An investigation of longwall failure using 3D numerical modelling – A case study at a copper mine

The studied object is located in the A5 region (Fig. 3). The A5/1 longwall panel lied at an average depth of 800 m, was 2 m high and 50 m long, and the continuous mining length was approximately 240 m. The P1, P2, P3 and P4 crosscuts were driven due to the thermal hazard. The roof of the gate roads (headgate and tailgate) was supported by four 1.6-m rockbolt elements in the 1.5 × 1.5 grid. In addition, the tailgate and crosscuts (P2, P3 and P4) were reinforced by applying two 5.2-m cable elements.

Figure 3

Roof rocks (behind the powered roof support) were continuously deflected on to the bearing support – the LINK-N-LOCK box crib with a cross section of approximately 1 m. Cribbing was performed in a 3.5 × (4.0 ÷ 8.0) m grid (Fig. 4).

Figure 4

The longwall face was supported using the powered roof support presented in Figure 5.

Figure 5

Copper deposit (copper seam) with an average thickness of 1.6 m (0.35 ÷ 3.2 m) and a dip of 15° contains 0.05 ÷ 0.6 m of cupriferous shale, 0.15 ÷ 0.5 m of silty dolomite, 0.8 ÷ 1.8 m of shale-streaked dolomite and 0.05 ÷ 0.8 m of limy dolomite. The roof rocks consist of dolomite, limestone and anhydrite. The floor rocks are sandstones (grey and red). The lithology of rock mass in the A5 region is shown in Table 1.

Tectonic disturbance: The rock mass in the A5 region is poorly involved tectonically. The presence of faults with small throws (0.4 ÷ 3.7 m) is confirmed. The strike line of the faults is NW–SE. There are a number of vertical and skew cracks filled with gypsum and calcite. Tectonic slips occur with truncated planes of 30°.

Hydrogeological conditions: In the A5 region, carbonate rocks are strong and poorly fractured. Rock mass in the A5 region is practically unsaturated. The thick layer of impermeable anhydrite and silty shale is a separating layer against the above-lying aquifers. Traces of UG (underground) water can be found in the floor rocks (sandstones). During mining operation, the occurrence of small fluxes is possible. Larger inflows into UG workings are not anticipated.

Mechanical parameters of the intact rocks in the A5 region are shown in Table 2.

The experimental longwall panel was subjected to 24-hour seismological monitoring. The basis for assessing the roof stability of the gate roads was regular visual observations of the state of the roof and support, as well as the side walls and floor. Visual observations were conducted once a day. In addition to visual observations of the rock mass behaviour, the following were also conducted:

convergence measurements in gate roads and CKN (Constrained Kohonen Network) (sensors at least twice a week),

Photoelectric sensors (DLN-76/4/500) were used to monitor trends and the speed of rock deformation surrounding the gate roads. They measure changes in distance ΔD between the opposite walls of a borehole with a diameter of D = 76 mm and deformation ΔL of the rock along the axis of the hole relative to the base with a length of L = 500 mm. These sensors were installed in the wall sides of the tailgate at a distance of 8–10 m from the end of each mining block.

Convergence sensors (CKN-2000) were used for measuring convergence in UG workings and checking the roof deflection in gobs. These sensors measure the change in distance ΔH between the roof and floor of gate roads. These convergence sensors were installed in the middle of each cross heading between mining blocks and in gobs.

Figure 6 shows the location of convergence measuring points in the headgate (1 ÷ 14), tailgate (15 ÷ 22) and in gobs (CKN 171, CKN 173, CKN 178, CKN 176 and CKN 172). In the headgate and tailgate, distance between measuring points was about 30 m.

Figure 6

Analysis of longwall failures was carried out using FLAC3D software (Itasca, 2012) [4]. The A5/1 longwall panel was modelled according to the geological and mining conditions mentioned in Section 3. Figure 7 shows the outline of the 3D numerical model and the location of the rock layers surrounding the A5/1 longwall. In all the models considered, the same block was cut out from the rock mass which is a three-dimensional deformation state. The model was divided into 632,000 elements and had dimensions of 500 m length, 250 m width and 100 m height (Fig. 7).

Figure 7

Hypothetically, the initial horizontal stress is equal to the vertical stress. The initial stress value was calculated by the formula presented by Biliński [21] that described the geological and mining conditions in Poland: (1) q=0.02⋅H⋅mc⋅cosα q = 0.02 \cdot H \cdot {m_c} \cdot \cos \alpha where q is the primary pressure (MPa), H is the mining depth (m), m_c is a factor dependent on the geological and mining conditions of the studied site and α is the coal seam inclination (°).

For the purposes of this work, numerical calculations were carried out using the plastic model group (Mohr–Coulomb model and strain-softening model). The failure envelope for these models corresponds to the Mohr–Coulomb criterion (function of shear yield) with tension cut-off (function of tensile yield). The Mohr–Coulomb model takes into consideration the plasticity of the rock mass, which is the non-linearity of its stress–strain characteristics. The strain-softening model enables representation of non-linear material softening behaviour based on prescribed variations of the Mohr–Coulomb model properties (i.e. shear strength) as functions of the deviatoric plastic strain [4]. The plastic flow law in FLAC3D rests on basic assumptions from plasticity theory that the total strain increment may be divided into elastic and plastic parts, with only the elastic part contributing to the stress increment by means of an elastic law. In addition, both plastic and elastic strain increments are taken to be co-axial with the current principal axes of the stresses. However, this is only valid if elastic strains are small compared to plastic strains during plastic flow. The flow rule specifies the direction of the plastic strain increment vector as being normal for the potential surface; it is called associated if the potential and yield functions coincide, and non-associated otherwise [4].

The model was originally developed as an elastic model to achieve the original stress state. Then, the displacement and velocity vectors were zeroed. During the next step, the ‘null’ model was assigned to the zones which corresponded to the extracted copper deposit, after which support applied for the longwall panel was assigned appropriate parameters and the model was recalculated.

Back analysis is a common technique for determining soil/rock properties by measuring the convergence of an underground opening. This approach has been widely applied to identify in situ stress field rock mass deformation modulus and strength parameters, rock mass hydraulic properties, rock mass zoning, boundary conditions, loads acting on tunnel linings, etc., through direct application of closed-form solutions or numerical methods [24,25,26,27,28,29,30,31,32,33,34]. Sakurai stated that long-term engineering experience, field measurements and back analyses are the most powerful tools in rock engineering practice [35]. Nowadays, back analysis has become an integral part of the observational method commonly used in rock engineering. To ensure excavation safety, field measurements are conducted during excavation operations to monitor possible differences between the conditions assumed in the design model and those actually observed in the field. If a difference is detected, a back analysis allows the design engineers to improve the values of the mechanical parameters of the rock and to modify, if necessary, the design or the extraction operation procedure.

Deformation monitoring data is usually used as the input data for back analysis because displacement measurements can be routinely conducted in situ and it is one of the most reliable measurable quantities in the field [24].

In this study, back analysis was conducted by matching the results of numerical modelling with the measured roof convergence as outcomes of the installed monitoring systems described in Section 3.3.

Figure 12 shows a comparison of the vertical convergence of in situ measurements and modelling results. The FLAC3D result values are close to in situ measurements, which indicates high accuracy of the trend function match.

Figure 12

Based on the results of back analysis, a set of mechanical parameters that can be adopted for further numerical modelling is shown in Table 4.

The longwall face was simulated using two material models mentioned in Section 3.1 (the classic Mohr–Coulomb model and the strain-softening model). It should be noted that these strong but thin rock layers (compact limy dolomite) located immediately above the copper deposit (Table 1) behave like elastic brittle or as strain softeners. This means that the post-failure characteristics can be calculated by reducing the rock mass strength values from the in situ values [36,37]. Mechanical parameters of rocks for the strain-softening model (the assumed plastic strain parameter at the residual strength was 0.1) are shown in Table 5.

Displacement contour and failures occurred around the longwall face with the two selected material models, which are shown in Figures 13–16. The size of the failure zone and the displacement values with the strain-softening model were greater than those found using the classic Mohr–Coulomb model. In the case of the strain-softening model, both active shear and tensile failure (marked in brown, violet and blue) occurred around the longwall face. These failures were the main cause of roof falls (Fig. 15). In addition, vertical displacement in the case of strain-softening model was approximately 90 mm in the place where active tensile failure occurred (Fig. 16).

Figure 13

Figure 14

Figure 15

Figure 16

Numerical simulation results using the strain-softening model are similar to the field observations, as shown in Figures 1 and 17. This means that simulation with the strain-softening model is highly capable of characterising the behaviour of brittle rock mass and the longwall failure mechanism in such a case study. This is also confirmed by many researchers [38,39,40,41,42]. In the following simulation, the strain-softening model is used to analyse the impact of selected factor on longwall failures.

Figure 17

Based on the geological and mining conditions in the 5A region, different operating factors were considered in order to investigate the longwall face failures and prevent those failures from threatening the future copper longwall operation. These different operating factors were: tip-to-face distance, load-bearing capacity of the powered roof support, spacing of the box crib and the roof control method. Therefore, different scenarios were used as shown in Table 6.

The plasticity indicator around the longwall face with different tip-to-face distances is shown in Figure 18.

Figure 18

It should be noted that shear failure only occurred in the case of tip-to-face distance of 1.5 m, while in the case of tip-to-face distance of 3 m, both shear and tensile failures of rock mass occurred around the longwall face, in contact with the powered roof support canopy. This means that the possibility of roof rock falls when there is a 3-m-long tip-to-face distance is higher than that when there is a 1.5-m-long tip-to-face distance.

The plasticity indicator around the longwall face with different box crib spacing is shown in Figure 19. The size of the failure zone above the longwall panel decreased due to the reduction of box crib spacing. Both active shear and tensile failures occurred in the case of 6-m box crib spacing (Fig. 19a). In the case of 1.5- and 3-m box crib spacing, only active shear failure occurred around the longwall face (Fig. 19b, c).

Figure 19

In mining practice, backfilling is considered to be the best option when it comes to improving the roof conditions. In the case of applying backfilling (e.g. sand), the size of the failure zone around the longwall panel decreased significantly. The size of the active failure zone around the longwall face also decreased (Fig. 20). Both roof and floor rocks behind the powered roof support will be evenly deflected in lower scale in comparison with the case of applying box crib (Fig. 19). Backfilling aims to minimise displacement of roof and floor rocks to the mined space. In consequence, it prevents failures around the longwall.

Figure 20

Figure 21 shows the plasticity indicator around the longwall face with different operating load-bearing capacities of the powered roof support. The size of the active shear and tensile failure zone decreased with higher load-bearing capacities. This means that an increase in the load-bearing capacity improves the roof conditions.

Figure 21

A simulation with lower tip-to-face distance (1.5 m), higher load-bearing capacity of the powered roof support (4000 kN) and lower spacing of the box crib (1.5 m) was conducted. The results showed that the size of the failure zone above the longwall panel decreased in comparison to the originally designed case, as shown in Figures 18a, 19a and 21a. In addition, the active failure zone around the longwall face also decreased significantly (Fig. 22). This means that changes in the selected factors combined clearly improve the roof conditions, avoiding possible failure around the longwall face.

Figure 22