1. Introduction
Pillars are the primary support systems in underground mines and are typically left between the openings to maintain their stability. The design and stability of the pillars are the two most complicated challenges in ground control studies. Unfeasible and incompetent direct loading tests on the pillars in underground mines lead to adopt empirical-based designs and back analysis. The empirical design theories are predominantly based on stable, unstable, and failed pillars and do not consider the different failure mechanisms of the pillars. The factors that influence the different failure mechanisms in pillars are:
This paper mainly focuses on the inclined pillars in orebodies with different orientations.
Following the catastrophic pillar collapse at the Coal Brook colliery on January 21, 1960, pillar stability and pillar design optimization has been more thoroughly investigated to obtain more reliable design approaches. One of the first empirical approaches was developed by Salaman and Munro [
1]. This empirical approach was later modified and applied for the Canadian Uranium mines by Hedley and Grant [
2], similarly based on stable, unstable, and failed pillars, to develop a relationship between the pillar strength and the geometric parameters of the pillar:
where
is the strength of the pillar (MPa),
k is the unit strength of the rock sample (MPa),
W is the width of the pillar, and
H is the height of the pillar, with constants
a and
b given as 0.5 and 0.75, respectively.
One of the commonly used empirical approaches is the confinement formula to determine the strength of hard-rock pillars, developed by Lunder and Pakalnis [
3] and based on 178 pillars, which is:
where
is the ultimate strength of the pillar (MPa),
K is the pillar size factor,
UCS is the uniaxial compressive strength of the intact rock (MPa),
C1 and
C2 are the empirical rock mass constants, and
is the friction term, which is calculated as:
where
is the average pillar confinement, and
Coeff is the coefficient of pillar confinement.
Few other researchers developed similar empirical relationships between pillar strength and pillar geometry [
4,
5,
6,
7].
Numerical tools were used by researchers to understand the failure mechanisms and to evaluate the strength of hard-rock pillars. The elastic-brittle plastic constitutive model in finite elements and boundary element modelling packages were developed to evaluate the strength of pillars [
8]. With the help of numerical modelling, the slender pillars in a limestone mine were classified as the pillars with highly variable strength, that depend on the structures which have little impact on higher width-to-height (W/H)-ratio pillars [
9,
10]. The failure mechanism in the slender pillars was described as brittle failure, where the failure plane passes through the centre of the pillar, while, in larger pillars, the failure mechanism was described as spalling followed by shear failure. Studies were conducted on joint spacing, joint length, and joint orientation to develop an understanding of the relationship between the area or volumetric fracture intensity and the strength of the pillars [
11,
12]. These studies were all based on a normal loading of the pillars, causing compression loads on them (
Figure 1a).
Inclined pillars undergo oblique loading, which is a combination of compressive and shear stresses, as shown in the
Figure 1b. Stress analysis studies were conducted on two pillars, one inclined along a 45° dip angle, and the other with normal loading, with the same extraction ratio. It was concluded that the failure in the dip pillar extended from one corner to the other, while the normal pillar was stable, which resulted in domino pillar bursting in Quirke mine [
13]. The progressive failure of 20°-inclined pillars was described by Pritchard and Hedley at Denison mine [
14]. It was described as spalling at the two sides of the pillars, followed by hour-glass fracture formation in an inclined fashion, which ultimately led to a complete failure in pillars with larger width-to-height ratios. Case studies conducted on inclined pillars and associated excavations which undergo oblique loading were described to be at higher risk of failure when compared to pillars with normal loading [
15,
16]. The case studies were only based on one pillar and its failure mechanism, while the strength of the pillar was not evaluated to develop pillars in an inclined fashion.
Two-dimensional finite-element numerical studies were conducted to evaluate the strength of hard-rock pillars under far-field principal stresses at different orientations, and it was concluded that the strength of the pillars with higher W/H ratios was highly affected by the orientation of the pillars [
17,
18]. Jessu and Spearing [
19] conducted similar numerical studies to that of Suorineni [
18] but with a finite difference code in a three-dimensional model to evaluate the strength and the failure mechanisms of pillars at different inclinations. It was concluded that, as the inclination increases, the pillars with higher W/H ratios have a significantly lower strength than the horizontal pillars. The study also described that the brittle failure mechanism is the dominant failure mechanism in inclined pillars, where failure starts from one corner of the pillar and proceeds to other corner. All the studies conducted were numerical modelling studies, no laboratory testing was conducted to show the effectiveness of the numerical modelling results.
This paper presents the laboratory testing of specimens under inclined loading and evaluates their reduction in strength in comparison to specimens tested under normal loading. The failure mechanisms were evaluated at the laboratory scale and were related to the cause of strength reduction. The results of the numerical modelling of pillars presented by Jessu and Spearing [
19] were further evaluated to describe the strength reduction factors.
2. Materials and Test Methods
Moulded gypsum and sandstone core specimens were tested under uniaxial and oblique loading conditions. These were selected by the authors, as moulded gypsum was extensively used by many researchers [
20,
21] as a representative of brittle rock, and both moulded gypsum and sandstone have lower strength, which is most suitable for the currently developed testing methodology.
Gypsum specimens were prepared by mixing the gypsum powder to water with a mass ratio of 100:35. PVC tubes of 50 mm inner diameter were used to cast the specimens. The PVC tubes were cut perpendicular to the tube length to cast the normal specimens, while for inclined specimens, the PVC tubes were cut at an angle on both ends. After pouring into the moulds, the mixture was stirred to remove bubbles and then placed in an oven at a temperature of 40 °C, until the specimens’ mass reached a constant value, which was attained in three days. All specimens were created in a single batch for consistency. The surface of the normal samples was made smooth and parallel according to the International Standards of Rock Mechanics (ISRM) standards with the help of a grinder, while, for the inclined specimens, the surfaces were polished with sandpaper, first with coarse grit #60 and then with fine grit #200. Gypsum specimens with three different inclinations and four different width-to-height ratios were prepared, as shown in
Figure 2a,b. Five specimens were tested in every test, therefore, a total of 60 gypsum specimens were tested.
A sandstone core of 42 mm diameter was used to conduct the test. The sandstone specimens were cut at different lengths to produce different width-to-height ratios (
Figure 2c). The sample ends were prepared to be parallel and straight, as specified in the ISRM standards, with the help of a grinder. The specimens were then loaded with inclined platens at different width-to-height ratios.
A uniaxial compression testing machine (
Figure 3a), which was controlled by a servo computer program GCTS CATS 1.8 software, was used to test the gypsum and sandstone specimens in normal and inclined fashions. Straight platens were used for the uniaxial compression testing of the normal specimens, while, for the inclined specimens, platens were manufactured at 10° and 20° angles. The inclined platens were fixed to the frame of the testing machine, and then the testing of the inclined sandstones specimens was done, as shown in
Figure 3b.
As the gypsum specimens were casted in the inclined fashion, they were placed directly at the centre of the straight platens, and a load of 0.2 kN was applied, such that the samples did not slip or tumble down the platens. For sandstone specimens, as they could be prepared (cut) with inclined ends, the straight specimens were placed on inclined platens in such a manner that the centre of the samples coincides with the centre point of the frame, as shown in
Figure 4b. After placing the specimens on the lower platens, a load of 0.2 kN was applied to hold the samples in position.
The machine recorded load and displacement data automatically at a rate of 600 samples/minute. A displacement-type loading rate was adopted, with a fixed loading rate of 0.12 mm/min, which was in accordance with the ISRM standards, as the gypsum and sandstone specimens with length-to-diameter (L/D) ratio of 2.5 reached their peak strength in 5 to 10 min. The testing materials and the loading conditions are specified in
Table 1.
Each test was conducted three or five times, and the results were averaged. Therefore, to verify if the average results were representative of the observed results, a statistical Pearson’s chi-square test was conducted for the strength values, elastic moduli, and equations proposed.
The Pearson’s chi-square test [
22] is a statistical test to determine if the observed values are significantly different from the expected value. Two hypotheses are made: the null hypothesis assumes there is no significant difference between the observed values and the expected values, while the alterative hypothesis suggests there is a significant difference. The test is conducted with the help of the formula:
where
is the chi-square goodness-of-fit test,
O is the observed value, and
E is the expected value. The degrees of freedom are determined with the help of the constraints and are one less than the number of the observed values. The chi-square and degrees of freedom are evaluated, and the probability is determined with the help of tables. If the probability is less than the significance level, which is generally taken as 0.05 or 0.1, then the null hypothesis is rejected. This shows that the expected value is significantly different when compared to the observed value.
Chi-square goodness of fit was used in this paper to determine if the average values were representative of the observed values. The chi-square values and the degrees of freedom are for all the tests conducted. The probability or the p-value is determined to show the non-significant difference between the observed values and the average values.
4. Numerical Modelling
FLAC
3D [
24], a geotechnical finite difference modelling package, was used to simulate the pillars, as shown in
Figure 10. A three-dimensional co-ordinate system was used, where the horizontal plane was represented in the x and y directions, and the vertical plane was represented in z-direction. For the inclined pillars, inclination was measured from the x direction, as shown in
Figure 10b. The model consists of the main floor, the main roof, and the pillar with a height of 4 m, considering the Lunder and Pakalnis database [
3]. The width of the pillars was varied to achieve the width-to-height ratio of the pillars. The extraction ratio was kept constant for all the vertical pillars at 75% and for the inclined pillars, the boundaries were established far enough to avoid boundary effects on the performance of the inclined pillars and their failure behaviour. The thickness of the roof and the floor was maintained at three times the pillar height in all models to avoid the boundary effects.
Boundary conditions such as fixed supports were placed at the bottom of the floor, which restricted the displacement and velocity in both parallel and normal directions. Roller supports were placed on the sides, thus restricting the velocity and displacement in the normal direction. Loading was applied in the form of uniform velocity on the top of the roof until the pillar entirely failed and the residual strength reached 50% of the peak strength, as recommended by Lorig and Cabrera [
25].
The bilinear strain hardening–softening ubiquitous joint constitutive model was applied to simulate the pillars. This model is based on the Mohr Coulomb strength criterion and the strain softening or hardening as a function of the deviatoric plastic strain [
24]. The elastic criterion was applied to simulate the roof and floor to ensure that the roof and floor were stronger than the pillar and failure was only induced in the pillar. The model properties are summarized in
Table 3 and
Table 4 [
26].
The critical parameter in defining the strain-softening properties is the model element size. The element size used was 0.5 m × 0.5 m × 0.5 m throughout the model, and cohesion softening was carried out to calibrate the horizontal the pillar results to those of Lunder and Paklnis [
3]. The horizontal pillars were simulated at four different width-to-height ratios (0.5, 1.0, 1.5, and 2.0). The numerical model results were found to be in the range of 2% of the theoretical results, as shown in
Figure 11a.
Jessu and Spearing [
19] simulated pillars at five different inclinations of 0°, 10°, 20°, 30°, and 40° with four different width-to-height ratios. It was concluded that the pillar strength increased with the increase of the width-to-height ratio at all the inclinations of the pillars, as shown in
Figure 11. The results presented in Jessu and Spearing [
19] for the numerical modelling of horizontal and inclined pillars were analysed further to determine the strength reduction factors.