University
stringclasses 19
values | Text
stringlengths 458
20.7k
|
|---|---|
Virginia Tech
|
5 Stress redistribution with the change of frequency of mining-induced
seismicity
5.1 Abstract
Mining-induced seismicity in hard-rock underground mines occurred in response to stress
generation due to excavations. The pattern of seismicity rates and how it changes with stress
transfer remains unknown. Double-difference tomographic studies were used to determine the
velocity structures, which served to illustrate stress distributions within the rock masses. Seismic
events from Creighton Mine and Kidd Mine were obtained and temporally divided into a number
of different groups, each of which including 1000 seismic events. By interpreting tomograms from
every group of seismic events, stress and mining-induced seismicity were compared with velocity
structures in the mining region. Velocity anomalies were examined with the distribution of
seismicity. Stress concentrations were assessed by combining the information from the velocity
imaging and seismicity. It has been observed that crustal earthquakes have a significant Gutenberg-
Richer relationship on frequency and magnitude of seismic events. By investigating the frequency
and magnitude of mining-induced seismicity, it is recognized that mining-induced seismicity of
Creighton Mine and Kidd Mine exhibited a similar pattern on frequency and magnitude with
earthquakes.
5.2 Introduction
The Creighton Mine and Kidd Mine are the sites of significant seismicity due to the excavation
of great depths. Large magnitude events in the underground mine cause destruction in rock mass
and infrastructures. Integration of seismic P-wave velocity imaging from microseismicity is useful
for evaluating geomechanical response of the rock mass (Young, Maxwell et al. 1992). In natural
earthquakes, it is believed that seismicity rate is enhanced by stress increases and suppressed by
stress decrease (Toda, Stein et al. 2005). An earthquake is able to enhance or suppress subsequent
seismicity, depending on locations and orientations of seismicity. Static stress changes influence
the space-time pattern of seismicity before and after earthquakes (Bowman and King 2001). It has
been generally agreed that events produce regional stress perturbations (King, Stein et al. 1994).
Seismic studies of the Creighton Mine have been conducted by Malek et al. (2008). They provided
evidence that most of the seismicity occur in proximity to excavations at Creighton Mine due to
the mining induced stress fracturing (Malek, Espley et al. 2008). It was concluded that there is no
distinct relationship between geological structure and seismicity. In addition, it is not recognized
59
|
Virginia Tech
|
that seismicity is correlated with shear zones in Creighton Mine as well (Snelling, Godin et al.
2013). Spatial correlations between stress and seismicity can be assessed by velocity imaging.
Velocity imaging is a useful tool to evaluate the correlation between stress and seismicity. In
velocity imaging techniques, double-difference tomography demonstrated substantial
improvement on velocity inversion and relocating seismic events. It is generally agreed that the
recorded seismicity is more likely to be confined to the high velocity areas (Wenzel, Sperner et al.
2002). It is necessary to note that seismicity locations and stress distributions in mining are not
exactly the same as the situations in earthquakes. Case studies combining velocity images and
induced seismicity of mining excavation were shown that the induced seismicity were evidently
located in an area of velocity transition between the high-velocity (high-stress) and low-velocity
(fractured zones) (Maxwell and Young 1996).
The goal of this work is to use double-difference tomographic studies for investigating and
evaluating the stress distribution and seismic rate change. Seismicity rate changes were correlated
with the regional stress change in underground rock mass. Spatial and temporal evidences from
Creighton Mine and Kidd Mine proved that seismicity was more likely to be located in transition
areas between high stress and low stress in seismicity active periods. The other purpose of this
study is to identify the relationship between frequency of events and magnitude. Frequency-
magnitude relationships of microseismic events were assessed based on the seismicity of two deep
underground mines.
5.3 Data and methods
The double-difference tomography code tomoDD was used to image 3D velocity structure
(Zhang and Thurber 2003). This technique incorporates the travel time difference of two paired
events and the locations of events. In Creighton Mine and Kidd Mine, the microseismic system
consists of triaxial sensors and uniaxial sensors, which can provide sufficient coverage of the
mining region. Figure 5.1 shows the sensors arrangements and raypaths coverage on some
microseismic events. Double-difference tomography performed velocity inversion using the
locations of seismic events and travel time from events to sensors. Data from Creighton Mine were
recorded between April and September, 2011. The selected data set consisted of 651,436 P-arrival
times from 47,428 microseismic events. The 47,428 microseismic events used in double difference
tomographic inversion were selected using the criteria that sensors picks were more than 8.
60
|
Virginia Tech
|
Similarly, data from Kidd Mine were recorded between March, 2007 and March 2012. The
selected data included 190,568 P-arrival times from 39,160 microseismic events.
The tomoDD uses initial velocity models to perform velocity inversion. The initial velocity
model was a 3D grid of points with a uniform P-wave velocity. We arranged 40 layers on depth in
the seismic tomographic model. Each layer was subdivided into 40 × 40 grids. To include all the
raypaths of Creighton Mine, the velocity model was applied over a 3300 ft (1000 m) × 3300 ft
(1000 m) area and the range of depth of the velocity model was from 4700 ft (1433 m) to 7800 ft
(2377 m). The distance between grid points was 82.5 ft (25 m) in the Northing and Easting
directions and 77.5 ft (24 m) in the Depth direction. The velocity model of Kidd Mine was applied
over a 680 m × 760 m area. The range on depth of the velocity model for Kidd Mine was from 400
m to 1400 m. The distance between grid points was 17 m in Easting direction, 19 m in Northing
direction and 25 m in Depth direction. The initial velocity models were assigned 6250 m/s for
Creighton Mine and 6000 m/s for Kidd Mine. Velocity inversion and location of events inversion
were conducted until achieving the reliable accuracy. Velocity structures were displayed after
interpolating 3D grid of points on last iteration of P-wave velocities.
Figure 5.1. Sensors location and raypaths coverage in Creighton Mine
The seismicity rates were uneven in both Creighton Mine and Kidd Mine due to the stress
change, which was affected by excavations and production blasts. The cumulative number of
microseismic events changed with the date in Creighton Mine as shown in Figure 5.2. There was
61
|
Virginia Tech
|
an average of 250 events per day in Creighton Mine. The periods with abundant microseismicity
can be recognized by the trend of the curve. It was identified that July 6th, July 10th, and August
4th experienced high seismicity rates. In tomographic studies on Creighton Mine, microseismicity
data was divided chronologically by number of groups and every group included 1000
microseismic events. The same number of events in every group provided the close density of
raypaths. Furthermore, the groups of 1000 microseismic events guaranteed that abundant raypaths
were accessible to generate the tomograms on high resolution.
Figure 5.2. Cumulative number of microseismic events in Creighton Mine
The seismicity rate in Kidd Mine was much lower than that in Creighton Mine (Figure 5.3).
The average number of microseismic events of Kidd mine was less than 20 per day from April,
2007 to March, 2012. Similarly, the data of Kidd Mine was divided into multiple groups, each one
of which includes 1000 microseismic events. Combinations of tomograms and locations of
seismicity from different groups are shown for comparison.
62
|
Virginia Tech
|
Creighton Mine
The seismicity rate changed significantly during the time intervals between March, 2011 and
September, 2011. It achieved high levels on July 6th and August 6th. The cross-section of
Creighton Mine shows the location of the series of tomograms (Figure 5.6). Seismicity distribution
and tomograms of level 7700 are shown in Figure 5.7 and Figure 5.8. Each tomogram was
generated based on 1000 events in the same group, which consists of microseismic events that
occurred within different time intervals. It is illustrated that microseismic events mainly
concentrated on the area with a center at Easting 4500 ft and Northing 6200 ft from June 30th to
July 11th. Some microseismic events scattered outside the drift area. It is noted that microseismic
events were more concentrated when seismicity rate increased. The peak seismicity rate was
achieved (193 events per hour) between 04:43 and 09:54, July 6th events (Figure 5.7. C). Also, the
seismicity rate stayed high (101 events per hour) during the next events group from 09:54 to 19:48,
July 6th (Figure 5.7. D). Meanwhile, the velocity images show that high velocity occupied the
studied areas when they reached the large seismicity rates. It is inferred that the maximum principle
stress strikes from NW to SE from the velocity images. Furthermore, stress increased when the
seismicity rate of the rock mass increased. The location of event groups failed to move with the
high stress region changing. Also, induced seismicity was not located in the high stress region, but
was more likely to concentrate on the area between the high stress and low stress zones. With the
decrease of seismicity rate after 19:49, 6th, July (Figure 5.7.E), the high velocity area started
declining until the seismicity rate switched to recover on 10th, July (Figure 5.7.G).
Table 5.1. Seismicity rate of July in Creighton Mine
Events Group Events Number Start End Hours Seismicity Rate (per hour)
C1 1000 6/30/2011 22:51 7/3/2011 3:20 52.97 18.88
C2 1000 7/3/2011 3:24 7/6/2011 4:43 73.32 13.64
C3 1000 7/6/2011 4:43 7/6/2011 9:54 5.18 192.93
C4 1000 7/6/2011 9:54 7/6/2011 19:48 9.90 101.01
C5 1000 7/6/2011 19:49 7/8/2011 19:57 48.12 20.78
C6 1000 7/8/2011 19:57 7/10/2011 3:38 31.68 31.56
C7 1000 7/10/2011 3:39 7/11/2011 4:41 25.03 39.95
C8 1000 7/11/2011 4:41 7/14/2011 3:42 71.02 14.08
65
|
Virginia Tech
|
E. Event Group C5 F. Event Group C6
G. Event Group C7 H. Event Group C8
Figure 5.7. Velocity on level and seismicity distribution in July of Creighton Mine
To evaluate whether the stress increases with larger seismicity rate, the other period with active
seismicity was also examined. The other active seismicity with high seismicity rate appeared on
4th, August in Creighton Mine. The tomograms and microseismicity distribution around the
seismic active periods were analyzed. The seismicity rate increased significantly and reached 291
events per hour on April, 4th, 2011. The tomograms of the same layer apparently reflect the velocity
change. In particular, it indicates that higher velocity occupied the region when the seismicity rate
was at its maximum (Figure 5.8.B). Events of this period were located on the intermediate velocity
area between the two high velocity areas. With the lowering of the seismicity rate, low-velocity
areas started to increase. The low stress is associated with the low-velocity area. After 14th, August,
low-velocity (< 5.7 km/s) areas occupied most regions of the tomographic studied area.
67
|
Virginia Tech
|
Figure 5.9 shows the locations of tomograms to be observed in Kidd Mine. Stress distribution
can be inferred through velocity distribution of the tomograms and locations of microseismicity is
indicated on tomograms. Furthermore, tomograms calculated from different events groups can be
compared to estimate the stress transfer as time periods changed. According to the seismicity rates
(Table 5.3), Event Group K8 (Figure 5.8.D), Event Group K9 (Figure 5.8.E) and Event Group K12
(Figure 5.8.H) experienced the highest seismicity rates.
Figure 5.9. Cross section of Kidd Mine
Tomograms calculated from Events Group K5 (Figure 5.10.A) to Events Group K14 (Figure
5.10.J) following the time sequence are shown in Figure 5.10. The tomograms are illustrated with
the microseismic events close to the level that tomograms are located on. From Events Group K5
(Figure 5.10.A) to Events Group K8 (Figure 5.10.D), the low velocity region diminished with the
raising of the seismicity rate. Although it is generally recognized that seismic events are more
likely to occur within a high stress region, microseismic events in this study close to the level of
tomogram are located in the vicinity of low velocity regions, and there are a few events located
between the borders of high velocity regions and low velocity regions. The largest change of
velocity distribution is observed from Events Group K8 (Figure 5.10.D) to Events Group K9
(Figure 5.10.E). The rise of high velocity regions is greatest. The areas adjacent to the entry of
NW-SE drift were enhanced. It is shown that the group of events within selected ranges of depth
70
|
Virginia Tech
|
was mainly located in areas of low and moderate velocity rather than high velocity (Figure 5.10).
However, the areas of high velocity grew with the increase of the total seismicity rate. It is inferred
that the increase of seismicity rate was associated with enhanced stress, implying that high stress
promoted the occurrence of these seismic events. In studies of natural earthquakes, it is recognized
that small events enhanced stress at the epicenters of them (Stein, King et al. 1992). Result of this
study agrees with the recognition in natural earthquakes that the small events increased stress.
However, whether the majority of seismicity occurred in regions where the stress had increased
needs further investigations. In addition to the change of the high velocity region, the low velocity
region displayed on NE of the drift reduced with the growth of the seismicity rate from Event
Group K5 to Event Group K8. Although the seismicity rate of Events Group 9 is lower than that
of Events Group 8, both tomograms (Figure 5.10. D and E) suggest that stress concentration
significantly enhanced along the NW-SE axis of the drift. Stress kept relieving until the time period
of Events Group K12 (Figure 5.10.H). The last high seismicity rate appeared over Events Group
K12. As a result of stress reduction before the time of Events Group K12, high velocity regions on
the center of drift decreased and low velocity regions developed to the east of the drift. Compared
with other Events Groups, the low velocity regions initially displayed in other event groups are
replaced by moderate velocity regions. For the event groups occurring after Events Group K12,
stress relieved on the east outside of the drift and low velocity regions recovered with the drop of
seismicity rate.
A. Event Group K5 B. Event Group K6
71
|
Virginia Tech
|
I. Events Group K13 J. Events Group K14
Figure 5.10. Tomograms of Kidd Mine
5.5 Summary and discussion
Mining-induced seismicity rate has shown to be correlated with stress changes on horizontal
cross sections. It was observed that high velocity region expands with the growth of the seismicity
rate. These findings confirm that the tomographic survey is a useful tool to investigate the
relationship between stress distributions and seismic rate changes in underground mines.
Measurements on the stress distribution with seismic rate changes provide a continuous estimation
on seismic risks since large magnitude events are not triggered in some circumstances even if there
is a high seismic rate in underground mines.
It was reported that a mining-induced seismicity sequence appeared on a high seismic velocity
region (Young and Maxwell 1992). However, this study’s findings show that microseismic events
are more likely to occur close to low velocity regions and the areas between low velocity regions
and high velocity regions. Although the location of the microseismicity groups is not the same as
the high velocity region, the stress accumulation still contributes to the increasing density of
seismicity groupings. It can be attributed to the fact that stress relaxes after the energy dissipation,
which is associated with the opening and closing of microcracks in the process of forming mining-
induced seismicity. Microseismic events are more widely distributed when the seismicity rate is
low. Conversely, microseismic events are more spatially concentrated during the time interval of
high seismicity rate.
73
|
Virginia Tech
|
6 Passive tomographic study on velocity changes in underground mines
6.1 Abstract
Double difference tomographic inversion on measurements of travel time and location are
performed to analyze the velocity structure within rock mass in underground mining. Residuals of
each iteration are estimated to evaluate the conversion of computation. Average wave propagation
speeds in tabulation areas are assessed to compare the velocity change affected by large magnitude
events. It is summarized that velocity increases before the occurrence of large magnitude events
and then reduces with temporal evolution after them. Possible explanations include static stress
buildup enhances the waveform propagation before the large magnitude events and static stress
reduction weakens the waveform propagation. Moreover, waveform propagation is attenuated by
the dynamic-shaking induced fractures and ruptures within rock masses. Velocity change is shown
to be of importance in assessing the stress redistribution and stability of rock masses.
6.2 Introduction
Passive tomographic model is widely used to characterize the structure of earth crust by
minimizing difference between simulated and observed seismic waveforms (Korenaga, Holbrook
et al. 2000, Tape, Liu et al. 2010). It is estimated that a velocity ranges from 6.0 to 7.0 km/s
traveling through the continental crust without deformation (Korenaga, Holbrook et al. 2000).
Iterative inversions on both locations of seismic events and the velocity structure are performed
till achieving the minimal error (Zhang and Thurber 2003). Analyses of velocity change associated
with earthquakes provide evidence that significant velocity tends to decrease after earthquakes
(Schaff and Beroza 2004). It is revealed that P and S wave velocities decrease with damaged rock
in the earthquakes and velocity recovered due to the healing effect with time (Li, Vidale et al.
2003, Li, Chen et al. 2006). Seismic imaging and microseismic monitoring are used to detect
highly stressed and failed regions of underground mines, especially in hard rock mines (Young
and Maxwell 1992). In situ stress redistribution influenced by mining excavation can be analyzed
by velocity anomalies, which are displayed by tomography. Studies suggested that reliable
microseismic monitoring systems are playing a key role for safety in mining (Urbancic and Trifu
2000); however, the knowledge of how to predict rock bursts by using microseismic events is still
not enough. It is found that velocity structure of rock masses can reveal the fracture-induced
anisotropy and burst-prone regions in mining (Young and Maxwell 1992, Wuestefeld, Kendall et
77
|
Virginia Tech
|
al. 2011). The aim of this paper is to investigate the velocity change affected by the occurrence of
seismic events with large moment magnitude (M > 1) in underground mining. A three-
n
dimensional tomographic model of the mining region is established to display the velocity change
due to the occurrence of large magnitude events.
6.3 Large magnitude events
In the underground mines, sensors of microseismic monitoring systems are installed at a great
depth to ensure accurate monitoring of microseismic events. Referring to the record of seismic
events in the data set, a considerable amount of seismic events occurred from April, 2011 to March,
2012. It exhibits that the seismic events at the depth of 7000 – 8000 ft take up over 90% of all the
seismic events (Figure 6.1). Creighton Mine experienced four large magnitude events in July, 2011
(Table 6.1). The location of microseismic events and large magnitude events provided the
reference to the dimension and spatial location for the velocity model.
Figure 6.1. Depth range of seismic events in Creighton Mine
Velocity distribution prior to large magnitude events and redistribution after them are
compared to discuss the velocity change affected by large magnitude events. The seismic network
of Creighton Mine provides a good quality seismic dataset for high resolution tomographic
inversion. Velocity profile within a certain spatial and temporal scale is accomplished based on
waveform propagation from seismicity to receivers. The period from July 6th to July 10th 2011 is
,
emphasized because the knowledge on whether some pronounced changes exhibit around the
occurrences of four major events would benefit seismic hazard assessment in underground mines.
78
|
Virginia Tech
|
Table 6.1. Times, Locations, and Magnitude of Major Events in Creighton Mine
Events Time North (ft) East (ft) Depth (ft) Magnitude
1 7/6/2011 8:41 6321 4589 7654 3.1
2 7/6/2011 8:46 6284 4850 7653 1.2
3 7/6/2011 8:47 6186 4782 7495 1.3
4 7/10/2011 2:44 6106 4540 7843 1.4
Seismic events were picked and compiled to groups of event-receiver pairs based on whether
they occurred prior to or after the large magnitude events. Tomographic results of them are
compared to evaluate the change between velocity distributions before and after the large
magnitude events. Each group includes 2000 events. Events in all groups are arranged sequentially
in time. A double difference tomography method is used to invert simultaneously the location of
seismicity and distribution of velocity. The comprehensive description of the tomographic method
and software package manual is given by Zhang (Zhang and Thurber 2003).
Table 6.2. Events Grouped for tomographic studies in Creighton Mine
Group From To Number of Events
First group before large magnitude events 6/29/2011 6:23:17 7/1/2011 16:58:54 2000
Second group before large magnitude events 7/1/2011 17:18:11 7/6/2011 8:40:32 2000
First group after large magnitude events 7/10/2011 2:44:58 7/12/2011 2:58:59 2000
Second group after large magnitude events 7/12/2011 21:08:51 7/19/2011 16:11:40 2000
Unlike all large magnitude events of Creighton Mine occurring within a short period range,
the two large magnitude events in Kidd Mine are in January and June. As a result of two different
periods including large magnitude events, 2000 events are split into four groups. As shown in
Table 6.2 and Table 6.4, each group of Creighton Mine consists of 2000 microseismic events,
while 500 events are included in each group of Kidd Mine. Two main factors are considered to
determine the amount of events in each group. First, the events could provide enough raypaths
traveling through the target areas. Then, periods based on the choice on the number of events
should be in reasonable ranges of several days in this tomographic study. The seismic rate of
Creighton Mine is higher than that of Kidd Mine, thus more microseismic events of Creighton
Mine and less microseismic events are included for the period range of several days. Monitoring
of seismicity at Kidd Mine shows that two major events occurred respectively in January and June,
2009 (Table 6.3). The number of events distribution on corresponding depth is shown in Figure
6.2.
79
|
Virginia Tech
|
Table 6.3. Times, Locations, and Magnitude of Major Events in Kidd Mine
Events Date Time North (m) East (m) Depth (m) Magnitude
1 January, 6th 2009 4:40 AM 65733 65686 1150 3.8
2 June, 15th 2009 7:01 PM 65861 65737 1035 3.1
Table 6.4. Events Grouped for tomographic studies in Kidd Mine
Events Groups From To Number of Events
First group before large magnitude event 1 10/07/2008 19:20:20 11/20/2008 04:23:44 500
Second group before large magnitude event 1 11/20/2008 05:56:53 01/06/2009 04:07:31 500
First group after large magnitude event 1 01/06/2009 04:57:40 02/05/2009 09:06:37 500
Second group after large magnitude event 1 02/05/2009 15:06:36 03/21/2009 03:50:20 500
First group before large magnitude event 2 05/10/2009 05:58:16 05/28/2009 16:44:05 500
Second group before large magnitude event 2 05/28/2009 21:18:27 06/15/2009 17:40:13 500
First group after large magnitude event 2 06/15/2009 19:19:49 08/10/2009 06:46:51 500
Second group after large magnitude event 2 08/10/2009 06:57:17 08/24/2009 16:53:46 500
Figure 6.2. Depth range of seismic events in Kidd Mine
6.4 Average velocity analysis in tabulation areas
To analyze the influence on velocity posted by large magnitude events, velocity results
computed by double difference tomography are visualized in tabulation areas. TomoDD is used to
perform velocity inversion on 3D nodes, which consists of 40×40×40 nodes (Figure 6.3). All nodes
are assigned with a same initial velocity value (Creighton Mine 6000 m/s; Kidd Mine 6025 m/s).
The initial velocities are estimated by linear fit on travel time and travel distance of raypaths
traveling through the rock mass.
80
|
Virginia Tech
|
Figure 6.3. Mesh grids of velocity model
The accuracy of tomographic studies for seismic events is determined by the density of
raypaths traveling through the geometry of velocity model. Travel time measurement and locations
of seismic events are combined using double difference tomographic inversion to compute updated
velocity models and relocations of seismic events at each iteration by LSQR algorithm (Zhang and
Thurber 2003). The conversion of the velocity models are validated by quantifying the residual of
each iteration in the inversion. It is indicated that the residuals of travel time estimation keep
decreasing with more iterations and converge after a certain amount of iterations for tomographical
studies of both Creighton and Kidd Mine (Figure 6.4). More accurate velocity distributions are
generated interacting with ultimate relocations of seismic events.
(a) (b)
Figure 6.4. Residuals of Creighton Mine (a) and Kidd Mine (b)
Creighton Mine
Velocity inversion is performed at each node. All nodes are divided by 5×5×5 mesh grids, as
shown in Figure 6.3. Average velocity of each unit cube is computed by all the velocity of nodes
81
|
Virginia Tech
|
in the same unit cube. Results of velocity distribution for mesh grids are exhibited in Figure 6.5
and Figure 6.6. Events 1, 2 and 4 are located in cubes on the level that ranges from 7546 to 7940
ft (Figure 6.5). Event 3 is located in cubes on the level ranges from 7152 to 7546 ft (Figure 6.6).
The most conspicuous feature is that significant velocity changes around the large magnitude
events.
It is observed in Figure 6.5 that the velocities within the objective region surrounding Event 1,
Event 2 and Event 4 are higher than the background velocity for all the results from all event
groups. High velocity anomalies are identified around Events 1, 2, and 4 before the large
magnitude events periods and the velocity increases significantly with the closer period to the
occurrence of large magnitude events (Figure 6.5a. and Figure 6.5b.). However, velocity trend
changes along with the occurrence of large magnitude events. The velocity of the tabulation area
especially including high velocities with events 1, 2 and 4 experienced a pronounced drop of
velocity in the postseismic periods (Figure 6.5b. and Figure 6.5c.). Eventually, the area with events
1, 2 and 4 continues to decrease to the level of background velocity. In addition to the fact that
the region with events 1, 2, and 4 indicates the most striking velocity change, increasing before
the seismic periods and decreasing in the coseismic and postseismic periods, other adjacent areas
experienced similar velocity changes and reaches low velocities as well. Similarly, a significant
velocity rising is manifested in the vicinity of Event 3 before the period ranges of large moment
magnitude (Figure 6.6a. and Figure 6.6b.). The region including Event 3 is identified by reductions
in average wave speed comparing the velocity distribution before the large magnitude events with
that after the large magnitude events, forming a slow velocity anomaly (Figure 6.6b. and Figure
6.6c.). Two possible causes, including static stresses concentration and shaking damage, could be
responsible for the velocity change. Static stress change can explain why the velocity anomalies
are located nearby the large magnitude events. The areas with static stress concentration are likely
to cause the large magnitude events. The high velocity areas before large magnitude events suggest
that a rock mass might be compacted due to the load force. A rock mass with a greater density is
easier for waveform propagation. The most likely explanation for the vicinity of large magnitude
events appearing close to high velocity anomalies is that stress concentration enhances density of
the rock mass. Bulk modulus of a rock mass is capable of increasing with volume decrease under
the pressure. Volumetric hardening is likely to be invoked if the isotropic pressure causes
irreversible volumetric compaction. Production blast activities are performed before the
82
|
Virginia Tech
|
occurrence of large magnitude events. It is inferred that production blast activities cause static
stress change. As a result of static stress change, the stress field fails to keep the originally balanced
state. Static stress change leads to uneven distribution in some regions. Abrupt energy release
might be triggered to form the source of large magnitude events.
a. First group before large magnitude b. Second group before large magnitude
e v e n t s (Layer with events 1, 2 and 4) events (Layer with events 1, 2 and 4)
c. First group after large magnitude d. Second group after large magnitude
events (Layer with events 1, 2 and 4) events (Layer with events 1, 2 and 4)
Figure 6.5. Average velocity of cubes by velocity inversion of Creighton Mine
After the large magnitude events, velocities in the vicinity of large magnitude events
experienced a significant reduction associated with decreasing in varying extent of the other areas.
A reasonable explanation is that the shaking effect by the large magnitude events leads to ruptures
and damage in nearby regions of rock mass. Opening of fractures by the shaking-induced damage
impedes the propagation of waveform in rock mass. Shaking-induced damage is inversely
correlated with the distance between its location and the hypocenter of large magnitude events.
83
|
Virginia Tech
|
This explains why the velocity drop is more evident on the regions that are closer to the location
of large magnitude events. The velocity around Event 3 changed more evidently than the region
of the large magnitude Event 1, 2, and 4 during the whole process. The region including Event 3
especially achieves a low velocity level right after the large magnitude events in postseismic
period.
a. First group before large magnitude b. Second group before large magnitude
events (Layer with event 3) events (Layer with event 3)
c . F i r s t g r o u p after large magnitude d. Second group after large magnitude
events (Layer with event 3) events (Layer with event 3)
Figure 6.6. Average velocity of cubes by velocity inversion of Creighton Mine
KIDD MINE
According to the large magnitude events of Kidd Mine, events group before and after the first
large magnitude event (Event 1) and the second major event (Event 2) are analyzed, respectively.
84
|
Virginia Tech
|
Analyses on microseismic events triggered before and after the seismic period in Kidd Mine are
conducted to further illustrate the velocity change associated with the large magnitude events.
There is a strong similarity of velocity change pattern between Creighton Mine and Kidd Mine. It
demonstrates that velocity around Event 1 grows and forms a high velocity anomaly in the region
before its occurrence (Figure 6.7a. and Figure 6.7b.). It is inferred that the stress in the vicinity of
major events increases to a higher level and the waveform propagation of seismic events is
strengthened before Event 1. The average velocity in the tabulation region of Event 1 also
experiences a significant drop after Event 1 (Figure 6.7b. and Figure 6.7c.). The average velocity
of the region of Event 2 keeps higher than the background velocity (6.3 km/s) all the time. There
is not a pronounced velocity change in the region of Event 2. However, the total high velocity
areas expand before the seismic period (Figure 6.8a. and Figure 6.8b.) and then they diminish
significantly in the postseismic periods (Figure 6.8c. and Figure 6.8d.). It might be because the
static stress is released and the propagation of dynamic shaking-induced fractures impair the
waveform propagation after the occurrence of either Event 1 or Event 2.
The most noticeable difference between the velocity change with Event 1 and Event 2 is that
the velocity change with Event 1 is strongly concentrated and intense in comparison of the region
of Event 2. The velocity changes with Event 2 are smoother and spread over larger areas. The
reasonable explanation is that the magnitude of Event 1 is greater than the magnitude of Event 2.
The hypothesis that the dynamic shaking-induced effect from Event 1 is stronger than that from
Event 2 is supported by the fact that the seismic energy of Event 1 is larger than Event 2.
b. Second group before large magnitude
a. First group before large magnitude
event (Layer with event 1)
event (Layer with event 1)
85
|
Virginia Tech
|
c. First group after large magnitude event d. Second group after large magnitude
(Layer with event 2) event (Layer with event 2)
Figure 6.8. Average velocity (Event 2) of 500 microseismic events of Kidd Mine
6.5 Summary and discussion
It is exploited that the velocity of waveform propagation experienced changes within rock
masses caused by large magnitude events in underground mining. It is useful to have a simplified
model that allows prediction of seismic risks in terms of large magnitude events. Data sets from
two hard rock underground mines are recorded and investigated. High seismic rates of each mine
provide a good raypath coverage for objective regions. Groupings, comprised of one thousand
events each, underwent velocity inversion in a double difference tomographic model to produce
the velocity structures. In the comparison of velocity distribution of multiple periods close to large
magnitude events, some findings are summarized.
The waveform propagation is enhanced in the large magnitude events prone regions before
their occurrence. It is inferred that high velocity anomalies are caused by the accumulation of static
stress in the regions including potential large magnitude events of the rock mass.
Velocity reduction after the large magnitude events in the vicinity of them implies that static
stress drop and dynamic-shaking induced fractures mutually lead to the weakness of waveform
propagation. The rate and magnitude of velocity change seems related to the depth of occurrence
and magnitude of events.
Double difference tomographic studies benefit the development of seismic hazard assessment
in underground mining. The results presented indicate that velocity anomalies within rock mass in
87
|
Virginia Tech
|
7 Statistical analyses of mining-induced seismicity from deep hard-rock
mines
7.1 Abstract
Previous studies have implicated changes of b-value associated with the occurrence of
mainshocks. However, whether changes of b-value can be used as a precursory for mainshocks
remains largely unknown. We compute the temporal changes of b-value based on a reliable
estimation of magnitude of completeness using mining-induced seismicity from a deep hard-rock
mine. The b-value analysis reveals a significant decrease of b-value with the occurrence of
mainshocks. To investigate behavior of aftershocks in mining-induced seismicity sequence, we
used a temporal decay model based on Generalized Omori’s law to examine temporal decay of
aftershock sequences. Results of temporal decay model indicated a close agreement between the
modeled temporal decay process and practical cumulative number of events with time. The
computed parameters conform to the empirical studies from crustal earthquakes, validating
effectiveness for mining-induced seismicity. Taken together, these results have important
implications for seismic hazard analyses in underground mines.
7.2 Introduction
Using mining-induced seismicity data to detect potential danger and mitigate hazard is a long-
term quest in mining safety research. Seismic monitoring system is a feasible and practical way to
monitor and record the seismic activities. Data describing the triggered time, locations and
magnitude of mining-induced seismicity contain the spatial and temporal information on the
occurrence of seismic events through periods of time. Whereas the average occurrence of rate of
seismicity is an important estimate of the potential dangers in underground mining, the average
occurrence of rate is insufficient to define the secular rate of seismicity. Seismologists have
devoted a significant effort on seismic hazard analysis by applying statistical scaling methods in
mainshock and aftershock sequence. Further, frameworks from crustal earthquake have been
proved that they can be used in mining-induced seismicity for improving safety (Young and
Maxwell 1992, Wuestefeld, Kendall et al. 2011).
Seismic hazard analyses of crustal earthquakes are mainly based on earthquake frequency
statistics on historical catalogues of seismicity. It is well known that aftershocks of a mainshock
satisfy Gutenberg-Richter frequency-magnitude scaling (Gutenberg and Richter 1956). Numerous
91
|
Virginia Tech
|
global and regional surveys have been performed to assess the empirical constant in validating the
Gutenberg-Richter law (Wesnousky 1999) (Shcherbakov, Turcotte et al. 2005). Isacks and Oliver
(1964) claimed that it is reasonable to use the hypothesis of constant b value to predict the
earthquakes by extrapolating to higher magnitudes based on frequency magnitude relations. The
b-value provides a new insight that can be used as an indicator of failures in a rock mass in the
laboratory and to forecast occurrences of earthquakes (Mogi 1963) (Smith 1981) (Lockner 1993).
Observations on b-values before large earthquakes in New Zealand, California and Venezuela
indicate that the mainshocks were preceded by periods of high b-values (Smith 1981). Evidences
support the view that the seismicity and rock deformation are shown to be closely related in space
and in time (McGarr and Green 1975). Merged with underground observations, seismic hazard
analyses on seismicity data indicate that seismicity in mining (mine tremors) obey the same
magnitude-frequency relation as crustal earthquakes (Boettcher, McGarr et al. 2009). Magnitude-
frequency data for events over 1 years at Harmony Gold Mine agree Gutenberg-Richter relation
very well (McGarr 1971). Accordingly, it is inferred that b-values of mining-induced seismicity
can be useful for seismic hazard assessment for underground mines.
Although b-values could be the determinants of seismic hazard analyses in different scales,
restraints and uncertainties involved in seismicity forecasting on historical data from different sites
can yield imprecise results. The spatial and temporal distribution of aftershocks, and the
dependence on the magnitude of the mainshock were assessed to provide reference for potential
danger and mitigating hazards (Krinitzsky 1993). Quantifying mainshocks and aftershocks can
improve the knowledge of correlations of seismic activities (Shaw 1993). The time dependence of
earthquake aftershocks is described in Generalized Omori’s law, which empirically gives the
temporal decay in the rate of aftershock occurrence (Shcherbakov, Turcotte et al. 2005).
Shcherbakov (2005) applied a scaling method on multiple aftershock sequences to scale the
parameters of Generalized Omori’s law. Based on two physical ingredients: rupture activation and
stress transfer, it is measured that seismic decay rates linearly increase with the magnitude of the
mainshock (Ouillon and Sornette 2005). Further, numerous studies are devoted to incorporating
the understanding of Generalized Omori’s law into seismic hazard analysis on mining-induced
seismicity. Vallejos interpreted a link between the productivity of seismicity and decay time of
seismicity to ensure the safe event rate for re-entry protocol in underground mines (Vallejos and
McKinnon 2011). Based on the Generalized Omori’s law, the aftershocks sequences from different
92
|
Virginia Tech
|
mine sites were assessed by a statistical analysis to establish the optimal re-entry protocol (Vallejos
and McKinnon 2010).
The purpose of this article is to present the change of b-value associated with mining-induced
seismicity sequences, to describe the relationships between b-values and magnitude of aftershocks,
and to establish time decay model of mining-induced seismicity from two mines. It has been
investigated and found that mining-induced seismicity from two mine sites agree the law of
Gutenberg-Richter frequency-magnitude very well. Then, average b-values are computed using
the seismicity that occurred in the same time length. The maximum likelihood method and least
square method are used to perform the regression on data of frequency-magnitude, respectively.
According to empirical studies on mining-induced seismicity, the decay rate of aftershocks is a
function of time as well. Statistical analyses are involved to model the predicted seismicity rate
since Generalized Omori’s law is validated on the basis of empirical investigation. Specifically,
secular rates of seismicity defined by the model can be used to assess long-term mining-induced
seismic hazards.
7.3 Data of mining-induced seismicity and analysis procedure
It is well known that events are triggered associated with the nucleation of microcracks in
rocks. Cracks are generated when the local stress exceeds the local strength (Kranz 1983). Mining
works can affect the stress regime. As a result of mine excavations, the rock mass in the proximities
of excavations loses its balance state of stress and stress concentration is generated in it. Rock
failures in the proximities of maximum stress concentration are usually found in the edges of
mining excavations. Associated with the occurrence of seismicity, fracture planes are formed
parallel to the stope faces (Cook 1976). Seismic events in mines mainly include fault-slip events
and strain-burst events. Whereas a fault stays in balance without change of load, it can be disturbed
by a mine’s excavation. Nucleation can be initiated when the stress is large enough. The length of
cracks and stress increase slowly during the first regime and then increase faster to achieve the
critical stress. A seismic event is triggered preceding the rupture development of the fault.
Specifically, the stress drop arises on the fracture plane associated with the seismic event, which
redistributes stress to vicinities (Shaw 1993).
Several mainshocks and aftershocks sequences recorded in the seismic networks of Creighton
Mine and Kidd Mine are compiled for this study. The data in Table 7.1 and Table 7.2 serve to
93
|
Virginia Tech
|
illustrate the time, location, and magnitude of mainshocks from Creighton Mine and Kidd Mine,
respectively. The moment magnitude of aftershocks is approximately from -2.5 to 0. The data
presented in this article was obtained from microseismic monitoring system and strong ground
motion system. Microseismic monitoring system is specifically for detecting the microseismic
activities, most of which are with moment magnitude less than zero. Strong ground motion system
is specially emphasized on recording mainshocks, there were approximately 40000 microseismic
events from April to September, 2011, in the Creighton Mine data sets, which includes source
parameters such as moment magnitude, focal mechanism, and ratio of E /E . Similar to the
S p
Creighton Mine data set, reliable information of seismicity is provided in the Kidd Mine data set.
The Kidd Mine data set cover seismicity from September, 2007 to March, 2012. The first
mainshock was a fault slip burst, but the second mainshock was a strain burst in Creighton Mine.
Both of the mainshocks in Kidd Mine were fault slip bursts.
Table 7.1. Times, Locations, and Magnitude of Mainshocks in Creighton Mine
Mainshocks Date Time North (m) East (m) Depth (m) Magnitude
1 July, 6th 2011 8:41 AM 1927 1399 2332 3.1
2 July, 10th 2011 2:44 AM 1853 1385 2392 1.4
Table 7.2. Times, Locations, and Magnitude of Mainshocks in Kidd Mine
Mainshocks Date Time North (m) East (m) Depth (m) Magnitude
1 January, 6th 2009 4:40 AM 65733 65686 1150 3.8
2 June, 15th 2009 7:01 PM 65861 65737 1035 3.1
An important empirical observation in seismology is the proportional relationship between the
magnitude M and cumulative number of seismic events with magnitude larger than M. In order to
check the existence of similar patterns in mining-induced seismicity sequences, mainshocks are
defined according to their magnitude. Then, aftershocks temporally following the mainshocks are
determined. The validity of the Gutenberg-Ritcher relationship of mining-induced seismicity in
both Creighton Mine and Kidd Mine is shown in Figure 7.1. The cumulative number of aftershocks
is given as functions of magnitude. Due to the fact that the average rate of seismicity in Creighton
Mine is approximately 60 times higher than that in Kidd Mine, different time scales are used for
these two mines. For Creighton Mine, seismicity within a time period of T = 3.5 days after the
mainshock are considered for the b-value measurements. For Kidd Mine, seismicity within a time
period of T = 90 days after the mainshock are performed. Note that b-value measurement by using
the least-square fit regressions are set to data with magnitude between -1.5 and 0. The cutoff
magnitude level of -1.5 is determined because only the data sets with magnitude above -1.5 are
94
|
Virginia Tech
|
strongly correlated and directly related to the aftershocks. The process of selecting magnitude of
completeness is discussed below.
(a) (b)
Figure 7.1. Cumulative numbers of aftershocks are given as functions of magnitude for (a) the first
mainshock in Creighton Mine and (b) the first mainshock in Kidd Mine. The solid straight line are the
best fit of Gutenberg-Ritcher relation.
The cutoff magnitude of aftershocks in these two mines is assessed by analyzing the magnitude
of completeness, which is the lowest magnitude at which 100% of the events are detected in
temporal and spatial scales (Woessner and Wiemer 2005). The estimate method of the magnitude
of completeness presented by Woessner (2005) is based on the self-similarity of seismicity
process, which implies a power-law distribution of seismicity. According to the implication of
power-law distribution of seismicity, the estimation method from Woessner (2005) considers the
estimate of magnitude of complexness and its influence on the b-value. The Monte Carlo
approximation of the bootstrap method is used in this method for calculating b-values and the
magnitude of completeness. In order to check the reliability of data regressions, maximum
likelihood estimation method is used to compare the result of it with the result obtained from the
least-square estimation method. Differences are found between the results of the least-square
estimation and maximum likelihood estimation method. As discussed in the results section below,
it is observed that the maximum likelihood method amplifies the variation of b-value, but the
results of both methods conform to the similar trend of variation. Figure 7.2 and Figure 7.3
illustrates the results of the magnitude of completeness and b-value by both the least-square
method and maximum likelihood method for aftershocks of Creighton Mine and Kidd Mine. It is
verified that -1.5 is the optimum magnitude of completeness for both Creighton Mine and Kidd
Mine (Figure 7.2 and Figure 7.3). First, it is well shown that better linearity is embodied at m > -
95
|
Virginia Tech
|
1.5 between cumulative number of aftershocks and magnitude of each aftershock sequence.
Second, b-values given by these two methods at m = -1.5 are conspicuously in closer agreement
than at most other magnitudes, whereas some results yielded by maximum likelihood method
fluctuate on some range of magnitudes. It thus confirmed the claim that m = -1.5 is the optimum
magnitude of completeness.
Figure 7.2. Change of b-values and number of aftershocks with the magnitude of completeness for (a)
first mainshock of Creighton Mine (b) second mainshock in Creighton Mine. The cumulative magnitude
distribution curve is approximately linear for M > -1.5 for both the aftershocks sequences. The b-values
calculated by the least-square method and maximum likelihood method at magnitude = -1.5 are
significantly closer than that with most other magnitudes.
Figure 7.3. Change of b-values and number of aftershocks with the magnitude of completeness for (a)
first mainshock in Kidd Mine (b) second mainshock in Kidd Mine. The cumulative magnitude
distribution curve is approximately linear for M > -1.5 for both aftershocks sequences. The b-values
calculated by the least-square method and maximum likelihood method at magnitude = -1.5 are
significantly closer than that with most other magnitudes.
96
|
Virginia Tech
|
The temporal decay of aftershock activity is described by the modified Omori’s law (Utsu and
Seki 1954). The applicability of the modified Omori’s law in mining-induced seismicity has been
proven (Vallejos and McKinnon 2010) (Vallejos and McKinnon 2011).
𝑑𝑁 𝐾
𝛾 ≡ = ( 7-1 )
𝑑𝑡 (𝑐+𝑡)𝑝
where t is the time elapsed since the mainshock and K, p, c are empirical parameters. Omori’s law
manifests the temporal correlations in aftershock sequences, which are relax processes after
mainshocks (Shcherbakov, Turcotte et al. 2005). The occurrence of a mainshock redistributes the
stress. During a main shock, the stress and strain are enhanced in some regions neighboring the
fault where the mainshock is located. The stress relaxation is associated with the occurrence of
aftershocks. Aftershocks assist to relieve the stress concentration arising from mainshocks. All
aftershocks contribute to the reduction in regional stress as a function of the magnitude of
aftershocks (Shcherbakov, Turcotte et al. 2005).
Since the value of parameters in Omori’s law analysis corresponds to different seismicity
sequences, the value of parameters K, p, and c are empirically determined. The exponent p is the
most important parameter among them. The dependence of p was investigated and found that the
exponent p increases as a function of the magnitude of mainshock. The physical mechanism of p
is explained in that aftershocks of the mainshock with a large magnitude decay at a faster rate than
aftershocks of the mainshock with a smaller magnitude. Previous studies demonstrated that the
average p value ranges from 0.9 to 1.2 for mainshocks with magnitudes going from 5 to 7.5
(Ouillon and Sornette 2005). By extending the Omori’s law in mining-induced seismicity, it is
measured that p is 0.4 for the aftershocks sequence that occurred in the Creighton Mine between
October 1997 and March 1998 (Marsan, Bean et al. 1999).
7.4 Results
The b-value variation in Creighton Mine
It has been discussed that the temporal and spatial changes of b-value have potentially
important implications for understanding seismicity patterns and forecasting. After selection of the
magnitude of completeness, a series of b-values of seismicity during the period March to
December 2011 are yielded. Figure 7.4 plots the variations of b-value with the occurrence of
mainshocks, which includes both the first mainshock and second mainshock. Note that the b-values
fluctuate significantly during the starting four weeks because of the large uncertainty of that period.
97
|
Virginia Tech
|
This phenomenon has also been found in the data set of arrival times of seismicity, noting that the
data of the four weeks are flawed. Both the results from one week and two weeks scales exhibit
the lowest b-value comparing with b-values of other periods.
(a) (b)
Figure 7.4. Temporal change of the b-values for (a) one week scale (b) two weeks scale. Continuous line
with open circles: b-values through all time periods; Asterisk: the specific time period with occurrence of
mainshocks. The period noted by an asterisk includes the time of the mainshocks in July, 2011.
According to the evidence of the temporal variation of b-values in crustal earthquakes, there
are mainly two patterns of b-value change associated with mainshocks (Smith 1981). First, the b-
value increases to a peak and then decreases preceding the mainshock. The mainshock occurs
during the decrease. Second, a high b-value appears prior to the mainshock. The b-value is thus
influenced by either the previous mainshock or the mainshock after the peak of b-value. The
investigation of b-value change in Creighton Mine demonstrates that b-value changing with the
occurrence of mainshock agrees with the b value change in the crustal earthquakes study discussed
above.
The b-value analyses of temporal change associated with the occurrences of mainshocks are
not performed, because the low seismic rate cannot yield reliable b-values if seismicity is divided
into small durations of the time periods.
Modeling fit for aftershocks in Creighton Mine using Omori’s law
Figure 7.5A and Figure 7.5B show the modeling fit for the aftershocks for the first and second
mainshocks, respectively. The statistical model used in this analysis is developed for aftershocks
in natural earthquake sequences by Woessner (2005). Bootstrapping was originally developed for
98
|
Virginia Tech
|
Figure 7.5. Creighton Mine cumulative number of aftershocks decay with time modeling for (a) the first
mainshock (b) the second mainshock. Solid lines: practical cumulative number of aftershocks; Dash lines:
fitted model using Generalized Omori’s law.
The fitted parameters of modeling for Generalized Omori’s law are computed. The p value is
0.81 and 0.87 for the first mainshock and the second mainshock, respectively. The fitted p values
indeed conform to the reasonable range from the natural earthquake study, which gives 0.14 < p <
1.20 for a considerable number of earthquakes between 1932 and 2003 (Ouillon and Sornette
2005). By comparing the yielded p value between the first mainshock and the second mainshock,
it is found that p (0.87) of the second mainshock is significantly close with that (0.81) of the first
mainshock. However, the p value usually increases with larger magnitudes in the study of crustal
earthquakes. It is interpreted as the magnitude dependence of p values because aftershocks of large
mainshocks decay at a faster rate than aftershocks of small mainshocks (Ouillon and Sornette
2005). However, a larger mainshock can trigger more events and then the decay of aftershocks
needs more time. Thus, a larger mainshock does not imply a large p value due to uncertainties of
the decaying process, whereas evidences are found in empirical studies of crustal earthquakes. As
a consequence, magnitude dependence of the p value fails to apply to mining-induced seismicity.
The close parameters for different aftershock sequences suggests that the fitted model is accurate
for application in the mainshocks of the same mine site.
100
|
Virginia Tech
|
Figure 7.6. Kidd Mine cumulative number of aftershocks decay with time modeling for (a) the first
mainshock (b) the second mainshock. Solid lines: practical cumulative number of aftershocks; Dash lines:
fitted model using Generalized Omori’s law.
7.5 Summary and discussion
Through b-values analyses and fitted model for aftershocks decay, we identified significant
changes of b-value associated with mainshocks and the pattern of aftershocks of decay. These
findings confirm the implication that b-value can be used to forecast potential mainshocks. In
addition, the fitted model for aftershocks decay provides a reference of decay process from
historical data. Indicators of potential hazards can be reported if the aftershocks fail to follow the
decay model in seismic hazard analysis. The comparison between observed and predicted decay
process may be applied to examine whether the aftershocks from rock bursts keeps a normal decay,
ensuring safety for restarting work in underground mines.
A reasonable examination of completeness of magnitude ensures accurate regression between
cumulative number of seismic events and the distribution of magnitudes. From our case studies, -
1.5 is a reliable completeness of these two mines. It is recommended that -1.5 be used as future
analyses of aftershocks in mining-induced seismicity.
Previous studies of b-value and fitted model of Generalized Omori’s law furnish a
comprehensive framework for seismic hazard analyses in mining-induced seismicity. However, it
102
|
Virginia Tech
|
8 Conclusions and Discussion
The study presented in this paper sought to develop adequate seismic hazard analysis for deep
hard-rock mines based on mining-induced seismicity. In order to complete this objective, double
difference tomographic studies, velocity fitting through raypaths, and statistical analyses on
aftershock modeling are designed to provide a comprehensive means for evaluating and
forecasting the potential seismic hazard in deep hard-rock mines. The fundamental concept used
in this paper is to investigate the changes associated with the occurrence of mainshock, and then
these changes discovered from historical data can be used as factors or indexes to assess stress
distribution and forecast potential seismic hazards in future mining processes. Geophysics
techniques used in crustal earthquakes provide underlying frameworks for mining-induced
seismicity. Velocity structure of rock masses, b-value change with mainshocks and temporal decay
of aftershocks give rise to comprehensive understanding of forecasting mainshocks in deep hard-
rock mines.
We have applied the double-difference tomography in data sets of mining-induced seismicity
within a certain space-time neighborhood of mainshocks (major events). It is observed that both
high velocity and low velocity anomalies are exhibited for all the periods. Prior to the occurrence
of mainshocks, high velocity body shifts toward the location of mainshocks. In addition, the
discrepancy between high velocity anomalies and low velocity anomalies concentrates in a larger
extent. Considerable consistence of the overall velocity distributions still exhibits before the
occurrence of mainshocks. Associated with the mainshocks, dramatic change of velocity
distribution is observed. A high velocity body dominates the vicinity of mainshocks, whereas low
velocity areas shift away with the mainshocks. The high velocity body experiences a trend of
moving beyond the center of drifts, and low velocity areas continue to approach the mainshocks
around drifts. It is explained that mainshocks are a possible driving force to alter the stress
distribution. Highly-stress approaches the potential mainshocks and high velocity anomalies are
found before the occurrence of mainshocks. Also, stress relief is found north of drift. Within two
weeks after the mainshocks, stress developed less concentrated around the drifts and low stress
areas were observed close to the drifts. Appearance of the high-velocity body and comparing
validity before and after the mainshocks proves the potential to use high velocity anomalies as a
precursor of mainshocks.
106
|
Virginia Tech
|
Besides the tomography studies on the velocity structures, the average velocities of Creighton
Mine and Kidd Mine are computed by robust linear fitting on distance and travel time pairs of all
the raypaths. Results show that coseismic and postseismic velocity decreases caused by the
mainshocks are observed in the rock mass adjacent to the mining region. The velocity change
suggests that velocity decreases due to the fractures, caused by regional static stress and dynamic
shaking from mainshocks and microseismicity, weakening the wave propagation in the rock mass.
The velocity eventually returns to the background values assigned in velocity models after
recovery due to the closure of the cracks (crack healing effects). Velocity reductions associated
with the increasing number of seismic events reflect that considerable fractures weakened wave
propagation. One possible explanation of velocity increases is an increase in bulk modulus due to
crack healing effects. Two factors lead to the increase in bulk modulus. First, dilatancy-induced
preexisting cracks are closed by enhanced stress, giving rise to an increase in elastic moduli.
Second, the bulk modulus increases when dilatancy-induced crack closure offsets the elastic
volume increase.
In addition to focusing the investigations on seismic analysis before and after mainshocks, the
whole data set of recorded microseismicity for both Creighton and Kidd Mine are sequentially
divided into different groups, each of which includes 1000 seismic events. Seismicity rates are
computed to analyze how the seismicity rates change with the velocity distribution in
corresponding tomograms generated from each group. It is summarized that microseismic events
are more likely to occur close to the areas between low velocity regions and high velocity regions.
Microseismic events spread to a larger extent when the seismicity rate is low. Conversely,
microseismic events tend to be spatially concentrated during the time interval of high seismicity
rate. Further, it is indicated that frequency and moment magnitude of seismic events in both
Creighton Mine and Kidd Mine follow the Gutenberg-Richter law, which is a fundamental
observation from empirical studies of crustal earthquakes.
By using a Matlab function “Pcolor”, the average velocities were calculated and plotted from
the velocity of nodes distributed in the same unit cube and give a quantitative comparison of the
velocity before and after mainshocks. The velocity structures of Creighton Mine and Kidd Mine
are computed and inversed by TomoDD. It is found that high velocity anomalies are caused by the
accumulation of static stress in the regions including potentially large magnitude events of the rock
mass. Velocity reduction after the mainshocks in the vicinity of mainshocks implies that static
107
|
Virginia Tech
|
stress drop and dynamic-shaking induced fractures mutually lead to the weakness of waveform
propagation. The results presented indicate that velocity anomalies within the rock mass in
underground mining are associated with the occurrence of large magnitude events. It is possible to
forecast the mainshocks by detecting whether the wave speed changes in a relatively large
amplitude compared to the historical data set.
An important index b-value is computed to illustrate its temporal change associated with
mainshocks on the premise that mining-induced seismicity conforms to the Gutenberg-Richter law.
The b-value change patterns of crustal earthquakes are introduced and referenced for feasible
discussion of b-value of mining-induced seismicity. The b-value change temporally around the
mainshock indicates that the dramatic increase of b-value can be used as a precursory signature
for assistance on forecasting the occurrence of mainshocks. In addition, another essential issue of
how to determine the completeness of magnitude is discussed based on least-square method and
maximum likelihood method. It is explained that -1.5 is a reliable completeness of magnitude for
mining-induced seismicity. Further, aftershock decay temporal processes are modeled based on
the statistical model of crustal earthquakes. Fitted models exhibit close agreements with real
cumulative numbers temporal decay. Detailed parameters of Generalized Omori’s law are
generated for extended use on other aftershock sequences for the same mine site. It is also
mentioned that the p value dependence of magnitude is not evident in aftershock sequences of
mining-induced seismicity. However, the p value dependence of magnitude has been discovered
in crustal earthquakes. Although developed specifically for the seismic risk analysis of individual
sites, the methods applied to these mines can be extended to other deep underground mines. More
case studies of mining-induced seismicity can be developed to ensure this property, which can be
applied to improve the accurateness of mainshock forecasting.
108
|
Virginia Tech
|
Remote Characterization of Underground Ventilation Systems using Tracer Gas and CFD
Guang Xu
ABSTRACT
Following an unexpected event in an underground mine, it is important to know the state
of the mine immediately to manage the situation effectively. Particularly when part or the whole
mine is inaccessible, remotely and quickly ascertaining the ventilation status is one of the pieces
of essential information that can help mine personnel and rescue teams make decisions. This
study developed a methodology that uses tracer gas techniques and CFD modeling to analyze
underground mine ventilation system status remotely. After an unanticipated event that has
damaged ventilation controls, the first step of the methodology is to assess and estimate the level
of the damage and the possible ventilation changes based on the available information. Then
CFD models will be built to model the normal ventilation status before the event, as well as
possible ventilation damage scenarios. At the same time, tracer gas tests will be designed and
performed on-site. Tracer gas will be released at a designated location with constant or transient
release techniques. Gas samples will be collected at other locations and analyzed using Gas
Chromatography (GC). Finally, through comparing the CFD simulated results and the tracer on-
site test results, the general characterization of the ventilation system can be determined.
A review of CFD applications in mining engineering is provided in the beginning of this
dissertation. The basic principles of CFD are reviewed and six turbulence models commonly
used are discussed with some examples of their application and guidelines on choosing an
appropriate turbulence model. General modeling procedures are also provided with particular
emphasis on conducting a mesh independence study and different validation methods, further
improving the accuracy of a model. CFD applications in mining engineering research and design
areas are reviewed, which illustrate the success of CFD and highlight challenging issues.
Experiments were conducted both in the laboratory and on-site. These experiments
showed that the developed methodology is feasible for characterizing underground ventilation
systems remotely. Limitations of the study are also addressed. For example, the CFD model
requires detailed ventilation survey data for an accurate CFD modeling and takes much longer
time compared to network modeling.
|
Virginia Tech
|
Attribution
Beside my committee members, one professor and three colleagues provided technical
and editorial input to different chapters in this dissertation. A brief description of their
contributions is included here.
Harold M. McNair is an emeritus professor in the chemistry department at Virginia Tech.
He is a co-author on Chapter 7 in this dissertation. He provided lots of technical support on gas
chromatography and edited this chapter.
John Bowling was a graduated master student in the department of mining and minerals
engineering at Virginia Tech. He is a co-author on Chapter 3 in this dissertation. He provided
technical and editorial input for this chapter.
Steve Schafrik is a Ph. D. student in the department of mining and minerals engineering
at Virginia Tech. He is a co-author on Chapter 4 in this dissertation. He provided technical
support on the high performance computer that was used for the CFD modeling and edited this
chapter.
Edmund Jong is a Ph. D. student in the department of mining and minerals engineering at
Virginia Tech. He is a co-author on Chapter 5, 6, and 7. He contributed editorial comments on
these chapters and helped with some of the experiments in Chapter 6 and 7.
vii
|
Virginia Tech
|
1 Introduction
Following an unexpected event in an underground mine, it is important to know the state
of the mine immediately, even with limited information, in order to manage the situation
effectively. When part or the whole mine is inaccessible, remotely and quickly ascertaining the
ventilation status is one of the critical pieces of information that can help mine personnel and
rescue teams make decisions. While underground communications systems are rapidly
improving and are designed for use post-incident, their survival cannot be guaranteed and it is
necessary to develop other methods to ascertain the mine status. Some alternate methods can be
used to gather information safely and remotely, including collecting air samples from bore holes,
inserting video cameras into bore holes to visualize underground status, and utilizing rescue
robots underground if possible. Most information is not clear before rescue personnel enter the
mine and none of the methods mentioned above could be reliable and efficient enough to stand
alone. In a word, in an emergency situation, any information is “gold”, not only to effectively
save miners’ lives, but also to help decision makers manage the emergency effectively, to
increase safety for the rescuers, and to advance the rescue operation.
To quickly characterize the ventilation system is one of the pieces of essential
information that can help mine personnel and rescue teams make decisions that save lives and
ensure the safety of responders. Especially in some incidents, such as an explosion,
communication systems may be severely damaged and collapse may occur with very little
information available at the surface. However, the airflow paths and ventilation patterns will
change according to the level and the location of damage. Therefore, the ventilation
characterization can be analyzed and predicted by monitoring and studying changes in the
ventilation system. Due to the complexity of the ventilation system, the use of the tracer gas is an
effective method and has been used in many situations where conventional techniques are
inadequate or cannot be effectively employed [1], [2]. Numerical simulations using
computational fluid dynamics (CFD) can be used to model the ventilation status, which can be
compared with the data from tracer gas measurement to further analyze, predict and confirm the
underground ventilation status.
The research described here aims to develop a new methodology that can characterize
underground ventilation systems using tracer gas techniques and CFD modeling. The overview
1
|
Virginia Tech
|
of the methodology can be seen in the flow chart shown in Figure 1. After an unanticipated event
that has damaged ventilation controls, the level of the damage and the possible ventilation
changes need to be estimated based on the available information. Then CFD models will be built
to model the normal ventilation status before the event, as well as possible ventilation damage
scenarios. At the same time, tracer gas tests will be designed carefully and performed on-site.
Tracer gas will be released at a designated location with constant or transient release techniques.
Gas samples will be collected at other locations and analyzed using Gas Chromatography (GC).
Finally, through comparing the CFD simulated results and the tracer on-site test results, the
general level of ventilation damage can be determined.
Preliminary estimation of the damage
level and possible ventilation scenarios
using available information
CFD modeling of normal ventilation
Tracer on-site experiments
status and possible damage scenarios
Compare on-sit experimental
results and CFD simulated results
Identify the level of
ventilation damage
Figure 1. Flow chart of the methodology
This dissertation consists of eight chapters. Each chapter, excluding the first and the last,
is a paper that is published in, or about to be submitted to, conference proceedings or peer
reviewed journals. This information is provided at the beginning of each chapter.
Chapter 1 is an introduction to the research and an overview of the developed
methodology. The literature review is detailed in Chapter 2 focusing on CFD applications in
mining. It is a standalone review that will be formatted for journal publication. Chapter 3
describes a preliminary laboratory experiment and CFD modeling with the objective of
evaluating the feasibility of CFD modeling of tracer gas in an experimental apparatus. Chapter 4
is a completed laboratory experiment with a detailed CFD modeling study. The aim of this study
is to use the experimental data to validate the CFD model, study the relationship between the
tracer concentration and the location of incident damages, and finally through analysis of air
2
|
Virginia Tech
|
samples and the CFD model results, determine the general location of ventilation damage.
Chapter 5 is a CFD modeling study for an actual size conceptual mine developed based on the
laboratory experimental mine model which primarily aims to review the best methodology for
full CFD mine simulations and the potential useful information that the simulation results can
provide. Chapter 6 describes the application of the developed methodology to an actual mine. It
proved that the methodology is feasible not only in the laboratory, but also in the field. Chapter 7
discusses some common problems encountered when using tracer gases in underground mines,
including tracer release methods, sampling and analysis techniques. Additionally, the use of CFD
to optimize the design of tracer gas experiments, which played an important role in this study, is
also presented. The aim of the chapter is to provide guidelines and recommendations on the use
of tracer gases in the characterization of underground mine ventilation systems. Finally, chapter
8 contains the conclusions and discussions of this study. Highlights of this research and
recommendations for future work are also included.
3
|
Virginia Tech
|
The following review paper will be submitted to a mining specific journal. It was entirely written by Guang
Xu with editorial input from Dr. Kray Luxbacher and Dr. Saad Ragab, and may be compressed to meet journal
requirements.
2 Review of CFD Applications in Mining
2.1 Introduction
The principles of fluid dynamics are widely applied to mine ventilation, including
methane control, fire development, explosion, dust movement, and ventilation efficiency.
Understanding of the mechanism of fluid dynamics is valuable for solving problems, especially
safety and health problems, in the mining industry. Due to the complexity of the phenomena
involved in mine fluid dynamics problems, Computational Fluid Dynamics (CFD) modeling has
been increasingly applied to the mining industry in recent years to accurately predict the flow
patterns, study the flow mechanism and results, and design equipment to improve the efficiency
and safety of mine industry. CFD modeling is especially useful when the comprehensive analysis
using physical experimentation requires expensive equipment, large amounts of time and
understanding of flow in inaccessible areas.
CFD is a tool with which one can carry out numerical experiments with the purpose of
determining indices that are impossible, or at least very difficult, to obtain from experiments.
The numerical experiments can not only be used to help interpret physical experiments, but also
to better understand phenomena that are observed during physical experimentation [3]. The
computational cost of CFD is dropping as a result of increasing speed of computers, and with the
cost of physical experiments generally increasing, these costs can be reduced considerably with
the use of CFD that can be used to better design physical experiments and increase efficiency.
CFD plays a strong role as a research and design tool and is a well-established technique
applied to a broad range of fields including aircraft, turbomachinery, automobile and ship design,
and meteorology, oceanography, astrophysics, biology, oil recovery, civil and architecture. Many
of today’s mining problems need both analysis and visualization of fluid flow behavior in
complex geometric domains making CFD a powerful application in both mining research and
design.
Because of the success of CFD, there are many publications of CFD application studies
in mining; however, very few include a comprehensive review detailing the current state-of-the-
art in research and development. The purpose of this paper is to present such a review which
provides current state-of-the-art information about the progress in CFD application in mining and
5
|
Virginia Tech
|
illustrates its capabilities by way of examples. The emphasis is on general-purpose commercial
CFD code methodology rather than specialized CFD software development. Previous research
into the area of CFD applications in mining is explored and summarized in this paper.
2.2 Principles of CFD
Computational fluid dynamics (CFD) is one of the branches of fluid mechanics. It began
evolving in the early 1970’s and employed physics, numerical mathematics and computer
sciences to simulate fluid flows. CFD deals with numerical solution of differential equations
governing the physics of fluid flow and the interaction of the fluid with solid bodies [4]. It uses
numerical methods and algorithms to solve and analyze fluid flow problems. Flows of gases and
liquids, heat and mass transfer, moving bodies, multiphase physics, chemical reaction, fluid-
structure interaction and acoustics can be simulated through computer modeling [5]. The
technique enables the user to predict what will happen under a given set of circumstances. The
following section will give a brief introduction of the governing equations, along with the
general methodology used in CFD.
2.2.1 Governing equations
CFD is based on the fundamental governing equations of fluid dynamics which express
the fundamental physical principles of fluid dynamics. These governing equations have four
different forms based on how they are derived: integral and partial differential form,
conservation and nonconservation form. They are not fundamentally different equations but the
same equation in four different forms [3].
The conservation of mass (The Continuity Equation) expresses the fact that mass cannot
be created or disappear in a fluid system, the net mass transfer to or from a system during a
process is equal to the net increase or decrease in the total mass of the system throughout the
process [4]. The partial differential equation form of the continuity equation is shown in
Equation 2-1 [3].
⃗⃗ ( 2-1 )
Where ρ is the density of fluid (kg/m3); t is time (seconds); is velocity vector (m/s);.
The conservation of momentum (Newton’s second law), as shown in Equation 2-2,
describes how the force action on the particle is equal to the mass of the particle times its
6
|
Virginia Tech
|
acceleration. When applied to the fluid element it states that the variation of momentum is
caused by the net force acting on a mass element. The conservation form, partial differential
equation, called the Navier-Stokes equation, is shown in Equation 2-3 [3].
( 2-2 )
Where F is body force (N), m is mass (kg), and a is acceleration (m/s2).
⃗⃗ ⃗⃗ ⃗⃗ ⃗ ( 2-3 )
Where in addition to the variables defined in equation 1 and 2, is viscous stress tensor
(newton) given by Equation 2-4, ⃗ is body force vector (newton), and μ is the molecular
viscosity coefficient.
( ⃗⃗ ( ⃗⃗ ) ) ( ⃗⃗ ) ( 2-4 )
The conservation of energy (the first law of thermodynamics) states that energy can
neither be created nor destroyed, but can only change forms. Specifically, it states that any
changes in time of the total energy inside the volume are caused by the rate of work of forces
acting on the volume and by the net heat flux into it [4]. The conservation form, partial
differential equation is shown in Equation 2-5 [3].
⃗⃗ ̇ ⃗⃗ ( ⃗⃗ ) ⃗⃗ ⃗⃗ ( ) ⃗ ( 2-5 )
Where ̇ is the rate of volumetric heat addition per unit mass , T is temperature, e is
internal energy per unit mass.
2.2.2 Steady and unsteady flow
The behavior of state of a fluid, such as velocity, pressure and density, generally vary
with respect to space and time. A steady flow is one in which the state of the fluid may differ
from point to point but do not change with time. Otherwise, if at any point in the fluid, the
conditions change with time, the flow is described as unsteady [6]. Realistically, there is always
slight variation in velocity and pressure in flow, but if the average values are constant, the flow
can be considered steady to study the problems effectively [6].
2.2.3 Laminar and turbulent flow
If the particles of fluid move in straight lines even though the velocity with which
particles move along is not necessarily the same, the fluid may be considered as moving in layers
7
|
Virginia Tech
|
and called laminar flow. Contrarily, if the paths of fluid particles are in a random and disorderly
manner and a thorough mixing of fluid takes place, the flow is considered turbulent [7].
Reynolds number (Re) is evaluated to determine whether the flow is laminar or turbulent.
It is calculated using Equation 2-6. For internal flow where ρ is the density (kg/m3), u is the
mean velocity over the cross section (m2/s), l is the pipe diameter or the hydraulic diameter for
non-circular ducts (m) and μ is the dynamic viscosity of the fluid (Pa.s). Under normal
engineering conditions, flow through pipes at Re< 2000 may be regarded as laminar, Re > 4000
may be taken as turbulent flow and 2000<Re<4000 are treated as transitional flow, which is a
mix of laminar and turbulent flow [6].
( 2-6 )
The flow in mine gob is treated as laminar flow in porous media in many studies.
However, the real flow inside gob is still not fully understood, and the laminar flow assumption
may not be valid [8], [9]. For most CFD codes, the modeling of transitional flow is usually not
provided. But since most times the transitional flow only covers a small region of the total flow
domain, it could be neglected and still allow for acceptable results [10], [11].
In underground mine ventilation, most flow states in mine openings are turbulent, which
allow for effective dispersion and removal of contaminants in the workplaces [12]. There are
many turbulence models available; however, none of the existing turbulence models is
universally accepted as being superior for all turbulent problems. Some of the well-known
turbulence models are discussed below.
Direct numerical simulation (DNS) solves the Navier-Stokes equations directly and
resolves all the scales of motion. It is the simplest approach and provides the most accuracy.
However, DNS requires very high grid resolution, thus, it has extremely high cost, and the cost
increases rapidly with the Reynolds number (approximately as Re3). For this reason, the DNS
approach was impossible until the 1970s when acceptable computational capability was
achieved. The DNS approach can provide valuable information for verifying or revising other
turbulence models, but the application is limited to fundamental studies, with low or moderate
Reynolds numbers flows [13–15].
However, for most engineering applications resolution of turbulent fluctuations in detail
is unnecessary. Therefore, some common turbulence models are based on the Reynolds averaged
Navier-Stokes equations (RANS) model which solves the Reynolds equations for the mean
8
|
Virginia Tech
|
velocity field through time averaging. The RANS models are determined by either the turbulent
viscosity hypothesis or modeled Reynolds-stress transport equations [13]. The turbulent viscosity
hypothesis is also called eddy viscosity hypothesis, and was introduced by Boussinesq in 1877. It
states that the deviatoric Reynolds stress 〈 〉 is proportional to the mean rate
of strain [13], which is shown below,
〈 〉 〈 〉
〈 〉 ( ) ( 2-7 )
Where is the turbulent viscosity.
Although the turbulent viscosity hypothesis has not been justified and has poor accuracy
for many flows [13], many of today’s most widely used turbulence models are based on it.
2.2.3.1 The standard k-ε model
The standard k- model is a two-equation model that computes the Reynolds stresses by
solving two additional transport equations, which are for the turbulence kinetic energy, k in
equation 2-8, and the dissipation rate of turbulence, in Equation 2-9 [16], [17].
[ ] ( 2-8 )
Where E is the mean rate of deformation tensor.
ij
[ ] ( 2-9 )
Finally the turbulent viscosity in Equation 2-10 is derived from both k and [16], [17]:
( 2-10 )
The model coefficients in the standard k- model are shown below [16], [17]:
( ) ( 2-11 )
The standard k- model is the simplest complete turbulence model and widely used in
the modeling of mining turbulent flow in broad range of applications. Yuan used the standard k-
model in the ventilation airways to study the flow path in the gobs [8], and the spontaneous
heating behavior in the gobs [18–21]. Similarly Ren [9] used the standard two equation k-
model to estimate the turbulent transport in his gob spontaneous combustion study. However,
these study results were not validated. Greg et al. [22] used the standard k- model in a coal dust
explosions study and the results were validated with test data from a coal dust explosion test
facility. Toraño et al. [23] used the standard k- model and Shear-Stress-Transport (SST) model
in their study to evaluate the wind erosion effect on different coal pipe. They finally chooses the
9
|
Virginia Tech
|
standard k- model with a certain wall roughness parameter due to its better agreement with the
US EPA wind tunnel measurements. Another study Toraño et al. [24] conducted used six
different turbulence models to simulate dust behavior in auxiliary ventilation in mining roadways
and compared their results with field measurements data, and the standard k- model provided
better results. Silverster [11] states that the standard k- model is a more general but
computationally intensive method and is favorable to be used in the field of mine ventilation.
This turbulence model was also successfully used in the CFD study of dust dispersion [25] and
minerals processing [26–29]. Many studies also indicate it can provide precise and good
correlation between the measured and the simulated results [30–34]. However, the standard k-
model is reported may produce inaccurate results under certain circumstances, especially for
flows with rotation, curvature, strong swirl, three dimensionality, and flows with strong
streamline curvature [5], [35]. This is partially because the turbulent viscosity hypothesis is not
valid if turbulence is not isotropic and the equation has many empirical constants which have
adverse effect on the predicted results [13].
2.2.3.2 The RNG k-ε model
The RNG k- model is an improvement on the standard k- model and it is derived from
the statistical methods used in the field of renormalization group (RNG) theory [36]. It is similar
in form to the standard k- model but includes modifications in the dissipation equation to better
describe flows in high strain regions, and a different equation is used for effective viscosity. The
turbulent kinetic energy and the dissipation rate equation are shown below:
( ) ( 2-12 )
( ) ( ) ( ) ( 2-13 )
Where √ , ( ), and , , , are derived from the RNG
theory [17].
2.2.3.3 The realizable k- model
The realizable k- model share the same turbulent kinetic energy equation as the standard
k- model, but a variable as shown in Equation 2-14, instead of a constant, to calculate the
turbulent viscosity using Equation 2-10 [17].
10
|
Virginia Tech
|
( 2-14 )
Where A =4.04, √ , √ , (√ ),
0
, and ̃ √ .
̃
Also a new transport equation is used for dissipation rate, which is shown in Equation 2-
15 [17].
[( ) ] ( 2-15 )
√
This model is better than other k- models for many applications, and has especially
improved the modeling of planar and round jets, boundary layers under strong adverse pressure
gradients or separation, and rotation and recirculation flows [17], [36].
2.2.3.4 Reynolds stress closure models (RSM)
The Reynolds Stress Model (RSM) is a higher level, elaborate model. The turbulent
viscosity hypothesis is not needed in this model and individual Reynolds stresses 〈 〉 are
directly computed from the model transport equations [13]. The advantage of RSM is that it
introduced terms accounting for anisotropic effects into the stress transport equations, which are
critical for flows with significant buoyancy, streamline curvature, swirl or strong circulation
[37]. More detail about this model can be found in Pope’s book [13] or Durbin’s study [38].
RSM can produce more realistic and rigorous solutions for complicated engineering flow, but it
requires more execution time and memory, and it can be difficult to achieve good convergence
behavior using this model[5].
2.2.3.5 Spalart Allmaras model (SA)
This model was developed by Apalart and Allmaras [39], and is a one equation model
first used in aerodynamic applications. In this model, the turbulent viscosity is solved by a
single model transport equation. The model equation is provided blow
̅
( ) ( 2-16 )
̅
Where the source term which depends on the laminar viscosity , turbulent viscosity
, the mean vorticity , the turbulent viscosity gradient | |, and the distance to the nearest
wall [13]. This model is intended for aerodynamic flow, such as transonic flow over airfoils,
11
|
Virginia Tech
|
and its application to aerodynamic flows has proved successful [13]. For more detail about the
model, one can refer to the original paper [39].
Wala et al. [40] used different turbulent models to simulate the air flow and methane
distribution in a ventilation test gallery. The results from the shear-stress transport (SST) and the
Spalart Allmaras (SA) model were presented. Both models were successful in predicting the
methane concentration and the airflow distributions, while one may better than the other in
different scenarios. Parra et al. [41] also applied the SA turbulent model in the study of deep
mine ventilation efficiency. Good velocity agreement was achieved comparing to the
experimental values.
2.2.3.6 Large eddy simulation (LES)
Large eddy simulation (LES) directly represent the larger three-dimensional unsteady
turbulent motions [13]. A filter operation is applied to the Navier-Stokes equations to eliminate
small scales of the solution. LES resolves large scales of the flow field solution and can be
expected to be more accurate and reliable than alternative approaches such as RSM and RANS.
It is especially much better suited to unsteady effects than RANS [42]. The computational
expense lies between RSM and DNS models [13]. It is also a very computational expensive
method and the prediction results may not improve for fully developed turbulent flow, compared
with the k- model [43]. Because it is at a much earlier stage of development than RANS
modeling, few applications were found in the mining related fields. One example is Edwards and
Hwang [44] used the LES method in Fire Dynamics Simulator (FDS) to study fire spread in the
mine entry. The results were compared with measured values and the differences were
reasonably interpreted.
2.2.3.7 Summary
There is no clearly superior model which works well over different applications. For
general engineering turbulent modeling, Bakker [17] recommend that start the calculation using
standard k- model. For very simple flows that contain no swirl or separation, converge the
calculation with second order upwind scheme. For flow involves jets, separation, or moderate
swirl, converge the solution with the realizable k- model and seconder order difference scheme.
If swirl dominates the flow, then RSM and a second order differencing scheme are
12
|
Virginia Tech
|
recommended. Other models should only be used if there is evidence from the literature proving
they are especially suitable for the interested problem.
2.2.4 Numerical analysis
All methods in CFD use some form of discretization which can be classified as finite-
difference, finite-volume, and finite-element. CFD can be approached using any of the three
main types of discretization mentioned above [3].
Finite difference method (FDM) is among the first approaches applied to the numerical
solution of differential equations and is widely employed in CFD. It is applied to the differential
form of the governing equations. It uses Taylor series expansion for the discretization of the
derivatives of the flow variables. Finite difference method is simple and allows for one to obtain
high-order approximation to achieve high-order accuracy. However, the application is restricted
because this method requires structured grids and can only be applied to simple geometries due
to the reason that it cannot be applied directly in body-fitted coordinates. Thus, finite difference
methodology is rarely used for industrial applications [4].
Finite volume method (FVM), which is derived from the finite difference method,
directly satisfies the integral form of the conservation law and uses the integral form of the
governing equations. It discretizes the governing equations by dividing the domain of interest
into several arbitrary polyhedral control volumes, and then integrates the differential form of the
governing equations over each control volume. Finite volume methods have two primary
advantages which make popular for use in CFD codes, including CFX, FLUENT, and
PHOENICS. The primary advantage is that the spatial discretization is accomplished directly in
the physical space. It naturally achieves the coordinate system transformation between the
physical and computational domain. Secondly, finite volume methods not only can be easily
implemented on structured grids, but also do not require a coordinate transformation in order to
be applied on unstructured grids. Therefore, the flexibility of finite volume methods are
particularly suitable for treating complex geometries [4], [45].
The finite element methods (FEM) need the governing equations to transform from
differential form to integral form and start with dividing the physical space into triangular or
tetrahedral elements. It is popular because it uses integral form and unstructured grids, which are
13
|
Virginia Tech
|
preferable for complex geometries. However, the popularity in solving the CFD governing
equations using these methods only started in the early of the 90’s [4].
2.3 Commercial CFD codes
Tremendous progress has been made in the development of CFD codes since the 1990s;
hence, the use of CFD codes has increased dramatically in the last few years. The commercial
CFD codes are the primary source of tools in use by the mining industry and other engineering
communities. The powerful application of these commercial codes to model complex flow in
many research and design fields makes them much more attractive. Some of the common
commercial codes are listed in Table 1. Most of the commercial CFD codes use the finite volume
method due to the fact that it satisfies the integral form and allows for treatment of complex
geometry. These codes employ graphical user interfaces and can be supported on the platforms
of UNIX, Linux and Windows on workstations or PCs.
Table 1. Commercial CFD code
CFD code Company Web site
http://www.ansys.com/Products/Simulation+Technology/Fluid+Dynam
FLUENT Anysys
ics/ANSYS+FLUENT
http://www.ansys.com/Products/Simulation+Technology/Fluid+Dynam
CFX Anysys
ics/ANSYS+CFX
PHOENICS CHAM http://www.cham.co.uk/
http://www.esi-group.com/products/Fluid-Dynamics/cfd-ace-
CFD-ACE ESI
multiphysics-suite
CFD 2000 Adaptive http://www.adaptive-research.com/cfd2000_software.htm
2.4 CFD commercial software analysis process
There are generally three stages to perform CFD analysis: preprocessing, solving and
postprocessing. Figure 2 shows the flow chart of CFD analysis process.
Preprocessing is the first step in building and analyzing a CFD model taking place before
the numerical solution process. The first step of the analysis process is to consider and
understand the flow problem. The second step is to create the geometry of the problem. CAD
geometries can be imported and adapted for CFD software. Approximations and simplifications
of the geometry may be needed to analyze the problem with reasonable effort. Then a suitable
computational mesh needs to be created and applied to the problem domain. After the mesh has
been developed, boundary conditions and initial conditions should be specified according to the
physical conditions which give the simulation a starting point. Finally, the flow problem is
specified by the fluid parameters, physical properties and solving techniques.
14
|
Virginia Tech
|
•Problem consideration
•Geometry creation and import
•Mesh development
•Boundary and initial conditions setup
Preprocessing •Specify fluid parameters, physical
properties and solving techniques
• Solve discretised equations until
iterative convergence and
required accuracy are obtained
Solving
• Analyze the computed results
numerically and visualizely
Post
processing
Figure 2. Flow chart of CFD analysis process
Iterative methods are usually used to solve the discretized equations until a predetermined
convergence and required accuracy are obtained.
Postprocessing is the final step in CFD analysis. It organizes and interprets the data
generated by the CFD analysis. The results can be analyzed both numerically and graphically.
Some powerful commercial CFD software not only creates visualization graphs, including
contour, vector, line plots and even animations, but also allows for export of CFD data to third-
party postprocessors and visualization tools such as TechPlot. The illustrative presentation of the
results allows the designer or researcher to have increased understanding of the interested
problems, thus, understand how the system responds to a variety of different operating
conditions [46].
After these three steps of processes, in order to better understand the possible differences
in the accuracy of results and performance of the computation with respect to physical properties
and important parameters such as flow conditions and boundary conditions, the process may
need to be repeat in order to exam the sensitivity of the computed results.
15
|
Virginia Tech
|
Although the commercial softwares are user-friendly, the simulation process, especially
analyzing the results, requires complete understanding of the underlying physics, and sometimes
a model needs reasonable assumptions and improved boundary conditions to make it
manageable. Therefore, dependable results cannot be achieved without specialized training and
sound engineering skills [46].
2.5 Quality control of CFD
As aforementioned, CFD is increasingly used in the research and design of ventilation
and other fluid systems with in the mining industry. Conscientious execution during the process
of CFD studies is of paramount important to ensure quality CFD results because modeling and
numerical errors and large deviations may occur in such studies. This section looks at the various
techniques that are necessary to improve the quality of CFD calculations. Finally, guidelines for
CFD quality control procedures are provided which are recommended in CFD-related studies.
2.5.1 Mesh quality and convergence
It is complicated work to discretize the computational domain into a suitable
computational mesh. Mesh generation may account for the majority of time spent on a CFD
study in order to generate a proper mesh that allows for a compromise between desired accuracy
and computational cost [47]. The following discusses reasonable mesh quality, mesh size, and
mesh convergence which ensure a high-quality computational solution.
2.5.1.1 Mesh convergence
It is important to conduct a mesh independence study before utilizing CFD results since
the numerical solution may depend on the mesh size if mesh independence is not reached [48].
However, obtaining a mesh-independent solution is almost impossible due to computational
expense. The mesh convergence, which is a relaxed criteria of mesh independence, states that the
solution asymptotically approaches the exact solution of the governing equations [43]. This is the
more practical method which requires the solution does not change significantly as mesh is
further refined.
By comparing the results of different mesh sizes, mesh convergence should be studied
considering different flow features and different representative locations. Flow features usually
16
|
Virginia Tech
|
used to check mesh convergence are velocity profile along a line, velocity contours and vectors
at interested locations, temperature distribution, and contaminant concentration.
Grid Convergence Index (GCI) [49] can be used for uniform grid refinement studies in
CFD. GCI provides a conservative estimate of the error between fine grid solution and the
unknown exact solution [43]. GCI is expressed as
| |
( 2-17 )
Where is the factor of safety and recommended by Roache for two grid comparisons; p is
the formal order of accuracy; is the grid refinement ratio (usually is 2) , in which
are mesh size for fine and coarse grid, respectively. is the relative error which is
shown below
( 2-18 )
Where and are any solution of interest, such as velocity, contaminant concentration,
temperature, of fine grid and coarse grid, respectively.
This approach is intentionally developed for uniform grids, and the calculation should be
within the asymptotic range of convergence [49]. Refer to the original paper for detailed
derivation and application of the method.
2.5.1.2 Near wall mesh size
The wall treatment in turbulent flow models is very important, because the wall is in the
viscosity-affected regions which have large gradients in the solution variables. A successful
prediction of wall bounded turbulent flows are determined by the accuracy of the near wall
region [47]. A y+ strategy can be used as guidance in selecting the suitable grid configuration and
corresponding turbulence models.
The wall y+ is a mesh-dependent dimensionless distance from the wall expressed as in
Equation 2-19:
( 2-19 )
There are three regions in the boundary layer [50]:
1. Laminar sublayer (y+ < 5)
2. Buffer region (5 < y+ < 30)
3. Turbulent region (y+ > 30)
A high degree of mesh refinement in the boundary layer is required for low Reynolds
number turbulence models, because it solves it solve the governing equation all the way to the
wall. The first grid normally should be located at [5]. While for high Reynolds number
17
|
Virginia Tech
|
models, an empirical law-of-the-wall relations for the flow regime of the boundary layer is used.
It does not consider the damping effects of a wall and the computation must start at a point in the
fully turbulent region. In this case, the mesh does not need to extend into the boundary layer
region, and the number of computational cells is consequentially reduced [5], [43].
Different turbulent flow models require different ranges of , and recommendations are
usually available for different CFD code. For example, Ansys Fluent recommend using either
very fine near wall mesh, on the order of , or coarse mesh that . These
recommendations should be used when using a specific CFD code [43].
Ariff et al. [51], [52] conducted a series studies using the wall y+ approach to compare the
influence of different near wall mesh sizes and different turbulent flow models. The value
(near wall mesh size) was chosen according to the Fluent User’s Guide. The study provided
guidance on selecting appropriate mesh configuration and turbulence model.
2.5.2 Solution convergence
The numerical solution is an iterative process. A steady-state solution requires the
solution converge to an accurate approximation of the exact solution. In order to monitor how
much the solution changes with each iteration, a residual is introduced, which is a quantity that
measures the unknown error. One definition of the residual is shown in Equation 2-20 [15], [53].
∑
√ ( 2-20 )
Where u is the solution of this iteration, u is the solution of last iteration, and N is the
i i-1
number of grids or cell in the calculation domain. The scaled residual is shown in Equation 2-21.
∑ ∑
√ ∑
(√ )⁄( ) ( 2-21 )
∑
The scaled residual is often used which is a relative measure to the average value
calculated in the domain, rather than an absolute measure.
Commercial CFD codes usually provide default convergence criterion, such as stop the
computation if the residual is reduced to a four-order-of magnitude. The solution is assumed
converged when the residual is below the default criterion. However, the default convergence
criterion is not always sufficient to ensure that the solution is converged. Some times smaller
convergence criterion need to be apply to achieve an accurate solution [43].
18
|
Virginia Tech
|
Sometimes the residual may reached the convergence criterion, but the solution still
changes with further iteration, which means a stable solution has not been reached. Therefore, in
addition to monitoring the residual, the selected solution variables must be monitored until they
no longer change with more iteration. A point monitor can be set up to monitor the solution
(velocity, temperature, etc.). If the solution profile indicates no change as the iterations proceed,
the solution is considered converged [15]. Monitoring selected variables at certain points is
recommended after the residual monitor, as part of the solution convergence assessment [15].
2.5.3 CFD verification and validation
Verification and validation are essential processes required to achieve reliable CFD
results. Verification and validation assess the credibility of the CFD results. Verification deals
with the mathematical correctness of a numerical solution, whereas validation deals with the
physical correctness of model. Verification and validation are extensive topics with much
literature devoted to them. This section will only briefly discuss these topics along with their
application to mining research.
2.5.3.1 Verification
Verification deals with the mathematical correctness of the CFD solution including two
topics: code verification and solution verification. Mining engineering usually uses commercial
CFD code or already developed and verified CFD code, which already ensures there are no
unknown errors (or minimal errors) in the computer code. In this situation, code verification is
not necessary. The main task in solution verification is error estimation. There are three sources
of numerical errors in CFD: the round off error, the iterative convergence error, and the
discretization error. The round off error is due to the finite precision arithmetic of computers. It
isusually negligible compared to the two other sources of error and will not be further discussed.
The iterative convergence error comes from the inexact solution of the algebraic system by some
iteration method. The discretization error is due to the replacement of the differential equations
by partial different equations. These two kinds of errors can be reduced by conducting the mesh
convergence study and solution convergence study mentioned previously. Verification must take
place before validation, otherwise the computed results may agree with the experimental results
only by chance [54], [55].
19
|
Virginia Tech
|
2.5.3.2 Validation
Validation deals with the physical correctness of the CFD model. It usually conducted by
comparing modeled data to experimental data [55]. Validation is very important in the field of
CFD studies because it provides the degree of confidence necessary for the CFD results
application. This section discusses different techniques that have been used in the validation of
mining CFD models.
Laboratory studies have traditionally been used to study mine ventilation and other fluid
problems in the field of mining. Jade [56] conducted a CFD study to estimate shock loss
coefficients in two-way splits and junctions in mine airways. The CFD results are reasonably
close to that of experiments conducted in a designed laboratory setup, which showed the
potential of CFD in predicting airflow and shock loss in mine airways. Collecutt et al. [22]
calibrated their CFD model by comparing to a laboratory scale experiment before modeling an
actual dust explosion in a tunnel. The laboratory studies usually use physical scale models. In
this case, the laboratory results are only valid when geometric and dynamic similarity are
achieved between the physical model and its prototype. Geometric similarity requires the model
to have the same linear scale ratio in three dimensions, and dynamic similarity requires the
model and the prototype to have the same length, time, and force scale ratio. For incompressible
flow with no free surface, this requires the Reynolds numbers (Re) to be the same; and if a free
surface is exist, the Reynolds (Re), Froude (Fr), and Weber (We) numbers need to be the same.
Compressible flow scenarios require the Mach number to be the same.[35]. However, the
dynamic similarity is often difficult or impossible to obtain. Moloney [35] built a 1/10th scale
model for the purpose of validating the CFD model, but the model could not meet dynamic
similarity, because it required a 62 m/s exit velocity, which is not practical in the lab. The same
challenge was faced by Ndenguma [32] when a 15% scale model required an impractical exit air
velocity of 152 m/s. Instead of meet the dynamic similarity, a percentage volume flow method
was used to scale the air flow.
The experimental results found in the literature could also be used to validate a CFD
model. In Toraño’s CFD study [23], which focuses on the effect of wind erosion on different
coal storage pile shapes, the numerical model was validated by the US EPA experimental
reference study to ensure that the CFD model was accurate and valid for situation. The Root
Mean Square values of deviation, shown in Equation 2-22, were used to quantitate the difference
20
|
Virginia Tech
|
between the CFD and the EPA experiment results, and a 3.75% deviation was found, which
indicated the CFD model was accurate enough.
√ ∑ ̂ ( 2-22 )
Another method used to validate a CFD model is to use the data obtained from field or on
site full scale experiment. For example, Peng et al. [57] developed a 2D dense-medium separator
CFD model to analyze the flow patterns and the mechanisms of particle separation in the
separator. The in-plant test results, which showed a close fit to the simulated results, were used
for the CFD validation. Similar validation can be found in Peng’s other work detailing CFD
studies of mineral separation [58]. Toraño et al. [59] used an anemometer and methane detector
to obtain velocity and methane concentration data which were than used for the CFD model
validation. The best CFD model was then chosen based on the least experiment and simulated
differences. Parra et al. [41] conducted a detail ventilation measurement in a real mine gallery
using an anemometer to validate the numerical model, and the grid size and the algorithm used
for the simulation were guaranteed acceptable by achieving good agreement. The validated
model was then used to simulate different combinations of blowing and exhaust ventilation
methods. However, sometimes it is hard to conduct accurate field measurement, and it is
common for the measured field data have error up to 20% [60].
The use of tracer gases started in the 1950s in building ventilation systems [61]. Tracer
gas techniques have been used in many situations where the standard ventilation survey methods
are inadequate [1]. The applications of tracer gases in underground mines include analyzing
ventilation patterns, measurement of air leakage rates, and evaluating dust control methods [62].
For this reason, tracer gas is sometimes used for CFD model validation. Krog et al. [63] used
CFD to study the airflow patterns around the longwall panels and used SF as a tracer gas to
6
validate the CFD model. Konduri et al. [64] used CO as a tracer gas in their field experiment to
2
determine effective ventilation air quantities when a jet fan was used for auxiliary ventilation,
and the measured results were used to compare with the CFD simulated results.
Flow visualization has long been used in the fluid flow research, and can be used for
CFD validation. A visual comparison of the results from the experiments and the CFD
calculation provides effective means of CFD validation [65]. There are three basic visualization
21
|
Virginia Tech
|
techniques: adding foreign material, optical techniques, and adding heat and energy [65].
Moloney et al. [35] conducted an experiment using the laser sheet flow visualization technique,
combined with the CFD studies to evaluate different ventilation methods. In another study, Wala
et al. [66] utilized a Particle Image Velocimetry (PIV) system to visualize the airflow patterns in
a physical scaled mine model and validate CFD models. PIV is an optical technique that can
measure flow components in a plane. Highly reflective tracer particles need to be added to the
flow field. The motion of the particles can be recorded by a camera and the tracer particles are
illuminated twice within one camera shot. The processed results can display velocity vectors of
the flow field and potentially be used to compare CFD velocity vectors in the same plane.
Ndenguma [32] used smoke to visualize the flow patterns in a scale mine heading model
ventilated by jet fan and scrubber. The flow patterns represented by the smoke which were
recorded using a camera were similar to those presented from the CFD results.
Velocity agreement between experimental and CFD results is a major indicator of
validated CFD model. Therefore it is necessary to accurately measure flow velocities. Moloney
et al. [35] conducted a 1/10th scaled mine auxiliary ventilated headings experiment to validate the
corresponding CFD model. Laser Doppler Velocimetry (LDV), which can measure two
components of velocity without disturbing the natural flow patterns, was used to measure flow
velocities and validate CFD model. Hargreaves and Lowndes [67] used a Trolex TX6522
Multisensor unit and TX5924 vortex shedding anemometer, which are intrinsically safe devices,
to measure and record air speed in an underground coal mine. However, because the anemometer
can only measure airflow perpendicular to the measurement gate, and airflow in the mine is
usually three dimensional, the measurement of the airflow and data interpretation is difficult.
Therefore, a multi-directional intrinsically-safe anemometer is needed to adequately map the
complicated flow patterns in mines. Taylor et al. [68] used an ultrasonic anemometer to measure
three dimensional air flow velocities in a simulated mine entry. The results of the ultrasonic
anemometer were used to generate the quantitative flow profile using vectors, which includes the
direction and magnitude of flow. These results were later used for CFD validation purposes [40].
In addition, the advanced experimental methods, such as LDV and PIV, have been widely used
to give detailed information on the turbulent flow field in stirred vessels and validation for the
CFD studies in the process industry [26], [69–72].
22
|
Virginia Tech
|
2.6.1 Mine ventilation airflow
CFD is widely applied to the study of ventilation to improve the quality, quantity and
control of ventilation, which can further assist in the improvement of gas, dust and climate
control [67].
Wala conducted a series of studies aiming at validating CFD code by comparing the CFD
results against mining-related benchmark experiment results [40]. He pointed out that although
significant studies have been conducted using full-scale field tests or scaled physical modeling to
evaluate the face ventilation system performance and have improved the ventilation
effectiveness, there are still some doubts on the results due to the complexity of the ventilation
and the limitation of experimental methods. Traditional theoretical and experimental methods
can produce valuable results, but they are limited in completeness and accuracy. CFD is a
promising tool which embraces a variety of technologies and can overcome the disadvantages
mentioned above if the CFD solution has been validated. CFD 2000 was used to determine the
optimum design of an upcast shaft and main fan ductwork arrangement [73]. The CFD model
was first validated by existing experimental data available in other literature with reasonable
agreement. Then the author designed four different shaft collar and shaft cover arrangements and
simulated them with the validated CFD model. A best design was determined by analyzing the
simulated results which could reduce the power cost due to less pressure loss. They took CFD
validation studies a step further by performing several experiments in a scaled physical face-
ventilation model and comparing a CFD model with the experimental results in order to validate
the work [40]. Methane concentration and three-dimensional airflow were measured and
compared with the CFD results. They conducted a grid independence study, used the SIMPLE
algorithm to achieve pressure velocity coupling of momentum and continuity equations and
compared two turbulence models: the Shear-Stress Transport (SST) model and the Spalat-
Allmaras (SA) model. These two models can be both used to simulate methane concentration
and airflow distributions, but the SST model achieved better results in the box-cut scenario,
while the SA model showed better agreement with the experimental results in the slab-cut
scenario. Furthermore, a CFD study was carried out to study the effect of scrubbers on the face
airflow and methane distribution during the box cut mining sequence [74]. The SST turbulent
model was used in the CFD simulation, and the methane concentration results were compared
under four different scenarios for CFD and experimental results. The results showed that the
24
|
Virginia Tech
|
scrubber improved the face ventilation. However, they indicated further study was needed
determine the reason for some of the differences between experimental and simulated data. Some
other CFD studies carried out by the same group of authors were presented [75], which showed
the great potential of CFD in studying the underground airflow and improving the health and
safety of miners.
Jade used a laboratory experiment and CFD simulation study to investigate the shock loss
at the 90 degree intersections of two-way splits and junctions [56]. The shock loss coefficients
(SLC) results showed that the CFD models are validated well with the experimental data. The
SLC results were compared with previous literature and found that the literature underestimates
the SLC by 50% or more for two-way junctions, and 20% for straight branch, thus, concluding
that the widely accepted methodologies significantly underestimate SLC of two-way splits and
junctions. The study also conducted regression analysis and obtained various equations for two-
way 90 degree splits and junctions.
Zheng studied the Diesel Particular Matter (DPM) in an underground metal/nonmetal
mine using CFD [76]. The study investigated the airflow and diesel exhaust propagation patterns.
The study assumed that DPM movement can be represented by the air flow pattern since a very
small fraction of DPM exists in the air. A model was built, which represents part of a mine in
Missouri with highly mechanized room-and-pillar mining. The main air flow was simulated with
and without stoppings. The model showed that although the DPM conditions are much improved
by stoppings, some places still need auxiliary ventilation for adequate dilution. These places are
dead end headings, cross cuts, and downstream of the backfill block. The study also evaluated
the effectiveness of both blower and exhaust system to reduce DPM problems. The CFD model
simulated a single heading with a loader and truck operating in the immediate face. The results
showed that with the blower system the DPM is distributed in a smaller space than the
exhausting system, but the loader driver in both systems would be working in a high DPM
environment. Therefore, they determined that other strategies are needed to improve the
situation. Overall, this study showed that CFD method can be used to simulate the airflow
patterns for the entire mine or part of it, the ventilation efficiency and different ventilation
measures can be evaluated.
Parameters such as velocity and contaminant concentration are broadly studied to
evaluate the underground ventilation system. However, the more restrictive parameters, such as
25
|
Virginia Tech
|
mean age of air, which are used for evaluating ventilation in public places, are not commonly
used to evaluate underground ventilation efficiency. Parra [41] points out that the mean age of air
and local levels of pollutants’ concentration in risk areas are better factors to qualify the
ventilation quality. He used a validated CFD model to evaluate the effectiveness of three
ventilation systems in deep mines: exhaust, blowing, and mixed, by analyzing the dead zones
and the local mean age of air. The Spalart-Allmaras turbulent model was used and Navier Stokes
equations for a three dimensional, steady, incompressible and isothermal flow are solved in the
model. Dead zones are regions where velocity is below the regulated minimum velocity, but the
dead zone criterion does not take into account the flow recirculation. The local mean age of air,
̅ , is obtained by solving Equation 2-23, [41]
̅ ̅
{( ) } ( 2-23 )
Where is the i component of the mean velocity, is laminar viscosity, is turbulent
kinematic viscosity, is laminar Schmidty number, and is turbulent Schmidty number. Low
air mean age indicates fresh air. The global efficiency, as shown in Equation 2-24, is used to
compare different ventilation systems. An efficiency value of 1 represents a perfect displacement
flow, and a value of 0.5 represents a perfect mixing flow.
̅
( 2-24 )
̅
Where ̅ is local mean age in the outlet section and ̅ is total local mean age.
Xicheng et al. used the same criterion, the dead zone and age of air, to study the
effectiveness of the push-pull auxiliary ventilation system [77]. They point out that the effective
range of a semi-confined jet can be determined by an equation, but there is no theoretical or
experimental equation available to calculate the effective range of an exhausting duct due to its
complexity. Thus CFD modeling is an approach to study it. The air age was calculated using
Equation 2-25, where is velocity, is the average air age, is density, is laminar viscosity,
is turbulent viscosity.
{ ( ) } ( 2-25 )
By examining the percentage of dead zones and the mean age of air of four different models, it
was concluded that once the forcing duct position is determined, there is an optimum position for
the exhausting duct in order to achieve the best efficiency.
26
|
Virginia Tech
|
Aminossadati et al investigated the effects of brattice length on fluid flow behavior in the
crosscut regions [78]. CFD-ACE (ESI Software) was used in their study and k-ε turbulence
model was employed. The results were compared with the results of FLUENT. The study
indicated that airflow into the crosscut region was improved due to the use of brattice.
CFD can also be used to evaluate the fan effectiveness. Konduri et al. used a two-
dimensional CFD model to simulate a jet fan for auxiliary ventilation and obtained similar results
with the experiments [64]. Ray et al. used CFD to simulate the performance of vertically-
mounted jet fans in a ventilation shafts [79]. The simulated results were compared with the
calculated results, with validation using field measurements left as future work.
Mining is a dynamic process and airflow changes are associated with advance and retreat
and the subsequent changes in mine geometry. However, it is particularly difficult to model a
time dependent mining step together with the airflow simultaneously using CFD. Hargreaves et
al. [67] use a series of steady-state computational models to represent the different stages of a
tunnel drivage cutting cycle in order to assess the effectiveness and ventilation flow patterns of
the force and machine mounted scrubber auxiliary ventilation system. The cutting cycle was
decomposed into several representative steady-state stages and 24 simulations were carried out to
replicate the whole cutting cycle. The simulation results were compared with the full scale
ventilation experimental data. This study shows that CFD modeling can improve the
understanding of auxiliary ventilation systems during different stages of cutting cycles and the
results can be used to improve the planning and operation of auxiliary ventilation systems. The
results of this study were later in conjunction with Virtual Reality (VR) technology to develop an
improved ventilation planning and training tool [11].
2.6.2 Spontaneous combustion
Spontaneous combustion often occurs in the gob areas where events are difficult to locate
and extinguish [8], and large-scale field experiments for the purpose of studying spontaneous
heating in underground mines are particularly difficult[80].
Yuan conducted a series of large-scale CFD numerical modeling studies on spontaneous
heating. He studied spontaneous heating in typical long wall gob areas with bleeder and
bleederless ventilation systems with a stationary longwall face [18], [21], [81]. The estimated
gob permeability and porosity profiles from a geotechnical model were used as inputs for the
27
|
Virginia Tech
|
CFD model using FLAC (Fast Lagrangian Analysis of Continua). The Kozeny-Carman equation,
as shown in Equation 2-26, was used to estimate the changes in permeability in the caved rock.
In the equation, n is the porosity and k is the permeability. The flow in the gob area is treated as
laminar flow while fully turbulent flow was applied to ventilation airways. The studies showed
the flow patterns inside the gob, and the effect of gob permeability, pressure at the bottom of
bleeder shaft, resistance at collapsed entries, nitrogen injection, apparent activation energy, coal
surface area, and critical velocity zones for spontaneous combustion were studied. The results of
the CFD studies were reasonable according to experience and data from previous experiments
and studies.
( ) ( 2-26 )
The effect of barometric pressure changes on spontaneous heating in longwall panels was
presented in another article by Yuan and Smith [19]. The actual recorded barometric pressure
variations were used in a bleederless ventilation model and the oxygen concentrations were
quantitatively examined. Results showed that the barometric pressure change will influence the
maximum temperature of the spontaneous heating in the gob, although the influence is not
significant. However, the influence was affected by the gob permeability and the coal oxidation
rate.
Another study by Yuan and Smith examined spontaneous heating in a coal chamber
utilizing CFD [80]. The results were validated by comparing results with a test from U.S. Bureau
of Mines experiments and the results indicated similar phenomena. Theydemonstrated that the
CFD model has the ability to reasonably reproduce the major characteristics of spontaneous
heating in agreement with experimental test results and that the model is useful for predicting the
induction time, which is key for prevention of spontaneous heating fires.
2.6.3 Mine fire
Mine fire is another challenging underground mine safety issue. The toxic gases,low
visibility, and open flame caused by fires create a hazardous environment underground. Miners
can be seriously injured by inhaling toxic products-of-combustion (POC), and the fire heat can
cause rib and roof collapse [44], [82]. Underground coal fires also produce large amounts of
CO2; for example, in one study in China, nearly 100 to 200 million tons of coal affected by
underground coal fires were calculated to produce 2-3% of the total world CO2 emission [83]. A
28
|
Virginia Tech
|
number of studies are related to the investigation of mine fire and its combustion products
utilized CFD.
A CFD study was conducted by National Institute for Occupational Safety and Health
(NIOSH) and Mine Safety and Health Administration (MSHA) to investigate the temperature
characteristics of mine fire [84]. The model was built using Fire Dynamics Simulator (FDS),
which is a CFD program developed by National Institute of Standards and Technology. The
model was built according to the deep seated coal fire test and the results will be used in the
follow-up fire experiments with remote fire suppression applications.
Edwards conducted some CFD studies to understand fire and smoke spread. A model was
made using FDS to simulate the 1990 fire at Mathies Coal mine [44]. The coal lined tunnel flame
spread rate was studied and they showed that it is not sensitive to the heat of pyrolysis but very
sensitive to the coal moisture content. The model also studied the flame spread in a tunnel lined
with Douglas Fir timber sets and along a conveyor belt.
Another model was built using CFD2000 to model buoyancy induced Product-Of-
Combustion (POC) spread from experimental fires in the laboratory and to analyze smoke flow
reversal conditions. The simulated POC spread rates and gas temperatures were higher than the
measured values. The reverse flow condition model had lower predicted critical velocity than
predicted by a Froude model analysis. The study illustrated the limitations of CFD models with
incomplete experimental conditions [85].
Ventilation control is a recommended method for the control of POC and smoke reversal,
but a quantified ventilation strategy is usually not available. Edwards conducted an experiment
and computational model to determine the critical ventilation velocity required to prevent smoke
reversal [82]. Fire smoke reversal experiments were conducted with different fire intensities and
it was determined that the critical velocity to prevent smoke reversal is proportional to the fire
intensity to the 0.3 power which is in agreement with the one-third law dependence theory posted
by other researchers [86]. The CFD model using FDS showed good agreement with the
experimental results, and provided a predictive method to simulate a range of fire intensities and
mine entry dimensions which is difficult to achieve experimentally.
Huang and others presented a CFD method using a two-dimensional model which is
based on the theory of natural convection and heat transfer in porous media to study the flow and
temperature fields in underground coal fires [83]. The solutions compared well with limited
29
|
Virginia Tech
|
available field data. The results showed that the fractures or high permeability are important
factors to enhance natural convection. In a uniform permeable stratum, air flows from the low
temperature zone to the hot area, but in a non-uniform permeable stratum, air flows from the
more permeable zone to the hot area and less permeable zone. The study also found that air
convection influences shallow coal seam fires more than deep coal seam fire and the gas
produced by secondary combustion in fractures can enhance the convection.
2.6.4 Methane flow and control
Methane in underground coal mining is a major safety issue. Methane is highly explosive
under certain concentrations and requires constant monitoring and control to maintain a safe
working condition. The gas flow in the gob and mine ventilation systems are complex and
difficult to measure. CFD can be used to better understand complex underground methane flow
and design ventilation methods to reduce the methane risks[87].
Ren et al. presented a CFD modeling study of methane flow around longwall coal faces
[88]. Due to the fact that methane to the working longwall face may be from source beds above
or below the working seam, this study constructed a model that included a methane bearing seam
80 m above the working seam. Laboratory results were used for the permeability values of the
roof strata, as well as the consideration of redistribution of stress field and the mining induced
fractures. The pressure and velocity contours were provided, which related to the methane
emission and migration. Although the CFD model provided practical results, validation from
field data is needed.
Toraño conducted CFD analyses of methane behavior in underground coal mine auxiliary
ventilation [59]. The conventional method calculates the average methane concentration without
considering different methane content in different zones. The study aims to analyze the evolution
of ventilation in different cross sections and in the roadway axis directions, and the influence of
time. A CFD model using Ansys CFX 10.0 and field experiment were carried out to study the
dead zone, airflow recirculation, and methane distribution. The study compared four different
turbulence models and selected the k-epsilon model which agrees with the field measurement
best. Both CFD and experiment results were compared with those calculated by conventional
methods. The study shows that it is necessary to analyze auxiliary ventilation systems by CFD
which helps identify potential dangerous zones and auxiliary ventilation design.
30
|
Virginia Tech
|
The coal mine production may be prohibited by the high methane content underground
when the ventilation is not sufficient to lower the content with normal ventilation systems. Oraee
used CFD to simulate a methane drainage system which can be used to reduce the ventilation and
development cost in a gassy mine [89]. The study evaluated the methane drainage system with
different vent hole spacing and the change of methane content with time. The study showed that
the CFD model can be used to improve the drainage system design to more effectively manage
methane underground. Balusu et al. presented an extensive study on the optimization of gob
methane drainage system [90]. Several techniques were used during the course of the project,
such as on-site monitoring, tracer gas tests, CFD simulations, and extensive field trials. The CFD
method was used to analyze the gas flow and buoyancy mechanisms in the gob. The models
were validated and calibrated using the field study results. The influence of different parameters,
such as face flow rate, drainage hole position and spacing, are extensively investigated. The CFD
results, in combination with other field investigations were used to develop optimum gob gas
control strategies. The gas drainage strategies developed by this study gained about 50% gas
drainage improvement compare to the traditional gob gas drainage strategy and greatly enhanced
the safety and productivity of underground coal mines.
2.6.5 Gob gas flow
It is important to understand the mechanics of gas flow inside the gob in order to develop
effective gas management and ventilation strategies [91]. It is hard to measure the air flow inside
the gob because much of the gob area is inaccessible. Therefore, the CFD modeling technique is
one reasonable way to investigate the ventilation in gob areas [18].
Permeability distribution in the gob is a key element of the gob gas flow model [91].
Esterhuizen and Karacan developed a methodology for calculating permeability variations in the
gob suitable for reservoir models or CFD models and simulated the leakage flow into the gob,
methane distribution within and effects of gob vent boreholes on flow patterns [92]. The
permeability changes were determined using FLAC3D numerical modeling program and the
results are used as input into the reservoir model. The simulated results were compared to
empirical experience and measurements and are consistent with empirical observations and
measurements reported in the literature.
31
|
Virginia Tech
|
A similar study was conducted by NIOSH [8]. The flow patterns inside the gob under
one-entry and two-entry bleederless systems, and a three-entry bleeder system were studied
using CFD. The gob permeability data were from the results of FLAC geotechnical modeling.
The study also discussed the possible location of critical velocity zones which support
spontaneous combustion.
2.6.6 Inertisation
The goal of gob inertisation is to lower the risk of potential explosions during longwall
panel sealing off periods. Gob inertisation has been widely used around the world to control fires
and spontaneous heating in underground coal mines. Effective inertisation can suppress the
development of potential gob heating and maintain a normal coal production rate [93]. High wall
systems have also effectively utilized inert gas to maintain safe methane levels [94].
Balusu et al. and Ren et al. presented their work using CFD to study the optimum
inertisation strategies which can achieve gob inertisation within a few hours of the sealing the
panel [93], [95]. The CFD model was first calibrated based on previous inertisation studies and
gob gas monitoring. The gob porosity parameter was from the results of geomechanics models.
The gob conditions before sealing off period were modeled by steady state modeling and then
the sealed gob atmosphere was modeled by transient modeling. The validated model was used
for parametric studies such as inert gas injection locations, inert gas flow rates, seam gradients,
and different inertisation strategies such as injection of inert gas through surface gob holes.
Results show that the inert gas composition is not the major factor in an inertisation process and
that injection of inert gas at 200m behind the face is more effective than at the location right
behind the face line. Several recommendations were provided to improve the inertisation strategy
and the strategy developed by this study was implemented and demonstrated in the field. The
new practice showed significant improvement by converting the gob into an inert atmosphere in
a few hours instead of two to four days by traditional methods.
Trevits and others conducted CFD modeling using FDS to study the effects of increased
inert gas (N ) injection [96]. The results of the model achieved good correlation with the field
2
test which used the pressure swing adsorption (PSA) N generation technology to inert a mine
2
sealed area. The results showed that the relationship between N gas injection rate and the time
2
needed to reduce the O level is not linear and the benefit of inert gas decreases as the injection
2
32
|
Virginia Tech
|
rate increases. The CFD results also showed that injection of N at two ventilation seals is more
2
efficient than at one seal location with double the injection rate.
Mossad et al studied the effectiveness of high wall mining inertisation using CFD [94].
The study focused on improving mine efficiency with regard to safety and production rates by
using inertisation to maintain methane concentrations within safe working limits. The model is a
2-D k-ε realizable turbulent model and the Semi-Implicit Method for Pressure-Linked Equations
algorithm was chosen for the velocity pressure coupling. The study indicated that applying the
inert gas at high angles of injection is more effective and CO2 is the most effective gas, when
applied at a 60 degree angle, compare to N2 and Boiler Gas. This work was described in detail in
Vella’s dissertation [97].
2.6.7 Dust dispersion and control
The amount of dust generated during mining is another major concern. Dust can cause
respiratory disease, contribute to the risk of underground explosion, and impede productivity
[98]. The airflow and dust dispersion are very complex and the standard mine ventilation
network analysis is not sufficient to analyze the detailed airflow patterns and dust distribution.
CFD is an attractive approach to develop and evaluate dust control methods.
Heerden and Sullivan completed a CFD study to evaluate the dust suppression of
continuous miners and roadheaders [98]. The study showed the steps during the CFD model
constructions, and plotted the velocity vectors and contours of the results. The dust particles are
assumed to follow the flow in the flow field, and the slow lines were used for qualitative
assessment of dust movement. The model was used to evaluate dust suppression under different
machine parameters and dimensions, such as the position of the continuous miners, the volume
of the flows, and different models of roadheaders. The effect of drum rotation, water sprays, and
air movers were also investigated. The methane concentration was added to the model later. The
authors indicated that the model was validated by comparing with the experimental data, but no
details were provided for the validation.
Srinivasa et al. studied airflow and dust dispersion at a typical longwall face using CFD
[99]. The study evaluated the air curtains, semi-see-through curtain and air powered venturi
scrubber dust control techniques. The effect of support legs and the shearer were also modeled
with simplified geometry. The dust is assumed inertialess and follows the air flow streams. The
33
|
Virginia Tech
|
dust was added in the model as a dust source at the model inlet and was assumed to be constant
and uniform across the inlet. The simulations were performed using Fluid Dynamics Analysis
Package (FIDAP) program. The flow field equations were solved independently from the
pollutant equation. The dust particles were calculated using Equation 2-27, where F refers to
fluid phase and P to the particle phase.
( 2-27 )
Where: = density, = time, = velocity, = body force, = gravity force, was given as:
( 2-28 )
Where = dynamic viscosity, = drag coefficient, = particle diameter.
The trajectory equation was given as the follows:
( 2-29 )
The advection-diffusion equation for the dispersed phase of dust particles was given as:
[ ] ( 2-30 )
Where, = dust concentration, = mass diffusivity, = source term.
The air velocity results and the dust concentration values using air curtain were compared
with field measurement. The predicted dust concentration was within 10% of the field values.
The simulation indicated that the air powered venturi scrubber is the most effective means to
control dust, with a 40-50% reduction within a distance of 3-4 m from the scrubber at 2.1 m/s
face air velocity. The study concluded that CFD can be used to model underground dust
dispersion and design dust control techniques [99].
Skjold et al. reported a CFD code DESC (Dust Explosion Simulation Code) which is a
simplified empirical based CFD code that can be used to simulate the dust lifting phenomenon
[31]. The empirical approach was used for the DESC code since the detail dust lifting
mechanisms, such as the Magnus forces, Saffman forces, and particle collisions, cannot be
feasibly modeled and is suitable for industrial applications. The DESC code is similar with the
CFD code FLACS (Flame Acceleration Simulator). The dust particles are assumed to be in
dynamic and thermal equilibrium with the fluid phase. The phenomena such as dust settling or
flow separation in bends and cyclones cannot be modeled because slip velocity is not included in
the code. The study simulated dust concentration for an experimental wind tunnel, as well as a
set of dust explosion experiments described in literature. Although the experiment technical
34
|
Virginia Tech
|
details were limited, the simulated dust layer results agreed well with the experimental data. As a
result of this study, Skjold et al. concluded that it is beneficial for the safety of coal mines or
other industrial field to use a simplified dust lifting model.
Ren et al. presented their work using CFD to develop a new dust control systems [100].
The geometry of their models were comprehensive, including not only the coal face and the
maingate, but also chocks, shearer, spill plate, BSL/crusher and conveyor, dust scrubbers, shearer
clearer, venturi sprays, and curtains. One particular example they showed is the use of CFD
modeling to design a new shearer scrubber system. By studying parameters such as the location
of the inlet and outlet, the capacity of the scrubber, and the face airflow rates, the study indicated
that positioning the scrubber inlet towards face ventilation can capture more dust particles. The
CFD modeling results were used to design a new shearer dust system, which achieved 43% to
56% more dust reduction.
Silvester et al. [33] presented a CFD study on the influence of underground mineral
tipping operations on the surrounding ventilation system and consequent dispersal of fugitive
dust. It used a two-phase continuum approach to describe the interaction between the falling
materials and the surrounding air. The standard k- model was used and wall roughness effects
were not considered since they were proven to have negligible impact on subsurface mine
ventilation modeling [101]. Different scenarios were modeled to investigate the influence of
different factors. A Lagrangian particle tracking algorithm was used to represent the dust flow
and plume dispersion. The CFD results were validated against experiments using scale models,
which used water as a substitutefor air to achieve adequate dynamic scaling, and used a dye
injection system to visualize the flow. Good qualitative agreement was achieved between the
experiment and the CFD results. However, the use of continuum dynamics to represent the
material as a granular fluid medium restricted the CFD model because they could only achieve
an approximation of the actual process and could not reveal the mechanisms of the process.
Another CFD study conducted by Silvester et al. on the dust control aspect was presented in
[102]. The dispersion and deposition of fugitive mineral dust generated during mining at a
surface quarry were studied. The influence of the mineral dust emission location, the wind
direction, and the in-pit ventilation flows were investigated, and the results can be used to assist
future quarry planning and blast operation to better control the dust emissions. However, the
CFD models were not validated.
35
|
Virginia Tech
|
Torno et al. developed a CFD model to simulate the dispersion of dust generated in
blasting in limestone quarries [25]. The standard k- model was used for turbulence modeling
and a Lagrangian particle tracking method was selected to model the air and dust multiphase
problem. The CFD model was validated by the experimental data using a trial-error method on
the value of dust injection. It has been proved that the use of a barrier placed downstream of
blasting can create 4.5% of dust emission retention.
2.6.8 Minerals processing
CFD has also been extensively used in recent years in the process industry for the
research and development of new and existing processes. CFD modeling results can help
researchers to gain detailed understanding of flow during minerals processing that can be used to
design and modify equipment to improve separation performance. Numerous studies are
available in the literature, but only two representative examples are shown here.
An initial study was presented by Lichter et al., which use the combination of CFD and
Discrete Element Modeling (DEM) to evaluate the performance of flotation cells [103]. The
CFD was used to simulate a flotation machine with different parameters, such as the size of the
flotation cells and inlet velocities. Slurry was treated as single phase newtonian fluid with
specified viscosity. The model did not include the air in the slurry system, but the author states
that it still can be used to compare one cell design with another. The CFD results were then
imported to a DEM simulation, which makes it possible to produce residence time distributions
as a function of size and evaluate the metallurgical performance. No final conclusion was made
on the relationship between parameters and the performance of the cell. However, this study
showed the potential of the combination use of CFD and DEM modeling to determine the
flotation cell operating and design parameters.
Peng used CFD to model the hindered-settling bed separator, which is used for size
classification or relative density separation [58]. Most studies of the separation mechanisms are
based on density and size difference without the particles-liquid interactions. Peng used the
Euler-Lagrange CFD approach in her study can model the physical effects influencing the
particle motion and predict liquid velocity profiles and solid particles movement. The 2-D model
was validated by comparison of the CFD results and the actual plant test. The flow pattern, effect
of feed system, and effect of operation parameters were investigated and discussed which
36
|
Virginia Tech
|
provided valuable information to better understand the detail separation mechanisms and to
predict and optimize the separation process.
2.7 Other applications
Applications of CFD have been found in other design and research areas of mining
industry, and some of them are reviewed here.
Toraño [23] used CFD to model different shapes of open storage systems for bulk
materials, such as coal and iron ore, to study the best operational and investment parameters to
reduce the airborne dust. The US EPA established a methodology to estimate the level of
airborne dust generated from an open pile. However, sometimes the existing methodologies or
standards do not match the way materials are actually stored due to reasons like area restrictions
and stacking means, and it is also not easy to carry out experiments to evaluate the level of
airborne dust to compare to the standards. Therefore, CFD could be used to predict the different
environmental impacts of various storage piles. Toraño used the commercial CFD software
Anysys CFX 5.7 to conduct the simulation. The model was validated by comparing results for
cone and flat top oval piles with the US EPA study, and then a semicircular pile model was built
and analyzed. Reynolds Averaged Navier–Stokes method and medium complexity turbulence
models were used. The model showed good agreement with the US EPA study with low root
mean square values. The semicircular pile model showed a lower emissions and wind erosions
level, but the wind direction is an important factor that would influence the results.
Berkoe et al. [46] presented several projects they have done which applied CFD to the
mining and metals field. These include quench cooler design to reduce process gas temperature,
solvent extraction settler design to achieve uniform flow, plume capture performance prediction
for different configurations of a fugitive emissions collection system, effect of wind on
operations facilities, performance of a ferronickel smelting furnace, and slurry flow distributor
design. Most of the models used the FIDAP software analysis package, except the slurry flow
distributor design used the FLUENT CFD and the discrete phase model. They especially
highlighted that the CFD study requires deep understanding of the underlying physics, and
usually needs to apply simplified assumptions and improve boundary conditions. Therefore, it is
important to have an person who can interface between the field engineering and the CFD
modeling functions to obtain reliable results.
37
|
Virginia Tech
|
CFD has also been used to study the water pollutants associated with mining. Doulati et
al. [104] used PHOENICS to study the acid mine drainage generation and subsequent pollutants
transportation. The chemical reaction process model was implemented to PHOENICS by
subroutines. Close agreements were achieved between CFD model and filed data. This study
illustrated the ability of CFD to study the groundwater pollution problems and better understand
pollution transport mechanisms.
CFD is also a promising method in the field of carbon capture and storage (CCS)
technology. Mazzoldi et al. [34], [105] presented a risk assessment study for CO transportation
2
using a commercial CFD software Fluidyn-PANACHE. In this study, models were built to
simulate an accidental release of CO from high pressure transportation facilities within CCS
2
projects. The results were compared with those using Gaussian/dense-gas models and they
demonstrated that CFD models are more reliable and produce more precise results, thus can
provide improved risk analyses. A similar study conducted by Dixon et al. [106] used the CFD
code CFX to predict the consequences of releases of CO from a liquid inventory. The
2
concentration of CO particles was modeled using both scalar equation and Lagrangian particle
2
tracking methods and good agreement with experimental data were observed.
2.8 Conclusions
The CFD concept, its application in the mining industry, specifically in ventilation, and
challenging issues have been discussed in order to provide insight into the current CFD research
activities in mining. It is evident in this review that the scope and the level of sophistication of
CFD studies in mining are increasing, especially with the continued high rate of advancement in
computer power. The application of CFD in the mining industry will allow for improved
understanding of the fluid problems that can enhance safety and optimize layout and equipment
design.
Turbulence models are widely used since most flows in mining are turbulent. It is clear
that the standard k-ε model has been commonly used as the most acceptable general purpose
turbulence model. However, the quality of the solution is dependent on the turbulence model.
Therefore the selection of turbulence models should consider physical models and flow features
of specific problems.
38
|
Virginia Tech
|
Mesh independence studies must be included in the construction and analysis of a CFD
model. This paper summarized the general procedures and methods to conduct the mesh
independence study, which assure that solutions are not significantly dependent on the mesh.
Because CFD uses approximate approaches and some assumptions, validation of the CFD
studies is necessary to ensure the simulated results are within an acceptable level of accuracy.
The validation studies are generally conducted by comparing the results obtained from laboratory
or full scale experiments with the simulated results. Several techniques were used in the cited
work for CFD validation, such as tracer gas, 3D velocity measurement, and flow visualization.
Accurate measurement of flow features may be difficult due to the complexity of the flow
domain, such as underground mine working faces and flow in the gob. General agreement with
the experimental data was reported in many validation studies, whilst discrepancies were also
noted in some studies, which indicated a requirement for model improvement and accurate
measurement of experimental flow parameters.
Overall, this paper reviewed the current state of research of CFD modeling in mining.
Examples discussed in this paper and numerous studies that can be found in the literature showed
that the potential benefits from the CFD simulations are enormous if the problem setup is
addressed carefully and proper model verification, such as mesh and solution convergence, and
model validation are conducted.
39
|
Virginia Tech
|
This paper was presented at the 2011 SME annual meeting in Denver, and is included in the meeting
preprints (Feb. 27-Mar. 02, 2011, Denver, CO, Preprint 11-121). Guang Xu conducted the majority of the
experimental and CFD modeling work and wrote the paper with technical and editorial input form coauthors: John
R. Bowling, Dr. Kray D. Luxbacher, and Dr. Saad Ragab. Please cite this article as: Xu, G., Bowling, J. R.,
Luxbacher, K. D., & Ragab, S. (2011). Computational fluid dynamics simulations and experimental validation of
tracer gas distribution in an experimental underground mine. 2011 SME Annual Meeting (p. Preprint 11–121).
Denver, CO (USA).
3 Computational Fluid Dynamics Simulations and Experimental
Validation of Tracer Gas Distribution in an Experimental
Underground Mine
3.1 Abstract
Following a disaster in a mine, it is important to understand the state of the mine damage
immediately with limited information. Computational fluid dynamics can be used to simulate
and ascertain information about the state of ventilation controls inside a mine. This paper
describes a simulation of tracer gas distribution in an experimental mine with the ventilation
controls in various states. Tracer gas measurements were taken in the lab experimental apparatus,
and used to validate the numerical model. The distribution of the tracer gas, together with the
ventilation status, was analyzed to understand how the damage to the ventilation system related
to the distribution of tracer gases. This study will be used in future research in real mine
measurements to compare collected and simulated profiles and determine whether damage to the
ventilation system has been incurred during an emergency situation, the nature of the damage
and the general location of the damage.
3.2 Introduction
There is a lack of knowledge about the state of ventilation controls in a mine following
the event of a significant incident such as a roof fall, bump, or explosion which requires
immediate action. Currently, some information may be gathered safely from the surface, but
most information regarding the state of the ventilation controls cannot be known before rescue
personnel enter the mine. Having quick access to more information will help decision makers to
more effectively manage a mine emergency and increase safety for rescue personnel.
It is essential to model ventilation patterns and the mine environment following an
incident in a mine. Tracer gas techniques and numerical simulations using computational fluid
dynamics (CFD) can be used to ascertain and simulate information about the state of ventilation
controls inside a mine. Tracer gas measurement is an effective method to detect air flow routes
41
|
Virginia Tech
|
and estimate air flow quantity and the rates of dilution and dispersal of contaminants in
underground mine ventilation systems [107], [108]. Air flow directions and quantities can be
estimated by analyzing the tracer gas concentration. Dispersion of tracer gas in underground
ventilation system may be very different depending on the location of damage after incident.
The use of tracer gases started in the 1950s in building ventilation systems [61]. Tracer
gas techniques have been used in many situations where the standard ventilation survey methods
are inadequate [109]. The applications of tracer gases in underground mines include analyzing
ventilation patterns, measurement of air leak rates, and evaluating dust control measures [62].
Sulfur hexafluoride (SF ) is widely used as a tracer gas and is ideally suited for use in the
6
underground environment. SF is not normally found in the underground environment and it is
6
inert, nonflammable, nonexplosive and non-toxic which makes it safe for use in underground
mining and other industrial environments. Most importantly, current technology makes it
possible to detect very low concentrations of SF (in the parts per billion or trillion range) [2],
6
[110].
Computational Fluid Dynamics is a tool which can approximate numerical solutions in
cases where experimental solutions are impractical or impossible. With the recent advances in
computer technology and the success of CFD，the application of CFD has become increasingly
attractive in modeling the ventilation systems in underground mines. It has been used in
simulations of explosions [31], methane control [94], [111], ventilation system improvement [41]
, gob inertisation methods [93], and spontaneous combustion and mine fires [20], [85].
A combination of experimental data and a CFD model of tracer gas dispersion has been
used to study airflow and contaminant transport in indoor environments [112–114], pollutant
dispersion [115], and other industrial applications. Little research has been done to simulate
tracer gas dispersion in underground mines, especially using tracer and CFD simulation to
predict the status after emergency in underground mines.
This paper presents both the experimental and numerical results of ventilation status and
tracer gas (SF ) dispersion in an experimental laboratory scale mine model with the ventilation
6
controls in various states. Valves are used in the experimental mine model to simulate different
ventilation statuses after “incidents” cause changes to the ventilation. Several passive area
sources with constant emissions of a tracer gas (SF ) were designed to simulate constant tracer
6
injections in the experiment. The objectives of the experiment were to collect data for evaluating
42
|
Virginia Tech
|
the influence of different locations of damage after incidents and to validate the CFD model.
This study indicates that tracer gas concentrations in a mine can be accurately modeled with
prior knowledge of the ventilation system. It is the first step toward the research using tracer gas
measurements to compare measured and simulated profiles and determine whether damage to the
ventilation system has been incurred during an emergency situation, the nature of the damage
and the general location of the damage.
3.3 Experimental measurements
3.3.1 Experimental apparatus
A simple typical mine layout was designed for experimental purpose. As is shown in
Figure 1, it includes one gob panel, one active panel, one stopping, and two regulators. Three
possible incidents locations are also designed including explosion damage to the stoppings and
causing short circuiting of the airflow between the main entries, a roof fall in the active panel
which will block the airflow across the working face, and an explosion in the gob which will
block the airflow through the gob. Two boreholes are present: one rescue borehole on the tailgate
of the active panel and another borehole in the gob. The normal air flow paths are also shown in
Figure 4. The experiment is not set up to include flow through the gob, simply around it.
The experiments were conducted using the experimental underground mine shown in
Figure 5, which is built according to the mine layout shown above. The experimental
underground mine is composed of 2 inch (0.0508 m) inside diameter PVC pipes with the
maximum dimensions of The PVC pipes were labeled with numbers for convenient reference.
The experimental system has one intake and the exhaust is hooked up to an exhaust fan shown in
Figure 6. Five valves (Figure 7) were used to simulate the stopping within the main entry,
regulators, and roof fall/explosion damage.
43
|
Virginia Tech
|
3.3.2 Experimental procedure
Although the five valves’ states can be changed to simulate different situations and
ventilation statuses, only two experiments were performed in this work. Case #1, defined as
having only valve 1 closed, simulates normal ventilation status with the air flow paths shown in
Figure 1. Case #2, with all the valves open, simulates the situation in which an explosion has
damaged stoppings in the main entries. Air flow becomes short circuited due to the damage so
that relatively little air reaches the panels, most intake air flows directly from the intake entry,
through the crosscuts where stoppings were damaged, and is exhausted directly. The air flow
paths in case #2 were shown in Figure 9.
Figure 9. Flow path of case 2 after the stopping was damaged by explosion
Before releasing SF , the exhaust fan was turned on allowed to run until the flow reached
6
a steady-state, marked by the air velocities no longer changing. The tracer gas was released just
inside the inlet at a constant rate of 1 liter per minute. Air samples were taken after ten minutes
had elapsed while tracer gas was released to ensure a stable airflow and tracer gas distribution.
Air samples were drawn through septa at four different sample points which are shown in Figure
1 and for each location three measurements were repeated.
46
|
Virginia Tech
|
the standard k-ε turbulence model were employed to predict the incompressible turbulent airflow
and user-defined scalar transport without chemical reaction and heat transfer was performed to
predict tracer gas dispersion. The model was selected because it achieves reasonable accuracy
over a wide range of turbulent flows in industrial flow simulations.
The inlet and the outlet of the model were specified as velocity inlet and pressure outlet,
respectively. 403.38 Pa gauge pressure was applied to the outlet according to the experimental
measurement. 18.0 ft/s (5.5 m/s) and 22.0 ft/s (7.0m/s) were applied to the inlet for Case #1 and
Case #2, respectively. All of the other surfaces are treated as stationary walls with no slip. Both
air and wall temperatures are assumed constant.
The numerical simulations in this study were conducted using the commercial CFD
package, ANSYS FLUENT 12.1, to simulate the airflow and tracer gas dispersion for the same
scenarios used during the laboratory tests. A first order upwind scheme was used for variables
including pressure, momentum, turbulent kinetic energy and turbulent dissipation rate.
Discretized airflow equations were solved with the SIMPLE algorithm in the CFD program to
couple the pressure, velocity, momentum and continuity equations.
3.5 Results and analyses
Air velocities were measured at four sample points and were used to calibrate the CFD
model. Table 2 shows the measured and simulated velocities. Generally the computed airflow
velocity agreed qualitatively with the experimental data. However, obvious errors exist in
quantitative comparison. For example, in Case #2, simulated velocities at Point 2 and Point 3 are
less than the measured data at the respective points. As we know in this case, the airflow was
short-circuited, so the velocities at Point 2 and Point 3 should be small. We can conclude that it
is very possible the measured velocity is not accurate. The difference between measured and
simulated data may be mainly caused by three factors: the precision and error of the differential
pressure transducer, the leakage of the experimental model, and the boundary conditions of the
computer model. Since the study is the first step of the project, the data are accepted for now
before further improvements are made.
SF was only used in Case #2. Air samples were taken three times at each sampling
6
location and the average concentrations were calculated to compare with the simulated result.
During the experiment, SF was released at a constant rate of 1 L/min through a ¼-inch inside
6
diameter tube and placed 10cm inside the air inlet. In the computational model, SF was released
6
48
|
Virginia Tech
|
from a point source (¼-inch cube) at the same location as the experiment, with a mass flow rate
of 401 kg/m3*s which is equal to the 1L/min SF flow rate. For Case #1 the measured SF
6 6
concentration is not available, but computer simulation was conducted.
Table 3 shows the measured and simulated SF concentrations for Case #2, also shows
6
the simulated SF concentration for Case #1. There are differences between measured and
6
simulated SF concentration. The measured results are generally larger than the simulated results.
6
This is probably due to absorption of SF to the PVC pipes, although the parameters and
6
boundary conditions used to simulate SF also need calibration. Figure 12 shows the SF
6 6
distribution at a cross-section of Sample Point 1 in Case #1 and Case #2. From the contours one
can see the CFD model can compute the diffusion of tracer gas and visualize the distribution.
Because in Case #1 the velocity at the inlet is less than that of Case #2 (5.5 m/s and 7.0 m/s
respectively), SF was diffused less in Case #1 than in Case #2 and has a different distribution
6
over the cross-section.
Table 2. Measured and simulated velocity at four sample points
Point 1 Point 2 Point 3 Point 4
Measured 6.9 m/s 3.8 m/s 3.7 m/s 7.0 m/s
Case 1
Simulated 6.8 m/s 3.6 m/s 3.5 m/s 7.0 m/s
Measured 8.2 m/s 2.0 m/s 1.9 m/s 8.5 m/s
Case 2
Simulated 8.3 m/s 0.4 m/s 1.5 m/s 8.8 m/s
Table 3. SF measured and simulated concentration
6
Point 1 Point 2 Point 3 Point 4
Test 1 3.00mg/L 7.41mg/L 6.81mg/L 9.61mg/L
Test 2 3.67mg/L 3.00mg/L 4.80mg/L 6.31mg/L
Case 2 Test 3 6.33mg/L 6.65mg/L 6.65mg/L 6.96mg/L
Average 4.67mg/L 5.69mg/L 6.09mg/L 7.40mg/L
Simulated 3.60mg/L 4.00mg/L 4.00mg/L 4.00mg/L
Case 1 Simulated 4.60mg/L 5.10mg/L 5.10mg/L 5.10mg/L
3
Case #1 Case #2
Figure 12. SF Distribution contours in the CFD model in cross-section at sample point 1
6
49
|
Virginia Tech
|
3.6 Conclusions and future work
This study investigated airflow and SF transport in an experimental coal mine through
6
both experimental measurements and numerical simulation with CFD under two different cases
(airflow patterns). An experimental coal mine, based upon a simple typical mine layout, was
built using PVC pipes. Pitot tubes, differential pressure transducers, a computerized data
acquisition system, and a gas chromatograph were used to measure the air velocity and tracer gas
distributions throughout the simulated mines. The numerical simulations used CFD with the
standard k-ε turbulence model and user-defined scalars to simulate airflow and tracer gas (SF )
6
distribution.
Measured data were used to calibrate the CFD model and the simulated results were
compared with the measured results. The velocities and the SF diffusion results were acceptable
6
while there are differences between the computed and measured results. Errors exist in both the
physical experiment and the CFD model and further experimental improvement and validation of
CFD model are needed.
The present study is the first step toward research intending to use tracer gas
measurements to compare measured and simulated profiles and determine whether damage to the
ventilation system has been incurred during an emergency situation as well as the nature and the
general location of the damage. Results showed that the methods used are feasible although
improvements are needed. Further work will include: (1) Experimental measurement validation
and design improvement including calibrating the velocity measurement results, controlling and
analyzing the errors from the differential pressure transducers, and improving the location of
velocity measurement. (2) The PVC pipes may also need to be replaced with a material that is
less prone to SF adsorption. (3) Further calibrating the CFD model, especially the boundary
6
conditions, diffusivity of SF , mass flow rate of SF . Also, it may be helpful to using the second
6 6
order upwind scheme to achieve more accurate results. (4) Studying more cases under different
airflow patterns to find the optimum location to release the tracer gas and techniques to release
tracer gas which include the tracer dilution method, the constant injection method, or other
methods will be constructive. (5) Finally, future experiments will use multiple tracer gases and
comparing the efficiency over the use of single tracer gas.
50
|
Virginia Tech
|
This paper is published in the journal Tunnelling and Underground Space Technology. The experimental
and CFD modeling work and writing was primarily completed by Guang Xu with editorial and technical input from
coauthors: Dr. Kray D. Luxbacher, Dr. Saad Ragab, and Steve Schafrik. Additionally, Steve Schafrik was
instrumental in the logistical details associated with running CFD on a high performance computer (HPC). The
paper can be found at the link: http://www.sciencedirect.com/science/article/pii/S0886779812001551. Please cite
this article as: Xu, G., Luxbacher, K. D., Ragab, S., & Schafrik, S. (2012). Development of a remote analysis method
for underground ventilation systems using tracer gas and CFD in a simplified laboratory apparatus. Tunnelling and
Underground Space Technology, 33, 1–11.
4 Development of a Remote Analysis Method for Underground
Ventilation Systems using Tracer Gas and CFD in a Simplified
Laboratory Apparatus
4.1 Abstract
Following a disaster in a mine, it is important to understand the state of the mine damage
immediately with limited information to manage the emergency effectively. Tracer gas
technology can be used to understand the ventilation state remotely where other techniques are
not practical. Computational fluid dynamics is capable of simulating and ascertaining
information about the state of ventilation controls inside a mine by simulating the airflow and
tracer distribution. This paper describes a simulation of tracer gas distribution in a simplified
laboratory experimental mine with the ventilation controls in various states. Tracer gas
measurements were taken in the laboratory experimental apparatus, and used to validate the
numerical model. The distribution of the tracer gas, together with the ventilation status, was
analyzed to understand how the damage to the ventilation system related to the distribution of
tracer gases. Detailed error analysis was performed and the discrepancies between experimental
and simulated results were discussed. The results indicate that the methodology established in
this study is feasible to determine general ventilation status after incidents and can be transferred
to field experiment. Because it is complex to simulate the actual condition of an underground
mine in a laboratory, the model mine used is simplified to simulate the general behavior of
ventilation in a mine. This work will be used to inform planned on-site experiments in the future
and the proposed methodology will be used to compare collected and simulated profiles and
determine the general location of ventilation damage at the mine scale.
4.2 Introduction
After a severe underground mine incident, such as a roof fall, outburst, water inrush, or
explosion that may cause tunnel collapse, underground information must be gathered
immediately to estimate the extent of damage for rescue and recovery operations. In these
51
|
Virginia Tech
|
situations, communications between underground miners and rescuers on the surface may be
tenuous at best, because very few commercially available communications systems have been
proven to meet the basic requirements for emergency communication [117]. Some alternate
methods can be used to gather information, such as collection of air samples from boreholes,
utilization of a video camera via borehole to visualize underground status, and utilization of
rescue robots underground if possible. However, none of these methods are reliable and efficient
enough to stand alone. In an emergency situation, accurate information regarding the mine status
is invaluable not only to save miners’ lives, but also to help decision makers manage the
emergency effectively, and to increase safety for rescuers.
In some incidents, such as explosions, all the communication lines maybe damaged and
collapse may occur with the location difficult to pinpoint from the surface. However, the airflow
paths and ventilation patterns will change according to the location of damage. Therefore, the
location of damage can be approximately determined by remote measurement of ventilation
parameters. Due to the complexity of the ventilation system, employment of the tracer gas
method is an effective means and has been used in many situations where conventional
techniques are inadequate or cannot be effectively employed [1], [2]. Numerical simulations
using Computational Fluid Dynamics (CFD) can be used to model the ventilation status and the
data from tracer gas measurement allow for further analysis, prediction, and confirmation of the
underground ventilation status together with the location of the damage.
Tracer gas was first used in the building ventilation systems in the 1950s [61] and has
been widely used for ventilation analysis both in buildings and underground mines [118]. Tracer
gas based ventilation measurement is an effective method to detect air flow routes, estimate air
flow quantity, and other complex ventilation problems [119], [120]. Sulfur hexafluoride (SF ) is
6
widely accepted as a standard mine ventilation tracer [118], because it can be detected in low
concentrations, is nontoxic, odorless, colorless, chemically and thermally stable, and does not
exist naturally in the environment [1]. The applications of tracer gases in underground mines
include measurement of turbulent diffusion [107], methane control [121], study of mine
ventilation recirculation of return into intake air, transit flow times through stopped areas,
effectiveness of auxiliary fans, and estimation of volumetric flow rates [1], [122] air leakage
investigation, and evaluation of dust control measures [62].
52
|
Virginia Tech
|
In recent years, Computational Fluid Dynamics (CFD) has become a powerful tool and
has been commonly used to model the underground mine air flow behavior and solve relative
problems [67], [100]. It has been used in a number of areas, including modeling ventilation
airflow patterns [67], [98], study and control of coal spontaneous heating and underground fire
[9], [20], [83], optimization of gob inertisation [95], dust control [98], and methane management
[91]. The combination of experimental measurement and CFD modeling of tracer gas has been
used to study airflow and contaminant transport in indoor environments and other industrial
applications [114], [123], but little research has been done to model underground tracer gas
applications, especially the use of these techniques to model and analyze the ventilation and
mine environment following an incident that alters the ventilation system.
CFD has also been used in many studies to investigate underground tunnel risks. Hua et
al. [124] used CFD model to develop an optimal smoke control strategy for tunnel fire. The
model was validated using the test results from a similar tunnel and an optimal smoke control
strategy was found based on the model results. Se et al. [125] used CFD model investigated the
effect of active fan group on the airflow structure and temperature distribution in a tunnel with
varied fire source. Gao et al. [126] used Large Eddy Simulation to study the dispersion of fire-
induced smoke in a subway station and the influence of natural and mechanical ventilation was
investigated. Risks challenging underground mine are also faced by underground tunneling and
constructions, particularly the methane explosion described in [127], the fires scenarios
mentioned above, flood, and earthquake. The proposed methodology can also be potentially
applied to those situations to better understand the ventilation status remotely, and thus manage
the emergency effectively with significant impacts on safety. Also, the tracer gas test, sampling,
and analyze techniques used in this study can be applied to underground tunnel ventilation
survey to investigate ventilation efficiency, flow path, and other related issues.
A CFD approach was used in this study due to the relative simplicity of the experimental
apparatus. CFD can resolve details of flow features and tracer distributions. These will help,
when we move our test to the field scale, to determine the optimum method and place to release
and sample tracer gas. However, it is not practical to apply CFD to the entire mine due to its
heavy demand on computational time. Ventilation network modeling is more practical in this
situation, but it cannot resolve the detail of tracer gas behavior at the micro scale. Although the
53
|
Virginia Tech
|
focus of this study is CFD modeling, a hybrid scheme will be investigated when this work is
applied to the field, which combines the benefits of CFD and network modeling.
In this study, tracer gas (SF ) was used in an experimental laboratory simplified model
6
mine which was built according to a conceptual mine layout. A CFD model was developed to
simulate the laboratory apparatus. Various states of ventilation patterns were controlled by
valves in the experimental mine to simulate different ventilation scenarios after incidents. Tracer
gas was released to the model mine at a constant rate. Air samples were analyzed to test the
tracer concentration at different locations. The aim of this study is to use the experimental data to
validate the CFD model, study the relationship between the tracer concentration and the location
of incidents, and finally, through analysis of the air sample and the CFD model result, determine
the general location of the ventilation damage. A preliminary version of this study was presented
by Xu and others [128].
4.3 Experimental setup and measurements
4.3.1 Experimental apparatus
The laboratory mine model represents a simple conceptual mine shown in Figure 13, in
which the arrows indicate the normal air flow path and this state is referred as Case 1. It has one
active panel and one gob panel, two regulators (V2 and V4) regulating the air flow into the
panels, one stopping (V1) between the main entries, and five boreholes (P1 - P5). Stopping
damage at V1, a roof fall in the active panel at V3, and explosion damage in the gob panel at V5
are three possible incidents will be investigated in this study. The air flow to the gob was
simplified in both the laboratory experiment and the CFD model in that the air simply flows
around the gob; a permeable gob was not studied. It should be noted that this simplified
experimental model mine was not used to represent a full scale mine, but rather to validate the
CFD model, choose the best sample methods, and test the effectiveness of our methodology,
which can later be used to conduct field experiments.
54
|
Virginia Tech
|
P1
Intake
V1
V4 V2
Exhaust
P4
Active
Gob Panel
V3
P2 P5
P3 V5
P1: velocity monitor point, P4: velocity monitor and sample points; P2: velocity monitor and tracer gas release point, represent a rescue borehole;
P3: velocity monitor and gas sample point, represent an existing borehole; P5: velocity monitor and gas sample point, represent a borehole drilled
specifically for gas sampling; V1:valve represents a stopping; V2 and V4: valves represent regulators; V3: valve represents roof fall; V5: valve
represents explosion damage
Figure 13. Typical coal mine layout used in this study (Case 1)
The experimental mine model was built, as shown in Figure 14, using 0.05m (2 in.)
inside diameter PVC pipes and allows for experiments representing general flow paths of a
typical coal mine shown in Figure 1 under different ventilation statuses. The general dimension
of the model is 6.63 m in length, 0.51 m in height, and 0.36 m in width. Air exhausts from the
apparatus via a variable speed exhaust fan. Valves were used to represent stoppings, regulators,
and damage due to incidents.
Figure 14. Configuration of the experimental underground mine model
55
|
Virginia Tech
|
Pitot tubes and an electrical manometer were used to measure differential pressure at
points P1 - P4, and the results were used to calculate velocities at those points using the
following equation [129]:
√ ( 4-1 )
Where h is differential pressure measured by pitot tube and manometer in kPa, and d is
kpa COR
corrected air density in kg/m3, which can be calculated from the following equation [129]:
𝐻
=3.4834×𝑃 𝐵 × 1 (0.3783× 100×𝑃 𝑠 ) ( 4-2 )
𝑂 𝑃
𝐾 𝐵
Where P is barometric pressure in kPa, T is absolute temperature in Kelvin, RH is relative
B K
humidity. P is partial pressure of water vapor at T , and can be calculated using the following
s K
equation:
𝑃 5 8 5 5 56/ 𝐾 ( 4-3 )
During this experiment, the measured barometric pressure is 94732 Pa, the temperature is
295 K, and the relative humidity is 29%. Therefore, using equations shown above, the calculated
air density is 1.114 kg/m3. This value is used in the experiment velocity calculation and the CFD
model input.
4.3.2 Gas release and sampling
There are basically two categories of tracer gas release techniques: transient techniques
and the constant injection rate techniques [130]. Because of the small dimensions of the model
mine and the high rate of air velocity, the air in the model mine will be totally replaced in 10 - 20
seconds. This makes it impractical to use transient release techniques since the sampling methods
we investigated could not sample quickly enough to resolve a useful profile. Therefore, a
constant injection rate technique was used when releasing SF . Tracer gas was released
6
continuously at a controlled rate of 40 standard cubic centimeters (SCCM), into the model mine
at point P2. In the experiment, tracer gas was released 10 min before sampling. An electrical
mass flow controller was used to control and measure the tracer gas (SF ) flow rate.
6
The experiment requires a sampling technique that is relatively simple with minimum
leakage. Four sampling methods were considered including glass syringe, plastic syringe,
vacutainer, and glass vial. The 50 μl gas tight glass syringe and 3 ml gas tight disposable syringe
56
|
Virginia Tech
|
are shown in Figure 15a and b. Glass syringes are expensive and relatively fragile, so they are
not appropriate for large samples or transport underground. Experimentation showed that
although plastic syringes are inexpensive and convenient, they yielded higher relative standard
deviation (RSD) values, which measure the precision and repeatability of gas sample.
Blood collection vacutainers, shown in Figure 15c have long been used for sampling
mine air and products of combustion because they are convenient and result in high precision
even after one - two weeks of storage [131], [132]. The 10 ml vacutainer was chosen as the
sampling method for the experiment. The vacutainers were re-evacuated in the laboratory to
improve the sampling accuracy which improved the capability of vacutainers.
Crimp top vials, shown in Figure 15d, were also evaluated in a similar manner as the
vacutainers. However, it was determined that the integrity was largely dependent on the crimping
technique which was not consistent among vials.
Considering the observation mentioned above and comparison between different gas
sampling methods, 10 ml blood collection vacutainers were chosen as a proper sampling method
for the experiment.
Figure 15. Gas sampling methods
4.3.3 SF measurement
6
A gas chromatograph equipped with an electron capture detector (ECD) was used to
analyze the concentration of SF in collected samples. Calibrations are required before testing
6
and the accuracy and precision of the calibrations are critical for quantitative analysis. A series of
SF standards were made, at concentrations of 10, 20, 50, and 100 ppm, to create a calibration
6
chart that could be used for the analysis of SF between 10 and 100 ppm. The standards were
6
made by injecting 2.75 μl, 5.5 μl, 13.75 μl, and 27.5 μl pure SF into a 275ml glass bulb which
6
was full of ultra-pure nitrogen to make 10 ppm, 20 ppm, 50 ppm, and 100 ppm standards,
57
|
Virginia Tech
|
respectively. 20μl injections of those standards were made to the GC using the 50μl gas tight
glass syringe shown in Figure 15a. The complete calibration chart is shown in Figure 16.
)e
120
m
u 100
lo
V
y
80
b
M 60 y = 6E-06x + 0.2148
P
P R² = 0.9809
( n 40
o
ita
20
rtn
e 0
c
n
o 0.00E+00 5.00E+06 1.00E+07 1.50E+07 2.00E+07
C
GC Respond Peak Area (μv2×1010)
Figure 16. GC calibration chart for SF
6
As noted, units of ppm were used for the SF concentration, indicating parts per million
6
by volume or by mole, which are identical for an ideal gas, and it has the same value at both
actual and standard conditions because the temperature and pressure changes affect the
denominator of the ideal gas law proportionally [133].
After gas samples were collected from the laboratory apparatus using vacutainers, 20 μl
of the gas sample was taken from the vacutainer, and injected to the GC using a 50 μl gas tight
glass syringe. Three samples were taken at the same sampling location, and the average SF
6
concentration was used as the final result.
4.3.4 Experimental ventilation status
The experimental mine model was operated under four different conditions by closing
and opening different valves, which represent the normal case, roof fall occurrence in an active
panel, explosion in the gob panel, and stopping damage. These scenarios were chosen because
they are likely to disrupt ventilation and have potential for remote characterization by tracer gas.
The air flow path under normal case is shown in Figure 13 while other cases are shown in Figure
17. For convenience, the normal ventilation condition, roof fall in active panel, explosion in the
gob panel, and stopping damage, will be referred as Case 1, Case 2, Case 3, and Case 4,
respectively.
58
|
Virginia Tech
|
P1 P1
Intake Intake
V1 V1
V4 V2 V4 V2
Exhaust Exhaust
P4 Active P4 Active
Gob Panel Roof Fall Gob Panel
(Valve 3 Close) V3
P2 P5 P2 P5
P3 V5 P3 Explosion Damage
(Valve 5 close)
Case 2: Roof Fall in Active Panel Case 3: Explosion Damages in the Gob Panel
P1
Intake Stopping
Damaged
(Valve 1 open) V4 V2
Exhaust
P4 Active
Gob Panel
V3
P2 P5
P3 V5
Case 4: Stopping Damage
Figure 17. Air Flow path of different ventilation conditions
As can be seen, air flow paths under the four ventilation conditions are different and
should result in varied distribution of SF . Although ventilation parameters, such as velocity and
6
pressure, are different for each case, it may be difficult to measure those parameters after an
incident in the field. However, air samples are routinely taken from boreholes when a mine is
inaccessible and the tracer concentration can be analyzed.
4.4 CFD model setup
4.4.1 Hypothesis
Approximations and simplifications of the actual problem are needed to construct the
CFD study, which allows for analyzing the problem with reasonable effort. The following
assumptions are made in this study:
1) No leakage in the model mine.
2) Mine air is incompressible.
3) PVC pipe surface is smooth.
4) The flow in the model mine is fully turbulent.
5) No heat transfer during the procedure and the wall and air temperatures are constant.
6) The gravity influence on SF is not considered.
6
7) Introduction of SF will not influence the final steady state of the air flow.
6
These assumptions are made based on our preliminary study focuses. For example,
assumption 1 is not realistic because small leakage could exist at the connections of the pipes.
But the influence of this small leakage on the final experimental results is minor or predictable.
59
|
Virginia Tech
|
A better understanding and interpretation of the actual problem and the results also make the
assumptions reasonable. Take assumption 7, for example, the influence of releasing small
amount of SF to the air flow is very minor compared to the quantity of air flowing through the
6
model mine.
4.4.2 Governing equations
CFD is based on the fundamental governing equations of fluid dynamics, including the
continuity equation, momentum equation, energy equation, and transport equation, which
express the fundamental physical principles of fluid dynamics. The energy equation was not used
in the model since the modeling fluid was assumed to be incompressible and there is no heat
transfer. The governing equations can be expressed in the conservation form of transport
equation [78]:
𝜑
div ⃗⃗⃗ 𝜑 div Γ grad𝜑 ( 4-4 )
𝜑𝑡 𝜑𝑡
Where, 𝜑 is general variable of interest, is air density, Γ is diffusive coefficient, and is
𝜑𝑡 𝜑𝑡
source term [78].
In the CFD model, SF was released at a constant rate of 40 SCCM at a point
6
corresponding to P2 in Figure 13, which is the same release rate and location as the laboratory
experiment. It was simulated using user defined scalar, and scalar transport equation (Equation
4-4) was solved to calculate the velocity, pressure, and other quantities of SF . However, in
6
turbulent flows, the diffusion, which is the third term of Equation 4-4, was programmed as a user
defined function as follows [134]:
Γ 𝜑𝑡 Γ 𝜑𝑡 ( 4-5 )
𝑡
where Γ is the diffusion coefficient of SF in air, μ is the turbulent viscosity, and S is the
φt 6 t Ct
turbulent Schmidt number. Tucker et al. [135] provided an equation to calculate Γ of one gas
φt
in another gas resulting in a diffusion coefficient of SF in air 8.96 × 10-6 m2/s. Bai et al. [136]
6
used a value of 9.7 × 10-6 m2/s in their study; while Ward and Williams [137] reported the
diffusion coefficient of SF in air is between 5.9 × 10-6 m2/s and 7.3 × 10-6 m2/s. These values do
6
not differ substantially, especially when considering the term μ /S , which determines turbulent
t Ct
diffusion, generally overwhelms laminar diffusion, which is determined by Γ ρt [50].
φt
Therefore, the final result is not significantly sensitive to the range of Γ cited in the literature.
φt
60
|
Virginia Tech
|
This is also proved by the parameter sensitivity study in section 6. A value of 5.9 × 10-6 m2/s is
used in this study. The Schmidt number is a dimensionless parameter which is the ratio of
diffusion of momentum to the diffusion of mass. For gases, it is approximately 0.7 [50].
4.4.3 Mesh and boundary conditions
Commercial drafting and meshing tools were used to generate the three-dimensional
geometry and mesh. An unstructured, hexahedral mesh was generated to represent the size and
geometry of the lab experimental mine model. The “O” grid is used on the pipe cross section
with fine mesh near the pipe wall and coarse mesh in the center. This can help resolve the rapid
variation of flow variables near the pipe wall with reasonable mesh size and compute time. Due
to the complexity of the model, the whole geometry was divided into four parts before generating
the mesh. Then four mesh parts were connected by interface boundary condition. Figure 18
shows the CFD model and the O grid mesh can be seen in Figure 19.
Figure 18. The 3D CFD model and meshing
Figure 19. Cross section of different mesh size (from left: coarse mesh, medium mesh, and fine mesh)
The inlet and the outlet of the model were specified as velocity inlet and pressure outlet,
respectively. The measured outlet pressures for the four cases were applied to the corresponding
model pressure outlet boundary. However, because the pitot tube can only measure the
maximum velocity when placed at the center of the pipe, the measured inlet velocities were
calibrated so that the velocities at the other four measure points approach the measured value.
61
|
Virginia Tech
|
Table 4 shows the detail boundaries condition for all cases. All of the other surfaces are treated
as stationary walls with no slip. Both air and wall temperatures are assumed constant.
Table 4. Boundary condition
Case Number Pressure Outlet (Pa) Velocity Inlet (m/s)
1 335 5.6
2 336 5.5
3 336 5.08
4 332 7.15
4.4.4 Numerical details
The numerical simulations in this study were conducted using the commercial CFD
package, ANSYS FLUENT 12.1, to simulate the airflow and tracer gas dispersion. A standard
two equations k- ε turbulence model was employed to simulate the air flow and SF transport.
6
The standard k- ε model is the simplest complete turbulence model and widely used in the
modeling of mining turbulent flow in broad range of applications [8], [9], [22]. A second order
upwind scheme was used for variables including pressure, momentum, turbulent kinetic energy
and turbulent dissipation rate, which ensures the higher order of accuracy results. Discretized
airflow equations were solved with the SIMPLE algorithm in the CFD program to couple the
pressure, velocity, momentum and continuity equations.
4.5 Mesh independent study
4.5.1 Mesh quality and size
In order to achieve results that are independent of mesh size, three different mesh size
models were developed and analyzed, including coarse mesh, medium mesh, and fine mesh. Two
factors, determinant and angle, were utilized as a measure of the mesh quality. A determinant
value above 0.3 is acceptable for most solvers and the minimum angle value above 18 degree is
acceptable for Fluent [50], [138]. The generated mesh quality meets the criteria mentioned
above.
The number of nodes are approximately doubled progressively, which is about 10
million, 20 million, and 40 million, for coarse, medium, and fine mesh, respectively. The meshes
were generated to improve the node density both on the pipe’s cross sections and along the pipe.
Figure 19 shows the cross section mesh for coarse, medium, and fine mesh.
62
|
Virginia Tech
|
4.5.2 Solution convergence
Some criteria need to be met to achieve converged numerical solutions. Two criteria are
used to check the solution convergence for each mesh [139]. The first criteria is residuals of each
conservation equation, which is a specified tolerance defined by Fluent. The criterion used in this
study is continuity residual value equal to or less than 10-5.
Sometimes the residual may reach the convergence criterion, but the solution still
changes with more iterations. Therefore, a second criterion is used, which monitors variables of
the solution until it no longer changes with more iterations. In this study a point monitor was
created on the outlet surface, shown as black dots in Figure 19, and the velocity at these points
were monitored. The final velocity values for coarse, medium, and fine mesh size models at the
monitor point are 6.716 m/s, 6.717 m/s, and 6.670 m/s, respectively. The velocity differences at
the monitor point between medium and fine mesh is less than 1%, which is acceptable and
indicates, from one aspect, that mesh independence has been achieved. The mesh independence
will be discussed in further detail in the next section.
4.5.3 Mesh independence study
It is important to conduct the mesh independence study before using the CFD results
since the numerical solution may depend on the mesh size if mesh independence is not achieved
[48]. As the mesh becomes finer, the numerical solution will asymptotically approach the exact
solution of the governing equations [43]. Mesh independences are studied considering different
flow features and at different locations.
A line at the centerline 0.36 m away from the outlet was created for each case and three
velocity profiles across the center line were plotted on Figure 20. It is apparent by comparing the
shape of the profile and the predicted velocity that the solution is not changing with the mesh
size since the profile points are practically on top of each other, which indicates that the solution
is mesh independent. The result differences are within 1% as indicated by the calculation in the
previous section. This also indicates that the medium mesh is sufficient for a robust solution and
could be used for further modeling.
63
|
Virginia Tech
|
8
7
6
ms) /5
CoarseMesh
y ct( i4 M Fine edi Mum esM hesh
o
Vel
3
2
1
0
-0.1 -0.09 -0.08 -0.07 -0.06 -0.05
PositioninModelCoordinate
Figure 20. Velocity profile at the line monitor
Velocity contours and vectors are also used for visual comparison of different mesh size
results. Due to limited space, they are not shown here. But the contours and vectors between
different mesh sizes are generally the same with very small different, and the general high and
low velocity zone are pretty similar.
The computed velocities at four velocity monitor points (P1 to P4) are compared with
experimental results. The comparison is shown in Figure 21, which shows that the computed
results for each mesh size are very close to the experimental results. Based on the comparison,
the medium mesh can be chosen with 8.2% RMS (root mean square) deviation compared to
experimental results. The differences between the computed and experimental results are due to
errors, such as the accuracy of the measured velocity, which will be discussed in detail in section
4.7.
8.0
) 6.0
s / Experiment
m
( y 4.0 Coarse Mesh
t
ic
o Medium Mesh
le
V 2.0
Fine Mesh
0.0
Point 1 Point 2 Point 3 Point 4
Figure 21. Comparison of computed and experimental velocity
64
|
Virginia Tech
|
4.6 Results and discussions
4.6.1 Velocity validation
The CFD models were first validated using the experimental velocity results before
modeling the tracer. The comparisons with the errors are shown in Table 5. The CFD simulation
data were taken at the same locations as the experimental data were measured. It can be seen that
the simulated velocity in all cases are in good agreement with the experimental results with no
more than 2% error. The differences are larger in Case 4 at points P2 and P3. This is because the
manometer used in the experiment is limited in range at pressures equivalent to a velocity of less
than 1.5 m/s. Generally, the comparisons indicated that the CFD models are valid and can be
used to conduct tracer modeling.
Table 5. Velocity results comparison between experimental measurements and CFD simulations
P1 P2 P3 P4
Experiment 7.17 m/s 3.66 m/s 3.66 m/s 7.17 m/s
Case 1 Computed 7.23 m/s 3.70 m/s 3.55 m/s 7.28 m/s
Error 0.68% 0.74% 3.13% 1.46%
Experiment 6.69 m/s 0 m/s 3.34 m/s 6.85 m/s
Case 2 Computed 6.73 m/s 0 m/s 3.29 m/s 6.78 m/s
Error 0.58% n/a -1.52% -1.1%
Experiment 6.52 m/s 3.34 m/s 0 m/s 6.69 m/s
Case 3 Computed 6.58 m/s 3.36 m/s 0 m/s 6.63 m/s
Error 0.89% 0.60% n/a -0.84%
Experiment 9.10 m/s 0 m/s 0 m/s 9.34 m/s
Case 4 Computed 9.16 m/s 0.89 m/s 0.87 m/s 9.18 m/s
Error 0.69% n/a n/a -1.68%
4.6.2 SF concentration results
6
SF concentrations obtained from the experiments and from the validated CFD models
6
are shown in Figure 22. Because the GC calibration curve is only valid when the SF
6
concentration is between 10 ppm to 100 ppm, results out the range only show approximate
values, for example, results above 100 ppm only indicate the results are more than 100 ppm and
the extent above it.
Case 1 Case 2 618 1000
n o ita
rtn
e c n o Cm p p ()
8
6
4
20
0
0 0
74
55
76
58
E Cx Fp Deriment n o
ita
rtn e c n o)m p p (
10 50
0 52
5E C 8x F p Derime 5n 7t
61
6 F 0 1 0 C 6
S F S 0
P3
P4 P3
P5 P4
P5
65
|
Virginia Tech
|
Case 3 Case 4
165 232
n o ita
rtn e c n o C
6 F
S)m p p (
8
6 4
20
0 0 0
0
0
61 82 64 E
C
1x
F
p Deri 0m ent n o ita
rtn e c n o C
6 F
S)m p p (
10 50
0
0
59 46
E
C
1x
F
p Der 0im ent
P3 P3
P4 P4
P5 P5
Figure 22. The comparison of experiment and CFD SF concentration results (Values above 100 ppm do not
6
indicate the exact value)
The largest difference is at point P3, Case 3. The CFD result shows 61 ppm while the
experimental result is zero. This is under the situation that the explosion damage blocked the
entry around the gob and theoretically no airflow exists in the entry. CFD calculations are based
on ideal conditions of no air leakage and no flow in this entry. Due to the diffusion caused by the
SF concentration gradient between the main entry and the gob entry, the SF concentration
6 6
eventually reaches an equilibrium state where the concentration is very close to that of the main
entry. Therefore, the CFD results showed that the main entry SF concentration at point P4 is 64
6
ppm and the gob entry at point P3 is 61 ppm. However, the laboratory model mine is not under
ideal conditions and it has minor leakage at the connections. Since a negative pressure is created
by the exhaust fan in the laboratory apparatus, the leakage can cause fresh air leak into the gob
entry and this purge effect overcomes the SF diffusion and causes the experiment result at point
6
P3 to be 0 ppm instead of a value close to that of point P4, which is 82 ppm. The same
phenomena caused the CFD result for Case 2 at point P5 to be above 1000 ppm, although the
experimental result at the same point is lower. The errors between the experiment and CFD
results at other points will be analyzed in section 6. However, except for the large disagreement
mentioned above, the experiment and CFD results at other points generally agree with each
other, especially in that they display the same trend, as can be seen in Figure 22 (please note that
values above 100 ppm do not indicate the exact value).
4.6.3 Ventilation status prediction
From section 4.6.2, one can see that the CFD predicted SF results are in agreement with
6
the experimental results if reasonable explanations are provided, meaning that SF concentration
6
results under different ventilation scenarios can be predicted ahead of time. Four different
ventilation scenarios assumed in the experiment led to four dramatically different combinations
of SF concentrations at different sample points, as shown in Figure 23. This indicates that it is
6
66
|
Virginia Tech
|
possible to use the known SF concentration at different sample points to predict the ventilation
6
scenarios. For example, if we have a result that the SF concentration at point P5 is very low
6
(near zero), but point P3 is much higher than P4, using Figure 23 we can predict that the
ventilation status is similar to Case 4, where the stopping between the main entries was damaged.
The relationships between the concentrations and the ventilation scenarios are quite
straightforward due to the relative simplicity of the mine model.
Case 1
Case 2
n Case 3
o ita 100 Case 4
rtn)m Case 4
e c np p 50 Case 3
o C 6( 0 Case 2
F Case 1
S P3
P4
P5
Figure 23. The comparison of SF concentration under different ventilation scenarios
6
4.7 Error analysis
Velocity and SF concentration are the two key parameters measured in the experiment
6
and calculated in the CFD model. Because the CFD models calculate the results based on
assumptions made in section 4.4.1 and measured parameter inputs, discrepancies exist when
comparing with the measured results.
Part of the error results from the experimental instrumentation. First, the pitot tube is
supposed to be installed against the airflow and at the center of the pipe, but this is not easy to
ensure since the PVC pipe is not transparent. The displacement from the center pipe line will
cause inaccurate differential pressure reading, resulting in an inaccurate velocity. Second, air
density was used in the velocity calculation, but it was calculated based on barometric pressure,
temperature, and relative humidity, all of which can introduce measurement error that result in
the final velocity calculation error in the experiment. Third, the sensitivity and the accuracy of
the manometer also can cause velocity error. The sensitivity of the manometer used is 0.005 in.
of water, which corresponds to 1.5 m/s air velocity. The model mine leakage, although
minimized by sealing of the apparatus, may also influence the velocities and SF concentrations
6
as well, especially at certain locations where the leakage overcomes the SF diffusion in dead
6
end scenarios. All of these errors may cause a final SF concentration calculation error in the
6
CFD model. Last, the GC calibration curve is a key factor to an accurate SF concentration
6
reading. The procedure described in section 2.3 may introduce error to the calibration curve that
67
|
Virginia Tech
|
affect the accuracy of the air sample analyses results although the RSD values for the GC work
were well within acceptable range.
Parameters input into the CFD model, such as the air density, viscosity, and diffusion
coefficient are chosen by the authors from calculation (air density) or from the literature. The air
density error was mentioned in the previous paragraph. Air viscosity and diffusion coefficients
also vary under different temperatures and in different literature. Therefore, parameter sensitivity
studies were conducted using the Case 1 model, which is the normal ventilation status, with
parameters chosen in this study, as the control model. Five more runs were calculated with
different combinations of parameters, which can be seen in Table 6. From the comparison
results, which are shown in Table 7, we can see that a 10% air density deviation has little
influence on point velocities, but will cause about 10% SF concentration deviation. The air
6
viscosity deviation has little influence on either velocity or SF concentration. The diffusion
6
coefficient was increased 64% percent to another value found in the literature [136], but the
subsequent influence on velocity was no more than 1.15% and SF concentration no more than
6
4%. This is a minimal effect and this finding is consistent with the discussion about the diffusion
coefficient in section 4.4.2. Overall, the three parameters studied: air density, air viscosity, and
diffusion coefficient, cannot cause velocity results to err substantially. However, the SF
6
concentration is sensitive to air density, but not to air viscosity and diffusion coefficient.
Table 6. Parameters details for sensitivity study
Air Density (kg/m3) Air Viscosity (kg/(m•s)) Diffusion Coefficient (m2/s)
Control model 1.1137 1.983×10-5 5.9×10-6
Run #1 (air density increased 10%) 1.2251 1.983×10-5 5.9×10-6
Run #2 (air density decreased 10%) 1.0024 1.983×10-5 5.9×10-6
Run #3 (air viscosity increased 10%) 1.1137 2.181×10-5 5.9×10-6
Run #4 (air viscosity decreased 10%) 1.1137 1.785×10-5 5.9×10-6
Run #5 (diffusion coefficient increased 1.1137 1.983×10-5 9.7×10-6
64%)
In conclusion, the CFD model results can be improved by reducing the instrument errors
and providing more accurate parameters for the CFD input. These include precise placement of
the pitot tube in the center of the pipe, use of manometers that have higher sensitivity and
accuracy, reduction of human errors introduced when calibrating the GC by using accurate tracer
gas standard. Also since the tracer gas concentration results are very sensitive to air density,
ensuring accurate measurement of barometric pressure, temperature, and relative humidity, can
improve the accuracy of the CFD calculated results.
68
|
Virginia Tech
|
The results indicate that the tracer gas concentrations can be predicted using CFD
modeling. Different ventilation statuses will result in substantially different tracer gas
distribution if tracer gas experiments were carefully designed. The methodology established in
this study is feasible to determine general ventilation status after incidents in subsurface
excavations. This requires that a detailed ventilation survey be conducted under normal status in
order to establish and calibrate the CFD model. Tracer gas experiments need to be designed and
performed carefully in order for ventilation status to be rapidly determined after an incident.
CFD models can predict tracer gas distribution results under different ventilation situations and
those results should be substantially different if the tracer gas release and sample locations are
optimized. By comparing the tracer gas experiment and the CFD predicted tracer gas distribution
results, the actual mine ventilation status should be the one with the similar CFD modeled
results. Further studies are needed, especially field trial, utilizing the tracer gas method outlined
in conjunction with CFD. The real mine experiment allows the use of transient tracer gas release
techniques, which are expected to be more efficient and achieve more definitive results.
Due to the complexity involved in simulated the conditions of an underground mine in a
laboratory, the model mine apparatus used in this study was simplified and built with PVC pipe.
It should be noted that ventilation network modeling can serve the purpose of this study equally
well at the macro scale and network modeling can be easily applied to full scale mines while
CFD cannot mainly due to its heavy demand on computational time and initial boundary
specification. However, network modeling cannot resolve the detail of tracer gas behavior, such
as where tracer gas is fully mixed with mine air, layering effects of the tracer, and how a tracer
concentration is distributed over entry cross sections. These factors are important for the tracer
experiment regarding the best location to release and collect gas samples. The focus of this study
is CFD modeling, but as this work is applied in the field, a hybrid scheme will be investigated. A
hybrid scheme should combine the benefit of CFD and network modeling. CFD will only be
used in critical areas where mixing and diffusion of the tracer gas within the airflow are
questionable, while most parts of a mine will be modeled using network modeling to save
computational time with equally effective results.
As stated earlier, the transient tracer gas release techniques will be used in the field
experiments. The expected results should be similar to the CFD modeling results presented by
Xu et al. at 2012 SME [140]. For the purpose of explanation, tracer gas’s concentration profile
70
|
Virginia Tech
|
under assumed normal and roof fall ventilation scenarios from that paper’s results are presented
in Figure 24. As can be seen, different ventilation scenarios result in different tracer profile in
terms of arrival time, number of peaks, and peak height. Because tracer gas field experiments
may be time and resource consuming, such simulated results are valuable for on-site tracer
experiments design since if the desired results are not achieved, it takes a period of time for
tracer to be cleaned out of the mine so it does not interfere the next experiments. For example,
the simulation can help to determine how much tracer gas needs to be released in order to
achieve a concentration at the monitor point practically detectable, that is to say, we want the
concentration fall into the range of the GC calibration curve so the gas samples can be directly
injected to GC without dilution or concentration. The optimal time interval for gas sampling can
also be determined before the experiment in order to adequately resolve each peak shown in the
figure. Overall, the established gas sampling and analysis method in the laboratory can be used
in the next stage field experiments. Numerical modeling with CFD or hybrid (CFD and network
modeling) approach can not only predict the general ventilation scenarios but also helps the
design of tracer gas tests to get the expected results and save time.
450
)b
400 Normal ventilation case
p 350 Ventilation after active panel roof fall
p
(
n 300
o
ita 250
rtn
200
e
c n 150
o
C
6
100
F
S 50
0
0 10 20 30 40 50 60 70 80 90 100
Flow Time (min)
Figure 24. CFD simulated SF concentration at a point monitor of different ventilation status
6
71
|
Virginia Tech
|
This paper was presented at the 2012 SME annual meeting in Seattle, Washington on February 22, 2012,
and is included in the meeting preprints(Feb. 19 – 22, 2012, Seattle, WA, Preprint 12-051). Guang Xu conducted the
majority of the work and wrote the paper with technical and editorial input form coauthors: Edmund Jong, Dr. Kray
D. Luxbacher, and Dr. Saad Ragab. Please cite this article as: Xu, G., Jong, E., Luxbacher, K., & Ragab, S. (2012).
Computational fluid dynamics study of tracer gas dispersion in a mine after different ventilation damage scenarios.
SME Annual Meeting (p. Preprint 12–051). Seattle, Washington (USA).
5 Computational Fluid Dynamics Study of Tracer Gas Dispersion
in a Mine after Different Ventilation Damage Scenarios
5.1 Abstract
Tracer gases are an effective method for assessment of mine ventilation systems, but their
dispersion characteristics can differ substantially as ventilation parameters, such as flow path and
velocity, vary. This research utilizes Computational Fluid Dynamics (CFD) to model a simplified
full scale model mine, details a sensitivity study examining mesh size for an underground coal
mine simulation, and examines gas dispersion parameters to determine the optimal model
methods for simulation of tracer gases in underground coal mines. These models can be used to
determine how a given tracer gas profile might be generated in a mine or areas of a mine that are
not accessible, for example, immediately following a mine disaster. Accurate simulation
scenarios can allow for the remote determination of the status of the ventilation network, but the
sensitivity of the simulation at mine scale must be carefully examined.
5.2 Introduction
There is a need to immediately know the underground status right after severe coal mine
incidents, such as roof falls, dust, and gas explosion, outbursts, and water inrush. In these
situations, measurement of many parameters are necessary to estimate and evaluate the
underground situation while organizing rescue operations, and managing the emergency
situation. Several techniques can be used for information collection purposes, including
collecting air samples from boreholes, inserting a video camera into boreholes to visually
monitor underground status, and deploying specialized robots. However, none of these methods
is sufficient to stand alone and more methods need to be developed in order to quickly and
accurately gather information that could help decision makers manage the emergency effectively,
increase safety for rescuers, and advance the rescue operation.
Tracer gas can be used to assess mine ventilation systems, and the dispersion of tracer gas
will change according to changes of airflow paths and ventilation patterns, as a result of an
incident. For this reason, tracer gas can be used to gather information, compare the results with
73
|
Virginia Tech
|
computational models and determine the general location of damaged ventilation systems.
Numerical simulations using computational fluid dynamics (CFD) can be used to model the
ventilation status and the data from tracer gas measurement can be utilized to further analyze,
predict and confirm the underground ventilation status and the location of the damage.
The use of computational fluid dynamics (CFD) to simulate flow problems in the mining
field has risen dramatically in recent years. CFD has become a cost effective research and design
tool with the increasing speed of high performance computers and more advanced computational
methods. CFD is applied to a wide range of industrial and research fields, such as aerodynamics
of aircraft, automotive, pollution control, agriculture, food science, power plant, civil
engineering, hydrology and oceanography, and medical science [3], [10]. It also has been used in
a number of mining areas, including modeling ventilation airflow patterns [67], [98], study and
control of coal spontaneous heating and underground mine fire [9], [20], [83], optimizing gob
inertisation [95], dust control [98], and methane management [91].
An underground mine tracer gas test needs to be carefully designed before the release of
any tracers. Knowing the expected results can help optimize the design of the tracer gas test,
such as the release location and the rate, and the location and time interval for sample collection.
This paper examines a simple full scale model mine CFD study, and primarily aims to review the
best methodology for full CFD mine simulations and the potential useful information that the
simulation results can provide. Although experimental data are not available to validate the
model, several studies were carried out to control the quality of the numerical model results, such
as the result convergence study and mesh convergence study. SF was then introduced to the
6
verified models which have different ventilation patterns due to different locations of damage.
The SF concentration at the outlet was monitored and compared for different models. The
6
relationship between the tracer’s concentration and ventilation status differences are also
examined, which demonstrated the feasibility of our methodology to clearly determine different
ventilation status after mine incidents using the tracer gas test and CFD simulated results.
5.3 The model mine
A full scale model mine was designed in this study based on the layout of a scaled
experimental study presented in a previous paper [128]. As shown in Figure 25 (entries are
represented by single lines), the designed mine model has one active panel and one gob panel.
Two regulators are used to control the air quantity that goes to the active panel and the gob. The
74
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.