Bearing characteristics of prestressed full-length anchoring rockbolt based on Wavelet Neural Network

Abstract

In order to improve bearing capacity of rockbolts in deep-buried coal mine roadways, orthogonal tests were conducted to study influencing factors of rockbolt anchoring effect. Wavelet neural network model was introduced to predict the pull-out force of rockbolt. The activation and output functions of the wavelet neural network were improved, and the scaling and translation parameters were also modified by using the gradient descent method. These improvements enhanced the approximation rate of the wavelet neural network model, and solve the problem that the wavelet transform method is monotonous and difficult to adapt to the complex and variable engineering conditions. Research results illustrated that The value of the ultimate pull-out force is positively correlated with the strength of the specimen and pre-tension value of the specimen. According to the test results, the coal mine roadway support scheme was optimized, and the high prestress full-length anchoring rockbolt support technology was proposed. The effectiveness of research was verified through the engineering applications and in-situ monitoring results.

Keywords

Rockbolt improved Wavelet Neural Network high prestress pull-out test field performance

1 Introduction

Rockbolt support has been widely used in underground engineering. Compared with other support technologies, anchor support has the advantages of small disturbance to the surrounding rocks, convenient construction safety, fast construction speed, low construction cost and simple construction technology, and has achieved great economic benefits. Rockbolts must provide sufficient anchor capacity for safe construction. With the increasing depth of underground engineering, the demands on the surrounding rock supports have increased significantly. Consequently, the requirements for anchoring bearing capacity provided by a single rockbolt were also increased.

As a result, a large amount of research has been carried out by several researchers in the field of rockbolt anchoring strength improvement. Huang used a combination of theoretical analyses, laboratory tests, numerical simulations and underground measurements to study the performance of resin anchored rockbolt and the factors that affect them [1]. Yu and Qi studied the anchoring effect of prestressed rockbolts by means of experiments and numerical simulations and concluded that prestressed rockbolts could effectively inhibit the opening deformation of the fracture surface of the surrounding rocks [2, 3]. Kang applied different preloads to the rockbolts and obtained that the anchor preload had the effect of enhancing the strength and stiffness of the anchored surrounding rocks [4]. Yu and Høien discussed the shear stress and axial stress distribution law of prestressed anchor section under load from the theoretical analysis, and obtained the influence of different influencing factors on the stress distribution law of anchored section of rockbolt [5, 6]. Yang and Aghchai analyzed the shear stress and axial force of rockbolts under different anchoring lengths from experimental analysis to improve the bearing capacity of the anchoring body [7, 8]. Vlachopoulos and Yang carried out the anchoring mechanics mechanism analysis by rockbolt pull-out test [9, 10]. Bastami, Thenevin and Li changed the anchoring mode of anchor by fast and slow curing resin to improve the anchoring support effect [11 –13].

In the field of civil engineering, there are numerous factors that influence the strength of engineering structures, and these factors exhibit a certain degree of randomness. Therefore, some researchers have started using different artificial intelligence models to predict the strength of engineering structures. Najafzadeh employed the neuro-fuzzy based group method of data handling to predict the flushing depth of bridge piers and pile groups [14, 15]. Movahed proposed a modified group method of data handling by utilizing the extreme learning machine and applied it to predict the longitudinal dispersion coefficient of water supply pipes [16]. Yao Predicted the settlement of metro stations by using improved BP neural network model [17]. Zou develops an integrated optimization control model which combines the utilization of back propagation artificial neural network (BP-ANN), Elman neural network (ENN), and adaptive neuro fuzzy inference systems (ANFIS) for obtaining high-quality tunnel smooth blasting [18]. Tan proposed a predictive model of rock creep parameters by means of gene expression programming (GEP) [19]. Asare established a prediction model which combined autoencoder neural network and multivariate adaptive regression spline for uniaxial compressive strength of rocks [20]. Armaghani predicted the rock brittleness by applying support vector machine (SVM) based models [21]. Currently, artificial intelligence technology has been widely applied in the field of civil engineering, and many researchers have developed rock strength prediction models based on artificial intelligence technology. However, the application of artificial intelligence in the field of anchor bolt anchoring and rock support is relatively limited. Although previous researchers have made many valuable research achievements, the research on improving the anchoring bearing capacity by anchoring influencing factors still needs to be improved, and further experimental work and optimization of rockbolt support need to be carried out.

In this paper, the orthogonal test of rockbolt pull-out under different anchoring parameters was carried out. Through the study of the whole process of rockbolt pull-out, the whole process of rockbolt displacement and rockbolt load change during the rockbolt pull-out process was analyzed. The influence law of the four factors of specimen strength, pre-tension torque, anchoring thickness (anchor grouting ring thickness) and anchoring section length on the ultimate pull-out force of rockbolt was analyzed based on improved Wavelet Neural Network. According to the test results, the anchoring support design of coal mine roadways was optimized. The optimized roadway support scheme was applied at the construction site of coal mine roadway and the load of anchoring was monitored. The monitoring results verified the correctness of the research conclusions.

2 Rockbolt pull-out mechanical tests

2.1 Test scheme

The pull-out force of rockbolt refers to the force that prevents the rockbolt from being pulled out of the rock mass, and it is a key parameter determining the anchoring effect of the rockbolt. In this test, the orthogonal test scheme was used to analyse the effects of test block strength, prestress level, anchorage thickness and anchorage section length on the rock mass anchored by the anchors. The test scheme is shown in Table 1. To facilitate the test operation, the preload torque was recorded instead of the prestress.

Table 1
Test scheme of rockbolt pull-out tests

Specimen number Specimen strength (MPa) Anchoring length (mm) Anchoring thickness (mm) Pretension torsional moment (N·m)

1-1 10 60 2 30

1-2 10 80 4 40

1-3 10 100 6 50

2-1 30 60 4 50

2-2 30 80 6 30

2-3 30 100 2 40

3-1 50 60 6 40

3-2 50 80 2 50

3-3 50 100 4 30

Specimen number	Specimen strength (MPa)	Anchoring length (mm)	Anchoring thickness (mm)	Pretension torsional moment (N·m)
1-1	10	60	2	30
1-2	10	80	4	40
1-3	10	100	6	50
2-1	30	60	4	50
2-2	30	80	6	30
2-3	30	100	2	40
3-1	50	60	6	40
3-2	50	80	2	50
3-3	50	100	4	30

The tests was carried out in order to understand the main forms of anchorage damage, to analyse the main characteristics of the anchor solid in each phase under tension load; to analyse the role of anchorage factors on the mechanical form of the anchor solid interface and the effect on the failure of the anchor interface; to study the anchorage effect produced by its different factors and to determine which influencing factors are the main ones.

2.2 Test equipment and materials

2.2.1 Test equipment

The maximum load of the WAW-3000 universal testing machine is 300 kN and the loading speed is adjusted to 0.5 mm/min. The load and displacement changes can be recorded automatically, see Fig. 1a. The test will be stopped if the following conditions are encountered: (1) the rockbolt is broken or the epoxy is pulled out; (2) the test block as a whole is damaged; (3) the displacement at the loading end does not converge; (4) the test data fluctuates greatly, the load will no longer be increased and the test will be repeated.

Fig. 1

Test equipment and tools. (a) WAW-3000 universal testing machine and specimen; (b) Torque wrench.

Tailor-made fixture was used to fix the test specimens (shown in Fig. 1a). Four threaded rods of 18 mm diameter and 200 mm length are used to fix the upper and lower clamping plates to the test block by means of nuts. Four threaded rods of 18 mm diameter and 200 mm length are used to fix the upper and lower clamping plates to the test specimen using nuts. To ensure the rigidity of the fixture, the upper and lower clamping plates are made from 300 mm×300 mm×30 mm steel plates, a square hole of 130 mm×130 mm is machined in the middle of the upper plate, and a 20 mm diameter metal rod is welded in the middle of the lower plate. The upper and lower clamping plates are used to hold the test specimen. The surface of the upper and lower plates must be smooth and flat, and to ensure accurate experimental results, petroleum jelly is applied to the surface of the test specimen to smooth and flatten the surface of the test specimen. During the experiment, the lower side chuck of the test machine holds the welded rod of the fixture and the upper side chuck holds the exposed end of the rockbolt of test specimen. In order to prevent test failure due to the tilt of the specimen bottom surface or the deviation of the rockbolt during loading, the exposed section of the rockbolt and the welding rod must be on the same line, thus ensuring that the testing machine loading always follows the direction of the rockbolt axis.

A torque wrench is used to apply a predetermined torque to the rockbolt nut to achieve the pre-stressing effect, and the torque applied by the torque wrench is 30N·m, 40N·m, and 50N·m. The torque wrench used in test has a maximum range of 210N·m and a minimum loading error of 1N·m, which can meet the requirements of the loading pre-stressing (Fig. 1b).

2.2.2 Test materials

Test specimens include materials such as test block, anchor, epoxy, rockbolts and gaskets. In order to reflect the effect of different material strength, test blocks were made of gypsum (axial compressive strength f_c = 6 N/mm², tensile strength f_t = 0.78 N/mm²), C30 cement (f_c = 14.3 N/mm², f_c = 1.43 N/mm²) and C60 cement (f_c = 27.5 N/mm², f_c = 2.04 N/mm²), and the size was 150 mm×150 mm×150 mm. The processed test specimens are shown in Fig. 2. After 28 days of curing, the diameters of the central drill holes of the test blocks were 20 mm, 24 mm and 28 mm respectively, and the depths were 60 mm, 80 mm and 100 mm respectively. In order to ensure the accuracy of the test, three specimens of each type were made in the test program.

Fig. 2

Test specimens.

The rockbolt model was made of φ16 mm threaded steel bars with the same material as the real rockbolt. In order to verify the influence of anchorage length on the mechanical behavior of the anchorage section interface, the anchorage lengths were taken as 60 mm, 80 mm and 100 mm respectively, as shown in Fig. 2.

Epoxy resin A and epoxy resin B were mixed in a ratio of 2:1 to form the anchoring material. After accurately positioning the rockbolt in the base, the anchoring grouting was carried out smoothly.

The epoxy resin was first used to anchor the rockbolt to half its depth, and after it had completely solidified, a gasket was placed on the surface of the specimen. A common φ16 mm coarse thread nut was used to apply pre-stress.

3 Test results

3.1 Failure mode of rockbolt pull-out

The rockbolt and anchoring grout are usually referred to as the anchoring body, and the rockbolt pull-out force is the most intuitive way to reflect the anchoring bearing capacity. Therefore, the distribution form and size of the rockbolt pull-out force can effectively describe the anchoring effect of the anchoring body.

From the test results, the pull-out failure modes mainly include the cracking of the specimen base and the complete pull-out of the anchoring body from the specimen base, indicating that the damage of the anchoring body and the hole wall interface is the main cause of the overall failure of the anchoring system. Figure 3a, Fig. 3b and Fig. 3c respectively show the pull-out of the anchoring body from the specimens of gypsum, C30 cement and C60 cement. The gypsum specimen and C30 cement specimen firstly show the cracking phenomenon of the specimen, and then the anchoring body is pulled out. The situation of the base damage caused by the pull-out of the anchoring body from the base is shown in Fig. 4a and Fig. 4b. With the pull-out of the anchoring body, small cracks appeared around the anchoring body and quickly developed into larger cracks, eventually leading to the failure of the anchoring section of the gypsum or cement blocks and part of the base material attached to the rockbolt was pulled out. When the C60 cement specimen was damaged, a crisp sound was emitted, and the rockbolt and epoxy resin were pulled out together, as shown in Fig. 4c.

Fig. 3

Failure characters of test specimens in rockbolt pull-out tests. (a) gypsum specimen. (b) C30 cement specimen. (c) C60 cement specimen.

3.2 Analysis of full loading-displacement of rockbolt pulling

Table 2 shows the full load-displacement data of each specimen. The damage of specimens 1-1, 1-2, 1-3 and 2-1, 2-2, 2-3 was caused by the cracking of the anchor segment rock mass, and the anchor body did not reach the desired pulling force in gypsum and C30 cement. However, the anchor body of specimens 3-1, 3-2, 3-3 was pulled out integrally, and the anchor body reached the corresponding bearing peak value as shown in Table 2.

Table 2
Test results of rockbolt pull-out test

Specimen number Specimen strength (MPa) Pretension torsional moment (N·m) Anchoring thickness (mm) Anchoring length (mm) Peak pull-out force (kN)

1-1 10 30 2 60 3.585

1-2 10 40 4 80 8.33

1-3 10 50 6 100 7.4

2-1 30 50 4 60 13.7

2-2 30 30 6 80 23.8

2-3 30 40 2 100 31.3

3-1 50 40 6 60 38.9

3-2 50 50 2 80 50.75

3-3 50 30 4 100 50.6

Specimen number	Specimen strength (MPa)	Pretension torsional moment (N·m)	Anchoring thickness (mm)	Anchoring length (mm)	Peak pull-out force (kN)
1-1	10	30	2	60	3.585
1-2	10	40	4	80	8.33
1-3	10	50	6	100	7.4
2-1	30	50	4	60	13.7
2-2	30	30	6	80	23.8
2-3	30	40	2	100	31.3
3-1	50	40	6	60	38.9
3-2	50	50	2	80	50.75
3-3	50	30	4	100	50.6

When the rockbolt is pulled, the interface layer of the anchor body deforms and expands due to the constraint of the base material. The higher the strength of the specimen, the higher the frictional resistance obtained, thus the greater the anchoring force. It can be seen that the strength of the base material has a significant effect on the bearing capacity of the anchor body.

The load-displacement characteristic curve recorded by the test machine is shown in Fig. 5. Under the action of load, the interface between the anchor body and the hole wall undergoes a process of elastic bonding, plastic deformation and detachment sliding from the beginning of deformation to the final failure. Taking specimen 2-2 as an example, as shown in Fig. 5b, the anchor body goes through the elastic bonding stage (0-a), the elastic-plastic transformation stage (a-b), the plastic deformation development stage (b-c), the initial detachment sliding stage (c-d), and the detachment sliding stage (d-e-f) in the pulling process. However, different influence factors play different roles in different test schemes, resulting in different loading-displacement curves, which are manifested in different failure forms of specimens in the test. When the influence of the anchoring rock body is small and the pulling force does not reach the maximum load, the specimen as a whole has already cracked, showing the sudden failure of the specimen. When the influence of the anchoring rock body is large and the pulling force reaches the maximum anchoring load, the specimen will not suddenly break, but the anchor body is gradually pulled out until it fails, showing the stable failure of the anchor body.

Fig. 5

Pull out force-displacement curve of test specimens. (a) Pull out force-displacement curve of all test specimens. (b) Pull out force-displacement curve of specimen 2-2.

As the load increases gradually, the relative displacement of the specimen base and the anchoring body on both sides of the interface layer increases, going through the elastic stage and plastic deformation stage, and continuing to develop. At this time, the interface layer will produce swelling, and the near end will first show debond phenomenon. If the influence of the anchoring rock body is small, especially when the strength of the specimen is small, the restraint of the base material is lost, and the specimen cracks and causes sudden failure, the tensile force will suddenly drop before reaching the maximum value. If the influence of the anchoring rock body is heavy, the restraint of the base material is strengthened, the plastic zone and elasticity will continue to move to the far end of the anchoring body, and the debond section will continue to expand downward. Therefore, in this stage, the anchoring body is stably pulled out, that is, stable failure. It should be emphasized that the influence of the anchoring rock body is not unlimited, because the increase of anchoring force is very limited, which indicates that when the anchoring influence factors are combined to a certain stable range, the anchoring effect has the optimal results.

4 Wavelet neural networks prediction model of rockbolt pull-out force and stochastic improvements

The conventional BP neural network has a wide range of applications, characterized by its high-level nonlinear mapping and generalization capabilities. It has a high fault tolerance rate and data collection ability in the process of working investigations, and can use its computational sequence to summarize and organize the input data, achieving the goal of self-learning. However, its disadvantages are also apparent: the self-learning rate of the BP neural network is relatively low, and when adjusting the function weights within the neural network, the threshold setting should not be too large, otherwise it will lead to many limiting factors in the calculation process of the BP neural network, resulting in low efficiency when repeatedly measuring similar information. There are many factors affecting the anchor pull-out force, and among these factors, the specimen strength has certain randomness due to the random distribution of cracks in the rock. Obtaining the specimen strength requires destructive testing, so it is impossible to accurately determine the strength of each specimen.

The wavelet neural network has adaptability and can automatically adjust the network structure and parameters according to the characteristics of the data. This adaptability makes the wavelet neural network more flexible and robust when dealing with complex and highly nonlinear problems. At the same time, the wavelet neural network can perform wavelet transformation and coefficient compression on the data, achieving dimensionality reduction and noise reduction. This compression capability makes the wavelet neural network more efficient when dealing with large-scale or high-dimensional data. Therefore, compared to traditional BP neural networks, the wavelet neural network has better capabilities and advantages in handling non-stationary signals, adaptability, and data compression.

4.1 Wavelet Neural Network

Artificial neural network algorithms often exhibit instability issues in practical applications, characterized by the network easily getting trapped in local minima. The neural network algorithm calculates the adjustment amount for the next iteration of weights based on the accumulated mean squared error from the previous iteration, along with the first-order partial derivatives with respect to each weight and threshold. However, the mean squared error function can sometimes be located at a local minimum in a particular weight adjustment direction, which can slow down the convergence speed of the network and adversely affect the accuracy of network predictions. In some cases, it can even prevent the network from converging.

A Wavelet Neural Network (WNN) is a type of artificial neural network that uses wavelets as activation functions in its hidden layers. It combines the ability of wavelets to analyze signals at different scales with the capability of neural networks to learn from data, making it suitable for signal processing and pattern recognition tasks [22]. Wavelet analysis has excellent time-frequency localization properties and properties of multi-resolution approximation. By using wavelet basis functions instead of the Sigmod function as the activation function, combining the characteristics of wavelet analysis with the self-organizing learning ability and adaptability of neural networks, the predictive capability and accuracy of the network algorithm are enhanced. The structure of WNN is shown in Fig. 6.

Fig. 6

Wavelet neural network.

As shown in Fig. 6, the activation function of the hidden layer can be represented as:

$g_{j} (x) = \prod_{i = 1}^{P} Ψ (\frac{x_{i} - b_{ij}}{a_{ij}})$ (1)

Where, Ψ represents the corresponding wavelet operation; x is the network input; i represents different input wavelet elements in the network; j is the intermediate layer code in the network; a_ij and b_ij represent new scaling and translation parameters after transformation, respectively.

Therefore, the output function of wavelet neural network can be expressed as: $f_{i} (x) = \sum_{j = 1}^{h} ω_{ij} g_{j} (x)$ (2)

Where, h is the number of levels in the wavelet network; ω _ij indicates the weight of the output.

4.2 Stochastic improvements of WNN

Although the functions and parameters of traditional wavelet networks are obtained through mathematical transformations of wavelets, their transformation methods are fixed and may not be able to adapt to the complex and changing conditions in practical engineering applications. This can also lead to a decrease in the approximation rate of the algorithm. Therefore, the random improvement of the wavelet neural network is considered from the following two aspects.

(1) In order to solve the approximation rate problem, the improved excitation function and output function are respectively:

${g^{'}}_{j} (x) = \prod_{i = 1}^{P} Ψ (\frac{\sum_{k = 1}^{n} ω_{j k} x_{k} (t) - b_{j}}{a_{j}})$ (3)

$f_{j}^{'} (x) = \sum_{j = 1}^{N} ω_{ij} \prod_{i = 1}^{P} Ψ (\frac{\sum_{k = 1}^{n} ω_{jk} x_{k} (t) - b_{j}}{a_{j}})$ (4)

Where, k represents code of the output layer, N is the total number of wavelets, P is the number of training sample spaces.

The error function can be expressed as: $E = \frac{1}{2} \sum_{p = 1}^{P} \sum_{i = 1}^{N} {(d_{i}^{p} - y_{i}^{p})}^{2}$ (5)

Where, d represents the mathematical expectation of the output value; y_i represents the actual network output value.

(2) At the same time, using the method of gradient descent, the expansion and migration parameters are improved: $a_{j} (t + 1) = a_{j} (t) + η_{a} Δ a_{j} (t)$ (6) $b_{j} (t + 1) = b_{j} (t) + η_{b} Δ b_{j} (t)$ (7)

Where, η _a and η _b are training factors, and the gradient indexes of Δaj(t) and Δbj(t) are respectively the stretching parameters and the displacement parameters. The gradient indexes of the improved two parameters are obtained through Equations (9) respectively. $Δ a_{j} (t) = - \frac{\partial E}{\partial a_{j} (t)}$ (8) $Δ b_{j} (t) = - \frac{\partial E}{\partial b_{j} (t)}$ (9)

Through the improvement of the above two aspects, the stochastic wavelet neural network can solve the problem of the approximation rate as a whole, but also can adapt to the complex and changeable random conditions, and become an effective tool for artificial intelligence prediction.

4.2 Prediction model of rockbolt pull-out force by stochastic wavelet neural network

The pull-out force of rockbolt is affected by anchoring length, rock strength, anchoring thickness and nut torque, so the above 4 parameters are taken as the input of the stochastic wavelet network rockbolt pull-out force prediction model, and the rockbolt pull-out force is taken as the output.

In wavelet neural network, the selection of hidden layer element is very important. The number of hidden cells is too small, and the whole network cannot handle the information well. Too many hidden units will directly lead to structure redundancy and local minimization. In order to balance the relationship between the two, the following formula is usually used to determine the number of hidden units in the wavelet neural network. $Z = \sqrt{mn + 1.68 n + 0.93}$ (10)

Where, Z is the number of hidden layer units, n is the number of network input, m is the number of network outputs.

Considering the number of input and output of the rockbolt pull-out force prediction model, n = 4 and m = 1 were substituted into Equation (10), and Z = 3.41 was obtained. Therefore, the number of hidden layer elements of the foundation pit surface settlement prediction model based on stochastic wavelet network was set as 3.

In order to verify the accuracy of prediction model of rockbolt peak pull-out force, the predicted results were compared with experimental results (shown in Table 3). The predicted results indicated that the prediction model of rockbolt peak pull-out force satisfies the demands of engineering applications.

Table 3

Prediction results of peak pull-out force

Input				Output
Specimen strength (MPa)	pretension torsional moment (N·m)	Anchoring thickness (mm)	Anchoring length (mm)	Experimental Peak pull-out force (kN)	Predicted Peak pull-out force (kN)	Error(%)
10	30	2	60	3.585	3.423	–4.52
10	40	4	80	8.33	8.724	4.73
10	50	6	100	7.4	7.835	5.88
30	50	4	60	13.7	14.012	2.28
30	30	6	80	23.8	25.126	5.57
30	40	2	100	31.3	32.965	5.32
50	40	6	60	38.9	40.021	2.88
50	50	2	80	50.75	52.358	3.17
50	30	4	100	50.6	53.618	5.96

4.3 Influence factor sensitivity analysis of rockbolt anchoring

The rockbolt peak pull-out force prediction model was used to get the values of peak pull-out forces of rockbolt under different variates, and the influence factor sensitivity were studied. It can be seen from Fig. 7 that the peak pull-out force of the anchor body increases with the increase of block strength and prestress, decreases with the increase of anchoring thickness, and there is a critical length for anchoring length. Within the critical length, the peak pull-out force of the rockbolt rod increases with the increase of anchoring length, and beyond the critical length, the pull-out force of the rockbolt rod decreases with the increase of anchoring length.

Fig. 7

Analysis of the influence of anchorage factors. (a) Specimen strength; (b) Prestress; (c) Anchoring thickness; (d) Anchoring length.

The magnitude of the pull-out force is related to the mechanical properties of the specimen strength. With the increase of the strength of the test specimen base, the pull-out force of the rockbolt rod increases. The higher the strength of the test specimen base, the larger the internal cohesion and internal friction angle, and the smaller the shear stress of the anchoring section of rockbolt rod during the pull-out tests. The larger the pull-out force of the rockbolt rod, the stronger the anchoring bearing capacity. Similarly, for the test specimen base with weaker strength, such as gypsum, the internal friction angle and cohesion are much smaller than that of C60 cement cohesion and internal friction angle. During the pull-out of the rockbolt rod, the shear force at the anchoring interface is greater than the tensile strength of the gypsum, and the influence of the shear force at the anchoring interface gradually decreases along the transverse direction of the test specimen. When the pull-out force reaches the tensile strength of the gypsum, The bottom fracture surface of the anchor body presents a bowl-shaped profile.

A comparison of the peak pull-out force of the anchor body under different pre-tightening torques shows that the pull-out force of the anchor body is the greatest when the pre-tension torque is 50N·m. After the peak strength of the specimen with a pre-tension torque of 30N·m, the axial stress dropped rapidly and the residual strength was low, showing a significant brittle failure characteristic. After the peak strength of the specimen with a pre-tension torque of 50N·m, the full load-displacement curve showed a “stair-like” downward trend, and the residual strength increased significantly, indicating that the rockbolt applied pre-stress to produce compressive stress in the specimen, and the specimen was compressed and compacted to a certain extent. Due to the limited volume of the specimen, the compaction effect of pre-stress can only be produced within a certain space. With the increase of pre-stress, the trend of rock reinforcement increases gradually, which is consistent with the result of Fig. 7b.

When the anchoring length exceeds the critical value, further increasing the anchoring length will cause the shear stress distribution of the anchoring section to move towards the far end of the interface, making the distribution more uniform. However, with the increase of anchoring length, the mechanical properties of the anchoring section did not change. When the rock near the anchoring section was detached, the tensile load had exceeded the chemical bonding of the anchoring section, and the tensile force did not increase effectively.

5 Engineering application of pre-stressed full-length anchoring rockbolt

5.1 Analysis of pre-stressed full-length anchoring rockbolt and surrounding rock support

From the previous test results, it can be seen that improving the rock strength and increasing the pre-stress can effectively improve the rock bearing capacity. Achieving the goal of increasing rock strength while increasing anchor pre-stress is the key to support of surrounding rocks. In order to realize the pre-stressed full-length anchoring process, self-drilling pre-stressed full-length anchoring rockbolt is designed, as shown in Fig. 8.

Fig. 8

Self drilling pretension full length anchoring rockbolt.

The pre-stressed full-length anchoring rockbolt is hollow in the rockbolt rod, a mixer is set at the front end of the rod body, a drill bit is installed at the end of the rockbolt rod, a resin outlet is set at the position of the drill bit, a water separator is installed at the end of the rockbolt rod, and the nut as well as rockbolt rod are connected together by pins. When installing the rockbolt, the bolter drives the rockbolt rod to rotate, and drills the rockbolt into the surrounding rocks. When the rockbolt is completely drilled into the surrounding rocks, the high-pressure water channel is opened, and the piston is driven to make the slow anchoring resin and fast anchoring resin in rockbolt pass through the inner cavity of the rockbolt in sequence, and then discharged from the resin outlet through the mixer. After the grouting resin is injected into the rockbolt borehole, the torque of the bolter is increased, and the pin on the nut is broken to tighten the nut and apply pre-tension to the rockbolt. during the slow anchoring resin curing process, the bolter is continuously operated to make the nut continue to apply pre-stress full-length anchoring of rockbolt is achieved. To prevent the drill bit from overheating, cooling water is passed through the splitter on both sides of the drill bit to cool and flush the drill hole during the full process of the rockbolt drilling operations.

5.2 Field application of pre-stressed full-length anchoring rockbolt

The field test site was selected at the ventilation roadway of the 17102(3) longwall mining face of the Pansan coal mine, Huainan, China. The roadway surrounding rocks consists of mudstone, fine sandstone and coal seam. The tip-anchored rockbolts have diameter of 22 mm and length of 2400 mm were installed within the working site with the interval of 750×800 mm. Due to the fracturing of the roadway surrounding rocks, high stress in the deep strata of the coal mine, combined with the excavation disturbance of adjacent roadways and other factors, the serious deformation of the roadway surrounding rocks was encountered.

In this test, each section of the roadway was supported in a different way and each test section was 50 m long. Force-measured rockbolts and displacement gauges were used for monitoring. In the field test, different support methods are used for each roadway section, and each test section is 50 m long. The first section uses the original roadway support design, the second section uses half-length anchoring rockbolts, the third section uses full-length anchoring rockbolts, the fourth section uses prestressed full-length anchoring rockbolts. The rockbolts and anchoring resins are used in underground working site are shown in Fig. 9.

Fig. 9

Rockbolt and anchoring resin used in underground coal mine roadway. (a) Rockbolt; (b) anchoring resin.

The displacement gauges were installed in the roadway to measure convergence value between roadway roof and floor. As shown in Fig. 9, with the original support scheme, the roadway roof sank faster within 5 days and stabilised after 30 days of rockbolt installation, with the roadway roof and floor converging to a deformation of over 500 mm. With the use of half-length and full-length anchoring rockbolts, the roadway deformation was 230 mm and 180 mm. With the use of prestressed full-length anchors, the roadway deformation was close to 110 mm, only 1/5 of the original support scheme. Deformation control of the surrounding rock is relatively satisfactory. The project effectively demonstrates that the self-drilling prestressed full-length rockbolt has a significant effect on roadway deformation control.

5 Conclusion

(1) From the rockbolt pull-out orthogonal test, it can be concluded that the destruction of the bonding interface between the anchoring resin and the test specimen is the main cause of the failure of the anchored structure. From deformation to failure, anchor solids go through elastic deformation stage, elasto-plastic deformation stage, plastic deformation extension stage, initiation slip stage and debonding slip stage in succession.

Fig. 10

Roof deformation measurement result of underground coal mine roadway.

(2) The rockbolt peak pull-out force prediction model was established based on improved wavelet neural networks and the prediction model was verified by experimental test results. The peak pull-out forces under various conditions were predicted and the influence factors of peak pull-out force was studied based on improved Wavelet Neural Network model. The factors that contribute to the ultimate pull-out force of the rockbolt are, in descending order of influence, test specimen strength, prestressing force, anchor thickness and anchor length. The strength of the specimen has a much higher influence on the ultimate pull-out force of rockbolt than the other factors.

(3) Self-drilling pre-stressed full-length anchoring rockbolt have been developed to improve the strength of the rock by applying pre-stress to a high stress level in to produce an effective compressive stress zone in the rock, thus improving the bearing capacity of the surrounding rocks.

(4) Engineering practice shows that for deep-buried soft rock coal mines roadways, pre-stressed full-length anchor support must be used to effectively control the crushing and swelling deformation of the surrounding rock.

Footnotes

Acknowledgments

This research was funded by Fund of State Key Laboratory of Mining Response and Disaster Prevention and Control in Deep Coal Mines (SKLMRDPC21KF13 and SKLMRDPC20ZZ04), Natural Science Foundation of Anhui Province, China (2208085ME118), Natural Science Foundation of Education Department of Anhui Province (KJ2021ZD0049).

References

Shunjie

, Guangming

, Xiangrui

, Xiang

, Wensong

, Gang

and Shikui

, Study of Prevention and Control Technology for Roadway Excavation under the Soft and Extra-Thick Coal Roof in Luling Coal Mine, Processes (9) (2022). doi: 10.3390/PR10091835.

Shuisheng

, Leilei

, Jin

, Yawei

and Honghao

, Study on Bond Defect Detection in Grouted Rock Bolt Systems under Pullout Loads, Advances in Materials Science and Engineering (2022). doi: 10.1155/2022/3282211.

Minjie

, Guangming

, Wensong

, Xiang

, Chongyan

, Zhixi

and Xukun

, Influence of Height–Diameter Ratio on Rock Compressive Failure Characteristics and Damage Evolution Law, Energies (15) (2022). doi: 10.3390/EN15155557.

Hongpu

, Jianzhong

, Guiyang

, Fuqiang

, Xiaoqing

and Jinghe

, Mechanisms of rockbolt support for highly fractured rock masses–insight from physical and numerical modeling, Bulletin of Engineering Geology and the Environment (5) (2022). doi: 10.1007/S10064-022-02696-X.

Shuisheng

, Wancheng

and Leilei

, Experimental and numerical evaluation on debonding of fully grouted rockbolt under pull-out loading, International Journal of Coal Science & Technology (1) (2022). doi: 10.1007/S40789-022-00482-4.

Høien

A.H.

, Li

C.C.

and Zhang

, Pull-out and Critical Embedment Length of Grouted Rebar Rock Bolts- Mechanisms When Approaching and Reaching the Ultimate Load, Rock Mechanics and Rock Engineering(prepublish). (2021). doi: 10.1007/S00603-020-02318-6.

Hao

, Wu

, Ranjith

P.G.

, Zhang

, Hao

and Teng

, A novel energy-absorbing rock bolt with high constant working resistance and long elongation: Principle and static pull-out test, Construction and Building Materials(C). (2020). doi: 10.1016/j.conbuildmat.2020.118231.

Aghchai

M.H.

, Moarefvand

and Rad

H.S.

, Rock Bolt Pull Tests Performance in an Underground Powerhouse Complex: A Case Study in Sri Lanka, Geotechnical and Geological Engineering: An International Journal (2) (2020). doi: 10.1007/s10706-019-01159-7.

Vlachopoulos

, Cruz

, Anthony Tatone

B.S.

and Carrapatoso

, The Performance of Axially Loaded, Fully Grouted Rock Bolts Based on Pull-Out Experiments Utilizing Fiber Optics Technology and Associated Numerical Modelling of Such Support Elements, Geotechnical and Geological Engineering: An International Journal (7) (2020). doi: 10.1007/s10706-019-01098-3.

10.

Hao

, Wu

, Li

, Tao

, Teng

and Hao

, Non-destructive inspection on anchorage defect of hollow grouted rock bolt using wavelet transform analysis, EURASIP Journal on Image and Video Processing (1) (2018). doi: 10.1186/s13640-018-0385-4.

11.

Bastami

, Shahriar

and Ghadimi

, Verification of the Analytical Model for Fully Grouted Rock Bolts Based on Pull-out Test (Case Study: Tabas Coal Mine), Procedia Engineering (2017). doi: 10.1016/j.proeng.2017.05.280.

12.

Thenevin

, Blanco-Martín

, Hadj-Hassen

and Wrana.

, Laboratory pull-out tests on fully grouted rock bolts and cable bolts: Results and lessons learned, Journal of Rock Mechanics and Geotechnical Engineering (5) (2017). doi: 10.1016/j.jrmge.2017.04.005.

13.

Charlie Li

, Kristjansson

and Høien

A.H.

, Critical embedment length and bond strength of fully encapsulated rebar rockbolts, Tunnelling and Underground Space Technology incorporating Trenchless Technology Research. (2016). doi: 10.1016/j.tust.2016.06.007.

14.

Najafzadeh

and Mohammad

, Azamathulla, Group method of data handling to predict scour depth around bridge piers, Neural Computing and Applications (7-8) (2013). doi: 10.1007/s00521-012-1160-6.

15.

Najafzadeh

and Azamathulla

H.M.

, Neuro-Fuzzy GMDH to Predict the Scour Pile Groups due to Waves, Journal of computing in civil engineering (5) (2015).

16.

Saberi-Movahed

, Najafzadeh

and Mehrpooya

, Receiving More Accurate Predictions for Longitudinal Dispersion Coefficients in Water Pipelines: Training Group Method of Data Handling Using Extreme Learning Machine Conceptions, Water Resources Management (2) (2020). doi: 10.1007/s11269-019-02463-w.

17.

Yao

, Zhang

, Wang

, Li

and Wei

, Fuzzy Random Prediction Model of Frost Heave Characteristics of Horizontal Frozen Metro Contact Channel in Coastal Area, Geofluids. (2022). doi: 10.1155/2022/6198660.

18.

Zou

, Wang

, Luo

and Hu

, Intelligent Control of Smooth Blasting Quality in Rock Tunnels Using BP-ANN, ENN, and ANFIS, Geofluids (3) (2021), 1–24.

19.

Hua

, Sirong

, Shilin

and Pin

, Predictive Model for Creep Behavior of Composite Materials Using Gene Expression Programming, Applied Composite Materials (3) (2023). doi: 10.1007/S10443-023-10109-9.

20.

nana Asare

, Affam

and Ziggah

Y.Y.

, A hybrid intelligent prediction model of autoencoder neural network and multivariate adaptive regression spline for uniaxial compressive strength of rocks, Modeling Earth Systems and Environment (2023). https://doi.org/10.1007/s40808-023-01717-2

21.

Armaghani

D.J.

, Asteris

P.G.

, Askarian

, Hasanipanah

, Tarinejad

and Huynh

V.V.

, Examining Hybrid and Single SVM Models with Different Kernels to Predict Rock Brittleness, Sustainability (6) (2020). doi: 10.3390/su12062229.

22.

Chengxiang

, Tianyu

, Mengke

, Ye

, Xiaobo

, Yijing

and Xinyue

, Reliability Prediction of the Distribution Network Based on Wavelet Neural Network with Quantum Particle Swarm Optimization Algorithm, Electric Power Components and Systems (4) (2023). doi: 10.1080/15325008.2023.2173828.