Arab J Geosci DOI 10.1007/s12517-013-1145-5

ORIGINAL PAPER

Memory cutting of adjacent coal seams based on a hidden Markov model Wei Li & Chengming Luo & Hai Yang & Qigao Fan

Received: 29 May 2013 / Accepted: 26 September 2013 # Saudi Society for Geosciences 2013

Abstract Height adjustment of a shearer cutting drum is one of the key processes involved when the shearer swings its cutting drum up and down on a fully mechanized mining face. Direct sensors are used to recognize the coal–rock interface for adjusting the shearer cutting drum; however, these sensors exhibit poor reliability and accuracy. A traditional memory cutting method is applied to avoid the deficiencies of direct sensors, but this method results in large residual errors and frequent adjustments of the shearer cutting drum. This paper proposes a hidden Markov model (HMM) memory cutting method for the shearer. The height of the shearer cutting drum is modeled by describing the collaborative automation of the shearer, scraper conveyor, and hydraulic supports. After analyzing the principle of traditional memory cutting for the shearer cutting drum, HMM memory cutting is developed by employing data correlation of adjacent coal seams. Moreover, the effectiveness of HMM memory cutting is compared with traditional memory cutting. Results indicate that HMM memory cutting effectively predicts the accurate height of the shearer cutting drum and reduces its adjustment frequency. The HMM memory cutting method tracks the coal–rock interface efficiently and enables the shearer to cut more coal seams and less rocks. Keywords Shearer . Memory cutting . Hidden Markov model . Adjacent coal seam . Data correlation

Introduction Coal is one of the most important energy resources in the world (Benndorf 2013). An upsurge in coal output leads to W. Li : C. Luo (*) : H. Yang : Q. Fan School of Mechatronic Engineering, China University of Mining and Technology, Xuzhou 221116, Jiangsu, China e-mail: [email protected]

irreconcilable challenges between coal productivity and coal safety. Less miners or unmanned mining is the key to achieve high productivity, high efficiency, and safe mining. Automated mining is regarded as the Holy Grail in the international mining industry (Ju and Xu 2013). Fully mechanized mining face is the core segment of automation mining. This form of mining is equipped with large machines, including the shearer, scraper conveyor, and hydraulic supports. Mining equipments of fully mechanized mining face undertake a complicated chain of coal mining activities, including coal breaking, coal charging, and roof supporting (Paul et al. 2012; Wang et al. 2012). In almost all cases, coal mining activities result in various accidents such as potential rock falls, presence of toxic gasses, the poor ventilation, and machinery failure (Singh et al. 2013; He et al. 2012). Many attempts on the automation, information, and integration of coal mine have been made to reduce mine disaster and ensure mine safety (Hosseini et al. 2013; Ashgari and Esfahani 2013). The cooperative movement among the shearer, scraper conveyor, and hydraulic supports is the basis of automation mining. The movement of shearer plays an important role in the cooperative movement. Scholars have invested large efforts to make the shearer cut the optimum path and work at a reasonable velocity, etc. (Ayhan and Eyyuboglu 2006). However, adaptive adjustment of the shearer mode, particularly to make the shearer cutting drum swing up and down is extremely difficult to achieve because of the complicated and volatile environment. Adaptive height adjustment of the shearer cutting drum has not been resolved effectively. The shearer driver always causes the shearer drum to cut the roof rock or the floor rock and mistakenly regards rocks as coal. The most direct solution to this problem is studying coal–rock interface identification, which will make the shearer automatically track the coal–rock interface. Direct sensors have been investigated by numerous authors. Existing techniques for identifying coal–rock interface include the ray probe

Arab J Geosci

(Asfahani and Borsaru 2007), radar detection (Strange et al. 2005), pick stress (Ren et al. 2010), and photoelectric sensing (Ratikanta 2011) methods. However, unsound mining conditions, such as diverse coal characteristics, variable coal mining routes, and unstable sensors, result in numerous deficiencies and limit the applications of the aforementioned methods. The demands for reliability and wide applications make the existing sensors unsatisfactory.

Related works West German scholars firstly proposed a memory cutting method that avoided the difficulties of direct sensors in recognizing coal–rock interface by borrowing ideas from teaching and tracking technologies in robotics (Alford 1985). The memory cutting method has been applied to new pattern shearers of the US JOY Company, Germany Eickhoff Company, and DBT Company. Adaptive height adjustment technology that adopts the memory cutting method has several applications in stable coal seams. Coal seam thickness memorized mismatches with actual coal seam thickness when coal seam changes. At this time, if the shearer still swings up and down its cutting drum in accordance with the memorized coal seam thickness, then the shearer may cut more rocks and less coal. As a result, the shearer driver needs to adjust the height of the shearer cutting drum according to actual coal seam thickness to prevent damage to the cutting drum caused by cutting rocks over a long period. The adjusted height of shearer cutting drum must be memorized periodically to ensure adaptive adjustment in the next cutting cycle. Adaptive height adjusting technology based on memory cutting method addresses the deficiencies of direct sensors in recognizing coal–rock interface. Efforts have been devoted to developing shearer memory cutting methods. Liu et al. (2004) proposed a memory program control mode for automatic reappearing. Once the height of shearer cutting drum was sampled in real time and stored in a computer, an automatic controller adjusts the cylinder displacement of shearer cutting drum. Li et al. (2011) presented an adaptive memory cutting method based on the greyMarkov model. The predicted height data of shearer cutting drum were used to build the state transfer probability matrix of Markov chain to achieve better precision and stability. A selfadaptive memory cutting strategy using fuzzy theory was presented, which can automatically adjust traction speed and cutting drum height as well as judge whether the shearer is cutting rocks (Xu et al. 2011). The aforementioned memory cutting methods are available for the adaptive adjustment of shearer cutting drum. However, none of these methods satisfies the standards of high accuracy. It is because that the existing methods have some disadvantages of analyzing the coal seam thickness characteristics from a global perspective

and ignoring local similarities. Therefore, a thorough investigation of coal seam thickness characteristics among adjacent memory cutting cycles is of great significance in improving the efficiency and accuracy of memory cutting. Several studies on data correlation are found in literature (AlSaud 2008; Roy and Leiva 2012; Nganje et al. 2011). Yuan et al. (2011) divided an entire data trajectory into several trajectory clusters using data structure correlation and obtained structural similarity by analyzing the structure features of every trajectory cluster. Yan et al. (2007) applied data correlation to studying speech recognition on the adjacent speech space. Meanwhile, several scholars used intelligent theory to research data correlation. Hassan et al. (2013) forecasted nonlinear time series data based on hidden Markov model (HMM) and combined it with adaptive fuzzy inference system. Hong et al. (2012) researched speech synthesis system using HMM, focusing particularly on quality degradation of the synthesized speech. Abarghouei et al. (2013) employed Artificial Neural Networks to predict drought and proved that the method was feasible by calculating the correlation coefficient. Direct sensors exhibit poor reliability and accuracy in adjusting the shearer cutting drum. Thus, we focus mainly on indirect methods for the shearer. The concept of teaching and tracking methods has been applied successfully in robotics. We borrow the ideas from this field and focus on the memory cutting method by employing data correlation of adjacent coal seams with the aid of the HMM approach. We then propose a HMM memory cutting method and apply it to a stable coal seam. The results indicate that the HMM memory cutting method can track coal–rock interface efficiently and allows the shearer to cut more coal seams and less rocks.

Fully mechanized mining face Collaborative automation of coal mining machines A fully mechanized mining face is equipped mainly with coal mining machines, including the shearer, scraper conveyor, and hydraulic supports, as shown in Fig. 1. The shearer moves and cuts coal cutting on the scraper conveyor at a certain speed. Then, the coal is delivered to the transshipment point by the scraper conveyor. Next, the hydraulic supports advance the scraper conveyor, support the goaf, and maintain the stability of the surrounding rocks. The parameters such as shearer mobile location, shearer attitude, shearer haulage speed, and hydro cylinder displacement affect the effect of memory cutting. But the height of shearer cutting drum is one of the main parameters. A mathematical model of the height of shearer cutting drum is developed, as shown in Fig. 2. Angle between shearer and horizon line is β, angles between two rock arms and horizon line is γ, length of arm is L,

Arab J Geosci

tracking processes from four or five to two or three cycles. The thickness of a stable coal seam always changes gradually between teaching and tracking processes, whereas the discontinuity geology can lead to a significant change in coal seam thickness. Thus, memory cutting is more applicable to a stable coal seam than to a coal seam with dirt band, coal discontinuity, or rocks. The principle of memory cutting for the shearer is described as follows.

1-Shearer. 2-Scraper conveyor. 3-Hydraulic supports Fig. 1 Collaborative automation of coal mining machines on fully mechanized mining face

and center distance is H b between shearer drum and shearer base. The height of shearer cutting drum H is given by (Li et al. 2011) H ¼ H b þ L  sinðγ − βÞ

ð1Þ

where γ and β can be obtained by tilt sensors. Memory cutting principle of the shearer The mining method of fully mechanized mining face is often summarized as a single directional or a bidirectional demonstration (Snopkowski 2009). This paper primarily adopts the single directional demonstration in investigating memory cutting. As shown in Fig. 3, the x-axis represents the forward direction of working face; the y-axis indicates the propelling direction of working face, and the z-axis denotes the height data of shearer cutting drum. Memory cutting consists of teaching and tracking processes. The height data of the shearer cutting drum are recorded and stored during the teaching process and used in four or five tracking processes. A significant change in coal seam thickness shortens the number of

(1) In teaching process H i , the shearer driver records the height data {Hi 1,Hi 2,…,Hi m} of the shearer cutting drum in advance, according to coal seam thickness. Then, the shearer cutting drum swings up or down according to the recorded height data in the next four or five tracking processes. (2) After four or five tracking processes, the shearer driver manually adjusts the height of shearer cutting drum and refreshes all the height data in response to the changed coal seam thickness. Thus, the new height data {Hi1 + 1, Hi 2 + 1,…,Hi m+ 1} of shearer cutting drum forms new teaching process H i+1 All height data in the new teaching process H i +1 must be newly sampled when coal seam thickness changes based on the description of traditional memory cutting. This process increases the workload of the shearer driver and weakens the effect of memory cutting. The coal seam thickness generally changes continuously because of the distribution characteristics of the coal seam. Variations in the height of shearer cutting drum are consistent between adjacent teaching processes H i and H i +1 based on the smooth variations of coal seam thickness. In adjacent coal seam, the increase or decrease in teaching process H i +1 has the same rate as that in teaching process H i . Thus, predicting the height of shearer cutting drum during the teaching process H i +1 is feasible using height data correlation between the teaching process H i and H i +1. We can use height data in teaching process H i to predict part of the height data in teaching process H i +1 based on the smooth variations of coal seam thickness. As a result, the shearer driver no longer needs to adjust all height data in teaching process H i +1, thus reducing adjustment frequency in adjacent teaching processes and easing the workload of the shearer driver. Therefore, learning how to exploit data correlation fully among adjacent teaching processes and improving the practicality of memory cutting are important.

HMM memory cutting for shearer The HMM principle

Fig. 2 Shearer uplink cutting schematic

HMM is a probability model that indicates random process statistical properties. HMM is a five tuple λ =(S ,O , A ,B ,π ), where S ={s 1,s 2,…,s N } is a finite set of states;

Arab J Geosci Fig. 3 Memory cutting principle of the shearer

O ={o 1,o 2, …,o N } is a finite observation alphabet set; π ={π 1,π 2,…,π N } is the probability distribution of initial states; A =(a i ,j )N * N is a probability distribution on state transitions; B =(b i ,j )N * M is a probability distribution on state symbol emissions (De et al. 2012).

The cutting segment t in the teaching process H i is only related to the cutting segments before and at cutting segment t. Then the following conditions need to be satisfied: condition a:       p H i;t H i;1 ; H i;2 ; …; H i;v ¼ p H i;t H i;1 ; H i;2 ; …; H i;t−1 ð3Þ

HMM memory cutting In this paper, HMM adds some links among the cutting points in new teaching process H i +1, which not only makes the teaching process H i affect the teaching process H i +1 but also −1 makes the height data Hi j+1 affect the height data Hij+1 . Therefore, all height data before the height data Hij+1 in the teaching process H i +1, in collaboration with the corresponding recorded height data in teaching process H i , affect the value of height data Hi j+1 , as shown in Fig. 4. Suppose fully mechanized mining face contains i times teaching process (i =1,2,…,l ); every process H i contains j cutting points (j =1,2,…,m ), and every cutting point j corresponds to height data Hi j ; every teaching process H i can be divided into t cutting segments (t =1,2,…,v); every cutting segment contains n t cutting points (n t =n 1,n 2,…,n v ). So the height data of teaching process H i can be described as

Hi ¼

8 > > > > > > > > > <

H 1i;1 ; H 2i;1 ; …; H ni;11 ; H 1i;2 ; H 2i;2 ; …; H ni;22 ; …; H 1i;v ; H 2i;v ; …; H ni;vv > |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} > > > > n1 n2 nv X X X > > > j j j > tð2Þ¼ H i;2 t ðvÞ¼ H i;v : tð1Þ¼ H i;1 j¼1

j¼1

j¼1

condition b: The cutting point j in cutting segment t is related to cutting points before and at cutting point j.     p H i;tj jH 1i;t ; H 2i;t ; …; H ni;t ¼ p H i;tj jH 1i;t ; H 2i;t ; …; H i;tj−1 ð4Þ Setting the height data {Hi 1,Hi 2,…,Hi m } in the teaching process H i , the absolute value of height variation between every two height data is h i ={hi1,hi 2,…,hi m −1}; setting the height data {Hi1+1,Hi2+1,…,Hi m+1} in the teaching process H i + 1, the absolute value of height variation between every two height data is h i +1 ={hi1+1,hi2+1,…,him+1−1}. So the predicted

9 > > > > > > > > > = > > > > > > > > > ;

ð2Þ

Fig. 4 HMM memory cutting

Arab J Geosci

height of teaching process H i +1 by HMM memory cutting can be expressed as   PðH i ; H iþ1 Þ arg max P H iþ1 jH i ¼ arg max P ðH i Þ H H

The height data of cutting segment t from the teaching process H i and teaching process H i +1 is used to obtain

ð5Þ

      j j j−1 j−1 p H 1iþ1;t ; …; H iþ1;t jH 1i;t ; …; H i;tj ¼ p H iþ1;t jH 1i;t ; …; H i;tj ; H 1iþ1;t ; …; H iþ1;t jH 1i;t ; …; H i;tj p H 1iþ1;t ; …; H iþ1;t

ð6Þ

1 −1 ,…,Hi,tj ,Hi1+1,t ,…,Hi j+1,t ) using the Bayes For p(Hi j+1,t |Hi,t rules, we can get

  j j−1 p H iþ1;t jH 1i;t ; …; H i;tj ; H 1iþ1;t ; …; H iþ1;t     j j j−1 p H i;tj jH 1i;t ; …; H i;tj−1 H 1iþ1;t ; …; H iþ1;t jH 1i;t ; …; H i;tj−1 ; H 1iþ1;t ; …; H iþ1;t p H iþ1;t   ¼ j−1 p H i;tj jH 1i;t ; …; H i;tj−1 ; H 1iþ1;t ; …; H iþ1;t

because all the information of p (Hi ,tj |Hi1,t , …,Hi ,tj− 1 , Hi1+ 1,t , …,Hi j+ 1,t ) comes from teaching process H i , the height variation probability has been integrated into Hi j+ 1 .



j−1 j p H i;tj jH i;tj−1 ; H iþ1;t ; H iþ1;t



So p (Hi j,t |Hi1,t ,…,Hi ,tj− 1,Hi1+ 1,t ,…,Hij+ 1,t ) can be simplified 1 as p (Hi j,t |Hi j,t− 1,Hij+−1,t ,Hij+ 1,t ). Applying this modification, we can acquire

    j j−1 j−1 p H iþ1;t jH i;tj−1 ; H i;tj ; H iþ1;t p H i;tj jH i;tj−1 ; H iþ1;t   ¼ j j−1 p H iþ1;t jH i;tj−1 ; H iþ1;t

Substituting formula (8) into formula (7)   j j−1 jH 1i;t ; …; H i;tj ; H 1iþ1;t ; …; H iþ1;t ¼ p H iþ1;t     j j−1 j−1   p H iþ1;t jH i;tj−1 ; H i;tj ; H iþ1;t p H i;tj jH i;tj−1 ; H iþ1;t j j−1     p H iþ1;t jH 1i;t ; …; H i;tj−1 H 1iþ1;t ; …; H iþ1;t j j−1 j−1 p H iþ1;t jH i;tj−1 ; H iþ1;t p H i;tj jH 1i;t ; …; H i;tj−1 ; H 1iþ1;t ; …; H iþ1;t ð9Þ

but 

j−1 jH 1i;t ; …; H i;tj p H 1iþ1;t ; …; H iþ1;t



  j−1   p H i;tj jH 1i;t ; …; H i;tj−1 ; H 1iþ1;t ; …; H iþ1;t j−1   p H 1i;t ; …; H i;tj−1 ; H 1iþ1;t ; …; H iþ1;t ¼ p H 1i;t ; …; H i;tj ð10Þ

ð7Þ

ð8Þ

Arab J Geosci

substituting formula (9) and formula (10) into formula (6) and doing some arrangement

p

    j j−1 j−1 p H iþ1;t jH i;tj−1 ; H i;tj ; H iþ1;t p H i;tj jH i;tj−1 ; H iþ1;t      ¼ j j−1 p H iþ1;t jH i;tj−1 ; H iþ1;t p H 1i;t ; …; H i;tj     j j−1 j−1 jH 1i;t ; …; H i;tj−1 ; H 1iþ1;t ; …; H iþ1;t  p H 1i;t ; …; H i;tj−1 ; H 1iþ1;t ; …; H iþ1;t p H iþ1;t

  j H 1iþ1;t ; …; H iþ1;t jH 1i;t ; …; H i;tj

1 −1 In formula (11), p(Hi,t …Hi,tj ) and p(Hi,tj |Hi,tj−1,Hij+1,t ) are previous knowledge in the teaching process H i , whose value can’t change in the teaching process H i +1; the



j p H 1iþ1;t …H iþ1;t jH 1i;t …H i;tj



Fig. 5 Algorithm for hidden Markov model memory cutting of the shearer

probability p (Hi1,t … Hi ,tj− 1, Hi1+ 1,t … Hi +j −1,t1 ) doesn’t depend on the height data Hi j+ 1,t . So their values are one.

  j j−1   p H iþ1;t jH i;tj−1 ; H i;tj ; H iþ1;t j j−1   p H iþ1;t jH 1i;t ::H i;tj−1 H 1iþ1;t ::H iþ1;t ¼ j j−1 p H iþ1;t jH i;tj−1 ; H iþ1;t

The absolute value of height variation between every two height data is small, owing to the stable coal seam. Once height data hi1+1,t is given, the absolute value of height variation hi j+1,t can be resolved by using continuous iterative e j can be obtained based solution. Then new height data H iþ1 j on the height data Hi and the height variation h i and h i +1. In order to evaluate the HMM memory cutting, several measures of accuracy are employed to see how far the model is capable of predicting the height of shearer cutting drum. For

ð11Þ

ð12Þ

this reason, the models are evaluated by four parameters containing the residual error of single cutting point, the residual error of every teaching process, the model credibility of every teaching process, and the ±σ confidence interval of single cutting point. The aforementioned parameters are obtained by: Residual error of single cutting point is given by j e j −H j εiþ1 ¼H iþ1 i¼1

ð13Þ

Arab J Geosci

Residual error of every teaching process is given by εiþ1 ¼

m  1 X  e j j  H i¼1 −H iþ1  m j¼1

ð14Þ

Model credibility of every teaching process is given by ρiþ1 ¼ ð1−εiþ1 Þ  100%

ð15Þ

±σ confidence interval of single cutting point is given by  j j H iþ1 ðσÞ ¼ 0:95*H iþ1 1:05*H iþ1 ð16Þ Algorithm for HMM memory cutting of shearer We propose HMM memory cutting for the shearer based on the above mentioned equations and analysis, as shown in Fig. 5. Fig. 7 Residual error for HMM memory cutting

Simulation and analysis To evaluate the accuracy and practicality between traditional and HMM memory cutting, we conduct a simulation based on the height data collected from a virtual coalfield. It is assumed that the simulation is deployed on a medium coal thickness with steady roof and floor. The simulation test has a number of characteristics as follows. The depth of shearer cutting is 1 m, and the length of working face forward direction is 50 m. The number m of sampling points is 50. The error threshold value is 0.1 m and confidence interval σ is ±95 %. Based on the teaching process H 0, we simulate and analyze the teaching processes {H 1,H 2,…,H 6}. The residual errors of teaching processes {H 1,H 2,…,H 6} for traditional and HMM memory cutting are shown in Figs. 6

and 7, respectively. The HMM memory cutting developed in this paper results in a smaller residual error, whereas traditional memory cutting has a larger residual error at most cutting points. The maximum residual errors are 0.53 and 0.43 m for traditional and HMM memory cutting, respectively. Consequently, HMM memory cutting has higher accuracy and practical applications than traditional memory cutting. The results of teaching process H 2 for traditional and HMM memory cutting are depicted in Figs. 8 and 9, respectively. The performance of traditional memory cutting along the forward direction of working face is presented in Fig. 8. The figure shows that traditional memory cutting cannot adjust the height of shearer cutting drum efficiently according to seam thickness. This process is called passive memory cutting. By contrast, HMM memory cutting enables dynamic

initial height traditioanl memory cutting

5.5

height data/m

5

4.5

4

3.5

3

2.5

0

10

20

30

40

50

working face forward direction/m

Fig. 6 Residual error for traditional memory cutting

Fig. 8 The results of teaching process H 2 for traditional memory cutting

Arab J Geosci 5.5

±95 % confidence interval traditional memory cutting

5

height data/m

4.5

4

3.5

3

2.5

0

10

20

30

40

50

working face forward direction/m

Fig. 9 The results of teaching process H 2 for HMM memory cutting

heigh data of H 5/m heigh data of H /m heigh data of H /m 3 1

initial height HMM memory cutting

HMM memory cutting

6

4

2

0

10

20

30

40

50

0

10

20

30

40

50

0

10

20

30

40

50

6

4

2 6

4

2

working face forward direction/m

adjustment of the shearer cutting drum height according to the previous cutting height, as illustrated in Fig. 9. This process is called active memory cutting. Therefore, traditional memory cutting results in a larger residual error when the shearer cuts steep coal seams, whereas HMM memory cutting does not suffer from such restriction. Figure 10 compares the residual errors of teaching process H 2 for traditional and HMM memory cutting. The average residual errors of teaching process H 2 for traditional and HMM memory cutting are 0.17 and 0.05 m. The model credibility of teaching process H 2 is 83 and 95 %, respectively. Moreover, traditional memory cutting has large residual errors at several cutting points, such as the −0.44 m residual error at point H232. By contrast,

0.18

traditional memory cutting HMM memory cutting

0.16

0

0.14

-0.5

0

10

20

30

40

50

HMM memory cutting

0.5

0

average residual error/m

residual error/m

HMM memory cutting results in smaller residual errors at several cutting points, such as the 0.29 m residual error at point H233. The preceding analysis demonstrates that HMM memory cutting exhibits higher accuracy and better reliability than traditional memory cutting. The curve distribution of teaching processes H 1, H 3, and H 5 in the confidence interval for traditional and HMM memory cutting is illustrated in Fig. 11. The black curve represents the results for HMM memory cutting, and the red curve represents the results for traditional memory cutting. The light green area represents the ±95 % confidence interval. Most height data for the traditional method is restricted in the ±95 % confidence interval, whereas a small amount of height data is

traditional memory cutting

0.5

residual error/m

Fig. 11 The ±95 % confidence interval of teaching processes H 1, H 3, and H 5

0.12 0.1 0.08 0.06 0.04 0.02

-0.5

0

10

20

30

40

50

working face forward direction/m Fig. 10 Residual errors of teaching process H 2 for traditional and HMM memory cutting

0

0

1

2

3

4

teaching process/n

Fig. 12 The average residual error

5

6

7

Arab J Geosci 15

adjusting frequency/n

traditional memory cutting HMM memory cutting

10

5

0

0

1

2

3

4

5

6

7

teaching process/n

Fig. 13 Adjustment frequency of shearer cutting drum

outside the confidence interval. By contrast, all height data for HMM memory cutting are distributed in the ±95 % confidence interval, with even a small amount of data on the center line of the confidence interval. Figure 12 exhibits the average residual errors of every teaching processes for traditional and HMM memory cutting. The average errors of teaching processes H 1, H 2, H 3, H 4, H 5, and H 6 are 0.06, 0.08, 0.08, 0.06, 0.05, and 0.07 m, respectively, for HMM memory cutting. The average errors for traditional memory cutting are 0.16, 0.17, 0.17, 0.15, 0.15, and 0.16 m, respectively. We can find that the predicted height data of shearer cutting drum for HMM memory cutting have higher accuracy than those for traditional memory cutting based on the preceding observations. Meanwhile, variations in residual error are consistent for the two memory cutting processes. The preceding analyses expound the accuracy of traditional and HMM memory cutting. In addition, Fig. 13 depicts the 8

residual coal-traditional memmory cutting redundant rock-traditional memmory cutting residual coal-HMM memmory cutting redundant rock-HMM memmory cutting

7 6

volume/m3

5 4 3 2 1 0

1

2

3

4

5

6

teaching process/n

Fig. 14 Average residual coal and redundant rock for traditional memory cutting and HMM memory cutting

adjustment frequency performance of shearer cutting drum for the two memory cutting methods. Traditional memory cutting needs to make 10, 15, 11, 14, 14, and 14 adjustments for teaching processes H 1, H 2, H 3, H 4, H 5, and H 6. HMM memory cutting needs to make 10, 6, 6, 7, 5, and 7 adjustments for the aforementioned teaching processes, respectively. HMM memory cutting is devoted to reducing the adjustment frequency and increase practicality. Less adjustment frequency can weaken the workload of shearer driver. Several observations are described in Fig. 14. The shearer cannot always track the coal–rock interface efficiently regardless of which memory cutting method is used. The key target for the shearer is cutting more coal and less rocks, although residual coal or redundant rock remains. The average residual coal and redundant rock are 5.36 and 5.08 m3 by traditional memory cutting. By contrast, the average residual coal and redundant rock are 1.40 and 1.63 m3 by HMM memory cutting. Furthermore, Fig. 14 proves that the average residual coal by HMM memory cutting is 3.96 m3 more than that by traditional memory cutting. Multiplying coal seam density 1.4 kg/m3 by coal seam volume 3.96 m3 provides the coal seam quality of 5.54 kg for fully mechanized mining face with an area of 50×1 m. For example, if the area of the fully mechanized mining face is 150×2,000 m, then the coal seam is 33.24 ton. The economic benefit of residual coal for HMM memory cutting is $2,326.80 based on the $70.00 price per ton of coal seam. The mining efficiency of coal resources can be increase by cutting more coal, thus resulting in more economic benefits.

Conclusion In this paper, the height adjustment of the shearer cutting drum is investigated to track the coal–rock interface. We apply the memory cutting method in adjusting the shearer cutting drum to address direct sensor deficiencies by considering the stable characteristics of coal seam thickness. The height data correlation of adjacent coal seams is investigated, and a new memory cutting method that adopts HMM is proposed. The performance of HMM memory cutting in adjusting the shearer cutting drum is evaluated via simulation and compared with traditional memory cutting. The results indicate that the shearer cutting drum can dynamically track the coal–rock interface via HMM memory cutting when coal seam thickness changes gradually. A marked improvement exhibited by HMM memory cutting is the predicted height data of the shearer cutting drum, which enables the shearer to cut more coal and less rocks. Moreover, HMM memory cutting aims to reduce the adjustment frequency and to decrease the workload of the shearer driver. HMM memory cutting is demonstrated to outperform traditional memory cutting and is suitable for application to fully mechanized mining face.

Arab J Geosci

In our future research, we will study the proposed HMM memory cutting method in relation to complex coal geological conditions, such as coal seams with dirt band, coal discontinuity, etc. We will also extend our experiments by using actual height data of the shearer cutting drum based on the mobile location of the shearer. Acknowledgments This work was done with support by the Fundamental Research Funds for the Central Universities (2012LWB37) and A Project Funded by A Priority Academic Program Development of Jiangsu Higher Education Institutions. The authors would like to thank the anonymous reviewers for their helpful comments which have improved the quality of the paper.

References Abarghouei HB, Kousari MR, Zarch MAA (2013) Prediction of drought in dry lands through feedforward artificial neural network abilities. Arab J Geosci 6(5):1417–1433 Alford D (1985) Automatic vertical steering of ranging drum shearers using MIDAS. Min Technol 4:125–129 AlSaud M (2008) Structural mapping from high resolution aeromagnetic data in west central Arabian Shield, Saudi Arabia using normalized derivatives. Arab J Geosci 1(2):129–136 Asfahani J, Borsaru M (2007) Low-activity spectrometric gamma-ray logging technique for delineation of coal/rock interfaces in dry blast holes. Appl Radiat Isot 65(6):748–755 Ashgari O, Esfahani NM (2013) A new approach for the geological risk evaluation of coal resources through a geostatistical simulation case study: Parvadeh III coal deposit. Arab J Geosci 6(3): 929–943 Ayhan M, Eyyuboglu EM (2006) Comparison of globoid and cylindrical shearer drums loading performance. J S Afr I Min Metall 106(1):51–56 Benndorf J (2013) Application of efficient methods of conditional simulation for optimising coal blending strategies in large continuous open pit mining operations. Int J Coal Geol 112:141–153 De LPL, Pucher M, Yamagishi J, Hernaez I, Saratxaga I (2012) Evaluation of speaker verification security and detection of HMM-based synthetic speech. IEEE T Audio Speech 20(8):2280–2290 Hassan MR, Ramamohanarao K, Kamruzzaman J (2013) A HMM-based adaptive fuzzy inference system for stock market forecasting. Neurocomputing 140:10–25 He XQ, Song L (2012) Status and future tasks of coal mining safety in China. Saf Sci 50(4):894–898

Hong DH, Sung JS, Oh KH, Kim NS (2012) Outlier detection and removal for hmm-based speech synthesis with an insufficient speech database. IEICE T Inf Syst E95D(9):2351–2354 Hosseini N, Oraee K, Shahriar K, Goshtasbi K (2013) Studying the stress redistribution around the longwall mining panel using passive seismic velocity tomography and geostatistical estimation. Arab J Geosci 6(5):1407–1416 Ju JF, Xu JL (2013) Structural characteristics of key strata and strata behaviour of a fully mechanized longwall face with 7.0-m height chocks. Int J Rock Mech Min 58:46–54 Li W, Fan QG, Wang YQ, Yang XF (2011) Adaptive height adjusting strategy research of shearer cutting drum. Acta Mont Slovaca 16(1): 114–122 Liu CS, Hou QQ (2004) A memory program-control mode of reappearing automatic lifting for shearer drum. Collied Mech Elect Technol 22(4):22–25 Nganje TN, Adamu CI, Ugbaja AN, Ebieme E, Sikakwe GU (2011) Environmental contamination of trace elements in the vicinity of Okpara coal mine, Enugu, southeastern Nigeria. Arab J Geosci 4(1– 2):199–205 Paul A, Singh AP, John LP, Singh AK, Khandelwal M (2012) Validation of RMR-based support design using roof bolts by numerical modeling for underground coal mine of Monnet Ispat, Raigarh, India—a case study. Arab J Geosci 5(6):1435–1448 Ratikanta S (2011) Application of sensor in mining machinery to recognize rock surfaces. Int J Comp Commun Technol 2(4):126–132 Ren F, Liu ZY, Yang ZJ (2010) Dynamics analysis for cutting part of shearer physical simulation system. Proceedings of 2010 I.E. International Conference on Information and Automation, Harbin, China, pp 260–264 Roy A, Leiva R (2012) Classification rules for multivariate repeated measures data with equicorrelated correlation trajectory on both time and spatial repeated measurements. Commun Stat-Theor M 41(8): 1411–1420 Singh TN, Pradhan SP, Vishal V (2013) Stability of slopes in a fire-prone mine in Jharia Coalfield, India. Arab J Geosci 6(2):419–427 Snopkowski R (2009) Stochastic model of the longwall face excavation using two-way shearer mining technology. Arch Min Sci 54(3):573–585 Strange AD, Ralston JC, Chandram V (2005) Near-surface interface detection for coal mining applications using Bispectral Features and GPR. Subsurf Sens Technol Appl 6(2):125–149 Wang F, Tu S, Bai Q (2012) Practice and prospects of fully mechanized mining technology for thin coal seams in China. J S Afr I Min Metall 112(2):161–170 Xu ZP, Wang ZB, Mi JP (2011) Shearer self-adaptive memory cutting. J Chongqing Univ 34(4):134–140 Yan BF, Zhu XY, Zhang ZJ, Zhang F (2007) Robust speech recognition based on neighborhood space. J Softw 18(4):878–883 Yuan G, Xia SX, Zhang L, Zhou Y (2011) Trajectory clustering algorithm based on structural similarity. J Commun 32(9):103–110