Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
PII: S0013-7952(16)30849-3
DOI: doi: 10.1016/j.enggeo.2017.08.004
Reference: ENGEO 4613
To appear in: Engineering Geology
Received date: 21 December 2016
Revised date: 13 July 2017
Accepted date: 1 August 2017
Please cite this article as: Jianqin Chen, Xiaojun Li, Hehua Zhu , Geostatistical method for
predicting RMR ahead of tunnel face excavation using dynamically exposed geological
information. The address for the corresponding author was captured as affiliation for all
authors. Please check if appropriate. Engeo(2017), doi: 10.1016/j.enggeo.2017.08.004
This is a PDF file of an unedited manuscript that has been accepted for publication. As
a service to our customers we are providing this early version of the manuscript. The
manuscript will undergo copyediting, typesetting, and review of the resulting proof before
it is published in its final form. Please note that during the production process errors may
be discovered which could affect the content, and all legal disclaimers that apply to the
journal pertain.
ACCEPTED MANUSCRIPT
PT
Road, Shanghai 200092, China
c
College of Engineering, Tibet University, Tibet 850000, China
* Corresponding author
RI
SC
Abstract:
Rock Mass Rating (RMR) is a rock mass classication system that is often used
NU
to select appropriate excavation methods and rock support systems in tunnel projects.
MA
This paper presents a geostatistical method for predicting RMR values quantitatively
ahead of excavation of the tunnel face. The study makes full use of geological
ED
rock mass quality and later predicts the RMR value using the kriging method.
EP
Predictions are constantly updated during the tunnel construction process. The
C
advantages of the proposed method are as follows: (1) The RMR prediction
AC
therefore, the resulting prediction can consider the geological conditions of the worst
variogram model that is updated by observation data on the new excavated faces, and
as the tunnel advances, the RMR prediction accuracy improves; and (3) The
proposed method is applied to a rock tunneling project, the Mingtang tunnel in Anhui
therefore, it has high potential as a tool for predicting RMR information ahead of
PT
Keywords: Rock Mass Rating Prediction; Spatial Variability; Model Uncertainty;
RI
Dynamic Updating
SC
NU
MA
T ED
C EP
AC
2
ACCEPTED MANUSCRIPT
List of Tables
Table. 1 Measured RMR values provided by the excavated tunnel faces during
construction.
estimation methods.
PT
Table 3. Cross-validation criteria for two estimation methods.
RI
Table 4. Period of the variogram models for two kriging estimation methods using
SC
nested hole-effect model. NU
Table 5. Cross-validation criteria for two estimation methods considering periodicity.
MA
T ED
C EP
AC
3
ACCEPTED MANUSCRIPT
List of Figures
PT
the measured RMR values provided by excavated tunnel faces.
RI
Fig. 4. RMR estimation interval.
SC
Fig. 5. RMR prediction with increasing number of data; the actual positions of the
NU
circles are detailed in Table 1.
MA
are calculated with the number of data equal to eight, fourteen and twenty; for
fitted using the lower bound of 2f , the top line represented by green is fitted
using the upper bound of 2f and the middle line represented by blue is fitted
C
AC
by automatic fitting; (a) K21+170; (b) K21+422; (c) K21+612; and (d)
K22+381.
Fig. 7. Kriging prediction interval considering model uncertainty; the blue line and
black line are the estimation results using the upper and lower bounds of the
variogram models; the red line is the true value calculated by the in situ tunnel
face mapping; (a) K21+170; (b) K21+422; (c) K21+612; and (d) K22+381.
4
ACCEPTED MANUSCRIPT
Fig. 8. Kriging variance at four prediction positions; the blue line and black line are
Kriging variance using the upper and lower bounds of the variogram models; (a)
uncertainty and spatial variability; the blue line is the upper bound of the
PT
estimation, the black line is the lower bound of estimation and the red line is the
RI
true value calculated by in situ tunnel face mapping; (a) K21+170; (b) K21+422;
SC
(c) K21+612; and (d) K22+381. NU
Fig. 10. RMR prediction intervals; (a) Model One: model uncertainty; and (b) Model
Fig. 11. Variogram model fitting; (a) RMRbasis; (b) joint condition (JC); (c) RQD; (d)
joint and bedding spacing (JS); (e) uniaxial compressive strength (UCS); and (f)
ED
Fig. 12. Scatterplot of the measured values and predicted values for two prediction
EP
methods.
C
Fig. 13. Variogram model fitting considering periodicity; (a) RMRbasis; (b) joint
AC
condition (JC); (c) RQD; (d) joint and bedding spacing (JS); (e) uniaxial
Fig. 14. Spectral analysis on RMRbasis and RMR components; (a) RMRbasis; (b) joint
condition (JC); (c) RQD; (d) joint and bedding spacing (JS); (e) uniaxial
5
ACCEPTED MANUSCRIPT
1. Introduction
are often limited and suffer from rough topography and difcult logistic conditions in
PT
mountainous terrain (Petronio et al., 2007). This may lead to large uncertainty and
RI
major geological risk in the construction phase of tunneling. Therefore, the rock mass
SC
classication systems known as empirical methods are still used in current
NU
engineering design and construction.
One of the most common classification systems is the Rock Mass Rating (RMR)
MA
that was developed by Bieniawski (1973, 1989). The RMR system incorporates six
Designation (RQD), joint and bedding spacing (JS), joint condition (JC), groundwater
T
condition (GW), and orientation of discontinuities with respect to the opening axis.
EP
Each of these parameters is classied and given a value or rating to express its
C
inuence on tunnel stability (Palmstrm, 2009). The RMR system is effective for
AC
evaluating the quality of rock mass surrounding the tunnel. It can be used to assist
process of tunneling.
actual rock mass class may not be consistent with the design value because of the
6
ACCEPTED MANUSCRIPT
variability of rock mass quality. Therefore, predicting the RMR value ahead of tunnel
plan a safe and economically efficient excavation method. Typically, RMR prediction
values range from 10 to 50 meters ahead of the tunnel face, and this is significant
because about one week is needed to adjust the excavation method to the geological
PT
conditions ahead. Currently, tunnel engineers usually rely on the seismic prediction
RI
method or ground penetration radar to predict rock mass quality and hazardous
SC
geological structures ahead of tunnel excavation. However, the inspection results of
NU
the seismic prediction method or ground penetration radar are qualitative, so they
observe and sample the rock mass on the tunnel faces. The RMR values can be
ED
constantly. However, this information has not been put into good use.
C
site characterizations. For tunnel site characterization, most current research uses
drillhole data (Kaewkongkaew et al., 2015) to interpolate RMR values directly for the
whole domain in the exploratory phase of tunnel project. Ferrari et al. (2014)
performed the ordinary kriging estimation to a valley area. These studies focus mainly
on the exploratory phase of a project. This means that the point-wise measurements
7
ACCEPTED MANUSCRIPT
from drill holes or geological eld surveys are used to interpolate the whole domain.
As the estimation process is conducted only once, the variogram model is constant.
advances. Therefore, the variogram model can be updated by using observation data
obtained from newly excavated tunnel faces. In addition, model uncertainty that is
PT
introduced in the process of experimental variogram fitting is not considered in the
RI
conventional analysis, which generates prediction uncertainty. The lower bound of the
SC
RMR estimation value represents the possible worst scenario in tunneling. Tunnel
NU
excavation plan based on this result would be considered extremely conservative. The
upper bound of RMR estimation value represents the best possible scenario, so tunnel
MA
(Pyrcz and Deutsch, 2014). The variogram fitting of RMR and its components must
EP
predict RMR values ahead of the excavation face during tunnel construction. This
research offers several contributions: (1) The RMR prediction is based on dynamic
observations of tunnel faces during excavation; (2) The RMR prediction uncertainty is
quantified by taking into account model uncertainty and spatial variability; and (3)
8
ACCEPTED MANUSCRIPT
quantitatively ahead of the tunnel face in Section 3, discussing the application of the
PT
2. Site Description
RI
To demonstrate application of the geostatistical method for RMR prediction for
SC
tunnel engineering, the Mingtang Tunnel is selected as a case study. Mingtang Tunnel
NU
is a rock tunnel situated in Yuexi County, Anhui Province, China (Fig. 1). This is part
of the Yue-wu highway that connects Wuhan city and Shanghai city. The tunnel was
MA
excavated using the drill and blast method. The total length of the tunnel is 7.548 km,
and the maximum burial depth is approximately 562 m. As shown in Fig. 2(a), the
ED
geology of the site area is mainly composed of granite and gneiss. Three major faults
T
transect rock formations in the tunnel alignment. The study area is in the interval
EP
is sampled from the excavation faces in forty-three discrete positions, and their
AC
corresponding rock mass quality is evaluated. The positions and measured RMR
values are shown in Table 1. Fig. 2(b) shows the positions of the tunnel face relative
3. Methodology
9
ACCEPTED MANUSCRIPT
Dynamic drill and blast excavation provides a useful opportunity for engineers
and geologists to observe and sample rock mass. Most difficult, however, is obtaining
the discontinuity information safely and efficiently. In our study, RQD value, joint and
bedding spacing (JS) and joint condition (JC) (except inlling and weathering) of
PT
actual RMR values can be obtained using the photogrammetry-based tunnel face
RI
mapping technique (Haneberg, 2008; Sturzenegger et al., 2009; Chen et al., 2016; Li
SC
et al., 2016; Zhu et al., 2016). Uniaxial compressive strength (UCS) of rock mass is
NU
measured through the uniaxial compression test. Groundwater condition (GW),
Adjacent areas show similar geological conditions, so the RMR data sampled
ED
from locations closer together tend to be more similar than those far apart. This spatial
T
parameters of the variogram best fitting the data are first conducted using R software
AC
(R. Core Team, 2015) with the GSTAT package (Pebesma, 2004). Then, Kriging is
The RMR values range from 44 to 74 with a mean of 61.5 and a standard
10
ACCEPTED MANUSCRIPT
RMR distribution is veried using the Shapiro-Wilk test (Shapiro and Wilk, 1965).
Moreover, trend analysis results show that there is no trend in the RMR. This
PT
3.2.2.1 Model uncertainty
RI
Construction of the semivariogram, which is a mathematical model that captures
SC
the spatial correlation among data, is a very important step in any geostatistical
NU
analysis. The semivariogram is a measure of variability that increases as samples
become more dissimilar. The variogram is dened as the expected value of a squared
MA
fitting with different parametric variogram models, such as the Spherical model,
Exponential model, and Gaussian model. There are four main approaches for
estimating the parameters of the variogram model: visual, least square regression
(Trangmar et al., 1985), weighted least squares (Cressie, 1985), and likehood methods
(Cressie, 1991). However, the process of variogram model fitting introduces some
11
ACCEPTED MANUSCRIPT
The method of least squares is used to fit the rising limb of the parametric
variogram to the experimental variogram, while holding the sill value equal to the
sample variance, so that range can be calculated. The sample variance is treated as a
PT
random variable. Hence, the uncertainty in determining the range is also considered
RI
by least square regression. Estimation of the sample variance is subject to uncertainty,
SC
which is estimated by Rehfeldt et al. (1992):
2 4
NU
Var ( 2f ) f (2)
N*
where 2f is the sample variance and N * is the number of independent
MA
measurements that can be obtained by dividing the tunnel into independent segments.
Two theoretical variogram models can be fitted by using the lower and upper
AC
upper and lower bounds of RMR values ( RMRupper and RMRlower ) that accounts for
12
ACCEPTED MANUSCRIPT
estimation is a measure of the possible variation in value, and hence the uncertainty of
the estimated value. If the standard error is small, then the possible statistical variation
When only considering the spatial variability, the estimation interval of RMR can
PT
be expressed as following with 68% confidence:
RI
RMR RMR RMRPred RMR RMR (4)
SC
where RMR is the mean and RMR is standard deviation of the kriging estimation.
NU
This prediction result only considers the spatial variability.
As is shown in Fig. 4, when considering both the model uncertainty and spatial
min{RMRlower lower , RMRupper upper } RMRPred max{RMRlower lower , RMRupper upper } (5)
T
where RMRupper and RMRlower are the kriging estimated values using the bounds of
EP
theoretical variogram models (defined in Eq. 3), and upper and lower are the
C
The spatial variability can be updated using observation data on newly excavated
tunnel faces. To evaluate the dynamic update effect of RMR prediction, different
number of observation data were used to construct the variogram model. Then, the
13
ACCEPTED MANUSCRIPT
predictions are made and green circles represent the excavated tunnel positions with
field measured RMR values. The actual positions of the circles are detailed in Table 1.
The positions of the circles relative to the tunnel are shown in Fig. 2(b). There are
three clouds with different colors next to each of the red circles, and each cloud
PT
contains different numbers of observation data ranging from eight, fourteen and
RI
twenty. Three variogram models are constructed based on the observation data in the
SC
clouds and they predict the RMR values at the position of the red circle. The goal is to
NU
show how the RMR prediction changes as the number of measurements increase. The
prediction position advances to different red circles, such as tunnel excavation. For
MA
each prediction, the number of data surrounded by the clouds is the same.
3.2.4 Two kriging estimation methods: Direct Method and Indirect Method
ED
Two types of prediction strategies are conducted in our study. The first approach
T
is called the Direct Method. In this approach, the base RMR value is fit with a
EP
variogram model and then estimated directly. The base RMR is composed of five
C
parameters:
AC
The last parameter (orientation of discontinuities with respect to the opening axis) is
excluded from the kriging estimation because it depends on the characteristics of rock
14
ACCEPTED MANUSCRIPT
The second approach is called the Indirect Method. In this approach, parameters
estimated RMR is obtained by adding the ratings of each variable. The kriging
PT
RMR
2
2
UCS
2
RQD
2
JS2 JC
2
GW
2
RI
(8)
2cov( RQD, JS ) 2cov( RQD, JC ) 2cov( RQD, GW )
SC
2cov( JS , JC ) 2cov( JS , GW ) 2cov( JC , GW )
where UCS
2
, RQD
2
, JS
2
, JC
2
and GW
2
are the kriging variances for each component.
NU
MA
Fig. 6 shows the fitted results of variogram models at four positions, respectively.
T
For each prediction position, three experimental variogram models are calculated with
EP
the number of observation data equaling to eight, fourteen and twenty. The aim of
C
these models is to investigate how prediction accuracy changes with the number of
AC
observation data.
variogram models. The bottom line (represented by purple) is generated using the
lower bound of 2f , the top line (represented by green) is generated using the upper
fitting. The bottom line and top line almost encapsulate the variation range of the sill
15
ACCEPTED MANUSCRIPT
measurement data. In Fig. 6(a), the sills of the variogram models decrease gradually.
The sills in Fig. 6(b) and Fig. 6(c) increase first and decrease later. In contrast, the
sills increase gradually in Fig. 6(d). Therefore, with newly added data, the variogram
PT
model is more consistent with the actual situation, meaning that the uncertainty of the
RI
variogram model decreases.
SC
Fig. 7 shows the Kriging estimation result with a different number of observation
NU
data. The blue line and black line are the estimation results using the upper and lower
bounds of the variogram models. The red line is the true value calculated by the in
MA
situ tunnel face mapping. The prediction interval considers the variogram model
uncertainty described in Section 3.2.2. The prediction interval covers the true value
ED
gradually with an increase of the measured data, which indicates that prediction
T
accuracy increases.
EP
Fig. 8 shows the corresponding kriging variance using upper and lower bounds
C
of variogram models. The upper and lower bounds of kriging variance are used to
AC
quantify spatial variability. The final RMR estimation interval allows for both model
uncertainty and spatial variability, as shown in Fig. 9. When the spatial variability is
incorporated into the prediction result, the prediction interval increases to cover the
true value.
The prediction result of the eight positions is summarized in Fig. 10. Model One
16
ACCEPTED MANUSCRIPT
refers to prediction that considers only model uncertainty. Model Two refers to
prediction that considers both model uncertainty and spatial variability. Fig. 10(a)
shows the RMR prediction intervals for Model One. Fig. 10(b) shows the prediction
intervals for Model Two. In Fig. 10, the prediction intervals decrease with an increase
of measurement data, especially at the position K21+170. For the position K21+170,
PT
when the number of measurement data is eight, the upper bound of 2f is large,
RI
which leads to a large RMR estimation interval with the lower bound of the estimated
SC
RMR value close to zero. When the number of measurement data increases to twenty,
NU
the upper bound of 2f decreases, which leads to a smaller RMR estimation interval
with the lower bound of estimated RMR value approximating to the true value.
MA
Reduction of the prediction interval does not decrease the prediction accuracy.
For the Indirect Method, the components of RMR are estimated separately as
T
individual variables and fitted with separate variogram models from Fig. 11(b) to Fig.
EP
11(f). The parameters of these variogram models are listed in Table 2. For the
C
variogram model of each component, the range of GW is the largest, the ranges of
AC
UCS are secondary, and the ranges of JC, RQD and JS are the smallest. Because
ground water can flow in the fractures of rock mass, the influence area is larger when
comparing the fracture parameters (JC, RQD and JS) and strength parameter (UCS).
However, the fractures are generated by tectonic actions and thus show larger
variability and heterogeneities in the domain. Moreover, the sill of the GW is largest
17
ACCEPTED MANUSCRIPT
because three faults traverse the Mingtang tunnel and dramatically change the
groundwater conditions. There is abundant ground water neighboring the faults, while
there is limited ground water in the regions far away from the tunnel.
In the Direct Method, the basic RMR value (defined in equation 6) is fitted with
a variogram model, as is shown in Fig. 11(a). The sill of RMRbasis (67.00) is affected
PT
by the largest sill of its component, which is GW with the value of 55.55. In addition,
RI
the range of RMRbasis (171.33) is affected by the smallest sill of its component,
SC
which is JC with a value of 175.45. NU
Fig. 12 shows the scatterplot between measured values and predicted values for
two estimation methods. It reveals a near- linear positive relationship between the
MA
measured value and predicted value for the two models. The correlation coefficients
for the Indirect Method and Direct Method are 0.73 and 0.7, respectively. Three
ED
performance metrics, the mean error (ME), the root mean squared error (RMSE) and
T
the mean standardized prediction error (MSPE), are used to evaluate the overall
EP
predictive accuracy. The ME should be close to zero. The RMSE measures the
C
difference between the predicted values and measured values, and this should be as
AC
An accurate model should have an MSPE close to zero. As shown in Table 3, the
Indirect Method has the smaller absolute values of ME and RMSE, with the MSPE
value being closer to zero. Therefore, the Indirect Method has better estimation
18
ACCEPTED MANUSCRIPT
phenomenon is sometimes referred to as the hole effect (Journel and Huijbregts, 1978;
PT
and is characterized by undulations in the variogram. Therefore, nested variograms
RI
composed of the Spherical model, Periodic model and Hole-effect model are
SC
constructed to fit the experimental variogram models of the whole RMR and each
NU
component of RMR.
Spectral analysis (Lomb, 1976) is used to detect cyclicities in the RMRbasis and
MA
RMR component. As shown in Fig. 14(a), the power spectral density of RMRbasis is
dominated by three significant peaks at frequencies of 0.0003, 0.0006 and 0.0021 that
ED
correspond to periods of 1421.8, 507.8 and 122.6 meters, respectively. In the RMR
T
component, there are also strong cyclicities indicated by peaks in the power spectral
EP
density. As shown in Table 4, the periods computed by the variogram model are
C
similar to the periods computed by the spectral analysis. This cyclicity suggests a
AC
strong inuence of tectonic actions of the geological body on the quality of rock mass.
Cross validation is performed to test the fit of the nested models to the data.
Their correlation coefficients both increase to 0.76. Three performance metrics are
used to evaluate the overall predictive accuracy. As is shown in Table 5, the Indirect
and Direct Methods have almost similar estimation accuracies for ME, RMSE and
19
ACCEPTED MANUSCRIPT
MSPE values after considering the periodicity in the variogram models. Furthermore,
prediction results of the Indirect Method, with the exception of RMSE, is better after
when the Indirect Method considers periodicity, the variability of the prediction
PT
increases as each component is predicted separately.
RI
SC
5. Conclusions NU
This paper presents a geostatistical method for predicting RMR values
quantitatively ahead of tunnel face excavation by making full use of the geological
MA
information collected from the tunnel face and considering the dynamic construction
variability and model uncertainty, newly added geological data, and periodicity of the
T
geological condition.
EP
variability and model uncertainty. Model uncertainty is derived from the variogram
AC
model fitting process, and spatial variability originates from variation in the estimated
value. The resulting RMR prediction interval incorporates the best scenario case and
worst scenario case. The tunnel excavation method is based on the limits of the
interval representing different levels of risk. Furthermore, considering the model and
20
ACCEPTED MANUSCRIPT
Second, RMR prediction accuracy increases with newly added geological data.
and Indirect Method. The Direct Method estimates the RMR value directly, while the
Indirect Method estimates the components constituting the RMR system separately. In
PT
our case study, the Indirect Method shows better estimation accuracy than the Direct
RI
Method. Moreover, the nested variograms composed of Spherical model, Periodic
SC
model and Hole-effect models perform relatively well in terms of prediction accuracy
NU
for both the Indirect Method and Direct Method. When the Indirect Method considers
Acknowledgments
ED
Foundation of China (NSFC 41272289, 41130751), the Science and Technology Plan
EP
Research Funds for Central Universities. Special thanks are due to Prof. Yoram Rubin
AC
from University of California, Berkeley for his valuable instructions regarding the
geostatistical method.
References
Alimoradi, A., Moradzadeh, A., Naderi, R., Salehi, M.Z., Etemadi, A., 2008.
21
ACCEPTED MANUSCRIPT
23(6), 711-717.
PT
of the South African Institution of Civil Engineers, v. 15, pp. 335-344.
RI
Bieniawski, Z.T., 1989, Engineering rock mass classifications: a complete manual for
SC
engineers and geologists in mining, civil, and petroleum engineering, New York,
NU
Wiley, xii, 251pp.
Cressie, N., 1985. Fitting variogram models by weighted least squares.Journal of the
MA
Cressie, N., 1991. Statistics for spatial data. John Wiley & Sons, NY, pp.900.
ED
Chen, J., Zhu, H., Li, X., 2016. Automatic extraction of discontinuity orientation from
T
rock mass surface 3D point cloud. Computers & Geosciences 95, 18-31.
EP
Choi, J.Y., Lee, C.I., 2007. An estimation of rock mass rating using 3D- indicator
C
Proceedings of the 11th congress of the international society for rock mechanics,
Choi, Y., Yoon, S.Y., Park, H.D., 2009. Tunneling analyst: a 3D GIS extension for
rock mass classification and fault zone analysis in tunneling. Computers &
22
ACCEPTED MANUSCRIPT
Exadaktylos, G., Stavropoulou, M., 2008. A specific upscaling theory of rock mass
1102-1125.
Ferrari, F., Apuani, T., Giani, G.P., 2014. Rock Mass Rating spatial estimation by
PT
geostatistical analysis. International Journal of Rock Mechanics and Mining
RI
Sciences 70, 162-176.
SC
Gholami, R., Rasouli, V., Alimoradi, A., 2013. Improved RMR rock mass
NU
classification using artificial intelligence algorithms. Rock mechanics and rock
Haneberg, W.C., 2008. Using close range terrestrial digital photogrammetry for 3-D
rock slope modeling and discontinuity mapping in the United States. Bulletin of
ED
New York, USA: Oxford University Press. Jones DR, A taxonomy of global
C
23, 345-383.
Journel, A.G, Huijbregts, C.J., 1978. Mining Geostat. NewYork, NY: Academic Press,
Inc.
Kaewkongkaew, K., Phien-Wej, N., Kham-ai, D., 2015. Prediction of rock mass along
23
ACCEPTED MANUSCRIPT
La Pointe, P.R., 1980, January. Analysis of the spatial variation in rock mass
PT
Palmstrm, A., 2009. Combining the RMR, Q, and RMi classification systems.
RI
Tunnelling and Underground Space Technology, 24(4), 491-492.
SC
Pebesma, E.J., 2004. Multivariable geostatistics in S: the gstat package. Computers &
NU
Geosciences, 30(7), 683-691.
Pyrcz, M.J. and Deutsch, C.V., 2003. The whole story on the hole effect.Geostatistical
MA
Pyrcz, M.J. and Deutsch, C.V., 2014. Geostatistical reservoir modeling. Oxford
ED
Santos, V., da Silva, A.P.F., Brito, M.G., 2015. Prediction of RMR Ahead Excavation
24
ACCEPTED MANUSCRIPT
Trangmar, B.B., Yost, R.S. and Uehara, G., 1986. Application of geostatistics to
You, K., Lee, J. S., 2006. Estimation of rock mass classes using the 3-dimensional
PT
Technology 21(3), 229-229.
RI
Zhu, H., Wu, W., Chen, J., Ma, G., Liu, X., Zhuang, X., 2016. Integration of three
SC
dimensional discontinuous deformation
NU analysis (DDA) with binocular
Li, X., Chen, J., Zhu, H., 2016. A new method for automated discontinuity trace
mapping on rock mass 3D surface model. Computers & Geosciences 89, 118-131.
ED
Petronio, L., Poletto, F., Schleifer, A., 2007. Interface prediction ahead of the
T
Egaa, M., Ortiz, J.M., 2013. Assessment of RMR and its uncertainty by using
AC
83-90.
Rehfeldt, K.R., Boggs, J.M., Gelhar, L.W., 1992. Field study of dispersion in a
25
ACCEPTED MANUSCRIPT
persistence on high mountain rock slopes and large landslides using terrestrial
remote sensing techniques. Natural Hazards and Earth System Science 9(2),
267-287.
Pinheiro, M., Vallejos, J., Miranda, T., Emery, X., 2016. Geostatistical simulation to
PT
map the spatial heterogeneity of geomechanical parameters: A case study with rock
RI
mass rating. Engineering Geology 205, 93-103.
SC
Webster, R., Oliver, M.A., 2007. Geostatistics for environmental scientists. John
NU
Wiley & Sons.
Zhang, L., Einstein, H.H., 1998. Estimating the mean trace length of rock
MA
Rubin, Y., 2003. Applied stochastic hydrogeology. Oxford University Press, New
ED
26
ACCEPTED MANUSCRIPT
PT
RI
SC
NU
Fig. 1. Location of the Mingtang Tunnel in China.
MA
T ED
C EP
AC
represented by cyan, and gneiss is represented by yellow); and (b) Positions of the
27
ACCEPTED MANUSCRIPT
PT
RI
SC
Fig. 3. Frequency distribution histogram of the Rock Mass Rating.
NU
MA
T ED
C EP
AC
28
ACCEPTED MANUSCRIPT
Fig. 5. RMR prediction with increasing number of data; the actual positions of
PT
the circles are detailed in Table 1.
RI
SC
NU
MA
T ED
EP
Fig. 6. Variogram fitting at four positions; (a) K21+170; (b) K21+422; (c) K21+612;
C
29
ACCEPTED MANUSCRIPT
PT
RI
SC
Fig. 7. Kriging prediction interval considering model uncertainty; (a) K21+170;
NU
(b) K21+422; (c) K21+612; and (d) K22+381.
MA
T ED
C EP
AC
Fig. 8. Kriging variance at four prediction positions; (a) K21+170; (b) K21+422;
30
ACCEPTED MANUSCRIPT
PT
RI
SC
Fig. 9. Prediction interval at four prediction positions with consideration of
NU
model uncertainty and spatial variability; (a) K21+170; (b) K21+422; (c) K21+612;
31
ACCEPTED MANUSCRIPT
PT
RI
SC
NU
MA
T ED
EP
Fig. 10. RMR prediction intervals; (a) Model One: model uncertainty; and (b)
C
32
ACCEPTED MANUSCRIPT
PT
RI
Fig. 11. Variogram model fitting; (a) RMRbasis; (b) joint condition (JC); (c)
SC
RQD; (d) joint and bedding spacing (JS); (e) uniaxial compressive strength (UCS);
Fig. 12. Scatterplot of the measured values and predicted values for two
prediction methods.
33
ACCEPTED MANUSCRIPT
PT
RI
Fig. 13. Variogram model fitting considering periodicity; (a) RMRbasis; (b) joint
SC
condition (JC); (c) RQD; (d) joint and bedding spacing (JS); (e) uniaxial
Fig. 14. Spectral analysis on RMRbasis and RMR components; (a) RMRbasis; (b) joint
AC
condition (JC); (c) RQD; (d) joint and bedding spacing (JS); (e) uniaxial
34
ACCEPTED MANUSCRIPT
Table. 1 Measured RMR values provided by the excavated tunnel faces during
construction.
Number Position Measured RMR value
1 K17+599 64
2 K17+620 56
3 K17+640 61
4 K17+681 59
5 K19+676 59
6 K20+124 54
PT
7 K20+205 63
8 K20+241 69
9 K20+347 66
RI
10 K20+375 66
11 K20+402 62
SC
12 K20+432 68
13 K20+456 68
NU
14 K20+475 74
15 K20+583 60
16 K20+650 74
MA
17 K20+880 74
18 K20+905 74
19 K20+930 72
20 K20+960 74
ED
21 K20+977 60
22 K21+037 50
T
23 K21+062 60
EP
24 K21+170 57
25 K21+200 57
26 K21+405 64
C
27 K21+422 64
AC
28 K21+503 44
29 K21+547 45
30 K21+612 47
31 K21+955 54
32 K21+987 52
33 K22+381 61
34 K22+533 59
35 K22+575 57
36 K22+881 52
37 K22+939 58
38 K23+790 68
35
ACCEPTED MANUSCRIPT
39 K24+090 65
40 K24+120 65
41 K24+149 65
42 K24+177 65
43 K24+708 59
PT
Parameters in Direct Indirect Method
estimation Method
RI
Methods
Model parameters RMRbasis JC RQD JS UCS GW
SC
Model Sph* Sph* Sph* Sph* Sph* Sph*
Sill 67.00 16.55 1.81 5.17 1.04 55.55
NU
Range 171.33 175.45 260 163.68 350 1107.85
Nugget 0 0 0 0 0 0
MA
Criterion
EP
ME 0.0558 -0.2035
RMSE 1.1179 5.4462
C
Table 4. Period of the variogram models for two kriging estimation methods
using nested hole-effect model.
Parameters in Direct Indirect Method
estimation Method
Methods
Period (m) RMRbasis JC RQD JS UCS GW
36
ACCEPTED MANUSCRIPT
PT
MSPE 0.0011 -0.0051
RI
SC
NU
MA
T ED
C EP
AC
37
ACCEPTED MANUSCRIPT
Highlights
PT
The RMR prediction uncertainty is quantified by accounting for spatial
variability and model uncertainty.
The prediction of RMR values allows for periodicity of geological conditions.
RI
SC
NU
MA
T ED
C EP
AC
38