International Journal of Mining Science and Technology xxx (2017) xxxxxx

Contents lists available at ScienceDirect

International Journal of Mining Science and Technology

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

Design of crown pillar thickness using finite element method

and multivariate regression analysis
Kumar Hemant , Deb Debasis, Chakravarty D.
Department of Mining Engineering, Indian Institute of Technology (Indian School of Mines), Dhanbad 826004, India
Department of Mining Engineering, Indian Institute of Technology, Kharagpur 721302, India

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

Article history: Minerals are now being extracted from deep mines due to drying up of resource in shallow ground. The
Received 15 November 2016 need for suitable supports and ground control mechanisms for safe mining necessitates proper pillar
Received in revised form 8 February 2017 design with filling technology. In addition, high horizontal stress may cause collapse of hanging wall
Accepted 10 March 2017
and footwall rocks, hence designing of suitable crown pillars is absolutely necessary for imposing overall
Available online xxxx
safety of the stopes. This paper provides a methodology for the evaluation of the required thickness of
crown pillars for safe operation at depth ranging from 600 m to 1000 m. Analyses are conducted with
Keywords:
the results of 108 non-linear numerical models considering Drucker-Prager material model in plane
Cut and fill mining
Crown pillar
strain condition. Material properties of ore body rock and thickness of crown pillars are varied and safety
FEM factors of pillars estimated. Then, a generalized statistical relationship between the safety factors of
Regression model crown pillars with the various input parameters is developed. The developed multivariate regression
Design chart model is utilized for generating design/stability charts of pillars for different geo-mining conditions.
These design charts can be used for the design of crown pillar thickness with the depth of the working,
taking into account the changes of the rock mass conditions in underground metal mine.
2017 Published by Elsevier B.V. on behalf of China University of Mining & Technology. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction induced by deep high-stress conditions have exceeded 40% of the

Most of the underground metal mines in India and abroad are
being operated at depths beyond 600 m from the surface and some
of the mines in India are being planned to mine at even greater
depths, beyond 1000 m. The depth of a few mines in South Africa
is more than 1000 m. As for example, the Anglogold mine has
reached 3700 m, and West Driefovten golden mine arrived at a
depth of 6000 m. In addition, in Canada, Australia, United States
and other countries, minerals are being mined at a depth over
one thousand meters [1]. At greater depth high stress conditions
exist and hence, the roof and walls of drives may be damaged
due to high convergence and side spalling. Crown pillars are
reported to be cracked or damaged due to the excessive load of
overlying strata and high horizontal stress. About 43% of the fatal-
ities in South African Gold mines are related to the depths
exceeded 2500 m. In those conditions, rock mass are subjected to
very high states of rock stress causing seismicity and rock bursting
[2]. Accidents due to severe damage in galleries and rock burst
Corresponding author at: Department of Mining Engineering, Indian Institute of
Technology (Indian School of Mines), Dhanbad 826004, India.
E-mail address: hemantismd@gmail.com (H. Kumar).
total coal mines accidents in China [3].
The stability of deep high stress rock masses is influenced by
many factors, among which the structural characteristics and
mechanical states of rock masses are the primary parameters. Sta-
bility of structures in deep underground mines can be broadly
divided into three different categories; global, regional and local
depending on volume of rock involved [4]. The major factors which
play important role in pillar stability are: (1) effect of depth of cover;
(2) effect of size of excavation; (3) effect of horizontal stresses; (4)
effect of rock mass properties; (5) effect of backfilling; (6) effect of
reinforcement; and (7) effect of the orebody dipping [513].
Although a thick crown pillar provides support for the hanging
wall and aid with overall stability of the stopes, but at the same
time it may be uneconomical from the mineral conservation point
of view. Thus, optimization of pillar dimension is very important
for metalliferous mines [14]. Estimation of the optimum thickness
of crown pillar is complex, and is generally based on practical
experience and various empirical techniques [6,15]. The stability
of pillars in deep underground metal mines depends upon the
strength or stability of the rock mass surrounding the excavations
and upon the stresses induced in this rock. These induced stresses
are a function of the shape of the excavations and the in-situ

http://dx.doi.org/10.1016/j.ijmst.2017.06.017
2095-2686/ 2017 Published by Elsevier B.V. on behalf of China University of Mining & Technology.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
2 H. Kumar et al. / International Journal of Mining Science and Technology xxx (2017) xxxxxx

stresses which existed before the creation of the excavations. The

magnitudes of pre-existing in-situ stresses have been found to be
varying widely, depending upon the geological history of the rock
mass in which they are measured [16,17].
In India, as well as in abroad, there has been very few works
devoted for the comprehensive design of dimension of crown
and sill pillars and also to perform stability analysis of the entire
stope for deep seated deposits. Existing design methods are limited
in scope and the mostly used ones are the empirical techniques or
the rules of thumb approach. Some researcher suggested a
methodology for the evaluation of sinkhole hazard in soft carbon-
ate rocks, combining seismic and mine engineering methods [18].
In the literature it is found that seismic P-wave velocity analysis
can be used with a revised relationship that accounts for long-
term surface stability of the crown pillar. The empirical analysis
on the stability of the crown pillar has been performed considering
the crown pillars thickness and its span as variables of function of
its length and width respectively [18]. There is hardly any study
devoted for parametric analysis of crown pillar design in cut and
fill stopes in deep mining conditions considering several parame-
ters as mentioned above. This study precisely focused on scientific
design of horizontal pillars (crown pillars) considering the chang-
ing rock mass conditions, variable width of the excavations and
increasing depth of the excavations.
The main purpose of this research is to develop a statistical
model for the safety factor of crown pillars followed by develop-
ment of simple design charts which can be used for the design of
crown pillar thickness with the depth of the working, taking into
account the changes of the rock mass conditions in the studied
underground metal mine. The charts is prepared in such a way that
it is easy to use for field personnel and contains rock mass condi-
tions, thickness of ore body, thickness of pillars, rock mass proper-
ties and depth variables. In this study, numerical analysis are
conducted based on parametric study considering a steeply dip-
ping ore body having reducing width with depth of mining (the
most common type of orebody shape in India), changing rock mass
conditions such as geological strength index (GSI), uniaxial com-
pressive strength (UCS or rci), modulus of elasticity (E), and
variable thickness of crown pillars. These analyses have been
conducted on 108 non-linear numerical models with varying
geo-mining conditions with reference to a case study mine located
in the eastern part of India. This study develops a generalized sta-
tistical relationship (multivariate regression model) between the
safety factors of crown pillars with the above mentioned input
parameters for further utilization in generating design charts of
pillars for different geo-mining conditions.
2. Description of the case study mine and models parameters
The ore body of the case study mine has varying dip as well as
width as shown in Fig. 1 (transverse section of case study mine).
Finite element models of the stopes and pillars are developed
below 596 m from the surface at various levels namely 685, 750,
815, and 880 meter level (mL) with a level interval of approxi-
mately 65 m. The solid model (Fig. 1), the meshed model with
loading conditions (Fig. 2) and the in-situ models are developed
considering the actual dip of the ore body, thickness of ore body
at different levels and levels at designated depth. Stopes starting
from 685 mL, and 750 mL are filled up by mill tailings in the exca-
vated zone in between the in-situ stope pillars (rib pillars) left dur-
ing the ore extraction. The excavated areas in the stopes starting
from 815 mL and 880 mL are completely filled up without leaving
rib pillars due to narrow ore body. It is noticed that an excavated
height of 4.5 m is left after the last slice below the crown pillar
to analyze the worst possible stress conditions in the pillars. In
levels 815 mL and 880 mL, post pillars are not needed since width
Fig. 1. Geometric model of case study mine (transverse section).
of the ore body is less than 8 m as suggested by Directorate General
of Mines Safety (DGMS) India. For example, if the thickness of sill
and crown pillar is 6 m each, horizontal pillars (Crown of the
stopped out area) of 14.8 m (6
2 + 2.8) are left considering level
drive of height 2.8 m. The general conventions in Indian metal
mining industry is that the portion of the horizontal pillar above
the level drive is known as sill pillar and the portion of horizontal
pillars below the level drive is known as crown pillar [19,20]. Sim-
ilarly, separate solid models are developed to represent a sill and
crown pillar thickness of 4 m, 5 m, and 7 m keeping all other
dimensions same. This variation in geometrical model has been
done to analyze the stability of crown pillar with variation of other
parameters viz. depth of mining, GSI, UCS, and E.
3. Geotechnical study and rock mass properties of a case study
mine
Geotechnical study of the drives, drifts, and pillars has been
conducted to determine the condition and spacing of joints so that
RMR of hanging wall, footwall and orebody rock mass can be
obtained. It may be noted that hanging wall of drives and pillars
are cracked and fractured at the deeper level as observed visually
in the mine as shown in Fig. 3a and b. These conditions are the evi-
dence of stress concentrations or relaxations around the stopes and
pillars.
Geotechnical study shows that the variation of GSI of the ore-
body rock mass ranges between 42 and 75, uniaxial compressive
strength lies between 46.23 MPa and 86.73 MPa, and modulus of
rigidity ranges between 9.43 GPa and 16.22 GPa. These data sug-
gest variation in strength properties of the rock mass and hence
accordingly this study guides wide range of rock properties as
modeling parameters. Three variations of GSI were considered for
the study, viz., 50, 60 and 70. Similarly, the uniaxial compressive
strength (UCS) of orebody is varied as 50, 65 and 75 MPa, modulus
of elasticity of intact rock is varied as 10, 15 and 20 GPa. The
strength and modulus of rigidity of rock mass are estimated based
on UCS, E, GSI and mi [21,22]. Altogether, 108 finite element mod-
els have been developed, based on all possible interactions
(namely, modulus of orebody (3)
thickness of sill/crown pillar
(4)
RMR/GSI of orebody (3)
uniaxial compressive strength
(3)). Apart from this, 27 in-situ or pre-mining finite element mod-
els are also developed by varying rock mass parameters. These

Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
 H. Kumar et al. / International Journal of Mining Science and Technology xxx (2017) xxxxxx 3

(a) Finite element meshes of transverse section of orebody.

(b) Zoomed view of 685 meters level.

Fig. 2. Finite element meshes showing transverse section of orebody below 596 m.

this model resembles the final stage of stoping operation where

maximum ore recovery has been done between levels, and stope
above is approximately filled with backfill material.

4.1. Meshed model and loading conditions

Fig. 3. Drive in weak rock condition and supported by concrete arching and rock
bolting.
models are analyzed in plane strain conditions considering non-
linear material behavior based on Drucker-Prager failure criteria.
4. Finite element models
Rock mass, orebody and openings are modeled with 6-noded
quadratic triangular elements. These elements have two degrees
of freedom at each node: translations in coordinate axes, i.e., x
and y directions. The finite element model of case study mine rep-
resents the vertical transverse-section along, approximately the
middle portion of W4# stope block and hence, plane strain consti-
tutive material behavior is assumed. It is important to note that
The solid models as well as the finite element meshes are devel-
oped using ANSYS software tool. The bottom boundary of the mod-
els is roller-constrained. A uniform load distribution of gradient
0.053 MPa/m is applied on the sides to simulate the in-situ stress
condition of the mine. Fig. 2 shows the meshed model of ore body
and its surrounding area below 596 mL, depicting loading and
boundary conditions. The meshing of complete model of 6 m sill/
crown pillar thickness produced an average of 12,081 6-noded tri-
angular elements and 24,439 nodes. A quadratic triangular ele-
ment consisting of 6 nodes is mostly suitable for two
dimensional (2D) stress analysis with material non-linearity. In
general, finer mesh is developed in the stoping zone for better eval-
uation of displacements, stresses and strains. Coarse mesh is devel-
oped in the rock mass away from the mining effected zones.
4.2. In-situ model
In-situ model is developed to analyze the stress conditions of
rock prior to excavation of the mine. For this purpose, in-situ mod-
els of same dimension as the solid models (previously mentioned)
have been developed, which includes orebody, hangwall and foot-
wall. In-situ models provide the ideas about stress and displace-
ment of rock mass prior to mining or excavation. Openings or

Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
4 H. Kumar et al. / International Journal of Mining Science and Technology xxx (2017) xxxxxx
excavation in rock mass causes re-orientation of stress regime and
change in displacement in the rock mass. A total of 27 in-situ finite
element models are developed by varying the material properties
of rock mass. An additional pressure of 16.98 MPa is applied uni-
formly on top of the model to consider the load of overlying strata.
A gradient-horizontal pressure is applied from left to right and also
from right to left the in meshed in-situ model.
5. Analysis of the finite element modeling results
Results of finite element models are analyzed in terms of stres-
ses, displacements and extent of yield zones in the pillars. Based on
these results, safety factor are determined in the middle of crown
pillars and relationship between safety factors and input parame-
ters are determined using multiple regression analysis technique.
5.1. Stress distribution in pillars and stopes
The major and minor principal stresses represent the maximum
and minimum normal stresses at a particular point in the rock
mass. The values of major and minor principal stresses at a point
in the rock mass along with mechanical and rock mass properties
determine whether yielding or failure may occur at that point.
Fig. 4ad shows the distribution of major principal stresses
obtained from each finite element model at depth D = 685 m for
various thickness of crown pillars. Similarly, data are obtained
for all other depths, thickness of pillars and other parameters.
Results of this study suggest that yielding or failure occurs at the
boundary of excavations especially for low values of GSI and UCS
Fig. 4. Major principal stress distribution around excavations and pillars for different pillar thickness T (Geo-mining conditions: GSI = 50, UCS = 75 MPa, D = 685 mL and
E = 20 GPa).
Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
of ore body. This is expected for any excavation made in rock mass
in a deep underground mine. However, if the safety factor at the
middle of the crown or sill pillar is found to be less than unity, it
can be assumed that the complete pillar may fail under the prevail-
ing stress conditions. Due to this reason, principal stress distribu-
tions have been estimated along the paths A-A0 and B-B0 (Fig. 4a)
for determination of safety factor at the middle of the pillars.
Major Principal stress profiles along the paths A-A0 and B-B0
are obtained for the above geo-mining conditions and tabulated
in Table 1. Table 1 summarizes the average principal stresses in
crown and sill pillars at D = 685 m and compares changes
with respect to thickness of crown pillar T = 5 m. The parameter
ji in the table denotes stress concentration factor measured as
ji = ri-induced/ri-in-situ. It is worth noting that similar data are tabu-
lated for other depths of workings and are not mentioned in this
paper.
5.2. Distribution of plastic strain intensity factor in pillars and around
the excavated areas
Yield or failure zones are defined based on the plastic strain
intensity factor around an excavation or in the pillars. In general,
higher plastic strain intensity value represents more severe yield-
ing or failure condition. In this study 4 m pillar and working depth
of 750 mL are considered for comparative study of yielding or fail-
ure of rock mass around the openings or level drives. The failure or
yield zone of rock mass has been analyzed based on the results
obtained from different numerical models.
 H. Kumar et al. / International Journal of Mining Science and Technology xxx (2017) xxxxxx 5

Table 1
Variation of principal stresses (ri, i = 1 for major principal stresses and i = 3 for minor principal stresses) and average stress concentration factors with pillar thickness T (Geo-
mining condition: GSI = 50, D = 685mL, rci=75 MPa and E = 20 GPa).

Pillar thickness T (m) Major principal stress r1, MPa

Along the path A-A0 (crown pillar) Along the path B-B0 (sill pillar)
Max. Min. Avg. % Change of Avg. Max. Min. Avg. % Change of Avg.
(a) Major principal stress r1 (MPa)
4 50.0 28.8 37.8 (1.70) 2.4 58.3 29.0 38.4 (1.76) 10.6
5 49.4 29.2 36.9 (1.66) 0.0 43.6 28.2 34.7 (1.59) 0.0
6 43.7 28.1 35.5 (1.60) 3.7 36.9 28.7 31.8 (1.46) 8.3
7 40.0 26.2 35.4 (1.59) 4.0 35.4 26.3 30.5 (1.39) 12.1
Pillar thickness T (m) Minor principal stress r3, MPa
Along the path A-A0 (crown pillar) Along the path B-B0 (sill pillar)
Max. Min. Avg. % Change of Avg. Max. Min. Avg. % Change of Avg.
(b) Minor principal stress r3 (MPa)
4 17.9 2.8 9.5 (1.32) 1.0 19.0 9.5 12.3 (1.74) 31.1
5 11.6 4.8 9.5 (1.31) 0.0 12.7 7.7 9.3 (1.32) 0.0
6 12.1 7.5 10.1 (1.40) 6.3 10.9 7.7 9.0 (1.27) 3.7
7 10.7 9.0 9.8 (1.37) 4.0 9.9 8.0 8.8 (1.24) 6.0

5.3. Effect of pillar thickness on plastic strain intensity and height of
failure zone
Fig. 5ad are plotted by varying T as 4 m, 5 m, 6 m and 7 m
respectively. It can be visualized that intensity and extent of plastic
strain has reduced as the T increases from 4 m to 7 m. These figures
indicate that thickness of pillar definitely influences the extent of
yield zone. To investigate this further, the maximum extent of yield
zone is calculated from these figures and tabulated in Table 2.
Similarly effect of other parameters like effect of depth of working
(D), geological strength index (GSI), uniaxial compressive strength
(UCS) and modulus of rigidity (E) have been carried out and signif-
icant effect on plastic strain intensity and height of yield zone
around the excavation was observed.
5.4. Development of multivariate regression model of pillar safety
factor
The novelty of the study lies in developing relationship between
input parameters such as UCS, T, GSI, D, E and safety factor (SF) of
crown pillars. A step by step procedure is outlined to describe this
relationship and then multiple regression analysis technique is
applied to estimate the co-efficient of the variables. After this rela-
tionship is built, design charts are prepared to demonstrate the
efficacy of the proposed method. Safety factors along the middle
of crown pillar are determined from 108 finite element models
for four depths of workings. A total of 432 sets of safety factor data
are analyzed based on the corresponding input parameters to
obtain a multivariate regression model that fits the data
coherently.
5.4.1. Estimation of safety factor of crown and sill pillars
Fig. 5ad shows that pillars does not show any failure or
yielding along the middle of crown pillar (Path A-A0 ) and middle
of sill pillar (Path B-B0 ) due to Drucker-Prager failure criteria
assumed in numerical modeling. Most of the safety factors along
the path A-A0 and B-B0 are found to be more than unity. A Hoek
and Brown failure criterion is assumed for the calculation of safety
factor along the middle of the pillars.

Fig. 5. Yielded zones around level drive and excavated areas for four values of sill and crown pillar thickness.

Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
6 H. Kumar et al. / International Journal of Mining Science and Technology xxx (2017) xxxxxx

Table 2
Extent of yield/failure zone on sill pillar and crown pillar for different pillar thickness T (Geo-mining conditions: GSI = 50, UCS = 65 MPa, E = 15 GPa and D = 750 mL).

Pillar thickness T (m) Approximate height of yield/failure zone on rock mass

Extent of yield zone in crown pillar hc (m) Extent of yield zone in sill pillar hs (m) hc/[hc (T = 5)] hs/[hs (T = 5)]
4 2.53 1.56 1.33 3.90
5 1.90 0.40 1.00 1.00
6 1.95 0.40 1.02 1.00
7 1.70 0.50 0.89 1.25

In general, the safety factor (SF) is defined as the ratio of the
strength of pillar Sp estimated based on Hoek and Brown yield cri-
terion, and the stress induced on pillar rp, as given below [23]:
Sp r3 rci mb rrci3 sa
SF
rp r1
where rci is compressive strength of intact rock; mb, s and a are
Hoek and Brown rock mass parameters. A pillar is considered to
be stable if SF > 1.0.
5.4.2. Development of a multivariate regression model of safety factor
Fig. 6 plots depth versus safety factors for different GSI and UCS
of ore body for pillar thickness of 6 m. A closer look of the safety
factor data reveals that SF values are significantly influenced by
four parameters: Depth Index (DI = D/W, where D is the working
depth, and W is the width of excavation), GSI, rci and T. It is found
that elastic modulus (E) of orebody has negligible effect about
2.59% change in SF if E of orebody changes from 10 GPa to
20 GPa. As a result, multivariate regression model is developed
based on 144 data sets considering E = 10 GPa only. In the follow-
ing, a multivariate regression model of SF is developed based on
the above four parameters considering three options:
5.4.2.1. Effect of depth of working and width of excavation. Fig. 7
shows a relationship of SF as function of DI for different thickness
of pillar T for GSI = 50, rci=50 MPa and E = 10 GPa. From Fig. 7, it is
found that a power law relationship provides a good estimate of SF
for each T and can be represented in general term as:
1
SF aDIc 2
where a and c are constants that vary linearly as a function of thick-
ness T as shown in Fig. 8, and can be expressed as:


a a1 b1 T
3
c a2 b2 T
where a1, b1 and a2, b2 are constants. Combining Eqs. (2) and (3), it
can be written as:
SF a1 b1 TDIa2 b2 T 4
Table 3 extracts 16 data set related to UCS = 50 MPa, GSI = 50
and E = 10 GPa from 144 data set. The concept and theory of
non-linear least square method is followed for determination of
the constants a1, b1, a2 and b2. From the regression analysis, values
of a1, b1, a2 and b2 are obtained as given in Eq. (5). The R2 value of
this relationship is found to be 0.881.

(1) Combined effect of depth and width of excavation for differ-
ent thicknesses having geo-mining conditions as GSI = 50
and UCS = 50 MPa is assumed to be the reference. These
cases provide the lower safety factors in the pillars. All
together there are 16 such cases out of 144 data sets.
(2) Incremental effect in SF due to change in GSI from 50 keep-
ing UCS = 50 MPa. There are 32 different cases are available
in the data set which meets this condition out of 144 geo-
mining conditions.
(3) Incremental effect in SF due to change in UCS from 50 MPa
keeping GSI = 50. In this case also, 32 different set of geo-
mining conditions are available.
The rest 64 cases of geo-mining conditions are the variations of
both GSI and UCS from 50 MPa.
SF 5:417  0:379TDI0:4300:035T 5
5.4.2.2. Variation of SF with change in GSI. In the previous section, it
is mentioned that GSI of rock mass directly influences the extent of
yield zone around an excavation. Similar influence of GSI on SF
estimated in the middle of the crown and sill pillars has been
observed.
To investigate this further, change of safety factor i.e., DSF is cal-
culated for each geo-mining condition only by varying GSI from
GSI = 50, i.e., DGSI = GSI-50. Fig. 9 plots DSF versus DGSI for differ-
ent T and DI. Figures show that rate of change of SF with respect to
change in GSI (i.e., DSF/DGSI) is different for different values of DI
for each T. It is clear that for all values of T and DI, DSF varies lin-
early with DGSI without any interception. This can be expressed
mathematically as:
DSF mDGSI mGSI  50 6

Fig. 6. SF along the middle of crown pillar. Fig. 7. Variation of safety factor with the variation of depth index (DI).

Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
 H. Kumar et al. / International Journal of Mining Science and Technology xxx (2017) xxxxxx 7

Fig. 8. Linear variation of a and c with pillar thickness T.

Fig. 10. Variation of slope m (DSF/DGSI) with the variation of DI.

Table 3
Safety factor and input data for UCS = 50 MPa, GSI = 50 and E = 10 GPa.
DSF GSI a3 b3 T c1 T 2 a4 b4 T c2 T 2 DI a5
DI T SF
55.65 4 1.2809 b5 T c3 T 2 DI2 GSI  50 9
55.65 5 1.3152
55.65 6 1.3476 where DSFGSI is the change in safety factor due to change in GSI.
55.65 7 1.3575
93.75 4 0.9368 5.4.2.3. Variation of SF with change in UCS. Variability of SF (DSFUCS)
93.75 5 1.0290
respect to change in UCS of ore body from UCS = 50 MPa is found to
93.75 6 1.0470
93.75 7 1.1328 be similar in nature as mentioned for GSI. The details of the process
181.11 4 0.8401 are omitted from this paper. Superposing both the effects simulta-
181.11 5 0.9452 neously in Eq. (5), regression model of SF for varying effect of DI, T,
181.11 6 1.0006 GSI and UCS can be given as:
181.11 7 1.0339
257.31 4 0.8354 SF a1 b1 TDIa2 b2 T a3 b3 T c1 T 2 a4 b4 T
257.31 5 0.9083
257.31 6 0.9547 c2 T 2 DI a5 b5 T c3 T 2 DI2 GSI  50 a6
257.31 7 1.0340
b6 T c4 T 2 a7 b7 T c5 T 2 DI a8 b8 T
c6 T 2 DI2 UCS  50 10

where slope m (m = DSF/DGSI) depends on DI and T. In order to
determine the variability of m with DI, slope versus DI is plotted
for different values of T as shown in Fig. 10. This figure shows that
slope m varies quadratically with DI and can be expressed using
mathematically as:
 
DSF
m P Q DI RDI2
DGSI
where P, Q, and R are constants to be determined based on the pillar
thickness T. From Fig. 7, values of the above constants are plotted
with T as shown in Fig. 11. It is obvious that each constant varies
quadratically with T with a general relationship as:
8 SF REG SF FEM e
>
< P a3 b3 T c1 T
2
where ai, bi, (i = 6, 7, 8) and cj (j = 4, 5, 6) are constants to be deter-
mined by regression analysis using least square method.
5.4.2.4. Regression model of SF. As mentioned before, from the 108
finite element models, 432 SF data are extracted with respect to
three variations of GSI, three variations of UCS, four variations of
pillar thickness T, four variations of depth index (DI) and three vari-
7
ations of modulus of elasticity (E). Since the effect of E has been
omitted in this analysis 144 data sets have been used for determin-
ing the coefficients of Eq. (10) with the help of least square method.
From the above discussion, SF can be estimated with a random
error e as:
11

Q a4 b4 T c2 T 2 8 where SFFEM is the observed (FEM model) data and SFREG is esti-
>
: mated from the regression model given in Eq. (10). The constants
R a5 b5 T c3 T 2
of the model need to be determined based on the data obtained
where ai, bi (i = 3, 4, 5) and cj (j = 1, 2, 3) are constants to be deter- from numerical models. It may be noted that constants a1, b1, a2
mined by regression analysis using least square method. Now com- and b2 have already been determined as given in Eq. (5) and hence,
bining Eqs. (7) and (8) into Eq. (6), we get: the first term of Eq. (10) is constant quantity only depending on DI

Fig. 9. Change in SF of T = 4 m, 5 m, 6 m, 7 m crown pillars with the change in GSI.

Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
8 H. Kumar et al. / International Journal of Mining Science and Technology xxx (2017) xxxxxx

Fig. 11. Variation of P, Q and R with the variation of pillar thickness T.

and T. The rest of the constants a3, b3, c1 a4, b4, c2, a5, b5, c3, a6, b6, c4, 5.5. Validation of results obtained from regression model
a7, b7, c5, a8, b8, and c6 are determined using the least square method
and reported in Eq. (12). In the following, several examples are presented to validate the
results obtained from the regression model. Results of regression
SF REG 5:417  0:379TDI0:4300:035T model have also been compared with those obtained from FEM
2 3
25:627
103 57:051
105 T 44:12
106 T 2 models in terms of effect of DI, GSI and UCS and T.
6 7
6 5 6 8 2 7
4 24:196
10 15:58
10 T  81
10 T DI 5GSI  50
5.5.1. Comparison of model output with finite element results data
10:5
107  15
108 T 108 T 2 DI2
2 3 4 4 2
3 5.5.1.1. Effect of DI. Based on the regression model, safety factor in
62:371
10  86:099
10 T 17:1066
10 T the middle of the crown and sill pillar can be forecasted for various
6 7
6 5 5 6 2 7
4 91:872
10 30:510
10 T  27:38
10 T DI 5UCS  50 combinations of, DI, GSI, rci, T and E. It may be noted that these
29:4
107  97
108 T 86:7
109 T 2 DI
2 models will provide comparable results if DI increases with depth.
In order to compare the results of the developed mathematical
12
model with that of FEM generated data, Fig. 12 plot SF verses DI
2
The R of this regression analysis is found to be 0.94 with for two different geo-mining conditions. It is seen that models pre-
F-statistic value of 127.36. The model is found to be significant dict reasonably well for higher values of T. The model could follow
with 95% confidence interval. the pattern of the data, i.e., SF decreases with the increase in depth.

Fig. 12. Comparison of the results of regression model with FEM generated data.

Fig. 13. Comparison of the results of regression models with actual data.

Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
 H. Kumar et al. / International Journal of Mining Science and Technology xxx (2017) xxxxxx 9

Table 4 three zones viz, ZONE-1, ZONE-2 and ZONE-3 are defined, describ-
Approximate excavation width of mine with depth of working. ing the state of stability conditions of crown pillars. The descrip-
Depth of working (m) Width of excavation (m) DI tions of each of these zones are as follows.
500 20 25.0 ZONE-1 represents unstable conditions (SF < 1.0) including (1)
530 20 26.5 Pillar is likely to experience a stress induced sever yielding condi-
560 17 32.9 tion or fall of roof in excavated zone; (2) Limited local support or
590 17 34.7 roof bolting may not be effective in maintaining stability; (3) The
620 14 44.3
650 14 46.4
magnitude and direction of the induced stress is likely to alter with
680 12 56.7 the increase in excavated area.
710 10 71.0 ZONE-2 represents potentially unstable conditions (SF between
740 10 74.0 1.0 and 1.5) including (1) Extra ground support may be required to
770 8 96.3
prevent possible fall of rock; (2) Wall rock movement is likely; (3)
800 8 100.0
830 6 138.3 Continuous monitoring of ground movement is required.
860 6 143.3 ZONE-3 represents stable conditions (SF > 1.5) including (1) No
890 5 178.0 uncontrolled fall of ground is likely to take place; (2) No significant
920 4.5 204.4 rock movement of back is likely to take place; (3) No extraordinary
950 4.5 211.1
support measures are required.
Development of stability/design chart (Example 1. An ore body
having decreasing width with depth of working similar to the case
However, for DI more than 181.11 (working depths 815 meters study mine).
level and below) SF remain constant or rather increases slightly. Lets consider two different geo-mining conditions, i.e., (1)
This has happened due to the decrease in width of excavation at GSI = 52, UCS = 48 MPa, and (2) GSI = 72, UCS = 55 MPa matching
deeper levels. with the geo-mining conditions of the case study mine. In both
the cases, E value of orebody ranges between 10 GPa and 20 GPa.
5.5.1.2. Effect of GSI and UCS. SF estimated from the mathematical The thickness and depth of the orebody is varying approximately
model and those obtained from FEM are plotted in Fig. 13ab for as per the data provided in Table 4.
two geo-mining conditions to show the effect of GSI and UCS. It From the regression models Eq. (12) the SF for each DI and T
can be seen that model provides reasonably accurate estimate of (4 m, 5 m, 6 m and 7 m) are estimated for crown pillars as shown
data with variation of GSI and UCS. However, for T = 4 m, predic- in Fig. 14. Figures show that SF values first decrease with increasing
tion of model is around 1.72% to 5.05% lower as compared to the DI for each T and then remain almost constant. Figure shows that
FEM generated data. For T = 7 m, the model predicts with an accu- the crown pillars of thickness form 4 m to 7 m are stable for Depth
racy of 99.47%. index (DI) approximately less than 30 and it becomes unstable at
DI greater than 30 for the first geo-mining condition. At 810 m
5.6. Stability chart for design of crown pillars based on regression depth (DI = 115) or below SF of 4 m crown pillar lies in ZONE-1
analysis results causing complete collapse, whereas 6 m or 7 m pillar remain in
ZONE-2 up to a depth of 890 m. For the second geo-mining condi-
A stability chart signifies the relationship between SF and the tion SF of 4 m, 5 m, 6 m and 7 m crown pillars enter into ZONE-2 at
corresponding input parameters. It shows regions of instability, if DI = 80 (D = 720 m), DI = 100 (D = 800 m), DI = 125 (D = 875 m), and
any, for given values of parameters. Based on this chart, design DI = 175 (D = 880 m), respectively. In this case, SF of crown pillar
input parameters of a structure are determined for a pre-defined thickness of 4 m to 7 m never lies in ZONE-1. This study finds that
SF. In the previous sections, multivariate regression models of SF considering SF values between 1.0 and 1.5 as potentially unstable
of crown pillars is established and their efficacy in forecasting zone, 4 m and 5 m pillars are potentially unstable for deeper work-
the same has been shown for various combinations (examples) of ings; whereas 6 m and 7 m pillars may stand up to depths of 875 m
input parameters and ore body width in the following sections. to 900 m which can be verified with the field data. Field data
obtained from the case study mine reviles that the thickness of
5.6.1. Interpretation of stability charts in terms of safe zones crown pillar at a depth 750 mL and below was 69 m (i.e. thickness
In the following, stability charts are prepared as SF versus DI for of horizontal pillar was 1216 m) without any disturbances in the
each T and for various combination of GSI and UCS. In every chart, competent rock conditions.

Fig. 14. Stability charts of crown pillars for an orebody with decreasing width of excavation in deeper levels.

Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
10 H. Kumar et al. / International Journal of Mining Science and Technology xxx (2017) xxxxxx
6. Conclusions
The motivation of this study has been obtained from the
demand for development of a suitable design method of crown pil-
lars especially for deep underground hard rock mining in varying
geo-mining conditions. This study has comprehensively analyzed
stresses, displacements, yield zones for a variety of geo-mining
conditions pertaining to deep hard rock mining in India and devel-
ops multivariate regression models for generating design charts of
crown pillars. It has been observed that a 5 m pillar may be safe
with a rock bolt support of 3 m length in crown pillar and 2 m
length in sill pillar, under the condition of UCS > 65 MPa, and
GSI > 70. The pillar thickness of 4 m is not suggested for any of
the geo-mining conditions considered in this study, since severe
yielding has been observed especially in the lower levels. In gen-
eral, 6 m crown pillar and sill pillars are suitable up to a depth of
1000 m for GSI of orebody over 60. Rock bolts of at least 3 m length
will be required to stabilize the boundary of the stope wall and
roof. Apart from the rock bolt support, additional supports like
chock mate, cement concrete, cross beams etc. may be required
in deeper working levels especially in weak and jointed rock
conditions.
This study develops multivariate regression models based on
results obtained from series of finite element models. This develop-
ment is unique in the sense that the both finite element and
regression models incorporate mining and geological parameters
such as depth of the workings (D), width of excavation (W), GSI,
UCS, and E of ore body rock and thickness of the pillar. Using
regression model, design charts of crown pillars are developed.
The design charts obtained from this study helps the mine opera-
tors as a quick reference for selecting the appropriate mining
parameters in order to optimize the pillar design and recovery of
ore from underground.
References
[1] Cao RH, Cao P, Lin H. Support technology of deep roadway under high stress
and its application. Int J Min Sci Technol 2016;26(5):78793.
[2] Koldas KS. Rock-related accidents, investigations and inquiries in South African
mines. In: Proceedings of 17th international mining congress and exhibition of
Turkey. Morgantown: West Virginia University; 2001. p. 10515.
Please cite this article in press as: Kumar H et al. Design of crown pillar thickness using finite element method and multivariate regression analysis. Int J
Min Sci Technol (2017), http://dx.doi.org/10.1016/j.ijmst.2017.06.017
[3] He Y, Shao P, Yue F. Deformations failures of deep high stress rock masses and
its stability. London: Taylor & Francis Group; 2004.
[4] Das KC. Enriched finite element method and applications in reinforced jointed
rock mass. Kharagpur: Indian Institute of Technology; 2013.
[5] Deb D. Finite element method: concepts and applications in
geomechanics. New Delhi: Prentice Hall of India Pvt. Ltd.; 2006.
[6] Tavakoli M. Underground metal mine crown pillar stability analysis. New
South Wales: University of Wollongong; 1994.
[7] Lu TK, Guo BH, Cheng LC. Geotechnical aspects of underground construction in
soft ground. London: Taylor & Francis Group; 2009.
[8] Gabriel SE, Dennis RD, John LE, Leonard JP. Pillar and roof span design
guidelines for underground stone mines. Pittsburgh: Department of Health
and Human Services (NIOSH); 2011.
[9] Edelbro C. Rock Mass strength A review. Technical Report Lulea University of
Technology: Department of civil engineering division of rock mechanics; 2003.
[10] Kvapil R, Blake W. Geometry and stability determination of large dimension
cut and fill rooms at Kamoto. In: Proceedings of the jubilee symposium on
mine filling. Kamoto: Australasian Institute of Mining and Metallurgy; 1973. p.
14754.
[11] MOSHAB. Geotechnical considerations in underground mines (Guidelines).
Western Australia: Department of industry and Resource; 1997.
[12] Deb D, Das KC. Enriched finite element procedures for analyzing decoupled
bolts installed in rock mass. Int J Numer Anal Meth Geomech 2011;35
(15):163655.
[13] Esterhuizen GS, Iannacchione AT. Effect of the dip and excavation orientation
on roof stability in moderately dipping stone mine
workings. Pittsburgh: Department of Health and Human Services (NIOSH);
2011.
[14] Bakhtavar E, Oraee K, Shahriar K. Determination of the Optimum Crown pillar
thickness between Open - pit and Block Caving. In: Proceedings of 29th
international conference on ground control in mining. Morgantown: West
Virginia University; 2010.
[15] Mark Christopher. Science of empirical design in mining ground control. Int J
Min Sci Technol 2016;26(3):46170.
[16] Herget G. Stresses in rock. Rotterdam: Balkema; 1988.
[17] Martin CD, Kaiser PK, Christiansson R. Stress, instability and design of
underground excavations. Int J Rock Mech Min Sci 2003;40(78):102740.
[18] Leucci G, De Giorgi L. 2D AND 3D seismic measurements to evaluate the
collapse risk of cave in soft carbonate rock. Central Eur J Geosci 2015;7:8494.
[19] Deshmukh DJ. Elements of mining technology. Nagpur: Central Techno
Publications; 1998.
[20] Hartman HL. SME mining engineering handbook. Colorado: Society for Mining,
Metallurgy, and Exploration; 1992.
[21] Gogoi MP. Estimation of rocks strength parameters for middle Siwalik,
Arunachal Pradesh. J Front Res Arts Sci 2011;1:949.
[22] Hoek E, Diederichs MS. Empirical estimation of rock mass modulus. Int J Rock
Mech Min Sci 2006;46(2):20315.
[23] Hoek E, Carranza-Torres C, Corkum B. Hoek-Brown criterion-2002 edition. Proc
NARMS-TAC Conf 2002;1(20):26773.