E3B -0.016
-0.018
Itasca Consulting Group, Inc.
Minneapolis, MN USA -0.020
-0.022
Measured FLAC FLAC No Injections
Figure 15. Location of the rod extensometers.
Extensometer E4
Date Extensometer E10
10/10
11/10
12/10
13/10
14/10
15/10
16/10
17/10
18/10
19/10
20/10
21/10
22/10
23/10
Date
10/9
11/9
12/9
13/9
14/9
15/9
16/9
17/9
18/9
19/9
20/9
21/9
22/9
23/9
24/9
25/9
26/9
27/9
28/9
29/9
30/9
1/10
2/10
3/10
4/10
5/10
6/10
7/10
8/10
9/10
10/10
11/10
12/10
13/10
14/10
15/10
16/10
17/10
18/10
19/10
20/10
21/10
22/10
23/10
10/9
11/9
12/9
13/9
14/9
15/9
16/9
17/9
18/9
19/9
20/9
21/9
22/9
23/9
24/9
25/9
26/9
27/9
28/9
29/9
30/9
1/10
2/10
3/10
4/10
5/10
6/10
7/10
8/10
9/10
0.002 0.002
0.000 0.000
-0.002
Vertical displacement (m)
- 0.002
-0.004
Vertical displacement (m)
- 0.004
-0.006 - 0.006
-0.008 - 0.008
-0.010 - 0.010
-0.012 - 0.012
-0.014 - 0.014
-0.016
-0.016
-0.018
- 0.018
-0.020
- 0.020
-0.022
- 0.022
Measured FLAC FLAC No Injections Measured FLAC FLAC No Injections
Figure 16. Results obtained for extensometer E4. Figure 18. Results obtained for extensometer E10.
0.0
maximum values of 19 mm in the hypothesis without
Vsubsidence/Vexcavated (%)
-0.3
-0.5
NoNoinjections
injections
Injections
Injections
treatment and 15 mm in the one with compensation
-0.8
-1.0
grouting. In the same figure the corresponding hori-
-1.3 zontal strains have been represented too. These
-1.5
-1.8 horizontal strains have been calculated as:
-2.0
-2.3
-2.5
1668 1673 1678 1683 1688 1693 1698 1703 1708 1713 1718 1723 1728 1733 1738 1743 1748
Chainage (4)
Quaternary
Quaternary
Natural
Natural where L is the initial distance between two points
A.
A.Granite
GraniteVV Fills
Fills
along the gallery, and L is the distance once the dis-
placement has occurred. The strains show low values
A.A.Granite
GraniteIIVV
in both cases, although the induced tensile strains are
higher in the hypothesis with injections.
A.A.Granite
anit II–III
Granite III
4 CONCLUSIONS
Figure 19. Volume loss analysis.
The main conclusions that can be obtained from this
analysis are:
0 0.050
-6 -0.025
lower than the actual volume loss (1–2%).
Strain (mm/m)
-8 -0.050
-0.075
– The expected volume to inject (13.7 m3) was much
-10
lower than the actual volume injected (68 m3). Still
-12 -0.100
only partial compensation was achieved.
-14 -0.125
– According to the comparison between the model
-0.150
-16
with compensation grouting and the model with-
-18 -0.175
out the treatment area has been insufficient.
-20
-50 -45 -40 -35 -30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30 35 40 45 50
-0.200
– Numerical models should play an important role
Distance to the cross measured on the ATLL (m) in the design of compensation grouting providing
Injections No Injections Strain I. Strain N.I.
accurate estimates of the ground loss and of the
required treatment area.
Figure 20. Horizontal strains in the ATLL gallery bottom.
A.K. Chugh
Bureau of Reclamation, Denver, CO, USA
T.D. Stark
University of Illinois, Urbana, IL, USA
ABSTRACT: An automated procedure is presented to generate a 3-dimensional mesh for numerical analysis
of engineering problems. The procedure is simple, effective and efficient, and can be applied to represent complex
geometries and material distributions. A listing of the program that was used for the sample problem of a landfill
slide is included.
One of the essential tasks in a 3-dimensional (3-D) The conceptual model for the generation of a 3-D
numerical analysis is to represent the geometry and dis- mesh follows the conventional procedure of portraying
tribution of materials in the numerical model. FLAC3D spatial variations of materials in 3-D via a series
provides means to facilitate mesh generation and the of 2-dimensional (2-D) cross-sections. This technique
built-in programming language FISH can be used to is commonly used by engineers and geologists in
develop and implement additional program instruc- constructing visual models of complex geologic sites
tions during execution of a data file. where a number of 2-D cross-sections are used to repre-
In geotechnical engineering, surface geometry, sent the field conditions. In these representations, linear
distribution of materials, and water table conditions variations between material horizons in consecutive 2-
usually vary from one location to the next and pose a D cross-sections are used to depict the 3-D spatial vari-
difficult set of conditions to represent in a numerical ability of a site. The accuracy of the representation is
model. In order to facilitate the analysis of landslides, improved by using closely spaced 2-D cross-sections.
a simple procedure was devised to represent complex The 3-D mesh generation procedure presented
surface geometry, subsurface material horizons, and herein follows the conventional practices used by engi-
water table conditions. The objectives of this paper are neers in constructing 2-D numerical meshes by hand
to present: for geotechnical problems to be solved using methods
other than FLAC3D. For example, in the creation of a
1 a simple method to describe field geometry and
2-D numerical model of a slope to be analyzed using
conditions for a 3-D numerical model of a slope
a limit-equilibrium based procedure, it is a common
problem;
practice to define profile lines via a set of data points
2 a simple procedure for automatic generation of a
followed by specifications of their connectivities.
3-D mesh; and
Also, in the creation of a 2-D model of a continuum to
3 an illustration of the use of the procedure for analy-
be solved by a finite-element based procedure, it is a
sis of a large slide in a landfill.
common practice to discretize the continuum into
A listing of the program for the landfill slide is included a network of zones; assign identification numbers
in the paper. This program listing is in the FISH lan- to the grid points; define the coordinates of the
guage and uses some of the functions available in the grid points; and then specify the connectivity of
FISH library. grid points.
Thus, in the conceptual model for the generation of each profile line and the water table (for parallel
a 3-D mesh in FLAC3D, use is made of defining a 2-D cross-sections, y-coordinate shall have same
series of 2-D cross-sections at representative locations constant value between two consecutive cross-
of a site; defining each of the 2-D sections as an sections).
assemblage of data points with line-segment connec- 2 The following steps are used for creating similar
tions; and organizing the data for an efficient and sets of data at each of the 2-D cross-sections:
effective discretization of the volume. a From the data in step 1(c) above, select control
points that are of significance in defining the
profile lines in all of the 2-D cross-sections.
3 WATER TABLE Tabulate the x-coordinates of these control points
in increasing order. For reference purposes, this
The water table surface is specified using the water table is referred to as Table 100.
table data of individual 2-D cross-sections and through b Use of the “Interpolate” function expands the 2-D
the use of 3-point planar polygons between consecutive cross-sectional data of step 1(c) by linear inter-
2-D cross-sections. This scheme allows incorporation polation for all of the control points listed in
of non-coplanar variations in the water table surface Table 100 for all of the profile lines and stores
in the entire 3-D model. the data in separate tables; assigns Table num-
bers in increasing order starting with the user
4 DESCRIPTION OF THE PROCEDURE specified starting number and incrementing it by
1; assigns an identification number to each
In geotechnical engineering, the ground-surface point; and positions the points in the 3-D model
geo-metry is obtained using contour maps that are space. These tables contain the (x,z) coordinates
prepared from land or aerial survey of the area. The of expanded 2-D cross-sectional data. A sample
subsurface material horizons are estimated from geo- listing of the “Interpolate” function and its
logic data and information obtained from exploratory dependency function “zz” in FISH language is
boring logs. The subsurface water conditions are esti- given in Figure 1. The starting table number used
mated from field observations, piezometers installed in the sample problem data file is 200.
at various depths, and/or from water levels in borings. 3 The following steps are used for creating zones in
Subsurface data are used to develop contour maps of the 3-D model space:
the subsurface geology and water conditions.
From these contour maps, the region-of-interest,
and the locations of significant cross-sections are def zz
identified; information for 2-D cross-sections are zz=table(t_n,xx)
end
read and tabulated; and 2-D cross-sections are drawn
for an understanding of the site details and preparation def interpolate
of input data for a 2-D analysis. In general, the cross- loop j (js,je); profile line #s -
sectional data for a site varies from one location to ; js is for the bottom, je is for top
the next. These variations may be caused by changes dt_n=dt_n_s+j; dt_n is destination table number
in the ground surface and (or) in subsurface material loop i (is,ie); is is the first interpolation #,
; ie is the last interpolation #
horizons, discontinuity of some materials, or a com- xx=xtable (100,i); x-coordinate of the
bination of these or some other variations. ;interpolation point
In the proposed procedure, the following steps are command
followed: (For ease of presentation, 2-D cross-sections set t_n=j
are assumed to lie in x-z plane and the x,y,z coordi- end_command
nate system follow the right hand rule.) table(dt_n,xx)=zz
id_pt=id_pt+1
1 The following steps are used for creating an x_pt=xtable(dt_n,i)
orderly assemblage of field data for 3-D discretiza- y_pt=y_pt
tion of the continuum of the region-of-interest: z_pt=ytable(dt_n,i)
command
a On the site map, select values of x, y, and z coor- generate point id id_pt x_pt y_pt z_pt
dinates that completely circumscribe the 3-D end_command
region-of-interest; endloop
b Mark locations of all significant 2-D cross- endloop
sections oriented in the same and preferably par- end
allel direction;
c For each 2-D cross-section, tabulate (x,y,z) coor- Figure 1. Listing of the “Interpolate” function and its
dinates of end-points of all line segments for dependency function “zz” in FISH language.
10
a Tabulate the y-coordinates of the 2-D cross- zones desired for each interval in the x-direction.
sections in increasing y-direction. For reference Tabulate these values for all of the intervals in
purposes, this table is referred to as Table 101. the increasing x-direction. For reference purposes,
The number of entries in Table 101 should equal this table is referred to as Table 102. The number
the number of 2-D cross-sections marked in of entries in Table 102 should be one less than
step 1(b). those in Table 100.
b Considering the spacing of x-coordinates of the c Considering the spacing between the 2-D cross-
control points in step 2(a), select the number of sections in the y-direction, select the number of
zones desired for each interval in the y-direction.
Tabulate these values for all of the intervals in the
def fill_grid increasing y-direction. For reference purposes,
i_n=table_size(102) this table is referred to as Table 103. The number
j_n=table_size(103)
of entries in Table 103 should be one less than
k_n=table_size(104)
loop jy (1,j_n) the number of 2-D cross-sections.
ny=xtable(103,jy) d Considering the spacing of the profile lines in
p0_d=(jy-1)*(i_n+1)*(k_n+1) the z-direction, select the number of zones desired
loop kz (1,k_n) for each material horizon in the z-direction.
nz=xtable(104,kz) Tabulate these values for all of the intervals in
if kz=1 then the increasing z-direction. For reference purposes,
material='shale' this table is referred to as Table 104. The number
endif
if kz=2 then of entries in Table 104 should be one less than the
material='ns'; native soil number of profile lines.
endif e Use of the “Fill_grid” function generates a
if kz=3 then brick mesh and assigns a group name to each 3-
material='msw'; municipal solid waste D volume zone. A sample listing of the
x_toe=xtable(105,jy) “Fill_grid” function in FISH language is given
endif in Figure 2.
loop ix (1,i_n)
if kz=3 then
xx_toe=xtable(100,ix)
if xx_toe < x_toe then 5 COMMENTS
material='mswt'
endif 1 Use of a Brick mesh with an 8-point description is
endif versatile and allows for creation of degenerated brick
nx=xtable(102,ix) forms through the use of multiple points with dif-
p0_d=p0_d+1 ferent identification numbers occupying the same
p3_d=(p0_d+i_n+1)
p6_d=(p3_d+1) (x,y,z) coordinate location in the 3-D model space.
p1_d=(p0_d+1) 2 During the development of the grid, it is possible to
p2_d=((i_n+1)*(k_n+1)+p0_d) assign group names to different segments of the
p5_d=(p2_d+(i_n+1)) model. This information can be useful in modifying
p7_d=(p5_d+1) the generated grid.
p4_d=(p2_d+1) 3 Expanding the (x,y,z) location data for all 2-D cross-
command sections to a common control number of locations
generate zone brick size nx,ny,nz ratio 1,1,1 &
p0=point (p0_d) p3=point (p3_d) &
via interpolations facilitates the programming of
p6=point (p6_d) p1=point (p1_d) & the automatic grid-generation procedure.
p2=point (p2_d) p5=point (p5_d) & 4 In engineering practice, it is generally desirable
p7=point (p7_d) p4= point(p4_d) group material to analyze a few 2-D cross-sections at select loca-
end_command tions prior to conducting a 3-D analysis. Because
if kz=3 then development of data for 2-D cross-sections is
material='msw' one of the steps for use of the proposed procedure,
endif
end_loop
it is relatively easy to conduct a 2-D analysis using
p0_d=p0_d+1 the 2-D cross-sectional data and the program
end_loop FLAC.
end_loop 5 The program instructions listed in Figures 1 and 2
end can be modified to accommodate geometry and
other problem details that are different or more
Figure 2. Listing of “FILL_GRID” function in FISH complex than those encountered in the sample
language. problem described in Section 6.
11
6 SAMPLE PROBLEM
12
“Fill_grid”. For each interval in the location of cross- x-direction (Table 102), the values of number of
sections in the y-direction (Table 103), and for each zones desired in the x, y, and z-direction and the id #s
material horizon between the profile lines in the z- of points in the 3-D model space are used in the
direction (Table 104), and for each interval in the “GENERATE zone brick p0, p1, … p8” command of
13
; station at y=15.24 m
set y_pt=15.24
table 2 erase
table 3 erase
table 4 erase
set dt_n_s=dt_n
table 2 0,223.60 163.07,223.60 306.02,238.84
table 2 348.08,240.67
table 3 0,228.60 163.07,228.60 306.02,243.84
table 3 348.08,245.67
Figure 6. Partial listing of the data file for the sample problem for FLAC3D.
14
7 ADVANTAGES OF THE PROPOSED Eid, H.T., Stark, T.D., Evans, W.D. & Sherry, P.E. 2000.
PROCEDURE Municipal solid waste slope failure. II Stability analyses.
Journal of Geotechnical and Geoenvironmental Engi-
neering 126(5): 408–419.
1 The proposed procedure for describing 3-D field Stark, T.D. & Eid, H.T. 1998. Performance of three-
conditions utilizes 2-D cross-sections, which are dimensional slope stability methods in practice. Journal
essentially the same as commonly used by geolo- of Geotechnical and Geoenvironmental Engineering
gists and engineers to describe the field conditions. 124(11): 1049–1060.
15
ABSTRACT: A new constitutive model is proposed based on the Hoek-Brown failure criterion. This model
incorporates a plasticity flow rule that varies as a function of the confining stress level. For a low confining
stress, at which a large rate of volumetric expansion at yield is anticipated, an associated flow rule is applied.
For high confining stress, at which the material no longer dilates at failure, a constant-volume flow rule is pre-
scribed. A composite flow rule, which provides a linear variation from associated to constant-volume limits, is
used between the low and high confining stress states. Using an appropriate softening relation, the model can
also represent the transition between brittle and ductile rock behavior. The new model is programmed in
C
and compiled as a DLL file (dynamic link library) that can be loaded directly into either FLAC or
FLAC3D. This paper describes the model and its implementation as a DLL. Physical justification is provided for
the formulation and, specifically, the representation of the volumetric behavior during yield, which depends on
confining stress. A verification example is provided.
17
(2)
where E1 K
4G/3 and E2 K 2G/3 and (e1,
e2, e3) is the set of principal strain increments. If
the yield criterion (equation 1) is violated by this set
of stresses, then the strain increments (prescribed as
independent inputs to the model) are assumed to be
composed of elastic and plastic parts, i.e.,
(3)
(4)
Eliminating the current stresses, using equations
(2) and (4),
(5)
(6)
It should be noted that the failure criterion (equa- At yield, equation (1) is satisfied by the final
tion 1) does not depend on the intermediate principal stresses; that is,
stress, 2; thus, the failure envelope is not isotropic.
Assume that the current principal stresses are (1,
2, 3) and that initial trial stresses (1t , 2t , 3t ) are (8)
calculated by using incremental elasticity, i.e.,
18
By substituting values of 1f and f2 from equation 3.2 Constant-volume flow rule
(7), equation (8) can be solved iteratively for ep3, As the confining stress is increased, a point is reached
which is then substituted in equation (7) to give the final at which the material no longer dilates during yield.
stresses. The method of solution is described later, but A constant-volume flow rule is therefore appropriate
first the evaluation of is discussed. when the confining stress is above some user-
prescribed level, 3 cv3. This flow rule is given by
We need to consider an appropriate flow rule, which The constant-volume flow rule defined by equation
describes the volumetric behavior of the material dur- (11) is represented graphically by point C in Figure 1a.
ing yield. In general, the flow parameter will depend The normal to the vector ep at point C has a slope equal
on stress, and possibly history. It is not meaningful to to unity, and therefore the rate of volumetric expansion
speak of a dilation angle for a material when its confin- in the plastic regime is null.
ing stress is low or tensile, because the mode of failure
is typically by axial splitting, not shearing. Although
the volumetric strain depends in a complicated way on 3.3 Radial flow rule
stress level, we consider certain specific cases for which Under the condition of uniaxial tension, we might
behavior is well known, and determine the behavior for expect that the material would yield in the direction
intermediate conditions by interpolation. of the tensile traction. If the tension is isotropi-
Three cases are considered below. cally applied, we imagine (since the test is practi-
cally impossible to perform) that the material would
deform isotropically. Both of these conditions are
3.1 Associated flow rule
fulfilled by the radial flow rule, which is assumed
It is known that many rocks under unconfined com- to apply when all principal stresses are tensile. For a
pression exhibit large rates of volumetric expansion flow-rate vector to be coaxial with the principal stress
at yield, associated with axial splitting and wedging vector, we obtain
effects. The associated flow rule provides the largest
volumetric strain rate that may be justified theoretically. (12)
This flow rule is expected to apply in the vicinity of
the uniaxial stress condition (3 ≈ 0). An associated
flow rule is one in which the vector of plastic strain The radial flow rule defined by equation (12) is
rate is normal to the yield surface (when both are represented graphically by points D1, D2 and D3 in
plotted on similar axes). Thus, Figure 1b. The directions of vectors ep at these points
intercept all the origin of the diagram.
(9)
3.4 Composite flow rule
We propose to assign the flow rule (and thus, a value
where the subscripts denote the components in the for ) according to the stress condition. In the fully ten-
principal stress directions, and F is defined by equa- sile region, the radial flow rule, rf, will be used. For
tion (8). Differentiating this expression, and using compressive 1 and tensile or zero 3 the associated
equation (6), flow rule, af, is applied. For the interval 0 3 cv3,
the value of is linearly interpolated between the asso-
ciated and constant-volume limits, i.e.,
(10)
(13)
19
The composite flow rule defined by equation (13) high confining stresses, the iteration converges in one
in the case of compressive stresses is represented step, but at low confining stresses, up to ten steps are
graphically by point B in Figure 1a. The slope of the necessary (the limit built into the code is presently 15).
normal to ep at point C is the linear interpolation of the
slopes at points A and B.
5 MATERIAL SOFTENING
20
(17)
(18)
(19)
(20)
(21)
21
Figure 2c, depends on the flow rule assumed for the From equation (23), again considering 3 0, the
material as follows, critical value of plastic strain is,
ep3 cr (0) 0.013
In the FLAC model, the tables for the softening
(23) parameters should be defined as follows:
For interest, we list here the expressions for the In addition to the table above, a table defining the
slopes corresponding to the elastic, softening and resid- relationship between the multiplier and the confin-
ual branches in the e3 vs. e1 – e3 diagram of Figure 2c, ing stress 3 will be normally defined. The type of
relationship to consider depends on how the drop mod-
(25) ulus of the softening branch is assumed to vary with
the level of confinement.
To illustrate the definition of the multiplier we
(26) consider first the case in which the drop modulus of
the softening branch, 2G, is maintained for increasing
values of confinement 3 (see line OPR in Figure 2b).
For this case, the multiplier is defined as follows,
(27)
(28)
We consider now a practical case of definition of
softening parameters in a FLAC model.
Let us assume the following values for the param-
eters that control the response of the material in Assuming an upper limit for the confining stress
Figure 2: equal to 10 ci, and taking 5 points to represent this
relationship, the definition of the multiplier in
ci 0.1 MPa FLAC will be as follows:
mb 5
s 1
a 0.5 3 [MPa]
0.5
0.00 1.0000
0.2 (for 3 0) 0.25 3.6742
E 100 ci 0.50 5.0990
0.3 0.75 6.2048
0o 1.00 7.1414
[Note that the condition 0o implies that the
material does not dilate in the plastic regime; in the Note that in the table above, the second column is
FLAC model this condition is satisfied by specifying computed using equation (28).
cv3 0.] As a second example of the definition of the mul-
For the value of defined above, the residual tiplier we consider now the case for which the drop
parameters Rci and mRb are computed with equations modulus of the softening branch, 2G, decreases (in
(19) and (20) and result to be, absolute value) for increasing values of confinement
3 (see line OPR in Figure 2b). To achieve the ductile
Rci 0.05 MPa behavior ( 0) at the confinement level 3 dc 3,
mRb 2.5 we can use the following relationship between the
From equation (23), and considering 3 0 the critical multiplier and the confining stress 3,
value of plastic shear-strain for which the residual
stage is achieved is, (29)
cr
p (0) 0.039
22
(30)
3 [MPa]
0.000 1.0
0.033 1.5
0.067 3.0
0.100
1.000
6 VERIFICATION
23
necessary to compare the response of the new model A.1 Plastic region, r Rpl
with actual measurements of rock behavior, both in the The critical internal pressure below which the failure
laboratory and in the field, and calibrate the parameter zone develops is computed from the following tran-
cv
3 from observations of volumetric strain. Further, the scendental equation,
softening behavior is assumed to depend on the confin-
ing stress, not on deviatoric stress, which is the more
usual assumption. This decision was made on a gen- (A.1)
eral knowledge of rock behavior, but it will need to be
verified (or falsified) by comparing model predictions
with measurements of rock response under post-peak
conditions. The extent Rpl of the failure zone is,
REFERENCES
24
(A.14)
(A.7)
(A.15)
(A.16)
(A.8) (A.17)
(A.10) (A.18)
(A.11)
The solution for the hoop stress field is
The boundary conditions to integrate the differential
equation (A.5) above are
(A.19)
(A.12)
The solution for the radial displacement field is
and
(A.20)
(A.13) The plastic and elastic solutions for the field quanti-
ties r and ur presented above are continuous at the
elasto-plastic boundary (i.e., at r Rpl). The solution
for the field quantity is discontinuous when there is
In the case of associated flow rule, the solution for a jump of strength from peak values (ci, mb, s and a) to
the radial stress field is obtained from integration of residual values (Rci, mbR, sR and aR) – see Figure 3b.
25
J. Parra
PDVSA – Intevep, S.A., Los Teques, Venezuela
ABSTRACT: The occurrence of thin localized bands associated with concentration of compressive strain has
recently been reported in very porous rocks, both in field and laboratory settings. These structures exhibit a
reduction of porosity, and are of importance to the petroleum industry because they can impact reservoir perme-
ability. Compaction bands have been the object of both theoretical and experimental studies by Olsson (1999),
Issen & Rudnicki (2000), Bésuelle (2001) and others. In this paper, we examine, in a simple theoretical framework,
the basic conditions for a band to appear. We consider the case of the Double-Yield model, identify conditions for
localization related to the volumetric cap, and give examples of numerical simulations that illustrate band formation.
27
where pc is the cap pressure, and F v
0 for elastic The coefficient a is the hardening modulus (positive
conditions. for softening) which is a function of total plastic volu-
The cap F v 0 is represented by a straight line in metric strain. An example of volumetric hardening
the plane of shear stress, q, versus mean pressure, p, behavior is represented in Figure 2.
shown in Figure 1. The flow rule gives the direction of plastic strain rate,
The flow rule for volumetric yielding is associated; which is parallel to the gradient of Gv in stress space:
thus, the potential function is:
(7)
(2) .
The plastic multiplier
, gives the magnitude of plas-
tic strain rate. It may be found from the consistency
The total strain rates are partitioned into elastic and condition:
plastic parts:
(8)
(3)
Substitution of the expression 2 for plastic potential
The stress-strain relations are, in rate form: in Equation 7 gives, after differentiation:
(9)
(4)
Using Equation 9 for the plastic strain rate, the
hardening rule in Equation 6 takes the form:
(10)
Figure 1. Volumetric cap for the Double-Yield model. Figure 2. Example of volumetric hardening rule.
28
After substitution of this expression in the consis- met within the material sample. First, equilibrium at
tency condition, we obtain: the interface of the band only allows a stress disconti-
nuity for the direct stress parallel to the band. Second,
the direct strain parallel to the band must remain con-
(11)
tinuous.
We analyze the situation at the onset of band forma-
The plastic strain increments in Equation 9 may tion, and denote as ni the unit normal to the potential
now be expressed as: planar band in which localized deformation occurs.
Mathematically, the condition for non-uniqueness
translates as (see e.g. Issen & Rudnicki 2000):
(12)
(17)
From Equation 3, elastic strain rate may be where the components of the stiffness matrix may be
expressed as total rate minus plastic rate: found in Equation 14.
We look at the case when the out of plane component
(13) of the normal to the band is zero, or n3 0. Band for-
mation is predicted to occur when the condition
Finally, after substitution of Equation 13 in
Equation 4, using Equation 12 and some manipulations,
the cap constitutive behavior may be written in the form:
(18)
(19)
(15)
(16)
29
Figure 4. Example of cap softening behavior. Figure 5. Plateau corresponding to pore collapse in a
schematic porous material stress–strain curve.
3 NUMERICAL EXPERIMENTS
This condition can only be satisfied if a
0, that
is when softening of the cap occurs, see Figure 4. Our theoretical derivation shows that cap softening
Physically, cap softening can correspond to grain col- is a necessary but not a sufficient condition for the
lapse or breakage of cemented grains. formation of compaction bands. A relation between
By definition, a pure compaction band is oriented hardening modulus (a) and stiffness properties (Kc,
perpendicular to the maximum compressive stress. Gc) must also be satisfied for the bands to appear. In an
According to our convention, we must have: n1 0, oedometric test, two sets of bands (one horizontal, and
in which case Equation 19 implies: one vertical) are predicted to occur for
(24)
(26)
Finally, using Equation 5 for 1, and Equation 15
for b, the localization condition may be expressed as:
Numerical experiments are carried out to validate
(25) this prediction, and illustrate band formation.
Numerical simulations of an oedometric test are per-
formed using the finite difference code FLAC. The con-
From the above consideration it follows that soften- figuration is axi-symmetric. The grid contains a total of
ing of the cap is a necessary but not sufficient condition 400 elements. The boundary conditions correspond to
30
31
4 CONCLUSIONS
ACKNOWLEDGEMENTS
32
33
F. Laigle
Electricité de France, Hydro Engineering Centre, France
ABSTRACT: A new viscoplastic constitutive model has been developed by EDF-CIH. Its aim is able to take
into account delayed behavior of rock materials in the framework of nuclear waste repository studies. In this
case, it’s important to predict the rock damage evolution in time in the neighboring storage tunnels. The main
assumptions of the constitutive model are presented in this paper. One application to the Mine-by-Test done by
the AECL in the Lac de Bonnet granite is shown. The low field strength of the granite in comparison with
laboratory measurements may be justified by the delayed behavior of this granite. Failure with v-shape notches
is well shown by the simulation. A prediction of the hydraulic permeability increasing around the tunnel versus
time is presented.
1 INTRODUCTION The Mine-by test has been done at the level – 420, in
an undamaged granite mass (Lac de Bonnet Granite).
In the framework of studies of underground nuclear This experiment consists of digging a gallery in well-
waste storage, it’s important to predict the evolution known conditions in a previously monitored part of
in time and at very long term of the ground surround- the rock mass. The major aim is to observe the behav-
ing the excavations. One objective of these studies ior of the granite during the excavation phase and at
is to assess the evolution of the EDZ (Excavation long-term.
Damage Zone) in the time. This EDZ is assumed to The direction of the gallery has been defined in
be the zone where rock is fractured and where the accordance with initial stresses in the ground. At this
permeability increasing is large. depth, the major compressive principal stress is more
In this aim, a viscoplastic constitutive model has or less horizontal, and is about 55 MPa. This stress
been developed by EDF and integrated in FLAC. This is 3.9 times the vertical stress corresponding at the
model will be briefly presented in the paper. weight of overburden. The intermediate principal
The application is about the Mine-by-Test done in stress is about 48 MPa.
the AECL-URL. The evolution in time of the failure
zone in the roof and invert of the gallery is well sim-
ulated. Assumptions of this model allow to assess the
damage zone (fissured rock) and the fractured zone
(continuous fissure) associated with a strong perme-
ability increasing.
35
Figure 2. Location of v-shaped notches in the Mine-by test Figure 3. Stress–strain curve for Lac de Bonnet granite.
tunnel.
36
2.3 Definition of the in situ strength The elastoplastic strain component ensues from the
In the case of an undamaged rock like the Lac de following non-associated rule function:
Bonnet Granite, the apparent in situ strength under a
specific confinement state is function of the loading
rate. As this loading rate is much lower than these (3)
applied in laboratory conditions, it seems reasonable
to assume that the in situ strength could be assimi-
lated to the long term strength of the rock. From lab- The hardening of the elastoplastic mechanism is only
oratory tests, this “long term strength” could have negative. Some specific hardening laws were defined
several definitions: for each of the internal parameters m, s and a, allow-
– From Sangha et al. (1972), the long term strength ing to describe the evolution of the rock sample
corresponds to the volumetric strain reversal. strength from maximum peak value to the residual
Above this threshold, the crack growth is assumed state. The softening behavior domain reached after
to be “unstable”. This notion of “instability” seems the peak strength, is assumed to be divided in three
totally subjective because is related to the delay phases:
allowed by experimental testing in laboratory 1. The first phase of softening corresponds to a dete-
conditions. rioration of the rock’s cementation illustrated by
– From Morlier (1966) or Wiid (1970), the long term a progressive disappearance of the cohesion at the
strength of rocks is assumed to be correlated to the macroscopic scale. This first phase is associated
beginning of the crack initiation. Assuming that with an increasing of the dilatancy.
the time behavior of cohesive materials like rocks 2. The second phase corresponds to the shear of an
is associated with a crack growth, this definition induced fracture. It’ s associated with a decreasing
appears more physical. It’s in accordance with of the dilatancy at the macroscopic scale.
experimental results carried out by Schmidtke 3. Finally, the last domain corresponds to a purely
which show that the long term strength of the Lac frictional behavior, which defines the residual
de Bonnet granite could be estimated to 90 to strength. The shear occurs without any volumetric
100 MPa. This threshold corresponds to the crack strain.
initiation under unconfined conditions.
A viscoplastic version of the model has been devel-
oped after that. This version is based on the Perzyna’s
theory, which assumes that the viscoplastic strain
3 CONSTITUTIVE MODEL
rate is a function of the distance between the loading
point representative of the state of stresses and a yield
3.1 General principles
viscoplastic surface, in accordance with the following
A new constitutive model has been suggested by flow rule:
Laigle (2003) aiming at accurate simulation of the
rock behavior in the averaged and large strains
domain. A first version of the model has been initially
developed in the framework of the elastoplastic theory. (4)
In this case, the yield surface corresponds to a gener-
alized form of the Hoek-Brown criterion. Internal
parameters “m”, “a” and “s” change according to an O(F) is a flow function and F is the overstress func-
hardening variable p: tion. Their expressions are followings:
(5)
(1)
(2)
(6)
37
An important assumption is to assume that the hard- Figure 7 shows simulations of a triaxial test with
ening parameter rate p corresponds to the total irre- 10 MPa of confinement at several strain rates. In
versible strain as follows: accordance with the increasing of the strain rate, we
may observe both an increasing of the peak strength
(7) and a changing of the behavior. It appears that the
lower the rate, the more the rock behaves like a duc-
tile material.
and the global strain rate is the following:
1000
(8) 900
80
Deviatoric stress level (%)
30
– Evolution of the strength in accordance with the
20
loading rate (Martin & Read 1992). Comparison of
10
these experimental results and simulations is
0
shown in Figure 5. 1 10 100 1000 10000 100000 1000000 10000000 1E+08 1E+09
– Evolution of the creep time to failure versus the Time to failure (s)
deviatoric creep level (Schmidtke & Lajtai 1985).
The adjustment of the theoretical curve is shown in Figure 6. Time to failure vs. deviatoric stress level –
Figure 6. comparison simulation–experience.
38
200 .
ε. =1.5.10-6/s
ε =1.5.10-7/s
.
150 ε =1.5.10-8/s
.
ε =1.5.10-9/s
100
50
0
0 0,1 0,2 0,3 0,4 0,5
Axial strain(%)
39
300 Simulation
Deviatoric stress (MPa)
0
0,1 1 10 100 1000 10000
Relative permeability k/k0
Fissured rock
2 months
Fractured rock
Fissured rock
Tension zone
2 years
40
expansion of the fracturing in the vicinity of the the fracturing process over several years. From these
tunnel in time. This expansion is located in highly theoretical results, it’s only after several hundred
stressed zone. Table 2 presents the depth of the frac- years that this fracturing will stop.
tured zone at the crown versus time. From available information, it seems that the
In accordance with rheological parameters esti- expansion rate of the fractured zone is too slow and
mated before, the simulation shows an evolution of that this evolution has been more rapid (apparently,
few months). However, we have to precise that only
one simulation has been done using set of parameter
Fractured rock coming from an adjustment with few laboratory tests.
It could be now possible to do a back analysis and
to adjust some parameters like “n” and “A” which
govern the kinematic viscoplastic strains creation.
Despite this, after several years, when the stabilization
is reached, the shape of the fractured zone is similar to
these which has been observed on site (Fig. 14).
Tension zone Figure 15 shows the evolution of the parameter “s”
versus time. This variable represents the cohesion of
Fissured rock
4 years
Fractured rock
Tension zone
Fissured rock
10 years
Figure 14. Theoretical damage state of rock at long term.
Figure 13. Theoretical damage state of rock after 10 years.
1
0,9
Parameters ⇔ Damage indicator
0,8
Table 2. Depth of the fracture zone versus time.
0,7
41
the rock. As long as it isn’t null, the rock is intact or 4.3 Evolution of the permeability in time
only fissured. So, this value is a damage indicator.
Damage of the rock results in a local increasing of the
According to rheological parameters retained, Figure 15
rock macroscopic hydraulic permeability. This increas-
show that the damage continues to involve during a
ing is in accordance with the suggested Equation (9).
long time. This is the case at a distance of 71 cm
Considering the parameter , which has been adjusted
above the crown. Even if the kinematic is not correct
on triaxial test results, this increasing could be about
in our simulations, the results are similar to monitor-
several order of magnitude in the fractured zone.
ing results. Several years after the end of excavation,
Figures 16–19 show increasing of the permeability
some acoustic emissions have still been registered
around the mine-by test tunnel.
while fracturing process is stopped.
20 days
2 ans
Log(k/k0) Log(k/k0)
Figure 16. Increasing of the hydraulic permeability after Figure 18. Increasing of the hydraulic permeability after
20 days. 2 years.
Log(k/k0) Log(k/k0)
Figure 17. Increasing of the hydraulic permeability after Figure 19. Increasing of the hydraulic permeability after
2 months. 1000 years.
42
REFERENCES
43
ABSTRACT: Since the past 30 years, the research for estimating an accurate prediction of deformations
induced by tunneling has been a major engineering challenge all around the world. The in situ measurements
have shown that deformations of the soil on the vicinity of a tunnel show a strong evolution with time. Three
essential phenomena, actually, can be related to this evolution: the evolution with time of the distance to the
working face, the distance of the lining to the working face and the viscous effects occurring in the soil.
The objective of this paper is to propose a procedure for predicting the deformations induced by tunneling,
by taking into account these three essential phenomena, particularly the third phenomenon. Therefore, in this
study, a constitutive model for the time-dependent behavior of cohesive soil has been developed within the
framework of elastoplasticity–viscoplasticity and critical state soil mechanics. The consideration of viscoplas-
tic characteristic sets the current model apart from the CJS model, and introduces an additional viscous mech-
anism. The evolution of the viscous yield surface is governed by a particular hardening called “viscous
hardening” with a bounding surface.
To describe this procedure and the capability of the model, a comparison between numerical calculations and
monitoring the progressive closure of tunnel conducted in the TGV tunnel of Tartaiguille, is performed. The
finite difference software, Fast Lagrangian Analysis of Continua (FLAC), has been used for the numerical sim-
ulation of the problems. The comparison results show that the observed deformations could have been reason-
ably predicted by using the proposed excavation model.
45
(5)
(6)
Figure 1. Plastic deviatoric mechanism and plastic
isotropic mechanism in CJS.
(4) (10)
46
c
where SII represents the second invariant of the devia- SII
toric part of stress in the characteristic state and is a
parameter of the model. The characteristic surface is
defined by:
(12)
47
48
3 CALCULATIONS
49
50
51
4 CONCLUSIONS
52
by Mr. Alain Robert and Mr. Adrien Saïtta from Lunardi, P. 2000. The design and construction of tunnels
CETU (Centre d’Etudes des Tunnel), Lyon, France. using the approach based on the analysis of controlled
deformation in rocks and soils, ADECO-RS.
Maleki, M. 1998. Modélisation hiérarchisée du com-
portement des sols, Phd. Thesis, École Centrale de Lyon.
REFERENCES Panet, M. & Guenot, A. 1982. Analysis of convergence
behind the face of a tunnel, Tunneling’ 82: 197–204.
Adachi, T. & Oka, F. 1982. Constitutive equations for nor- Peila, D. 1994. A theoretical study of reinforcement influ-
mally consolidated clays based on elasto-viscoplasticity, ence on the stability of a tunnel face, Geotechnical and
Soils and foundations, Vol. 22, No. 4: 57–70. Geological Engineering, 12.
AFTES. 2002. La méthode convergence-confinement, Perzyna, P. 1966. Fundamental Problems in viscoplasticity.
Tunnels et ouvrages souterrains, No 170: 79–89. Advances in Applied Mechanics, Vol. 9: 243–377.
Cambou, B. & Jafari, K. 1987. A constitutive model for gran- Roscoe, K. H. & Burland, J. B. 1968. On the Generalised
ular materials based on two plasticity mechanisms. Stress-Strain Behavior of ‘Wet Clay’, Engineering
Constitutive equations for granular non-cohesive soils, Plasticity, J. Heyman and F. A. Leckie (Eds). Cambridge:
Saada & Bianchini (Eds), Balkema, Rotterdam: 149–167. Cambridge University Press: 535–609.
Kaliakin, N. & Dafalias, F. 1990. Theoretical aspects of the Sekiguchi, H. 1984. Theory of undrained creep rupture
elastoplastic-viscoplastic Bounding surface model for of normally consolidated clays based on elasto-
cohesive soils, Soils and foundations, Vol. 30, No. 3: 11–24. viscoplasticity, Soils and foundations, Vol. 24, No. 1:
Katona, M. G. 1985. Evaluation of viscoplastic cap model, 129–147.
Journal of Geotechnical Engineering, Vol. 110, No. 8: Serratrice, J.F. 1999. Tunnel de Tartaiguille (Drôme) TGV
1107–1125. Méditerranée, Essais de laboratoire sur la marne, LRPC
Lade, P. V. 1998. Experimental Study of Drained Creep d’Aix en Provence.
Behavior of Sand, Journal of Engineering Mechanics,
Vol. 124, No. 8, August: 912–920.
53
J.M. Rodríguez-Ortiz
Gamma Geotécnica SL, Madrid, Spain
ABSTRACT: Anhydrite and rocks containing argillaceous minerals experience swelling phenomena when
they come into contact with water. In tunneling, this can lead to a strong heave of the floor and to a high level
of stresses in the lining. Although characterization of swelling potential, monitoring of swelling process, and a lot
of relevant case histories of tunnel construction in swelling rocks are currently available, the design of the support
in swelling rocks usually do not consider an accurate stress–strain relationship for the swelling. Current trend
in tunneling design considers numerical modeling of the rock-support interaction, but the available geotechni-
cal codes do not include the swelling formulation. This paper presents the implementation in FLAC (via FISH
routines) of the analytical stress–strain formulation for the swelling presented by Wittke (1999) and the valida-
tion of this algorithm against the swelling tests carried out by different authors and presented by Wittke (1999).
1 DESCRIPTION OF THE ANHYDRITE created at the surface of the grains inhibits the water
SWELLING PHENOMENA penetration stopping the process. In the case of inter-
bedded anhydrite-mudstone the swelling process
1.1 Chemical description leads into the disintegration of the rock, reducing its
strength. Steiner (1993) quantifies this reduction of
Calcium sulphate appears naturally as two different
strength with an angle of friction of 20°.
minerals: gypsum (CaSO4.H2O), in which water
appears within the crystalline structure, and anhydrite 1.2 Characterization of the swelling behavior
(CaSO4). The hydration of anhydrite is a complex
process that depends on the pressure and the temper- The International Society of Rock Mechanics has pro-
ature. For ambient conditions the chemical reaction is posed a set of tests to quantify the swelling of argilla-
illustrated in Table 1. ceous rocks: the Maximum Axial Swelling Stress test,
With an external inflow of water, the volumetric the Axial and Radial Free Swelling Strain test, and the
increment associated to this process is presented in Axial Swelling Law test (axial swelling stress as a
Equation 1: function of axial swelling strain, or Huder-Amberg
swelling test).
(1) An illustration of the results from the Huder-
Amberg swelling test (total vertical strain of the sample,
z in %, versus vertical load, z in kPa) is presented in
Transformation of anhydrite into gypsum can be Figure 1, taken from Wittke (1999). Stages 1, 2 and 3
inhibited at 20°C with a pressure of 1.6 MPa; this pro correspond to the initial loading phase with 2 load
cess is reversible, being necessary a pressure of 80 MPa cycles; stage 4 corresponds to the watering of the
at 58°C to transform gypsum into anhydrite. sample (no stress increment but strain increment),
Previous data refer to pure anhydrite, but in case of and finally stage 5 corresponds to the different points
interbedded mudstone-anhydrite the maximum swelling of the unload-swelling process. The swelling strain
volume is lower but the swelling stress is larger (in the equals the total strain (stage 5 of the test) minus the
order of 2 to 5 MPa). elastic strain (stages 2 and 3 of the test).
In the case of pure anhydrite, as the hydration pro If the strain due to swelling is plotted against
cess begins, the thin layer of impervious gypsum the stress in a semi-logarithmic scale (Figure 2) a
55
Table 1. Transformation of anhydrite into gypsum. swelling process); Kq swelling deformation parame-
ter; z axial effective stress; and 0 axial swell-
Anhydrite water Gypsum
ing stress (as showed in Figure 2, it is intersection of
Equation CaSO4 H2O CaSO4.2H2O the straight line with qz 0).
Mass (gr) 136.14 36 172.14
Density (gr/cm3) 2.96 1 2.32
Volume (cm3) 46.2 36 74.3 2 ANHYDRITE SWELLING LAW
(3)
(4)
Figure 2. Axial swelling law (Grob 1972, in Wittke 1999).
56
57
associated to these strains are calculated with the fol- To determine a value of t small enough, only a
lowing lineal elastic relationship, Equation 6: fraction of the stress increment associated to the
remaining swelling strain should be “injected” in the
elements of the model.
A criterion of a maximum of 1% of the current
(6) stress state has been adopted to determine de frac-
tion of i to “inject”. The minimum fraction obtained
from the 3 principal directions in each element is
adopted. These relationships are illustrated in
where and G are constants known as Lamé’s Equation 11 for every element in the model.
parameters ( K
2/3G; K is the bulk modulus and
G is the shear modulus).
Previously to the calculation of the swelling time
parameter the plastic component of the volumetric (11)
strain has to be determined.
The total volumetric strain, addressed in FLAC with
a FISH variable, is the sum of the following compo-
nents, Equation 7:
where ri fraction of the stress increment i;
(7) i stress state; and r minimum fraction of the
3 principal directions.
From Equations 6 and 11 the fraction r of the stress
where totv total volumetric strain; v elastic vol-
el0
increment that are going to be “injected” in the ele-
umetric strain produced in the model previous to any ments of the model can be expressed with the follow-
calculation; elv elastic volumetric strain produced ing Equation 12:
during the calculation; it can be calculated as
vel (1 2 3)/(3 K), being K the bulk modulus;
qv swelling volumetric strain accumulated during (12)
calculation, qv q1 q2 q3; therefore, the plastic
volumetric strain, plv, can be calculated with the fol-
lowing Equation 8: and therefore, the timestep for each element can be
obtained as, Equation 13,
(8)
(14)
and therefore,
The “injection” of stresses associated to the swel
(10) ling behavior can be expressed with the following
Equation 15,
58
(16)
(17)
This procedure is repeated until the accumulated Figure 4. Swelling pressure test (in Wittke 1999).
swelling time reaches the expected age of the swelling
simulation.
500
400
4 VALIDATION OF THE ALGORITHM
(kN/m )
2
300
The algorithm introduced in previous paragraphs has
been implemented in FLAC via FISH routines. Now, 200
the validation of this algorithm against the swelling
tests presented in Wittke (1999) is presented. 100
59
-24
-20
-16
-12
ez (%)
-5 4
0.1 1 10 100
sz (MPa)
-4
Ensayo: Carga-descarga inicial Ensayo: hinchamientos
Wittke elástico Wittke plástico
FLAC elástico FLAC plástico
ε zq (%)
-3
0 with the results from the tests and the values fitted by
0 5 10 15
t (días)
Wittke (1999).
sz = 520 kN/m2 sz = 260 kN/m2 sz = 130 kN/m2 sz = 65 kN/m2 sz = 32.2 kN/m2
FLAC sz=520 kPa FLAC sz=260 kPa FLAC sz=130 kPa FLAC sz=65 kPa FLAC sz=32.2 kPa
60
Elastic Plastic
1a
2 0.0018 0.0 40.0 0.1
Figure 10. Combined swelling pressure and swelling strain 3 0.0018 0.0 2.0 0.1
test on a sample from Gypsum keuper (in Wittke 1999).
5
was reached. Starting phase 1b it was allowed for a
small vertical strain (that is not recognizable in Figure 4
10 because of the chosen scale) that results in a reduc-
tion of the vertical stress to approximately 3.8 MPa.
z(MPa)
3
Following to this, the vertical deformation of the
sample was again prevented, and consequently, the
vertical stress increased again to the same value of 2
4.2 MPa.
The course of phase 1c was equivalent to the one of 1
phase 1b. During phase 1d it was allowed for a vertical
strain slightly larger than during the preceding phases.
0
At the beginning this led to decrease the vertical stress 0 5 10 15
to less than 0.5 MPa. Subsequently, the vertical stress t (años)
was increased to 2.5 MPa over a period of 0.3 year Ensayo FLAC
without stabilization of the vertical stress.
During phase 2 of the test, the vertical stress was Figure 11. Combined swelling pressure–strain test simu-
lowered to 0.5 MPa keeping it constant for more than lated with FLAC; comparison of stresses.
5 years. The vertical strain was measured as a function
of time. Phase 2 was stopped after a vertical strain of
approximately 28% had occurred without stabilization of the coefficient avp from 40 year
1 to 2 year
1, that
of the deformations. is equivalent to a reduction of the permeability of the
During the phase 3 of the test, a further increase sample during phase 3 due to the increment of the
of the vertical strain was prevented and the increase vertical stress in this phase of the test.
of the vertical stress was registered as a function of Changes in permeability of the sample during
time. The slope of the stress–strain curve decrease the load process are not taken into account in the
continuously with time and after a period of 5.7 years a formulation of the kinetic, thus Wittke (1999) suggests
vertical stress of 4 MPa was measured. that the coefficient avp should vary during the calcula-
Wittke (1999) reproduced this test with the elastic– tion for an accurate simulation of the swelling process.
plastic properties presented in Table 2. The test described in this paragraph has been sim-
For the swelling parameters, Wittke (1999) uses ulated with FLAC, considering the same parameters
the following values, 0 16 MPa and Kq 15%. (in Table 2 and Table 3). Figure 11 presents the evo-
Nevertheless, regarding on the kinetics of the swelling, lution of the vertical stress (in MPa) versus time (in
Wittke found necessary to change the swelling time years), comparing the results from FLAC simulation
parameter during the course of the test. The parameters against the test.
proposed are presented in Table 3. Figure 12 presents the evolution of the vertical
As Wittke (1999) refers, to reproduce accurately the strain (in %) versus time (in years), comparing the
phase 3 of the test it is necessary to reduce the value results from FLAC model against the test.
61
0
0 5 10 15 REFERENCES
t (años)
Ensayo Cálculo FLAC ISRM 1989. Suggested Methods for Laboratory Testing of
Argillaceous Swelling Rocks. In Int. J. Rock Mech. Min.
Figure 12. Combined swelling pressure–strain test simu- Sci. & Geomech. Abstr, Vol. 26, No. 5: 414–426.
lated with FLAC; comparison of strains. Huder J. & Amberg G.1970. Quellung in Mergel, Opalinuston
und Anhydrit. Schweizer, Bauzeit, 83: 975–980.
5 CONCLUSIONS Noorany I., Frydman S. & Detournay C. 1999. Prediction of
soil slope deformation due to wetting, In Detournay & Hart
(eds), FLAC and Numerical Modeling in Geomechanics:
The formulation for the swelling behavior presented 101–107. Rotterdam: Balkema.
by Wittke (1999) has been reviewed and a calculation Saïta A., Robert A. & Le Bissonnais H. 1999. A Simplified
algorithm, based in this formulation, has been imple- Finite Element Approach to Modeling Swelling Effects
mented in FLAC, via FISH routines. This algorithm in Tunnels. In Alten et al. (eds), Challenges for the 21st
allows the simulation of the swelling behavior with Century: 171–178. Rotterdam: Balkema.
FLAC code. Steiner W. 1993. Swelling Rock in Tunnels: Rock Charac-
The algorithm has been checked against different terization, Effect of Horizontal Stresses and Construction
swelling tests presented by Wittke (1999), and the Procedures. In Int. J. Rock Mech. Min. Sci. & Geomech.
Abstr. Vol. 30. No. 4: 361–380.
results from the model fit quite well to the results of Wittke W. 1999. Stability Analysis for Tunnels. Fundamentals.
the different tests. Therefore, these routines can be used Geotechnical Engineering in Research and Practice.
to simulate the swelling behavior of expansive grounds WBI-Print 4. Ed. WBI Prf.Dr.Ing. W. Wittke. Consulting
in real engineering problems. engineers for Foundation and Construction in Rock Ltd.
Nevertheless, when using these routines to simu- Verlag Glückauf GmbH. Essen.
late a swelling behavior, the following limitations of
the formulation have to be remembered:
– The direction of the principal stresses does not
change during the swelling process.
62
ABSTRACT: We present the use of FLAC3D in conjunction with Gocad and the CSIRO-developed software
3DMACS to model deformation – fluid processes during mineralization. Simple idealized geological models
containing one planar fault are used to determine the effects of a number of deformation scenarios on volume
strain, pore pressure and resultant fluid flow patterns. Our results show that whether fluid flows up or down the
fault is strongly controlled by the dip of the fault; steep faults cause fluids to flow up the fault, while more shal-
low faults dilate and draw in fluid from the overlying sedimentary unit as well as the surrounding host rocks.
Geometrically complex models, which more closely resemble the geology surrounding an ore deposit, are
aimed at determining how the shape of a doubly plunging dome affects fluid flow patterns and the location of
sites of dilation. Areas of maximum dilation occur on the flanks of the dome near its crest. Complex meshes are
constructed using Gocad, which is then translated into FLAC3D using 3DMACS. This software may also be used
to set model parameters and properties, and for coupling FLAC3D to other numerical codes.
63
Figure 1. Screen shot of web-browser interface of 3DMACS showing how properties and boundary conditions are applied.
so that the zones become parallel to the surfaces. The – Allows the user to set group properties imported
CSIRO-developed software 3DMACS (Fig. 1) is then from an external properties database, which can
used to import this model, along with a set of proper- then be edited by the user. These properties can be
ties, into FLAC3D. from any of the 4 domains above.
– Allows the user to set model parameters and choose
2.2 The 3DMACS software suite visualization outputs.
– Due to its underlying XML character, users can
Primarily, 3DMACS is used for the importation, use a web browser (or the built-in 3DMACS GUI)
parameter-selection/editing and running of 3D geolog- from any machine connected to the internet and run
ical models. It is a suite of software modules that at their simulations remotely via 3DMACS. Multiple
its core, leverages an XML data model. It harnesses processes can be distributed amongst various
various vendor-provided software, such as FLAC3D, machines.
to provide the background simulation capabilities. – Allows for the storage of all user parameters and
Overall, it provides the following functionality: selected properties within a nominated repository,
– Allows for the importation of 3D models such as so that the user can re-run prior defined problems.
those produced by Gocad or FracSIS. The above functionality allows the simulation to be
– Can “couple” a simulation across 4 phenomeno- fully specified within the user-domain, rather than the
logically distinct domains: mechanical/deformation, process domain, which normally requires specialized
thermal, fluid and chemical. Currently FLAC3D knowledge of syntax and macro languages such as
and FastFlo (a CSIRO package for partial differential FISH. By providing basic problem “templates” for
equation solving) are used to provide modeling scenarios such as mechanical/fluid, mechanical/fluid/
across these domains. thermal and mechanical/fluid/chemical modeling,
64
2.3 Visualizing results Figure 2. Typical simple fault model. Arrows indicate
applied fluid discharge.
3DMACS provides for the visualization of FLAC3D
results in Gocad or the commercial software FracSIS,
via the export of scalar and vector data as 3D point-
cloud sets. Gocad allows for the creation of isosur-
faces from scalar point data. Both Gocad and FracSIS
can be used for volume rendering of scalar data.
FracSIS also allows the user to control the opacity of
certain color values in both scalar and vector data. By
“hiding” certain values we are able to see inside the
FLAC3D model more easily and are not required to Figure 3. Examples of initial geometries of simple fault
use cross-sections or cut planes. models.
Using FISH from within FLAC3D, we are also able
to create VTK files of scalar and vector data, which
are used by the freeware software MayaVi. MayaVi is
able to visualize isosurfaces, and scalar and vector cut
planes as well as fluid flow vectors so that their color
varies with magnitude.
All of these software packages are able to create
VRML files which, given the appropriate plug-in,
allows one to use a web-browser to view results.
65
of the model is allowed to move in the vertical direc- of extension pore pressure is 30 MPa lower than those,
tion, and which are essentially compressional. In all models,
– transpression and transtension, which are similar contours of pore pressure in the sedimentary unit slope
to the reverse and normal models but contain a towards the right due to the application of a fluid flux
strike slip component of movement. (discharge) at the left boundary.
By changing the dip and strike of the fault we have 3.2.2 Volume strain
a number of scenarios which range from a model with In all models the fault region is an area of high positive
a shallow dipping faults with a dip of 30° and com- volume strain (dilation) and a zone of significant dila-
pression at right angles to the strike of the fault to a tion propagates from the tip of the fault into the sedi-
model with a steeply dipping fault (60°) where the mentary (Fig. 5). In the compression, reverse and
compression direction is at 45° to the strike of the fault. transpression models this zone is oriented roughly
Constant fluid fluxes of 1m/yr are applied at the parallel to the strike of the fault. In the extension, nor-
base of the fault and the left boundary of the sedi- mal and transtension models this zone is much steeper,
mentary unit. Permeability is isotropic and remains and in the strike-slip models it is close to vertical.
constant during deformation. Mechanical anisotropy Positive volume strain in the fault is greatest in
is modeled using the ubiquitous joints constitutive models with an extensional component. In the reverse
model. In the sedimentary unit these are oriented hor- model only the fault is a region of significant positive
izontally and represent bedding, while in the base- volume strain and therefore is also a region of low
ment units they are oriented roughly parallel to the pore pressure relative to the other basement units. The
contacts of the units and represent a pervasive cleavage. transpression model is similar; however, it contains
These fabrics are given 90% of the strength (cohe- regions of dilation in both the hangingwall and foot-
sion, tensile strength) of the rock type. Mechanical wall. In the normal and transtension models the fault
properties are listed in Table 1. The size of the model is also a region of significant dilation (higher than the
varies depending on the dip of the fault. All models reverse and transpression models). As with volume
are 2 km tall (z-direction) and 2 km deep (y-direction) strain (dilation/contraction) the fault region records
but the width (x-direction) changes. In all cases the the highest shear strain in all models. The location
bottom of the fault is a minimum of 1500 m away and orientation of the zones of high shear strain are
from either boundary. All models are deformed to 5% coincident with those of significant dilation in all
shortening or the equivalent amount of displacement models.
for those models with a strike-slip component.
3.2 Results
3.2.1 Pore pressure
Models which have an extensional component of
deformation (extension – pure shear, normal faulting,
transtension) cause the greatest decrease in pore
pressures because they have the greatest dilation
(positive volume strain). The normal and transtension
models also have the steepest pore pressure gradients
and therefore fluid flow rates in the fault are highest in Figure 5. Volume strain increment and fluid flow vectors in
these models. Pore pressure at the bottom of all models compression model (cross-section view through middle of
is similar; however, in the models with a component model). Maximum fluid flow velocities are 2.02 108 m/s.
Table 1. Mechanical and fluid flow properties used in simple fault models.
66
3.2.3 Fluid flow vectors more because they are oriented at an angle which is
Fluid flows towards the center of the fault in the com- parallel to the direction of maximum compression.
pression, reverse and transpression models (Fig. 6). Steeply dipping faults and those oriented normal to
This is a result of the fault being an area of significant the maximum compression direction are more likely
dilation, and low fluid pressure, relative to the sur- to contract.
rounding rocks. In the extension, normal and transten- In models where the compression direction is nor-
sion models, fluid flows up and out of the fault into mal to the strike of the fault, the dip of the fault has
the hanging wall. This occurs despite the fault being little affect on the orientation of the high strain zone,
an area of higher positive volume strain (dilation) which propagates into the sedimentary unit. In all
than the surrounding rocks. Fluid flows up because models where the compression direction is normal to
pore pressure gradients are much steeper than in the the strike of the fault, this zone of high strain takes on
compressional models. In the strike-slip models fluid a dip of 45°. In models where the fault dips 60° and
flows up and parallel to the boundaries of the fault. In the hanging and footwall are both pelitic gneiss the
the extension, normal and transtension models fluid orientation of the high strain zone in the basement is
on the right side of the sedimentary unit flows to the also 45°. When the footwall rock type is made for
right towards the zone of significant dilation which rigid, the high strain zone in the basement is nearly
propagates from the tip of the fault. parallel to the dip of the fault. In models where the
fault dips 30° and the high strain zone and the foot-
3.2.4 Dip and strike of fault wall is more rigid the high strain zone in the basement
The dip angle of the fault strongly controls whether is parallel to the fault. This high strain zone becomes
fluid flows up and parallel to the fault boundaries. steeper (close to 45°) in the sedimentary unit.
Models with the fault dipping 60° allow fluid to flow Changing the orientation of the fault with respect
up the fault. Fluid will flow up a steep fault whether to the model boundaries and direction of compression
the footwall rocks are quartzite or pelitic gneiss. results in different orientations for the high strain
Similarly when the fault dips 45°, fluid flows into the zones. In the models where the strike of the fault is ori-
fault from the foot and hanging walls whether the ented at 45° to the maximum compression direction
footwall rocks are quartzite or pelitic gneiss. In mod- (and the dip is 60°), close to the fault the high strain
els where the fault is steep (60°), the fault does not zone is parallel to both the dip and strike of the fault
dilate as much and therefore pore pressure does not in both the basement units and the sedimentary unit.
decrease in the fault as much as it does in models Away from the fault zones, high strain zones form
where the fault has a shallower angle. Decreasing the with a strike normal to the maximum compression
dip angle of the fault to 30° causes the fault to dilate direction and a dip of 45°. When the angle between
more than models where the fault dips at 45°. This the strike of the fault and the maximum compression
however does not significantly affect the pore pressure direction is increased to 67.5°, the orientation of the
values in the fault and fluid still flows towards the high strain zone in the basement is nearly parallel to
center of the fault from the hanging and footwalls. the fault. As this zone propagates into the sedimen-
Moderate to shallowly dipping faults are able to dilate tary unit its orientation rotates towards a strike which
is normal to the maximum compression direction and
a dip closer to 45°. Therefore, both the strike and dip
of the fault, with respect to the maximum compres-
sion direction as well as the strength of the rocks, may
control the orientation of the high volume and shear
strain zones.
In models where the direction of maximum com-
pression is oriented less than 90° to the strike of the
fault, fluid flow vectors change along a line stretching
from the top west end to the bottom east end of the
fault. On the west side flow is directed up and out into
the hanging wall, while on the east side fluid flows
down and into the footwall side of the fault.
Increasing the permeability of the fault marginally
does not change the values of volume strain or pore
pressure in a significant manner. Fluid flow patterns
remain the same however fluid flow velocities are
increased slightly.
Figure 6. Fluid flow vectors in and around the fault Decreasing the strength of bedding and cleavage
(compression model). fabrics from 90 to 75% of the strength of the host
67
4 COMPLEX GEOMETRIES
4.2 Results
The altered metasedimentary unit contains regions of
negative volume strain (contraction) on the flanks of
the basalt dome where the dip is steep and at a high
angle to the compression direction. Towards the top
of the dome (but not at the crest) the weak altered
metasedimentary unit contains regions of high posi-
tive volume strain (dilation) above the areas of con-
traction (Fig. 9).
This causes fluid flow rates to be highest close to
the top of the dome where areas of contraction and
maximum dilation are in close proximity (Figs. 10 &
11). Contraction occurs within the matrix above the
highest point of the dome. Regions of high positive
volume strain are also regions that have failed in
tension. These areas are more likely to have formed
quartz veins, which commonly host gold.
68
Table 2. Mechanical and fluid flow properties used in geometrically complex model.
69
70
M. Souley
INERIS, Ecole de Mines de Nancy, Nancy Cedex, France
K. Su
ANDRA, Châtenay-Malabry, France
M. Ghoreychi
INERIS, Parc Technologique ALATA, Verneuil-en-Halatte, France
G. Armand
ANDRA, Laboratoire de Recherche Souterrain Meuse/Haute-Marne, Bure, France
ABSTRACT: This paper deals with numerical implementation of non-linear constitutive models of rock mass
and its verifications and validations. In the 3-dimensionnal code, FLAC3D, an elasto-damage-plastic model
(damage is approached through the theory of plasticity) for Hoek-Brown media has been implemented.
Simulations of triaxial compression tests provide a verification of the numerical implementation with a good
agreement between predictions and theoretical values of peak and residual strengths. The applicability of the
implemented model to predict the damage and/or failure development around a circular opening is checked.
Finally a validation of poroplastic calculations based on the drainage of a cylindrical hole in poroplastic media
is achieved by comparison to an existing semi-analytical solution.
71
2 MECHANCICAL SHORT TERM BEHAVIOR For instance, the initiation of damage (Fsend), the
peak (Fsrup) and residual (Fsres) strengths are given by:
2.1 Brief mathematical description
Based on several triaxial laboratory tests performed
on the argillite rock samples, the typical characteris-
tics of stress–strain curves are displayed in Figure 1:
Phase 1: linear isotropic and elastic behavior after
a short non-linear phase corresponding to the closure
of microcracks;
Phase 2: strain-hardening in the pre-peak region
corresponding to the initiation and the growth of
microcracks assumed to be described by plasticity,
contrary to the concept of effective stress and the
hypothesis of strain equivalence (Lemaitre 1995, Ju (1)
1989) that is generally used;
Phase 3: softening after the peak (failure) associated where mend, send and mrup, srup are Hoek-Brown con-
with a progressive loss in material cohesion and then a stants respectively corresponding to onset of damage
decrease in strength; and the peak; cend and crup are uniaxial compressive
Phase 4: residual phase where the rock strength strength at the onset of damage and peak; uniaxial
remains practically constant. residual strength; 3bd confining pressure for
Based on these observations, a constitutive model brittle/ductile transition; 1 and 3 major and minor
for this material was firstly developed in the framework principal stresses (compressive stress is negative and
of the European project: EURATOM MODEX-REP 1 2 3).
and recently compiled by Su (2003).
The features of this model are: 2.2 Constitutive equations
(a) linear elasticity to model the Phase 1; In order to obtain a simple but general constitutive
(b) damage initiation and growth are approached by a model, an extended Hoek-Brown yield function is used.
strain-hardening based on Hoek-Brown criterion The general form of the yield function is expressed in
where the Hoek-Brown constants and the uniaxial the following equation:
compressive strength are plastic strain dependent;
(c) the peak, post-peak (Phase 3) and residual (Phase
4) are also based on Hoek-Brown criterion with (2)
respect to brittle/ductile transition in accordance
with the experimental data.
where softening flow function (parabolic form
with respect to the internal plastic variable, in phase
3, and null elsewhere); m, s Hoek-Brown constants
(linearly varying with in phase 2, and constants
1 2 3 4 in phase 3 and 4); c uniaxial compressive strength
(linearly varying with in phase 2, and constant in
Peak
phase 3 and 4).
Deviatoric stress (σ1-σ3)
where d
p
increment of total damage/plastic strain
tensor in phases 2, 3 and 4.
Axial strain( ε1) For simplicity, an associated flow rule is used (the
plastic potential is identical to the yield function given
in Equation 2). In addition, in order to take into account
Figure 1. A typical stress–strain curve provided by a triaxial the geometry of stresses (compression differing to
test. extension), the previous yield function is generalized
72
in terms of the three stress invariants (J1, J2 , J3 ). By substituting Equations (4), (7) and (5) in Equation
Then, principal stresses are expressed in terms of mean (8), we can express the plastic multiplier:
stress (p), generalized deviatoric stress (q) and Lode’s
angle () according to:
(9)
where
(10)
and compiled
The incremental expression of Hooke’s laws in terms as DLL file (dynamic link library) that can be loaded
of generalized stress and strain tensors has the form: whenever it is needed.
(7)
2.4 Verification and validation
where C isotropic linear tensor. In order to verify the implemented model, seven triaxial
compression tests with confining pressures of 2, 5,
The consistency condition, dFs() 0 leads to:
10, 12, 16, 20 and 25 MPa have been simulated. They
are the part of the wide number of triaxial compression
(8) tests used to characterize the non-linear behavior of
the studied materials.
73
mend send cend (MPa) mrup srup crup (MPa) 3bd (MPa)
55 80
25 MPa
Deviatoric stress 13(MPa)
50
20 MPa 45
70
Principal major 1
16 MPa 40 60
35 50
12 MPa
30
(MPa)
25 40
10 MPa
20 30
5 MPa
15
20
10 2 MPa
5 10
Lateral strain [-]
0
Axial strain [-] Principal minor stress 3 (MPa)
-0.03 -0.02 -0.01 0 0.01 0.02 0.03
0
-10 0 10 20 30
Figure 2. Numerical result of triaxial compression tests. Peak [Eq. 1]
Residual [Eq. 1]
Onset of damage [Eq. 1]
Peak - Flac3d
The input parameters needed are derived from Residual - Flac3d
Onset of damage - Flac3d
standard laboratory tests and based on the identification
of model parameters. These parameters are summarized
in Table 1. The isotropic elastic characteristics used Figure 3. Onset of damage, peak and residual strengths:
numerical and analytical solutions.
are E 4500 MPa and 0.3. The generalized peak
and residual plastic strain were also identified from
triaxial laboratory tests. These are: rup 0.0063 and the beginning of phase 2). This validates the numerical
res 0.0175. implementation of the elasto-damage-plastic model
FLAC3D simulations are carried out on a single in FLAC3D.
zone of unit dimensions: fixed normal displacements
are applied to 3 perpendicular planes (bottom and two 2.5 Application to a circular opening
perpendicular vertical faces). First, the three other
The aim of this section is to provide a verification of
faces of model are subjected to an isotropic stress state
the implementation for non-triaxial stress paths and to
corresponding to the given confining pressure.
show numerically the ability of the implemented
Secondly, the deviatoric stress is exerted by applying
model to evaluate the extent of damaged and/or failed
a constant displacement rate at the top of model.
zones around a circular underground excavation.
Figure 2 presents the deviatoric stress–axial and lat-
We then consider an infinite circular opening in an
eral strain curves for different confining pressure. From
infinite elasto-damage-plastic medium initially sub-
this figure, we note that the post-peak behavior is con-
jected to an anisotropic initial stress in order to maxi-
fining pressure dependent: the transition stress between
mize the deviatoric stress and then, the risk of damage
brittle failure and ductile behavior is clearly marked
and/or failure. The axis of gallery is parallel to the
and the numerical transition stress, bd 3, is approxi-
horizontal minor stress leading to a maximum devia-
mately about 20 MPa. These curves are qualitatively
toric stress in the gallery section.
similar to the experimental ones (not reported herein).
The 2D-plane strain geometry as well as the initial
Figure 3 shows a comparison in terms of the damage
in situ stresses and model geometry including a cir-
threshold, the peak and residual strengths between the
cular gallery are plotted in Figure 4. The modeling
predictions (corresponding values in Fig. 2) and theory
sequence was performed as follows:
(Eq. 1).
The match is very good as may be seen in this fig- (1) the model without excavation was consolidated
ure, where numerical and analytical solutions coincide. under the previous in situ stresses, and
More precisely, the relative error for peak and residual (2) the circular excavation was carried out using
strengths is less than 0.3%, and 0.9% for the onset of roller boundaries to the model sides respectively
damage (dependent on the magnitude of loading at parallel to x- and z-axis for seeking symmetry.
74
v v 0
v r*
-2 0 2 4 6 8 10
radius of gallery: 3 m -4 radial
model length : 30 m -6
model heigth : 30 m
-8
model thickness : 1 m
gallery axis : // à h(// à Oy) -10 axial
H
H -12
v= h= 10.8 MPa
H= 15.1 MPa -14
-16 orthoradial
-18
Stresses (MPa) (a)
-20
0
0 2 4 6 8 r* 10
-5 radial
Figure 4. Model geometry, initial stress state and boundary -10
conditions. axial
-15
-20
-25
orthoradial
-30
Stresses (MPa) (b)
-35
75
3 HYDROMECHANICAL VALIDATION 35
3.1 Brief mathematical description of mechanical This formulation slightly differs from the elasto-
model damage-plastic model detailed in section 2 by the
The mechanical behavior is described by an elasto- absence of hardening in the pre-peak region and brit-
plastic model with a post failure softening phase. The tle/ductile transition. Based on the previous imple-
model is based on the Hoek-Brown criterion with mentation, this variant of the elasto-damage-plastic
associated plastic potential. The main characteristics model is implemented in FLAC3D. As verification, tri-
of this model are: (a) linear and isotropic behavior in axial com-pression tests were simulated. The results
the pre-peak region; (b) peak strength governed by the are shown in Figure 7. In addition, the corresponding
Hoek-Brown criterion; (c) a softening phase based on a numerical residual and peak strengths are represented
Hoek-Brown yield function and an associated flow rule; in Figure 8 and compared with the analytical expres-
(d) a perfectly plastic behavior in the residual phase. sions (Eq. 12 & 13). From Figure 7, it should be noted
Assuming that compressive stress is negative and that for a given level of confining pressure, the three
1 2 3, the peak strength and residual strength phases (elastic before failure, softening for post-peak
are given by: behavior and perfect plastic for residual behavior) are
clearly distinguished.
The match is very good as may be seen in Figure 8,
(11) where numerical and analytical solutions coincide.
The relative error for strengths is less than 0.5%.
76
50 Initial conditions
Major principal stress 1
P9 E0 (MPa) 0 c (MPa) m R
5800 0.3 14.8 2.62 0.015 0.01
r30 h (m/s) b M (MPa) w (kN/m3)
9º
10 12 0.8 6000 10
3m PY
30 m
PZ1 and pore pressure p, along the inner wall are expressed
as follows:
r3
PZ0 (15)
0,1 m
(16)
subjected to a uniform and isotropic stress state and a
uniform pore pressure. The induced mechanical and
hydraulic perturbations are examined during and after where t time; T 1.5 106 s represents the exca-
excavating. The main assumptions are: vation duration.
A semi-analytical solution of this H-M 1D problem
– gravity forces are neglected; has been developed in the framework of the European
– mechanically, the medium behaves as an isotropic project: EURATOM MODEX-REP (Su 2002).
and elastoplastic material according to the model Finally, the geometry shown in Figure 9, initial and
described in §3.1; boundary conditions reported in Table 2 are used in
– hydromechanical coupling process is expressed by our FLAC3D model. Hydromechanical properties are
Biot’s theory; shown in Table 3, where E0 and 0 denote the drained
– hydraulic and mechanical boundary conditions at the elastic properties; h is the hydraulic conductivity;
hole walls are time-dependent: continuous reduc- b the Biot coefficient; M the Biot modulus and w
tion of normal stress and pore pressure at the hole the specific weight of water.
boundaries from their initial values to zero. For both semi-analytical and numerical solutions
The geometry of this 1D problem is shown in the required results are:
Figure 9. It consists of a thick wall cylinder with – the radial displacement;
internal radius of 3 m and external radius of 30 m. The – the pore pressure;
initial and boundary conditions are summarized in – the radial; orthoradial and axial effective stresses.
Table 2.
As previously mentioned, the hydraulic and mechan- as a function of radial distance from the hole center
ical boundary conditions along the inner wall are time- (r ranged from 3 to 30 m) and time (ranged from 0
dependent. More precisely, the total radial stress r to 100 Ms) in this paper.
77
Height radial distances are considered for output. to negative pore pressure) are well reproduced by
They are: 3, 3.05, 3.1, 3.2, 3.5, 3.7, 5, and 10 m. Ten numerical results for radial distance and time ranged
time periods are also considered for result output. The respectively from 3.05 to 3.2 m, and from 1.5 to
involved times are: 1.2, 1.5, 1.6, 2.5, 10, 50, and 1.6 Ms (corresponding to the start of full drainage).
100 Ms (million of seconds). Due to null and negligible values of pore pressure
In the case of the semi-analytical, all of these in the vicinity of the inner radius; relative errors
required results are given at the previous radial between semi-analytical and numerical solutions are
distances. Because of displacements and pore pres- not evaluated for radial distance inferior to 3.7 m; so
sure are gridpoint variables whereas stresses are zone the difference in results of pore pressure does not
variables and evaluated at the zone centroid, numerical exceed 0.02 MPa. For radial distance superior to 3.7,
solutions are checked at the following set of radial the maximum relative error between semi-analytical
distances: and numerical solutions is about 1.2%.
Comparison of radial and orthoradial effective
– 3, 3.05, 3.1, 3.2, 3.5, 3.7, 5, and 10 m for radial dis-
stresses between the semi-analytical and numerical
placement and pore pressure;
solutions is plotted in Figures 12 & 13. It should be
– 3.0125, 3.0625, 3.1125, 3.2125, 3.5625, 3.725,
noted that the profiles of principal effective stresses
5.05, and 10.05 m (centroid of the closest zone) for
are qualitatively returned.
stresses.
From a quantitative point of view and for a radial
Therefore, small differences in the results of stresses distance superior to 3.1 m; the absolute error on the
compared to the semi-analytical solution are to be
expected. r=3 - Anal
5 r=3,05 - Anal
r=3,1 - Anal
r=3,2 - Anal
3.3 Comparison with the semi-analytical solution 4 r=3,55 - Anal
r=3,7 - Anal
Pore pressure (MPa)
r=3,05 - Anal
r=3,1 - Anal
-2 -2
Radial displacement (mm)
r=3,1 - Anal
r=3,2 - Anal
r=3,2 - Anal
-3 r=3,55 - Anal
r=3,55 - Anal
r=3,7 - Anal
-4 r=3,7 - Anal -4
r=5 - Anal
r=5 - Anal
-5 r=10 - Anal
r=10 - Anal
r=3 - Flac3D -6 r=3,013 - Flac3D
-6 r=3,063 - Flac3D
r=3,05 - Flac3D
r=3,1 - Flac3D r=3,113 - Flac3D
-7
r=3,2 - Flac3D -8 r=3,213 - Flac3D
-8 r=3,55 - Flac3D r=3,563 - Flac3D
r=3,7 - Flac3D r=3,725 - Flac3D
-9 r=5,05 - Flac3D
r=5 - Flac3D -10
-10 r=10 - Flac3D r=10,05 - Flac3D
Figure 10. Numerical and semi-analytical solutions: radial Figure 12. Numerical and semi-analytical solutions: radial
displacement. effective stress.
78
79
agreement between predicted peak and residual Martin, C.D. & Chandler, N.A. 1994. The progressive fail-
strengths and theoretical ones. ure of Lac du Bonnet granite. International Journal of
Comparison of normal displacement between the Rocks Mechanics and Mining Sciences. 31(6): 643–659.
semi-analytical solution and the numerical ones indi- Shao, J.F., Chiarelli, A.S. & Hoteit, N. 1998. Modeling of
coupled elastoplastic damage in rock materials. Interna-
cates a very good agreement: the relative error is glob- tional Journal of Rocks Mechanics and Mining Sciences.
ally about 0.2%. Also, for a given radial distance, 35(4–5): Paper No. 115.
both numerical and semi-analytical solutions are Sharan, S.K. 2003. Elastic-brittle-plastic analysis of circu-
quantitatively and qualitatively similar in terms of lar openings in Hoek-Brown media. to appear in
pore pressure results with a neglected difference International Journal of Rocks Mechanics and Mining
(0.02 MPa) compared to the initial field of pore pres- Sciences & Geomechanics Abstracts.
sure (4.7 MPa). Souley, M., Homand, F., Hoxha, D. & Chibout, M. 1999.
In terms of radial and orthoradial effective stresses, Damage around a keyed URL excavation: change in per-
the semi-analytical and numerical solutions are qual- meability induced by microcracks growth. In Detournay
& Hart (eds), FLAC and Numerical Modeling in Geo-
itatively the same. Far to the gallery wall, the stress mechanics: 205–213. Rotterdam: Balkema.
difference does not exceed 0.3 MPa (that corresponds Souley, M., Hoxha, D. & Homand-Etienne, F. 1998. Distinct
to a relative error of 2%). element modelling of an underground excavation using a
continuum damage model. International Journal of Rocks
Mechanics and Mining Sciences. 35(4–5): Paper No. 6.
Steif, P.S. 1984. Crack extension under compressive load-
REFERENCES ing. Engineering Fracture Mechanics. 20(3): 463–473.
Su, K. 2002. Analysis of the capacity of numerical models
Carranza-Torres, C. & Fairhurst, C. 1999. The elasto-plastic to simulate excavation in deep argillaceous rock,
response of underground excavations in rocks masses 5th EURATOM framework programme, MODEX-REP
that satisfy the Hoek-Brown failure criterion. Interna- project contract FIKW-CT2000-00029 – Deliverable 1,
tional Journal of Rocks Mechanics and Mining Sciences. August 2002.
36(6): 777–809. Su, K. 2003. Constitutive models of the Meuse/Haute-
Detournay, E. & Cheng, A.H.-D. 1993. Comprehensive Rock Marne Argilites, MODEX-REP project contract FIKW-
Engineering. Pergamon Press Ltd. CT2000-00029 – Deliverable 2&3, February 2003.
Detournay, E. & Cheng, A.H.-D. 1988. Poroelastic Response Voller, V., Peng, S. & Chen, Y. 1996. Numerical Solution of
of a Borehole in a Non-Hydrostatic Stress Field. Interna- Transient, Free Surface Problems in Porous Media.
tional Journal of Rocks Mechanics and Mining Sciences. International Journal of Numerical Methods in Engi-
25(3): 171–182. neering. 2889–2906.
Homand-Etienne, F., Hoxha, D. & Shao, J.F. 1998. A contin- Vouille, G., Tijani, M. & Miehe, B. 2001. Hydro-mechanical
uum damage constitutive law for brittle rocks. Computers theoretical problem: Drainage of an infinite medium
and Geotechnics. 22(2): 135–151. by a cylindrical hole. In EC-5th EURATOM framework
Ju, J.W. 1989. On the energy based on coupled elastoplastic programme 1998–2000 MODEX-REP project: contract
damage theories: constitutive modeling and computa- FIKW-CT-200-00029, NOT-EMP-01-02, Technical Note,
tional aspects. International Journal of Solids Structures. fevrier 08.
25(7): 803–833. Wong, T.F. 1982. Micromechanics of faulting in Westerly
Lemaitre, J. 1985. A course on damage mechanics. 2nd edi- granite. International Journal of Rocks Mechanics and
tion. Springer. Mining Sciences. 19(1): 49–62.
80
Slope stability
ABSTRACT: Slope failures can cause delay in the production schedule and the loss of life and equipment. In
this study, slopes excavated in very weak rock masses where expected failure mode is circular failure have been
studied. The effects of slope height, slope angle, water saturation, cohesion, internal friction angle and density
on slope stability under circular failure conditions have been investigated by three methods: Hoek and Bray stabil-
ity diagrams, Bishop’s simplified method of slices, and finite difference numerical code, FLAC3D (Itasca 1997).
Safety factor calculations have been carried out for the various values of parameters and obtained values are
compared with each other. However, the main focus is on the results of the numerical modeling. The presence
of correlation between the studied parameters and the factors of safety obtained from numerical models has
been searched and the fitted equation has been given.
83
determined from the chart given by Hoek and Bray Table 1. Parameter values used in all approaches.
(1981) was chosen as a basis for comparison purpose.
In other words, safety factors have been calculated for Parameters Values Parameters Values
this critical failure surface by these three methods.
Cohesion (kPa) 50 Cohesion fully 50
However, the main focus of this study is on the results
90* saturated (kPa) 90
of numerical modeling. 130 130
Geomechanical properties of rock and the shape of 170 170
the slope used in this study can be listed as:
Internal friction 20 Slope angle (°) 30
– Cohesion angle (°) 25* 40
– Internal friction angle 30 50
– Density 35 60*
40 70
– Water condition
– Slope angle Density (t/m3) 1.6 Slope height (m) 20
– Slope height 1.9 50*
2.2 80
All these factors have been taken as parameters 2.5* 140
and safety factors have been calculated. For all three 2.8 200
methods, the value of the parameter whose effect on
safety factor will be investigated has been changed *Constant values.
while the other parameters have been kept constant.
The parameters and values used are given in Table 1.
84
each zone can be calculated from the Equation 1 where si shear strength; n normal stress;
(Brady & Brown 1993): internal friction angle.
Local safety factor is described as the ratio of shear
strength to shear stress developed on the critical plane
(1)
for each zone:
85
properties used in numerical modeling are given in models has been searched. For this purpose, a R factor
Table 2. k ratio, the ratio of horizontal stress to vertical has been proposed and calculated from the Equation 11:
stress, has been found from the Equation 9:
(11)
(9)
86
These indicators are shown in Figure 4 for one of the In the light of the evaluation of the numerical models
numerical models. The displacement vectors and the having the mohr-coulomb plasticity model, safety fac-
contours of shear strain increment of the same numeri- tors of numerical models have been classified in terms
cal model are given in Figure 5. of failure, and given in Figure 6.
Safety factor values obtained from the models hav-
ing different Poisson ratios for cohesion parameter
Table 4. Safety factors and failure conditions of numerical are given in Figure 7.
models in FLAC3D. Results of the safety factor calculation for numeri-
cal models with different tensile strengths calculated
Parameters Values Safety factor Failure according to Equation 10 are shown in Figure 8 for
the cohesion parameter.
Cohesion 50 0.551 YES A correlation has been established between R fac-
(kPa) 90 0.705 YES
130 0.859 NO
tor and safety factors of numerical models. According
170 1.013 NO to this, the obtained linear regression model and cor-
relation coefficient are as follows:
Internal fric- 20 0.641 YES
tion angle (°) 25 0.705 YES
30 0.750 NO (12)
35 0.788 NO
40 0.821 NO
Density (t/m3) 1.6 0.900 NO where F safety factor; r correlation coefficient.
1.9 0.815 NO It can be said that there is a very strong positive
2.2 0.753 YES linear relationship between R factor and safety factor
2.5 0.705 YES and it is shown in the Figure 9.
2.8 0.668 YES
Cohesion 50 0.103 YES
(fully satu- 90 0.141 YES
rated) 130 0.180 YES 5 CONCLUSIONS
(kPa) 170 0.218 YES
For all parameters high safety factor values have been
Slope angle 30 1.450 NO
(°) 40 1.038 NO
given by Bishop Approach and it has been followed
50 0.856 NO by Hoek and Bray and FLAC3D approaches. But in
60 0.705 YES full saturated condition, high safety factor values are
70 0.551 YES given by Hoek and Bray approach and it is followed
Slope height 20 1.094 NO by Bishop and FLAC3D approaches.
(m) 50 0.705 YES It was considered that the reason for low safety
80 0.597 YES factor values in numerical models was the assumption
140 0.533 YES of no tensile strength. Then new numerical models in
200 0.472 YES which tensile strength was calculated depending on the
cohesion and internal friction angle have been run and
87
Figure 5. The displacement vectors and the contours of shear strain increment of the numerical model.
1.6 1.8
1.4 1.6
1.2 1.4
Safety Factor
Safety Factor
1.0 1.2
1.0
0.8
0.8
0.6 0.6
0.4 Failure 0.4 στ = 0
0.2 No failure 0.2 στ # 0
0.0 0.0
1 2 3 4 5 6 7 8 9 10 11 12 13 50 90 130 170
Cohesion (kPa)
Figure 6. Safety factor values classified in terms of failure.
Figure 8. Safety factors at two different tensile strength
1.4
conditions.
1.2
2
1.0
Safety Factor
1.5
Safety Factor
0.8
0.15 1
0.6 0.20 0.5
0.4 0.25
0
0.2 0.30 0 0.005 0.01 0.015 0.02
0.35 R Factor
0.0
50 90 130 170
Figure 9. The relationship between R factor and safety factor.
Cohesion (kPa)
88
be useful during the design stage. But it should be Hoek, E. and Bray, J.W. 1981. Rock Slope Engineering.
noted that this regression model should be used with London Institution of Mining and Metallurgy, 358 p.
great caution. The reason for this is that safety factors Itasca Consulting Group, Inc. 1997. FLAC3D – Fast
calculated by using equation 12 will depend on numer- Lagrangian Analysis of Continua in 3 Dimensions, Version
2.0 User’s Manual. Minneapolis, MN: Itasca.
ical models formed in this study. Because, equation 12 Kourdey, A., Alheib, M. and Piguet, J.P. 2001. Evaluation of
has been obtained from regression analysis carried out Slope Stability by Numerical Methods, 17th Int. Mining
on the results of numerical modeling studies. These Congress and Exhibition of Turkey, IMCET 2001. Ankara.
results are greatly affected by numerical discretization Sjöberg, J. 1999. Analysis of Large Scale Rock Slopes,
chosen for numerical models in this study. Doctoral Thesis, Lulea University of Technology.
Stacey, T.R. 1968. Stability of Rock Slopes in Open Pit Mines.
National Mechanical Engineering Research Institute.
Council for Scientific and Industrial Research, CSIR
REFERENCES Report MEG 737, Pretoria, South Africa, 66 p.
Starfield, A.M. and Cundall, P.A. 1988. Towards a Method-
ology for Rock Mechanics Modeling. Int. J. Rock Mech.
Aksoy, M. 2001. A Study on the Effect of Parameters
Min. Sci. & Geomech. Abstr. 25(3): 99–106.
Affecting safety Factor of Slopes under Circular Failure
Condition, MSc Thesis, Osmangazi University, Turkey.
Brady, B.H.G. and Brown, E.T. 1993. Rock Mechanics for
Underground Mining. London Second Edition, Chapman &
Hall.
89
S.A. Bastani
Leighton Consulting, Inc., Irvine, California, USA
B.L. Kutter
University of California, Davis, California, USA
ABSTRACT: Several earth dams, tailings dams, and slopes failed or were severely damaged due to liquefaction
during or after earthquakes. In seismic areas, earth structures such as embankments may be subjected to two forces:
the static loads due to gravity and the inertia forces caused by earthquakes. In a significant number of cases, lique-
faction-induced failure of embankments occurred from seconds to hours after an earthquake. In these cases, lique-
faction reduced the material strength and the failure occurred only under static loads. A finite slope was modeled at
UC Davis National Geotechnical Centrifuge to evaluate post-earthquake deformations with an injection-induced
liquefaction system. A constitutive model was developed to capture the behavior of sands with a minimum number
of physically meaningful parameters to enable prediction of post-earthquake liquefaction and/or seepage-induced
liquefaction. This constitutive model is based on the Mohr-Coulomb constitutive model and the Critical State con-
cept by adding three parameters to the conventional Mohr-Coulomb model. The constitutive model performed ade-
quately for modeling the sand behavior under monotonic drained and undrained triaxial loading and water injection
for a simple shear test under a constant shear stress. Using the new constitutive model, the failure mode of the cen-
trifuge model due to seepage-induced liquefaction was studied utilizing FLAC. Stress and strain paths for specific
elements in the embankment are studied and presented in this paper.
91
(15,19)
(40,16)
where e is the void ratio, (ecs)a is the critical state void
(34,12)
(15,16) (28,16)
ratio at one atmosphere,
is the virgin compression
(18,4)
(20,13) Yolo Loam
slope, p is the mean effective stress, and pa is the
(13,9) atmospheric pressure.
(18,6) As explained by Been & Jefferies (1985), the grad-
Nevada Sand (47,6)
ual change from dilative to contractive behavior can be
quantified in terms of the state parameter . The dila-
tion angle (dilation) was modified based on the state
Monterey Sand
parameter and its changes according to the equation:
0 0.5
Grid plot
(2)
Figure 1. FLAC grid.
92
0.92
300 (')cs 300
(')cs+(')dilation
0.88
250 250
0.84
e
200 200
0.8
q (kPa)
q (kPa)
150 150
0.76
100 100 1 10 100
p'(kPa)
50 50
0 0
0 50 100 150 200 250 300 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
p'(kPa) γ
150 0.04
100 0.03
Pore Water Pressure (kPa)
-50 0
Cavitation
-100 -0.01
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
γ γ
continued up to the critical state condition. Finally, the Table 1. Model parameters.
model behavior was studied under a constant shear
Parameters Value
stress and pore water pressure increase, modeling a
simple shear test with pore fluid injection. The consti-
* 0.022
tutive model slightly dilated prior to reaching the fail- (
/5) 0.0044
ure envelope; thereafter, the sample dilated with the cs (degree) 32
increase of pore water pressure and the stress path (dilation)max (degree) 10
approached the origin along the failure envelope in the 0.25
p -q space until it reached the critical state condition e0 (initial void ratio) 0.77
similar to the stress path suggested by Boulanger (ecs)a* 0.809
(1990). The dilation rate was less than that shown by Atmospheric pressure, pa (kPa) 101.2
his experiment (Boulanger, 1990); however, the stress
* Archilleas et al. 2001.
path, boundary condition, and initial condition of the
experiments performed by Boulanger prior to water
injection into his simple shear tests were not known, dynamically under the unbalanced external loads.
and therefore were not completely simulated by this Some oscillation is observed in the q– curves at shear
calibration. As expected, the water injection to the strains greater than 0.12, but the softening behavior can
element led to an unstable condition when the strength still be clearly observed during the dynamic collapse.
of the element dropped below the applied shear The parameters used for this calibration and later
stress. Continued softening caused the stress path to in the numerical modeling based on this constitutive
drop toward the origin while the sample collapsed model are provided in Table 1.
93
4 FLAC MODEL BEHAVIOR for the two runs are plotted on Figure 3. This figure
indicates the following behaviors:
FLAC version 4.0 was utilized to model the centrifuge
1. The mobilized friction angle was reduced along
test. The numerical model was run twice. The first
the Nevada Sand interface elements by the new
run used the conventional Mohr-Coulomb constitu-
constitutive model and along a deeper seated fail-
tive model, while the second run utilized the new con-
ure plane as shown by the new constitutive model;
stitutive model (discussed in Section 3) to model the
2. Volumetric strains were concentrated along the
Nevada Sand behavior. This numerical model was
interface of Nevada Sand and Yolo Loam in both
bounded with its and the constitutive model’s limita-
numerical models. However, deeper volumetric
tions; however, it was successfully used to observe the
strains were observed in the modified constitutive
general mechanism of localized increase in void ratio
model, which coincided with the friction angle and
just beneath the less permeable clayey silt layer, and
shear strain patterns;
the failure mechanism; exact predictions were not
3. Shear strains were also concentrated at the slope
expected. The FLAC runs were performed with the
interface within the Nevada Sand layer. Similarly a
large-strain mode.
deeper shear zone was predicted by the new consti-
Figure 1 shows the grid utilized in this model. The
tutive model matching the volumetric strain and
grid nodes and elements are identified in the subse-
mobilized friction angle reduction patterns;
quent figures with their column and row numbers
4. Sand and clay layers moved downward at the slope,
(i,j). The column and row numbers increase from left
which was translated to vertical uplift at the toe.
to right and bottom to top, respectively.
Contours of mobilized friction angle, volumetric It is worthwhile to mention that the pore water pres-
strain, and shear strain and grid deformation patterns sure was mainly increased from the slope toe within
Figure 3. Friction angle, volumetric and shear strains, and deformation patterns at 13 seconds of seepage.
94
the Nevada Sand layer toward the slope crest and with the second run due to the ability of the modified con-
a slower rate from the back of the slope crest toward stitutive model to withstand a mobilized friction angle
the slope. greater than the critical state friction angle during dila-
It should also be noted that the development of a tion. The majority of volumetric strains of elements
deep failure mechanism, or not, was affected by the were induced when the mean effective stress (p )
rate at which the water was injected relative to the became less than 10 kPa and close to zero. The volu-
permeability of the soils. For somewhat slower injec- metric strains were stabilized wherever the injection
tion, the deeper mechanism would disappear and slid- did not cause the strength to fall below the applied
ing along the bottom interface of the Yolo Loam shear stress.
would be apparent. For much greater injection rates, a Predicted pore water pressures and deformations
failure mechanism at the interface between the coarse are compared with the experimental results in Figures
Monterey Sand and the fine Nevada Sand was 5 & 6, respectively. In general the predicted pore water
observed (Bastani 2003). pressures are in good agreement with the experimen-
Stress/strain paths of several elements at the toe, tal results for both constitutive models. However, the
along the slope, and at the slope crest are plotted on initial rate of pore water pressure increase is in better
Figure 4. Effective stresses of slope/leaning elements agreement with the new constitutive model. Similar
reduced while oscillating around constant shear stresses trends were obtained by the numerical model, such as
up to the failure envelope. However, shear stresses of stabilization of pore water pressure at the beginning
carrying elements along the slope and its toe increased and its further increase for PPT# 5296 (Fig. 5). Other
during the failure of leaning elements until reaching than deformation at the middle of the slope (LVDT
the failure envelope. Stress paths moved toward the #3), where the experimental result indicated bulging,
origin after reaching the failure envelope and strain deformation rates and magnitudes were predicted very
softening was observed. In general, the elements at the well by both models.
toe and along the slope showed higher shear strengths The numerical model successfully predicted the
prior to their stress paths diving toward the origin in dilatancy to cause a very loose layer of sand below
Figure 4. Stress path and behavior of elements. Solid and dashed lines refer to the results of the new and Mohr-Coulomb
constitutive models, respectively.
95
5 CONCLUSION
96
presented here clearly demonstrate the possibility that Boulanger, R.W. 1990. Liquefaction Behavior of Saturated
layers that impede drainage may cause a significant Cohesionless Soils Subjected to Uni-Directional and Bi-
localized zone of softened material that should be Directional Static and Cyclic Simple Shear Stresses.
considered a possibility in seismic design. To deter- Dissertation presented to University of California, at
Berkeley, in partial fulfillment of the requirements for
mine induced deformations due to local drainage of a the degree of Doctor of Philosophy.
system, a material model that captures this process Castro & Poulos (ASCE paper circa 1984).
should be incorporated in the numerical model. The Dobry, R. & Alvarez, L. 1967. Seismic Failure of Chilean
mode of failure and local drainage of the centrifuge Tailing Dams. Journal of Soil Mechanics and Founda-
test presented here was successfully predicted utiliz- tions Division, Proceeding of the American Society of
ing the modified Mohr-Coulomb constitutive model Civil Engineers 93(SM6): 237–260.
in conjunction with FLAC numerical framework. Finn, W.D. 1980. Seismic Response of Tailing Dams.
Presented at Seminar on Design and Construction of
Tailing Dams, Colorado School of Mines, Denver,
ACKNOWLEDGEMENT Colorado, pp. 76–97.
Itasca Consulting Group, Inc. 2001. FLAC – Fast Lagran-
The authors would like to thank Dr. Ben Hushmand, gian Analysis of Continua, Ver. 4.0 User’s Manual.
James Ward, and Vivian Cheng for reviewing this Minneapolis, MN: Itasca.
Kokusko, T. & Kojima, T. 2002. Mechanism for Postlique-
paper and providing constructive comments. faction Water Film Generation in Layered Sand. Journal
of Geotechnical Engineering, ASCE 128(2): 129–137.
Malvick, E.J., Kulasingam, R., Boulanger, R.W. & Kutter, B.L.
REFERENCES 2003. Analysis of a Void Ratio Redistribution Mech-
anism in Liquefied Soil. To be Published in Proceedings
Archilleas, G.P., Bouckovalas, G.D. & Dafalias, Y.F. 2001. of the June 2003 Soil and Rock America Conference.
Plasticity Model for Sand Under Small and Large Cyc- Okusa, S., Anma, S. & Maikuma, H. 1978. Liquefaction of
lic Strains. Journal of Geotechnical Engineering, ASCE Mine Tailing in the 1978 Izu-Ohshima-Kihkai Earth-
127(11):.973–983. quake, Central Japan. Engineering Geology Vol. 16, pp.
Bastani, S.A. 2003. Evaluation of Deformations of Earth 195–224, Elsevier Scientific Publishing Co.
Structures due to Earthquakes. Dissertation presented to Poulos, S.J., Castro, G. & France, J.W. 1985. Liquefaction
University of California, at Davis, in partial fulfillment of Evaluation Procedure. Journal of the Geotechnical Engi-
the requirements for the degree of Doctor of Philosophy. neering Division, ASCE 111(6): 772–792.
Been, K. & Jefferies, M.G. 1985. A State Parameter for Seed, H.B., Lee, K.L., Idriss, I.M. & Makdisi, F.I. 1975. The
Sands. Geotechnique 35(2): 99–112. Slides in the San Fernando Dams during the Earthquake
Bolton, M. 1991. A Guide to Soil Mechanics. Published by of February 9, 1971. Journal of the Geotechnical Engi-
M D & K Bolton, Printed by Chung Hwa Book Company, neering Division, ASCE 101(GT7): 651.
pp. 63–92.
97
ABSTRACT: The stability of slopes may be estimated using 2D limit equilibrium methods (LEM) or numerical
methods. Due to the rapid development of computing efficiency, several numerical methods are gaining increasing
popularity in slope stability engineering. A very popular numerical method of slope stability estimation is the shear
strength reduction technique (SSR). It’s a well known fact that for a simple slope factor of safety (FS) obtained from
SSR is usually the same as FS obtained from LEM. However for slopes of complex geology, considerable differ-
ences between FS values may be expected. Application of SSR for such slopes is usually restricted to the weakest
link estimation – that part of the slope with the lowest FS. Finite Difference Method code, FLAC (Itasca 2000), gives
the opportunity to analyze several slip surfaces by using the modified SSR technique (MSSR). The method is based
on reducing shear properties of soils after identification of the first slip surface. MSSR allows a complete estima-
tion of stability for any type of slope.
25 m
99
2 2
1.9
1.9
1.8
1.7
1.8
FS
1.6
FS
1.7 1.5
1.4
1.6
1.3 Weak layer 5 m thick
Weak layer 1 m thick
FLAC
FLAC
Fellenius
1.2 Fellenius
1.5 Bishop
Bishop
1.1 Janbu
Janbu
1.4 1
0 10 20 30 40 0 10 20 30 40 50
Distance of weak layer from slope crest
Distance of weak layer from slope crest
100
of the slope with the lowest FS. However the Finite classical SSR technique – especially with application
Difference Method code FLAC gives the opportunity of any Finite Element Method code. However FLAC is
to analyze several slip surfaces using modified shear created especially for modeling physical instability (in
strength reduction technique – MSSR (Cala & Flisiak this case – physical instabilities would be better term).
2003a, b). This allows to continue shear strength reduction
This method is based on reducing shear properties of and to identify another possible slip surfaces. In ana-
soils after identification of first slip surface (FS1). It lyzed case, next identified failure surface is located
is simply the continuation of classic SSR, but after in the upper part of the slope. FS2 1.00 calculated
first instability occurrence. It is possible only using by MSSR is again very close to FS 1.008 given by
Finite Difference Method. The FLAC program uses Bishop’s method. And finally application of MSSR
the explicit, Lagrangian calculation scheme. The full allowed to evaluate FS for entire slope – FS3 1.24 is
dynamic equations of motion are used, even when also very close to FS 1.228 given by Bishop’s
modeling systems that are essentially static. This method.
enables FLAC to follow physically unstable processes It seems that FS calculated with MSSR are within a
(i.e. several processes simultaneously) without numer- few percent of the FS obtained from LEM for simple
ical distress. In fact, FLAC is most effective when cases. It must be however underlined that effectiveness
applied to nonlinear or large-strain problems, or to sit- of MSSR must be verified on real cases.
uations in which physical instability may occur. This
may lead to identification of several other slip sur-
3.2 Large scale, complex geology slope
faces. The same criterion is used to identify secondary
stability case
(and further) failure surfaces. The primary and the fol-
lowing identified failure modes are constantly active Let’s consider a slope consisted of eight different geo-
(not suppressed) during entire calculation process. logical units (from a Polish lignite open pit mine).
Let’s consider benched slope stability (Fig. 5). The mechanical properties of the soil units involved
Figure 6 shows the slip surfaces identified in in the slope are given in Table 1.
benched slope by MSSR and LEM. Failure of the lower Figure 7 shows geometry and geology of the ana-
part of the slope was detected first. FS1 0.90 cal- lyzed slope. The overall sloping angle was equal
culated by SSR is very close to FS 0.921 given by ! 7.477°.
Bishop’s method. And precisely here ends the range of Figure 8 presents the slip surface identified by
MSSR and LEM. Again SSR finds the location of the
lowest safety factor FS1 0.67. Application of MSSR
identifies four new slip surfaces in several parts of the
10 m
40º slope. FS2 0.87 also shows the local failure surface
which, in fact, does not affect the overall slope stabil-
10 m
3
= 20 kN/m
45º
= 20º ity (precisely like previous one). Another possible
C = 10 kPa
failure surface with FS3 1.02 is based on layer 5
(very thin and weak one) and broken line upward.
15 m
101
2 3 4
168 m
63 m 5 6
7 8
800 m
FS5 = 1.29
Figure 8. FS values and critical slip surfaces identified with MSSR and LEM.
102
C. Peybernes
Electricité de France, Centre d’Ingénierie Hydraulique, France
ABSTRACT: The construction and first filling of a 150 m high zoned dam are modeled with FLAC. The site is
a deep and curved canyon under the dam. The aim of this study is the understanding of the dam behavior and the
assessment of the dam safety. A lot of attention is put on the hydraulic fracturing risk during construction or during
first filling due to the core arching in the deep canyon. 2D and 3D models are compared. The 2D model is unable
to explain the monitoring data, but the 3D model fits more accurately the monitoring measurement. Although a
high contrast of modulus exists between core and shell, no hydraulic fracturing is observed in the core.
103
– Settlements in the core measured at the end of the 4 CONSTITUTIVE EQUATIONS FOR SOIL
construction (up to 80 cm).
– total stresses in the core Two different constitutive models are used:
– pore water pressures in the core
– deformations in the downstream shoulder recorded 1. elastic model, for first parameter setting,
by three tassometers and two extensometers. 2. plastic law: Mohr-Coulomb.
5 SCENARIO OF LOADING
5.2 Construction
The construction period is 4 years and the embankment
Figure 3. Global sight of the model. is placed in 4 m layers.
Figure 5. Section right bank – left bank. Figure 6. Boundary conditions to construction.
104
Two storage versus time curves are used: one for 6 COMPARISON OF 2D AND 3D MODELS
the downstream shell and one for the upstream storage
(Fig. 7). In this section, the results of the model analyses are
compared to measurements at the end of construction
5.3.2 Mechanical boundary conditions (Fig. 9).
On the upstream and downstream faces of the dam,
mechanical pressure caused by the impounding is
modeled by normal stresses on the dam faces. The 6.1 2D Model
specific weights of materials are modified when they
are saturated. The calculated stresses are too high at the base of
the core compared to the measured ones. A parametric
First filling dates
study of mechanical properties could not solve the dis-
crepancy. It is speculated that the arching effect unloads
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
1000
the central section and transfers stresses to the banks.
No realistic calculation could be reached using the 2D
980 model.
960
cote (m)
940
6.2 3D Model
920
For this model, the calculated settlements in the core
900 are in good agreement with the measured ones. The
stresses are smaller than the 2D problem, because of
880
upstream downstream arching. Nevertheless, the deformation of the shells is
still larger than measured. A parametric study of core
and shell moduli was undertaken to reconcile the
Figure 7. Curves of upstream and downstream fillings. discrepancy.
105
106
8 3D MODEL FOR FIRST FILLING CASE two reservoirs, and finally a steady state was reached
(Fig. 14).
8.1 Pore pressures The values from the simulation and from measure-
The pore pressures calculated in the core at the level ments are rather close until the date 1981 for the
925 are close to the measured values (Fig. 13). It cells to the level 925, and 1983 for the cells of level
was difficult to get agreement between the measured 905 (Figs. 15 & 16). These dates correspond to the
and calculated values below this level. Variations of sudden drop of measured stress values. Then the sim-
permeability in the core were not represented in the ulations and measured values disagree. It is specu-
modeling. Spatial variation of the hydraulic properties lated that this phenomenon was induced by water
in the core may improve the correlation, but this was infiltration at the cell level, collapsing the clay and
not done. lowering the stress. This phenomenon should be inte-
grated in the future and modeled by the clay collapse
8.2 Stresses after wetting.
Some hydraulic fracturing can be observed in the
Generally the shapes of the calculated and monitored upstream shell, but this is not of concern. The shell is
stresses versus time were similar. The construction drained and the water tightness of the core is not
phase is apparent in the plots, then the filling of the altered.
420
380
340
300
260
220
180
140
100
60
20
-20
01/01/76 31/12/77 01/01/80 31/12/81 01/01/84 31/12/85 01/01/88 31/12/89 01/01/92 31/12/93 01/01/96 31/12/97 01/01/00 31/12/01
125-E11-905 E11-calcul elastique
125-E15-914 E15-calcul elastique
Figure 13. Pore pressures in kPa in the core (calculation: full features).
107
31/12/77
01/01/80
31/12/81
01/01/84
31/12/85
01/01/88
31/12/89
01/01/92
31/12/93
01/01/96
31/12/97
01/01/00
31/12/01
01/01/76
31/12/77
01/01/80
31/12/81
01/01/84
31/12/85
01/01/88
31/12/89
01/01/92
31/12/93
01/01/96
31/12/97
01/01/00
9 CONCLUSION
according to this calibration, hydraulic fracturing of
Detection of hydraulic fracturing in the core is the core was not indicated.
the objective of the analysis of this zoned dam. The
three-dimensional modeling appears to be the only
way to model the problem. The real geometry of the REFERENCES
foundation has to be carefully modeled to represent
the phenomenon of stresses transferred to the banks Varona P. 2001. Curso de FLAC3D, Itasca Consultant Spain,
and the unloading of the core in the canyon. January 15–19, 2001.
The parametric study of the mechanical properties Laigle, F. & Boymond, B. 2001. CERN-LHC Project –
of the materials was required to fit the monitoring Design and excavation of Large-Span Caverns at Point 1,
data and ensure the accuracy of the analysis. Finally, EDF-CIH France.
108
ABSTRACT: Results of FLAC’s strength reduction technique are compared with conventional limit equilib-
rium analysis for both linear and non-linear strength criteria. The comparison includes both safety factors and
failure modes (critical slip surfaces and normal stress functions). The collection of FLAC’s plastic points is not
a reasonable criterion for estimating the potential failure zone and it is necessary to establish this zone by post-
processing FLAC’s results. It is shown that failure modes implied by FLAC analysis are sensitive to mesh geom-
etry effects and, in order to obtain reasonable results, it is necessary to use meshes consisting of nearly square
elements. Safety factors are much less sensitive to mesh geometry effects than failure modes. FLAC’s mesh sen-
sitivity is more pronounced for non-linear failure criterion than in the linear case. Using acceptable mesh geom-
etry, FLAC’s strength reduction technique and limit equilibrium procedures yield comparable results (failure
modes and safety factors) for both linear and non-linear strength criteria. Engineering implications of linear and
non-linear failure criteria in the context of slope stability analysis are presented and discussed. It is shown that
equally valid interpretations of the same experimental information may, under certain conditions (e.g. steep
slopes), lead to very different engineering implications. Under such conditions the choice between alternative
strength models must be based on the practical implications of these laws.
109
110
4 FLAC IMPLEMENTATION OF
H-B CRITERION
(4.2)
111
compared with FMC 1.95). The previous discus- the collection of all such elements represents a L-E
sion showed that these two strength models are definition of the critical slip surface function implied
supported equally well by the available experimen- by the S-R technique. Mesh elements defined by the
tal information (Fig. 1). Faced with a situation in above process are shown as the open circles in
which two material models are equally justified by Figures 2 & 3, and critical slip surfaces defined by
the data, the choice between these models must be this process are shown as the heavy dashed lines in
based on their engineering consequences. In the those figures. In principle, the above identification of
present problem the H-B model delivered smaller failed elements is not different from FLAC’s defini-
safety factors than M-C and this model should be tion of plastic points. Nevertheless the set of failed
used in order to ensure a safe design. elements shown in Figure 3b is quite different from
2. For the M-C criterion the two meshes result in prac- the set of FLAC plastic points. The source of this
tically the same safety factor. The corresponding difference is probably related to an internal pro-
difference for the H-B model (F 0.1) is small, gramming detail in the FLAC program. More detailed
but it is not negligible. investigation appears to indicate that the internal
3. Figures 2 & 3 show that the square and inclined FLAC criterion used for definition of plastic points
meshes resulted in very different failure mecha- employs a too-large tolerance in the definition of
nisms. This difference is seen in terms of both dis- these points, resulting therefore with inclusion of ele-
tribution of FLAC’s plastic point, and contours of ments which are not really at failure. Controlling the
shear strain increments. Plastic points identify ele- accuracy with which the tangency requirement is
ments, which are at yield (failure). However yield- enforced provides a convenient numerical mechanism
ing of a particular element does not imply that the eliminating at least some failed elements, which are
slope as a whole is at failure. Consequently, the not relevant for definition of global slope failure (crit-
collection of FLAC’s plastic points does not pro- ical slip surface).
vide a clear indication of the global failure mecha- The following comments are relevant with respect
nism. In particular, some of the plastic points for to the process of identifying failed elements:
the inclined mesh are located deep in the interior of
1. Inferred critical slip surfaces defined by the above
the slope, and they are obviously not relevant for the
process are consistent with the shear strain incre-
purpose of identifying the slope’s failure mecha-
ment contours shown in Figures 2 & 3, but they pro-
nism. The distribution of plastic points and shear
vide a clearer definition of the global failure mode.
strain increments in the square meshes appears to
2. In some cases the set of failed elements includes a
indicate that very large area of the slope is in a
group of elements located in the vicinity of the
state of simultaneous failure. The inclined meshes
high entry point of the critical slip surface. This
imply failure mechanisms of the type postulated in
group represents elements failing in tension rather
conventional L-E calculations; namely an essentially
than shear. The L-E critical slip surface is not well
rigid body sliding along a narrow transition zone.
defined in such zones.
It is noted however that the “critical slip surface”
3. Tangency points between Mohr circles and the
implied by Figure 3b emerges above the toe of the
mobilized strength envelope define the normal
slope. From a L-E perspective, such a surface can-
stress acting on the critical slip surface passing
not be critical, corresponding essentially to a slope
through a given element. Consequently, the above
with a “reduced height”. Yet, this “unreasonable”
process results with L-E definition of both critical
slip surface is associated with a smaller safety fac-
slip surfaces and normal stress functions.
tor than the reasonable (but ill-defined) critical slip
surface in Figure 3a. The inferred L-E critical slip surfaces and normal
stress functions resulting from the above process are
shown in Figures 4 & 5, which correspond to Figures
6 FAILURE MODES IMPLIED BY FLAC’S 2 & 3 respectively. In those figures we have super-
S-R TECHNIQUE imposed also critical slip surfaces and normal stresses
functions resulting from the following approximate
Following a FLAC run, the state of stress (Mohr circles) L-E analyses:
is known at each element of the mesh. The state of
stress in failed elements satisfies Equation 1, and such 1. Simplified Bishop’s method. The original formula-
stress states are represented by Mohr circles, which tion of this procedure was based on the linear M-C
are tangential to the mobilized strength envelope. The strength functions. For the present purpose we
tangency requirement has to be satisfied with a cer- have modified this classical procedure incorporat-
tain tolerance in order to prevent exclusion of all ele- ing in it also the H-B criterion.
ments. Each tangential Mohr circle is associated with 2. The local linear approximation (LLA) technique
a certain mesh element, which can be identified; and presented by Baker (2003b). This approximation is
112
113
114
115
REFERENCES Griffith, D.V. & Lane, P.A. 1999. Slope Stability analysis by
finite elements. Geotechnique, 49(3): 387–403.
Baker, R. & Garber, M. 1978. Theoretical analysis of the sta- Hoek, E. & Brown, E.T. 1980. Empirical strength criterion
bility of slopes. Geotechnique, 28(4): 395–411. for rock masses. ASCE Jour. Geotech. Eng., 106(9):
Baker, R. 2003a. Inter-Relation between experimental and 1013–1035.
computational aspects of slope stability analysis. Inter. Jiang, J.C., Baker, R. & Yamagami, T. 2003. The effect of
Jour. Numer. Anal. Meth. Geamech. 27(5): 379–401. strength envelope nonlinearity on slope stability compu-
Baker, R. 2003b. Non-linear strength envelopes based on tri- tations. Can. Geoteh. J., 40(2): 308–325.
axial data. Accepted for publication in J. Geotech. And Matsui, T. & San, K.C. 1992. Finite element slope stability
Geoenvir Engrg., ASCE. analysis by shear reduction technique. Soils and
Baker, R. 2003c. Sufficient conditions for existence of phys- Foundations, 32(1): 59–70.
ically significant solutions in limiting equilibrium slope Maximovic, M. 1979. Limit equilibrium for non-linear fail-
stability analysis. Accepted for publication in Inter. Jour. ure envelope and arbitrary slip surface. Proc. 3rd Intr.
of Solids and structures. Conf. on Numerical Methods in Geomechanics, 769–777.
Bishop, A.W., Webb, D.L. & Lewin, P.I. 1965. Undisturbed Maximovic, M. 1989. Nonlinear failure criterion for soils.
samples of London clay from the Ashford common shaft: J. Geoth. Eng. ASCE, 115(4): 581–586.
strength-effective normal stress relationship. Geotech- Naylor, D.J. 1962. Finite element and slope stability. Nume,
nique, 15(1): 1–31. Meth. un Geomech., Proc. NATO advanced study institute.
Charles, J.A. & Soares, M.M. 1984. The stability of slopes Lisbon, Portugal, 229–244.
with nonlinear failure criterion. Cand. Geoth. J., 21(3): Penman, A. 1953. Shear characteristics of saturated silt in
397–406. triaxial compression. Geotechnique. 15(1): 79–93.
Day, R.W. & Axten, G.W. 1989. Surficial stability of com- Perry, J.A. 1994. A technique for defining non-linear shear
pacted clay slopes. J. Geoth. Eng. ASCE, 115(4): 577–580. strength envelopes and their incorporation in slope stabil-
Dawson, B.M., Roth, W.H. & Drescher, A. 1999. Slope sta- ity method of analysis. Quart. J. of Eng. Geology, 27(5):
bility factors of safety by strength reduction. Geotechnique, 231–241.
49(6): 835–840. Ugai, K. 1989. A method of calculation of total factor of
Donald, I.B. & Giam, S.K. 1988. Application of nodal dis- safety slopes by elasto-plastic FEM. Soils and Foundations,
placement method to slope stability analysis. Proc. 5th 29(2): 190–195.
Australia-New Zealand Conf. on Geomech., Sydney, Ugai, K. & Leshchinsky, D. 1995. Three-dimensional limit
Australia, 456–460. equilibrium and finite element analyses: a comparison of
Frydman, S. & Samoocha, Y. 1984. Laboratory studies on results, Soils and Foundations, 35(4): 1–7.
Israeli clays for reservoir embankment design. Proc. 5th Zienkiewicz, O.C., Humpheson, C. & Lewis, R.W. 1975.
Inter. Conf. on Expansive soils, Adelaide, South Associated and non-associated visco-plasticity and plas-
Australia, 94–98. ticity in soil mechanics. Geotechnique, 25(4): 671–689.
116
ABSTRACT: The Boinás East open pit mine has been exploited by the gold producer Río Narcea Gold Mines
in Northwest of Spain. Considering the influence of the radius of curvature on the factor of safety for slope
angle, and taking into account that a small increase in the overall angle will result in a very high increment in
the total amount of the ore mined, the slope stability analysis was made using the finite difference code FLAC3D
to calculate the factor of safety by reducing the rock shear strength. Due to the existing complicated geology
and the complexity of the 3D geometrical modeling, a “FISH routine” was used to import the block model of
the mine into the FLAC3D program. This block model is the database commonly used in the mine works and was
generated with Datamine. This method is a very good tool to generate a complex model in FLAC3D.
117
rocks, to temperatures between 150°C and 250°C and Table 1. Feasibility study data.
smaller pressure of 0.2 kbar. This process gives rise to
the formation of oxidized and very brecciated materi- Overall slope angle(°) Factor of safety
als, with bad geotechnical quality and that approxi-
mately constitute 90% of the operated material in the 55 0.91
50 1.03
open pits.
45 1.15
The gold mineralization remained hidden by Tertiary
lacustrine deposits, which as well were partially hidden
by Alpine thrusts that placed an important repetition
of limestone of the Láncara Formation above the The analyses were carried out using the XSTABL
Tertiary and the sandstone of the Oville Formation. software over sections with simplified geology and
Figure 1 shows a scheme of the zone of the deposit El considering different overall slope angles. Considering
Valle-Boinás and Figure 2 a scheme of the whole Rio the slope totally drained, 330 meters of vertical height,
Narcea Gold Belt. the Bishop’s method and circular surfaces of failure,
were obtained the factor of safety showed in the Table 1
(SRK 1996, 1999).
3 PRELIMINARY ANALYSIS The geology and the rock mass properties used by
SRK in their analysis are summarized in Table 2.
From the geotechnical data of the different materials
taken in from the geological exploration holes, the
shear strength of the different lithologies were deter-
mined following the Bieniawski (1989) classification 4 NUMERICAL MODELING WITH FLAC
and the Hoek-Brown 99 methodology.
With the obtained values SRK Ltd. carried out the The holes drilled during the year 1998 defined a new
feasibility study in October 1996, updating it in 1999. mineralized zone amenable to extraction by open pit
118
Table 2. Rock mass properties in the SRK model. Table 3. Factor of safety obtained with FLAC.
K G Cohesion Density
Lithology (GPa) (GPa) (°) (kPa) (t/m3)
119
(2)
conical and the different lithological units have a very
marked dip towards the East.
For the final design of the stable slope in the The reduction in the shear strength properties is made
Boinás East pit, the FLAC3D software (Itasca 1997) simultaneously for all materials.
120
121
Face angle
Deepest Type of Factor
Phase A1 B2 level Program analysis of safety
5 RESULTS AND CONCLUSIONS It must be take into account that with this system
we are going to have free faces at 90º and the high of
Once analyzed the different proposed models, it was the block size, so we must confirm that the factor of
proposed an open pit with lower level at 340 meters safety obtained corresponds to the slope factor and
ASL, a maximum slope height (in the East wall) of not to the brick face.
345 meters. A factor of safety of 1.45 was obtained
with slope angles of 70° in the dolomite above the main ACKNOWLEDGEMENTS
thrust, 60° in upper sandstone, fresh granite, fresh
skarn, tertiary, and black skarn, and 50° in the rest. The authors would like to thank Mr. Manuel G.
A comparison between the factors of safety Fernández of Río Narcea Gold Mines for his great
obtained for the different models and programs used work with Datamine, and Mr. Alan Riles, COO of Río
are provided in Table 5. Narcea Gold Mines Ltd. for the valuable help in the
The minimum factor of safety values are obtained translation of the paper. Finally the authors are grate-
in the East wall of the pit. Figure 9 shows the FLAC3D ful to Mr. Pedro Varona of Itasca Consultores, S.L. for
model with the zone of maximum displacement. These his technical support. Thanks are extended also to Mr.
zone correspond with a convex geometry in the slope, Pedro Velasco and Ms. Montserrat Senís for their help.
therefore steeper slopes could have been considered
in areas where the pit slopes were concaves, but these
was not our case. We only have changed the face angle REFERENCES
according with the geology and not with its geometry.
As we can see in the factor of safety obtained in the Cepedal, A., Martin-Izard, A., Fuertes, M., Pevida, L.,
different model, the influence of the radii of curvature Maldonado, C., Spiering, E., Gonzalez, S. & Varela, A.
in the global stability of a pit is very considerable, 1998. Fluid Inclusions and Hydrothermal Evolution of
specially taking into account that a small increasing the El Valle-Boinas Copper-gold Deposits. In Arias, A.,
in the overall face angle results in a very large amount Martin-Izard, A. & Paniagua, A. (eds), Gold Exploration
of ore recovered, as it was in our case. and mining in NW Spain: 50–58. Oviedo.
On the other hand, the hard work required to design Cepedal, M.A. 2001. Geología, Mineralogía, Evolución y
a model for a complex deposit in three dimensions is Modelo Genético del yacimiento de Au-Cu de El Valle-
Boinas. Belmonte (Asturias). Ph.D. thesis, University of
avoided when we import the block model into FLAC3D.
Oviedo.
With the routine described above, it is easy to create a Itasca Consulting Group, Inc. 1997. FLAC3D (Fast Lagrangian
block model with DATAMINE, or whatever other Analysis of Continua in 3 Dimensions), Version 2.0.
program, and delete the block above the surface we Minneapolis: Itasca.
want to analyze, and them import all the model to Itasca Consulting Group, Inc. 1998. FLAC (Fast Lagrangian
FLAC3D and obtain a factor of safety. Analysis of Continua), Version 3.4. Minneapolis: Itasca.
122
new
def creamalla
array aa(11488);(number of lines in file *.txt)
status open(‘BE16x16a.txt’,0,1)
status read(aa, 11488)
loop k(2, 11488); no lee la primera línea
xx parse(aa(k),1)
yy parse(aa(k),2)
zz parse(aa(k),3)
xxmax max(xx,xxmax)
mat parse(aa(k),4)
xc (xx-1)*16
662;put blocks in x local
yc (yy-1)*16
9745;put blocks in y local
zc (zz-1)*16
x0 xc
x1 xc
16
x2 xc
x3 xc
x4 xc
16
x5 xc
x6 xc
16
x7 xc
16
y0 yc
y1 yc
y2 yc
16
y3 yc
y4 yc
16
y5 yc
16
y6 yc
y7 yc
16
z0 zc
123
ABSTRACT: Rock behavior around tunnels excavated in the same rock with various inclinations and “k” ratios
has been investigated by numerical analysis employing the FLAC3D finite difference code. Stress distributions and
displacements at the critical points of the tunnels in underground mining have been examined. Observed stress and
displacement values with respect to the change in tunnel inclination and “k” ratio have been analyzed by using sta-
tistical methods employing “multiple regression analysis” in order to find out a meaningful correlation between the
stress, displacement values and the inclination and “k” ratios. Statistical analyses have presented meaningful corre-
lations giving mathematical equations whose dependent variable is displacement or stress and independent variables
are tunnel inclination and “k” ratio.
127
Geomechanical
properties Values Rating
Property Values
128
0.2
0.0
0
5
-1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2
10
5
1. 15 Observed Values (m)
20
25
0
30
1.
35
Incline (o)
40
-0.2
Predicted Values (m)
-0.4
-0.6
0.0
-0.8
-0.3
-0.6 -1.0
-0.9 -1.2
Zdisprf
1.2
(m) - -1.4
-1.5 -1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0
-1.8 Observed Values (m)
0.5
50
45
1.0
35
30
25
k-ratio 1.5
20
15
Incline (˚)
10
2.0
5
129
0.7
0.6
0.5
2.0
3 0 Observed Values (m)
3 5 1 .5
2 0
Incline (˚) 2 15 k-ratio
1 .0
0.6
Predicted Values (m)
0.5
0.4
1.8
1.5 0.3
1.2
Zdispfl 0.2
(m) 0.9
0.6 0.1
0.1 0.2 0.3 0.4 0.5 0.6 0.7
0.3
Observed Values (m)
50
45 0
4 35 Figure 10. Observed vs predicted values in z-direction
2.0
30 5
2 0 (floor).
1.5
Incline (˚) 2 5
1 0
1.0
1 k-ratio
5
5.0
0
where Ydispfl y-displacement at the floor; k “k”
Figure 8. z displacements vs “k” ratio and inclination (floor). ratio and I inclination of the tunnel (°).
The predicted values versus observed values in the
floor in the y-direction are shown in Figure 9. This
of “k” ratio and tunnel inclination. As can be seen regression model shows a strong correlation between
from Figure 7, the horizontal (y) displacements the observed and predicted values (r 0.99).
increase with an increase in both the “k” ratio and The regression equation proposed for predicting
tunnel inclination. the z displacements in the floor is as follows:
Figure 8 shows that the magnitude of the vertical
displacements in the floor increased with an increase
in the “k” ratio. On the contrary, the vertical displace-
ments decrease as the tunnel inclination increases.(4)
The regression equation proposed for predicting
the displacements in the y-direction in the floor is where Zdispfl z-displacement at the floor; k “k”
given below: ratio and I inclination of the tunnel (°).
The predicted values versus observed values in the
floor in the z-direction are shown in Figure 10. This
regression model also shows a strong correlation
(3) between the observed and predicted values (r 0.99).
130
0.0 0.60
-0.2 0.45
.4 Zdispsdw 30
Xdispsdw -0 0.
(m)
(m) -0.6
0.15
-0.8
4 50 50 5
4 5
0.5 35 0 4 0
4 5 2.0
1.0 2530 3 0
1.5
2 3 5
k-ratio 1.5 1 0 2 0
10 5 Incline (˚) Incline (˚) 2 15 0 1.0
5
2.0 0 1 5 0.5 k-ratio
0
0.0
-0.1
0.20
Predicted Values (m)
0.15 -0.2
0.10
Ydispsdw -0.3
(m) 0.05
-0.4
0.00
-0.05 -0.5
50 5
4 0
4 35
2.0
-0.6
30 5 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0
1.5
2 0
2 Observed Values (m)
Incline (˚) 1510
1.0
5
k-ratio
0
0.
(sidewall).
Figure 12. y displacements vs “k” ratio and inclination
(sidewall).
131
The regression model recommended to predict the where Zdispsdw z-displacement at the sidewall;
y displacements in the sidewall is given below: k “k” ratio and I inclination of the tunnel (°).
The predicted values versus observed values in the
sidewall in the z-direction are shown in Figure 16.
This regression model also shows a strong correlation
(6) between the observed and predicted values (r 0.99).
The correlation coefficients of all the proposed
where Ydispsdw y-displacement at the sidewall; regression models for displacements have been sum-
k “k” ratio and I inclination of the tunnel (°). marized in Table 3.
The predicted values versus observed values in the
sidewall in the y-direction are shown in Figure 15. 3.2 Evaluation of models in terms of stresses
This regression model shows a strong correlation As the observed principal stresses are examined, no
between the observed and predicted values (r 0.99). strong relationship between the principal stresses and
The regression model proposed for predicting the the “k” ratio or the tunnel inclination has been found
z displacements in the sidewall is as follows: except for the principal stresses in the tunnel floor.
Therefore, only principal stresses in the floor are taken
into consideration.
(7) The graphs of maximum and minimum principal
stresses are given in Figures 17 and 18 respectively.
The regression equation proposed for predicting
0.20 the magnitude of 1 in the floor and the correlation
coefficient are given below:
0.15
Predicted Values (m)
0.10
(8)
0.05
r 0.95
0.00
Table 3. Correlation coefficients for displacements.
-0.05
Correlation coefficient (r)
-0.10
-0.10 -0.05 0.00 0.05 0.10 0.15 0.20 Direction Roof Floor Sidewall
Observed Values (m)
x – – 0.99
Figure 15. Observed vs predicted values in y-direction y 0.99 0.99 0.99
(sidewall). z 0.98 0.99 0.99
0.5
0.4
Predicted Values (m)
0.3 -315
-330
0.2 -345
1 fl (kPa) 60
-3
0.1 -375
0
5
2.0 1510
0.0 2
1.5 25 0
0.0 0.1 0.2 0.3 0.4 0.5
k-ratio 1.0 3 30
Observed Values (m) 40 5 Incline (˚)
4
50 5
0.5
Figure 16. Observed vs predicted values in z-direction
(sidewall). Figure 17. 1 vs “k” ratio and inclination (floor).
132
-10
-15
-360
The correlation coefficients of all proposed regres-
-380 sion models have been summarized in Table 4 for
maximum and minimum principal stresses. As seen,
-400 there is only a meaningful correlation for the history
-400 -380 -360 -340 -320 -300
point in the floor.
Observed Values (kPa)
133
134
ABSTRACT: Shallow tunneling performed by a Tunnel Boring Machine (TBM) induces volume loss, mainly
due to the conical shape of the machine and the consolidation of the injected grout. This excavation volume loss
causes ground movements at the surface that can induce damages to surrounding structures. However, to know
the influence on structures, it is not sufficient to apply the Greenfield strains because of the influence of the
structure’s stiffness. Therefore a computational three-dimensional soil-structure interaction analysis is required
to take into account all the complexity of the problem. This paper presents a FLAC3D analysis of the interaction
between shallow tunnel excavation and surface buildings, applied to the case of the Lisboa subway. A simpli-
fied simulation of TBM tunneling is adopted by imposing volume loss. To highlight the influence of settlements
on a six-floor structure, different cases of volume loss are studied from 0.5 to 5 % of the total excavated volume.
135
136
0
Step n
Z
Surface settlements (cm)
-0.05
-0.1
20 m
Y
-0.15
-0.2
-0.25
-0.3 Tunnel face
-0.35
-20 -15 -10 -5 0 5 10 15 20
Distance to tunnel axis (m) Step n+1
20 m
Figure 3. Transverse settlement trough.
phases induced by the boring machine. The hypothe- 20 110 m which corresponds to the entire model
ses are as follows: the soil displacements at the tunnel length added with the distance between the tunnel
face are blocked, simulating a perfect equilibrium of face and the position where the lateral displacement is
confinement pressures. The lateral soil displacements constant.
137
(1)
Distance to tunnel axis (x/D)
0 1 2 3 4 5 6 7 8 9
0.05
Where S settlement at distance x of the tunnel
axis, Smax maximum settlement obtained in the
Horizontal strain (%)
0.00
trough center, i distance from the inflection point
of the trough to tunnel axis. In this case, i 1.8 D. -0.05
-0.10
4.2 Horizontal surface displacement
Figure 7 presents the horizontal surface soil strains in -0.15
a transverse section, at the final state. In this figure, compression extension
-1
-1 Vt
-1.5
0.5%
0.5%
-2
-2 1%
1%
-2.5
3%
3%
-3
5%
5%
-3 -3.5
-4
-4 -4.5
138
5 D/2
Maximum settlement (cm)
Tunnel
4 18m
Smax = 0.82Vt
4m
3
4m
2 Y
1 12m
0 X
0 1 2 3 4 5
Tunnel volume loss (%)
0.00
Vt
-0.05
5%
-0.10 3%
1%
-0.15 0.5%
-0.20
139
The forces in the structural elements are not initial- Distance to tunnel axis (x/D)
ized: this step represents the initial state in terms of 0 1 2 3 4 5
internal forces in the structure. Finally, the obtained 0
Building
-2 Greenfield
140
60 20
Column
Longitudinal bending
50 My Tx A
moment (kN.m)
Mx Ty 15 B
40 C
F/Fini
D
30 10
20
10 5
0
0 10 20 30 40 50 60 70 80 90 100 110 0
Excavation length (m) 0 20 40 60 80 100
Excavation length (m)
28
Initial state
D Excavation of 45m
B Column height (m) 24 Excavation of 77m
Y Final state
20
16
X A
C 12
8
Tunnel axis 4
0
Figure 17. Position of studied columns. -20 -15 -10 -5 0 5 10
Longitudinal bending moment (kN.m)
40 A
moment (kN.m)
B
30 C
D The most affected column in both directions is col-
20 umn C. Bending moments and shear forces distribution
in these columns are analyzed for different excavation
10 lengths.
Figures 19 and 20 present the repartition of the
0 longitudinal forces in column C. The most prejudicial
0 2 40 60 80 100
excavation step corresponds to the excavation length
Excavation length (m)
of about 80 m. The maximum longitudinal bending
moment is 19 kN.m, reached between levels 3 and 4
Figure 18. Maximum transversal bending moment in the
(Fig. 19). For a length bored of 45 m, figure 19 shows
four columns.
that the column is affected in the opposite direction,
with a value of 7 kN.m between levels 2 and 3. The
maximum longitudinal shear force is equal to 2.8 kN,
bending moment is studied, because shear forces vary reached at level 5. Level 3 is also affected with
in the same way. TY 2.6 kN (Fig. 20).
Figure 18 presents the evolution of the maximum Figures 21 and 22 present the repartition of the
value of MY for the different studied columns. It can be transversal forces on column C. The most prejudicial
noted that more columns are far from the tunnel axis, excavation step is at the final state. The initial state is
more columns are affected in the transverse direction. not represented because of negligible values. This fig-
Figure 19 presents the evolution of the maximum ure show that the most affected levels in the trans-
value of Mx for the different studied columns. It verse direction are levels 1 and 2. The maximum
seems that the longitudinal stresses are similar in all value of MY is 44 kN.m (Fig. 21) and the maximum
the building’s columns. value of TX is 11 kN (Fig. 22).
141
28 28
Excavation of 45m
24 24
Column height (m)
Figure 21. Longitudinal shear force (TY) on column C. Figure 23. Transversal shear force (TX) on column C.
28
Distance to tunnel axis (x/D)
24 Excavation of 45m
0 1 2 3 4 5 6 7
Column height (m)
Final state 0
20 -0.5
16 Surface settlement (cm) -1
Vt
12 -1.5
-2 0.5%
8 -2.5 1%
4 -3 3%
-3.5 5%
0
-10 0 10 20 30 40 50 -4
Transverse bending moment (kN.m) -4.5
Building
-5
142
30 1%
5
4 0.5% The analysis of the volume loss influence showed a
3 linear relation between volume loss in tunnel and forces
2 induced in the structure columns during tunneling.
1
0
0 10 20 30 40 50 60 70 80 90 100 110
Excavation length (m) REFERENCES
Figure 27. Evolution of maximum longitudinal forces in AFTES 1999. Recommandations pour les tassements liés au
columns. creusement des ouvrages en souterrain. Tunnels et
Ouvrages Souterrains: 106–128.
Benmebarek, S., Kastner, R. & Ollier, C. 1998. Auscultation
60 et modélisation numérique du processus de creusement à
l’aide d’un tunnelier. Géotechnique 48 (6): 801–818.
50 Tx - final state Dias, D., Kastner, R. & Maghazi, M. 1999. Three dimensional
y = 1032x
40 Ty - excavation of 77 m simulation of slurry shield tunneling. In International
Symposium on Geotechnical Aspects of Underground Con-
F/Fini
143
Oteo, C.S. & Sagaseta, C. 1982. Prediction of settlements Sagaseta, C. 1987. Evaluation of surface movements above
due to underground openings. In Int. Symp. On numeri- tunnels, a new approach. In Colloque International ENPC
cal Models in Geomechanics, Zurich, 653–659. Interactions sol/structure, Paris, Presses ENPC, 445–452.
Panet, M. 1995. Le calcul des tunnels par la méthode Standing, J.R., Gras, M., Taylor, G.R., Gupta, S.C., Nyren,
convergence-confinement. Paris: Presses de l’ENPC. R.J. & Burland, J.B. 2002. Building response to tunnel
Peck, R.B. 1969. Deep excavation and tunnelling in soft step-plate junction construction – the former Lloyds
ground, State of the art report. In 7th International Bank building, St James’s, London. In 4th Symposium
Conference on Soil Mechanics and Foundation Engi- Geotechnical Aspects of Underground Construction in
neering, Mexico, 225–290. Soft Ground, Toulouse, France, 3.
Potts, D.M. & Addenbrooke, T.I. 1997. A structure’s influence Swoboda, G., Mertz, W. & Schmid, A. 1989. Three dimen-
on tunnelling induced ground movements. In Instn Civil sional numerical models to simulate tunnel excavation.
Engineers in Geotechnical Engineering. 125, 109–125. Numerical Models in Geomechanics NUMOG III.
Ribeiro e Sousa, L., Dias, D. & Barreto, J. 2003. Lisbon Elsevier. 581–586.
Metro Yellow Line extension. Structural behaviour of the
Ameixoeira Station. In 12ª Conferência Panamerican on
Soil Mechanics and Geotechnical Engineering, Boston.
144
ABSTRACT: For the safe design of a deep underground high-level radioactive waste (HLW) repository, it is
important to understand the thermal–mechanical behavior of the engineering barriers and rock mass around the
repository influenced by the high stress and the heat generated from the waste. In this study, thermal–mechani-
cal coupling analysis was carried out to investigate the reliability of the Korean HLW repository concept using
FLAC3D with the thermal and mechanical properties of rock and rock mass measured at two drilling sites. For
effective thermal–mechanical coupling, a FISH routine was developed and used for the modeling of different
conditions. By using FLAC3D with the FISH routine, the evaluation of the thermal–mechanical stability of the
preliminary disposal concept could be done successfully.
145
40m
6m
7m
6m
2.2m
BT
TB
Backfill
Buffer
Canister Outshell
Spent Fuel
Deposition
Hole
FT
material because of its low hydraulic conductivity, high Table 1. Material properties of fuel part and outshell.
sorption capacity, self-sealing characteristics, and
Unit Fuel part Outshell
durability in nature.
In the Korean repository concept, the mixture of Material type Fuel
Stainless
bentonite and crushed rock will be used as the back- cast iron steel
filling material. The deposition tunnels are 6 m wide Model type Elastic Elastic
and 7 m high. The canister containing spent fuel is E GPa 190 200
assumed to be emplaced in the vertical boreholes 0.3 0.3
drilled along the center line on the floor as shown in Density Kg/m3 6500 8000
Figure 1. Thermal conductivity W/m°K 43 15.2
Specific heat J/Kg°K 424 504
Thermal expansion /°K 1.2e-5 8.2e-6
3 FLAC3D MODELING
3.1 Materials in the model laboratory tests using Korean bentonite (Kyungju
3.1.1 Fuel part and outshell bentonite), which is considered as a candidate buffer
Four PWR assemblies are inserted in a canister with material for the Korean repository. The other material
outshell thickness of 5 cm. The mechanical and thermal properties, which could not be determined from tests,
properties of the fuel part, which represents the part were chosen from literature review and listed in Table 2.
inside of the outershell, were determined with the In this study, the buffer and backfill materials were
assumption of that the fuel and cast iron were uniformly modeled with a Drucker-Prager plastic model. The
mixed. The average properties of the fuel part were Drucker-Prager parameters in Eq. (1) for buffer and
calculated based on volume ratio and listed in Table 1. backfill could be determined from the triaxial com-
Among the candidate material types for the outshell, pression tests under different confining pressures.
stainless steel was considered in this study. The diam-
eter of the canister is 1.22 m and the length is 4.78 m. (1)
The thermal–mechanical properties of stainless steel
are also listed in Table 1.
where, ! and kshear are material parameters, J1 is
3.1.2 Buffer and backfill the first invariant of the stress tensor, and J2D is the
Some of the thermal and mechanical properties of second invariant of the deviatoric stress tensor
buffer and backfill material could be determined from (Desai & Siriwardane 1984).
146
Rock
3.1.3 Rock properties waste. Subsequent heating of the rock mass by the
It is assumed that the underground repository is con- heat-generating waste would increase the stresses in
structed in a granite body at 500 m below surface. The the buffer, canister, and rock mass because of thermal
mechanical and thermal properties of the granites expansion (Simmons & Baumgartner 1994). The
from two drilling sites, Kosung and Yusung, are listed thermal stress due to the thermal expansion can be
in Table 2. The two sites are representing the east and calculated as follows:
west sides of Korean Peninsula. NX size rock cores
were retrieved from the drill holes reached up to 500 m (4)
below surface.
The influence of discontinuities is considered indi-
where, is increase in stress due to the expansion
rectly using the equations proposed by Fossum (1985)
of rock, is thermal expansion coefficient, T
to calculated the modulus of randomly jointed rock
is temperature increase, E is Young’s modulus, and
mass. The effective bulk and shear moduli can be writ-
is Poisson’s ratio. FLAC3D has functions for coup-
ten in terms of the intact and joint properties. Effective
ling behaviors such as hydraulic–mechanical,
bulk and shear moduli are
thermal–mechanical, and thermal–hydraulic couplings.
In FLAC3D, the thermal–mechanical coupling occurs
(2) only in one direction: temperature changes cause ther-
mal strains to occur which influence the stresses, while
the thermal calculation is unaffected by the mechanical
changes taking place (Itasca 1996). As normal in most
modeling situations, the initial mechanical conditions
correspond to a state of equilibrium which must first
be achieved before the coupled analysis is started.
(3) There are the following three suggestions for ther-
mal–mechanical coupling in the FLAC3D manual.
where, E is Elastic modulus of rock, is Poisson’s 1. A thermal only calculation is performed until the
ratio, S is joint spacing, and kn and ks are normal and desired time and then the thermal calculation is to
shear stiffness of joint. Thermal logging was carried out be turned off and the mechanical calculation is per-
to find the geothermal gradients at the two sites (Park formed. When the mechanical equilibrium is
et al. 2001). reached, thermal calculation is performed again.
2. For each thermal time step, several mechanical
steps are taken until detecting equilibrium condition.
3.2 Modeling method
3. The STEP command is used while both mechanical
Thermal–mechanical coupling is important due to the and thermal modules are on. In this approach, one
thermal stress developed by the decay heat from the mechanical step will be taken for each thermal step.
147
The first approach is useful for thermal–mechanical In the model, 5 different materials, rock, buffer, back-
analysis of an elastic model. For non-linear models fill, outshell, and fuel part, were included. In the
such as plastic models, the thermal change must be Korean preliminary disposal concept, the backfill
communicated to the mechanical module at closer thickness L1 1 m, upper buffer thickness L2 1.5 m,
time intervals to respect the path dependency of the bottom buffer thickness and side buffer thickness
system. In this case, a certain number of mechanical L3 0.5 m.
steps are taken for each thermal step to allow the sys-
tem to adjust. In this approach, the transition from 3.3.2 Initial and boundary conditions
thermal to mechanical calculation is based on time The in situ stress was assumed to be hydrostatic in
instead of temperature variation. Since the heat gen- this study based on the fact that the stress ratios in
eration is varying with time, the transition based on Yusung and Kosung sites are more or less 1.0 at 500 m
temperature variation is more reasonable in the early depth. The initial temperature in the model was calcu-
stage of the repository. The second and third approaches lated with the geothermal gradients of Yusung and
may be more accurate than the first approach, but the Kosung sites. It was assumed that the average surface
problem is that the calculation will take a long time to temperature is 20°C.
model the long-term behavior of repository. In order
to overcome the disadvantages of the three approaches, 3.4 Decay heat
another technique for thermal–mechanical coupling Decay heat is the thermal energy resulting from the
was developed. In the new approach, the transition radioactive decay of the radioactive materials in the
from thermal to mechanical calculation is based on spent fuel discharged from reactors. In Korea, the PWR
temperature change. A FISH program for the new spent fuel with 45,000 MWd/tHM is now considered
approach was developed for PWR spent fuel. Figure 2 as the reference PWR spent fuel, because that type of
shows the flow chart of the thermal–mechanical cou-
pling adapted in this study.
Figure 2. Flow chart of the TM coupling. Figure 3. Model mesh around the tunnel and deposition hole.
148
spent fuel occupies 64% of all spent fuel from Korean be higher than 100°C and thus the disposal design can
reactors. Even though significant amount of CANDU satisfy the thermal criteria.
spent fuel with 7500 MWd/tHM is generated from Because of the higher geothermal gradient in
CANDU reactors, CANDU spent fuel is not considered Kosung area, the temperature around the repository in
in this study, because of its much lower burnup com- the Kosung case is higher than in the Yusung case.
pared to PWR spent fuel. For the spent fuel with 45,000 From Figure 4, it is possible to see when the transitions
MWd/tHM, the heat decays exponentially with time between mechanical and thermal steps had happened.
as following: At the 500 m deep location, the initial temperature in
the Kosung case was 38.75°C while the temperature
(5) in the Yusung case was 32.5°C. The initial tempera-
ture difference due to the difference in geothermal
where, t is time (year) after discharge from reactors. gradient was about 6°C. Table 3 lists the temperatures
at the checking points at 20 years and 200 years after
4 FLAC3D MODELING RESULTS the emplacement of canister. The difference in tem-
perature after 20 years in the Kosung and Yusung
4.1 Temperature cases ranges from 8 to 12.4°C, which is higher than
the initial temperature difference. The increase of
In order not to lose the required properties of ben- temperature difference is due to the lower thermal
tonite buffer, the maximum temperature at the canister conduction in Kosung case, which has lower thermal
surface and throughout the buffer must not exceed conductivity than that in Yusung case. With increase
100°C. As shown in Figure 4, the highest temperature in time, the temperature difference between the two
in the Kosung case was 92.97°C at 15 years after cases decreased and it was about 7–9°C at 200 years
emplacement, while it was about 80.71°C in 20 years after emplacement.
after emplacement in Yusung case. Since the highest
temperatures in the model in both cases are lower
than 100°C, the peak temperature in the buffer cannot 4.2 Displacement
In the deposition hole, the heat from the waste will
lead to thermal expansion of the canister, buffer, and
100
95
backfill. The displacements around the tunnel will also
Temperature (deg. C)
Table 3. Temperatures (°C) at the checking points and different time for Kosung and Yusung
cases.
149
5 CONCLUSIONS
150
Figure 7. Principal stress contours at the canister for Yusung case, 200 years after emplacement.
80
70 Kosung
Yusung
Mises stress (MPa)
60
50
40
30
20
10
0
1 2 3 4 5 6 7 8 9
Check Points
Figure 8. Comparison of von-Mises stress at different locations for Kosung and Yusung cases.
– From the fact that the peak temperature around the – FLAC3D with FISH function could be successfully
repository is reached in several tens of years after applied to evaluate the thermal–mechanical stabil-
the emplacement of canister, modeling up to sev- ity of the Korean preliminary repository design in
eral tens of years are good enough for the sensitiv- deep underground rock.
ity analysis, which is mainly for investigation the
relative influence of design parameters.
– The stress distribution on the canister surface REFERENCES
shows that the tensile stress is highest at lower part
of the canister. The maximum principal stresses Berge, P.A. & Wang, H.F. 1999. Thermomechanical Effects
were tensile and it was up to 26 MPa and 28 MPa in on Permeability for a 3-D Model of YM Rock, Bernard
Yusung and Kosung cases, respectively. Such a Amadei et al. (eds), Proceedings of the 37th U.S. Rock
tensile stress may induce catastrophic failure of the Mech. Symp., Vail, Colorado, June 1999, Vol. 2: 729–749.
outshell and thus needs to be carefully analyzed. Rotterdam: Balkema.
151
Desai, C.S. & Siriwardane, H.J. 1984. Constitutive laws for Park, J.H., Kuh, J.E. & Kang, C.H. 1998. An examination of
engineering materials, Prentice-Hall, Inc., Englewood thermal analysis capability of FLAC3D on the near field
Cliffs, NJ. of high level radioactive waste repository, KAERI/TR-
Fairhurst, C. 1999. Rock Mechanics and Nuclear Waste 1187/98, KAERI
Repositories. S. Saeb and C. Franke (eds), Proceedings Park, B.Y., Bae, D.S., Kim, C., Kim, K.S., Koh, Y.K. & Jeon,
of the International Workshop on the Rock Mechanics of S.W. 2001. Evaluation of the Basic Mechanical and
Nuclear Waste Repositories, Vail, Colorado, June 1999: Thermal Properties of Deep Crystalline Rocks, KAERI/TR-
1–43. Alexandria, Virginia: American Rock Mechanics 1828/2001, KAERI.
Association. Park, J.H., Kwon, S., Choi, J.W. & Kang, C.H. 2001.
Fookes, P.G. 1995. Aggregates: a review of prediction and Sensitivity analysis on mechanical stability of the under-
performance, Proceedings of STATS 21st Anniversary ground excavations for a high-level radioactive waste
conference, London, UK: 91–170. repository, KAERI/TR-1749/2001, KAERI.
Francke, C.T., Saeb, S. & Carrasco, R.C. 2001. Three- Patchet, S.J., Carrasco, R.C., Francke, C.T., Salari, R. &
Dimensional Analysis of Nuclear Waste Disposal in Saeb, S. 2001. Interaction Between Two Adjacent Panels
Horizontal Boreholes, Proceedings of the 38th U.S. Rock at WIPP,” in Rock Mechanics in the National Interest,
Mechanics Symposium, Washington, D.C., July 2001, Vol. 1: Proceedings of the 38th U.S. Rock Mechanics Symposium,
497–503. Lisse, The Netherlands: Swets & Zeitlinger B.V. Washington, D.C., July 2001), Vol.: 517–523. Lisse, The
Itasca Consulting Group, Inc. 1996. FLAC3D – Fast Netherlands: Swets & Zeitlinger B.V.
Lagrangian Analysis of Continua in Three-Dimensions, Selvadurai, P.S. & Pang, S. 1990. Mechanics of the interac-
Ver 1.1 User’s Manual. Minneapolis, MN: Itasca. tion between a nuclear waste disposal container and a
Johansson, E. & Hakala, M. 1995. Rock Mechanical Aspect buffer during discontinuous rock movement, Engineering
on the Critical Depth for a KBS-3 Type Repository Based Geology, Vol. 28: 405–417.
on Brittle Rock Strength Criterion Developed at URL in Simmons, G.R. & Baumgartner, P. 1994. The disposal of
Canada, SKB, AR D-95-014, SKB. Canada’s nuclear fuel waste: Engineering for a disposal
Kwon, Y.J., Kang, S.W. & Ha, J.Y. 2001. Mechanical struc- facility, AECL Research, AECL-10715, AECL.
tural stability analysis of spent nuclear fuel disposal
canister under the internal/external pressure variation,
KAERI/CM-440/2000, KAERI.
152
F. Laigle
Electricité de France, Hydro Engineering Centre, France
A. Saïtta
Centre d’Etudes des Tunnels, Lyon, France
ABSTRACT: In this paper, a numerical model is used to predict radial bolting performance in soft rocks crossed
by the Tartaiguille TGV (high speed railway) tunnel located on the new TGV French southeast line. First, a brief
description of results of field investigations and monitored displacements during excavation of this under-
ground tunnel are done. The support system is constituted by shotcrete and radial grouted bars. Because of too
high monitored displacements in some sections in these marls, a modification of the support system has been
decided on site. This modification consists in an increasing of the density of bars in a specific zone. Back analysis
confirmed the very strong efficiency of these additional bars in this case. However, conventional continuous
modeling of the tunnel done up to now, strongly underestimated the real contribution of these extra bolts. Some
new numerical simulations have been done using a constitutive model proposed by Laigle. This constitutive model
integrated in FLAC focuses on the post-peak behavior of rocks. It’s based on a simple and physical description of
the behavior of ground in this domain, with accurate evolutions of the cohesion and the dilatancy. This paper
describes numerical results obtained using this new constitutive model applied to the Tartaiguille tunnel case. The
significant monitored effect of additional grouted bars is well shown by this computation.
153
154
-40 S11 the bolts because this assures that the evolution of the
S12
S13
convergences does not result from a modification of the
-60 S14 ground.
S15
S16
-80 S17
S18
3 NUMERICAL SIMULATIONS
-100 S19
S20
S21
The aim of these numerical simulations was to simulate
-120 S22 in the framework of usual continuous modeling the
strong effect of a grouted bars system on the stability
-140 conditions of the Tartaiguille tunnel. The particularity
of this work is to use a new constitutive model devel-
Figure 3. Measurements of convergence of cords BC. oped by Laigle. These numerical simulations are done
using the elastoplastic version. A more general elasto-
70 plastic–viscoplastic version has been developed and
: Average convergence -46% presented in another paper (2003).
60 Following phenomena are studied in the frame-
Zone 3
Convergences (mm)
Figure 4. Measurements of the convergences at 30 m from In general, at least for deep and no-urban tunnels, the
the face. goal of a support system composed with shotcrete and
grouted bars is to prevent mechanical failure within a
So we interested ourselves, for each of the mea- sufficient safety level. In cohesive rocks, which can be
surement sections, in these values, but also in the con- considered as a continuous material, this failure is asso-
text in which they were obtained, that is, everything ciated with the development of a fracturing process
that could have an influence on the results of the induced by the excavation. So, it seems necessary to
measurements. focus the simulation of the mechanical behavior both
The following graphs present the results, which on the pre-peak behavior and the post-peak behavior.
come directly from the worksite of the convergences This new constitutive model has been written in the
of the cords BC for sections S07 to S22 (Fig. 3). framework of the elastoplastic theory The expression
of the yield surface corresponds to the generalized
Hoek and Brown criterion. This surface is governed
2.4 Study of the results of the upper half-section by 4 parameters, which are the unconfined compressive
The calculation of the average convergence at 30 m for strength and three other parameters “m”, “s” and “a”.
each of the zones (see Fig. 4) shows a reduction of 37% These 3 last parameters change in accordance with an
between Zone 1 and Zone 2, and 46% between Zone 3 internal variable p, which is the irreversible shear
and Zone 4. The increase in the number of bolts at the strain defined below:
sidewalls (from 1 bolt/3.3 m2 to 1 bolt/1.6 m2) therefore
resulted in a reduction of at least 37% in short-term hor-
izontal convergences of the BC cord. We may show that (1)
this reduction is directly due to the bolting. All the other
155
Deviatoric stress
(2)
post-peak domain
pre-peak domain
damage rock sample crossed by an induced shear Domain 4: Fractured rock
Volumetric strain
fracture. In these conditions, cohesion of the rock
at a macroscopic scale is assumed to be null.
Axial strain
– The last criterion corresponds to the residual strength
criterion, which is reached at very large shear strains.
Some specific hardening laws are suggested for each
parameter “m”, “s” and “a” allowing to describe the
evolution of the yield surface from one threshold to Figure 5. Schematic behavior of a rock sample during a
another (Laigle 2003). In the softening domain, the triaxial test.
negative hardening is assumed to be divided into three
phases:
5
– The first phase of softening corresponds to a dete-
4.5
Maximal principal stress (MPa)
3.2 Behavior of the tunnel without support 3.3 Behavior of the tunnel with initial support
A first simulation has been done without considering A second simulation has been done considering the ini-
any support system. The excavation is simulated by tial support system (Zone 1, ring of grouted bars every
156
Time (days)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
0
-1
Deconfinement: 97,8 %
Development of the
Figure 7. Physical state of rock without any support.
induced fractured
zone
Fractured rock
Strains of
bolts (Domain 4)
Intact rock (Domain1) Figure 10. Physical state of rock during the bench excava-
tion and failure mechanism.
Figure 8. Physical state of rock at the end of the upper-half convergences versus time, assuming an advancing rate
excavation. of 2 m/day. These simulations don’t take into account
the time. Convergences are drawn versus time only to
2 meters). Bars are simulated using structural cable compare with monitored values.
elements and a shotcrete layer using structural beam Without any modification of this initial support, the
elements. These bars and beams are installed after a excavation of the bench can’t be finalized. Because of
deconfinement of 70%, so approximately in the first the bench excavation, the fracturing phenomenon takes
2 meters behind the face heading. off again toward the roof. The initial support is not suf-
Bars are linked to some beam nodes with the goal ficient to stop the development of this damage zone.
to simulate face plates. Despite this, we will observe The stability cannot be demonstrated from this numer-
that the maximum tensile stain is located in the ical simulation (Fig. 10).
ground and not near the wall (Fig. 8) These results seem in accordance with the site
With this support, the stability of the tunnel during engineer’s decision to adapt the support system. Con-
the upper-half excavation can be theoretically justi- sidering high measured displacements and the devel-
fied, even if a damage zone exists near tunnel the foot opment of a local failure of the shotcrete layer at the
of the tunnel (Fig. 8). Figure 9 presents horizontal roof, he decided to reinforce this initial support system.
157
Time (days)
0 10 20 30 40 50 60 70 80
0
Excavation of vault Excavation of bench
-10
-25 Failure
Rock at residual state
(Domain 5) Figure 12. Influence of reinforced bolting pattern of hori-
Fissured rock in zontal convergences.
post-peak domain
(Domain 3)
Fissured rock
in the pre-peak shown by numerical computations. With the aim of
domain reinforcing the rock mass, the grouted bars system
(Domain 2) has to be installed as early as possible, before any crit-
Intact rock ical increasing of monitoring displacements. This is
(Domain 1) because the goal of this type of support is to protect
and help the rock to keep a sufficient shear strength to
ensure the global stability.
Figure 11. Physical state of rock at the end of the bench
excavation, considering a reinforced support system.
3.5 Effect of a local failure of the shotcrete lining
During excavation of the vault, a crack appeared and
3.4 Behavior of the tunnel with a reinforced
developed in the shotcrete at the crown of the
support
Tartaiguille tunnel. This same phenomenon has been
The reinforcement of the support system corresponds observed during excavation of one large cavern of
to the installation of additional bolts on sidewalls of the CERN-LHC project in Geneva (Laigle 2002).
the tunnel. In this zone, the bolting density is double. Depending on the support design, a shotcrete failure
In the framework of numerical simulations, several could be critical for the global stability of the tunnel.
patterns are considered: However, in these two previous cases, the support has
been designed with the consideration that one major
– 2 additional grouted bars are installed on each side-
component is the grouted bars system and not only
wall at 6 meters behind the face heading (90% of the
the shotcrete layer. If the shotcrete keeps an essential
vault deconfinement).
function, it can’t be assimilated in these cases to a
– 1 additional grouted bar is installed on each side wall
continuous shell like in the SCL approach.
as soon as possible, about 2 meters from the face
In the case of Tartaiguille tunnels, it was interest-
heading (70% of the vault deconfinement).
ing to know if this crack in the shotcrete was really
In these 2 last cases, it appears that theoretically, critical from a global stability point of view. A simu-
the stability is demonstrated both during the vault and lation has been done, considering the reinforced bolt-
the bench excavation. The local increase of the bolt- ing system on sidewalls. Cracking of the shotcrete
ing density prevents the development of induced frac- layer has been simulated by deleting some structural
tures, which appear on the sidewall during the vault elements near the crown. This deletion is done at 90%
excavation. The precise location of these additional of the vault deconfinement.
bolts is essential. We understand that an increase of Figure 13 shows the physical state of the rock mass
the bolting density on the roof is not useful for the at the end of the tunnel excavation. The local failure
goal to delay and stop the observed mechanics on site. of the shotcrete generates new damage and a frac-
A second interesting aspect is the effect of the time tured zone above the tunnel roof. The growth of this
at which additional bolts are installed. Figure 12 new fractured zone is stopped by grouted bars.
shows that the final horizontal convergence is smaller Figure 14 presents the evolution of the horizontal
with only one additional grouted bar installed earlier convergence in accordance with time, with and with-
rather than 2 bars added later. out failure of the shotcrete. An increase of displace-
These simulations highlight very well what project ment appears at the time of the failure but a new
engineers already knew but which has never been stable configuration is reached after that.
158
Time (days)
0
0 10 20 30 40 50 60 70 80
-5
-10
Time (days)
0 4 CONCLUSIONS
0 10 20 30 40 50 60 70 80
-5
The study of monitored convergences in the Aptian
Horizontal convergence (mm)
-10 Without failure of the shotcrete layer marls of the Tartaiguille Tunnel has made it possible
-15 With failure of the shotcrete layer to approach quantitatively the effect of bolting on the
-20 structure’s stability. We were able to confirm the very
considerable efficiency that a few extra bolts bring to
-25
a mass of non-fractured soft rock. This back analysis
-30 has made it possible to quantify the effect of reinforc-
-35 ing the sidewalls on the reduction of convergences in
-40
the Tartaiguille tunnel.
In the framework of a back analysis, some numeri-
Figure 14. Influence of a local shotcrete failure on horizontal cal computations have been done considering a new
convergences. constitutive model well adapted to underground engi-
neering expectations. Goals of these simulations were
to find with a suited numerical tool major behaviors
observed and monitored during the Tartaiguille tunnel
3.6 Justification of the length of bars digging.
Some computations have been done considering sev- These simulations allow us to identify:
eral lengths for grouted bars. Figure 15 shows hori- – The potential major failure mechanism of the tun-
zontal convergences versus time, in accordance with nel. The knowledge of this mechanism is essential
these lengths. If the length is greater than 4 m, global both during the design phase and during excava-
stability is assured. On the contrary, if this length is tion process.
2 m or less, a stable configuration can’t be obtained – The significant efficiency of a grouted bar system
and stability of the tunnel can not be justified during on tunnel stability conditions.
the bench excavation. – The effect of a local reinforcement of the bolting
From these results, we may conclude that there is system on displacements and safety level during
an optimal length for grouted bars, depending on the the digging.
potential failure mechanism of the tunnel during the – The limited effect of a local shotcrete failure if the
excavation process. These numerical results confirm grouted bar system is sufficient and if the stability
usual formulas, which provide an estimation of the is not only ensured by a shotcrete shell.
159
160
ABSTRACT: The ability to predict ground movements caused by tunneling is becoming increasingly impor-
tant as more tunnels are constructed in urban areas. It is generally recognized that the ground surface settlement
trough above a tunnel is well represented by a gaussian curve. Data is available from many projects around the
world, which provides guidance on the values to use in the gaussian curve. However, if novel forms of tunnel-
ing are used, if unusual ground conditions are present or if horizontal or subsurface ground movement predic-
tions are required, then these empirical methods are not suitable.
This paper presents the results of a suite of numerical analyses carried out to identify the most appropriate soil
model to use for the prediction of surface settlement troughs. As shown by many other authors linear elastic or
linear elastic perfectly plastic soil models are unsuited to the prediction of realistic surface settlements. Most of
the analyses carried out in the study predict a settlement trough that is wider than observed despite modifica-
tions to the size of the model, the boundary conditions, the in-situ stress conditions and the initial small strain
stiffness. The constitutive model that predicts the most realistic settlement trough was a non-linear anisotropic
soil model with a higher horizontal than vertical stiffness. The soil non-linearity was based on the approach
suggested by Jardine but modified for anisotropy. The shape of the trough was found to be sensitive to the value
of the vertical to horizontal shear modulus and the ratio of the horizontal and vertical Young’s moduli.
161
162
A 1104 1260
B 1035 1143
C 5.00E-06 1.00E-06
R 275 618
S 225 570
T 2.00E-05 1.00E-05
! 0.974 1.335
0.94 0.617
1.044 2.069
0.98 0.42
smin 8.80E-06 1.40E-05
smax 3.50E-03 2.00E-03
vmin 2.10E-05 1.00E-04
vmax 2.00E-03 2.00E-03
0.2
0.4
0.6
0.8
and p is the current mean effective stress.
t1
Throughout the analysis the stiffness was continu- 1 t2
ally updated. Up to a specified minimum strain (smin Gauss Curve
or vmin), the stiffness varies only with p , but thereafter 1.2
163
0 10 20 30 40 50 0 10 20 30 40 50
0 0
0.2
0.2
0.4
0.4
0.6
0.6
0.8
t3
0.8 1 t5
t1 t6
t3 t7
1 t4 1.2
Gauss Curve
Gauss Curve
1.2 1.4
settlement is still not located on the tunnel centerline, In both of these analyses the increase in the soil stiff-
but is offset by about 10 m. The trough is significantly ness made the shape of the settlement trough worse.
narrower than the linear elastic case, but is still much In the third analysis (t7) the small strain stiffness was
broader than the gaussian curve. To investigate the increased by 50%, but the shape of the stress–strain
influence of the high horizontal stress on the shape of curve was the same after the end of the initial plateau
the settlement trough, an analysis was run using a k0 region as in analysis t3. The modifications to the
profile based on the approach suggested by Mayne & small strain stiffness had only a modest influence on
Kulhawy (1982). The k0 profile assumed that 170 m the shape of the settlement trough.
of overburden had been removed from the top of the
London Clay prior to the deposition of the Terrace 2.3 Anisotropic soil model
Gravel. The k0 was taken as 0.4 in the Terrace Gravel.
The shape of the k0 profile is given in Figure 1 and A number of authors have indicated that only with an
shows lower k0 values particularly in the top ten anisotropic soil model can a realistic shaped settlement
meters of the London Clay than the profile used in trough be obtained (Simpson et al. 1996, Addenbrooke
analysis t3. The resulting settlement trough from et al. 1997). There is good evidence that the behavior
analysis t4 is shown in Figure 4. Although the low of London Clay is anisotropic with a higher hori-
point of the settlement trough is nearer to the tunnel zontal than vertical Young’s modulus (Bishop et al.
centerline and the trough is generally narrower, the 1965). Data also exists for other overconsolidated
overall shape of the settlement trough is only slightly clays (Lings et al. 2000).
different. The anisotropic elastic model was modified to allow
To consider the effect that fixity conditions on input of non-linear elastic behavior. No anisotropic
the far boundary have on the shape of the settlement elastoplastic soil model currently exists in FLAC,
trough, analyses were run with the far boundary fixed however with a volume loss of only 0.8%, the strains
both horizontally and vertically and also with a stress in the ground surrounding the tunnel are only suffi-
boundary condition. Neither analysis gave an improved cient for very local plastic yielding of the ground to
shape of settlement trough. occur and therefore there should be only a small error
Analyses were also carried out with wider meshes to in the adoption of an elastic model. In the absence
see if a boundary width of 4 tunnel depths was inade- of any definitive anisotropic constitutive soil model
quate. Analyses were carried out with a mesh width of for the London Clay, the basic Jardine equation was
150 m and 1000 m. The effect of an increased mesh modified as follows:
width was minor with a small reduction in the settle-
ment at 50 m from the tunnel centerline, but a corre-
sponding increase in the settlement 5 m from the tunnel
centerline.
Analyses were carried out to investigate the effect
of modifying the shape of the non-linear model and
the results are presented in Figure 5. In the first analy-
sis (t5) the stiffness was increased by 50% at all
strains compared to the model prediction. In the next
analysis (t6) the strain limit for the plateau region where X is defined as above. The values of the
of the model was extended to a higher strain level. constants in the above equations are given in Table 3.
164
0.2 0.2
0.4 0.4
0.6 0.6
0.8 0.8
t9
t8 t11
1 t9 1 t12
t10 t13
Gauss Curve Gauss Curve
1.2 1.2
The small strain stiffnesses are slightly lower as they Distance from centre-line (m)
now relate to the vertical stiffness rather than an 0 5 10 15 20 25 30 35 40 45 50
0
165
Distance from centre-line (m) The stress–strain behavior of most rocks is con-
0 5 10 15 20 25 30 35 40 45 50 trolled not by the elastic behavior of the intact mate-
0
Settlement / Centre-line Settlement .
166
167
ABSTRACT: A FLAC3D analysis was conducted for the investigation of the failure and the design of remedial
works for one tunnel in Melbourne Australia. Failure of an approximately 8 m length of the sidewall of the formed
concrete arch occurred in mid February 2000. The original design of the un-reinforced concrete lining was such
that it just sits on the flat upper surface of the approximately 1.8 m-thick concrete invert. It was generally believed
that the failure was primarily due to the compressive stresses across the arch/invert joint being substantially low,
which means that the compressive stresses in the arch lining induced by the groundwater pressure at time of failure
must have been transferred by 3D action to the west and east of the failure. The purpose of the numerical mod-
eling is to return the failed section to a fully functional arch/invert concept. Key factors in the remediation are
the width and sequence of removal of the panels, which were investigated in details in this paper.
1 INTRODUCTION and 2 are 4 m high, Panels 3 and 4 are 3.2 m high and
Panel 5 is 1.6 m high. The model geometry is shown
The remedial concept is to remove the formed arch in Figures 1a, b & 2.
concrete within and immediately around the failure area
and reinstate the original design. As a precursor to the
3D analysis, 2D analyses using Phase II was carried 2.2 Interfaces
out in PSM office to assess the likely compressive
Interfaces are planes within a FLAC3D model along
and tensile stresses generated above and around panel
which sub-grids can interact, slip and/or separation is
cut-outs, and the effects of flat jack stressing. The
allowed. A total of eight interfaces shown in Figure 3
results were used as a guide in selecting the 1.5 m
are modeled as:
panel width and the excavation sequence proposed in
the design. – Interface 1: between arch and invert.
A FLAC3D model includes a 36 m length of the tun- – Interface 2: between concrete arch/invert and rock
nel and the surrounding rock, which contains interface simulating the membrane.
elements between the arch lining and rock, between – Interface 3: between arches A and B.
the arch lining and floor, and between arch pours A – Interface 4: between panel 1 and arch B.
and B. The model allows an initial 3 mm gap to exist – Interface 5: between panel tops and concrete arch A.
at the arch/invert interface within the modeled failure – Interface 6: between back of panels 1, 3 and 5 and
area, prior to application of groundwater pressures. rock surface simulating the membrane.
– Interface 7: between back of panels 2 and 4 and
rock surface simulating the membrane.
2 NUMERICAL MODELING – Interface 8: between panels and concrete arch A.
169
2.4 Modeling sequence was initially applied at the interface between the
concrete arch/invert and rock surface. To do so, a
The following stages were developed:
virtual interface inside the tunnel surface has to be
– Stage 0: Initial condition. A 3 mm gap was modeled. set up in order to store the face list for applying the
The pore pressure of pre-leak value of 470 KPa equivalent normal stress to the rock surface.
170
– Stage 1: Reduce the pore pressure around the tunnel – Stage 2: Remove the Panels 1, 3 and 5. The corre-
to pre-repairing condition that is zero behind the sponding interface element at the back of the pan-
failure area and increases linearly to 470 KPa at 30 m els should be removed from the list and the same
away from the failure. principal applies to the corresponding rock surface
171
172
Figure 4. Contour of smin at stage 0 showing the arching effect due to 3 mm gap between arch and invert.
Figure 5a. Contour of smax at stage 2 where panels 1, 3 and 5 (from the right) were excavated.
and develops between panels 4 and 5, where a stress Figure 9 shows the cracking pattern after all
concentration is noticed due to the difference in flat jacks are stressed to 4 MPa with groundwater
height. It should be noted that tensile cracking in pressure at low values, corresponding to the pro-
these locations is not of a particular concern. cess during the repair work. In general, the stresses
173
Figure 5b. Contour of smin at stage 2 where panels 1, 3 and 5 (from the right) were excavated.
are benign and there is no new cracking. Figure 10 arch about 2.5 m above the panels vertically. This
shows the crack pattern when the groundwater indicates the need to increase the flat jack pressures
pressures are returned to a high value at 470 KPa. progressively as the groundwater pressures are
A tensile crack is predicted on the rock side of the allowed to recover.
174
Figure 7a. Contour of smax at stage 4 where panels 2 and 4 (from the right) were excavated.
Figure 7b. Contour of smin at stage 4 where panels 2 and 4 (from the right) were excavated.
As an alternative, one more model was run from hydrostatic load around the tunnel to a lower and uni-
the end of stage 6. Instead of increasing the hydro- form load of 400 KPa. Then increase the hydrostatic
static load to 470 KPa in one go, a progressive load to 425 KPa, 450 KPa and finally to 470 KPa
approach was adopted this time. First adjust the progressively. An improved cracking was noticed as
175
176
(excavate panels 1, 3 & 5 first and then panels 2 & 4) is recommended that the measures of progressively
is reasonable design assumptions. Tensile cracking is increasing the flat jack pressures as well as slowly
predicted at the eastern top corner of Panel 1 adjacent recovering the groundwater pressure to a full uniform
to the frictional joint between arches pours A and B. It value of 470 KPa are to be taken to minimize the
177
tensile cracks in the concrete arch region above the assistance in developing FISH coding. The authors also
repaired section. benefited from many discussions with him as well.
It also demonstrates that the FLAC3D is a useful
tool and can be well applied to solve the complicated
engineering problem. REFERENCES
178
Mining applications
ABSTRACT: The paper presents the numerical analyses performed with FLAC3D to study the stability of the
rock mass surrounding the stopes at the Niobec mine in Chicoutimi, Québec. Since the mine expansion is done
at depth, the stability of the planned stopes had to be evaluated in order to determine the support requirements
linked to an increase of the in situ stresses. The paper focuses on the dimensioning of the sill pillar between
mining blocks 3 and 4 using FLAC3D. A rock mass characterization of the site has been made prior to the numer-
ical modeling. The characterization consisted of structural geological mapping, laboratory testing of rock sam-
ples and in situ stress measurement. All the stopes were then modeled and the parameters obtained from the site
characterization were used in the FLAC3D model. Various alternatives have been analyzed, including the use of
backfill and variations of pillar thicknesses for the third pillar.
181
182
183
speed up the run time of the analyses. In order to eval- 4 FLAC3D SIMULATIONS
uate the influence of the principal stress tensor reori-
entation, two analyses were done using the Examine 3D Before running the analyses used to estimate what sill
boundary element program developed by Rocscience. pillar size would be optimal, various scenarios were
This software allows us to apply the in situ major studied in order to perform the numerical analyses
principal stresses with any orientation relative to the more efficiently and to verify if the use of certain
model. In one analysis, the measured in situ stress support elements such as backfill would have an
tensor (Table 1) was input and in the other, the modi- effect on the local and overall stability of the mine. It
fied stress tensor with 1 and 2 in the horizontal should also be stated that a validation of the model by
plane was used. As expected, the second model gave comparing its results with in situ observations was
lower strength factor values, which confirms the sim- difficult for various reasons. First, no in situ monitoring
plifying assumption related to the reorientation of the of displacements was available. Secondly, as the rare
stresses puts the model on the safe side. The strength instabilities around the excavations in the mine are
factor given by Examine 3D compares the stress state mostly controlled by the presence of discontinuities
at every point of the model with the strength enve- which are not considered in FLAC3D, it is difficult to
lope. A strength factor of 1.0 means the stress state is perform a direct comparison between the extent of
on the strength envelope while a lower value means instabilities in the numerical model and the ones
the stress state exceeds the strength of the rock mass. observed in the field. This only emphasizes the fact
These strength factors must be considered with care that in the absence of field monitoring and in the
since they tend to underestimate the extension of presence of a good quality rock mass, validation of
potentially failed zones, since the elastic analysis the numerical models is difficult.
software performs does not consider the post failure
stress redistribution.
4.1 Mining sequence
3.3 Constitutive laws and material properties The influence of the mining sequence (excavating
In the analyses shown in this paper, the rock mass was the stopes in the same sequence they were mined out
assumed to present an elastic perfectly plastic behav- in blocks 1 and 2) on the results of the analyses was
ior, so no post peak strength values are given. Table 4 studied. This was an important point to verify since the
shows the strength and deformability parameters of mining sequence for the new mining blocks (3 and 4)
the rock mass used in the model. These parameters was unknown and excavating the stopes all at once in
were obtained by combining the laboratory test results the numerical model was an interesting alternative as
with the rock mass classification parameters presented it would allow important time savings. Consequently,
in section 2.4. two analyses were run, one by excavating the stopes
The authors are aware that for fragile hard rock one after the other and waiting for the unbalanced
such as the one found at the Niobec Mine, perfect forces to stabilize in between and the other by nulling
plasticity is a not realistic assumption, but using a the elements in the stopes all at once. The comparison
strain softening constitutive law would have slowed between the two runs is done by taking the number of
down the runtime of the analyses which already took failed elements in each simulation as shown in Table 5.
over 5 days to run on a 1.0 GHz Pentium PC. Nonethe-
less, now that the model is built, it would be a simple Table 5. Comparison of the number of failed zones on
matter to implement the strain softening parameters different sections of the model for the analyses with and
and perform a sensitivity analysis by varying the post- without sequential mining.
peak strength parameters. The authors are also aware
that perfect plasticity will underestimate the exten- Mining all at Sequential
once (failed mining (failed
sion of eventual failure zones. Section elements) elements) Difference
4410 36 37 1
Table 4. Material parameters used in the 4465 74 80 6
model to simulate the rock mass behavior. 4530 65 64 –1
4575 70 81 11
Uniaxial compressive strength 34.4 MPa 4625 58 71 13
Tensile strength 2.1 MPa 4675 45 35 –10
Cohesion 10.5 MPa 4730 41 37 –4
Friction angle 38.5° 4795 87 92 5
Young’s modulus 47.3 GPa 4830 37 43 6
Poisson’s ratio 0.254 4900 14 15 1
184
Most of the failed elements are located around the any significant stresses, so it would not serve the pur-
stopes and their dimensions are small compared to pose of reducing the size of the third horizontal pillar.
the 20 m 20 m 20 m mesh used in areas remote
from the excavations. The section heading refers to var- 4.3 Design of the third horizontal pillar
ious sections in the model that cut through the stopes in
The main objective of this project was to find the
the model. It is considered that the excavation sequence
optimal thickness of the third horizontal pillar. The
has a negligible influence and that no clear pattern is
first pillar between mining blocks 1 and 2 is 30 m
observed. Consequently, it was concluded that for the
thick, the second between mining blocks 2 and 3 is
following simulations, the stopes could be excavated all
45 m thick and the third one between mining blocks 3
at once without much effect on the outcome of the runs.
and 4 was also planned to be 45 m thick. Since the use
of backfill was found of little use, only three simula-
4.2 Influence of backfill tions were made. One optimistic analysis with a 30 m
As the mine had never used backfill, it was decided to thick pillar, another with a 45 m thick pillar identical
see if the use of such support would allow minimizing
the occurrence of local failure around certain stopes. Table 6. Comparison of the number of failed zones on dif-
To estimate the influence of backfill on the stability of ferent sections for the simulations with and without backfill.
the openings, the same methodology as used to evalu-
Without With
ate the influence of the mining sequence was adopted.
Section backfilling backfilling Difference
In these simulations, after a stope was mined out and
the unbalanced forces had stabilized, the null elements 4410 37 37 0
were replaced by zones having properties matching 4465 80 77 3
a backfill with 8% cement. The next stope was then 4530 64 61 3
excavated and the sequence was repeated for all the 4575 81 79 2
openings. The number of failed zones with and without 4625 71 67 4
backfill are presented in Table 6. These simulations 4675 35 33 2
showed the fill to have no significant influence on the 4730 37 34 3
4795 92 88 4
stability of the mine. It is believed that the very low
4830 43 43 0
stiffness of the backfill compared to that of a good qual- 4900 15 15 0
ity rock mass makes it almost impossible for it to absorb
Figure 5. Failure zones for the 30 m (100) pillar for section 4795.
185
Figure 6. Failure zones for the 45 m (150) pillar for section 4795.
Figure 7. Failure zones for the 60 m (200) pillar for section 4795.
186
to the one between blocks 2 and 3 and finally a last Also, if one wishes to fine-tune the model, an opti-
simulation with a 60 m pillar. mization of the element size around the excavations
As the stope layouts for the third and fourth mining could be made.
blocks were not available at the time the simulations In conclusion, the work presented in this paper
were run, the geometry of these blocks was assumed showed that modeling a complete mine with FLAC3D is
to be similar to that of blocks 1 and 2. quite an undertaking since there are no simple ways to
The simulation with a 30 m pillar showed that the create the geometry, although there is now an interface
pillar would be stable, but that only a 20 m thickness with AnsysCivilFEM which should facilitate this
would remain intact (Fig. 5). The two other simula- task (it was not available at the time the project started).
tions showed that a 40 m thick zone would remain There are also difficulties in the application of the
intact with the 45 m pillar (Fig. 6) and 53 m would be boundary conditions as mentioned in section 3.2 since
free from failed zones for the 60 m pillar (Fig. 7). the authors we unable to apply shear stress gradients
The number of failed zones in the pillars were then to the model boundaries. Besides these difficulties,
compared. The direct comparison can be done since once the model is built, it is interesting to be able to
the number of zones remained constant between sim- perform sophisticated sensitivity analyses by modify-
ulations and only the zone thicknesses were changed ing the parameters of various constitutive laws.
to modify the pillar thickness. The analyses showed
the 30 m pillar to have 657 failed zones on a total of
4425 zones in the pillar, while the 45 m pillar had 603 ACKNOWLEDGEMENTS
failed zones and the 60 m pillar showed 583 failed
zones. Although the 30 m sill pillar showed an overall The authors wish to acknowledge Martin Lancet and
stability, the intact thickness is considered too small the personnel of the Niobec Mine who have contributed
as the presence of planes of weakness not considered to the success of this M.A.Sc project. They also want to
in the analyses may cause important instabilities. acknowledge the National Research Council of Canada
Bearing this in mind, the 45 m pillar would leave an for its financial support (grant # OGP0089752).
intact rock section considered more adequate. The
results show the 60 m thick pillar would not increase
the overall safety factor significantly and the side REFERENCES
effects of having a pillar which is too thick, is the ore
loss and also the fact that a thicker pillar will expose Arjang, J. 1986. In situ stress measurement at Niobec Mine,
Canmet Laboratory Report.
the stopes in the fourth mining block to higher in situ Corthésy, R. 2000. Mesure des contraintes in situ, mine
stresses (due to their increased depth) causing Niobec. CDT report, Ecole Polytechnique.
unwanted dilution. Corthésy, R., Gill, D.E., Leite, M.H. 1993. An integrated
approach to rock stress measurement in anisotropic non
linear elastic rock, Int. J. Rock Mech. Min. Sci., Vol. 30,
5 DISCUSSION AND CONCLUSIONS no. 3, pp. 395–411.
Leeman, E.R., 1967. The doorstopper and triaxial rock stress
The proposed pillar design presented in this paper measuring instruments developed by the CSIR, J. of the
should, prior to accepting it, be analyzed using a more South Afr. Inst. of Mining and Metall., Vol. 69, no. 7,
1967, pp. 305–339.
realistic constitutive law than perfect plasticity for hard Leite, M.H., Corthésy, R., Gill, D.E., St-Onge, M., Nguyen, D.
rock. The strain-softening model available in FLAC3D 1996. The IAM – A down-the-hole data logger condi-
should be tested with various post-peak strength tioner for the modified doorstopper technique. 2nd North
parameters in order to perform a sensitivity analysis of American Rock Mechanics Symposium, Montréal,
the excavation response to these parameters. pp. 897–904.
187
ABSTRACT: Mines are often faced with sill extraction situations, and one technique that can be used to extract
a sill is to leave a consolidated fill mat in the sill cut. Several Canadian mines employ this sill extraction technique,
and in this paper the stability of these sills was modeled using two-dimensional FLAC numerical simulations. The
objective of this parametric study was to derive relationships between the required strength of the sill mats and
the maximum stable unsupported undercut span for various orebody dips. The footwall-to-hangingwall spans
described in this paper were 1.2 m (4 ft), 2.4 m (8 ft), 3.6 m (12 ft), 4.8 m (16 ft), 6.0 m (20 ft) and 10.5 m (35 ft), with
mining dips of 60°, 70° and 80°. The range of fill cohesive strength used was from 100 to 500 kPa, a typical range
for most hydraulic or paste fills. Stability charts were derived (one per ore body dip), that can be used to select
the minimum fill strengths required (in terms of cohesion) to maintain stability for different combinations of
spans and dips.
2 FLAC MODEL
189
fill and each wall, in order to allow movement of the Table 1. Mechanical properties used for the host rock and
backfill along this surface. the ore.
190
1,800 Table 2. Strength and elastic properties used for the back-
1,600 fill sill mat.
1,400
Cement Elastic Shear Bulk
1,200
UCS (kPa)
0.70
the purpose of this study, it was assumed that the cohe-
0.60
Confinement
during test
sion and, hence, tensile strength, would decrease lin-
0.50
(kPa)
800
early from their maximum value at zero plastic strain,
0.40
400
Average
down to zero at a cumulative plastic strain of 1.5%.
0.30 The internal angle of friction was set to also vary lin-
0.20 early, but from its maximum value of 33° at zero plas-
0.10 tic strain, down to 30° at a cumulative plastic strain of
0.00 1.5% and beyond. Neither the shear nor the bulk
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
Cement Content (%) moduli are affected by plastic strain and were thus left
unchanged.
Figure 3. Typical stiffness properties for backfill (based on The older waste fill above the sill mat was subjected
triaxial lab tests carried out on a typical paste fill). to the same plastic strain-dependent weakening process.
Cohesion and tensile strength were also decreased lin-
early from their maximum value (set, as mentioned pre-
6. calculate the corresponding shear modulus G and viously, at 90% of those of the sill mat) at zero plastic
bulk modulus K, assuming a Poisson’s ratio of strain, down to zero, also at a cumulative plastic strain
0.30 (for our example G would be 206 MPa while of 1.5%. Similarly to the sill mat, the internal angle of
K would be 442 MPa); and, friction of the weaker fill material was set to decrease
7. use these values as input to the FLAC model. For linearly from its maximum value of 33° at zero plastic
the example, these inputs would be: cohesion strain, down to 30° at a cumulative plastic strain of
200 kPa, tensile strength 100 kPa, G 206 MPa, 1.5% and beyond.
K 442 MPa.
The nominal friction angle was chosen at 30° (friction 2.6 Interface between rock and fill
angle values will be discussed in more detail later). As previously discussed, interfaces between the host
Table 2 summarizes the properties used for the various rock and the fill, along both the footwall and hanging
fill strengths considered. The density of the backfill was wall, had to be specified due to the continuum nature
assumed to be 2.0 t/m3. of the finite difference approach used in FLAC. During
The mechanical properties of the surcharge fill the initial compaction stage of each run, both the inter-
placed on top of the sill mats were lowered by 10% to nal angle of friction and the cohesive strength of these
account for the lower quality of fill typically placed interfaces were set to zero in order to prevent artifi-
on top of sill mats. The properties affected were the cial stresses from developing along them as the back-
cohesion, tensile strength, shear modulus and bulk fill settled. During the later stages of the runs these
modulus – the friction angle was kept the same. values were reset – the internal angle of friction was
As mentioned, the constitutive model used for the set equal to the internal angle of friction of the sill
consolidated fill material was a strain-softening one. In mat, and the cohesion was set equal to the cohesion of
order to simulate this behavior, the decrease in strength the sill mat.
191
JOB TITLE : Undermined Sill Mat Stability Analyses FLAC (Version 3.30)
+05
(10 )
FLAC (Version 3.30) LEGEND
(10+05 )
LEGEND
step 9000 5.000
step 9000 5.000 HISTORY PLOT
HISTORY PLOT Y-axis : 4.000
Y-axis : Max. unbal. force
4.000
Max. unbal. force X-axis :
3.000
X-axis : Number of steps
Number of steps 3.000
2.000
2.000
1.000
1.000
1 2 3 4 5 6 7 8 9
+03
(10 )
1 2 3 4 5 6 7 8 9 Itasca Consulting Canada Inc.
+03
Itasca Consulting Canada Inc. (10 )
Figure 6. Example of a FLAC “y-displacement” (vertical) history plot at the various control points located within the back-
fill column, for a stable configuration. Note that the maximum displacement is finite at about 45 mm.
192
Figure 7. Example of a FLAC “y-displacement” (vertical) history plot at the various control points located within the back-
fill column, for an unstable configuration. Note that the maximum displacement increases without bound (up to 1.4 m at the
end of 12,000 cycles in this case).
Figure 8. Example of a FLAC grid plot showing failure of the sill mat at mid span, and especially at the hanging wall
contact. Note that the fill displaces downward by up to 500 mm, indicating failure.
193
Figure 9. FLAC results for a 70° dip, a 400 kPa cohesion sill mat and a 6.7 m (20 ft) span. Most of the movement occurs in
the lower part of the sill mat, and the whole arrangement remains stable.
600 600
500 500
Average Cohesion (kPa)
300 300
200 200
0 0
0 5 10 15 0 5 10 15
Stope Span (m) Stope Span (m)
Stable 60 Failed 60 Stable 80 Failed 80
Figure 10. Stability Chart for 60° dipping ore showing Figure 11. Stability Chart for 80° dipping ore showing
failed cases and stope spans. Also shown is a crude contour failed cases and stope spans. Also shown is a crude contour
separating the failed cases from the stable cases. separating the failed cases from the stable cases.
Back support in the form of mat reinforcing (e.g., and dips. Since fill strength is normally measured in
screen placed on the floor of the stope to fill, together terms of uniaxial compressive strength, this can be
with vertical bolts) will stabilize the local back when estimated by multiplying the cohesion by a factor of
the sill stopes are extracted and the fill mat is under- 4, i.e. for a cohesion of 200 kPa, a uniaxial compres-
mined. The aim of this study is limited to the overall sta- sive strength of 800 kPa will be necessary. Note that
bility of the backfill sill mats, and excludes minor falls the FLAC analyses as presented do not incorporate
of fill from the back that must be expected to occur any Factors of Safety – appropriate Factors of Safety
unless appropriate mat reinforcing techniques are used. must therefore be applied to the fill strength for design
It is understood that the support of the immediate back purposes. This will depend on the quality and degree of
will be ensured by mines through the appropriate use of uniformity of the fill as placed. Note also that the charts
this type of reinforcing. suggest that it is not possible (for the fill strengths
These charts can be used to select fill strengths (in considered) to undermine a fill panel of 10.5 m width
terms of cohesion) for different combinations of spans without some form of additional reinforcement.
194
195
A. Mortazavi
Previously with Itasca Consulting Canada, Inc., Sudbury, Ontario, Canada
B. Simser*
Falconbridge Ltd., Sudbury Mines/Mill Business Unit – Craig Mine, Onaping, Ontario, Canada
P. George*
New Brunswick, Canada
*previously with Noranda, Inc., Brunswick Mine, Bathurst, New Brunswick, Canada
ABSTRACT: In early 2001 Itasca Consulting Canada Inc. was contracted by Noranda Inc. to assist in the
design of the first two drifts that were going to be driven through paste backfill at Brunswick Mine in the south
end of the 1000 m Level in order to create alternate accesses to the western ore zones. A numerical stability
analysis of the proposed tunnels was carried out by means of two-dimensional FLAC simulations, which took
into account different fill strengths, alternate tunnel geometries, various floor conditions and the presence of
unconsolidated plugs of waste rock within the paste backfill at close proximity to the tunnel in one area. The
main objectives of this work were to investigate the self-standing characteristics of the exposed paste material,
evaluate the deformations expected as a result of tunneling through it and recommend adequate ground support
alternatives. This paper describes the modeling approach used, the results obtained and how they corresponded
to the behaviors later encountered underground during the excavation of the tunnels.
197
198
3 FLAC SIMULATIONS dry rockfill placed between 1000-3 sub and 1000-2 sub
in the secondary stopes on both sides of this access
Simulations were run for the following cases: had run into the area due to local caving on the 1000-2
(1) a horseshoe shape drift driven with loose material sub horizon. Despite repeated attempts to remove this
on both sides; (2) a flat back drift also developed waste material, uncontrolled runs of fill had resulted
with loose material on both sides; (3) a horseshoe in significant amounts of loose material being present
shape drift driven without any loose material on either above the 1000-2 sub elevation at the time the paste
side; and, (4) a horseshoe shape drift driven without backfill was poured in the 236-8 Access. This, as
any loose material on either side, but with a horizon- shown in Figure 1, resulted in loose material being
tal discontinuity in the paste material directly above it located on both sides of the future drift, at a repose
(in order to investigate potential large-scale slabbing angle of about 45°.
of the paste into the tunnel). Figure 1 also shows the numerical grid, geometry
and boundary conditions of the model constructed for
3.1 Case 1 – arched back drift, with loose waste this first series of runs. Paste material cohesion values
material on both sides of 50, 100 and 400 kPa were investigated – the results
This situation prevailed in the 236-8 Access on 1000-2 indicated stable self-standing conditions for a paste
sub, where caving had occurred along a band of weak material with a cohesive strength greater than 50 kPa.
waste rock inside the original drift, and some of the As shown in Figure 2, a 50 kPa cohesion resulted in
the prediction of a maximum displacement of over
31 cm 12,000 cycles into the simulation. Further
stepping of the model (to 15,760 cycles) confirmed
the complete failure of the paste backfill material and
clearly described the paste failure mechanism taking
place under the simulated conditions.
It is interesting to note that the back failure was not
Paste Fill as pronounced as that of the walls. As intuitively
expected, the simulation confirmed that, under vertical
(gravity) loading conditions, most of the vertical load
around the excavation is deflected and concentrated in
the drift walls.
Loose Waste The existence of weak contacts between the paste
backfill and the loose waste material towards the bot-
Rock fill tom and on both sides of the drift initiated a deforma-
tion of the paste along this contact. This, in turn, led to
the shearing of paste material on both sides of the drift.
After the side wall failure, the process propagated
upwards and led to the shearing of the paste material
above the drift, as shown in Figure 3. A maximum dis-
placement of as much as 1.20 m was observed at this
later stage. Moreover the unbalanced force history
within the model showed that after the initiation of fail-
ure the unbalanced force continued to increase, indicat-
ing that a progressive failure kept on evolving within
the model as no state of equilibrium was being reached.
199
Figure 2. Displacement vectors at step 12,000. A maximum displacement of over 31 cm was predicted in this case.
200
3.3 Case 3 – arched back drift, with no loose waste sub-level elevation. Three different cohesive strength
material on the sides values were again considered for the paste backfill,
The aim of this set of analyses was to investigate the which were 25, 50 and 100 kPa in this case.
stability of a drift driven through paste backfill under No major displacement was predicted to occur in
“normal” conditions, i.e., with no loose waste material the drift for the cases of the 50 and 100 kPa cohesion.
in the pasted region and with the paste material poured For the 50 kPa cohesion case a maximum displacement
directly on compacted rockfill. This situation, illus- of just under 3.7 cm was predicted, whereas this maxi-
trated in Figure 5, corresponded to the situation in the mum displacement was predicted at just over 3 cm for
129-7 Access in the south end of the 1000 m Level #1 the 100 kPa cohesion case.
201
Figure 4. Displacement vectors at step 12,000 (50 kPa cohesion, flat back drift profile).
202
Figure 6. Displacement vectors for a cohesion of 25 kPa. This situation evolved into the complete collapse of the tunnel.
203
Figure 8. Displacement vectors for a cohesion of 50 kPa. A maximum displacement of just over 3.6 cm was predicted in
this case.
204
ABSTRACT: On November 27, 2002 Agnico Eagle’s Laronde Mine experienced a magnitude 2.6 seismic
event. As part of the investigation into the cause of the burst, a high-resolution FLAC3D model was created to
determine the effect of the mining sequence on stresses in the area of the burst. At this mine, secondary stopes
are intended to fail following primary stope extraction. The numerical simulations showed that where remnants
were left with non-ideal geometry (through unfortunate but necessary mining decisions), these remnants could
be too strong to yield as intended.
The FLAC3D model showed that one such 3-wide pillar centered at the location of the burst was subject to a
local high stress concentration. This provided a unique opportunity to confirm calibration of the FLAC3D model.
Further modeling also highlighted other areas of the mine where pillars were in a high stress state and recom-
mendations were made to alter the mining sequence to prevent future events.
205
Figure 1. A typical cross-section of the Laronde orebodies looking East showing the different rock elements.
206
20-North Ore-
152-20-64 body. Mined and
Foliated Zone
Stope Backfilled.
149 Level
Shear stress on
the foliations
152 Level
Stress trajectories
Figure 3. Simplified cartoon view of the intersection of Figure 5. Looking East on the 152 Level. The damage is
the foliated zone with the levels involved in the burst. more severe especially to the east. A large amount of mate-
rial up to 2 meters deep is seen in the main entrance to the
level on the lower left.
4 FLAC3D MODEL
207
Figure 6. Maximum principal stress plot looking North showing the location of the three high stress pillars and the location
of the burst.
Figure 7. A cross section looking East through the region of the burst (section A-A in Fig. 6).
vertical component appears to have provided enough although not as critical as the one that caused the
localized stress on the foliated zone to cause the slip. burst, were cause for concern. The short term mining
From the mining plan, there are two other three- plan did not include these particular stopes although
wide pillars that appear to be concentrating stresses. after the modeling, recommendations were made to
A series of modeling runs was conducted in order to mine these stopes as early as possible.
determine how the short-term mining plan needed to The additional scenarios showed that the mining
be adjusted to prevent additional stress building up in of adjacent stopes created incremental increases in
these areas. Figure 6 shows these areas above and to the the stresses in these areas, so although they need not
East of the location of the burst. These two stopes, be mined immediately, a rapid development and
208
Figure 8. Maximum principal stress plot showing the changes caused by mining 143-20-65 stope.
Figure 9. Maximum principal stress along a long section looking North following the mining of 155-20-59 stope in the bottom
left.
production schedule should be implemented in order made because it was postulated that the burst had
to avoid further seismic events in the region. likely dissipated some of the stored energy in 152-20-
Using the simulations for the short term mining 64 stope and it was therefore unlikely that a second
plan of the area, it was determined that the best sce- seismic event would occur in the short term. Also
nario would involve mining the 143-20-65 pillar first. some significant rehabilitation on 152-level would
Figure 8 shows the stresses around the 152 mining delay the development and production of this stope by
horizon after mining out of 143-20-65 stope. The deci- a couple of months. The mining of 143-20-65 would
sion to recommend mining this stope first was it was not shed much additional stress on the other stopes but
209
210
ABSTRACT: Numerical tools can be very useful to investigate the mechanical response of backfilled stopes. In
this paper, the authors show preliminary results from calculations made with FLAC. Its use is illustrated by showing
the influence of stope geometry. The results are compared with analytical solutions developed to evaluate arching
effects in backfill placed in narrow stopes. Some common trends are obtained with the two approaches, but there
are also some differences regarding the magnitude of the stress redistribution induced by fill yielding.
211
dh
2 ARCHING EFFECTS
H C dh W C
Arching conditions can occur in geomaterials such as S S
soil, jointed rock mass and backfill, when differential
B
straining mobilizes shear strength while transferring
V + dV
part of the overburden stress to stiffer structural
components.
Arching typically occurs when portions of a fric-
rock mass B
tional material yield while the neighboring material
stays in place. As the yielding material moves between
Figure 1. Acting forces on an isolated layer in a vertical
stable zones, the relative movement within the former
stope.
is opposed by shear resistance along the interface with
the latter. The shear stress generated along the contact
area tends to retain the yielding material in its original
with
position. This is accompanied by a pressure reduction
within the yielding mass and by increased pressure on (2)
the adjacent stiffer material. Above the position of the
main arch, a large fraction of the total overburden where vh and hh are the vertical and horizontal
weight in the yielding mass is transferred by frictional stresses at depth h, respectively; represents the unit
forces to the unyielding ground on both sides. weight of the backfill; is the effective friction angle
Investigations on models and in situ measurements between the wall and backfill, which is often taken as
have shown that the magnitude of the stress redistribu- the friction angle of the backfill, bf. Equations 1 and
tion depends to a large extent on the deformation of the 2 constitute the Marston theory solution. In this repre-
walls confining the soft material (e.g. Bjerrum 1972, sentation, K is the reaction coefficient corresponding to
Hunt 1986). the ratio of the horizontal stress hh to vertical stress
A few analytical solutions have been developed to vh. This reaction coefficient depends on the horizontal
analyze the pressure distribution since the pioneering wall movement and material properties. When there is
work of Janssen (1895) (see Terzaghi 1943 for early no relative displacement of the walls, the fill is said to
geotechnical applications). Among these, the Marston be at rest, and the reaction coefficient is usually given
(1930) theory was proposed to calculate the loads on by (Jaky 1948):
conduits in ditches (see also McCarthy 1988). The
authors have used this theory to develop an analytical (3)
solution for arching effects in narrow backfilled stopes
(Aubertin 1999). where bf is the friction angle of the backfill. For typ-
Figure 1 shows the loading components for a verti- ical fill properties ( bf ≅ 30° to 35°), K0 is much
cal stope. On this figure, H is the backfill height, B the smaller than unity.
stope width, and dh the size of the layer element; W If the walls move outward from the opening, the
represents the backfill weight in the unit thickness horizontal pressure might be relieved, and the reaction
layer. At position h, the horizontal layer element is sub- coefficient tends toward the active pressure coefficient,
jected to a lateral compressive force C, a shearing force which can be expressed as (Bowles 1988):
S, and the vertical forces V and V
dV.
The force equilibrium equations for the layer ele- (4)
ment provide an estimate of the stresses acting across
the stope (Aubertin et al. 2003). From these, the vertical with Ka K0. If an inward movement of the walls
stress can be written as follow: compress the fill, it increases the internal pressure.
Then, the reaction coefficient tends toward the passive
condition, which becomes (Bowles 1988):
(1)
(5)
212
0.5m
0.5 void space
shh Marston theory v
h = 2
backfill
0.4 svh
overburden
shh rock mass
stress (MPa)
H = 45 m
0.2 E = 30 GPa
ã = 1800kg/m3
= 0.3
0.1
' = 30°
= 2700 kg/m3
0 c = 0 kPa
0 1 2 3 4 y
h/B B=6m
with Kp
1
K0.
In the above equations, it is assumed that cohesion is
low in the backfill; the fill then behaves as a granular 1
material. Based on limit equilibrium, it can be antici-
pated that a cohesion would increase Kp but decrease
Ka. However, more work is needed to investigate its
influence on arching effects and stress distribution.
Figure 2 shows values of vh and hh calculated with
Equations 1 and 2 (with K K0 0.5), and calculated
with the overburden pressure (i.e. vh h and
hh K0 vh). It can be seen that the overburden stress
represents the upper-bound condition, which is appli-
cable for low fill thickness (or for wide stopes). Typi-
cally, when H 2 to 3B, the pressure near the bottom
of the stope becomes more or less independent of the
(b)
1
depth of the fill, in accordance with measurements in
bins (Cowin 1977).
Figure 3. (a) Narrow stope with backfill (not to scale)
used for modeled with FLAC; the main properties for the rock
mass and backfill are given using classical geomechanical
3 NUMERICAL CALCULATIONS notations; (b) Schematic stress-strain behavior of the backfill
(available as a material model in FLAC).
3.1 Vertical stope
Recently, the authors have shown some preliminary
calculation results obtained with a finite element by considering the overburden weight (for an overall
code (Aubertin et al. 2003). Significant differences depth of 250 m). The natural in situ horizontal stress h
have been revealed between the Marston theory and is taken as twice the vertical stress v, which is a typi-
these numerical calculations, which may be cal situation encountered in the Canadian Shield. The
explained, in part, by different assumptions associ- rock mass is homogeneous, isotropic and linear elastic,
ated to the two approaches. In this paper, the same while the granular backfill obeys a Coulomb criterion.
geometry and material properties (Fig. 3a) are used for Figure 3b shows the stress-strain relation used with the
the basic calculations made with FLAC. The dimen- Coulomb plasticity model available in FLAC. This con-
sions of the opening are H 45 m and B 6 m, with stitutive behavior is different from the one used for the
a void of 0.5 m left at the top of the stope. The natural finite element calculations presented in Aubertin et al.
in situ vertical stress v in the rock mass is obtained (2003), which was of the elastic-brittle type. There are
213
overburden stress
0.6 Marston theory
yy (MPa)
K = 1/3
0.4
K = 1/2
K=3
0.2
0
0 9 18 27 36 45
(a) h (m)
0.2
xx (MPa)
K=3
0.1
K = 1/2
K = 1/3
0
0 9 18 27 36 45
(b) h (m)
214
σyy (MPa)
K = 1/3 K = 1/3
0.4 K = 1/2
K=3 K = 1/2
K=3
0.4
0.2
0
0 9 18 27 36 45 0
(a) h (m) 0 2 4 6
(a) x (m)
K=3 0.2
σxx (MPa)
σxx (MPa)
K = 1/2 0.1
K = 1/3
K = 1/3
0
0 9 18 27 36 45 0
(b) h (m) 0 2 4 6
(b) x (m)
Figure 6. Comparison of the stresses on the wall calculated
at different elevations h, with the analytical and numerical Figure 7. Stresses calculated at the bottom of the vertical
solutions: (a) vertical stress yy; (b) horizontal stress xx. stope, with the analytical and numerical solutions; (a) verti-
cal stress yy; (b) horizontal stress xx.
215
B=6m
4.1 Influence of mining sequence
rock mass In the numerical calculations presented in Aubertin
et al. (2003), the mining sequence was not taken into
stope v account, so the wall convergence due to elastic strain-
y
h = 2v ing of the rock mass was imposed on the fill. This
created an increase of the mean stress in the fill, while
60˚ vertical and horizontal stresses locally exceeded the
overburden pressure and the Marston theory solution
depth = 250 m (near mid-height of the stope).
x Modeling in this manner implies that the backfill is
placed in the stope before wall displacement takes
Figure 8. The inclined backfilled stope modeled with
place. For a single excavation stope, this is not a real-
FLAC (properties are given in Fig. 3).
istic representation (at least for hard rock masses).
However, with a cut-and-fill mining method where the
mining slices (or layers) are relatively small compared
to the whole height of the stope, filling is usually
0.2 modeling with FLAC-2D made quickly after each cut. In this case, wall conver-
overburden gence after each cut compress the fill already in place
Marston theory (Knutsson 1981, Hustrulid et al. 1989). The inward
movement of the walls may then create a condition
xx (MPa)
216
K
h (m)
0
0.3 active
at floor
σxy (MPa)
0 9 18 27 36 45
-0.02
at 1/2H modeling with FLAC
-0.04 at 1/2H
at 3/4H
-0.06 0.1
K = 1/3
K=3 0 2 x (m) 4 6
-0.08 K = 1/2
Figure 12. Reaction coefficient K obtained with analytical
Figure 10. Comparison of shear stress distribution along and numerical solutions across the full width of the vertical
the wall. stope at different elevations h.
217
218
Itasca Consulting Group, Inc. 2002. FLAC – Fast Lagrangian Nantel, J.H. 1983. A review of the backfill practices in the
Analysis of Continua, User’s Guide. Minneapolis, MN: mines of the Noranda Group. In S. Granholm (ed), Mining
Itasca. with Backfill: Proceedings of the International Symposium
Jaky, J. 1948. Pressure in silos. Proceedings of the 2nd Inter- on Mining with Backfill: 173–178. Rotterdam: Balkema.
national Conference on Soil Mechanics and Foundation Ouellet, J. & Servant, S. 2000. In-situ mechanical characteri-
Engineering, 1: 103–107. Rotterdam: Balkema. zation of a paste backfill with a self-boring pressuremeter.
Janssen, H.A. 1895. Versuche über Getreidedruck in CIM Bulletin, 93(1042): 110–115.
Silozellen. Zeitschrift Verein Ingenieure, 39: 1045–1049. Richards, J.C. 1966. The Storage and Recovery of Particulate
Knutsson, S. 1981. Stresses in the hydraulic backfill from Solids. Institution of Chemical Engineers, London.
analytical calculations and in-situ measurements. In O. Spangler, M.G. & Handy, R.L. (1984). Soil Engineering.
Stephansson & M.J. Jones (eds), Proceedings of the Harper & Row.
Conference on the Application of Rock Mechanics to Cut Take, W.A. & Valsangkar, A.J. (2001). Earth pressures on
and Fill Mining: 261–268. Institution of Mining and unyielding retaining walls of narrow backfill width.
Metallurgy. Canadian Geotechnical Journal, 38: 1220–1230.
Li, L. & Aubertin, M. 2003. A general relationship between Terzaghi, K. 1943. Theoretical Soil Mechanics. John Wiley &
porosity and uniaxial strength of engineering materials. Sons.
Canadian Journal of Civil Engineering (in press). Udd, J.E. 1989. Backfill research in Canadian mines. In F.P.
Marston, A. 1930. The theory of external loads on closed Hassani, M.J. Scoble & T.R. Yu (eds), Innovation in Mining
conduits in the light of latest experiments. Bulletin No. 96, Backfill Technology: 3–13. Rotterdam: Balkema.
Iowa Engineering Experiment Station, Ames, Iowa.
McCarthy, D.F. 1988. Essentials of Soil Mechanics and
Foundations: Basic Geotechnics. 4th edition, Prentice Hall.
219
ABSTRACT: Agnico-Eagle’s Laronde Mine is currently investigating mining options down to 3000 meters
below surface. FLAC3D was used to simulate the entire current mine from 1340 m to 2150 m below surface as
well as the potential future expansion. FLAC3D allowed each of the three main ore lenses to be modeled together,
providing information on the interaction between lenses which had never been available before. The model provides
information useful for determining the ideal stope dimensions, mining method, mining sequence, support options
for large excavations, as well as the best option for the shaft location.
221
Figure 1. Cross-section looking East of the Laronde orebody showing the different rock units included in the modeling.
CANMET Stress Model Used to Obtain 180 MPa. The orebody itself is competent, also with a
Gradients for Flac 3D Model UCS of 180 MPa. Both of these values come from
0
CANMET test results (Labrie 2000a).
S3 = 0.027 * Depth The stresses at Laronde were measured by
200
CANMET on the 146 level and the 150 level (Labrie
S2 = 0.0362 * Depth 2000b). The stress gradients are shown in Figure 2.
400
S1 = 0.0437 * Depth
600 3.1 Overall mine model
Depth (m)
800
The overall mine model was used for a number of pur-
CANMET @ 150
S1 Average = 70.2
poses. First, it was used as a calibration of the stress
1000
S2 Average = 61.18 and material properties based upon information col-
S3 Average = 47.61 lected from site visits and previous experience at the
1200
CANMET @ 146
S1 Average = 59.31
mine. Secondly, it was used to determine how much of
S2 Average = 48.13 an interaction between ore zones was likely taking
S3 Average = 25.08
1400 place. Finally, it was used to determine the in-situ
and post-mining stresses along the potential shaft loca-
1600 tions and the overall stress regime throughout the
0 20 40 60 80 100 entire mine.
Stress (Mpa)
CANMET S1 CANMET S2 CANMET S3
Stress Gradient (S1) Stress Gradient (S2) Stress Gradient (S3)
3.1.1 Calibration
The calibration of the material properties became easier
Figure 2. Stress gradients used in the FLAC modeling at to perform after a seismic event occurred on the 152
Laronde Mine based upon CANMET stress measurements. Level of the mine. Using the existing model as a frame-
work, an investigation showed – and the model con-
firmed – that there was a large stress concentration
3 FLAC3D MODELING centered right around the location of the burst caused
by a 3-wide pillar being formed by retreating stopes
The use of FLAC3D in this project was a logical choice towards the central access for the level.
based on the geometry of the orebody and work previ- In March 2003, a large fall of ground occurred in a
ously performed by Itasca for the mine (a reasonably stope on the 215 Level where a double width stope
good calibration of the model, material properties, and had been taken. Again, the model was able to show a
stresses had already been performed). To extend upon similar pattern in the stresses and failure zones in this
this base information, new stress data and core testing area. Between these two events, a comfortable degree
by CANMET resulted in a more refined picture of the of confidence was gained that the model reflects realis-
complex interactions between the different rock struc- tic stresses and failure zones based upon the known
tures and stresses. geology and geometry of the orebody at depth.
Around each ore lens is a layer of highly sheared
schist that varies in thickness up to 5 m on both the 3.1.2 Ore lens interactions
hangingwall and footwall. Surrounding the schist is a In all of the previous work performed, the investigation
garnetiferous tuff material that is relatively strong and of the impact of the different ore lenses on each other
stiff with a uniaxial compressive strength (UCS) of was not considered relevant because of the distance
222
223
Figure 5. Failure plot on a long section looking North of the overall mine model as mining approaches a sill. Blue blocks are
intact, red and light blue blocks have failed in shear and green blocks also have a tensile failure component.
Figure 6. Maximum principal stress plot on a long section looking North through the orebody. Peak stresses in a couple of
areas exceed 250 MPa.
of the mine. The model used a simplified representation the thicknesses of the different units and their location
of the orebody to give a constant strike and thickness of in the model could be defined, as well as the depth at
the ore and schist zones to eliminate geometrical effects which the simulation was to occur. This automation
in the comparison. A total of four simulations were run, reduced the turnaround time between model runs to
three with different ore thicknesses ranging from 10 m only a few minutes.
up to 30 m, and a fourth simulation in which the stope
size was reduced to determine the impact on the stope 3.2.1 Results
stability. Some results from the different ore thicknesses
To simplify the running of multiple scenarios with are shown below, in each case, the early stages of
similar geometry, a FISH function was written in which mining are shown when only four stopes have been
224
Figure 7. Maximum principal stress plot on a long section looking North through the orebody with a 10 meter ore thick-ness
showing some pinching of high stresses above the trailing primary stopes. Secondary stopes are in post-failure even in narrow ore.
Figure 8. Maximum principal stress plot on a long section looking North through the orebody with a 30 meter ore thickness
showing that the destressed zone is much larger due to the extra freedom provided by the larger stoping spans. Overall, the
stresses are more spread out and peak stresses are predicted to be much lower in this case.
mined.A 10 m and 30 m ore thickness are shown for stronger secondary stopes carrying more loading
comparison. and the smaller failure zones. In contrast, the 30 m
Looking at the principal stress plots it can be ore zones (Fig. 8) result in a very large and smooth
seen that there is a definite pinching of the stresses stress distribution with lower peak stress levels.
with the narrower ore geometry Fig. 7) due to the This pattern of pinched stresses and higher peak
225
Figure 9. Cross-section looking North showing failed regions in the model with 10 meter ore thickness. The impact of indi-
vidual stopes can be seen by the irregular shape of the failure zone. Most of the blocks are failed in shear (light blue and red
blocks, although some tensile failure is evident along the stope boundaries (green blocks).
Figure 10. Cross-section looking North showing failed regions in the model with a 30 meter ore thickness. The failure region
is much larger and more even with the wider ore at these great depths. The color coding is the same as in Figure 9.
levels continued throughout the simulation as mining impact of a single stope is lost in the overall picture.
proceeds. In the narrower orebody, the impact of individual
Looking at the failure plots (Figs. 9 & 10) the rea- stopes on the overall shape of the failure zone is still
son for the pinching of the stresses from the pre- apparent.
vious plots (Figs. 7 & 8) can be seen. The failure From these results it can seen that there is a signifi-
zones are much larger with the thicker ore- cant impact on the stresses and failure zones with
body; this creates a more even shell of failure around increasing ore thickness as would be expected with the
the stopes which becomes spherical in shape. The change in the pillar width-to-height ratio.
226
Figure 11. Grid used in the FLAC modeling of a 9 meter diameter shaft with liner support.
Figure 12. Maximum principal stress plot around a 9 meter diameter shaft at a depth of 2000 meters.
227
Figure 13. Predicted failure envelope around a 9 meter diameter shaft at 2000 meters subjected to in-situ stresses. The depth
of failure predicted from this model is around 3 m.
Figure 14. Maximum principal stress plot at shaft location #3 at a depth of 2100 meters when subjected to mining induced
stress changes.
In order to determine the stresses around a large Figure 14 shows an example of the #3 shaft loca-
selection of shaft locations and depth/stress condi- tion at a depth of 2100 m (which corresponds to the
tions, the stress ratio plot (see Fig. 4) was used to largest increase in stress ratio). It can seen that the
determine the best and worst conditions that may be stresses have rotated clockwise about 45 degrees,
expected from each shaft location. Six shaft models which is shown in both the stress plot and the plastic-
were run – this included three generic in-situ runs at ity plot (Fig. 15). These results correspond very well
2000, 2500, and 3000 m. Another set of three models with what was intuitively expected based on the
were used to represent post-mining conditions where geometry of the region. At this elevation, the shaft is
the largest increases, decreases and rotations of the just passing through the western abutment stress of
stress tensor were occurring. the 215 mining horizon, which is reflected by the
228
Figure 15. plot for shaft location #3 at a depth of 2100 m at the post-mining stage. The rotation of the stresses at this location
has had a significant impact on the failure zones with an increased depth of failure along a line running North-West to South-East.
Figure 16. Stresses in the 12 inch concrete liner for post-mining stresses of shaft #3 at 2100 m. Peak stresses approach 4 MPa
in the liner in this case.
slight increase in stresses and the general rotation of difficulties are anticipated beyond those expected
the stress tensor. with mining at extreme depth.
Looking at the stresses in the liner (see Fig. 16),
it can be seen that the stresses in the liner due to
4.2 Generic large excavation model
mining-induced stress are quite small with a peak stress
of around only 4 MPa, which is well below the strength The large excavation model provided some general
of the concrete to be used in the liner. From this analy- guidelines that can be used in the design of infrastruc-
sis it was determined that there should be no excessive ture in the mine. To do this, an arched back drift was
stresses or failure zones that cannot be designed for created in FLAC with a span of 11 m and a height of
with current technology. Barring any poor geological 6.5 m, as shown in Figure 17. The drift was set up at a
horizons through the shaft locations, no significant 3000 meter depth using the in-situ stress. Post-mining
229
Figure 18. Large excavation running parallel to the maximum principal stress. Peak stresses approach 200 MPa with a failure
zone that extends around 5 m into the back of the drift.
stresses were not considered since they are location- resulted in a peak stress of around 200 MPa and a fail-
dependant and no information was available for place- ure zone that extended up to 5 meters into the back
ment of infrastructure in the deep mine. The first run of (Fig. 18). This represented a very extensive shell of
the model assumed that the drift ran parallel to the failed material, which, depending on geology, could be
maximum principal stress (North–South), while the difficult to support. The ground at Laronde tends to
second model ran perpendicular to the maximum prin- involve some significant displacements which make
cipal stress (East–West) in order to provide informa- stiff support such as shotcrete a less attractive support
tion on these two extreme situations. system as it cannot accommodate much displacement.
With the drift set up to run parallel to 1, the back is With the alternative scenario, which had the drift
shielded from the effect of the highest stresses. As a running perpendicular to 1, the drift was exposed to
result, the stress seen in the drift was a combination of the full impact of the highest stress component (see Fig.
2 and the overburden-related vertical stress. This 19). As a result, the peak stresses across the back of the
230
Figure 19. Large excavation running perpendicular to the maximum principal stresses. In this case the peak stresses are about
25% higher and the failure zone extends nearly 6 m into the back.
stope reached 250 MPa (which is about 25% higher were encountered. These problems will be magnified
than the previous case). The failure zones extend about at depth.
15% farther into the back that in the previous case,
making for an even more difficult ground support
requirement. ACKNOWLEDGEMENTS
231
ABSTRACT: This paper describes a FLAC3D model for a typical deep two-entry longwall coal mine. The coal
seam is modeled as a strain softening material to attain a representative analysis of stresses and deformations
experienced by the coal ribs and yielding chain pillars corresponding to various loading stages. The strain soft-
ening parameters are established by calibrating separate test pillar models to common empirical pillar strength
formulas. The test pillar models showed that strain softening material behavior results in lower pillar strengths
than the traditional Mohr–Coulomb models based on constant peak cohesion and friction values. The longwall
model incorporates compaction simulations of the gob material in the back area. Two algorithms for representing
gob compaction are described.
1 INTRODUCTION
233
2 LONGWALL MODEL
234
Loading 30
2.5
2.4
2.2
35 MPa/εp
25 2
Cross-cut 3.25 m 1.8
1.6
Entry 15
6m 10
0
0 1 2 3 4 5
w/h
30
2.5
2.4 50 MPa/εp
25 2.2
2
1.8
Figure 5. Test pillar model geometry. Peak strength (MPa) 1.6
20 1.5
Salamon
Bieniawski
and/or friction drop as a function of plastic strain 15
in the post-peak region. The determination of the MC
and MCSS parameters for a rock mass is a difficult 10
task, but can be carried out empirically by performing
back-analyses. In this study, the parameter determina- 5
tion is based on the two most commonly used empirical
pillar strength formulas given by Salamon (1967) and 0
Bieniawski (1984). 0 1 2 3 4 5
A FLAC3D model of a single test pillar was developed w/h
to establish the most suitable combination of coal
30
MCSS parameters for replicating pillar strength val- 2.5
ues based on empirical formulas. Figure 5 shows the 2.4 100 MPa/εp
25 2.2
FLAC3D model of the test pillar in a room and pillar 2
1.8
environment. By considering symmetry conditions, one
Peak strength (MPa)
20 1.6
quarter of the pillar is modeled. The vertical walls of 1.5
Salamon
the model are set as frictionless by fixing the normal Bieniawski
15
displacements on them, except for pillar sides when
they are formed. The model is loaded along the top 10
boundary using a constant displacement of 2 107 m
per FLAC step. 5
The floor material is modeled as an elastic layer
having a 20 GPa elastic modulus. The MC interface 0
between the pillar and floor has strength parameters 0 1 2 3 4 5
of 0.5 MPa cohesion and friction angle of 23 degrees. w/h
For all pillar test simulations, the friction and dila-
tion angles are held constant at 30 and 15 degrees, Figure 6. Model pillar strength versus empirical pillar
respectively. strength at cohesion drop rates of 35,50,100 MPa/p
Four pillar width-to-height (w/h) ratios (1, 2, 3, (Strength formulas: Salamon: 9(w0.46/h0.66), Bieniawski:
and 4) were modeled. For each w/h ratio, the numerical 9(0.64
0.36 w/h) in MPa; assuming a coal cubic strength
model was run with different combinations of a peak value of 9 MPa).
cohesion and cohesion drop rate.
The strengths established from the test pillar models of 2.2 MPa and cohesion drop rate of 50 MPa/p is
are plotted against the empirical pillar strength for- considered suitable for modeling yielding of the
mulas in Figure 6 for the cohesion drop rates of 35, chain pillars.
50, and 100 MPa per plastic strain (p) increment. The test pillar models were repeated using the MC
Based on the trends of these plots, a peak cohesion failure criterion with the same peak cohesion, friction
235
2 / MC
2.2 Gob compaction
15
The gob compaction process is an essential part of the
3 / MCSS
1 / MC longwalling process since it can alter the pillar and
10 abutment loads by acting as an additional support for
2 / MCSS the system. The gob behavior is based on the following
5 “compaction” model: vertical stress (v) in the gob
1 / MCSS
increases with increasing vertical strain (v) according
to the relationship given by Salamon (1990),
0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014
Strain (1)
Figure 7. The vertical stress–strain curves of MC and
MCSS pillars.
where “a” is gob initial deformation modulus; and “b”
is the limiting vertical strain. Based on studies carried
50 out at the USBM on gob behavior, the values for the
45 MC models peak
Salamon's constants were taken as a 3.5 MPa and b 0.5
Bieniawski's
strength could not MC models (Deno & Mark 1993).
40 MCSS models
be defined beyond Two different algorithms are considered for imple-
35
this point mentation of the gob behavior of Equation 1 in the
Peak Strength (MPa)
30
FLAC3D model. In the first algorithm, referred to as
the “nodal force”, the compaction load is modeled as
25 the sum of vertical forces applied at the grid points of
20 the roof elements in the back area after mining. After
each mining step, the vertical strain in a particular
15
zone within the gob area is used to calculate the verti-
10 cal stress according to Equation 1. Grid reaction forces
5 are then calculated by multiplying vertical stress by the
corresponding area of the roof element. In the second
0
0 1 2 3 4 5 6 7 8 method, the gob is modeled as a non-linear elastic
w/h
material. Its bulk modulus is continually increased as
function of vertical strain within the gob area. The
Figure 8. Pillar strength determination from numerical algorithm for this “modulus updating” method uses the
modeling and empirical formulas (refer to Figure 6 for bulk modulus K for each gob element:
empirical strength formulas).
236
20 2.5 50
Analytical solution Cohesion
18 Nodel force method 45
Friction
16 Modulus updating method Dilation
14
35
Cohesion (MPa)
12
1.5 30
10
25
8
1 20
6
4 15
2 0.5 10
0 5
0 10 20 30 40 50 0 0
% closure 0 0.02 0.04 0.06
Plastic strain
Figure 9. The gob stress-closure results from the analytical
solution and two FLAC3D algorithms.
Figure 10. MCSS parameters used for modeling of the
coal material.
The gob compaction curves for the analytic solu-
tion (Salamon 1990) and the two FLAC3D algorithms Table 1. Material properties used in longwall simulations.
are compared in Figure 9. As shown, both nodal force
and modulus updating algorithms compare well with Property Values
the analytical model. Since the nodal force algorithm
requires longer running time, the modulus updating Miscellaneous
method was embraced as the gob model for the Seam depth 680 m
FLAC3D longwall simulations performed in this study. Stress gradient 0.025 MPa/m
x, y and z 17 MPa
Coal properties
3 RESULTS Coal elastic modulus 3 GPa
Coal Poisson’s ratio 0.25
Coal strength 7.6 MPa
Figure 10 defines the MCSS material parameters used Coal density 1313 Kg/m3
in the model, which are also summarized in Table 1.
Roof properties
For the coal seam, these parameters correspond to an
Elastic modulus 20 GPa
MCSS material having a cubic strength of about Poisson’s ratio 0.25
9 MPa, friction angle of 30 degrees, and cohesion Density 2500 Kg/m3
drop rate of about 50 MPa/p.
Interface properties
The model of the longwall layout described in Type Mohr–Coulomb
Section 2 is brought to equilibrium elastically to hori- Cohesion 0.5 MPa
zontal and vertical virgin stress conditions of 17 MPa Friction angle 20°
at the coal seam level. The elastic coal seam is then
replaced by a MCSS material prior to development. The
entries are developed with the right entry leading the and then averages these values to produce an average
left entry by 9 m. The entries advance by 3 m in each vertical pillar stress–strain curve.
mining step. A cross-cut is then mined when the trailing Figure 11 shows a typical pillar stress–strain curve
entry is 9 m ahead. Mining of the longwalls is carried obtained from the FLAC3D simulation. The vertical
out starting at the right panel. The longwall advances dashed line on the left shows the pillar loading at the
initially in steps of 50 m and then the steps are reduced end of entry development.
to 10 m in the fine-meshed central region of the model. At this stage, the pillar is at or close to its peak
After each longwall advance the area behind the long- capacity. The pre-peak stress drops indicate sidewall
wall face is changed to “gob material” and the model failures experienced by the pillar during entry devel-
is brought to equilibrium. The pillar response to mining opment. As the longwall approaches, the pillar initially
is monitored using a FISH algorithm. The algorithm sheds load slowly and subsequently rapidly, eventually
keeps a record of the vertical stress and vertical strain reaching eight per cent compression. At its residual
histories of all zones comprising the top of the pillar, strength, the pillar carries a vertical stress of 4 MPa,
237
24 25
20 20
18
Figure 11. Complete average vertical stress–strain curve Figure 12. Vertical stress and closure induced at a point in
of the yielding chain pillar in modeled longwall layout. the gob.
238
of Health and Human Services, CDC. Support provided Bieniawski, Z.T. 1984. Rock Mechanics Design in Mining
by Department of Health and Human Services, CDC, is and Tunneling. A.A. Balkema, p. 1–272.
greatly acknowledged. The work presented is part of the Deno, M.P. & Mark, C. 1993. Behavior of Simulated Longwall
Health and Safety research activities currently carried Gob material. United States Department of the Interior,
Bureau of mines, Report of investigation No. 9458.
out at Western Mining Resource Center (WMRC) at the Itasca Consulting Group, Inc. 2002. FLAC3D – Fast
Colorado School of Mines. Lagrangian Analysis of Continua in Three Dimensions,
Ver. 2.1. Minnesota: Itasca.
Salamon, M.D.G. 1990. Mechanism of caving in longwall coal
mining. Paper in Rock Mechanics Contributions and
REFERENCES Challenges Proceedings of the 31st US Symposium, Ed.
W. Hustrulid and G. A. Johnson. Denver, Colorado, June
Badr, S.A., Schissler, A., Salamon, M.D.G. & Ozbay, U. 18–20, 1990. A.A. Balkema, 1990, p. 161–168.
2002. Numerical Modeling of Yielding Chain Pillars in Salamon, M.D.G. 1967. A study of the strength of coal pillars.
Longwall Mines. Proc. of the 5th North American Rock Journal of South Africa Institute of Mining and Metallurgy,
Mechanics Symposium, Toronto, Canada, pp 99–107. v. 68, p. 55–67.
Badr, S.A. 2003. Numerical Analysis of coal yield pillars at Schissler, A. 2002. Yield pillar design in non-homogenous
deep longwall mines. Ph.D. Thesis in preparation. and isotropic stress fields for soft minerals. Ph.D. Thesis.
Department of Mining Engineering, Colorado School of Department of Mining Engineering, Colorado School of
Mines, Golden, Colorado (To be submitted.). Mines, Golden, Colorado.
239
ABSTRACT: FISH functions were developed to generate FLAC3D elements and to accurately present numer-
ical simulation results for underground mining situations whereby the ore lenses have complex geometries and
erratic distributions. The element-generating FISH functions for FLAC3D proved flexible and provided signifi-
cant timesavings during the model construction stage. Furthermore, these functions made it easier to modify
a model, to achieve a high degree of resolution in the domains of interest and to create a minimum number of
elements in order to minimize the computational power required to run the model.
Long, narrow and winding ore bodies with complex geometries do not lend themselves well to the represen-
tation on longitudinal sections of the modeling results, mainly because the rendering planes wander in and out
of the ore body. Such ore bodies are however often visualized and managed based on their longitudinal appear-
ance, which is typically projected and simplified on an idealized plane. This difficulty of longitudinally show-
ing the predicted stresses and displacements within the ore lens can impair the full and clear understanding of
the modeling results. This paper describes two means of solving this problem based on FISH functions. The
FISH functions presented in this paper have been applied to, and validated by, a FLAC3D modeling exercise
carried out at the Falconbridge Thayer Lindsley Mine near Sudbury, Ontario, Canada.
1 GENERAL INSTRUCTIONS Narrow and undulating ore lenses also make it dif-
ficult to represent the simulation results on longitudi-
There are essentially two ways to build a FLAC3D nal views. The existing FLAC3D “plot” command can
numerical model: one is to generate regular elements prove inadequate to illustrate load and deformation
over the entire domain and then structure the desired results because longitudinal sections generated through
geometry and geology around them, the other is to set the approximate center of a given narrow and undu-
FLAC-provided blocks for specific objects to simulate lating ore lens typically wanders in and out of it. This
and to assemble these blocks into the model. In under- made it difficult to visualize the stress redistribution
ground mining numerical modeling applications the and deformation everywhere within the ore lens itself.
first approach is generally used because there is usually Two approaches can be used to solve this problem.
no need to account for topographically irregular ground One consists of extracting from save files the stress
surfaces or very complex and precise excavations, as is and deformation data at each point along a curved
often the case in civil engineering applications. This surface centered in the middle of the undulating nar-
method, although quite versatile, can however result in row body of interest, and to generate iso-contour plots
the creation of a large number of elements in order to with specialized software (such as Goldsoft Surfer®,
achieve the desired degree of resolution, particularly for example). This approach has the advantage of
when the geometry of the ore lenses is complex, or allowing the user to extract and plot any desired
when multiple independent ore lenses are present. A parameter or criterion, such as factors of safety or
large number of elements can, in turn, result in exces- custom-defined stress ratios, and to clearly represent
sively long running times and even prevent a model their variation. Another way is to define a thin central
from running if the computer platform is insufficiently zone in the middle of the undulating narrow body as a
powerful. In such cases, the second strategy may not be FLAC3D Group or a FLAC3D Range, which can subse-
adequate either, due to the irregularity of the geometry quently be used to represent a true longitudinal sec-
of the ore lenses. Furthermore, it is usually more time- tion. This approach allows use of existing FLAC3D
consuming to build a model using the second approach. commands and functions to generate the plots. Both
241
dxyz(1,3)
tion) and 300 m-long (in the East–West direction)
section of the mine, centered on the #2 Zone, repre- dayz(i,1)
Coordinates at the point O: dxyz(i,J)
sented the region of interest for the FLAC3D simu- Index j (1,2,and3) corresponds to x,y, and z.
Coordinates at point O:
lations. A resolution of 5 meters in the East–West dxyz (i, j)
Boundary Internal model Index j (1, 2, and 3) cor-
direction, by 2 meters in the North–South direction, block (outer responding to x, y, and z.
by 5 meters in the vertical direction was considered a model)
242
τ
Hanging wall
Strength envelope
Element
A A
B B
3 1 1'
C C
Figure 4. Illustration of the user-defined stress ratio where
Footwall 1 and 3 are the major and minor principal stress of rock
mass, respectively, at a given point as computed by FLAC3D.
Figure 3. Sketch illustrating the concept of constructing a 1 is the major principal stress at this point assuming that the
non-planar longitudinal section from an arbitrary narrow rock mass is undergoing yielding under the same confine-
and undulating ore lens. ment conditions.
243
Figure 5. Longitudinal section looking North illustration of the Surfer™-generated contours of a user-defined stress ratio
based on the customized FISH functions described in the previous sections.
Figure 5 shows a simplified composite longitudinal and FLAC3D users can develop very specific functions
section of the T.L. Mine #2 Zone. The small squares to solve specific problems.
along the edge of the excavations represent the state
of the local rock mass at a given mining step, as
ACKNOWLEDGEMENTS
inferred from the FLAC3D results. The overlaid contours
of the user-defined stress ratio were generated based
The authors would like to thank Scott Carlisle1 for
on the stresses at each element extracted with the FISH
reviewing, and Falconbridge Limited for granting
function mentioned previously. In the areas where the
permission to publish this paper and for the use of
rock mass is still in its elastic state, which are beyond
Thayer Lindsley data.
the small squares as shown in Figure 5, the low values
of the stress ratio refer to a low likelihood that stress-
induced problems will arise at this mining stage. With
the help of the FISH function, the state of the rock REFERENCE
mass can be illustrated quite precisely.
In Figure 5, all the numerical elements from which Itasca Consulting Group, Inc. (1997) FLAC3D – Fast
the stresses were extracted are those located the clos- Lagrangian Analysis of Continua in 3 Dimensions,
est to the middle of the #2 Zone ore lens at T.L. Mine, Version 2.0. Minneapolis, MN: Itasca.
or those extending into the surrounding rock mass
away from the east and west boundaries of the lens.
These contours present quite a bit of information APPENDIX I – FISH FUNCTION FOR THE
about the state of the rock mass, both in the main lens GENERATION OF THE ELEMENTS OF A
and the surrounding rock. Furthermore, how far away FLAC3D MODEL
the elastic state rock is from the onset of yielding can
be readily estimated and displayed by the contours ; Define blocks in the internal region of the model
generated by the customized FISH functions. ; num_box10 blocks currently limited
Currently, this cannot be achieved with any built-in def Ore_box
FLAC command. array xyz(10,3),dxyz(10,3),p_xyz(10,3)
loop i (1,num_box)
P0_xxyz(i,1)
5 CONCLUSIONS P0_yxyz(i,2)
P0_zxyz(i,3)
User-defined FISH functions can be a powerful tool p_xyz(1,1)p0_x
to solve various FLAC or FLAC3D numerical modeling p_xyz(1,2)p0_y
problems. The FISH function presented in this paper p_xyz(1,3)p0_z
for the generation of regular elements is applicable to
many scenarios where similar geometrical issues are
present. The other FISH function discussed in this 1
Falconbridge Limited, Sudbury Mines/Mill Business Unit –
paper is a good example of how experienced FLAC Mining Services, Onaping, Ontario, P0M 2R0, Canada.
244
P_xyz(2,1)p0_x
dxyz(i,1) P4_xx0
dx
x_len
P_xyz(2,2)p0_y P4_yy0
dy0
y_len
P_xyz(2,3)p0_z P4_zz0-z_len
P_xyz(3,1)p0_x P5_xx0
dx
P_xyz(3,2)p0_y
dxyz(i,2) P5_yy0
dy0
P_xyz(3,3)p0_z P5_zz0
dz
P_xyz(4,1)p0_x P6_xx0
dx
x_len
P_xyz(4,2)p0_y P6_yy0-y_len
P_xyz(4,3)p0_z
dxyz(i,3) P6_zz0
dz
z_len
P_xyz(5,1)p0_x
dxyz(i,1) P7_xx0
dx
x_len
P_xyz(5,2)p0_y
dxyz(i,2) P7_yy0
dy0
y_len
P_xyz(5,3)p0_z P7_zz0
dz
z_len
P_xyz(6,1)p0_x Y_S_boun p1_y
P_xyz(6,2)p0_y
dxyz(i,2) Z_B_boun p1_z
P_xyz(6,3)p0_z
dxyz(i,3) command
P_xyz(7,1)p0_x
dxyz(i,1) gen po id 1011 p0_x p0_y p0_z
P_xyz(7,2)p0_y gen po id 1012 p1_x p1_y p1_z
P_xyz(7,3)p0_z
dxyz(i,3) gen po id 1013 p2_x p2_y p2_z
P_xyz(8,1)p0_x
dxyz(i,1) gen po id 1014 p3_x p3_y p3_z
P_xyz(8,2)p0_y
dxyz(i,2) gen po id 1015 p4_x p4_y p4_z
P_xyz(8,3)p0_z
dxyz(i,3) gen po id 1016 p5_x p5_y p5_z
loop n (1,8) gen po id 1017 p6_x p6_y p6_z
id_pn
8*(i-1) gen po id 1018 p7_x p7_y p7_z
P_xp_xyz(n,1) end_command
P_yP_xyz(n,2) end
P_zP_xyz(n,3) def Back_box
command P0_xx0
gen po id id_p p_x p_y p_z P0_yy0
dy0
end_command P0_zz0
end_loop P1_xx0
dx
end_loop P1_yy0
dy0
end P1_zz0
; Building boundary blocks P2_xx0-x_len
def right_box P2_yy0
dy0
y_len
; calculate the dimensions of this box in x, y and z P2_zz0-z_len
; xyz_ratio grid length ratio P3_xx0
; n_grid number of grids P3_yy0
dy0
; x_1, y_1 and z_1 element length along P3_zz0
dz
; corresponding directions in boundary blocks P4_xx0
dx
x_len
; x0, y0 and z0 coordinates of point O of the P4_yy0
dy0
y_len
; internal model as shown in Figure 2. P4_zz0-z_len
P5_xx0-x_len
x_lenx_1*(1-xyz_ratio^n_grid)/(1-xyz_ratio)
P5_yy0
dy0
y_len
y_leny_1*(1-xyz_ratio^n_grid)/(1-xyz_ratio)
P5_zz0
dz
z_len
z_lenz_1*(1-xyz_ratio^n_grid)/(1-xyz_ratio)
P6_xx0
dx
P0_xx0
dx P6_yy0
dy0
P0_yy0 P6_zz0
dz
P0_zz0 P7_xx0
dx
x_len
P1_xx0
dx
x_len P7_yy0
dy0
y_len
P1_yy0-y_len P7_zz0
dz
z_len
P1_zz0-z_len X_E_bounp7_x
P2_xx0
dx Y_N_bounp7_y
P2_yy0
dy0 Z_T_bounp7_z
P2_zz0 command
P3_xx0
dx gen po id 1021 p0_x p0_y p0_z
P3_yy0 gen po id 1022 p1_x p1_y p1_z
P3_zz0
dz gen po id 1023 p2_x p2_y p2_z
245
246
247
T. Allen
Washington State Department of Transportation, Washington, USA
ABSTRACT: The paper describes a FLAC numerical model that was developed to simulate the construction
and measured response of large-scale geosynthetic reinforced soil walls that were constructed at the Royal
Military College of Canada (RMC). The reinforced soil structures were constructed with three different poly-
meric reinforcement configurations. The backfill strength properties and reinforcement material properties
were determined from conventional laboratory tests. The soil elastic modulus values were back calculated from
surcharge loading tests on the wall backfill. The numerical models were able to capture the observed differences
in wall behavior due to different reinforcement configurations to within the accuracy of the measurements.
Reinforcement strain magnitudes and distribution were more accurately predicted using a stress-dependent
model for the soil backfill compared to a linear elastic model.
251
reactions and reinforcement strain distributions. The wall was a nominally identical structure except that
comparisons are based on both linear elastic and non- the reinforcement stiffness and strength of the geogrid
linear hyperbolic models for the soil backfill. were reduced by 50% by removing every other longi-
tudinal member in each layer. Wall 3 was nominally
identical to Wall 1 except that only four reinforce-
2 PHYSICAL TEST MODELS ment layers were used in the wall at a vertical spacing
of 0.9 m. In each structure, the wall facing consisted
Figure 1 shows a front view of an RMC test wall with of a column of discrete, dry-stacked, solid masonry
a modular block (segmental) facing. The walls were concrete blocks with continuous concrete shear keys.
3.6 m high with a facing batter of 8° from the vertical. The wall facing was built with three discontinuous
The first wall (Wall 1 – control) was built with six vertical sections with separate reinforcement layers in
layers of weak polypropylene geogrid (PP) reinforce- plan view. The width of the instrumented middle
ment placed at a vertical spacing of 0.6 m. The second section was 1 m. The backfill was a clean uniform
size rounded beach sand (SP) with a flat compaction
curve. The sand was compacted to a unit weight of
16.7 kN/m3 using a lightweight vibrating mechanical
plate compactor. The friction between the backfill
8° soil and sides of the test facility was minimized by
24 rows of segmental blocks
0.3 m
Facing blocks
6
Facing
potentiometer
Strain gauge
5
Extensometer
Connection load 4
rings
3.6 m
3
0.15 m
2 Reinforcement layer
Horizontal toe
load ring 1
2.52 m
Figure 2. Schematic instrumentation layout of the test walls used in calibrating the numerical model (Walls 1 and 2). Note:
Wall 3 is constructed with four reinforcement layers with a vertical spacing Sv 0.9 m.
252
facing column. Horizontal toe loads were measured approaches: 1) a linear elastic (perfectly plastic)
using load rings (a horizontal restrained toe boundary model, and 2) the stress-dependent hyperbolic model
condition). Vertical toe loads were measured using proposed by Duncan et al. (1980; also see Itasca
load cells supporting a double row of steel plates, 1998). Bathurst & Hatami (2001) and Hatami &
which were used in turn to seat the first course of Bathurst (2002) reviewed previous attempts reported
modular block units. A set of steel rollers was located in the literature to numerically model the response of
between the steel plates to de-couple the horizontal reinforced soil structures. Their survey showed that
and vertical toe load reactions. Reinforcement strains the stress-dependent nonlinear elastic model (hyper-
in the wall were measured using strain gauges that bolic model) proposed by Duncan et al. (1980), or
were bonded directly to the polypropylene geogrid variants, was the most common constitutive model
longitudinal members and extensometers attached to used to simulate the backfill response during con-
selected geogrid junctions. Backfill settlements were struction and under surcharge loading. However, no
measured using tell-tales and settlement plates. comparisons have been reported for simulations
Further details of the construction and monitoring using other models including a simple linear elastic
techniques used in the RMC test walls have been model. The backfill material properties used in the
reported by Bathurst et al. (2001). current study are reported in Table 1. The values of
soil hyperbolic parameters were determined by
3 NUMERICAL MODELING Table 1. Material properties for sand used in the numerical
model.
3.1 General
Value
The finite difference-based computer program FLAC
(Itasca 1998) was used to simulate the response of the Stiffness properties (Hyperbolic model)
reinforced soil test walls up to the end of construc- Kc (elastic modulus number) 2000
tion. Figure 3 shows the numerical grid used for the Kb (bulk modulus number) 2000
n (elastic modulus exponent) 0.5
segmental retaining walls.
m (bulk modulus exponents) 0.5
Rf (failure ratio) 0.73
3.2 Material mechanical models and properties (range of permissible Poisson’s ratio values) 0–0.49
0.3 m
Concrete facing blocks Interfaces
°
Sand backfill
24 x 0.15 = 3.6 m
3.6 m
0.6 m
Reinforcement
°
° 2.5 m
5.5 m
Figure 3. Numerical model of the segmental retaining walls (Walls 1 and 2). Note: Wall 3 is constructed with four
reinforcement layers with a vertical spacing Sv 0.9 m.
253
adjusting initial values taken from the results of triax- Table 2. Reinforcement stiffness and strength properties.
ial compression tests on backfill sand specimens. The
value of Poisson’s ratio for each soil zone during Polymer Number Stiffness Jt () Ty
analyses was determined from the calculated values Wall type of layers (kN/m)* (kN/m)
of the soil elastic modulus and bulk modulus from the
W1 PP 6 138–1698 14
hyperbolic model and hence was allowed to vary W2 PP 6 69–845 7
between values of 0 and 0.49 as noted in Table 1. The W3 PP 4 138–1698 14
modulus numbers Ke and Kb were increased to match
the measured settlement response of the backfill in * Equations valid for 2.5%
the retained zone behind the reinforced soil zone dur-
ing uniform surcharge loading. The backfill peak
plane-strain friction angle value was taken as Facing Soil-Block
Backfill
44° (Bathurst et al. 2001). The backfill dilation Modular Interfaces
Blocks (Continuum)
angle value from direct shear tests was found to vary Soil Column
Zones
from 9° to 12° for the range of confining soil Behind Facing
Nulled Zone
pressures in the test walls. The value 11° was (magnified)
used in the numerical models.
3.2.2 Reinforcement
The reinforcement layers were modeled with two-
noded elastic-plastic cable elements with a strain-
dependent tensile stiffness, J(), tensile yield strength,
Ty and no compressive strength. The reinforcement Block-Block
Interfaces Connection
load–strain response was modeled in parabolic form as: Beam
Backfill
Numerical
Elements Two-noded
Reinforcement Grid
(1) Elements
where T is axial load and is axial strain. This equa- Figure 4. Details of facing-backfill-reinforcement
tion is valid for 2.5% which captures the range of connection.
in-situ measured strains that correspond to the end-
of-construction working stress levels for the experi-
mental walls and is well below the reinforcement Table 3. Interface properties.
strain at yield. The strain-dependent, secant tensile
stiffness of the reinforcement, Js (), was calculated Value
from Equation 1 as:
Soil–Block
sb (friction angle) (deg) 44
sb (dilation angle)(deg) 11
(2) knsb (normal stiffness) (kN/m/m) 0.1 106
kssb (shear stiffness) (kN/m/m) 103
254
Elevation (m)
the parametric analysis. The magnitude of the normal
interface stiffness value was made as large as possible
to avoid the intrusion of adjacent zones but not to 2
cause excessive computation time. The wall deforma-
tion response was found to be relatively insensitive to
the value of inter-block shear stiffness for ksbb
50 MN/m/m. Smaller values of ksbb together with
1
material properties reported in Tables 1 to 3 were
shown to over-predict measured wall deformation
results. The value ksbb 50 MN/m/m gave the best
overall agreement with the measured data. This value
is also within the range of shear stiffness values back- 0
calculated from load-displacement results of labora- 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
tory interface shear tests on the block units (Hatami Facing displacement (mm)
et al. 2002). The interface shear strength was modeled
with the Mohr–Coulomb failure criterion defined
Figure 5. Measured and predicted facing displacements at
by interface cohesion and friction angle. The block– end of construction.
block interface peak friction angle and equivalent
cohesion values were determined from the laboratory
interface shear tests as bb 57° and cbb 45.7 kPa, Section 3.2.1, greater soil modulus values than those
respectively (Hatami and Bathurst 2001). The inter- obtained from laboratory triaxial compression tests
face friction angle, sb, between the backfill and fac- were used for the backfill model. With this approach,
ing blocks was back-calculated from measured toe negligible horizontal stress was needed behind the
reactions and the sum of measured connection forces facing panel to simulate backfill compaction in the
using the facing equilibrium analysis described by walls reported in this paper.
Hatami and Bathurst (2001). Their analyses demon-
strated that the soil-facing interface friction angle
value in the test walls was close to the magnitude of 4 RESULTS
the backfill peak plane-strain friction angle (i.e. sb
44°). 4.1 Calibration results
The response results for each of the three test walls in
3.2.4 Construction and boundary conditions this investigation were obtained by changing the rein-
Fixed boundary conditions in horizontal and vertical forcement stiffness (Table 2) or number of layers in
directions were assumed in the numerical model for the numerical model to match the physical test. The
gridpoints at the rigid foundation level, and in the material properties for all other wall components
horizontal direction at the backfill far-end boundary. were kept the same.
The toe boundary condition in the physical and num-
erical models is a reasonable approximation to the 4.1.1 Facing displacements
restraint that can be expected for the typical field case Figure 5 shows the measured and numerically calcu-
of a buried footing. The backfill and facing units were lated facing lateral displacement at potentiometer
placed in lifts of 150 mm (i.e. the height of one mod- levels at the end of construction. The measured dis-
ular block) and the reinforcement layers were numer- placement results are readings from the potentiome-
ically installed as each reinforcement elevation was ters that were mounted against the facing blocks at
reached. Backfill compaction during construction reinforcement layer levels during construction. The
was modeled by applying a horizontal stress compo- predicted results are obtained using the material prop-
nent on the back of the facing units as the soil layers erties shown in Tables 1–3. The results of Figure 5
and facing blocks were put in place and the model show satisfactory agreement between recorded and
solved to reach equilibrium. However, as described in predicted facing lateral displacements for all three
255
test walls. Both experimental and numerical results However, while not reported in this paper, the magni-
show greater facing displacement magnitudes for tudes of strain are very different between the three
Walls 2 and 3 constructed with lower stiffness rein- walls under surcharge loading at which time larger
forcement and fewer layers, respectively, compared to wall lateral deformations have occurred and the
the control wall (Wall 1).
Strain (%)
0.0
tion as large as $0.3% strain (Bathurst et al. 2003).
Hence, within the accuracy of the physical measure- 1.0
ments, the results shown in Figures 6–8 capture both Layer 3
0.5
the magnitude and distribution of strains in the meas-
0.0
ured data. The data show that strain magnitudes and
distributions are similar for all three walls. This can 0.8
Layer 2
be explained by the contribution of the very stiff con- 0.4
crete facing column that carries a large portion of 0.0
the lateral earth forces at the end of construction.
0.8
0.4 Layer 1
0.0 0.0
0.8 0.8
Layer 3 Layer 3
0.4 0.4
Strain (%)
0.0 0.0
0.8 0.8
Layer 2 Layer 2
0.4 0.4
0.0 0.0
0.8 0.8
0.4 Layer 1 Layer 1
0.4
0.0 0.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Distance (m) Distance (m)
Figure 6. Measured and predicted strain distributions at Figure 8. Measured and predicted strain distributions at
end of construction using hyperbolic soil model (Wall 1). end of construction using hyperbolic soil model (Wall 3).
256
tensile load capacity of the reinforcement layers is horizontal loads is the strong influence of the heavy
mobilized (Bathurst et al. 2001). The calculated max- facing column as noted previously.
imum reinforcement load in all wall models at end of
construction was less than 1 kN/m, which was well
4.2 Influence of soil model on predicted
below the yield strength of the reinforcement materi-
wall response
als (Table 2).
Selected response features of Wall 2 were examined
4.1.3 Toe reactions using a linear elastic soil model with the values of
Figure 9 shows the histories of the measured and cal- Young’s modulus and Poisson’s ratio taken as
culated horizontal and vertical toe loads for the test E 48 MPa and v 0.2, respectively. The value of
walls during construction. The figure shows a satis- Young’s modulus was back calculated from the mea-
factory agreement between the predicted and mea- sured pressure-settlement results of the backfill behind
sured horizontal and vertical toe reactions for the walls the reinforced soil zone during surcharge loading.
at the end of construction. The plots of measured hor- Figure 10 shows the measured response and the pre-
izontal toe load during construction of Walls 2 and 3 dicted facing displacement results using the linear
deviate from a smooth curve at early stages during elastic and hyperbolic soil models for Wall 2. The
construction. This is thought to be due to a local plotted values are deformations with respect to the
greater soil compaction effort at the back of the fac- time of installation of each displacement device.
ing units during construction. Otherwise the plots of Hence, these plots should not be confused with the
predicted and recorded horizontal toe loads are in actual wall deformation profiles at the end of con-
close agreement. The results shown in Figure 9 indi- struction. Both predicted curves capture the range of
cate that wall reinforcement stiffness or number of wall deformations recorded at the end of construc-
layers has a negligible effect on the magnitude of toe tion. The close agreement between the predicted fac-
loads during construction for the wall height and rein- ing displacement results in Figure 10 indicates that
forcement configurations examined. The reason for the values of soil hyperbolic model parameters reported
this consistent response, particularly with respect to in Table 1 are consistent with a constant soil modulus
value that was determined from the measured load-
settlement response of the backfill. This result may
50
not be unexpected since the hyperbolic modulus num-
a) Wall 1 vertical bers Ke and Kb in Table 1 were independently
40
measured
30 predicted 4
measured
20 hyperbolic soil model
horizontal
10 linear elastic soil model
0
3
Toe reaction (kN/m)
50
b) Wall 2 vertical
40
Elevation (m)
30
20
horizontal 2
10
0
50
c) Wall 3 vertical
40 1
30
20
horizontal
10
0 0
0 5 10 15 20 25 0 2 4 6 8 10
Number of facing units (blocks) Facing displacement (mm)
Figure 9. Measured and predicted toe reaction forces Figure 10. Measured and predicted facing displacements
during wall construction. of Wall 2 at end of construction.
257
obtained by matching the load-settlement response of magnitude of predicted strains in the reinforcement
the backfill retained zone under surcharge loading as layers at the end of construction for Wall 2.
explained in Section 3.2.1. Figure 11 also shows that Specifically, the predicted peak reinforcement strains
the predicted toe reaction responses were essentially using the soil linear elastic soil model are located far-
identical using both soil models. Taken together, the ther back from the facing compared to the peaks from
data in Figures 10 and 11 suggest that the simpler the hyperbolic model. Comparison with Figure 7
elastic soil model is sufficient to model these perfor- shows that the measured predicted peak strains are
mance features of the wall. However, the results using located close to the back of the facing column and not
the two soil models shown in Figure 12 illustrate a within the reinforced soil mass as predicted for all
subtle but important difference in the distribution and layers in Figure 12 with the exception of layer 1. The
absence of peak strain values within the reinforced
50 soil mass (as predicted using the linear elastic soil
Toe reaction (kN/m)
linear elastic vertical model) was corroborated by the lack of a visible shear
40
hyperbolic zone in the backfill at the time of careful soil excava-
30 tion of the wall. On the other hand, both the hyper-
20 bolic model results and the measured data show
horizontal relatively high reinforcement strain magnitudes at the
10 connections with the facing panel at end of construc-
0 tion, which are not captured using the linear elastic
0 5 10 15 20 25 model. It can be argued that horizontal stresses in the
Number of facing courses placed soil decrease locally behind the facing due to the out-
ward horizontal movement of the facing column dur-
Figure 11. Comparison of predicted toe reaction forces ing construction. As a result, the stress-dependent
using linear elastic and hyperbolic soil models during hyperbolic model predicts smaller soil stiffness val-
construction (Wall 2). ues behind the facing compared to the constant stiff-
ness model. Therefore, the predicted strain magnitudes
at the reinforcement connections with the facing can
linear elastic hyperbolic be expected to be greater (and hence more accurate)
0.2 using the hyperbolic model rather than the constant
Layer 6
0.1 stiffness (linear elastic) soil model.
0.0 Finally, the better match between the predicted and
measured wall response using back-fitted modulus
0.4 Layer 5 values from the measured load-settlement response of
0.2 the backfill in the actual physical tests highlights the
0.0 inability of conventional triaxial compression tests to
capture the backfill plane strain stiffness in the large-
0.6 scale wall tests.
0.4 Layer 4
0.2
Strain (%)
0.0
5 CONCLUSIONS
1.0
Layer 3
0.5 A numerical model has been developed using FLAC
0.0 to predict the measured response of carefully instru-
mented, large-scale geosynthetic reinforced soil mod-
1.0
Layer 2
ular block retaining walls during construction. The
0.5 numerical model accounts for staged construction of
0.0 the retaining walls and incremental lateral displace-
ment of the modular facing using FISH functions.
1.2
Additional subroutines are included in the program to
0.8 Layer 1
0.4 model the backfill stress-dependent stiffness proper-
0.0 ties and the nonlinear reinforcement strain-dependent
0.0 0.5 1.0 1.5 2.0 axial stiffness.
Distance (m)
The measured and numerical results for the con-
struction stage of each wall showed satisfactory agree-
Figure 12. Comparison of predicted strain distributions at ment for different response parameters including
end of construction using linear elastic and hyperbolic soil facing displacements, reinforcement strains and his-
models (Wall 2). tory of toe forces. In particular, reinforcement strain
258
distributions using a hyperbolic soil model were found Inter. Conf. Comp. Meth. Adv. Geomech. Invited Theme
to be in good agreement with the measured data. Paper, Tucson, AZ, USA, January 2001: (2) 1223–1232.
Bathurst, R.J., Walters, D.L., Hatami, K. & Allen, T.M. 2001.
Full-scale performance testing and modeling of rein-
forced soil retaining walls. Special Lecture, IS-Kyushu
ACKNOWLEDGEMENTS 2001. Fukuoka, Japan, November 2001.
Duncan, J.M., Byrne, P., Wong, K.S. & Mabry, P. 1980.
The financial support for this study has been provided Strength, stress-strain and bulk modulus parameters for
by the Natural Sciences and Engineering Research finite-element analysis of stresses and movements in soil
Council (NSERC) of Canada, 11 Departments of masses. Report No. UCB/GT/80-01. University of
Transportation in the USA, and grants from the California, Berkeley: Department of Civil Engineering.
Department of National Defence of Canada. Hatami, K. & Bathurst, R.J. 2001. Modeling static response
of a segmental geosynthetic reinforced soil retaining wall
using FLAC. Proc. 2nd Int. FLAC Symp. Numerical
Modeling in Geomechanics, Lyon, October 2001,
REFERENCES 223–231.
Hatami, K. Blatz, J.A. & Bathurst, R.J. 2002. Numerical
Allen, T.M., Bathurst, R.J. & Berg, R.R. 2002. Global Level modeling of geosynthetic reinforced soil retaining walls
of Safety and Performance of Geosynthetic Walls: A and embankments. Proc. 2nd Can. Spec. Conf. Comp.
Historical Perspective. Geosynthetics International, (9): Appl. Geotech., Winnipeg, MB, Canada, April 2002.
395–450. Hatami, K. & Bathurst, R.J. 2002. Numerical simulation of
Allen, T.M., Bathurst, R.J., Lee, W.F., Holtz, R.D. & Walters, a segmental retaining wall under uniform surcharge load-
D.L. 2003. A New Working Stress Method for Prediction ing. Proc. 55th Can. Geotech. Conf. Niagara Falls, ON,
of Reinforcement Loads in Geosynthetic Walls, Canadian Canada, October 2002.
Geotechnical Journal, (in press). Itasca Consulting Group, Inc. 1998. FLAC – Fast
Bathurst R.J. & Hatami, K. 2001. Review of numerical mod- Lagrangian Analysis of Continua, Ver. 3.40. Minneapolis,
eling of geosynthetic reinforced-soil walls. Proc. 10th MN: Itasca.
259
J.P. Hsi
SMEC Australia Pty Ltd, Sydney, Australia
M.A. Coulthard
M.A. Coulthard & Associates Pty Ltd, Melbourne, Australia
ABSTRACT: An excavation adjacent to a major bridge was carried out for the construction of a cut and cover
tunnel, Hawthorne Street Tunnel, as part of the South East Transit Project Section 2 in Brisbane, Australia. The
bridge was founded on shallow foundations whilst the excavation extended to below the foundation level. There
was limited tolerance for the bridge foundations to deflect resulting from the excavation, due to its structural
articulation. The support system for the excavation consisted of contiguous and scallop bored piles and ground
anchors. The excavation was carried out in stages taking advantage of the 3D effects to minimize ground defor-
mation. To predict the ground performance during excavation and to optimize the design of the ground support,
FLAC3D was employed to simulate the 3D effects, the construction sequence and the soil-structure interaction.
Field monitoring results showed performance comparable with that predicted by FLAC3D.
261
262
263
structural depth required. Fill has been maintained on 9. Installed and stressed the second (lower) row of
the structure for the location of services. ground anchors on the western wall of the south-
The western bored piles are located very close to ern half of the tunnel.
the existing abutment of the Hawthorne Street 10. As for 8, but for northern half of the tunnel.
Bridge. Each of the piles next to the bridge was to be 11. As for 9, but for northern half of the tunnel.
fitted with permanent ground anchors to maintain the 12. Excavated to the tunnel floor level.
required horizontal bearing pressure and reduce foun- 13. Installed the pre-stressed concrete planks over
dation movements for the Hawthorne Street Bridge. the headstocks.
Utilizing permanent ground anchors, the anchors can 14. Constructed cast in situ concrete slab over the
be stressed to counteract any predicted movement, tunnel floor.
and re-stressed if the measured movement is greater 15. Took deflection measurements frequently dur-
than that predicted. ing the above construction stages with record of
The eastern wall consisted of 0.9 m diameter piles at corresponding temperature and time.
1.5 m c/c spacing whilst the western wall comprised 16. At any time if the measured lateral deflection
1.2 m diameter piles at 1.77 m c/c spacing, except for of the bridge foundation was greater than 5 mm,
the section within 3 m from the existing bridge foun- contingency measures including further stressing
dation where the piles were at 1.25 m c/c spacing. All the ground anchors would be implemented.
the piles were socketed 0.5 m into slightly weathered
phyllite. The gaps between the piles were shotcreted
with fibrecrete. Two VSL permanent ground anchors
7 NUMERICAL MODELING
were installed on each of the piles adjacent to the bridge
abutment. Each anchor was socketed 10 m into slightly
7.1 Approach
weathered phyllite and prestressed to 1000 kN.
During construction the horizontal movement of The complex three-dimensional nature of the prob-
the abutment and vertical mid-span movement of the lem and the need to allow for possible yield of the
bridge were monitored. The measurements were com- various rock units and to account for a range of struc-
pared to the estimated values and adjustments made tural elements suggested that FLAC3D would be well
to the construction method and program if required. suited to the modeling. The work was performed in
Temporary struts and hydraulic jacks were specified 1999, using version 2.0 of that program.
as part of the contingency plan to help control pile The existing bridge was included only as a loaded
movement at the most critical section. foundation on one side of the new tunnel then the
construction sequence outlined above was repre-
sented in the numerical model. Coding in the in-built
6 CONSTRUCTION SEQUENCE
programming language FISH was used to manage
grid generation, excavation stages, installation and
To minimize ground movement associated with the
linking of piles, struts, crossbeams and anchors, and
construction work, excavation was carried out in
many other aspects of the modeling. The task of
stages, as follows:
development and testing would have been much more
1. Constructed bored pile walls and headstocks. difficult without this powerful feature.
2. Installed instruments, with minimum accuracy of
1 mm, for monitoring deflection of headstocks,
7.2 Geotechnical model
and abutments and mid-span joints of Hawthorne
Street Bridge. The geotechnical model adopted for the numerical
3. Undertook baseline readings of the instruments modeling included the subsurface stratigraphy and
without live load on the bridge and at relatively geotechnical parameters. Boreholes in this area indi-
similar climatic temperature during construction. cated that the subsurface consisted of residual soils
4. Excavated southern half of the tunnel to the level to 5–6 m depth underlain by extremely to highly wea-
of the first (upper) row of ground anchors, with thered (EW/HW) phyllite to depths ranging from 9 m
fibrecrete applied between piles progressively. near the eastern wall to 12 m near the western wall, over-
5. Installed and stressed the first (upper) row of round lying moderately weathered (MW) phyllite. Slightly
anchors on the western wall of the southern half weathered (SW) phyllite occurred at depths between
of the tunnel. 14 m at the eastern wall and 16 m at the western wall.
6. As for 4, but for northern half of the tunnel. A uniform subsurface profile based on the more
7. As for 5, but for northern half of the tunnel. critical profile on the western side of the excavation
8. Excavated southern half of the tunnel to the level was assumed and the Mohr-Coulomb soil/rock con-
of second (lower) row of ground anchors with stitutive model was adopted for each rock and soil
fibrecrete applied between piles progressively. unit. The assumed geotechnical model and material
264
parameters are given in Table 1. The tensile strength analysis of the existing bridge and (b) the bearing
for each rock unit was taken to be 25% of the cohe- pressure specified on the original bridge drawing.
sion, and zero for the soil. Loading (a) was significantly smaller than (b), which
In situ horizontal stresses were assumed to be half is understandable as the foundation pressure shown in
the vertical in the soil layer and equal to the vertical in the drawing would generally include a factor of safety.
the rock units, where an initial approximation to the For prudent and conservative reasons, the design of the
vertical stress was computed from the above layering supporting structures to the excavation was based on
and densities. case (b). The total applied vertical and horizontal loads
were 31.55 MN and 21.86 MN respectively, where the
horizontal load was taken to act in the direction of
7.3 Bridge footing
the short axis of the footing, i.e. at 25° to the normal
The Hawthorne Street Bridge was represented simply to the busway walls. These loads were converted to
via loads applied to the footing shown in the central part equivalent Cartesian stress components, which were
of the grid in Figure 4. Two loading cases were con- applied to the upper surface of the footing.
sidered, based on (a) SMEC’s independent structural
7.4 Retaining system
Table 1. Geotechnical model adopted for analysis.
The retaining system consists of a line of piles on
Soil/Rock Depth t c E each side of the busway and two rows of ground
Type (m) (kN/m3) (kPa) (deg.) (MPa) anchors attached to the piles on the western side to
support the existing bridge footing. Temporary struts
Residual 0–6 18 10 30 40 0.35 on the northern half of the busway were also modeled
EW/HW 6–12 20 500 35 50 0.30 in some cases but proved to be ineffective.
MW 12–16 22 750 40 200 0.25
SW
16 25 2500 45 400 0.20
The piles on the eastern side are 0.9 m diameter,
installed at 1.5 m center spacing. Those on the western
Figure 4. Rock units and zoning in inner section of FLAC3D grid, before tunnel construction.
265
Table 2. Piles (pile structural elements). that a coarsely zoned region was attached at the base
of the finely-zoned part shown, to provide a better
Parameter Value representation of the rock mass response at depth.
Young’s modulus 31 GPa
Poisson’s ratio 0.2 7.6 Modeling strategy
Density 2400 kg/m3
Shear/normal coupling spring stiffness 50 MPa The excavation and support sequence as modeled in
Shear/normal coupling spring cohesion 1 MN/m FLAC3D was as follows, where the computation was
Shear/normal coupling spring friction angle 30° stepped to equilibrium at each stage:
0. Apply boundary conditions to far boundaries
(fixed horizontal displacements on vertical sides
Table 3. Ground anchors (cable structural elements).
and fixed all displacement components on base),
Parameter 8-strand 27-strand initialize approximate in situ stresses, apply grav-
ity and footing load.
Young’s modulus 200 200 GPa 1. Excavate entire busway to 3 m depth.
Density 7860 7860 kg/m3 2. Install both sets of piles then excavate more distant
Tensile/compressive yield strength 1.19 4.0 MN half of busway (relative to view in Figure 4) a fur-
Grout stiffness 10 9 GPa ther 3 m.
Grout cohesive strength 1.13 1.7 MN/m 3. Install upper ground anchors, attached to piles
Grout friction angle 0° 0° between previous excavation and bridge footing,
and excavate 2 m from nearer half of busway.
4. Install crossbeams along lines of piles and struts
side are of 1.2 m diameter, at 1.25 m c/c spacing adja- across excavation (not in all models), and exca-
cent to the nearest corner of the bridge footing and vate more distant half of busway a further 3 m.
1.77 m c/c away from that corner. All piles are sock- 5. Install lower ground anchors attached to piles
eted 0.5 m into SW phyllite. between stage 2 excavation and bridge footing,
The ground anchors for the tunnel as built are and excavate further 3 m from nearer half of
8-strand cables of 15.2 mm diameter, installed in busway.
0.145 m diameter holes and pre-tensioned to 1 MN 6. Remove struts (if installed at stage 4), install
after installation. They dip at 45° and are socketed upper ground anchors attached to piles between
10 m into SW phyllite but are ungrouted over the stage 3 excavation and bridge footing, and exca-
remainder of their lengths. Other forms of anchor vate 2 m more from further half of busway.
were considered in several of the numerical models, 7. Excavate another 2 m from nearer half of busway.
e.g. varying numbers of strand, hole diameter and 8. Install lower ground anchors attached to piles
pre-tension force. between stage 3 excavation and bridge footing
Non-geometric property values for the FLAC3D and excavate final 1 m from nearer half of busway.
structural elements representing the piles and ground The final model configuration is shown in Figure 5,
anchors, are given in Tables 2 and 3 respectively. where the view is as in Figure 4.
Implementation of this construction sequence in a
7.5 Finite difference grid FLAC3D model was largely straightforward, with
As shown in Figure 2, the tunnel was to run obliquely FISH routines controlling the various excavation
under the existing bridge, with a bridge foundation stages and the placement of all types of structural ele-
immediately adjacent. In addition, a soil slope around ments. However, one aspect of the modeling proved
the bridge foundation was to be replaced, as the tun- to be unexpectedly complex, viz. the setting of links
nel was excavated, by a reverse-angle slope that was between the many structural nodes and rock zones or
not constant in profile along the excavation. The grid other structural nodes.
was generated in sections, some of which had to be The key constraint (D. Potyondy, private communi-
joined via “attach” commands, then the entire model cation) is that each structural node in a FLAC3D
was transformed to create the correct skew angle model can only be the source of one link. This link
between the bridge and tunnel. A view of the inner may provide either a node-to-zone connection or a
part of the pre-construction grid is given in Figure 4. node-to-node connection. This means that, where
The cut-and-cover tunnel was to be constructed cables or beams are linked to piles at sub-surface nodes,
within the finer-zoned region to the left of, and paral- then multiple nodes must be created to manage the
lel to, the slope shown in Figure 4. Note that the grid various links, and the direction of those links must
did not conform precisely to the assumed horizontal be carefully controlled. For example, creation of a pile
boundaries on either side of the MW rock unit, and automatically creates links from each sub-surface
266
Figure 5. Inner section of FLAC3D model, showing structural elements at completion of excavation and support.
structural node in the pile to the rock zone that con- associated structural nodes were not automatically
tains it. If a ground anchor is to be connected to that deleted by FLAC3D; this also had to be done explicitly
pile then the pile must be formed in such a way that in the data and FISH files.
there is a node at the intended connection point. A final complication arose from the fact that mul-
When the ground anchor is created as a series of cable tiple links at a point in space can only be distin-
elements, a duplicate node will exist at the same coor- guished via the link number that is set within FLAC3D
dinates as the pile node to which it is to be connected. when each particular link is created. Management of
By default, each cable node will also have a node-to- the links therefore required careful monitoring of the
zone link. These must be deleted at all nodes above the numbers of active structural nodes, elements and
grouted section of the cable, then a new node-to-node links and the highest id number for each. Again, this
link must be created, emanating from the node at the was handled effectively using FISH coding.
top end of the cable and ending on the correspond-
ing pile node (so that the pile node is the target of the 7.7 Cases modeled
node-to-node link from the cable node and the source
After a great deal of development and testing several
of the node-to-zone link to the rock).
production analyses were performed. As indicated in
A similar procedure had to be implemented to han-
section 7.4 above, temporary struts were predicted
dle links between piles and crossbeams and between
not to be effective, so the final analyses only included
crossbeams and struts, in cases where the latter were
piles and ground anchors, thus:
modeled. In that case, it was critical that the final set
of links be ordered thus: strut node → beam node → – bbhsu9c: 8-strand cables with 1 MN pre-tension;
pile node → zone. Rigid links were used for all con- case (a) footing loading from section 7.3;
nections between structural elements. – bbhsu9d: as for 9c except for higher case(b) foot-
Further, when structural elements were deleted ing loading;
from a model, such as when struts were removed at – bbhsu9e: as for 9d except 27-strand cables with
computational stage 6 above, then the links from the 3 MN pre-tension.
267
Figure 6. Induced horizontal displacements in footing at end of construction, from run bbhsu9d (upper plot) and bbhsu9e
(lower plot). Piles and ground anchors are shown as solid and dashed lines respectively; busway is to left of piles.
Figure 6 shows the predicted horizontal displace- i.e. they would overdo the support. In contrast, the
ments of the bridge footing at the end of runs 9d and 9e anchors in model 9d allowed the footing to relax
respectively. These results suggested that the heavier towards the excavation, but the maximum horizontal
ground anchors, with 3 MN pre-tension, would actually displacement was constrained to be less than 5 mm,
pull the footing away from the busway excavation, as required. Histories of x-, y- and z-components of
268
Copyright © 2003 Swets & Zeitlinger B.V., Lisse, The Netherlands
09069-31.qxd 08/11/2003 20:39 PM Page 269
Figure 7. Histories of induced x-, y- and z-displacements at two points on footing, during the stages of construction, as
computed in model bbhsu9d.
Figure 8. Axial forces in one set of ground anchors at end of construction, as computed in model bbhsu9d.
displacement of several points on the footing, for the The axial forces acting within one set of ground
same model 9d, showed that they would vary through anchors at the completion of construction are shown
the various computational stages but were also always in Figure 8; the vertical axis represents the vertical
predicted to be less than about 5 mm (see Fig. 7). coordinate (RL) along the cables and the horizontal
Some further representative results from run bbhsu9d axis the axial force (note that the sign convention for
are presented and discussed below. forces in structural elements in FLAC3D depends
269
Figure 9. Axial forces in some of the piles on the footing side of the excavation, at the end of construction, as computed in
model bbhsu9d.
Figure 10. Moments (my – upper and mz – lower) in the same set of piles as in Figure 9.
270
upon the relative orientation of the nodes). In the existing bridge were not disturbed. Analysis of the
grouted (lower) section, the developed forces are bridge indicated that lateral displacements of the foot-
close to the pre-tension force of 1 MN but they are ing must be constrained not to exceed 5 mm. Program
about 8% smaller in the ungrouted (upper) section. FLAC3D was used to simulate the complex construc-
This is consistent with the response expected as the tion sequence, including the placement of piles and
system is re-equilibrated after pre-tensioning of the ground anchors. FISH programming was used exten-
anchors. Examples of forces and moments generated sively to assist in generating the grid and in managing
in piles adjacent to the bridge footing are given in the links between the various structural elements.
Figures 9 and 10. In each case the vertical axis gives The results from a series of production analyses indi-
the RL (z-coordinate) of pile elements and the hori- cated that a design based on 8-strand ground anchors,
zontal axis the force or moment, in SI units. The axial pre-tensioned to 1 MN, would be satisfactory. The
forces in Figure 9 clearly show the effects of the con- predictions of FLAC3D have been confirmed by mon-
nection of ground anchors at two points in the upper itoring during construction.
sections of some of the piles.
ACKNOWLEDGEMENTS
8 FIELD PERFORMANCE
The authors wish to acknowledge many valuable
During the entire excavation process the measured
communications with Dr. D. Potyondy, Itasca, partic-
lateral deflection of the abutment of Hawthorne
ularly about the linking of structural elements in
Street Bridge was less than 5 mm. There was no dis-
FLAC3D.
tress of Hawthorne Street Bridge during and after
construction of the Hawthorne Street Tunnel.
REFERENCES
9 SUMMARY
Itasca Consulting Group, Inc. 1997. FLAC3D – Fast
Design and construction of the section of the Brisbane Lagrangian Analysis of Continua in 3 Dimensions,
Busway had to ensure that the adjacent footings of an Version 2.0. Minneapolis, MN: Itasca.
271
A. Klar
Technion – Israel Institute of Technology, Haifa, Israel
I. Einav
Centre for Offshore Foundation Systems, UWA, Australia
ABSTRACT: This paper presents a numerical simulation of pile installation using FLAC. A new contact for-
mulation between rigid and deformable bodies is employed. This formulation utilizes equations of motion to
describe the behavior of the deformable nodal point along the contact surface. Unlike FLAC’s own embedded
interface formulation, the new formulation does not encounter discontinuities problem along nonlinear or
piecewise linear surfaces.
273
are the ones that may be represented by the cavity 2.5 Large strain Lagrangian finite element
expansion solutions. The behavior in Zone I is highly analysis
affected by the soil surface. One of the purposes of
This method is the one employed in FLAC, in which
the work presented here is to evaluate the surface
the mesh is updated throughout the pile penetration
effect on the solution, and to discover the required
process. This method is the only one that in theory can
depth for which the evolution of the end bearing
capture the installation process as it is, while including
capacity factor Nc is redundant. To learn more about
the influence of both soil surface and changing prop-
the use of cavity expansions in the solution of pile and
erties with depth. It should be noted, however, that this
cone penetration, one may refer to the excellent book
method is associated with numerous numerical prob-
by Yu (2000).
lems, when simulating pile or cone penetrations, due
to the great deformations involved in the problem.
2.3 Strain path method
The strain path method was first suggested by Baligh 3 PENETRATION DIFFCULTIES IN FLAC
(1985). In this method, a flow field of soil is assumed
to exist around the pile. From this assumed flow field, At this current stage, if one chooses to utilize FLAC’s
strains are derived, while stresses can then be deter- embedded interface formulation to simulate the soil-
mined according to a particular constitutive relation. pile contact, he should acknowledge that on top of the
Two main drawbacks are associated with this method: problems due to the great deformations, he introduces
(a) equilibrium will not necessarily exist, and (b) the new problem. As in many other codes, in FLAC’s inter-
effect of the surface and/or changing properties with face formulation the two bodies are prevented from
depth cannot be included, i.e. it can represent only crossing each other. This leads to discontinuities in the
deep steady penetrations. The first limitation may be contact between the bodies (Itasca 2000), if nonlinear or
partly overcome by the use of the iterative procedures piecewise linear surfaces are involved. Figure 2 shows
suggested by Teh & Houlsby (1988). an example of the problem for piecewise linear rigid
contact surface. In Figure 2a the contact formulation
corresponds to that of FLAC; i.e. it does not allow for
2.4 Large strain Eulerian finite element analysis the deformable body (represented by the quadrilateral
In the Eulerian large strain finite element analysis, the elements) to overlap the rigid body (represented by the
material (soil) streams through fixed points in space thick black line), and therefore gaps between these two
[e.g. van den Berg 1994]. This method possesses the are developed near discontinuity points along of the
same limitation of the last methods; i.e. inability to rigid body. There are two kinds of gaps that may
model the soil free surface. develop between the rigid body and the deformable one.
The first kind, (noted as type I in Fig. 2), is a gap that
will always result when the deformable body is in con-
tact with a concave surface of a rigid body, and is fic-
tional because the lamped grid points are still in contact.
The second kind (noted as type II in Fig. 2) is true gap
Zone I
Figure 1. Cavity expansion zones along the pile. Figure 2. Contact problems along piecewise linear surface.
274
This behavior, shown in Fig. 2(a) for piecewise lin- defined in the fixed coordinate system. If considering
ear surface, will be more pronounced in nonlinear sur- the moving coordinate system then the deformable
faces, since, at least, every second grid point will be body motion is defined by a velocity vector of
out of contact with the rigid body in a convex surfaces. vL vD vR and acceleration vector of aL aD aR.
To overcome this problem of undesirable gaps, the Note the rigid body is stationary in the moving coordi-
contact formulation must allow the deformable body nate system. Figure 3 shows velocity and acceleration
to overlap the rigid body, as shown in Figure 2b. In this diagrams of a grid point located on the rigid body. This
“overlapping” formulation, the grid points, from body can be represented by a shape function x f (z)
which the strain increments are derived, travel along (x and z are the coordinates of the moving system).
the rigid body surface, unless some sort of contact Since the deformable body cannot enter the rigid
logic that allows separation is included. In the follow- body nor departure from it (unless tensile failure is con-
ing section, such a contact formulation is presented, sidered as will be discussed later) the motion of it can
and is used later for the analysis of pile installation. only be tangential to the rigid body; i.e. only the tangen-
tial components of both velocity vector and acceleration
need to be introduced in to the equations of motion. If
4 NEW CONTACT FORMULATION we consider an explicit time marching numerical
FOR FLAC scheme, the following expression can be written:
4.1 Formulation
A body may be defined as rigid if the distance between
any two points of it is constant with time. The motion
of a non-rotating rigid body can be described by two where v Lx, vLzand aLx, aLzare the components of veloc-
components, a velocity vector, vR and an acceleration ity and acceleration vectors in the moving coordinate
vector aR. The motion of each lumped mass located on system (x, z) at time t. vxD, vDz , aD x
, aDz , vxR, vRz , axR, azR
a grid point that represents the deformable body can are the components of the motion vectors of the
also be described by two vectors, vD and aD for veloc- deformable and rigid body in the fixed coordinate
ity and acceleration respectively. vD and aD are here system (x, z) at time t. If the motion of the rigid body
is prescribed (i.e. know a priori) then the motion of
the deformable body in the fixed coordinate system is
as follows:
Velocity Diagram Acceleration Diagram
(2)
νL
R
a
aL
aD that state variable are frozen at each step (dt); i.e. for
νD
each time step the rigid and deformable body are fixed
in space, and therefore, all values in the right side of
Z' Equation 1 are known. The value of aD in Equation 1 is
obtained from the assumption that the deformable body
X' is not in contact with the rigid body; i.e. the accelera-
tion is obtained from the forces acting on the grid point
Figure 3. Velocity and acceleration diagrams. due to the deformation of the deformable body. To
275
introduce some friction between the rigid and the 5.2 Assumptions
deformable body one can add frictional force in the
The numerical analysis was conducted under the fol-
tangential direction and re-obtain from it the value of
lowing assumptions:
aD. If a rigid-plastic tangential interface is desired then
one can define the friction forces direction simply 1. The material behaves elastic perfectly plastic and
according to the relative velocity between the soil and satisfies the von-Mises failure criteria. Since the
the pile. If an elastic plastic tangential interface is analysis is associated with undrained loading, and
desired a slightly more complicated formulation is the volumetric stresses are decoupled from the
required. Since in the current paper only smooth piles deviatoric ones in the considered constitutive
are considered, this kind of formulation is not pre- model, it is possible to perform a “Dry” simulation;
sented, although written and verified by the writers. i.e. to obtain the excess pore pressure value using
To consider possible separation between the rigid Skempton’s parameter, B 1 1/(1 (Kw/n)/ Ks),
body and the deformable body, a contact logic must utilizing the formula u B(ii ii0 )/3, where
be introduced. If, for example, the contact logic con- the superscript 0 denotes initial state.
siders zero tensile forces between the rigid and the 2. The undrained strength of the soil is defined
deformable bodies as condition for separation, then it according to the relation Cu 0.25v0 OCR0.95,
will occur once 180 a a. Whenever this v0 is the initial vertical effective stress and OCR
condition is satisfied, the grid point is solved accord- is the over consolidation ratio.
ing to aD; i.e. vD(t dt) vD(t) aD(t)dt. If during 3. The shear modulus is taken proportional to the
one of the following steps the grid point comes undrained strength Cu, and the bulk modulus was
in contact with the rigid body, Equations 1 & 2 are high enough so the material can be considered as
applied. Some small changes need to be introduced incompressible.
into Equation 1 if it is desired to apply one of FLAC’s 4. The analysis presented herein considered a con-
damping schemes which operates on grid points mass. stant OCR with value that equals 2.
If damping results only from the constitutive model, 5. Initial stress condition corresponded to K0 of 0.7.
then Equation 1 is satisfactory. One may refer to 6. To avoid the kinematic constrains of a fixed bound-
Einav & Klar (2003) where the above formulation is ary, and to allow approximate simulation of an infi-
extended to a more general case of three-dimensional nite soil layer, a prescribed boundary condition was
rotating rigid-deformable bodies in space. applied to the outer radius of the grid. The boundary
The described procedure is easily implemented in condition corresponded to the cylindrical cavity
FLAC using a FISH function, which is called during expansion solution of an incompressible elastic
each of the calculation cycles. Generally, since the material; i.e. the external pressure acting on the mesh
motion of the contact grid points is solved indepen- was defined by the analytical solution of the internal
dently (i.e. using Equations 1 & 2 rather than by FLAC),
they need to be fixed in both directions. Quantities
related to aD are extracted from FLAC’s gridpoint vari-
able xforce and yforce. Quantities related to velocities,
both readable and writeable, are manipulated using
FLAC’s gridpoint variable xvel and yvel.
5 NUMERICAL ANALYSIS
(3)
276
pressure of cavity expansion with an identical radius As discussed in section 2, only Lagrangian large
to that of the outer boundary. It was verified that the strain analyses can simulate the penetration of the pile
plastic zone did not reach the outer boundary, thus through the surface, and therefore are the only ones
the elastic cavity expansion solution was suited. that can evaluate the influence of the surface on the
cone penetration. Figure 9 shows the cone resistance
factor, Nc, for different rigidity indexes (G/Cu). The
6 RESULTS Nc factor was calculated according to:
Figure 5. Distorted mesh in steady penetration Figure 7. Development of excess pore pressure
(G/Cu 100). (G/Cu 100).
277
15
G/Cu=1118 7 SUMMARY AND CONCLUSIONS
14
13 G/Cu=500
D Zc
12 G/Cu=223 A contact formulation for interaction between rigid and
11 deformable bodies is presented. This approach over-
10 G/Cu=100 comes some of the problems associated with FLAC’s
Nc
278
F. Ma
Washington State Department of Ecology Solid Waste Program, Eastern Regional Office, Washington, USA
ABSTRACT: A finite difference analysis using the computer code FLAC was conducted of a municipal land-
fill expansion proposal at the Cedar Hills Regional Landfill (CHRL), King County, Washington State. The main
objective of the modeling efforts was to assess whether a standard design of a liner system would be adequate
to withstand typical loading conditions for municipal solid waste landfills. The loading conditions were: (1) the
gradual layered waste dumping up to 38 m (125) over the High Density Polyethylene (HDPE) liner; (2) dynamic
loadings caused by a shallow earthquake and a deep subduction zone earthquake; and (3) the simulation of a
cavity development in the old existing waste underneath the HDPE liner due to collapsing of some bulky items.
The FLAC (2D) analyses have revealed (1) the developments of the axial tensile stress and displacement in the
HDPE liner; (2) the stress and deformation developments in the municipal solid wastes; and, (3) the accumulative
and separate developments of stress and displacement of the landfill system under waste dumping, earthquakes
and cavity collapsing. The main conclusion was that the maximum axial tension in the 60 mil HDPE liner is higher
than the yielding strength of a GSE 60 mil HDPE liner (HDR/Golder 2001) under the proposed site, operational
and loading conditions. Thus some local reinforcement or stronger geomembrane liners will be needed.
2 SITE CONDITIONS
Figure 1. Cross section of the landfill including location of
Landfilling commenced at the CHRL site in the existing wastes, liner, new wastes to be disposed and foun-
early 60s. Those portions of the landfill started before dation soil.
279
280
Figure 2. Distributions of axial tension in the geomem- Figure 4. Distributions of axial tension in the geomem-
brane liner. brane liner due to static and dynamic loadings.
281
282
283
284
285
F. Ma
Washington State Department of Ecology Solid Waste Program, Eastern Regional Office, Washington, USA
ABSTRACT: The objective of the FLAC modeling effort was to analyze the stress–strain behavior of High
Density Polyethylene (HDPE) leachate collection pipes proposed for an eastern Washington State solid waste
landfill. Case one of the FLAC modeling assessed whether a 102 mm (4) diameter Standard Dimension Ratio
(SDR) 9 perforated HDPE leachate collection pipe could withstand the load of up to 64 m (210) high column
of solid waste. Similarly case two of the FLAC modeling predicted how a 305 mm (12) diameter perforated
SDR 11 HDPE pipe would perform under a solid waste load of as much as about 26 m (85). The FLAC analy-
ses allowed simulating the development of stresses and deformations in the HDPE leachate pipes as the solid
waste column grows. The model predictions were compared with results from a methodology included in the
Guidelines for HDPE Pipes in Deep Fills (Petroff 1998) used by CH2MHILL (2002) in their design. Using the
industry standards of (1) ring compressive stress, (2) pipe deflection and (3) wall buckling, the FLAC results
were very similar to the values in the CH2MHILL (2002) study. The main conclusion of the FLAC modeling is
that the proposed HDPE leachate collection pipes will be adequate to withstand the loadings associated with the
proposed solid waste column heights.
287
3.2 This study’s current modeling (1994) by Petroff (1998), the typical design values
of one-dimensional constrained modulus Ms of soil
The empirical equations provided in the Guidelines
increase linearly with the increase of the soil over-
for HDPE Pipes in Deep Fills (Petroff 1998) predict
burden pressure. This linear relationship (Petroff
the maximum ring compressive stress and the maxi-
1998) was used in the FLAC modeling and
mum pipe deflection. The FLAC modeling described
was related to bulk modulus K and shear modulus
herein went on to predict the distribution of the axial
Gs by (K Ms (1 )/(3 (1 )) and
stress, shear stress and bending momentum in the
Gs Ms(1 2)/ (2(1 ), respectively. The
pipes as well as the stress–strain or deformation rela-
elastic modulus Es is related to the constrained
tionship under plane strain conditions.
modulus Ms of the soil by Es Ms(1 )
The following simplifications or assumptions were
(1 2)/(1 )). The material properties of the
made in the FLAC modeling.
gravel layer are listed in Table 2.
i. The HDPE pipe is modeled with isotropic-elastic iii. The waste behavior is simulated by the elasto-
beam segments; its material properties per unit plastic Modified Cam-Clay model in FLAC. The
length are area A t, moment of inertia I t3/12, behavior of the HDPE pipes is the focus of the
elastic modulus under plane strain e ey / modeling effort here. The stress-strain response of
(1 2), where t is the pipe wall thickness, ey is the waste was of little interest. Thus, the waste
the Young’s modulus and the Poisson ratio. The properties approximated with values typical of
material properties of the 102 mm (4) SDR 9 and soft clay at a density of 1441 kg/m3 (90 lb/ft3) were
305 mm (12) SDR 11 pipes are listed in Table 1. used as the input parameters of the wastes. The
The effect of perforations in the HDPE pipes was model parameters are elastic shear modulus
modeled by reducing the wall thickness by one G 5.12 104 kPa (7430.6 psi), maximum
twelfth as typical perforations account for that elastic bulk modulus Kmax 1.53 105 kPa
much of the pipe mass. (22222.2 psi), density a variable with depth
ii. The gravel drainage layer is simulated as a perfect (Kavazanjian et al. 1995), slope of elastic swelling
plastic Mohr-Coulomb material. The material liner 0.03, slope of normal consolidation line
properties include bulk modulus K, shear modu- 0.13, frictional constant M 1.418, precon-
lus Gs, friction angle and density. The average val- solidation pressure pc 71.8 kPa (10.4 psi), refer-
ues of density and friction angle of gravel are ence pressure p1 71.8 kPa (10.4 psi), and
2.16 g/cm3 (135 lb/ft3) and 40 degree, respec- specific volume at reference pressure on normal
tively. As adapted and extended from McGrath consolidation lien 1.75.
d t A I e
Setting kg/m3 mm mm2 mm4 kPa
* Note: the values are obtained or deduced and deducted from Driscoppipe data sheets
and are values per 25.4 mm (1 inch) of the pipes.
Overburden Ms K Gs
kPa (psi) kPa (psi) kPa (psi) kPa (Psi)
*Note: the values are obtained or deduced and deducted from Petroff (1998) with
Poisson ratio 0.3.
288
iv. The HDPE pipe is connected to the surrounding of the pipe response. When executing the models, the
gravel layer using the FLAC interface. The inter- waste was added layer by layer to simulate a landfill
face allows the relative slip between the HDPE operation. As the waste pile grew, increasing the verti-
pipe and the gravel. The interface input parame- cal overburden pressure, the modulus of the gravel layer
ters per unit length are normal stiffness increased accordingly (Petroff 1998). Since the rela-
kn 411.6 N/mm (2341 lb/in), shear stiffness tively small sizes of the HDPE pipes, no further refine-
ks 107.8 N/mm (619 lb/in), and friction angle ments of the mesh around the pipe openings were done.
30°. Due to limited data availability on the
interface properties, more studies will be done in
4 MODELING RESULTS
some future researches.
The details of the model grids for the two scenarios 4.1 305 mm (12 ) SDR 11 HDPE LCRS header
are shown in Figures 1 & 2. The horizontal dimension pipe under solid waste load up to 26 m (85
)
of the grid was chosen such that a further increase in
The axial compression, shear and moment distribu-
width had no material impact on the modeling results
tion for the 305 mm (12) SDR 11 HDPE LCRS
sump/pump pipe under solid waste of 26 m (85) are
shown in Figure 3. Since the results for pipes with or
without perforations are very similar, only the latter
are shown here. Numerical values of the data shown
graphically in Figure 3 are presented in Table 3.
289
*Note: Elem ID 1 corresponds to the pipe segment at pipe crown with nods 1 and 2. Elem ID
24 to the pipe segment at pipe invert. F-shear and F-axial are the shear and axial forces of the
each pipe segment, respectively. Mom-1 and Mom-2 are the moments of both ends of each
pipe segment.
Figure 3a shows that the maximum hoop compres- and Figures 4a&c. As in the case of the 305 mm (12)
sive force occurs near the pipe springline and is SDR 11 pipe, the maximum shear force is predicted
approximately 50% larger than that at the pipe crown to occur at approximately 45 degrees below the pipe
and invert. The maximum shear force occurs at crown, see Figure 4b.
approximately 45 degrees below the pipe crown as in
Figure 3b. Figure 3c shows that the moment near the
pipe crown and springline are approximately 25% 4.3 Pipe crown deflections
larger than that at pipe invert. The predicted crown deflection of a buried pipe is one
of the key parameters in assessing the structural ade-
4.2 102 mm (4 ) SDR 9 HDPE LCRS header pipe quacy of a pipe in traditional pipe analyses. The pre-
under solid waste load up to 64 m (210
) dicted displacement histories of the pipe crown and
invert are shown in Figures 5 & 6.
The axial compression, shear and moment distributions
for the 102 mm (4) SDR 9 HDPE LCRS header pipe
predicted for 64 m (210) of solid waste are shown in 4.4 Comparing FLAC predictions to those of
Figure 4. As in the earlier case, the results for pipes with traditional empirical formulas (Petroff 1998)
or without perforations are very similar. Thus, only the
latter are shown here. The numerical results underlying According to general thin beam theory, the normal
the graphical data in Figure 4 are cited in Table 4. stress in the pipe wall is the combination of normal
Similar trends are evident in the loadings predicted stresses from the hoop force and the bending
in Figure 4 to those predicted for the 12 SDR 11 pipe moments. It is expressed as follows:
case. However, the differences between the values of
hoop compression and moments at the pipe crown, (1)
springline and invert are much smaller, see Table 4
290
291
292
293
G. Swan
Falconbridge Ltd, Sudbury Operations, Sudbury, Ontario, Canada
G. Doyle
3M Canada Co., Mining Division, London, Ontario, Canada
ABSTRACT: The current practice of bolting and screening of underground excavations is time consuming and
labor intensive and requires extensive materials handling. In pursuit of alternative rock support systems,
Falconbridge Ltd. has experimented with several different spray-on liners. In co-operation with 3M Canada, a thin
spray-on liner was developed with the intention of replacing screen and reducing the cycle time in rapid develop-
ment mining. FLAC was used on this project to provide an efficient method of investigating alternative material
properties without the expense that is typically incurred in full scale testing.
The implementation of high speed development tech- The spray on liner support system developed by 3M
niques necessitates the use of rapidly installed support Canada is a polymeric compound that contains approx-
in order to meet the desired cycle time. Current prac- imately 40% water when initially applied. As the liner
tices of bolting and screening are labor intensive and dries out (the rate is dependant on the ambient temper-
add significant time to the development cycle. The use ature, humidity, and air flow) the strength builds until
of spray-on-liners to act as membrane support in place after 24 to 72 hours it approaches its ultimate tensile
of screen is seen as the next step in the evolution of and adhesive strength. The time dependence of the
rapid development. strength of the liner is a critical aspect and one that is
The physical characteristics of membrane support very difficult to define in full scale testing apparatus
need to be extensively tested. Full scale physical testing since there are so many variables involved.
is expensive and time consuming and the number of Shotcrete is often sprayed in excess of 4, this liner
such tests needs to be limited. In order to fill the gap system only requires 2 to 3 mm final dried thickness to
between laboratory measurements of liner properties perform. When dealing with such a thin application,
and full scale trials of the material, a numerical model minor thickness variations become important.
of the testing apparatus was constructed. This provides Another important characteristic of this liner com-
a method by which a large number of potential liner pound is the elongation potential. Depending on the
formulations can be investigated without the expense of formulation in use, strains ranging from 100 to 600%
full scale testing. are possible. Having such a large capacity to deform
Using the cable element capability of FLAC, a model should help prevent violent failures due to stress
of the “baggage loading” testing apparatus was gener- build up. A stress–strain curve for some early liner for-
ated to allow for a large number of simulations to be mulations is shown in Figure 1.
carried out to cover the wide range of properties One final challenge in understanding this material is
observed in the liner material. that it does not yield in a linear fashion. Instead, there
295
System I
3 FLAC MODEL
296
7
6 3M System 1 - FLAC
3M System 2 - FLAC
5 3M System 1 Averaged
3M System 2 Averaged
4
3
2
1
0
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70
Strain
Figure 5. A stress–strain plot created with the FLAC FISH function compared to the actual strengths reported in material
tensile testing.
Stress (MPa)
Strain (mm/mm)
1000
DRYING TIME, hours
1
0.01 0.1 1 10 100 1000
STRENGTH/STIFFNESS, MPa; ELONGATION,%
297
4 RESULTS With this second test, the liner was stable under
gravity loading from the dead weight of the rock in the
Several model simulations were performed in order to testing frame. The total displacement under this static
calibrate the model against the observed and measured loading was 24 mm. This test was also the source of the
response of the liner. stress–strain data generated in Figure 6. This test pro-
One of the first full scale tests carried out was an vided valuable data for the calibration but also demon-
application of the liner after 4 hours with no adhesion to strated that there was another effect at work that was not
the rock (a silicone material prevented liner adhesion). being accounted for in the modeling. The model pre-
Immediately it became apparent that the adhesion of dicted a total displacement for this particular test at
the material was playing a much bigger role in the sta- 90 mm rather than the 24 mm observed in Figure 10.
bility of the baggage loading test than would be anti- In order to make up for the difference in displace-
cipated. Figures 8 & 9 show the actual results and the ments observed between the baggage loading tests and
FLAC modeling results respectively for a liner that has the FLAC models, an investigation took place in which
no adhesion to the rock. In both cases the liner quickly the stiffness of the material was increased until match-
deforms due to the low strength observed at this early ing results were observed.
time after spraying. Modifying the model properties resulted in a curi-
The second test in the series involved the baggage ous result. In order to match the performance of the
loading test with the material left to dry over a 24 hour baggage loading test, the stiffness of the material had
period. In this case, adhesion was allowed to take to be increased to near 20 times its original value.
place by lightly wetting the surface of the rock prior When these modified properties were used, the results
to application. matched quite well with the observed testing (Fig. 12).
JOB TITLE: Falconbridge-3M Baggage Load Testing Model: 3M System 2 JOB TITLE: Falconbridge-3M beggage Load Testing Model: 3M System 2
FLAC (Version 4.00)
FLAC (Version 4.00)
0.900
LEGEND 0.500 LEGEND
21-Mar-02 12:08
27-Mar-02 14:56 0.700
step 21000
step 19190 -2.222E-02 <x< 1.422E+00
-3 860E-01 <x< 1.786E+00 -4.222E-01 <y< 1.022E+00
0.200
-1.300E+00 <y< 8.715E-01
Grid plot Grid plot 0.500
0 0 2E-1
5E-1
Axial Force on
-0.200 Structure Max. Value 0.300
Axial Force on
Structure Max. value # 1 (Cable) -4.934E+03
# 1 (Cable) -1.100E+02
0.100
-0.500
-0.300
-1.000
-0.100
Figure 9. FLAC modeling results for the baggage loading Figure 11. FLAC results for the modeling of the baggage
tests with no adhesion. loading test with adhesion and modified material properties.
298
0.100
5 CONCLUSIONS
-0.100
The FLAC model has provided a valuable tool for
-0.300 reducing the cost of full scale testing. With this model,
Itasca Consulting Group, Inc.
Minneapolis, Minnesota USA
it is possible to anticipate the response of different
0.100 0.300 0.500 0.700 0.900 1.100 1.300
formulations of the liner at different thicknesses and
time frames. Given the high cost of full scale tests, it
Figure 13. FLAC output for baggage load testing with can be used to narrow down the testing regime to the
adhesion. Anticipated displacement was 90 mm instead of
the actual 24 mm measured.
most promising combinations of thickness, time, and
liner properties in order to maximize the data collected
during full scale tests. Further calibration against full-
Based upon the results observed in the full scale scale tests will help to reduce some of the uncertainty
testing, it was obvious that some mechanism must be involved with the model.
at work in order to account for the discrepancy in the
modeling. There are three possible sources of uncer-
tainty that would tend to artificially increase the stiff- ACKNOWLEDGMENTS
ness of the material, the adhesion, and the impact of
blocks interlocking with each other and the frame. The authors would like to thank all of those involved
Firstly, the FLAC model is a 2D model, while the in the preparation of the baggage load tests including
actual baggage load test is three-dimensional. 3M Canada, Falconbridge Ltd, and CANMET.
However, the actual discrepancy would appear to be
too large to be accounted for simply by this difference.
299
ABSTRACT: Piles are often placed in the base of deep excavations to carry future structure loads or to reduce
base heave. Where these piles are located close to retaining walls, they can provide additional resistance to the
movement of the embedded length of the retaining wall. This paper discusses a series two-dimensional and three-
dimensional analyses carried out using the finite difference programs FLAC and FLAC3D to investigate the increase
in the passive resistance in front of the embedded retaining wall due to the presence of these piles. Two passive fail-
ure mechanisms were identified; the first involved squeezing of the ground upward between the wall and the piles
and the second involved squeezing of the ground between the piles. The influence of pile and wall roughness, pile
spacing and pile to wall separation was investigated in order to define which of the two passive failure mechanisms
would govern and under what circumstances. Based on the results of the two-dimensional analyses a methodology
was developed to determine the limiting passive resistance allowing for the presence of the piles. Three-dimensional
analyses were carried out which showed a close agreement with the results of the two-dimensional analyses.
301
(1)
the undrained material. The limiting passive earth pres-
sure coefficients are also given in Table 2. When the
wall and piles were placed 20 m apart, the computed
(2) values of Kpc are very close to the theoretical values
given in CP2 and reproduced in Table 2. BS8002 sug-
gests that the passive resistance in a cohesive soil can
where Kpc and Kp are the passive earth pressure coef- be approximated by the following equation:
ficients associated with undrained and drained soil
materials respectively.
(3)
2.1.1 Undrained material
Figures 2 & 3 show the limiting horizontal passive
resistance and the limiting passive earth pressure coef- The passive earth pressure coefficients predicted
ficients respectively for a 10 m deep wall embedded in by this equation are also given in Table 2. The passive
302
-10
303
Kp
3
10 2.95 4.50 5.53 6.45 7.09
5 2.98 6.99 11.2 17.6 36.2
3 3.00 13.3 35.5 95.5 354 2
2 3.01 34.3 141 639 –
1 3.02 731 – – –
0.5 3.01 – – – – 1
CP2 3.0 4.0 – 4.9 5.8 Associated
Non-associated
0
0 20 40 60 80 100
50
phiw = 0 Wall friction (phi %)
phiw = phi/3
phiw = phi/2 Figure 7. Comparison of passive earth pressure coefficient
40 for associated and non-associated materials.
phiw = 2phi/3
phiw = phi
304
16
14
12
10
Ncp
8
6
D
4
smooth, 0 kPa
rough, 0 kPa
2 smooth, 200 kPa
S/2 rough, 200 kPa
0
1 2 3 4 5 6 7 8 9 10
S/D
305
smooth and rough pile interface analyses the limiting the piles is resisted by a passive wedge behind the piles.
Ncp values are 9.3 and 12.0 respectively. These are com- It is suggested that, except at very close pile spacings,
parable to the results given by Chen & Martin (2002). the wedge mobilized behind the piles is identical to that
Figure 10 shows the effect of varying the in situ which would have been mobilized if the piles had not
stress state for a constant pile spacing of 3.0 D. For the been present. The restraint provided by the piles is
smooth pile interface the Ncp value increases from therefore generally additive to the normal passive
4.63 to 9.1 as the in situ stress is increased from 0 kPa wedge. The piles provide resistance only in that portion
to 70 kPa and is then constant for higher in situ stress of the passive wedge through which it passes. For a
states. For the rough pile interface the response is sim- smooth wall the failure surface underlying the passive
ilar with the Ncp value increasing from 5.67 to 11.1 at wedge rises at 45° from the toe of the wall. For a wall
about 70 kPa with the value constant at higher in situ with friction the failure surface rises at a shallower
stresses. The reason for the change from a gradually angle (see Fig. 4). Conservatively it can be assumed
increasing Ncp values at low stress to a constant value that the pile penetrates through the passive wedge to a
at higher stresses can be resolved by inspection of the depth D–h. The total passive resistance, Ptotal, per meter
deformation pattern around the piles. At stresses less run provided by this failure mode can therefore be
than 70 kPa the ground is not in contact with the back expressed as follows:
of the pile whereas above this stress the ground is in
touch with the pile over the whole pile circumference.
(7)
As the stress is gradually increased from 0 kPa to
70 kPa the length over which the ground is not in con-
tact with the pile gradually reduces. To decide whether failure will occur by squeezing
These analyses would appear to suggest that the pas- of ground between the piles or by failure of ground in
sive resistance provided by the pile would vary with front of the piles the mode with the lower failure load
depth down the pile. Near the surface the restraint pro- must govern.
vided by the pile would be least and the ground move- As an example the force required to develop the two
ment would lead to a gap developing on the side of the failure mechanisms has been calculated for the soil con-
pile furthest from the wall. ditions described above with a 10 m deep wall with
1.5 m diameter piles at 4.5 m centers located 2 m in
2.3 Combined effect front of the wall. The force for the combined failure
mode is 1825 kN/m whereas for the failure mode in
The two dimensional horizontal section analyses have front of the piles, the force is 2103 kN/m. In this case
shown that under high stress ground can be forced failure by squeezing of the ground between the piles is
between the piles. However, the ground forced between more likely than failure by squeezing in front of the
piles. It is illustrative to note that passive failure would
12
have occurred at a force of 1297 kN/m if the piles had
not been present. This illustrates the significant increase
in the passive resistance caused by installing piles in
10 this location.
8 3 THREE-DIMENSIONAL ANALYSES
6
standing of the mechanisms involved with piles in the
passive zone of embedded retaining walls, but the
4 actual behavior is almost certainly more complex than
the two dimensional analyses can show. It is conceiv-
able that failure of a deeply embedded wall would occur
2 by a combination of both mechanisms. It is also diffi-
Smooth pile
Rough pile cult to judge the apparent horizontal stress acting in the
0 horizontal plane when assessing the resistance provided
0 50 100 150 200 by the piles.
In situ stress (kPa) To provide more guidance on the equivalence of dis-
crete passive-zone piles compared to an equivalent
Figure 10. Comparison of passive resistance coefficient continuous wall a pair of three-dimensional analyses
developed on smooth and rough piles for different in situ have been carried out using FLAC3D. Both analyses
stress states. considered clay with the properties given in Table 1.
306
The clay layer was taken to be 30 m thick overlying a The wall and pile properties are given in Table 4.
hard stratum. To reduce the size of the model only that The equivalent wall properties were derived using the
part of the construction below final excavation level following formulae, which ensured that the equiva-
was considered. The overall width of the model was lent wall had the same bending and axial stiffness as
15 m which was deemed to be sufficiently wide to pre- the discrete piles.
vent interaction of the boundary with the wall or the
pile. As in the two-dimensional analysis of the horizon-
tal slice, symmetry was adopted on vertical planes per-
pendicular to the wall through the center of the piles
and also through a point midway between the piles. (8)
The wall was embedded 10 m into the clay and was
modeled using liner elements which comprise triangu-
lar plate elements connected to the ground through an Where Ep and Es are the Young’s moduli of the pile and
interface with normal and shear elastic and plastic the equivalent wall respectively, and t is the equivalent
properties. Many propped embedded retaining walls smeared pile wall thickness.
undergo maximum horizontal displacement at around The model was initially brought to equilibrium under
final excavation (formation) level and therefore the the in situ stress conditions and by fixing the horizontal
wall was moved towards the soil by applying a hori- movement of the embedded wall. The reactions devel-
zontal force to the wall at formation level while also oped on the embedded wall during this stage were then
preventing rotation about a horizontal axis at this applied as a series of nodal forces acting on the wall for
point. In one of the analyses, 1.5 m diameter piles the remainder of the analysis. The analysis was contin-
spaced at 4.5 m centers were modeled with the pile axes ued by increasing the magnitude of the horizontal force
located 2.75 m in front of the wall, resulting in 2.0 m at the top of the wall in increments. After each increase
of soil between the wall and the nearest edge of the in the force the model was allowed to reach equilibrium.
piles. A close-up of this model is shown in Figure 11. Figure 12 shows the results of the two analyses. The
In the second analysis the pile was substituted by a solid symbols show the displacement of the top of the
continuous wall with equivalent smeared properties to wall versus applied force for the two analyses. In both
that of the discrete piles. The centerline of the equiva- cases the displacements are initially small as the force is
lent wall was also located 2.75 m from the embedded increased. Up to an applied force of 1500 kN/m the wall
wall. Both the discrete piles and the equivalent wall movement is very similar in the two analyses. However
were modeled using solid brick elements rather than as the force is increased above 1500 kN/m, the wall
structural elements. movement in the analysis with discrete piles increases
Both the discrete piles and the equivalent wall rapidly and appears to become unlimited at an applied
extended the full depth of the model and both were force of about 1850 kN/m. In the analysis with the piles
assumed to be rigidly fixed in a hard stratum at the represented by an equivalent wall, the wall movement
base of the model. Rough interface properties were does not increase as rapidly and only becomes unlim-
considered between the embedded wall, the pile, the ited as the applied force approaches 2000 kN/m. These
equivalent wall and the ground. limiting values are very similar to the theoretical values
calculated in section 2.3 above.
Also shown in Figure 12 as open symbols is the pile
head or equivalent wall top movement versus applied
force. The pile movement is very similar to the
embedded wall movement up to an applied force of
1500 kN/m. In the analysis with discrete piles the pile
starts to move less than the wall as the applied force is
increased above 1500 kN/m. In the analysis with an
equivalent wall, the equivalent wall only starts to move
less than the embedded wall when the applied load
Equivalent
Setting Wall Pile wall
307
1200 0
Discrete Piles
Equivalent Wall
Structural Elements
800 -10
600 -15
400 -20
308
elements have no physical thickness in the model piles the critical pile spacing appears to vary from 2
although their axial and bending stiffness is mod- pile diameters when the piles are located 2 m in front of
eled correctly by assigning appropriate values of area the wall to 4 diameters when the piles are located 5 m
and moment of inertia. The structural elements are in front of the wall.
generally placed along the centerline of the piles or
wall that they are intended to represent. To investigate
the effect that modeling the piles as structural elements 5 CONCLUSIONS
has on the predicted passive resistance the three-
dimensional model was rerun with the piles repre- The analyses have shown that piles installed in the
sented by shell elements with properties identical to passive zone of embedded retaining walls can signifi-
those of the equivalent wall given in Table 4. The shell cantly increase the passive resistance mobilized in
elements were rigidly connected to the mesh and there- front of the retaining walls. The passive resistance is
fore represented a rough wall. To be comparable to the sensitive to the distance of the piles to the wall and
other two analyses, the toe of the equivalent wall was whether the piles and wall are rough or smooth. For
fixed against horizontal movements and was prevented granular deposits the passive earth pressure coeffi-
from rotating around a horizontal axis. Because cient increases dramatically as the spacing between
the structural elements have no physical thickness in the walls and the piles reduces and it is suggested that
the model, 2.75 m of clay is now present between the passive failure is unlikely to occur in this material
embedded wall and the equivalent wall. unless the piles are widely spaced or the wall has only
The results of the analysis are shown in Figure 12 as a shallow embedment.
crosses. The greater thickness of soil between the Three-dimensional analyses showed a very similar
embedded wall and the equivalent wall gives a softer limiting passive resistance to a calculation based on
wall displacement response compared to the analysis two-dimensional analyses. However, the deflections to
with the equivalent wall modeled using solid elements. mobilize this passive resistance are large and may imply
The limiting passive pressure is also lower because of unacceptable movement of the retaining structure. The
the greater separation between the two walls. From piles undergo significant lateral deflection, localized
Table 2 it can be seen that increasing the wall separa- bending and axial tension due to the movement of the
tion from 2 m to 2.75 m for a 10 m deep wall has the embedded wall and therefore it is important that these
effect of reducing the passive earth pressure coefficient piles are designed for these additional forces.
from 6.4 to about 5.3. This results in a reduction in the Where the passive failure mechanism does not com-
limiting passive resistance from 2100 kN/m to about prise squeezing of the ground between the piles, two-
1835 kN/m which is very similar to the passive resist- dimensional plane-strain analyses in which the piles are
ance predicted by this three-dimensional analysis. represented by a wall with equivalent smeared proper-
ties will provide a safe estimate of the passive resist-
ance. The analyses suggest that where piles are spaced
4 DISCUSSION at 3 diameters or less in a cohesive deposit, squeezing of
the ground between the piles is unlikely to occur unless
There appears to be a good match between the passive the piles are located nearer to the wall than 30% of the
resistance obtained in the three-dimensional analysis embedded length. Where the piles are represented by a
and the predicted passive resistance made from the structural element, the additional soil present in the
results of the two-dimensional analyses despite the obvi- model between the embedded wall and the structural
ous limitations of these analyses. A sensitivity study member will lead to a conservative estimate of the
carried out using the results of the two-dimensional passive resistance.
analyses for the undrained material shows that squeez-
ing of the ground between the piles is more likely when:
1. The embedded retaining wall and piles are rough REFERENCES
rather than smooth.
2. The piles are spaced more widely. BS8002: 1994. Code of Practice for Earth Retaining
3. The piles are nearer to the wall. Structures. British Standards Institution.
Chen C-Y & Martin, G.R. 2002. Soil-structure interaction
Using the two-dimensional analyses it is possible to for landslide stabilizing piles. Computers and
identify the critical pile spacing defining the change Geotechnics 29: 363–386.
in passive failure mechanism from squeezing of the Civil Engineering Code of Practice No. 2 1951. Earth
ground between the piles to squeezing upwards in front Retaining Structures, Institution of Structural Engineers,
of the piles. For a 10 m long embedded rough wall and London.
309
ABSTRACT: A spate of fires in tunnels over the past decade, causing serious loss of life and significant
structural damage, has led to new safety concepts in tunnel construction. Nowadays these concepts are already
being considered during the design. In this context numerical methods represent a powerful tool for assessing
the structural forces in the lining and the change of material properties caused by thermal effects.
This article deals with the simulation of a fire within a tunnel by means of the program FLAC. The tunnel lin-
ing is modeled by four-node continuum elements for simulating non-linear and time dependent temperature
variation within the lining. The thermal effect is applied according to the fire load curve of the BEG-project,
a future major railway section between Italy and Austria passing the Alps, at the inside of the lining. The coeffi-
cient of thermal transmission between the thermal source and the lining is chosen such that the temperature-
field within the lining corresponds to experimental data. Non-linear material behavior due to thermal loading is
implemented by varying the coefficient of thermal expansion.
313
1000 sM = E . 1 = 3K . (1 − 2v) . 1
Temperature [°C]
800
ϑ0 ϑ1 ϑ2 ϑ
600 − +
elϑ
400
M
200
Figure 1. Fire load curve. parameters, until no further stresses can be taken, i.e.
material destruction. The presentation in form of a
usual stress–strain diagram is insufficient and inap-
Temperature of concrete during BEG-1 fire propriate as the process is still controlled by tempera-
1200 ture. To be able to define a material-law dependent on
1100 temperature, the thermal process is depicted by analogy
to the stress–strain behavior (dashed axes).
1000 180,ξ Due to material warming, the elementary volume
900 expands linearly proportional to the temperature
360,ξ
increase (el) and the temperature expansion coeffi-
Temperature of concrete [°C]
800
540,ξ cient, whereby the material behavior is temperature
700 independent (Fig. 3, 0 1 70°C). Due to
1800,ξ
restrained thermal expansion in the closed ring struc-
600
3600,ξ ture of a tunnel lining, the initial stress increase is cal-
500 culated as:
5400,ξ
400
7200,ξ (1)
300
10080,ξ
200
(2)
100
314
where
[Dij] deviatoric part of stress tensor;
[0ij] volumetric part of stress tensor;
[ij] total stress tensor;
[Dij] deviatoric part of strain tensor;
[0ij] volumetric part of strain tensor;
G, K shear and compression modulus, respectively
(Školska Knjiga 1996).
Based on the assumption that the deviatoric part of
the stress tensor is negligibly small in a closed tunnel
lining the total stress then equals the volumetric part
of the stresses:
(6)
(7)
315
At a temperature 2 700°C the stress in the ele- If the temperature in the elementary volume is greater
mentary volume has to be zero. than 700°C this effect is called as physical material
destruction and the elastic modulus decreases to zero.
(17) As a zero value of the elastic constants within a numer-
ical model is not possible a further correction of the
so that we can write: coefficient of expansion is necessary in order to achieve
a compensation of the increasing stresses, which would
be caused by static loads. The external load would cause
(18) a negative extension (εM, Fig. 3) and therefore gener-
ate a compressive stress (M, Fig. 3) which should be
(19) compensated by thermal expansion. The maximum
temperature in the element can theoretically reach the
value of the temperature source (3 1200°C, fire load
(20) curve). Again, we can derive a temperature depend-
ence of the coefficient of thermal expansion:
where I 1 major principal stress at a temperature
1 70°C and 1 equivalent coefficient of thermal (21)
expansion for temperature between 70°C and 700°C.
From Equation 20 we can easily obtain an equiva- (22)
lent coefficient of thermal expansion for the relax-
ation area:
JOB TITLE : G4126 BEG-Stans/Terfens, RQ 9a - km 53+500, Fire Load - 120 minutes (*10^2)
LEGEND
0 2E 1
-0.500
-0.700
-0.900
-1.100
Geoconsult ZT GmbH
Salzburg - Austria
-0.200 0.000 0.200 0.400 0.600 0.800 1.000 1.200
(*10^2)
316
linear-elastic material model. The lining and the adja- 5 VERIFICATION OF MATERIAL-MODEL
cent soil are linked with so called “Interface” ele- AND CALCULATION RESULTS
ments. For the tunnel lining a Young’s modulus of
E 30 GPa and a Poisson’s ratio of
0.2 has been The verification of the temperature fields at different
chosen. The soil is modeled according to an elastic– time steps is done by comparison of the experimental
plastic material model, the Mohr–Coulomb failure data with the temperature pattern within the tunnel
criterion. To keep the calculation time within reason- lining. As it can be seen in Figure 7 the numerical
able time limits only one half of the system is mod- results match well with the experimental values.
eled, introducing symmetry boundary conditions. For verification of the material law a beam (1 m
The horizontal displacements at the symmetry axis wide, 45 cm high) was modeled by using a 1 cm by
at the right boundary of the mesh are fixed as well as 1 cm zone size. A linear-elastic material model was
the vertical displacements at the bottom boundary of used with a Young’s modulus of 30 GPa and a Poisson’s
the model. At the surface a constant uniformly distrib- ratio of 0.25. Normal pressure of 3.0 MPa was applied
uted load of 100 kN/m2 is applied. The initial stress on the vertical boundaries and after static calculation
state is defined applying a lateral earth pressure coef- a thermal load according to the fire load curve (Fig. 1)
ficient of 40% of the vertical pressure. The imple- was applied on the bottom of the model. The “while-
mentation of the tunnel lining is performed without any stepping” loop was used for calculating the equivalent
relaxation immediately after excavation.
After the static analysis, i.e. equilibrium within the JOB TITLE : G4126 BEG-Stans/Terfens, RQ 9a - km 53+500, Fire Load - 120 minutes
to the literature. 5 10 15 20 25 30 35 40
The calculation was performed as a coupled time- GEOCONSULT ZT GmbH
-02
(10 )
Salzburg - Austria
dependent mechanical analysis where mechanical and
thermal time steps, which were calculated in real time,
changed cyclically. Figure 7. Depth of temperature penetration in FLAC.
LEGEND 0.900
19-May-03 8:03
step 45768
0.700
Thermal Time 7.2000E+03
-4.259E+00 <x< 1.019E+01
-4.270E+00 <y< 1.018E+01
0.500
Grid plot
0 2E 0
0.300
Applied Heat Sources
O Max Value = 1.302E+03
0.100
-0.100
-0.300
Geoconsult ZT GmbH
Salzburg - Austria
-0.300 -0.100 0.100 0.300 0.500 0.700 0.900
(*10^1)
317
JOB TITLE : 90 minutes JOB TITLE : G4126 BEG-Stans/Terfens, RQ 9a - km 53+500, Fire Load - 120 minutes
FLAC (Version 4.00)
FLAC (Version 4.00)
-01
+07 LEGEND (10 )
LEGEND (10 )
20-May-03 9:34
2.000
2-Jul-03 15:38 step 45768
2.400
step 10200 Thermal Time 7.2000E+03 1.500
Thermal Time 5.2000E+03 2.000 1.000
HISTORY PLOT
HISTORY PLOT Y-axis : 0.500
1.600
Y-axis : m_his_239 (FISH)
X-axis : 0.000
Rev_Prin. stress 1( 50, 2)
1.200 Number of steps
X-axis : -0.500
ztemp2 (FISH)
0.800 -1.000
-1.500
0.400
-2.000
0.000
39 40 41 42 43 44 45
+03
1 2 3 4 5 6 7 8 9 Geoconsult ZT GmbH (10 )
+02 Salzburg - Austria
GEOCONSULT ZT GmbH (10 )
Salzburg - Austria
39 40 41 42 43 44 45
Geoconsult ZT GmbH +03
(10 )
10 20 30 40 50 60 70 Salzburg - Austria
+01
Geoconsult ZT GmbH (10 )
Salzburg - Austria
39 40 41 42 43 44 45
Geoconsult ZT GmbH (10
+03
)
Figure 13. Time progression of normal force in roof.
Salzburg - Austria
318
loading are presented for sections in the roof and shoul- to the length of the element as well as the number of
der area. Until application of a fire load a negative the Gauss integration points is a very important factor
moment predominates (tension soil-sided) whereas in for the relation of the internal forces from the stresses
the roof a positive moment (tension cavity-sided) of continuum elements. Therefore an optimum number
predominates. of elements as well as geometry of the elements
If the boundary elements do not reach the so-called should be achieved in order not to falsify the calcula-
“Creep-Temperature” it is apparent that at the begin- tion results.
ning the negative moment in the shoulder increases Local phenomena like spalling and local loss of
significantly, but immediately afterwards the moment thermal protection, which lead to irregular temperature
decreases because of the loss of stress reception. variation within the cross-section of a concrete lining,
In the roof, constraint stresses develop from the are factors which influence the results considerably.
temperature. On account of this reason the tensile These effects are not considered in this paper.
stresses at the inside of the tunnel lining are reduced The internal forces, especially the moments are
or turn into compressive stresses. As a result of fire considerably influenced by the rate of temperature
loads the positive moment reduces as well until the spread. All calculations show that the moments are
elements achieve the relaxation temperature. Afterwards predominately effected within the first minutes of a
the moments change their direction in dependence on fire case as soon the concrete has the full stiffness and
the temperature in the next element rows. stress reception capability. The negative moments
The normal force in the lining is calculated over (tensile stresses soil-sided) increase rapidly and after-
the projection of the stresses normal to the cross-section wards decrease slowly because of the reduction of the
through integration over the thickness. From Figures 12 stiffness and loss of stress reception capability within
& 13, it is obvious that the normal force increases the heated zones of the lining. The positive moments
suddenly and decreases after failure of the material (tensile stresses cavity-sided) show the same trends
within a part of the lining elements. Afterwards the and can change their sign in dependence of the initial
normal force levels around the initial value and stays stress state. These extreme values can provide impor-
more or less constant. tant information for dimensioning the lining.
Within the first minutes of a fire load the normal
forces in the tunnel lining show the same trend as the
bending moments. As a result of fire loads a signifi-
6 CONCLUSIONS
cant stress increase at the inside of the tunnel lining
can speed up the spalling of concrete.
From the calculation results it is apparent that a
numerical simulation of thermo-mechanical processes
is possible. The quality of the results is dependent on
the quality of the implemented material-law and on REFERENCES
the amount of experimental data on which the material
model and the thermal process can be calibrated. Itasca Consulting Group, Inc. 2000. FLAC – Fast Lagrangian
The temperature pattern within the first centimeter Analysis of Continua, Version 4.0 User’s Manual.
of the section has the steepest temperature gradients Minneapolis: Itasca.
Školska Knjiga, 1996. Inženjerski Priručnik, Zagreb:
and is highly non-linear. In comparison to experimen- Stručno-Znanstvena Redakcija Biblioteke.
tal data the temperature interpolation in the center of Kusterle W., Waubke N.V., 2001. Baulicher Brandschutz –
the element (FLAC) of a discretisized model is linear. Betontechnologie, Innsbruck: Institut für Baustoffe und
Because of the accuracy of the calculated bending Bauphysik der Universität Innsbruck.
moments and normal forces the relation of the height Gresslehner K.H., 2001. Festlegung der BEG-1 Kurve.
319
M. Beaty
Senior Engineer, Calif. Dept. of Water Resources, Sacramento, CA, USA
ABSTRACT: A fully coupled effective stress dynamic analysis procedure for modeling seismic liquefaction
is presented. An elastic plastic formulation is used for the constitutive model UBCSAND in which the yield loci
are radial lines of constant stress ratio and the flow rule is non-associated. This is incorporated into the 2D version
of FLAC by modifying the existing Mohr-Coulomb model. This numerical procedure is used to simulate centrifuge
test data from the Rensselaer Polytechnic Institute (RPI). UBCSAND is first calibrated to cyclic simple shear tests
performed on Nevada sand. Both pre- and post-liquefaction behavior is captured. The centrifuge tests are then
modeled and the predicted accelerations, excess porewater pressures, and displacements are compared with the
measurements. The results are shown to be in general agreement when stress densification and saturation effects
are taken into account. The procedure is currently being used in the design of liquefaction remediation measures
for a number of dam, bridge, tunnel, and pipeline projects in Western Canada.
321
(a) σv0
u Increment Vector f
Shear Stress, τ
Elastic
Shear Strain, γ
(b)
Contraction Dilation Normal Effective Stress, s'
p
Plastic Volumetric Strain Increment, dεv
Pore Pressure, u
Shear Strain, γ
f
τ , dγ p
(c)
Plastic Strain
B
Effective stress, σv
Increment Vector
σ'v0
A Yield Locus
φd
p
s', dεv
Shear Strain, γ
Figure 3. UBCSAND model.
Figure 1. Undrained response of loose sand in simple shear:
(a) stress–strain, (b) pore pressure, and (c) effective stress
response. dilation angle, . This model is really too simple for
soils since plastic strains also occur for stress states
can cause high excess pore pressures resulting in a very below the strength envelope. The UBCSAND stress–
large reduction in shear stiffness. Large shear strains strain model described herein modifies the Mohr-
may occur, and the soil will dilate with these strains Coulomb model incorporated in FLAC to capture the
unless the soil is very loose. This dilation causes the plastic strains that occur at all stages of loading. Yield
porewater pressure to drop and the stiffness to increase, loci are assumed to be radial line of constant stress
which can limit the strains, induced by a load cycle. ratio as shown in Figure 3. Unloading is assumed to be
This behavior is illustrated in Figure 1 for monotonic elastic. Reloading induces plastic response but with a
loading. stiffened plastic shear modulus.
It is this tendency of the soil skeleton to contract and The plastic shear modulus relates the shear stress and
dilate that controls its liquefaction response. Once the the plastic shear strain and is assumed to be hyper-
skeleton behavior is modeled, the response under bolic with stress ratio as shown in Figure 4. Moving
drained, undrained or coupled stress-flow conditions the yield locus from A to B in Figure 3 requires a plas-
P
can be computed by incorporating the bulk stiffness and tic shear strain increment, , as shown in Figure 4,
flow of the pore fluid. and is controlled by the plastic shear modulus, GP. The
associated plastic volumetric strain increment, dPv, is
obtained from the dilation angle :
322
(5)
Dg p
GPi
Ge and
depends on relative density, is the
Plastic Shear Strain, g p current shear stress, f is the projected shear stress at
failure, and Rf is the failure ratio used to truncate the
Figure 4. Hyperbolic stress–strain relationship.
hyperbolic relationship.
The position of the yield locus d is known for each
element at the start of each time step. If the stress ratio
increases and plastic strain is predicted, then the yield
Dilation locus for that element is pushed up by an amount d
as given by Equation 6. Unloading of stress ratio is con-
Shear Stress,
(6)
323
Dr=44%
0.15 Dr=40%
20
15 (a) Drc=40% CSR=0.1
Shear Stress (kPa)
0.1
10
5
0 0.05
-5 Test: Dr=40%
-10 Test: Dr=44%
-15 0
Test Calibration 1 10 100
-20
0 20 40 60 80 100 No. of Cycles to Liquefaction
Vertical Effective Stress (kPa)
Figure 8. Predicted and measured liquefaction response of
20 Fraser River sand.
(b) Drc=40% CSR=0.1
15
Shear Stress (kPa)
10
5
of the test results and the UBCSAND calibration are
shown in Figure 8. The predicted and measured lique-
0
faction response for v0 of 100 and 200 kPa is in close
-5 agreement.
-10
-15 Test Calibration
-20 5 CENTRIFUGE TESTS
-20 -15 -10 -5 0 5 10 15 20
Shear Strain (%) A simulation using UBCSAND was made of 2 cen-
trifuge tests carried out at RPI as described in Table 1.
Figure 6. Stress path and stress–strain relationship In the centrifuge test, a small model is used that is
(CSR 0.1). subjected to a high acceleration field during the test.
324
This has the effect of increasing its stresses by the that is 120 (Model 1) or 60 (Model 2) times larger
ratio of the induced acceleration divided by the accel- than the actual model. Results from the centrifuge test
eration of gravity. This ratio or factor is 120 for can be presented at either the model or prototype scale.
Model 1 and 60 for Model 2 as indicated by Table 1. The prototype scale is used for this paper.
The centrifuge model under the increased acceleration While in flight, a motion simulating an earthquake
field can also be thought of as representing a prototype time history is applied to the base of the model. For
dynamic similitude at the model scale, the earthquake
time scale must be decreased by a factor of 120 (Model
Table 1. Centrifuge model tests. 1) or 60 (Model 2), and the earthquake acceleration
increased by the same factor. The engineering coeffi-
RPI Model 1 RPI Model 2 cient of permeability k will also increase by this same
factor due to the increased unit weight of the fluid. k
Test condition Level Slope should be decreased for hydraulic similitude, although
Dr 55% 40% it is not necessary to model a specific k. It is common
Centrifuge acc. 120 g 60 g
Max. v 380 kPa 100 kPa
to use a fluid in the test that is 30 to 60 times more vis-
Soil depth 38 m 10 m cous than water to prevent rapid rates of dissipation
Fluid viscosity 60 w 60 w that might unduly curtail liquefaction effects.
Nevada sand was used for these centrifuge tests
and its liquefaction and permeability (at 1 g using
water as pore fluid) properties were obtained from
0.5
UBCSAND laboratory tests (Arulmoli et al. 1992, Kammerer
Dr=44% et al. 2000, Taboada-Urtuzuastegui et al. 2002). Its
Cyclic Stress Ratio (CSR)
Z = 0.0 m
Ac6 Ac7 Z = 1.3 m
P7 Ac5 P8 Z = 6.3 m
P5 Ac4 P6 Z = 13.1 m
325
0.3
Centrifuge Input Numerical Input
0.2
Acceleration (g)
0.1
-0.1
-0.2
-0.3
0 10 20 30 40
Time (Sec)
60 Depth =13.1m
Layer kG kB (kPa) (m/sec) 40
20 -- σ' vo
0
Top 952 2856 0.44 0.2 E5 5 E-5 0 10 20 30 40
200
150
100 Depth = 24.8m
M (measurement) P (prediction) 50 -- σ' vo
0.4
M(depth=1.3m) P(depth=1.3m) 0
0.2
Acc (g)
0 10 20 30 40
0
-0.2
350
-0.4 300
0.4 250
EPP (kPa)
M(depth=6.3m) P(depth=6.3m)
0.2 200
Acc (g)
0
400
-0.2
EPP (kPa)
-0.4 300
-- σ' vo
0
0
-0.2 0 10 20 30 40
-0.4 Time (sec)
0.4
M(depth=30.8m) P(depth=30.8m)
0.2
Figure 12b. Measured and predicted excess pore pressures
Acc (g)
0
-0.2 of Model 1.
-0.4
0.4
M(depth=37.0m) P(depth=37.0m)
0.2 The container for Model 1 consisted of slip “rings”
Acc (g)
0
-0.2
that allowed differential horizontal displacements in
-0.4
0 10 20 30 40 10 20 30 40
the vertical direction but not in the horizontal. This
Time (sec) Time (sec) was simulated in the FLAC model by “attaching” the
vertical sides, Figure 10. The initial horizontal effec-
Figure 12a. Measured (left) and predicted (right) accelera- tive stresses were set to 0.5 times the vertical effective
tions of Model 1. stresses.
The measured and predicted excess pore pressures
in Figure 11 and consisted of 50 cycles with an ampli- and accelerations for various depths are shown in
tude of 0.2 g and a frequency of 1.5 Hz. The key inputs Figure 12. The predicted accelerations are initially
including water bulk stiffness (Bf) for different layers about the same at all depths and approximately equal to
in the numerical model are listed in Table 2. the base input value of 0.2 g. The accelerations decrease
326
over much of the model as the shaking continues. The Table 3. Key input for Model 2 numerical analysis.
decay of acceleration is most rapid in the upper layers
and can be explained in terms of the excess porewater Bf after
pressures shown in Figure 12b. A large drop in acceler- spinup Permeability
ation response occurs when the measured excess pore Layer kG kB
(kPa) (m/sec)
pressure reaches the initial vertical effective stress v0, Free field 867 2601 0.22 1.0 E5 2.1 E-5
which corresponds to a liquefied state. Measurements
show that liquefaction occurs first near the surface and
then progresses downward. The accelerations and
excess pore pressures predicted using UBCSAND are 5.2 Model 2
in generally good agreement with the measurements.
The analysis described above incorporates the effect The cross section for Model 2 is shown in Figure 13
of densification due to the increased acceleration field. and comprises a steep 1.5:1 slope in loose fine sand
If this effect is not considered, and a uniform Dr of with Dr 40% (Taboada-Urtuzuastegui et al. 2002).
55% is used in the analysis, then liquefaction is pre- The base motion consists of 20 cycles of 0.2 g at a fre-
dicted to occur first at the base of the model rather quency of 1 Hz. The container for model 2 was rigid
than at the surface. The higher Dr at the base reverses and this was simulated in the FLAC model by apply-
this trend and indicates the importance of stress den- ing the input motion to the vertical sides as well as the
sification in centrifuge tests. base. The key inputs for Model 2 are listed in Table 3.
Full saturation of the pores is difficult to achieve in a Pore pressures and accelerations were measured away
centrifuge test. The best fit with the data was obtained from the face of the slope, approximating free field
assuming an initial placement saturation, or Sr, of 98% conditions, as well as adjacent to the slope.
at atmospheric pressure. The pore pressure will increase The predicted and observed accelerations and pore
as the centrifuge acceleration is applied, and the result- pressures in the free field are shown in Figures 14 & 15.
ing increase in Sr is modeled using the gas laws. As expected, similar trends are seen as for the level
In summary, ground test of Model 1, i.e. ru of 100% and reduced
accelerations.
(a) UBCSAND provides a reasonable agreement to The accelerations and pore pressures near the
the test results, slope are shown in Figures 16 & 17. It may be seen in
(b) ru 1.0 and liquefaction can occur at depths of Figure 16 that there is little or no reduction in the
40 m in medium dense sand strata, accelerations. Instead, large upslope acceleration spikes
(c) a large reduction in the accelerations can occur occur. Excess pore pressures are shown in Figure 17.
upon liquefaction, Large negative excess pore pressure spikes occur that
(d) the effect of stress densification should be coincide in time with the upslope acceleration spikes.
included, and The slope is steep and the upslope acceleration of the
(e) the degree of saturation, Sr, must be considered. base tends to induce failure of the slope and relative
327
0.2 Input
0.2
0.0 0
-0.2
Input (a) -0.2
0.4
0.4 AH1
0.0
0
Acceleration (g)
-0.4
(a)
AH1 -0.4
0.8 0.8 AH5
0.4 0.4
0.0 0
-0.4 -0.4
AH5 (b)
-0.8 -0.8
0.8 (Taboada-Urtuzuastegui et al. 2002) 0.8 AH6
0.4 0.4
0
0.0
AH6 (c) -0.4
-0.4
-0.8
-0.8
0 5 10 15 20 5 10 15 20
Time (sec) Time (sec)
Figure 14. Measured (left) and predicted (right) accelerations at free field.
80
80
40 40
PP1 (a)
Excess pore pressure (kPa)
0 PP1
0
40 40
20 20
PP5 (b)
0 0 PP5
20 ru = 1.0 20
10 10
PP6 (c)
0 (Taboada-Urtuzuastegui et al. 2002) 0 PP6
-5
0 5 10 15 20 5 10 15 20
Time (sec) Time (sec)
Figure 15. Measured (left) and predicted (right) excess pore pressures at free field.
downslope movement. The soil dilates as it shears in downslope movement and gives rise to the accelera-
the downslope direction, producing negative pore tion spike (Taboada-Urtuzuastegui et al. 2002).
pressures which stiffen the shear modulus. Enough UBCSAND provides a reasonable prediction of
strength is mobilized through this dilation to arrest the the accelerations and pore pressure response for the
328
Input
0.2 0.2
0.0 0
-0.4 -0.4
AH2 (b)
0.2 0.2
0.0 0
-0.2 -0.2
-0.4 -0.4
-0.6 (c) -0.6 AH4
AH4
AH7
0.4 (Taboada-Urtuzuastegui et al. 2002) 0.4
0.0 0
-0.4 -0.4
AH7 (d)
-0.8 -0.8
0 5 10 15 20 5 10 15 20
Time (sec) Time (sec)
Figure 16. Measured (left) and predicted (right) accelerations near the slope.
20 PP2
20
10
10
0
PP2
0 (a) -10
30 30
20
Excess pore pressure (kPa)
20
10 10
PP3 PP3
0 (b)
0
20 20 PP4
0 0
-20 -20
PP4 (c)
ru = 1.0
10 10
0 0
-10
-10 PP7 -20
-20 (Taboada-Urtuzuastegui et al. 2002) (d) -30 PP7
-30
0 5 10 15 20 5 10 15 20
Time (sec) Time (sec)
Figure 17. Measured (left) and predicted (right) excess pore pressures near the slope.
329
10 10
8 8
6 6
4 4
2 2
0 0
50 45 40 35 35 30 25 20 15 10 5 0
Figure 18. Measured displacements for Model 2 from centrifuge test (Taboada-Urtuzuastegui et al. 2002).
free field. More significant differences are observed (e) the dilative spikes prevented very large displace-
for locations near the slope. Some of these differences ments from occurring in this homogeneous fine
are due to UBCSAND under predicting the dilative sand model.
spikes. This requires further investigation. The meas-
ured and predicted displacements after shaking are
shown in Figures 18 & 19. It may be seen that both the 6 SUMMARY
magnitude and pattern of displacements are in gen-
eral agreement. A fully coupled effective stress dynamic analysis
In summary, procedure has been presented. The procedure is first
(a) UBCSAND provides reasonable agreement with calibrated by comparison with laboratory element
this centrifuge test, although further study is test data and then verified by comparison with two
needed for locations close to the sloping face, centrifuge model tests.
(b) a decrease in accelerations after liquefaction was Model 1 represented a deep sand layer with a level
not observed near the slope, ground condition. This model showed that high excess
(c) a large upslope acceleration spikes occurred near porewater pressure and liquefaction can occur to
the slope, depths of 40 m in medium dense sands. Liquefaction
(d) a decrease in pore pressure due to dilation corre- first occurred at the surface and progressed down-
sponded with these upslope acceleration spikes, and ward under continued shaking. Accelerations above
330
the depth of liquefaction showed a significant Beaty, M. & Byrne, P. 1998. An effective stress model for
decrease. The numerical model results were in good predicting liquefaction behaviour of sand. ASCE Geot.
agreement with the measurement when stress densifi- Special Pub. No. 75: 766–777.
cation and saturation effects were included. Gonzalez, L., Abdoun, T. & Sharp, M.K. 2002. Modeling of
seismically induced liquefaction under high confining
Model 2 represented a steep slope condition in stress.
homogeneous loose fine sand. The results showed that Kammerer, A., Wu, J., Pestana, J., Riemer, M. & Seed, R.
large upslope acceleration spikes occurred near the face 2000. Cyclic simple shear testing of Nevada sand for
of the slope after liquefaction. These acceleration PEER Center project 2051999. Geotechnical Engineering
spikes corresponded with large negative excess pore Research Report No. UCB/GT/00-01, University of
pressure spikes associated with dilation. It is the California, Berkeley, January.
increase in effective stress associated with these nega- Puebla, H., Byrne, P.M. & Phillips, R. 1997. Analysis of
tive pore pressure spikes that curtails the displacements CANLEX liquefaction embankments: prototype and
and makes the slope more stable than might be expected centrifuge models. Can. Geotech. Journal, Vol. 34, No. 5:
641–657.
under cyclic loading. The overall pattern of predicted Skempton, A.W. 1986. Standard penetration test procedures
response is in reasonable agreement with the measure- and the effects in sands of overburden pressure, relative
ments, although both the acceleration and pore pressure density, particle size, ageing and overconsolidation,
spikes are under predicted by the UBCSAND analysis. Geotechnique 36, No. 3: 425–447.
A new series of centrifuge tests are planned at Taboada-Urtuzuastegui, V.M., Martinez-Ramirez, G. &
CCORE (Centre for Cold Ocean Research), Memorial Abdoun, T. 2002. Centrifuge modeling of seismic behav-
University, Newfoundland, which will permit further ior of a slope in liquefiable soil, Soil Dynamic and
verification and refinement of the numerical model. Earthquake Engineering, Vol. 22: 1043–1049.
Youd, T.L., Idriss, I. M., Andrus, R.D., Arango, I., Castro, G.,
Christian, J.T., Dobry, R., Finn, W.D.L., Harder Jr., L.F.,
Hynes, M.E., Ishihara, K., Koester, J.P., Liao, S.,
REFERENCES Marcuson III, W.F., Martin, G.R., Mitchell, J.K.,
Moriwaki, Y., Power, M.S., Robertson, P.K., Seed, R.B. &
Arulmoli, K., Muraleetharan, K.K., Hossain, M.M. & Fruth, Stokoe, K.H. 2001. Liquefaction Resistance of Soils:
L.S. 1992. VELACS laboratory testing program, soil data Summary Report from the 1996 NCEER and 1998
report. The Earth Technology Corporation, Irvine, NCEER/NSF Workshops on Evaluation of Liquefaction
California, Report to the National Science Foundation, Resistance of Soils. ASCE J. of Geot. and Geoenv. Eng.,
Washington D.C., March. Vol. 127, No. 10: 817–833.
331
R.A. Green
Department of Civil and Environmental Engineering, University of Michigan, Ann Arbor, MI, USA
R.M. Ebeling
Information Technology Laboratory, US Army Engineer Research and Development Center, Vicksburg, MS, USA
ABSTRACT: A research investigation was undertaken to determine the dynamically induced lateral earth pres-
sures on the stem portion of a concrete, cantilever, earth-retaining wall. In total, the wall-soil column system was
68.6 m in height, with the upper 6.1 m being composed of the cantilever wall retaining compacted backfill. A series
of numerical analyses were performed using FLAC. The analyses consisted of the incremental construction of the
wall and placement of the backfill, followed by dynamic response analyses, wherein the soil was modeled as elasto-
plastic. This paper outlines the details of the numerical model used in the analyses. Particular attention is given to
how the ground motion was specified, determination of the wall and soil model parameters, and the modeling of the
wall-soil interface. To benchmark the FLAC results, comparisons are presented between the FLAC results and the
results from simplified techniques for computing dynamic earth pressures and permanent wall displacement.
333
3
boundaries,” thus simulating radiation damping.
Using a free-field acceleration time-history recorded
3m
334
Error analyses were performed on the spectra corre- had little energy at higher frequencies. The interlayer
sponding to the different profiles and different types motion (at 9.1 m depth) computed using SHAKE was
of specified input motions. In the error analyses, the specified as an acceleration time-history along the
spectra for the SHAKE motions were used as the base of the FLAC model.
“correct” motions. The word “correct” does not imply
that SHAKE precisely models the behavior of an actual
2.4 Model parameters for soil
soil profile subjected to earthquake motions. Rather,
SHAKE gives the analytically correct motion for a The stress-strain behavior of the soil was modeled using
visco-elastic profile with constant damping applied to the Mohr-Coulomb constitutive model. Four parame-
all frequencies of motion. On the other hand, the FLAC ters are required for the Mohr-Coulomb model: effec-
models used in this study give numerical approxima- tive internal friction angle (); mass density (); shear
tions of the correct analytical solution. The errors in modulus (G); and bulk modulus (K). The first two
the FLAC spectral values were computed at a spec- parameters (i.e., and ) are familiar to geotechni-
trum of frequencies using the following expressions. cal engineers, where mass density is the total unit
weight of the soil (t) divided by the acceleration due
to gravity (g), i.e. t/g. As stated previously,
(1a) for the foundation soil was 40° and 35° for the back-
fill. These values are consistent with dense natural
deposits and medium-dense compacted fill. G and K
(1b) may be less familiar to geotechnical engineers; there-
fore, their determination is outlined below.
Several correlations exist that relate G to other
From the results of the parametric study, it was soil parameters. However, the most direct relation is
determined that the specification of the input motion between G and shear wave velocity (vs):
in FLAC in terms of stress time-histories gives the
least accurate results, wherein the stress time- (2)
histories were applied at a “quiet boundary” along the
base of the FLAC model. The errors corresponding to
specifying the motions in terms of acceleration and s may be determined by various types of site charac-
velocity time-histories were essentially identical and terization techniques, such as cross hole or spectral
considerably less than those associated with the stress analysis of surface waves (SASW) studies.
time-histories. Values for K can be determined from G and
Poisson’s ratio (v) using the following relation:
335
1.5 m 1 Beam
Beam Elements
Elements Interface
Elements
1.5 m 2
6.1m
6.1m
1.5 m 3
No Interface
Elements
1.5 m 4
5
4m 4m
Figure 3. Numerical model of retaining wall using elastic Figure 4. Location of interface elements in the FLAC model.
beam elements.
on the member. In dynamic analyses, it is difficult to
elements: cross sectional area (Ag); mass density (); state a priori whether the use of sectional properties
elastic modulus (Ec); and second moment of area (I ), corresponding to uncracked, fully cracked, or some
commonly referred to as moment of inertia. intermediate level of cracking will result in the largest
The basis for subdividing the wall into five seg- demand on the structure. However, I 0.4 Iuncracked
ments was the variation of the mechanical properties was used as a reasonable estimate for the sectional
in the wall. A wall having a greater taper or largely properties (Paulay & Priestley 1992).
varying steel reinforcement along the length of the
stem or base would likely require more segments.
For each of the segments, Ag and were readily 2.6 Model parameters for wall-soil interface
determined from the wall geometry and the unit weight Interface elements were used to model the interaction
of the concrete (i.e. 23.6 kN/m3). Ec was computed between the concrete retaining wall and the soil.
using the following expression (e.g. MacGregor 1992): However, FLAC does not allow interface elements to
be used at the intersection of branching structures
(5) (e.g. the intersection of the stem and base of the can-
tilever wall). Several approaches were attempted by
In this expression, f c is the compressive strength of the authors to circumvent this limitation in FLAC, with
the concrete (e.g. 4000 psi for the wall being modeled), the simplest and best approach, as found by the authors,
and both Ec and f c are in psi. Because the structure is illustrated in Figure 4. As shown in this figure, three
continuous in the direction perpendicular to the analy- very short beam elements, oriented in the direction of
sis plane, Ec computed using Equation 5 needed to be the stem, toe side of the base, and heel side of the
modified to account for plane-strain conditions. This base, were used to model the base-stem intersection.
modification was done using the following expression No interface elements were used on these three short
(Itasca 2000, FLAC Structural Elements Manual). beam elements. However, interface elements were used
along the other contact surfaces between the soil and
(6) wall, as depicted by the hatched areas in Figure 4.
A schematic of the FLAC interface element is pre-
sented in Figure 5. As may be observed from this figure,
where 0.2 was assumed for Poisson’s ratio for concrete. the interface element has four parameters: S slider
I is a function of the geometry of the segments, the representing shear strength; T tensile strength;
amount and location of the reinforcing steel, and the kn normal stiffness; and ks shear stiffness. The
amount of cracking in the concrete, where the latter in element allows permanent separation and slip of the
turn depends on the static and dynamic load imposed soil and the structure, as controlled by the parameters
336
τ (psf)
400 FLAC
kn hyperbolic model
300
model
zone
200
Side B of Interface
100
Figure 5. Schematic of the FLAC interface element (adapted
from Itasca 2000).
0.000 r 0.002 0.004 0.006 0.008 0.010
s (ft)
T and S, respectively. For the cohesionless soil being
modeled, T 0, while S was specified as a function
Figure 6. Calibration of the FLAC interface model to the
of the interface friction angle (
). For medium-dense
hyperbolic-type model proposed by Gomez et al. (2000a,b).
sand against concrete,
31° (Gomez et al. 2000b).
As a rule-of-thumb, the FLAC manual (Itasca 2000,
Theory and Background Manual) recommends that kn where,
be set to ten times the equivalent stiffness of the
stiffest neighboring zone, i.e.: (8b)
(8c)
(7)
337
(10b) (13a)
(10c) or
338
Permanent relative
displacement (m)
bound of the values for Rayleigh damping. Judgment 0.5
is required in selecting the damping ratio between the Newmark
0.4
lower and upper bounds; three-percent Rayleigh damp-
0.3
ing was used for most of the retaining wall analyses FLAC
0.2
performed by the authors. The central frequency cor-
0.1
responding to the specified damping ratio is typically
set to either the fundamental period (small strain) of 0.0
the system being modeled (an inherent property of -0.1
0 5 10 15 20 25 30 35 40
the wall-soil system) or predominant period of the
Time (seconds)
system response (an inherent property of the wall-soil
system and the ground motion). For the FLAC analyses
Figure 7a. Comparison of the permanent relative displace-
performed, the central frequency was set equal to the ments computed from the FLAC results and a Newmark slid-
small strain fundamental frequency of the retaining ing block analysis with N*g 0.22 g.
wall-soil system (i.e. ≈6 Hz).
0.7
3 DISCUSSION 0.6
Permanent relative
Several analyses were performed using the model of displacement (m) 0.5 Newmark
0.4
the wall-soil system described above, scaling the input 0.3
motion to different peak ground acceleration values. FLAC
0.2
To assess the adequacy of the model, the results from
0.1
the FLAC analyses were compared with the results from
simplified techniques for estimating the permanent 0.0
wall displacement and the dynamic earth pressures. -0.1
0 5 10 15 20 25 30 35 40
The purpose of the comparisons was only to provide a Time (seconds)
reality check of the FLAC results, while true validation
of the FLAC model would require a comparison with Figure 7b. Comparison of the permanent relative displace-
actual field observations. Comparisons of the results ments computed from the FLAC results and a Newmark slid-
are discussed in the following sub-sections. However, ing block analysis with N*
g 0.27 g.
the reader is referred to Ebeling & Morrison (1992)
and Green and Ebeling (2002) for more detailed dis-
1.0
cussions about the simplified techniques used.
Acceleration (g)
339
coefficient (K)
2.5
was observed in the deformed FLAC mesh. To account
for this additional resistance to sliding, N*
g was 2.0
recomputed assuming a friction angle of 35°, which is
between the interface friction angle (i.e.
31°) and 1.5
the of the foundation soil (i.e. 40°), with the revised 1.0
value of N*
g 0.27 g. A comparison of the perma- KAE
0.5
nent relative displacements computed from FLAC and
the sliding block analyses using the revised value of 0.0
N*
g is shown in Figure 7b. As may be observed from 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
this figure, the predicted displacements are in very kh
close agreement, thus giving credence to the validity
of the proposed FLAC model. Figure 9. Comparison of FLAC and Mononobe-Okabe
dynamic lateral earth pressure coefficients.
3.2 Dynamic earth pressures
The dynamically induced lateral earth pressures act- 1.0
ing on the stem of the wall were computed by FLAC.
N* = 0.27
The corresponding lateral earth pressure coefficients 0.5
(KFLAC) were computed from these stresses using the kh 0.0
following expression (Green et al. 2003): 0 5 10 15 20 25 30 35 40
-0.5 Time (seconds)
(14) -1.0
340
341
Newmark, N.M. 1965. Effects of Earthquakes on Dams and Terzaghi, K. 1943. Theoretical Soil Mechanics. New York:
Embankments. Geotechnique 15(2): 139–160. John Wiley and Sons, Inc.
Okabe, S. 1924. General Theory of Earth Pressures. Journal Wartman, J., Bray, J.D., & Seed, R.B. 2003. Inclined Plane
Japan Society of Civil Engineering 10(6): 1277–1323, Studies of the Newmark Sliding Block Procedure.
plus figures. Journal of Geotechnical and Geoenvironmental Engi-
Paulay, T. & Priestley, M.J.N. 1992. Seismic Design of neering 129(8): in press.
Reinforced Concrete and Masonry Buildings. New York:
John Wiley and Sons, Inc.
342