## Abstract

This paper presents the design optimization for the Xianglushan tungsten mine using the room and pillar mining technique. The influence of pillar height, pillar length, pillar width, room length and room width on the stability of room and pillars was examined through analyses using the orthogonal experimental design method. The pillar and room sizes were identified as the most critical factors. The types of rock masses in the Xianglushan tungsten mine were identified in terms of rock mass classification and the mechanical parameters of those rock masses were assessed. A numerical model was developed in the continuum-based finite-element program 3D-*σ* to simulate the excavation process. Given fixed mining and pillar heights, the maximum room width and minimum pillar width were calculated. A separate calculation was also conducted to assess the stable excavation span using the stability graph method. The results obtained from numerical parametric study and empirical-based analysis are consistent. A comprehensive field monitoring programme is introduced, including the data from acoustic emissions, stresses and displacements. The measured responses of pillars and roof rock demonstrate that the proposed design optimization is effective.

The room and pillar mining technique has been widely used in the extraction of underground minerals for the past few decades. In this technique, the mined material is extracted across a horizontal plane while horizontal arrays of rooms and pillars are created as the extraction proceeds. Pillars are untouched mined material that are intended to support the roof overburden, whereas rooms are open areas that are to be extracted underground. However, the success of mineral extraction using the room and pillar mining technique depends largely on the degree of understanding of the site geological conditions and rock mass properties, as well as the experience of designers. This means that a comprehensive site investigation and stability analysis of rock roof and pillars must be conducted to develop a rational design of mineral extraction with which the mined material can be safely extracted while the economic benefit is maximized. Furthermore, the rock roof and pillars should be closely monitored so that their stability conditions can be evaluated and updated on a timely basis. Failure to do so could result in consequences that are catastrophic; for example, the collapse of the CoalBrook coal mine at Clydesdale Colliery, South Africa in 1960, in which 438 miners were trapped and died (van der Merwe 2006).

Pillar size and excavation span (room size) are two important factors that must be determined with caution when using the room and pillar mining technique for mineral extraction. A general design philosophy is that excavated stopes should be dimensioned to have similar sizes among rooms and among pillars, as shown in Figure 1. This is because if one of the pillars fails owing to inadequate size, the loads it carries will transfer to adjacent pillars, which increases the risk of initiating progressive collapse failures. On the other hand, the mineral extraction will be less productive if a large amount of a mine is used as pillars and left untouched. Extensive studies have been conducted to address this issue.

Based on numerical analyses on the stability of unsupported rock pillars, Lee *et al.* (2013) concluded that a larger ratio of pillar width to pillar height is beneficial to increase the stability of pillar and room excavations. Ghasemi *et al.* (2014) developed an empirical method to predict the stability of pillars using artificial intelligence techniques in terms of fuzzy logic based on a database containing 399 room and pillar coal mines in the USA. They also identified the critical parameter that controls the initiation of rock falls during retreat mining using a probabilistic semi-quantitative analysis method (Ghasemi *et al.* 2012). Zipf & Mark (1997) compared three design methods to control ‘cascading pillar failure’ in room and pillar mines based on the local mine stiffness stability criterion. Bogert *et al.* (1997) evaluated the effects of pillar geometry on the design of room and pillar stopes in highly fractured areas using laboratory test results and modelled the behaviour of highly folded and fractured rock masses using numerical simulations. They proposed a normalized approach for determining pillar and room sizes taking into account pillar configurations and mining sequences. Nazarov *et al.* (2006) utilized convergence data for roof and floor rocks to estimate the rheological parameters of rocks, which were then included to estimate the service life of pillars.

Extraction of minerals beneath the earth surface will inevitably cause ground subsidence and disturbance to adjacent utilities, infrastructure and the environment; for example, house subsidence in the vicinity of mining zones (Marino & Gamble 1986). Because of complex geological and geotechnical settings for a mining site, predicting the surface subsidence to a level of precision that is satisfactory is very difficult before the mine activity is actually initiated and field monitoring data become available. Sheorey *et al.* (2000) compared the subsidence guidelines in the UK and India, and found that the excavation procedures for un-caved rocks and pre-existing caved goafs are different. Usually, a higher disturbance to ground surface should be expected for a rock mass with intense fractures. The current solution for prediction of ground subsidence or ground disturbance is to use field observations to back-calibrate a prediction model; for example, the influence function method to calculate the surface subsidence for shallow to moderately deep coal mines in India (Sheorey *et al.* 2000), methods combining fieldwork and numerical simulations to evaluate damage to infrastructure caused by mining cave-in in Lo Tacón industrial area, Murcia, Spain (Álvarez-Vigil *et al.* 2010; López Gayarre *et al.* 2010; Álvarez-Fernández *et al.* 2011), and the approach to investigate the bedding separation failure in roof strata for the Jinggonger coal mine in China (Yan *et al.* 2016).

Field measurements are often needed as geological conditions and rock mass properties are site-dependent and the rock mass properties and behaviour in the field cannot be fully represented by the responses of intact rock samples measured in the laboratory (Ni *et al.* 2017). The distribution of fractures within the rock mass could change the behaviour significantly. Field measurements can include the use of biaxial extensometers and inclinometers to detect cracks in adjacent infrastructure (Álvarez-Fernández *et al.* 2011). Tesarik *et al.* (2009) presented a comprehensive study on field tests at the Buick Mine near Boss, Missouri, USA for almost 16 years. They installed borehole extensometers and biaxial stressmeters in pillars and the surrounding rock, and used embedment strain gauges, extensometers and earth pressure cells in the cemented backfill. The details of calibration and measurements for all these sensors can benefit other similar field studies.

In this paper, the Xianglushan tungsten mine is used as an example to show the process of design optimization for room and pillar mines. The geological conditions of the site are reviewed and the stratigraphy of the area, the statistics of fracture apertures and the rock quality designation are summarized. These data are then used to evaluate the rock mass using empirical classification systems, such as the rock mass rate (RMR) system (Bieniawski 1989), the Q system (Barton *et al.* 1974) and the BQ system (GB/T 50218-2014, Ministry of Housing and Urban–Rural Development 2014). A series of laboratory tests, including uniaxial compression tests, Brazilian tests, direct shear tests and P-wave velocity tests, are performed to estimate the mechanical properties of intact rock samples obtained from field investigation. The measured mechanical parameters are scaled using the software Roclab (DeLeo 1993) based on rock mass classification for use in numerical analyses to account for the difference in behaviour caused by the presence of fractures (Ni *et al.* 2017). The influence of parameters in an equation for computation of the factor of safety on the stability of pillars is investigated using the orthogonal experimental design approach to assess the significance of each parameter. The factor of safety for all existing 450 pillars in the Xianglushan tungsten mine are calculated for comparison. A numerical model of room and pillar stopes is subsequently developed to analyse the stability of the system. The optimal room width and pillar width, and the stable excavation spans are evaluated. Finally, a field monitoring programme is introduced, including acoustic emissions, stress changes, and roof and floor displacements, to assess the efficacy of the proposed design optimization.

## Rock mass evaluation

### Xianglushan tungsten mine

The Xianglushan tungsten mine is a working underground mine located *c.* 35 km NW of Xiushui County, Jiangxi Province, China. The construction of the mine was started in 1993. The total mining area is about 5.3 km^{2}, and is primarily divided into five panels, as shown in Figure 2. The rock slope at the site, in general, ranges from 25° to 30°. However, the skarn scheelite deposit is a very thick flat-lying deposit; as a result, the room and pillar mining technique was chosen. Based on geological surveys and geotechnical investigations, four groups of rocks were identified (Table 1). On average, the joint density of the rock mass is between 7.9 and 8.3 fractures per m^{3}, and there are three dominant joint sets. The dip angle and the dip direction of the three joint sets are 79/252°, 71/130° and 16/332°.

### Rock mass classification

#### The rock mass rating (RMR) system

The rock mass rating (RMR) system was initially proposed at the South African Council of Scientific and Industrial Research (CSIR) based on data obtained from underground excavations in sedimentary rocks (Bieniawski 1976). This classification system was developed for geotechnical applications rather than mining activities. Bieniawski (1989) modified the system to make it less conservative for mining design. Six parameters are needed in the RMR system: the uniaxial compressive strength of intact rock, rock quality designation (RQD), joint spacings, joint orientations, joint conditions and groundwater conditions. Each parameter is quantified by a value, which is determined based on field surveys and laboratory element tests. The determined six values are then summed to give the RMR value, which represents the characteristics of the rock mass. The RMR value can also be used to approximate the geological strength index (GSI) (Marinos *et al.* 2007), useful in estimating the mechanical parameters of the rock mass (e.g. strength and deformation modulus). The calculated results using the RMR system for the four rocks are summarized in Table 2.

#### The Q system

The Q system was developed using a total of 212 tunnelling cases at the Norwegian Geotechnical Institute (NGI) to correlate the required support system and the rock mass quality (*Q*) (Barton *et al.* 1974; Barton 2002). This method considers the contributions from the RQD, joint set number (*J _{n}*), joint roughness number (

*J*), joint alteration number (

_{r}*J*), joint water reduction number (

_{a}*J*) and stress reduction factor (SRF) in the following form: (1)where the RQD/

_{w}*J*term represents the degree of jointing (or block size), the

_{n}*J*/

_{r}*J*term shows the effects of joint friction (inter-block shear strength) and the

_{a}*J*/SRF term corresponds to the active stress.

_{w}The maximum excavation span (*W*) without supports (Barton 2002) can be calculated empirically as follows:
(2)where ESR is used to define the excavation support ratio. For the permanent mine in the present study, the ERS value should be taken to be between 1.6 and 2.0. Table 3 presents the calculated rock quality and the associated excavation spans.

#### The BQ system

The GB/T 50218-2014 (Ministry of Housing and Urban–Rural Development 2014) design guideline provides a two-step rock mass classification method (i.e. the BQ system). A basic quality (BQ) can be calculated first as a function of the uniaxial compressive strength of intact rock (*R _{c}*) and the intactness index of rock mass (

*K*) as follows: (3)where the

_{v}*K*value is defined as the square of the ratio of P-wave velocity of rock mass to that of intact rock. When

_{v}*R*> 90

_{c}*K*+ 30, the parameters of

_{v}*R*

_{c}_{ }=

_{ }90

*K*+ 30 and

_{v}*K*should be used in the calculation. When

_{v}*K*> 0.04

_{v}*R*+ 0.4, the parameters of

_{c}*K*

_{v}_{ }=

_{ }0.04

*R*+ 0.4 and

_{c}*R*should be employed.

_{c}The rock quality should be determined based on a further modification to the BQ value by taking environmental factors such as the influence of groundwater conditions (*K*_{1}), joint orientations (*K*_{2}) and initial geo-stress field (*K*_{3}) into account:
(4)The evaluated results of rock mass quality using the BQ system are tabulated in Table 4.

#### Summary of classification

The rock mass quality at the site was evaluated using three rock mass classification systems and the analysis outcomes are consistent. Limestone and marble are identified as class II rocks, and the excavation span should be limited to 10–15 m conservatively. On the other hand, hornfels and granite are class I rocks, and an excavation span up to 20 m should be allowed.

### Estimating the mechanical rock mass parameters

#### Laboratory tests

Laboratory tests were conducted to measure the mechanical properties of the four types of rocks. Rock samples with dimensions of 200 mm (length) × 200 mm (width) × 150 mm (height) were taken from different boreholes in #4 panel of the room and pillar mine (Fig. 2), including five limestone blocks, four marble blocks, five hornfels blocks and four granite blocks. These samples were cut into three kinds of specimens as shown in Figure 3a for laboratory tests. For uniaxial compression tests and P-wave velocity tests, 20 long cylindrical specimens with a diameter of 50 mm and a height of 100 mm were prepared (Fig. 3b). Similarly, 20 short cylindrical specimens with an equal diameter and height of 50 mm were cut for Brazilian tests (Fig. 3c). The rest were cubic specimens with an equal edge-length of 50 mm for direct shear tests (Fig. 3d).

In uniaxial compression tests (Fig. 4a), strain gauges were attached along the specimen to measure the strain changes during the tests. The stress–strain curve can then be constructed from the axial loads recorded by a servo-controlled hydraulic actuator and strain measurements. A film of grease was applied between the loading platens and the specimen before the specimen was loaded. The application of the friction treatment at the boundaries was to eliminate the end effects (Wang *et al.* 2013*a*). Therefore, a splitting failure mode could be observed (the testing results could be influenced by the end effects if a cone failure mode was observed) (Kang *et al.* 2017). Brazilian tests (Fig. 4b) and direct shear tests (Fig. 4c) were less influenced by the end effects. Therefore, no further treatment on boundaries was implemented. Table 5 summarizes all measured mechanical parameters of intact rock in the laboratory.

#### Correlation of mechanical parameters between rock mass and intact rock

The randomness of structural discontinuities (e.g. joints) may result in a much lower strength for the rock mass (Wang *et al.* 2011, 2013*b*, 2017; Wang and Ni 2014). Hence, a representative elementary volume is often used to represent the rock mass as a homogeneous medium with ‘average’ properties for numerical analysis (Ni *et al.* 2017).

The Hoek–Brown failure criterion (Hoek & Brown 1980; Hoek & Marinos 2007) is used in this study to evaluate the reduction of the compressive and tensile strength of rocks based on their GSI values. The deformation modulus is determined by correlations with either the RMR values or the GSI values (these two calculations are equivalent in the present study, as shown in Table 2). Although the Hoek–Brown failure criterion has no direct correlation with the Mohr–Coulomb failure criterion, the Mohr's envelope was derived from triaxial compression testing data to compute the reduced parameters of cohesion and internal friction angle. All calculations were performed in the software Roclab (DeLeo 1993), and the derived reduced mechanical parameters of rock mass are listed in Table 6.

## Pillar stability analysis

### Influencing factors

The stability of pillars is critical in the design optimization of room and pillar mines, as the collapse of one pillar may induce a ‘domino-type failure’. The loads that act on a pillar can be simply computed based on the tributary area theory. Figure 5 shows schematic diagrams of design considerations for a pillar with dimensions of *L _{p}* (pillar length) ×

*W*(pillar width) ×

_{p}*h*(pillar height). The pillar will sustain the weight of roof rock (corresponding to the product of unit weight

*γ*and mining depth

*H*) acting on an area of (

*L*+

_{p}*L*

_{0}) × (

*W*+

_{p}*W*

_{0}). Here,

*L*

_{0}and

*W*

_{0}are the length and width of the room, respectively. The compressive stress

*σ*is calculated as follows: (5)The compressive strength of the pillar,

_{p}*S*, can be roughly correlated with the rock strength,

_{p}*S*, based on the

_{l}*in situ*testing data of 66 pillars (Bieniawski & Van Heerden 1975) as (6)

This equation is applicable to pillars with a ratio of pillar width to height between 0.5 and 34. When a pillar size exceeds a threshold value, the pillar strength will no longer increase. In general, a safety factor of 1.5–2.0 should be used for room and pillar mines. In 1981, Bieniawski (1981) improved the original calculation form as follows:
(7)where parameter *α* is dependent on the ratio of pillar width to height. For a pillar with a pillar width five times greater than its height, *α*_{ }=_{ }1.4 should be considered, whereas *α*_{ }=_{ }1.0 should be used in calculation for a pillar width less than five times its height.

The safety factor of a rectangular pillar can be determined (Yin *et al.* 2012) as follows:
(8)In total, there are eight factors influencing the computed safety factor for the stability of a pillar: mining depth *H*, pillar height *h*, pillar width *W _{p}*, pillar length

*L*, room width

_{p}*W*

_{0}, room length

*L*

_{0}, uniaxial compressive strength of pillar

*σ*and unit weight of roof rock

_{c}*γ*. In the present study, all pillars are assumed as squared pillars, eliminating two of the influencing factors in terms of pillar length

*L*and room length

_{p}*L*

_{0}, and equation (8) becomes (9)

### Orthogonal experimental design

A sensitivity analysis was performed using the orthogonal experimental design method (Cavazzuti 2013) to discover the most important factors. Each of the six factors (input parameters) that appear in equation (9) was assigned five levels (Table 7). The orthogonal array was constructed as L_{25}(5^{6}), where subscript 25 is the number of experimental runs, superscript six is the number of factors, and five is the number of levels. For each factor, an equal interval was set between two levels. These design parameters are normalized by following two steps: (1) subtract from all the five values the minimum value; (2) divide the subtracted five values by the difference between the maximum and minimum values. For example, the pillar height varies from 40 to 300 m with an interval of 65 m. After subtracting the minimum value of 40 m, the subtracted pillar heights range from 0 to 260 m. The subtracted pillar heights are then divided by the maximum height difference of 260 m to obtain a range of normalized pillar height varying from zero to unity with an interval of 0.25.

The safety factor was calculated in each orthogonal experimental run. The computed safety factors for each factor are averaged and plotted against normalized design parameters as shown in Figure 6a. Pillar stability decreases with increasing mining depth, room width, pillar height and unit weight of roof rock. Conversely, the larger the pillar width and compressive strength, the larger the safety factor. This is consistent with the observations of Martin & Maybee (2000) and Lee *et al.* (2013) that the use of a higher ratio of pillar width to height improves the stability of room and pillar underground excavations. Bieniawski (1992) suggested using a safety factor of 1.5 for room and pillar mines without retreat mining. It can be seen in Figure 6 that most of the combinations of design parameters result in a safety factor higher than 1.5. Although the mining depth may vary spatially at different panels, an appropriate selection of room and pillar sizes could guarantee the safety of the mine. The allowable room width is calculated as 21.03 m based on a safety factor of 1.5. Similarly, the minimum pillar width should be selected to be larger than 3.72 m.

Figure 6b shows the variations in the main interactions with design parameters. The main interaction is determined as the difference between safety factors at the highest and lowest levels (Cavazzuti 2013). This quantity provides a direct evaluation of the influence of design parameters on the stability of pillars. The significance of design parameters can be summarized as follows: room width > pillar width > mining height >_{ }pillar height > compressive strength of pillar > unit weight of roof rock. It also demonstrates that the size of room and pillar width should be controlled strictly during design optimization. Furthermore, the controlled explosion technique should be implemented during construction to minimize the damage to pillars.

### Comparison with *in situ* conditions of pillars

There are five panels with *c.* 450 pillars in the Xianglushan tungsten mine. Pillars with large diameters are excluded from the stability check, and only those pillars with small sizes and slight damage on the pillar surface are evaluated. This slight damage was found only in hornfels pillars. Based on local design experience, the uniaxial compressive strength of hornfels pillars was taken as 70% of the non-damage strength (Table 6); that is, 87.9 MPa. Safety factors were computed using equation (9) for all 450 pillars. A total of 13 of the 450 pillars had computed safety factors below 1.5 and were therefore potentially unstable (Table 8). Pouring concrete around them can be considered as an effective measure to strengthen the pillars (Tesarik *et al.* 2009).

## Design optimization of stopes

### Numerical modelling

In this investigation, the excavation process of the room and pillar stopes is modelled numerically using the software 3D-*σ* developed by Geoscience Research Laboratory Co., Ltd. Each pillar is assumed to have a height of 10 m and a square cross-section of 4 × 4 m. The room width is taken as 12 m. The average mining depth is *c.* 150 m below the ground surface. The excavation area varies from 300 to 4000 m^{2} and an excavation area of 76 × 60_{ }=_{ }4560 m^{2} is used in the numerical model. To diminish the boundary effects, the boundary was set at a distance that is five times larger than the excavation from the centre of the excavation zone (Ni *et al.* 2018). The final dimensions of the numerical model are 586 m (along the longitudinal direction of the excavation) × 556 m (along the transverse direction of the excavation) × 556 m (height).

The typical excavation process is simulated by nine steps. In step 1, an initial geo-stress field is simulated through a geostatic equilibrium analysis. In step 2, extraction of minerals is initiated and a room with a square cross-section of 12 × 12 m (excavation area of 144 m^{2}) is excavated. All following excavation steps are tabulated in Table 9.

Figure 7 shows the details of the numerical model. The lateral sides are constrained in the normal direction as smooth and rigid boundaries (Fig. 7a) and all degrees of freedom for the bottom boundary are fixed. Finer meshes are used near the excavated zone and coarser meshes at greater distance. The size of the finest mesh is determined by considering the calculation of the representative elementary volume (Ni *et al.* 2017) for the specific rocks. A 20-node brick element is used to provide better accuracy for analysis outcomes. The initial model before excavation has 30 560 elements and 129 371 nodes. The Drucker–Prager failure criterion is used for all rocks with parameters as listed in Table 6.

### Rock mass stability around stopes

For each excavation step, the stresses and displacements within the rock mass can be calculated. Figure 8 illustrates the maximum principal stress contours for four steps. The maximum principal stress increases as more volumes of minerals are extracted. The peak tensile stress occurs at the (hornfels) pillar mid-height. These tensile stresses in pillars are unfavourable to maintain pillar stability. The peak compressive stresses are primarily in the pillars, based on the contours of minimum principal stress. However, it should be noted that significant stresses are also concentrated in the limestone roof rock.

The maximum calculated stresses are plotted in Figure 9 against the excavation area. Two strength limits of limestone are also shown for reference. As given in Table 6, the compressive and tensile strength of hornfels are 125.5 and 2.66 MPa, significantly more than the strength of limestone, at 37.8 and 0.88 MPa, respectively. As a result, the stability of the room and pillar stopes depends on the stability of the limestone in this investigation. It is clear that failure in compression is not likely but rather a roof rock cave-in could occur.

Figure 10 presents the calculated maximum displacements of floor and roof rock. In general, the deformation is symmetric about the mid-height of pillars. At the end of excavation, the maximum deflection of 7.2 mm occurs at the floor, whereas the roof deforms by 7.0 mm.

### Influence of room width

The pillar heights of the Xianglushan tungsten mine are set at 5, 10 and 15 m for analysing the influence of room width through a parametric study. The values selected for the room width are shown in Table 10.

The minimum stresses obtained from numerical analyses are plotted versus normalized room width in Figure 11. The room width is normalized using the two-step normalization procedure noted above. A range of allowed room width can be determined by linear interpolation of normalized room width at the allowable tensile strength (0.88 MPa for the limestone roof). For instance, when the pillar height is 15 m, the lower and upper bounds of maximum room width are 13.65 and 15.46 m, respectively. The optimized room width should be less than the lower bound, and the calculated results are summarized in Table 10.

### Influence of pillar width

A parametric investigation was performed to determine the optimized pillar width as presented in Table 11. Figure 12 shows the changes in the calculated minimum stresses as a function of normalized pillar width ranging from zero to unity with an interval of 0.25. The pillar width that intersects with the tensile strength of 0.88 MPa, for a pillar height of 10 m and a room width of 13 m, is 6.92 m. To have a stable pillar, the optimized pillar width should be larger than the calculated value (Table 11).

### Stable excavation spans

Rock mass classification systems have normally been developed to allow a conservative estimation of stable excavation spans, especially for geotechnical applications at shallow to moderate depth. Mathews *et al.* (1981) proposed an empirical stability graph method for design optimization of room and pillar stopes at depths greater than 1000 m based on 50 case histories by correlating a stability number with a shape factor. Therefore, stopes can be dimensioned given the geometry of the stope surface and ground conditions. Potvin (1988) and Stewart & Forsyth (1995), and others, have collected further field data to confirm this relationship. Some recent analyses (Trueman *et al.* 2000; Mawdesley 2002) have extended the original Mathews *et al.* (1981) method by applying logistic regression to field mentoring results and redefined the difference between stable, failure, major failure and caving zones.

The stability number is computed as 12.14 for limestone, 12.41 for marble and 31.34 for hornfels, respectively. The shape factors of limestone, marble and hornfels are determined to be 7.8, 7.9 and 10.8 m correspondingly. The excavation area for limestone is 765 and 973 m^{2} for circular and squared stopes, and 784 and 999 m^{2} for marble. For hornfels, the excavation area is much larger, at 1466 and 1866 m^{2} respectively. The excavation area for limestone evaluated by the Mathews *et al.* (1981) method is consistent with the results of numerical analyses in the section ‘Rock mass stability around stopes’, in which tensile failure of roof rock could occur before step 4 with an excavation area of 784 m^{2}.

## Field monitoring

### Monitoring plan

The skarn scheelite deposit of the Xianglushan tungsten mine extends to a depth of *c.* 40–300 m below the ground surface. Underground excavations could induce surface subsidence, which may damage adjacent infrastructure. For the safety of the mine as well as minimizing subsidence-induced damage, the stresses, displacements and acoustic emission (AE) signals in pillars and roof rock are monitored using instruments as summarized in Table 12. In addition, other instruments, such as a total station, dumpy level and theodolite, are used to measure the ground surface deformations.

A set of early-warning alarm thresholds is proposed from this study, taking into consideration the instrument resolution, to avoid triggering a false alarm through measurement error. Ground surface displacement measurement errors are influenced by the hardness of rocks. For example, the resolution of horizontal and vertical deformations on hard rock is 2 and 4 mm, respectively, and values of 5 and 10 mm are measured for weathered rock. Depending on the type of rock, the threshold is set from 14 to 30 mm to identify the initiation of slope failure. Based on the correlation between the uniaxial compressive strength of rocks and AE events (Wang *et al.* 2013*a*; Kang *et al.* 2017), a threshold of 18 counts per min is selected for AE monitoring. For stress measurements, the threshold should be dependent on depth (Tesarik *et al.* 2009). When the deposit is at a depth of 100 m, the stress threshold is 5.5 MPa. The threshold values increase to 8.25 and 11 MPa for an excavation depth of 150 and 200 m, respectively. For metal mines in hard rocks, the occurrence of a cave-in is usually catastrophic with a roof displacement less than 20 mm (Álvarez-Fernández *et al.* 2011). Therefore, a roof displacement of 10 mm is used as the threshold.

### Monitoring results

The monitoring results for underground excavations between September 2008 and March 2009 are used to evaluate the design optimization for room and pillar stopes. No surface subsidence was observed.

Acoustic emissions were recorded on a daily basis between 4 a.m. and 6 a.m., as there was minimum disturbance to AE signals during this time period. In each borehole, the measurements were taken for 5 min and the average count number per minute was calculated. Figure 13 shows the maximum recorded AE events in each month for 12 representative AE sensors. Only the S23 sensor presented results that exceeded the threshold of 18 counts per min, where failure may occur. The AE events recorded by other sensors were all much lower than the threshold. A close-up check of AE events for the S23 sensor in September and October 2008 is presented in Figure 14. Only one discrete data point in each of the two months was observed to exceed the threshold. These accidental measurements are regarded as not representative as there was no accumulation of AE events afterwards.

Stress measurements were taken on seven occasions. Stress measurements from 12 representative stressmeters are plotted in Figure 15 with date. Six of the 30 stressmeters (Y1, Y2, Y3, Y11, Y21 and Y23 sensors) showed an increasing trend in stress, but the increments were insignificant from a practical point of view. All stress measurements were less than the minimum threshold value of 5.5 MPa.

Similarly, seven measurements of displacements using extensometers and separation sensors were conducted. The results are summarized in Table 13. The maximum displacement measured using extensometers was only 2 mm, which is much less than the threshold of 10 mm. In addition, separation sensors were used to indicate whether cave-in could occur or not. The maximum measured value of 0.2 mm on the roof is negligible.

Overall, it is reasonable to assume that the room and pillar mine is, in general, stable and the optimized design presented in this investigation is effective.

## Conclusions

Design optimization of the Xianglushan tungsten mine was performed in this investigation. Four types of rock masses are identified based on geological surveys and field measurements. The rock masses are evaluated using three rock mass classification systems, and the BQ system recommended in design guidelines was found to provide results consistent with those of the RMR system and the Q system. These data are used to scale the mechanical properties obtained from laboratory element tests to determine the parameters for use in numerical analyses.

For the orthogonal experimental design of stopes, the pillar and room sizes are the two most important factors in the stability of pillars. In numerical simulations, design optimization was carried out to determine the allowable (maximum) room width and minimum pillar width. The stable excavation span was also evaluated using the stability graph method. For room and pillar stopes with a pillar height of 5 m under a 150 m roof rock, the maximum room width and minimum pillar width are 12 and 5.5 m, respectively. Similarly, the combinations of 10 m pillar height, 13 m room width and 7 m pillar width and of 15 m pillar height, 13.5 m room width and 8 m pillar width should be used. Field monitoring results demonstrate the efficacy of the proposed design optimization.

## Funding

This work was financially supported by the National Natural Science Foundation of China (No. 51464015).

*Scientific editing by Jamie Standing; Richard Ghail*

- © 2018 The Author(s). Published by The Geological Society of London. All rights reserved