A method of regulating the wind field to reduce safety hazards of gas and dust in fully mechanized heading face based on dynamic data

Abstract

Accumulation of gas and dust in excavation work is a safety risk that remains to be solved. A substantial amount of gas and dust accumulate in fully mechanized heading face outlets, and the press-in tube jet cannot be adjusted dynamically based on the actual physical conditions. To solve the problem, a dynamic data-driven method for optimising gas and dust distribution by regulating the wind field is proposed. This method is based on the immune genetic algorithm and uses dynamic data as the incremental data with the historical data from the mine regulation rules to obtain the optimal incremental regulation rules (IRRs) for the wind field. The experiment was performed in Ningtiaota coal mine, Yulin city, northern Shaanxi, China. The observed wind velocity and gas concentrations were within the specification range when regulated by IRRs. The dust concentrations at the position of the driver and the average concentrations on the return air side decreased by 64.6 % and 56.5 %, respectively, when the outlet was 5 m from the head-on. When the distance of the outlet from the head-on was 10 m, the dust concentrations at the position of the driver and the average concentrations on the return air side dropped by 42.9 % and 68.6 %, respectively. The results from this study provide measures that would improve safety and efficiency during excavation of fully mechanized heading faces.

Fully mechanized heading face; Gas and dust measurements; Immune genetic algorithm; Incremental regulation rules for wind fields.

1. Introduction

Coal is a primary source of energy and plays an essential role in industrial production (Chen et al., 2022; Song et al., 2021); however, the dangerous nature of coal mining, which has resulted in many casualties, has long been a problem that needs to be addressed (Yin et al., 2017). A fully mechanized heading face is the main dust source, accounting for about 60% of the total amount of coal dust generated in mines (Zhang et al., 2022). An increase in tunnelling distance and speed associated with a fully mechanized mining face sharply increases gas and dust generation, which increases the likelihood of coal mine accidents (Lin et al., 2021; Seaman et al., 2020; Zhang et al., 2020a) and endangers the safety of miners (Liu and Liu, 2020; Paluchamy et al., 2021). Researchers worldwide found that about 12,489 miners died of 104 coal dust explosion accidents from 1900 to 2020. Asia had the highest number of deaths, followed by Europe, North America, Africa, Australia and South America (Khan et al., 2022). The proportion of gas accident deaths was between 20 % and 30 % from 2010 to 2020 according to the coal mine accidents statistics of China (You et al., 2021). According to the information released by National Health Commission, in 2020, 14,367 new cases of occupational pneumoconiosis were reported in China, accounting for 84.2 % of the total number of new reported occupational diseases, of which coal workers' pneumoconiosis patients accounted for more than 50 % (Peng et al., 2022). The prevalence of Coal Workers’ Pneumoconiosis (CWP) in miners in the United States has consistently increased from 2000 to 2017 and exceeds 10 % in workers with more than 25 years of exposure (Gonzalez et al., 2022; Song et al., 2022). Press-in ventilation is an essential component of safe tunnelling because it prevents gas explosions and reduces dust exposure to miners, thereby providing a more comfortable working environment. However, owing to a static air volume and ventilation tube position, the airflow tends to form vortices in the roadway, resulting in the accumulation of gas and dust. Therefore, the existing complete ventilation volume control mode must be modified and refined based on actual excavation requirements.

Scholars have studied the effects of the press-in ventilation parameters and tube placement on the distribution of gas and dust in fully mechanized heading faces. Zhou studied the wind velocity distribution with different layout parameters and observed that the wind velocity at the outlet and the roadheader size substantially influenced the wind velocity field (Zhou et al., 2013). Parra evaluated the influence of the location of the ventilation tubes on the gas field under different ventilation models (Parra et al., 2005). Zhou assessed the effect of varying inlet volumes on the dust diffusion law (Zhou et al., 2022). Gong et al. studied the effects of three parameters, namely diameter, right angle, and upward angle of the outlet, on the wind velocity and gas and dust concentrations. The optimal combination of these three parameters under different working conditions was used as the wind field regulation rules (Gong et al., 2019a). Furthermore, the wind field regulation device changed the airflow state at the outlet, which effectively diminished the gas and discharged the dust.

At present, wind field regulation rules are calculated using static mining data under different working conditions and do not consider the dynamic characteristics of the data (Gong et al., 2019b, 2021). The static method not only misses the first regulation time but also the effect of the regulation cannot meet the changing needs. Scholars have argued that new rules should be established for each additional dataset, and the models should be updated through incremental knowledge acquisition, rather than through repeated training, with an improved model obtained each time (Martínez-Plumed et al., 2015; Zhang et al., 2016). Therefore, to meet the dynamic ventilation requirements in excavation work, the regulation rules must be updated incrementally based on the real-time changes during the excavation process, including wind velocity and gas and dust concentrations (Yi et al., 2004; Zhu et al., 2003).

Among the various methods for acquiring incremental association rules, the immune genetic algorithm (IGA) (Wang et al., 1999), which imitates the immune system, was established and then verified through application in practical engineering problems. Zhou used IGA to design the optimal path acquisition method in uncertain environments, which improved the target searching efficiency of UAVs (Zhou et al., 2020). Tao used an adaptive IGA to improve the search ability and maintain population diversity, verifying that the IGA performance is superior to other algorithms (Tao et al., 2012).

A method based on IGA to obtain wind field incremental regulation rules (IRRs) is proposed to improve the updating efficiency of regulation rules and the accuracy of the results. The regulation rules under current working conditions were used as incremental antibodies, and the initial regulation rules under the same working conditions of the previous stage were taken as memory antibodies. The objective function was designed based on the dust concentrations at the position of the driver and along the return side as the antigen. The specified wind velocity range and gas concentrations were the constraint conditions. The memory antibodies in the memory library were then fused with the incremental mining data.

2. Acquisition method of the IRRs based on IGA

2.1. Condition and decision attributes of wind field regulation

The press-in tube was suspended near the roof and the side wall. The jet ejected from the outlet and returned after arriving at the head-on (Zhang et al., 2020b). The airflow was affected by obstacles in the return process, such as the roadheader, resulting in increased wind velocity in some corners and shelter areas were reduced, which generated eddy currents and accumulation of gas and dust (Figure 1).

Figure 1.

Schematic of the airflow distribution in a fully mechanized heading face.

The study demonstrated that the change of airflow state at the outlet significantly affected the airflow distribution (Gong et al., 2017). Therefore, the device was installed at the outlet, as shown in Figure 2, and the outlet parameters were adjusted based on the regulation rules to optimise the airflow, thereby optimising the distribution of gas and dust.

Figure 2.

Representation of the wind field regulating device in a fully mechanized heading face.

By analysing numerous roadways, the typical values of the initial conditions, including the digging method, roadway geometry, and tube location, that shape the wind field and remain unchanged, were selected as the initial condition attributes.

The analysis revealed that the influence on wind velocity and gas and dust distribution was greater when the parameters, namely the distance between the outlet and the head-on ($a_1$/(m)), calibre ($a_2$/(m)), horizontal right angle ($a_3$/(°)), and vertical deflection angle ($a_4$/(°)), were changed. These parameters were therefore used as attributes for regulating the wind field.

The safety rules implemented at the Ningtiaota coal mine stipulate that the wind velocity range of the fully mechanized heading face is 0.25–4 m/s, and when the gas concentrations exceed 1 %, work must immediately cease. Thus, the average wind velocity ($v_1$/(m/s)) on the return air side and at the position of the driver ($v_2$/(m/s)) and the gas concentrations in the corner zone ($c$ (%)) were taken as constraint conditions. The average dust concentrations ($d_1$/(mg/m³)) on the return air side and at the position of the driver ($d_2$/(mg/m³)) were taken as decision-making attributes.

2.2. Steps of IGA for obtaining the IRRs

The immune system was mimicked by IGA whereby the individuals in the population were used as antibodies and the target function as antigens. An information entropy approach was used to determine the degree of affinity between individuals. Individuals accelerated or were inhibited by their fitness and concentration, and the most superior individuals were added to the memory library. The newborn population was generated through genetic manipulation of the previous generation (Figure 3) (Zhu et al., 2009).

Figure 3.

IGA process for the acquisition of wind field IRRs.

Based on the IGA, the essential elements involved in acquiring IRRs for wind fields are as follows.

• (1)Antigen: under the premise that both wind velocity and gas concentrations are within the specified range, the dust concentration on the return air side (d₁) and at the position of the driver (d₂), should be reduced so that the weights of the two are the same. Therefore, the following objective function was designed as the antigen:

$$F\left(x\right)=\frac{d_1-d_{1_{\min }}}{2\left(d_{1_{\max }}-d_{1_{\min }}\right)}+\frac{d_2-d_{2_{\min }}}{2\left(d_{2_{\max }}-d_{2_{\min }}\right)}$$ (1)

• (2)Fitness function: evaluates the individual of the population. The greater the fitness, the more superior the individual, and the lower the dust concentration. The fitness function is as follows:

$$fit\left(x\right)=\frac{1}{F\left(x\right)}$$ (2)

• (3)Memory library: stores the superior individuals from the initial population. The immune system can respond quickly when incremental data is added to improve computational efficiency. In this study, the top 10 individuals in the current population in terms of fitness were added to the memory library.
• (4)Memory library updating: the expected value of individuals was calculated based on the population concentration and the affinity of the antigens and antibodies, and those with low values were inhibited. The next generation population was obtained by single-point crossover, uniform mutation, and other genetic operations, and the fitness of every individual was calculated. The 10 individuals in the new population with the most superior fitness were selected to calculate their affinity to the individuals in the previous generation memory library. The most compatible individuals in the top 10 of the new population fitness group replaced those in the previous generation memory library. The memory library updating process is displayed in Figure 4.
• (5)Affinity: refers to the matching degree between antibodies and antigens and the similarity between antibodies. In this paper, the affinity was measured using information entropy. Suppose the antibody has M genes, and the size of the character set used at each locus is S, and since the sample data is binary encoded, S = 2, the information entropy of the antibody is as follows (Yan et al., 2016).

$$H\left(\eta \right)=\frac{1}{M}\sum _{j=1}^M{H}_j\left(\eta \right)$$ (3)

where $H_j\left(\eta \right)=\sum _{i=1}^S-p_{ij}\lg p_{ij}$. H_j(η)j is the information entropy of the jth gene on the antibody and p_ij is the probability of the ith symbol appearing on the jth locus in the character set.

Figure 4.

The updating process of IRRs memory library.

The affinity between antibodies v and w is as follows:

$$A_{vw}=\frac{1}{1+H\left(2\right)}$$ (4)

where the range of A_vw is (0, 1), H(2) refers to the information entropy between the antibodies v and w and when H(2) = 0, the genes of v and w are the same.

Using the support and confidence in association rules to calculate the affinity between the antigen and the antibody. The degree of affinity between antigen Ag and antibody v is as follows:

$$a{x}_v=W_s\times \frac{\sup port\left(X\Rightarrow Y\right)}{{\sup }_{\min }}+W_c\times \frac{confidence\left(X\Rightarrow Y\right)}{con{f}_{\min }}$$ (5)

In the formula, W_s = 0.5, W_c = 0.5, Sup_min = 0.1, conf_min = 0.8.

• (7)Antibody concentration: describing the diversity of the population. For any antibody in the population, the concentration is as follows:

$$C_v=\frac{1}{n}\sum _{w=1}^{n}a{c}_{vw}，a{c}_{vw}=\{\begin{array}{l}0，A_{vw}<0.9\\ 1，A_{vw}>0.9\end{array}$$ (6)

• (8)Expectation: the expected value of antibody v is defined as:

$$S_v=\frac{a{x}_v}{C_v}$$ (7)

The probability of being selected is higher when the affinity of the antibody to the antigen is higher, and the concentration is lower. The high-affinity antibody is dominant, and the high concentration antibody is inhibited. The algorithm operates as follows:

The steps of obtaining the IRRs based on IGA are as follows:

Step 1: Obtain the initial data (Initial_Data) and incremental data (Add_Data).

Step 2: Pre-process data coding, whereby the sample data is binary coded based on different parameter segmentation points and then duplicate binary coding schemes are deleted.

Step 3: Set the size of antibody population size and the number of iterations (MAXGEN = 10), where the cross probability (pcross) is 0.7, probability of mutation (pmutation) is 0.01, and maximum (T_max) is the number of iterations when the condition is terminated.

Step 4: Identify the antigen: where the aim is to reduce the dust concentrations andunder the specified range conditions for wind velocity and gas concentrations.

Step 5: The initial population comprises antibodies obtained from the initial sample data. Upon adding the increment data, a new population is formed by the increment data and the antibodies from the initial population.

Step 6: Calculate the fitness, support, and confidence of each antibody in the new population.

Step 7: Update the memory library whereby the 10 most superior adaptive antibodies are added to the memory library, and the affinity of all antibodies is calculated. The antibodies with the highest affinity to these 10 antibodies are then deleted and replaced in the memory library.

Step 8: Promote and inhibit the antibody by calculating the expected value of each antibody in the population.

Step 9: The next antibody generation is produced using single-point crossover and uniform variation of the antibodies in the memory library.

Step 10: If the maximum number of iterations is reached, the encoding is decoded as a wind field regulation rule, and the antibody with the highest fitness is added to the regulation rule library Rules, and the algorithm operation ceases. If the maximum number of operations is not reached, then t = t+1, and the process is repeated from Step 6.

3. Acquisition of the regulation data

3.1. Numerical simulation for the initial sample data acquisition

Previously static mining data was obtained by numerical simulation. To ensure consistent data sources, the initial sample data under different working conditions was obtained via numerical simulation and included wind velocity and dust concentrations at the position of the driver and the return air side and the gas concentrations in the corners near the head-on.

The S1202 of Ningtiaota coal mine in Shaanxi Province, China, was the experiment object, and the height and width of the mine were 3.85 m and 5.9 m, respectively. The roadheader model was an EBZ-200 cantilever. The diameter of the press-in ventilation tube was 1 m, and the centre line was 3.05 m from the floor and 0.75 m from the left side wall. The measured wind velocity at the outlet of the tube was 7.9 m/s. The numerical model was established by taking the working condition example where the outlet is 5 m away from the head-on (Figure 5(a)). The x-axis, y-axis, z-axis indicate the directions from the tube side to the return side, from the floor to the roof, and from the head-on to the outflow, respectively. A mesh module was used to mesh the set model in Fluent, where a tetrahedral structured mesh was selected (Figure 5(b)).

Figure 5.

Geometric model and mesh generation for the numerical simulation of the press-in ventilation in the roadway.

The mesh model was imported into Fluent, and the boundary conditions were set. The inlet type was the inlet velocity, the end of the roadway was the free outflow, and the fluid was incompressible. Because gas and dust are mainly generated in the cutting process, the source was set at the head-on. Gas was selected as a gas component coupled with wind and dust simulation. The key parameters are presented in Table 1.

Table 1.

Setting of numerical simulation parameters.

Parameter type	Condition	Define
Parameters of calculation models	Solver	Steady
		Absolute Velocity
		Gravity（y = -9.81 m/s2）
	Viscous Model	Realizable k-ε
	Near-wall Treatment	Standard Wall Functions
Parameters of boundary condition	Turbulence Intensity	3.24%
	Hydraulic Diameter	1.0m
	Wall Shear Condition	No Slip
Parameters of discrete phase model	Interaction with Continuous Phase	On
	Saffman	On
	Diameter Distribution	Rosin-Rammler
	Min. Diameter	1e-06m
	Max. Diameter	1e-04m
	Mid. Diameter	6.03e-05m
	Spread Parameter	1.62
	Total Flow Rate	0.0047 kg/s
	Turbulent Dispersion	DRW model (Random orbit model)
	DPM Boundary Cond. Type	Floor as trap type, wall as reflection type

A grid independence test was performed to ensure the accuracy of the simulation results. The number of grid divisions of the original roadway model was 200,000. The control size of the grid division was adjusted twice based on the original roadway model, resulting in grid numbers of 1.22 million and 1.42 million. Three simulations were performed, Fig. 6(a) and (b) show the results of the grid independence test when the wind speed and dust concentrations in the respiration belt on the return side were used as the reference, respectively.

Figure 6.

Grid independence test.

The results show that the distribution of wind speed and dust concentrations of the different grids is consistent, and the distribution curve is almost coincident, demonstrating that the simulation results with these three grid numbers verified the model accuracy. Therefore, to reduce the number of numerical simulations and improve the calculation speed, 10.2 million grids were used before adjustment in the numerical simulation to ensure calculation accuracy.

The model accuracy was verified by comparing the physical measurements and simulation results of the wind velocity and dust concentrations under the same working conditions. The height of the pedestrian breathing zone on the return side (X = 5 m, Y = 1.5 m) was sampled to measure the wind velocity and dust concentrations to ultimately reduce the harm to miners caused by dust. Comparison of wind velocity and dust concentration obtained by the physical measurements and simulation results are showed in Fig. 7(a) and (b), respectively.

Figure 7.

Comparison of wind velocity and dust concentrations on the return side obtained by numerical simulation and physical measurements.

Because the primary underground gas monitoring positions are the three dead-angle areas close to the head-on, and the stability of the flow field in the first 5 m of the roadway is poor, the flow field in the roadway is static. Gas concentration data were collected in 10 s intervals at each measuring point near the head-on, namely, the lower right corner, the upper right corner, and the lower left corner. The average values were calculated and compared with the gas concentrations at the corresponding point of the numerical simulation to verify the accuracy of the simulated data (Table 2).

Table 2.

Comparison of gas concentration in the corner zones.

Measuring point	#1	#2	#3
Simulated concentration	0.81	0.43	0.33
Measured concentration	0.77	0.42	0.31
Relative error	5.19%	2.38%	6.45%

The maximum relative error of wind velocity was 9.92 % (Z = 10 m), the minimum error was 4.08 % (Z = 15 m), and the average error was 8.27 %. The maximum dust concentration relative error was 9.46 % (Z = 20 m), the minimum error was 0.57 % (Z = 7 m), and the average error was 4.76 %. The error between the simulated data and the measured data of the three monitoring points less than 10 %. In summary, a relative error of less than 10 % was consistently observed, which demonstrated that the simulation data were consistent with the actual physical situation, which verified the reliability and accuracy of the simulated sample data source.

3.2. Physical incremental sample data acquisition

Obtaining and processing data is challenging owing to the complexity of the underground environment. The wind velocity and gas and dust concentration monitoring were based on the similarity principle (Adams et al., 2022), whereby a 1:5 scale platform was designed and built according to the S1202 (Figure 8). Incremental sample data were obtained via physical measurements. The central line of the tube was 0.61 m and 0.15 m from the floor and sidewall, respectively, and the tube diameter was 0.2 m. The computer-controlled measurement device was installed at the end of the tube. The end face was equipped with a gas and dust generating device. The platform could synchronously simulate the working environment of the S1202 to obtain incremental sample data.

Figure 8.

Physical experiment platform for wind velocity and gas and dust concentration monitoring.

Sensors for real-time monitoring and collection of wind velocity and dust and gas concentrations were installed. The layout of the measuring points is presented in Figure 9. The sensors were installed at points 1–9, where the wind velocity and dust concentration on the return air side were captured at points 1–8. Point 9 was the wind velocity and dust concentration monitoring point at the position of the driver. The gas sensors were installed at points 10–12.

Figure 9.

Layout of wind velocity and gas and dust concentration sensors.

The wind velocity and dust concentrations under the same working condition were obtained on the platform and compared with the measured data. Because there were errors in the underground test equipment and the platform, the average value at each point was measured multiple times to minimise the error. Comparison of wind velocity and dust concentration on the return side obtained by the physical experiment and underground measurements are showed in Fig. 10(a) and (b), respectively.

Figure 10.

Comparison of wind velocity and dust concentration on the return side obtained by the physical experiment and underground measurements.

The gas concentrations in the three dead-angle areas near the head-on are compared in Table 3.

Table 3.

Comparison of gas concentration in the corner zones.

Measuring point	#10	#11	#12
Simulated concentration	0.71	0.39	0.30
Measured concentration	0.77	0.42	0.31
Relative error	-7.79%	-7.14%	-3.23%

The wind velocity maximum relative error, minimum error, and average error were 9.59 % (Z = 7 m), 2.98 % (Z = 10 m), and 5.61%, respectively. The dust concentration maximum relative error, minimum error, and average error were 8.27 % (Z = 25 m), 0.65 % (Z = 20 m), and 4.10 %, respectively. The error between the simulated data and the measured data from the three monitoring points was less than 10 %, and the relative error was also less than 10 %. Thus, the underground measured data can be replaced by the data obtained from the platform as the incremental sample data.

3.3. Sample data acquisition

According to the investigation of S1202 and the mining safety requirements, the distance between the outlet and the head-on should be controlled between 5 m and 10 m. To avoid overstatement, the method and process of obtaining the IRRs were demonstrated using the 5 m condition as an example. The $a_2$ values were 1.1m and 1.2m, the optimised range of $a_3$ was 0°–25°, and observing values at 5° increments resulted in values of $a_4$ being 0°, 3°, and 6°. The simulation experimental schemes were designed based on the orthogonal experiment method (Table 4). The wind velocity and gas and dust concentrations were used as the initial sample data (Table 5).

Table 4.

Numerical simulation schemes when the outlet is 5m away from the head-on.

Schemes	a2/（m）	a3/（°）	a4/（°）
1	1.1	0	0
2	1.1	5	0
3	1.1	10	0
…
36	1.2	25	6

Table 5.

Initial sample data after preprocessing.

Schemes	Wind velocity (m/s)		Gas concentration of the corner area (%)	Dust concentration (mg/m3)
	Driver's position	Return side		Driver's position	Return side
1	0.54	0.59	0.46	114.37	176.75
2	0.49	0.61	0.39	144.53	156.78
3	0.53	0.77	0.38	76.83	121.94
…
36	0.40	1.33	0.32	85.17	97.92

Because the platform comprises 1/5 of the actual tunnel, the length was reduced to 1/5 of the actual measurement, and the angle remained constant based on geometric similarity conditions (Adams et al., 2022). The experimental schemes are presented in Table 6. The wind velocity and gas and dust concentrations were used as the incremental sample data (Table 7).

Table 6.

The physical experiment schemes when the outlet is 5m away from the head-on.

Schemes	a2/（m）	a3/（°）	a4/（°）
1	0.22	0	0
2	0.22	5	0
3	0.22	10	0
…
36	0.24	25	6

Table 7.

Incremental sample data after preprocessing.

Schemes	Wind velocity (m/s)		Gas concentration of the corner area (%)	Dust concentration (mg/m3)
	Driver's position	Return side		Driver's position	Driver's position
1	0.52	0.53	0.38	72.35	119.72
2	0.57	0.58	0.31	86.71	95.75
3	0.48	0.62	0.29	71.16	86.37
…
36	0.38	0.43	0.39	79.25	93.22

4. Acquisition of IRRs for the S1202 wind field

4.1. IRR memory library

The data were encoded by four-segment binary encoding. The outlet and wind velocity parameters were determined using the equipartition method (Wang and Zhang, 2014). The k-means algorithm determined the gas and dust concentrations (Yang and Zhao, 2019). The first segment was the genome of the dividing points of regulation parameters for the wind field, and the second section determined the range of parameters at these points. The third section was the genome of the segmentation points of decision attributes, and the fourth section determined the parameter range at the segmentation points of decision attributes. Finally, the regulation parameters were divided using the equal division method according to the reasonable range (Table 8).

Table 8.

Break points for parameters of the initial sample data.

Attributes	Break points
Caliber (m)	0.8, 0.9, 1.0, 1.1, 1.2
Horizontal right angle (°)	0, 5, 10, 15, 20
Vertical upward deflection angle (°)	0, 3, 6
Wind velocity at driver’s position (m/s)	0.40, 0.70, 1.00
Wind velocity at return side (m/s)	0.40, 0.70, 1.00
Gas concentration in corner area (%)	0.35, 0.42
Dust concentration in driver’s position (mg/m3)	86, 92, 106, 126
Dust concentration at return side (mg/m3)	102, 131, 164, 175

An initial binary decision information table was developed using binary coding and reduction of the initial sample data (Table 9).

Table 9.

Initial decision information table when the outlet is 5m from head-on.

Schemes	Coding of control scheme	Coding of wind velocity, gas and dust concentration
1	00110100000100 00100000000000	1101100100110001 1001000100100001
2	00110110000100 00100100000000	1101101100010011 1001001000010010
3	00110011000100 00100010000000	1100111110001100 1000101000001000
...
36	00011000011011 00010000010010	1000011011001000 0000010010000000

The initial decision information table represented the initial population (initialpop), the capacity of memory library (overbest) was 10, number of iterations (MAXGEN) was 10, cross probability (pcross) was 0.7, and mutation probability (pmutation) was 0.01. The 10 superior fitness antibodies were then added to the initial memory library (Table 10).

Table 10.

Initial memory library when the outlet is 5m from head-on.

Schemes	Coding of control scheme	Coding of wind velocity, gas and dust concentration	Fitness
1	0001100011011011110111100100	11000110100010001000010000000000	34.30
2	0001100001101100010000010010	10000110110010000000010010000000	29.16
3	0011000110010000100001000000	11001111100011001000101000001000	28.88
…
10	0001100011011000010000100100	11000110011011001000010001001000	24.89

4.2. IRRs memory library

The incremental sample data binary encoding comprised two steps. The parameter partition points were determined in the first step, and in the second, a four-segment binary coding was adopted according to the parameter partition points obtained in step one. The split points are presented in Table 11.

Table 11.

Split points for parameters of the incremental sample data.

Attributes	Break Points
Caliber (m)	0.16, 0.18, 0.20, 0.22, 0.24
Angle of horizontal right deviation (°)	0, 5, 10, 15, 20, 25
Angle of vertical upward deflection (°)	0, 3, 6
Wind velocity at driver's position (m/s)	0.40, 0.70, 1.00
Wind velocity on return side (m/s)	0.40, 0.70, 1.00
Gas concentration of corner zone (%)	0.29, 0.37
Dust concentration at driver's side (mg/m)3	72, 77, 90, 98
Dust concentration on return side (mg/m)3	90, 96, 114, 120

An incremental binary decision information table is shown in Table 12.

Table 12.

Incremental decision information table when the outlet is 5m from head-on.

Schemes	Coding of control scheme (Antibody)	Coding of wind velocity, gas and dust concentration (Antigen)
1	1000010000010000000000000000	11011001110000111001000110000010
2	1000011000010000000100000000	11011011011011001001001001001000
3	1000001100010000000010000000	11011010100010001001000000000000
…
36	1000000001110000000000010000	10011001011011000001000101001000

The incremental decision-making information table and initial memory library were used as the initial population, and the memory capacity, iterations, crossover probability, mutation probability, and other parameters were the same as those in the previous section (Section 4.1). The calculated 10 most superior fitness antibodies were added to the incremental memory library (Table 13). The scheme with the highest adaptability was selected, and the wind regulation rules are provided in Table 14.

Table 13.

Incremental memory library when the outlet is 5m from head-on.

Schemes	Coding of regulating schemes	Coding of wind velocity gas and dust concentration	Fitness
1	0011001100010000100010000000	11011010100010001001000000000000	11
2	0011000110010000100001000000	11011010011011001001000001001000	9.34
3	0001100001110000010000010000	10011001011011000001000101001000	9.26
…
10	0001101100010000010010000000	11011011000100111001001000010010	7.44

Table 14.

Decoding of IRR for wind field when the outlet is 5m from head-on.

IRR of wind field		Fitness
Conditional attributes:	1.0m < a2 ≤ 1.1m, 5° < a3 ≤ 10°, a4 ≤ 0°	11
Constraints:	0.4 m/s < v2 ≤ 0.7 m/s, 0.4 m/s < v1 ≤ 0.7 m/s, 0.29% < c ≤ 0.37%
Decision attributes:	90 mg/m3 < d2 ≤ 98 mg/m3, 96 mg/m3< d1 ≤ 114 mg/m3

5. Experiment and verification of the regulation effect

5.1. Schemes and arrangement of measurement points

The wind velocity and gas and dust concentrations before and after regulation were compared to verify the advantage and regulation effect of the above method. The regulation rules obtained from the static mining data method under the conditions of 5 m and 10 m from the outlet to the head-on were as follows: $a_2$ = 1.2m, $a_3$ = 20°, $a_4$ = 3°; $a_2$ = 0.9m, $a_3$ = 15°, and $a_4$ = 3°. Combined with incremental data, IRRs under the two conditions were obtained from the same sample as follows: $a_2$ = 1.1m, $a_3$ = 10°, $a_4$ = 0°; $a_2$ = 0.8 m, $a_3$ = 10°, $a_4$ = 0°.

The device was installed in the S1202. Owing to the complex environment, gas accumulated predominantly in the corners near the head-on. The gas concentration cannot be measured during cutting; thus, the regulation effect of gas concentration was tested by numerical simulation and the experimental platform.

Based on the layout of roadway facilities, the mining activities were primarily located at the position of the driver and on the return air side thus the underground measuring points of wind velocity and dust concentration were designed on this basis, the breathing height of the return side and measuring points of the return side are showed in Figure 11(a) and (b), respectively. Because of safety issues, measuring points were 7 m from the head-on. Points 1 and 2 were 3 m apart, measuring points 2–8 were 5 m apart, and point 9 for monitoring wind velocity and dust concentration was at the position of the driver.

Figure 11.

Underground measuring points of wind velocity and dust concentrations.

The wind velocity and dust concentrations at the position of the driver and on the return air side were measured under different regulation schemes by adjusting the outlet parameters. Each measuring point was measured simultaneously by three CCZ-1000 dust detectors with the same model parameters so that an average dust concentration at the measuring point can be calculated to reduce the error. Figure 12(a) shows the installation of the regulating device in S1202, and Figure 12(b) shows experimenters simultaneously collecting data with three dust detectors.

Figure 12.

Underground wind field regulation experiment.

5.2. Gas field regulation effect based on IRRs

The numerical simulation results of the gas field at the outlet positions 5 m and 10 m away from the head-on are shown in Figures 13 and 14, respectively, and the corresponding gas distribution cloud images at a section 0.3 m from the head-on were obtained and compared at various stages.

Figure 13.

Comparison of the gas distribution clouds before and after regulation at a 5 m distance.

Figure 14.

Comparison of the gas distribution clouds before and after regulation at a 10 m distance.

To enhance the credibility of the results, the gas concentrations before and after regulating the same experimental scheme were compared through the experimental platform. As evident from Table 15, the gas concentrations in the corner areas generally decrease to less than 1 % owing to the wind field effect.

Table 15.

Gas field test verification under different working conditions.

Working conditions(m)	Gas concentration of corner areas（%）
	Before regulation	Regulated by static rule	Regulated by IRR
5m	0.84	0.53	0.32
10m	0.71	0.45	0.28

The results show that the original gas field under the two conditions gathers in the lower right corner near the head-on, and the volume reached 0.84 % and 0.71 %, which is close to the upper limit of the safety regulations and poses a high explosion risk. Through the regulation of static rules, the accumulated gas was blown away after the outlet deflected to the right. However, the deflection angle was small when the outlet was 5 m away from the head-on, and accumulated gas remained in the lower right corner. When the outlet was 10 m away from the head-on, the right deviation angle was too large, and the effective jet length was insufficient, resulting in gas accumulation in the lower left corner near the head-on. After the incremental regulation, the gas concentration in the lower right corner area decreased to 0.32 % by increasing the right deviation angle and narrowing the aperture when the outlet distance was 5 m from the head-on. Compared with the static regulation, the gas concentration decreased 0.21 %. When the outlet was 10 m away from the head-on, the right deviation angle decreased, and the diameter decreased to increase the effective jet length, which not only decreased the gas concentration in the lower right corner to 0.28 % but also eliminated the gas accumulation near the head-on. In summary, the regulation effect based on IRRs is evidently superior to that based on static rules.

5.3. Regulation effect of wind velocity and dust concentration based on IRRs

Underground experiments were conducted on the wind velocity and dust concentration before and after regulation under the two conditions. The measured wind velocity and dust concentration data and the comparison diagram at a 5 m distance are presented in Table 16 and Figure 15, the comparative wind velocity and comparative dust concentrations are showed in Figure 15(a) and (b), respectively. Table 17 and Figure 16 present the measured data at the 10 m distance, the comparative wind velocity and comparative dust concentrations are showed in Figure 16(a) and (b), respectively.

Table 16.

Wind velocity and dust concentration on the return side before and after regulation according to two rules when the outlet is 5m away from the head-on.

Measuring points (m)	Wind velocity on return side (m/s)			Dust concentration on return side (mg/m)3
	Before regulation	Regulated by static rule	Regulated by IRR	Before regulation	Regulated by static rule	Regulated by IRR
7m	1.04	1.63	1.97	277.53	206.05	145.80
10m	1.08	1.57	2.05	248.57	160.48	120.20
15m	0.70	1.22	1.81	213.87	143.45	113.89
20m	0.16	0.62	1.33	182.25	97.36	76.33
25m	0.19	0.48	1.29	144.18	71.53	62.42
30m	0.29	0.41	1.08	129.48	54.93	41.44
35m	0.37	0.37	0.86	96.97	36.47	28.25
40m	0.38	0.36	0.81	87.79	37.25	13.25
Mean Value	0.53	0.83	1.40	172.58	100.94	75.20

Figure 15.

Comparison of wind velocity and dust concentrations on the return air side before and after regulation at a distance of 5 m.

Table 17.

Wind velocity and dust concentration on the return side before and after regulation according to two rules when the outlet is 10m away from the head-on.

Measuring points (m)	Wind velocity on return side (m/s)			Dust concentration on return side (mg/m)3
	Before regulation	Regulated by static rule	Regulated by IRR	Before regulation	Regulated by static rule	Regulated by IRR
7m	1.84	1.97	1.95	276.03	138.77	95.16
10m	1.45	1.92	1.89	239.60	95.16	85.47
15m	0.89	0.67	1.50	148.73	85.47	83.36
20m	0.64	0.47	1.13	141.84	83.36	34.32
25m	0.69	0.61	1.03	87.34	34.32	33.82
30m	0.48	0.38	0.86	81.34	33.82	24.90
35m	0.42	0.49	0.59	101.94	24.90	5.72
40m	0.38	0.57	0.62	132.45	5.72	16.80
Mean Value	0.85	0.89	1.20	151.16	62.69	47.44

Figure 16.

Comparison of wind velocity and dust concentrations on the return air side before and after regulation at a 10 m distance.

The results presented in Table 16 and Figure 15 indicate that before regulation, the wind velocity measured at the 20 m and 25 m points were 0.16 m/s and 0.19 m/s, respectively, which are lower than 0.25 m/s. The air flowed slowly, and the capacity to discharge pollutants was poor. Figure 15(a) indicates that the wind velocity increased to 0.25–4 m/s after regulation according to the two rules. In addition, the increase of average wind velocity after regulation based on the IRR was 2.9 times that after regulation based on the static rule.

The dust concentration at the position of the driver before regulation was 337.24 mg/m³ and decreased to 145.76 mg/m³ (56.8 %) after static regulation and to 119.41 mg/m³ (64.6 %) after incremental regulation. The average dust concentration on the return air side was 172.58 mg/m³. After regulation by the static rules, the average concentration decreased to 100.94 mg/m³ (41.5 %) and after IRR regulation decreased to 75.20 mg/m³ (56.4 %). The latter increased dust fallout by 35.9 %.

The results in Table 17 and Figure 16 indicate that after regulation based on the wind field rules, the overall wind velocity in the roadway increased significantly, and the capacity to discharge pollutants increased. The dust concentration at the position of the driver before the regulation was 321.88 mg/m³ and decreased to 210.98 mg/m³ (34.4 %) after static regulation and to 183.87 mg/m³ (42.9 %) after incremental regulation. The average dust concentration on the return air side was 151.16 mg/m³. After regulation based on the static rule, the average concentration decreased to 62.69 mg/m³ (58.5 %). And after IRR based regulation, the average concentration decreased to 47.44 mg/m³ (68.6 %). The latter increased dust fallout by 17.2 %.

6. Conclusions

• (1)In the tunnelling process, numerous historical data reflect the correlation between the regulating parameters and the dynamic wind velocity and gas and dust concentration data. To effectively utilise these two data types, the distribution of gas and dust should be optimised to reduce the safety risks. In this study, the historical regulation rules obtained from historical mine records were used as memory antibodies, and the current wind regulation rules were used as incremental antibodies. Memory antibodies were incrementally mined using the wind velocity and gas volume dynamic data ranging from 0.25–4.00 m/s for wind velocity and less than 1 % for gas volume. A method of IRRs acquisition based on IGA has been proposed and established in this study, which can be used to update the regulation rules to dynamically regulate the wind field during excavation work, thereby avoiding gas and dust accumulation.
• (2)The underground mining static data regulation rules were applied and verified by incrementally updating S1202 at the Ningtiaota Mine. The results indicate that the wind velocity and gas concentration at the 5 m and 10 m conditions were controlled within the safety standard limits. After regulation based on static rules, the dust concentration at the position of the driver decreased by 56.8 % and 34.5 %, respectively, and along the return side decreased by 41.5% and 58.5%, respectively. Whereas after regulation based on IRRs, the dust concentration at the position of the driver decreased by 64.6 % and 42.9 %, respectively, and along the return side decreased by 56.4 % and 68.6 %, respectively. The amount of dust fallout increased by 35.9 % and 17.2 %, respectively, compared with regulation based on static rules. Therefore, for continuous work, the incremental mining rule method is more suitable for the dynamic regulation of coal mine wind fields than the static data association rule.

Declarations

Author contribution statement

Zhuangzhuang Liu: Conceived and designed the experiments; Analyzed and interpreted the data; Wrote the paper.

Xiaoyan Gong; He Xue: Conceived and designed the experiments; Analyzed and interpreted the data.

Long Chen; Zheng Han: Performed the experiments.

Minjie Wei; Yue Wu: Contributed reagents, materials, analysis tools or data.

Ao Cheng: Contributed reagents, materials, analysis tools or data; performed the experiments.

Yue Wu: Contributed reagents, materials, analysis tools or data.

Zheng Han: Performed the experiments.

Huming Niu: Conceived and designed the experiments.

Hao Li: Performed the experiments.

Funding statement

Xiaoyan Gong was supported by Research on Natural Science Foundation Project of Shaanxi Province [2021JLM-01], National Natural Science Foundation of China [51874235].

Data availability statement

The authors do not have permission to share data.

Declaration of interest's statement

The authors declare no conflict of interest.

Additional information

No additional information is available for this paper.