Professional Documents
Culture Documents
Journal of Rock Mechanics and Geotechnical Engineering: Kunui Hong, Eunchol Han, Kwangsong Kang
Journal of Rock Mechanics and Geotechnical Engineering: Kunui Hong, Eunchol Han, Kwangsong Kang
a r t i c l e i n f o a b s t r a c t
Article history: The geological strength index (GSI) system, widely used for the design and practice of mining process, is
Received 6 September 2016 a unique rock mass classification system related to the rock mass strength and deformation parameters
Received in revised form based on the generalized Hoek-Brown and Mohr-Coulomb failure criteria. The GSI can be estimated using
20 April 2017
standard chart and field observations of rock mass blockiness and discontinuity surface conditions. The
Accepted 10 May 2017
Available online 10 July 2017
GSI value gives a numerical representation of the overall geotechnical quality of the rock mass. In this
study, we propose a method to determine the GSI quantitatively using photographic images of in situ
jointed rock mass with image processing technology, fractal theory and artificial neural network (ANN).
Keywords:
Jointed rock mass
We employ the GSI system to characterize the jointed rock mass around the working in a coal mine. The
Geological strength index (GSI) relative error between the proposed value and the given value in the GSI chart is less than 3.6%.
Image processing Ó 2017 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Production and hosting by
Fractal dimension Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
Artificial neural network (ANN) licenses/by-nc-nd/4.0/).
http://dx.doi.org/10.1016/j.jrmge.2017.05.001
1674-7755 Ó 2017 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-
NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
K. Hong et al. / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 702e708 703
Fig. 1. Chart for determining GSI of jointed rock mass (Hoek and Brown, 1997).
determine the GSI values of a rock mass disturbed by underground In this paper, we propose a method to quantitatively determine
exploitation. the GSI by first detecting the joints in two-dimensional (2D) pho-
In recent years, several authors such as Han et al. (2014), Poulsen tographs of a rock mass surface using the image processing tech-
et al. (2015) and Wang et al. (2015a) have also suggested many nology, then determining the fractal dimension, and finally
methods to determine the strength and deformation parameters of predicting the GSI using artificial neural network (ANN). The
rock mass using GSI. All of the above-mentioned papers focused on applicability of the method proposed is verified through stability
quantifying the GSI chart to facilitate use of the system especially analysis of the working in a coal mine.
by inexperienced practitioners.
Many researchers (e.g. Crosta, 1997; Castleman, 2002; 2. Joint detection on the rock mass surface using image
Hadjigeorgiou et al., 2003; Lemy and Hadjigeorgiou, 2003; Lato processing
et al., 2009) have investigated digital face mapping as a practical
tool to characterize rock masses, which can significantly reduce the Fig. 2 shows the schematic flowchart for detecting joints on the
time required in the field and avoid exposure to potentially unsafe rock mass surface via image processing. The detailed steps for joint
conditions. detection on the rock mass surface are described as follows.
The digital rock mass rating (DRMR) developed by Monte (2004)
uses basic image processing procedures and calculations to esti- (1) Converting the color image of rock mass to black and white
mate a classification rating from digital images of rock masses. The one
rating system incorporates fracture information collected from a
discontinuity trace map (e.g. length, spacing, large-scale, rough- To detect the joints on the image of jointed rock mass, the
ness, rock bridge percentage, and block volume). contrast of the image should be analyzed. We convert the natural
704 K. Hong et al. / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 702e708
8 9
>
> 1 1 1 >
>
> 0 0 >>
>
>
> 52 26 52 >
>
>
> >
>
>
> 1 3 1 3 1 >>
>
> >
>
>
> >
>
> 52 52 13 52 52 >
>
>
>
< >
=
1 1 1 1 1
Hð5; 5Þ ¼ (3)
> 26
> 13 13 13 26 >
>
>
> >
>
>
> 1 1 >>
>
> 3 1 3 >
>
>
> >
> 52
> 52 13 52 52 >
>
>
>
> >
>
>
> 1 1 1 >
>
>
: 0 0 ;>
52 26 52
8 9
< 1 1 1 =
Hð3; 3Þ ¼ 1 8 1 (4)
: ;
1 1 1
s2B ðkÞ
hðkÞ ¼ (5)
s2T
Fig. 2. Schematic flowchart for detecting joints on the rock mass surface.
where
9
color image to black and white one using the gray level conversion ½mT uðkÞ mðkÞ2 >
s2B ðkÞ ¼ >>
function as follows (Castleman, 2002): uðkÞ 1 u ðkÞ >
2 =
(6)
L1 h
X i >
>
gðx; yÞ ¼ T½f ðx; yÞ ¼ ½Rðx; yÞ þ Gðx; yÞ þ Bðx; yÞ=3 (1) >
>
s2T ¼ ði mT Þ2 Pi ;
i¼0
where g(x,y) is the output black and white image; f(x,y) is the input
color image; T is the operator on the f(x,y); R(x,y), G(x,y) and B(x,y)
ni X
L1 X
L1 X
L1
are the RGB components of color image. Pi ¼ ; mT ¼ ðiPi Þ; uðkÞ ¼ Pi ; mðkÞ ¼ ðiPi Þ (7)
N i¼0 i¼0 i¼0
(2) Smoothing and sharpening
where N is the number of the whole pixels, L is the number of gray
To emphasize the joints, the image needs to be smoothed and levels, Pi is the density function, and ni is the number of pixels
sharpened using the corresponding masks. Smoothing and sharp- within the i-th level.
ening can be performed by image enhancement technique using From Eqs. (5)e(7), we first calculate s2B ðkÞ for all k ranging from 1
following equation (William, 2007): to L, to find the level k, at which s2B ðkÞ is maximized. Then, the
maximum s2B ðkÞ is used as a threshold for binary encoding.
ZþN ZþN
g 0 ðx; yÞ ¼ gðx; hÞHðx x; y hÞdxdh (2) (4) Noise removal
N N
An image may be influenced by noise and interference from
where g 0 ðx; yÞ is the output smoothed or sharpened image, gðx; hÞ is several sources, including electrical sensor noise, photographic
the input smoothed or sharpened image, and Hðx x; y hÞ is the grain noise and especially blast-induced cracks. Therefore, corro-
mask for smoothing or sharpening. sion and swelling operations in automatic and manual procedures
In Eq. (2), we need to choose the impulse response H with low- must be performed for the binary images in order to remove un-
pass or high-pass characteristics, respectively, for smoothing and necessary noise which may cause errors in joint detection. Partic-
sharpening. In detail, we use the common 5 5 low-pass and 3 3 ularly, the manual procedure removes the blast-induced cracks on
high-pass masks for smoothing and sharpening purpose, which are the joint trace maps, while comparing the results of automatic
defined by detection with those of in situ survey.
K. Hong et al. / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 702e708 705
index is changed from zero (very good, i.e. very rough and fresh Fig. 4 shows a change of the squared sum of learning errors
unweathered surface) to 100 (very poor, i.e. slickenside and highly according to the learning iteration.
weathered surface). As a result, we set up the mean square error as 103 and the
Here, the number of neurons of hidden layer is first set to three maximum learning number as 1500 times. After training 1028
and it is finally determined via learning procedure to establish the times, the mean square error of the ANN is reached to threshold
most effective ANN structure. As a result of learning with one by
one increasing the neurons number of hidden layer, the learning Table 2
Comparisons of GSI values by two methods.
error is minimized at 13 neurons, when the highest accuracy of
ANN learning is reached. The transfer function between input layer No. GSI value Relative error (%)
and hidden one is tansigmoid and that between hidden layer and Previous New
output one is purelin.
1 95 94 1.1
In general, the output vector, containing all yk of the neurons of 2 97 98 1
the output layer, is not the same as the true output vector y*k (i.e. the 3 89 89 0
GSI value obtained from Fig. 1). The mean square error (Ek) between 4 86 88 2.3
these vectors is made during processing the inputeoutput vector 5 75 75 0
6 82 81 1.2
pair and can be calculated as follows:
7 78 80 2.6
8 92 94 2.2
1X * 2 9 76 74 2.6
Ek ¼ yk yk (10) 10 79 81 2.5
2 11 58 58 0
12 73 72 1.4
Learning dataset involves 310 data made according to the rock 13 91 92 1.1
mass classification and surface conditions of rock mass in GSI chart 14 62 61 1.6
(Fig. 1) given by Hoek and Brown (1997) and 40 data among them 15 77 76 1.3
are used for checking the accuracy of the ANN. 16 59 58 1.7
17 68 69 1.5
18 77 78 1.3
19 56 55 1.8
20 65 66 1.5
21 76 74 2.6
22 86 86 0
23 58 56 3.5
24 76 75 1.3
25 65 64 1.5
26 93 95 2.2
27 83 82 1.2
28 66 68 3
29 49 50 2
30 55 56 1.8
31 70 72 2.9
32 77 78 1.3
33 97 96 1
34 82 83 1.2
35 84 85 1.2
36 86 88 2.3
37 74 73 1.4
38 68 66 2.9
39 85 85 0
Fig. 4. Change of squared sum of learning error according to learning iteration in BP
40 56 58 3.6
ANN.
K. Hong et al. / Journal of Rock Mechanics and Geotechnical Engineering 9 (2017) 702e708 707
error, and the relative error between the GSI value predicted by The considered working, as main haulage drift, is driven into
ANN and the GSI value given in the GSI chart is below 3.6% relatively stable rock mass, which mainly consists of sandstone as a
(Table 2). footwall of the target coal seam. The rock mass is slightly weath-
Fig. 5 shows the interface, coded by using built-in tools of ered with one or two groups of joint sets, and the uniaxial
MATLAB 7.0, such as image processing, fractal analysis and ANN, compressive strength of rock samples is approximately 88 MPa,
based on the proposed method. which corresponds to the class I rock according to rock mass clas-
sification determined from total ratings (Bieniawski, 1978).
Images of rock mass in a working of about 800 m long are taken
5. Application example
every 10 m spacing. In total, we have taken 240 images on rock
mass for left and right walls, and roof side. Fig. 6 shows an example
We present an example for determining the GSI value of rock
for some sections.
mass around the working at Kangdong Coal Mine Complex in
Results of determining the GSI value for every section are listed
Korean Peninsula. The Coal Mine Complex is exploiting anthra-
in Table 3. As shown in Table 3, GSI values vary between 88 and 93
cite seam, formed during the Permian period, upper Paleozoic
in the considered sections.
era.
Fig. 6. Images of rock masses around the working taken at (a) 400e410 m, (b) 410e420 m, (c) 420e430 m, and (d) 430e440 m from the entry.
Table 3
GSI values of jointed rock mass in different sections.
Size of lattice Number of joints in every class Fractal dimension GSI value
network
Section a Section b Section c Section d Section a Section b Section c Section d Section a Section b Section c Section d