Computers and Geotechnics 48 (2013) 134145

Contents lists available at SciVerse ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Surface subsidence prediction for the WUTONG mine using a 3-D nite
difference method
Nengxiong Xu a, Pinnaduwa H.S.W. Kulatilake b,, Hong Tian c, Xiong Wu d, Yinhua Nan a, Tian Wei a
a
School of Engineering & Technology, China University of Geosciences, Beijing 100083, China
b
Rock Mass Modeling and Computational Rock Mechanics Laboratories, University of Arizona, Tucson, AZ 85721, USA
c
Department of Geotechnical Engineering, RWTH Aachen University, Aachen 152074, Germany
d
School of Water Resources & Environment, China University of Geosciences, Beijing 100083, China

a r t i c l e i n f o a b s t r a c t

Article history: WUTONG coal mine is adjacent to an auxiliary dam of the Yuecheng Reservoir. In this paper, mining-
Received 18 April 2012 induced surface subsidence prediction is conducted by means of the nite difference method (FDM) to
Received in revised form 21 September 2012 judge whether the extraction of the coal seam will have a negative impact on the dam. First, the initial
Accepted 23 September 2012
values of the rock mass mechanical parameters are estimated using the available literature that relates
Available online 12 December 2012
intact rock and discontinuity properties to rock mass parameters. Then, based on available surface sub-
sidence monitoring data on WUTONGs mined areas, the main mechanical parameters of coal and rock
Keywords:
masses are determined by a back analysis procedure that combines an experimental design technique
Mining
Subsidence prediction
with numerical simulations. Finally, the surface subsidence results in the mining area are numerically
Back analysis predicted for four different mining scenarios (S1 through S4). Scenario S3 emerged as the best choice
Numerical modeling of these four scenarios. The predictions are: (1) the maximum surface subsidence within the mining area
is 2.14 m, with the maximum settlement point located in the mid-west area of the coaleld, and (2) the
nearest distance from the boundary of the surface movement area to the edge of the dam foundation is
35 m. Therefore, mining the coal seam will not cause damage to the dam.
2012 Elsevier Ltd. All rights reserved.

1. Introduction taken into account the effect of geologic structures, geomechanical

The WUTONG coal mine covers an area of approximately
37.36 km2 and is located in Handan City, Hebei Province, the Peo-
ples Republic of China. WUTONGs southwest side is adjacent to
the Yuecheng Reservoir (Fig. 1). As one of Chinas large-scale reser-
voirs, the Yuecheng Reservoir has a total storage capacity of up to
1.3 billion m3 and provides ood control, irrigation, water supply,
power generation, etc. A part of an auxiliary dam for the reservoir
is within the mine area (Fig. 1). It is essential to predict surface
subsidence before mining begins to prevent the dam from being
damaged from mining-induced surface subsidence and to maxi-
mize the mining area to extract the greatest possible amount of
coal.
Research on coal mining subsidence prediction has taken place
for nearly two centuries. Empirical formulas, curves, graphs and
inuence functions have been developed based on numerous mea-
surements of mining subsidence [19]. Although they are quite
simple to understand, such methodologies have not explicitly
Corresponding author. Address: The University of Arizona, 1235 E. James E.
Rogers Way, Mines Bldg., Rm. 131, Tucson, AZ 85721, USA. Tel.: +1 520 621 6064;
fax: +1 520 621 8059.
E-mail address: kulatila@u.arizona.edu (P.H.S.W. Kulatilake).
properties and in situ stress of the rock masses at the mine site on
subsidence prediction. Therefore, the empirical models are very
much site-dependent, and they can only be applied to similar sites.
In contrast, the numerical methods that are based on theories of
continuum and/or non-continuum mechanics predict subsidence
through a mechanistic approach, using rock and coal mass geolog-
ical, physical and mechanical properties. The suggested models
from this group are random eld theory based models [1011],
elastic or visco-elastic theory based models [1214], and contin-
uum or discontinuum numerical models [1519]. The boundary
element method mainly deals with the excavation surface without
modeling the complicated geology that exists due to the presence
of different lithology above the coal seam. Therefore, its capability
of subsidence prediction is not strong as the capabilities of the
nite difference method (FDM) and nite element method (FEM).
Both the FEM and FDM can handle nonlinear and non-
homogeneous characteristics of rock masses. Therefore, they have
become important tools for coal mining subsidence prediction. The
FDM uses an explicit solution procedure in contrast to FEMs
implicit solution procedure. An advantage of the explicit method
is that, because matrices are never formed as in the implicit proce-
dure, large displacements and complex constitutive models for
rock units are possible with no additional computing effort. The

0266-352X/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compgeo.2012.09.014
 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145
Fig. 1. Salient features of the WUTONG coal mine (contour interval is 30 m).
FDM seems to be an appropriate method to perform subsidence
modeling, especially because the Lagrangian formulation permits
materials to yield and ow and the grid to deform in a large-strain
mode [15]. To obtain accurate subsidence prediction through
numerical modeling, the following steps should be taken: (1) use
a simplied geologic system in the numerical model to capture
the essential features of the actual geologic system, including the
in situ stress; (2) select constitutive models that capture the salient
mechanical behavior of the rock masses studied in the numerical
model; (3) use physical and mechanical property values that reect
accurate behavior of the rock masses studied; (4) simulate mining
history reasonably well.
Two inch diameter samples used for intact rock in the labora-
tory contain micro-cracks. The spatial distribution of these cracks
varies from one sample to another. In addition, some material het-
erogeneity can exist in two inch diameter samples. The minor dis-
continuity geometry that exists in rock masses is highly variable.
The discontinuities that exist in a rock mass can be open, closed
or lled with gouge or cementing material. In addition, different
levels of roughness ranging from slickensided surfaces to very
rough surfaces can be expected on open discontinuities. Therefore,
unfortunately, signicant variability and uncertainty exist in esti-
mating rock mass mechanical properties, either through laboratory
or eld tests. Thus, for reliable subsidence prediction, rock mass
mechanical properties should be calibrated using available subsi-
dence monitoring data. Such a procedure is described in the paper.
In this paper, the mining-induced subsidence of the WUTONG
coal mine is predicted by FDM. FLAC3D4.0 software, developed by
the Itasca Consulting Group, Inc. [20], is used in this study. First,
the mechanical parameters of the coal and rock masses in the min-
ing area are estimated by a back analysis method that combines an
experimental design procedure with a numerical simulation. Then,
135
the surface subsidence outcomes induced by four different mining
scenarios are numerically predicted. Finally, after examining these
predictions, a suitable mining scenario for preventing damage to
the dam is proposed for future mining activities.
2. Geology and mining history of the site
2.1. Geology
2.1.1. Faults
Within the WUTONG coaleld, seven major faults exist: F1, F2,
F3, F5, F17, F25 and F31 (Fig. 1). The fault zones of these faults are
approximately 1030 m in width. They are composed of mudstone
and mylonitic rocks as well as breccia. The basic features of the
faults are summarized in Table 1.
2.1.2. Strata and coal seams
The WUTONG coal mine is located east of the Fengfeng coaleld
and is covered by Cenozoic strata. It has been revealed through
drilling that the stratigraphic sequence in WUTONG, in the order
from oldest to newest, is Ordovician, Carboniferous, Permian,
Tertiary and Quaternary, as described below.
(1) The Ordovician (O) stratum, 285654 m thick, is composed
of dark gray or blue-gray limestone. The limestone is pure in qual-
ity and high in strength, with ssures generally lled with calcite.
(2) The Carboniferous (C) stratum, 119149 m thick, is com-
prised of dark gray or black-gray sandstone and siltstone, with
interbeds of 68 thin limestone layers.
(3) The Permian (P) stratum has an average thickness of 1000 m.
The sequence is divided into the following four formations, from
the bottom to top.
136 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145
Table 1
Geometrical features of the faults.
Fault F1 F2 F3
Type Normal Normal Normal
Avg. dip direction () 93 114 110
Avg. dip angle () 73 77 81
Avg. fault throw (m) 200 60 300
Avg. fault thickness (m) 10 28 26
 The Shanxi formation is composed of light-gray to gray, med-
ium- to ne-grained sandstone, gray sandy mudstone and
siltstone.
 The Xiashihezi formation is mainly composed of gray, dark gray,
gray-green, and purple-mottled siltstone.
 The Shangshihezi formation is mainly composed of gray, light
gray, gray-green and yellow medium- to thick-bedded gravel
feldspar coarse sandstone.
 The Shiqianfen formation is mainly composed of purple, pur-
plish red or light purple-red medium- to thick-bedded ne-
grained sandstone.
(4) The Tertiary (T) stratum is divided into four groups: bottom,
lower, middle and upper. At the bottom, conglomerate with weak
bonding is the major unit. The lower group is composed of sand
mixed with gravel and ne-grained soil, while the middle group
consists of ne-grained soil, clay and sandy loam soil, including
lenticular bodies of ne-grained sands. The upper group is com-
posed of sand and gravel mixed with ne-grained soil.
(5) The Quaternary (Q) stratum is composed of scree, laterite,
clay, sandy loam soil, gravel, etc. and its contact with the Tertiary
is an angular unconformity.
The No. 2 coal seam, which is approximately 3.3 m thick on
average and is located in the Carboniferous stratum, is the main
minable seam in the WUTONG mine. As a result of faulting, the
Carboniferous and Permian strata on the east of F25 are split into
multiple terraces, whereas the strata between F25 and F2 form a
horst (Fig. 2ac). Correspondingly, the No. 2 coal seam is also cut
into terraces, resulting in an increase in the coal seam roof eleva-
tion from west to east (Figs. 2a and c) crossing F25. Within the
study domain, the elevation of the coal seam roof varies widely,
from the highest point of 200 m in the Midwest to the lowest
of 1300 m in the southeast of the coaleld. The depth of the roof
varies from 500 to 1400 m (Fig. 2b).
2.2. Mining history
From 2001 to 2007, seven mining panels in the No. 2 coal seam
were excavated in the north-central WUTONG mine, as shown in
the mined area on Fig. 2a. Each panel is 150 m wide and 500
1000 m long. The fully mechanized top-coal-caving technique
was adopted to extract the coal. In the mined area, the average
mining thickness was 3.3 m, and the mining depth varied in the
range of approximately 5001400 m. Table 2 lists the mining dates
of the seven mining panels.
The surface subsidence was monitored during mining. The max-
imum settlement measured was 1.62 m. However, because the
mined area was at least 3 km distance from the auxiliary dam,
the coal mining activity did not affect the dam at all.
Mining activity has not taken place in the region within a 3 km
distance of the dam due to concerns about surface subsidence
resulting in dam damage. However, from a dam-safety point of
view, it seems a little conservative to set the forbidden mining area
at a 3 km range. Therefore, numerical prediction of mining-induced
subsidence was performed for several selected mining areas to nd
an area that will not damage the dam while allowing the maxi-
mum possible coal extraction.
F5 F17 F25 F31
Normal Normal Normal Normal
109 93 295 145
83 78 80 79
210 30 315 85
32 14 30 15
3. FDM-based computational model
In this study, FLAC3D was utilized to conduct numerical predic-
tion of coal mining-induced rock movement. FLAC3D is a 3-D nite
difference computer program for solving rock and soil mechanics
problems.
3.1. Model domain
The rectangular region enclosed by the red lines shown in Fig. 1
was selected as the computational model domain based on the
spatial distribution of current and future areas for mining. As
shown in Fig. 1, the x-axis of the Cartesian coordinate system is
N 20E, and the z-axis points in the vertical upward direction.
The model dimensions are 10.5 km  9 km; the lower left corner
and the upper right corner have coordinates of (21100 m,
14500 m) and (31600 m, 5500 m), respectively. As the lowest
altitude of the coal seam roof is 1300 m, the simulated elevation
of the model ranges from 1550 m to 250 m. The nearest horizon-
tal distance from the mining boundary to the model boundary is
2 km.
3.2. Generalization of strata and faults
The Ordovician, Carboniferous, Permian, Tertiary and Quater-
nary strata, as well as the coal seam in the Carboniferous stratum,
are simulated in the model. The geometric shape of each stratum
interface was constructed based on sampling data. These data are
a series of spatial points located on geological interfaces and ob-
tained from geological maps, borehole logs and other resources.
The coal seam is represented by a thin layer that was 3.3 m in
thickness. The model also contains all the faults listed in Table 1.
Each fault is a fractured zone approximately 1030 m in width
and composed of mudstone, mylonitic rocks and breccia. Typically,
a fault can be simulated with interface elements using FLAC3D
when the thickness of the fault zone is small. However, when a
fault zone is more than 10 m in thickness, it is inappropriate to
simulate a fault using a single interface. Therefore, a thin layer with
a certain thickness is used to represent each fault in this paper
(Table 1). A layer for a single fault is composed of two boundary
surfaces. These surfaces are constructed based on sampling data
obtained from geological maps and borehole logs.
3.3. Computational mesh
The geological model was meshed using a set of triangular
prism elements. A triangular prism has two triangular ends and
three quadrangular sides. The triangular prism was selected in-
stead of the more typical hexahedral mesh because the elevation
of the strata in the WUTONG mine varies signicantly in magni-
tude, and the nodes in a hexahedral mesh can give rise to relatively
large displacements during the computing process. The displace-
ments may result in extremely distorted elements and may cause
computational problems. FLAC3D allows various options for shapes,
including hexahedrons, tetrahedrons, pyramids, triangular prisms,
etc. [20]. Hexahedral elements are not used in this study, although
they are normally the best choice for numerical simulation.
 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145 137

Fig. 2. (a) Elevation contours of the coal seam roof; (b) depth contours of the coal seam roof; (c) cross-section 11 (unit: m).

The detailed approach to prism meshing is as follows. First, the
model is divided into 28 layers vertically with each layer having a
thickness of 3.380 m. The closer a layer is to the coal seam roof,
the thinner the layer is set; the farther the layer is from the roof,
the thicker it is set. Then, in the horizontal direction, each layer
is meshed into 111,698 elements. The element size is made smaller
as it approaches the mining panel. Finally, a mesh is formed with a
total of 1,675,940 elements and 3,127,544 nodes. Fig. 3 shows the
computational mesh used with FLAC3D. The same gure also shows
an enlarged view of fault and surrounding rock elements around a
fault.
3.4. The constitutive model and yield criterion
The incremental elasticplastic constitutive model and Mohr
Coloumb yield criterion [20] are used in this study.
138 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145
Fig. 2. (continued)
Table 2
Mining dates of the seven mining panels located in the mined area.
Date of mining Panel01 Panel02 Panel03
Start date Jan. 1st, 2006 Nov. 1st, 2004 Oct. 1st, 2006
End date July 30th, 2007 Jan. 31st, 2006 Oct. 30th, 2007
Fig. 3. Computational mesh used with FLAC3D.
3.5. Boundary conditions and the initial state of stress
Displacement boundary conditions [20] are used in this study.
In the computational model (Fig. 1), the mesh points with coordi-
nates x = 21,100 m or x = 31600 m were not allowed to move along
the x-axis; the mesh points with coordinates y = 14,500 m or
y = 5500 m were not allowed to move along the y-axis, and the
mesh points with the coordinate z = 1550 m were not allowed
to move along the z-axis. In this study, the free surface has an irreg-
ular geometry, and the computational model is a non-uniform grid
[20]; therefore, gravity along with the lateral stress coefcient at
rest, k0, given by k0 = l/(1 l), where l is the Poissons ratio were
used to apply the in situ stresses.
4. Back analysis on mechanical parameters of the rock-masses
One of the key factors inuencing the reliability of a numerical
simulation is whether the mechanical properties of the rock
masses are estimated accurately. As stated earlier, signicant
uncertainty exists in estimating accurate values for rock mass
mechanical parameters. In this study, the rock mass mechanical
parameters are estimated in the following way:
Panel04 Panel05 Panel06 Panel07
Sep. 1st, 2002 Aug. 1st, 2005 May 1st, 2001 Jan. 1st, 2004
Nov. 19th, 2003 May 11th, 2006 June 28th, 2002 Mar. 31st, 2005
First, the values of the deformation modulus, E, Poissons ratio,
l, cohesive strength, C and the internal friction angle, / of rock
masses of different strata are estimated empirically based on the
available literature that relates intact rock and discontinuity prop-
erties to rock mass properties.
Then, the aforementioned mechanical parameter values are
determined by combining the back analysis with an orthogonal
experimental design technique (explained later) [21] and numeri-
cal simulation.
4.1. Determination of the initial values of rock mass mechanical
parameters
Rock mass is composed of intact rock and discontinuities. The
mechanical parameters of intact rock obtained by laboratory
experiments are quite different from those of rock masses [22
26]. Estimations by Mohammed et al. [23] provide the following
values for rock masses: a mean deformation modulus equal to
47% of the corresponding intact rock; a mean uniaxial compressive
strength equal to 28% of the laboratory intact rock strength, and a
mean tensile strength equal to 50% of the laboratory intact rock
tensile strength. For a granitic gneiss rock, Kulatilake et al. [24]
 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145 139

obtained a mean rock mass deformation modulus equal to approx-
imately 30% of the intact rock Youngs modulus. For Aspo diorite
rock located at a depth of 485 m at Aspo Hard Rock Laboratory,
Sweden, Kulatilake et al. [25] estimated the mean rock mass mod-
ulus to be 51% of the intact rock Youngs modulus. Additionally, the
mean rock mass Poissons ratio was estimated to be 21% higher
than the intact rock Poissons ratio. In a recent study of a limestone
rock mass, Wu and Kulatilake [26] have obtained a mean rock mass
strength equal to approximately 45% of intact rock strength. For
the same rock mass, they have obtained a mean deformation mod-
ulus equal to approximately 50% of the intact rock Youngs modu-
lus. The mean shear and bulk moduli for the same rock mass were
approximately 40% and 30% of the intact rock shear modulus and
bulk modulus, respectively. The mean Poissons ratio for the rock
mass was approximately 14% higher than the intact rock Poissons
ratio.
In this paper, the values of the deformation modulus and Pois-
sons ratio of rock masses in different strata are estimated empiri-
cally based on the aforementioned literature as well as by
comparisons with the mechanical parameter values of the coal-
elds adjacent to the WUTONG mine (Table 3). In this table, for
each rock formation, the value of Poissons ratio for rock mass is
approximately 1120% higher than that of the intact rock and
the value of the deformation modulus is approximately 4060%
of the intact rock Youngs modulus. The values of C and / of rock
masses are determined through comparisons with those collected
from other mines in the Fengfeng coaleld.
4.2. Back analysis on rock mass mechanical parameters
4.2.1. Back-analysis method
Back analysis has often been used to determine values for geo-
technical parameters [2732], especially with the development of
numerical simulation techniques that take measured displacement
as basic information. Currently, back analysis has been widely
adopted in research in combination with numerical simulation
and mathematical methods, such as orthogonal experimental de-
sign [21], uniform experimental design [21], neural networks
[33] and genetic algorithms [34]. In this paper, an approach com-
bining numerical simulation with an orthogonal experimental de-
sign [21] is used to back analyze rock mass mechanical parameters.
The detailed procedure is as follows:
(1) After analyzing the measured data for surface subsidence
within the mined area, the Maximum Surface Subsidence
Coefcient (MSSC), dened as the ratio of the maximum
surface subsidence to the thickness of the coal seam at the
corresponding position, is selected as the benchmark value
of the test indicator.
(2) Selecting the rock mass mechanical parameters as the exper-
imental design factors, a multi-factor and multi-level testing
program is formed using an orthogonal experimental design
technique, and all the testing schemes are listed in an orthog-
onal table.
(3) A series of numerical simulations are conducted using FLAC3D
with the parameter schemes listed in the orthogonal table.
(4) A range analysis is performed on the results and a relation is
established between the test indicator and each of the
experimental factors.
(5) Analyzing the results by means of range analysis, the rock
mass mechanical parameters are optimized.
4.2.2. Selection of experimental factors
The parameters E, l, C and / are selected as the experimental
factors in this study. It is observed from Table 3 that there are ve
strata and faults leading to 11 major rock formations, and the
mechanical parameter values of each formation are different from
each other. If E, l, C and u of all the rock formations are used as the
experimental factors, there are 44 factors in total. The back-analy-
sis calculation for such a system is very tedious and difcult. The
following simplied approach is used in this study:
First, the average values of E, l, C and / resulting from all the
rock formations given in Table 3 are calculated. Then, the initial
mechanical parameter value divided by the corresponding average
value is calculated for each mechanical parameter of each forma-
tion. These values are listed as kiE ; kiU , kiC and kiF , in Table 3, where
i represents the ith rock formation.
Second, in the orthogonal experiments, the average values of
the deformation modulus, Poissons ratio, cohesion and internal
 l
friction angle, E,  and /,
, C  representing all the rock formations
are taken as the experimental factors. Then, the mechanical param-
eters of the ith formation are given as ki E,  ki l  and ki /,
 , ki C, 
E l C /
respectively. It should be noted that the average values of the
mechanical parameters of the rock formations listed in Table 3
 l
are different from the experimental factors (E,  and /).
, C  The for-
mer are determined from the empirical formulae, while the latter
are variables that change with the orthogonal experimental
scheme. However, the parameter value ranges for the latter are
determined by the former.
4.2.3. Selection of the test indicator
A test indicator is a variable that assists in judging whether the
experimental results are reasonable. The indicator should satisfy
the following two conditions: rst, its value should change with
the values of the experimental factors, and second, its benchmark
value can be obtained by specic means, such as eld measure-
ments. The MSSC is used as the test indicator in this study because
it is very sensitive to changes in rock mass mechanical parameters,

Table 3
Initial values selected for mechanical parameters of the representative rock formations and faults.

No. Strata Rock formation E (GPa) l C (kPa) u () kiE kli kiC kiu

1 Q 0.06 0.36 56.8 22.6 0.011 1.558 0.069 0.773

2 T 1.95 0.35 209.8 20 0.356 1.515 0.254 0.684
3 P Shiqianfeng 6.48 0.16 963.5 34 1.185 0.693 1.166 1.162
4 Shangshihezi 6.55 0.13 985.3 34 1.197 0.563 1.193 1.162
5 Xiashihezi 6.23 0.15 969.8 34 1.139 0.649 1.174 1.162
6 Shanxi 6.46 0.18 966.9 37 1.181 0.779 1.170 1.265
7 C Taiyuan 5.28 0.21 865.3 27 0.965 0.909 1.047 0.923
8 Coal 0.53 0.31 171.6 20 0.097 1.342 0.208 0.684
9 Benxi 7.5 0.21 979.6 28 1.371 0.909 1.186 0.957
10 O 19.02 0.14 2867.6 40 3.477 0.606 3.471 1.367
11 Faults 0.1 0.34 52.3 25.2 0.018 1.472 0.063 0.861
140 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145
and its benchmark value can be determined by surface displace-
ment monitoring.
4.2.4. Surface subsidence monitoring data in the mined area
(1) Surface subsidence monitoring points
From 2001 to 2007, 7 mining panels (Fig. 4) in the central-north
area of WUTONG Mine were extracted. A total of 79 surface subsi-
dence monitoring points were set up in the mined area along the
measuring line L shown in Fig. 4 to record the values of surface
subsidence during the mining period starting 2001.
(2) Surface subsidence measurements
The engineers of the WUTONG Mine measured the elevation
and calculated the subsidence at each monitoring point every
6 months. Their measuring methodology was guided by the Code
of Measuring Practice for Coal Mine [35], which is a general stan-
dard followed with respect to measurements associated with coal
mines in China. Precise leveling instruments, with precision of
0.5 mm/km, were used to measure the elevations at the selected
points.
(3) Measured subsidence
From 2001 to 2008, the engineers had conducted subsidence
measurements about 12 times. However, because some of the data
have been lost, only ve surface subsidence proles along line L
were available for this study (Fig. 5). The data for these proles
come from ve measurements conducted on the dates of 6/18/
2003, 7/26/2004, 8/15/2006, 9/21/2007 and 8/28/2008, respec-
tively. The horizontal locations of Panel01 through Panel07 are
Fig. 4. Selected surface subsidence monitoring points along Line L in the mined
area.
shown at the bottom of Fig. 5. The ve subsidence proles are
briey explained as follows:
 Subsidence prole of June 18, 2003
Mining was completed at Panel06 in June 2002, and the surface
movement caused by Panel06 had reached a stable state. Excava-
tion at Panel04 had progressed about 60%, and its inuence on sur-
face movement had not reached a steady state. A distinct surface
settlement center appeared above Panel06 and Panel04 with a
maximum settlement of 0.42 m (Fig. 5).
 Subsidence prole of July 26, 2004
Mining was completed at Panel06 and Panel04, while about 45%
of Panel07 had been nished. Two surface settlement centers ap-
peared above the three mining panels, in which the maximum set-
tlement reached 0.82 m (Fig. 5).
 Subsidence prole of August 15, 2006
By August 15, 2006, the completed mined panels included Pa-
nel04, Panel06, Panel07, Panel05 and Panel02. The maximum set-
tlement reached 1.49 m above the mining panels Panel04,
Panel06, Panel07 and Panel05; while another center of settlement
with a maximum value of 1.05 m appeared above Panel02 (Fig. 5).
 Subsidence prole of September 21, 2007
Except for Panel01 and Panel03, mining was completed at all
other panels prior to June 2006; and the mining-caused settlement
was almost stable by September 2007. Four centers of settlement ap-
peared on the ground with a maximum settlement of 1.6 m (Fig. 5).
 Subsidence prole of August 28, 2008
By October 30, 2007, mining was completed at all panels, and
surface settlement caused by the excavation reached a stable state
by August 28, 2008 with a maximum settlement of 1.62 m (Fig. 5
and point Q in Fig. 6).
Since the mining-induced surface subsidence was basically sta-
ble by August 28, 2008, the maximum settlement on this date is
treated as the criterion for back analysis of the rock mechanical
parameters. According to the measured subsidence, within the
mined area, the measured maximum settlement is 1.62 m with
MSSC being 0.49. Therefore, the measured MSSC value of 0.49 will
be taken as the benchmark value of the test indicator of the orthog-
onal experiment.
4.2.5. The orthogonal experiment and analysis of results
4.2.5.1. Orthogonal experimental design. Four factors and ve levels
were selected to perform the orthogonal experiment. Following the
orthogonal experimental method [21], an orthogonal experimental
table was designed (Table 4). In Table 4, the second through fth
columns are the experimental factors. Table 4 lists 25 test
schemes; each row represents a level of the four factors. After
the level of each factor is set, the value of each factor can be deter-
mined according to Table 5. In this manner, the values of the four
rock mass mechanical parameters were determined.
4.2.5.2. Computational results of the orthogonal experimental
schemes. Using the values of each scheme listed in Table 4 as the
mechanical parameter values, Panel01 through Panel07 mining
was simulated numerically by one step using FLAC3D computa-
tional model described in Section 3. The last column of Table 4 lists
the MSSC-values obtained from the 25 experimental schemes.
 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145 141

Fig. 5. Measured and calculated surface subsidence proles along line L, shown in Fig. 4.

Table 4
Formed orthogonal experimental design table and results of test schemes.

Scheme 
E l 
C 
/ MSSCs

1 I I I I 0.536
2 I II II II 0.514
3 I III III III 0.493
4 I IV IV IV 0.486
5 I V V V 0.483
6 II I III II 0.510
7 II II IV III 0.391
8 II III V IV 0.340
9 II IV I V 0.486
10 II V II I 0.556
11 III I V III 0.297
12 III II I IV 0.598
13 III III II V 0.378
14 III IV III I 0.601
15 III V IV II 0.432
16 IV I II IV 0.564
17 IV II III V 0.265
18 IV III IV I 0.632
19 IV IV V II 0.294
20 IV V I III 0.661
21 V I IV V 0.242
22 V II V I 0.344
23 V III I II 0.690
24 V IV II III 0.641
25 V V III IV 0.266
Fig. 6. Calculated surface subsidence contours (at a 0.1 m interval) for the
verication example.

4.2.5.3. Range analysis performed on the orthogonal experimental

Table 5
results. The changing relation between the indicator and each of
Selected values for the ve levels of the experimental factors.
the factors is analyzed statistically according to the range analysis
method [21]. Taking the factor E  as an example, the procedure of Level Factor
range analysis is illustrated as follows:  (GPa)
E l  (kPa)
C  ()
/

I 1.85 0.27 449 19.5

 are set
 From Table 4, select the schemes in which the values of E II 4.16 0.25 680 22.4
as the value of the 1st level and calculate the average MSSC- III 6.47 0.23 911 25.3
value of these schemes. For convenience, this average value is IV 8.78 0.21 1142 28.3
V 11.09 0.19 1374 31.2
called the Average MSSC.
142 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145

 Calculate the difference between the maximum and the mini-

mum of the ve Average MSSC values. This difference is named

the range value of MSSC associated with E.

Using the above steps, the changing relation between the test
indicator and each of the experimental factors (Fig. 7) and the
range values of the MSSC of all the experimental factors were ob-
tained. Note the range values of E, l  and /
, C  are 0.066, 0.084,
0.243 and 0.163, respectively. Obviously, for the MSSC, the range
 and /
values of C  and l
 are much greater than those of E  , indicating
that changes in C and /
 have greater impacts on the MSSC than
changes in E and l .

4.2.6. Selection of the rock mass mechanical parameters

Fig. 7. Calculated average MSSC for the experimental factors.
The MSSC of the 3rd scheme is 0.493, and, of the 25 schemes, it
is the closest to that measured. Thus, the scheme 3 MSSC is se-
lected to estimate rock mass mechanical parameter values. Using
the MSSC of 0.493 in the statistical changing relations shown in
Fig. 5, the values of the four experimental factors are estimated
 = 2.4 GPa, l
as E  = 773 kPa and /
 = 0.24, C  = 22.

 Similarly, calculate the Average MSSC values when E  is set as 5. Verication test
the values of the levels II, III, IV and V, respectively. These Aver-
age MSSC values describe the relation between the indicator Another simulation within the mined area was conducted using
MSSC and E. the estimated rock mass parameter values in the same FLAC3D

Fig. 8. Vertical displacement contours of the rock mass in section I-I of Fig. 6.

Fig. 9. Boundaries selected for mining and boundaries of regions affected by mining for four scenarios. C1, C2, C3, C4 are four different boundaries of mining area of the four
scenarios; c1, c2, c3, c4 are boundaries of mining-affected regions of the four scenarios.
 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145 143

computational model to verify the reliability of the estimated rock
mass parameters.
Fig. 6 shows the surface subsidence contours produced by the
results of FLAC3D. As shown in Fig. 6, the measured maximum sur-
face settlement point Q (the green point), with a subsidence of
1.62 m, and the calculated maximum surface settlement point P
(the red point), with a value of 1.6 m, are very close to each other.
Fig. 5 is a comparative map between the measured and calcu-
lated surface subsidence curves on line L shown in Fig. 4. It can
be observed that on line L, the subsidence values of the calculation
tally closely with those measured on August 28, 2008.
Fig. 8 shows the vertical displacement contours of the rock for-
mations on section II, shown in Fig. 6. The maximum settlement
displacement of the coal seam roof is 3.3 m after mining, which
coincides with the phenomenon of the roof collapsing after mining.
The aforementioned analyses show that the rock mass parame-
ters estimated from the back analysis can be used in predicting
mining-induced subsidence in the mining area.
6. Surface subsidence prediction under different scenarios
To avoid mining-induced damages to the dam, the minimum
distance between the mining area boundary and the dam bound-
ary must be long enough. This minimum distance, as specied
by the owner of the dam, is approximately 3 km. However, the
mine owner advocates for this distance to be as short as possible
to extract the maximum possible amount of coal. To identify a
suitable mining area boundary, four mining scenarios, S1, S2,
S3 and S4, are considered in four respective mining area bound-
aries: C1, C2, C3 and C4 (Fig. 9). Using FLAC3D, the scenarios and
mining area boundaries predict surface subsidence due to min-
ing (Fig. 9).
In Fig. 9, the boundaries of the regions affected by mining for
each of the four scenarios are plotted as c1, c2, c3 and c4. In mining
scenarios S1 and S2, the values of the maximum surface subsidence
on the mining area are 2.15 m and 2.14 m, respectively. The re-
gions affected by mining of these two scenarios have covered part

Fig. 10. Contours of mining-induced surface subsidence for scenario S3.

Fig. 11. Contours of vertical displacement of rock strata along the cross-section IIII shown in Fig. 10.
144 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145
of the dam, and the values of the maximum dam subsidence are
0.06 m and 0.02 m. No dam subsidence is the precondition when
mining near the dam. Therefore, these two scenarios cannot be se-
lected. When mining with scenario S4, the maximum surface sub-
sidence on the mining area is 2.13 m, and the shortest distance
from the boundary of the mining-affected region to the dam foun-
dation is 187 m. S4 is a safe but conservative scenario. The shortest
distance between the boundary of the mining-affected area and
the dam foundation is 35 m when mining with scenario S3. There-
fore, S3 is the best of the four scenarios. The corresponding surface
subsidence results of the scenario are described below.
6.1. Surface movement results
In mining scenario S3, 40 mining panels, located in 7 regions
R1R7 (Fig. 10), are proposed in the mining area of the WUTONG
mine. Fig. 10 shows the subsidence contours resulting from FLAC3D
calculations. Among the mining regions, the maximum surface
subsidence of 2.14 m occurs within the mining region R2. The
corresponding MSSC value is 0.65; the depth of the coal seam cor-
responding to the maximum surface subsidence point is 483 m.
The closest distance from the boundary of R2 to the boundary of
the surface movement area is 770 m. The maximum surface subsi-
dence of R5 is 1.02 m, and the corresponding MSSC is 0.31. The cor-
responding coal seam roof depth is 1010 m; the closest distance
between the boundary of R5 and that of the surface movement
area is 1035 m. From Fig. 10, it can be concluded that the deeper
the coal seam roof, the smaller the MSSC, and the larger the min-
ing-induced surface movement area. Fig. 10 was used to estimate
the shortest distance from the boundary of the mining-induced
surface movement area to the dam foundation as 35 m. Thus, it
is concluded that coal extraction using this scenario will not cause
damage to the dam.
6.2. Movement results of rock strata on a typical cross-section
Fig. 11 provides the contours of the vertical displacement of
rock strata on the cross-section IIII shown in Fig. 10. On the sec-
tion, the surface settlement of point P1 is 1.69 m and that of P2
is 1.81 m, whereas the maximum settlement of the coal seam roof
is 3.3 m. P2 is located in the mined area, and the subsidence value
of P2 was 1.62 m before the coal was extracted in the mining area.
Therefore, the extraction of coal in the mining area has caused
more settlement in the mined area. Additionally, Fig. 11 shows that
the closer a rock stratum is to the coal seam roof, the larger its set-
tlement value.
7. Conclusions
Mining-induced surface subsidence predictions were performed
using FDM to nd a possible future mining scenario that would
prevent the Yuecheng Reservoir dam from being damaged due to
mining at WUTONG. The computational model included a
simplied geologic model constructed from available geologic
information for the site. Rock mass mechanical parameters for
the computational model were estimated through a back analysis
procedure that incorporated measured subsidence data, geome-
chanical parameters initially estimated through empirical and
engineering analogy methods, a statistical experimental design
technique and numerical modeling using FLAC3D.
Four possible mining scenarios with different mining area
boundaries were designed. The subsidence prediction in the min-
ing area was conducted for each of the predesigned mining sce-
narios using the estimated rock mass mechanical parameters in
the FLAC3D computational model. Scenario S3 turned out to be
the best one of the four scenarios to provide a solution to the
problem tackled in the paper. The predictions from this scenario
are as follows:
(1) Within the coaleld, the shallower the coal seam, the larger
the MSSC. The MSSC in the mining area is 0.65, and the max-
imum settlement point is located in the mid-west of the
coaleld, which is the shallowest area of the coal seam
depth, with a depth of 483 m.
(2) The nearest distance from the boundary of the surface move-
ment area to the edge of the dam foundation is 35 m.
Therefore, coal mining according to scenario 3 will not cause
dam damage.
Acknowledgments
This research was supported by the Natural Science Foundation
of China (Grant Numbers 40602037 and 40872183) and the Funda-
mental Research Funds for the Central Universities. The authors
would like to thank the editor and reviewers for their contributions
on the paper.
References
[1] Knothe S. Observations of surface movements under inuence of mining and
their theoretical interpretation. In: Proc Eur congr ground movement, Leeds;
1957. p. 2108.
[2] Zenc M. Comparison of Bals and Knothes methods of calculating surface
movements due to underground mining. Int J Rock Mech Min Sci
1969;6:15990.
[3] Kratzsch H. Mining Subsidence Engineering. New York: Springer-Verlag; 1983.
[4] Kumar B, Saxena NC, Singh B. A new hypothesis for subsidence prediction. J
Mines Met Fuels 1983;31:45965.
[5] Ren G, Reddish DJ, Whittaker BN. Mining subsidence and displacement
prediction using inuence function methods. Min Sci Technol 1987;5:89104.
[6] Ren G, Whittaker BN, Reddish DJ. Mining subsidence and displacement
prediction using inuence function methods for steep seams. Min Sci
Technol 1989;8:23552.
[7] Gonzez Nicieza C, lvarez Fernndez MI, Menndez Daz A, lvarez Vigil AE.
The new three-dimensional subsidence inuence function denoted by n-k-g.
Int J of Rock Mech Min Sci 2005;42(2005):37287.
[8] lvarez Fernndez MI, Gonzez Nicieza C, Menndez Daz A, lvarez Vigil AE.
Generalization of the nk inuence function to predict mining subsidence. Eng
Geol 2005;80(2005):136.
[9] Luo Y, Cheng JW. An inuence function method based subsidence prediction
program for longwall mining operations in inclined coal seams. Min Sci
Technol 2009;19(2009):5928.
[10] Litwiniszyn J. On certain linear and non linear strata theoretical models. In:
proc int conf strata control, New York; 1964. p. 83394.
[11] Torao J, Rodrguez R, Ramrez Oyanguren P. Probabilistic analysis of
subsidence-induced strains at the surface above steep seam mining. Int J of
Rock Mech Min Sci 2000;37(2000):11617.
[12] Astin J. A visco-elastic analysis of ground movement due to an advancing coal
face. J Eng Math 1968;2:922.
[13] Salamon MDG. The role of linear models in the estimation of surface ground
movements induced by mining tabular deposits. In: Geddes JD, editor. Proc
conf large ground movement and structures. Cardiff: Pentech Press; 1977. p.
187208.
[14] Berry DS. Progress in the analysis of ground movement due to mining. In:
Geddes JD, editor. Proc. Conf large ground movements and structures, Cardiff:
Pentech Press; 1977. p. 781810.
[15] Alejano LR, Ramrez Oyanguren P, Taboada J. FDM predictive methodology for
subsidence due to at and inclined coal seam mining. Int J Rock Mech Mining
Sci 1999;36(4):47591.
[16] Alejano LR, Alonso E. Subsidence prediction with FLAC. In: Detournay E, Hart R,
editors. FLAC and numerical modelling in geomechanics. Rotterdam: Balkema;
1999. p. 22543.
[17] Tajdus K. New method for determining the elastic parameters of rock mass
layers in the region of underground mining inuence. Int J Rock Mech Min Sci
2009;46(2009):1296305.
[18] Villegas T, Nordlund E. Numerical analysis of the hangingwall failure at the
Kiirunavaara mine. In: Shunnesson H, Nordlund E, editors. Proc MassMin
2008. Lule, Sweden: Lule University of Technology; 2008. p. 86776.
[19] Villegas T, Nordlund E. Numerical simulation of the hangingwall subsidence
using PFC2D. In: Shunnesson H, Nordlund E, editors. Proc MassMin
2008. Lule, Sweden: Lule University of Technology; 2008. p. 90716.
 N. Xu et al. / Computers and Geotechnics 48 (2013) 134145
[20] Itasca Consulting Group, Inc.. FLAC3D fast lagrangian analysis of continua in 3
dimensions, Version 4.0. (a) Users guide and (b) theory and, background;
2009.
[21] Taguchi G. Systems of experimental design. New York: Uni. Pub.; 1987.
[22] Hoek E, Brown ET. Practical estimates of rock mass strength. Int J Rock Mech
Min Sci 1997;34(8):116586.
[23] Mohammad N, Reddish DJ, Stace LR. The relation between in situ and
laboratory rock properties used in numerical modelling. Int J Rock Mech Min
Sci 1997;34(2):28997.
[24] Kulatilake PHSW, Wang S, Stephansson O. Effect of nite size joints on the
deformability of jointed rock in three dimensions. Int J Rock Mech Min Sci
Geomech Abs 1993;30(5):479501.
[25] Kulatilake PHSW, Park J, Um J. Estimation of rock mass strength and
deformability in 3-D for a 30 m cube at a depth of 485 m at sp Hard Rock
Laboratory, Sweden. Int J Geotech Geolog Eng 2004;22(3):31330.
[26] Wu Q, Kulatilake PHSW. REV and its properties on fracture system and
mechanical properties, and an orthotropic constitutive model for a jointed
rock mass in a dam site in China. Comput Geotech 2012;43:12442.
[27] Gens A, Ledesma A, Alonso E. Estimation of parameters in geotechnical back
analysis. Application to a tunnel excavation problem. Comput Geotech
1996;18(1):2946.
145
[28] Ledesma A, Gens A, Alonso E. Parameter and variance estimation in
geotechnical back analysis using prior information. Int J Numer Anal Meth
Geomech 1996;20:11941.
[29] Calvello M, Finno R. Selecting parameters to optimize in model calibration by
inverse analysis. Comput Geotech 2004;31(5):41125.
[30] Hashash Y, Jung S, Ghaboussi J. Numerical implementation of a neural network
based material model in nite element analysis. Int J Numer Meth Eng
2004;59(7):8981005.
[31] He J, Chen SH, Shahrour I. Back analysis of equivalent permeability tensor for
fractured rock masses from packer tests. Rock Mech Rock Eng
2011;44(4):4916.
[32] Miranda T, Dias D, Eclaircy Caudron S, Gomes Correia A, Costa L. Back analysis
of geomechanical parameters by optimisation of a 3-D model of an
underground structure. Tunn Undergr Sp Tech 2011;26(6):65973.
[33] Huang Y. Application of articial neural networks for prediction of aggregate
quality parameters. Int J Rock Mech Min Sci 1999;36(4):55161.
[34] Wang C, Ma GW, Zhao J, Soh CK. Identication of dynamic rock properties
using a genetic algorithm. Int J Rock Mech Min Sci 2004;41(Suppl 1):4905.
[35] The Ministry of Energy of the Peoples Republic of China. Code of measuring
practice for coal mines. Beijing: China Coal Industry Publishing House; 1989.