A single-frame coal drawing material connection control system and control method based on coal gangue multi-source information fusion identification

Abstract

The present application relates to a kind of single frame coal drawing material association control system and control method based on coal gangue multi-source information fusion identification, belong to coal gangue identification technical field, utilize the coal gangue identification of multi-source information fusion of internet of things structure, caving monitoring and tail beam control are realized to top coal.By vibration sensor, sound sensor, industrial camera simultaneously obtain vibration signal, sound signal, and image information, and obtain tail beam inclination sensor, plugboard displacement sensor data, utilize multiple convolutional neural network structure respectively to vibration, sound, image multi-source heterogeneous information is identified, and using D-S decision makes decision level fusion to multi-source heterogeneous information identification result, to accurately classify coal gangue according to different gangue content;And according to the control principle of the tail beam angle and the plugboard length corresponding to classification result makes control decision, realizes coal drawing control.The present application is fused by internet of things structure and multi-source information, is favorable to realize accurate coal gangue identification and coal drawing control.

Description

Technical Field

[0001] The present invention relates to a single-rack coal placing Internet of Things control system and a control method based on coal gangue multi-source information fusion identification, belonging to the technical field of coal gangue identification. Background Art

[0002] Fully-mechanized top-coal caving (FCC) is one of the primary mining methods for thick and extra-thick coal seams in my country. Currently, top-coal caving in FCC mining relies primarily on manual control. However, the mine environment is complex and harsh, and long periods of underground work can severely harm workers' physical and mental health. Furthermore, workers, influenced by subjective judgment, are prone to misjudgment, resulting in over- or under-caving. Coal gangue identification in top-coal caving is a key technology driving the development of intelligent control in FCC mining. Existing research on intelligent coal gangue identification in FCC mining relies primarily on single sensors, resulting in unstable identification results due to uncertainties in the mine. Direct control of the tail boom from opening to closing and the flapper from retracting to extending when the shutdown threshold is reached introduces hysteresis. Relying solely on controllers and microcomputers underground for coal gangue identification and decision-making carries a high risk of misjudgment. Therefore, this technology is still immature in terms of identification accuracy and top-coal caving control. With the continuous development of mobile Internet of Things systems, their reliability and stability are increasing, and they are now widely used in various fields. Internet of Things technology can realize the supervision of the entire top coal caving process, including top coal caving, coal gangue identification, and hydraulic support posture control. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention provides a single-frame coal caving IoT control system based on multi-source information fusion and identification of coal gangue, which can solve problems such as over-caving and under-caving in top coal caving operations and hysteresis in tail beam control, and realize real-time monitoring of the coal gangue mixing ratio and tail beam posture throughout the top coal caving process.

[0004] The technical solutions of the present invention are as follows:

[0005] The model utilizes the Internet of Things (IoT) architecture to achieve multi-source information fusion for gangue identification, top coal caving monitoring, and tail boom control. Vibration sensors, sound sensors, and industrial cameras simultaneously capture vibration signals generated by gangue particles striking the tail boom, sound signals generated by gangue falling, and image information of the coal flow and scraper conveyor surface. Data from the tail boom inclination sensor and blade displacement sensor are also acquired and processed by a controller before being sent to a MYSQL database for storage via the IoT module and mobile communication network. The model accesses the online database and utilizes multiple convolutional neural network structures to identify heterogeneous multi-source information, including vibration, sound, and image information. The DS decision-making process is then used to fuse these heterogeneous identification results at the decision-making level, accurately classifying gangue according to gangue content. The gangue content, tail boom angle, and blade extension length are uploaded to the cloud platform in real time. Control decisions are made based on the tail boom angle and blade length control principles corresponding to the classification results and sent to the controller. A solenoid valve controls the relief valve to change the oil pressure to achieve coal caving control, including tail boom angle and blade length. The present invention facilitates accurate coal gangue identification and coal caving control through the integration of the Internet of Things structure and multi-source information.

[0006] A single-rack coal placement IoT control system based on coal gangue multi-source information fusion and recognition, such as Figure 1 The main components shown are hydraulic supports, scraper conveyors, and multi-source information acquisition devices;

[0007] The hydraulic support is used to support the coal mining face and control the top coal caving. The hydraulic support includes a top beam, a shield beam, and a tail beam that are interconnected. A plug-in plate is provided at the end of the tail beam. The plug-in plate is used to adjust the size of the coal caving port to control the coal caving speed by its retraction and extension. A front column is provided below the tail beam. The hydraulic support is equipped with a coal caving control hydraulic system, which includes a solenoid valve, a relief valve, a hydraulic oil tank, a hydraulic cylinder, and a joint oil pipe. The hydraulic cylinder includes two groups, which are respectively provided at the bottom of the tail beam and inside the tail beam. The hydraulic cylinder is connected to the hydraulic oil tank through a joint oil pipe. The solenoid valve and the relief valve are connected and installed on the hydraulic oil tank to control the oil supply from the hydraulic oil tank to the hydraulic cylinder.

[0008] The scraper conveyor is installed under the hydraulic support to receive the dropped coal gangue mixture;

[0009] The multi-source information acquisition device includes an industrial camera, a vibration sensor, an inclination sensor, a sound sensor, and a displacement sensor; the industrial camera is installed on the front column of the hydraulic support using a camera fixing bracket. It is not affected by the top coal caving working conditions, but can obtain the latest top coal caving flow image information in a timely manner. The lens is facing the scraper conveyor to collect image information of the coal flow and the scraper conveyor surface; the vibration sensor is located at the bottom of the tail beam end, where the vibration sensitivity is highest, and is used to collect the vibration signal generated by the coal flow impacting the tail beam; the inclination sensor is set at the bottom of the tail beam, with its axis parallel to the tail beam axis, and is used to collect the angle between the tail beam and the horizontal plane at any time; the sound sensor is placed at the bottom of the tail beam or the shield beam, and is used to collect the sound signal generated by the coal gangue sliding during the top coal caving process; the displacement sensor is set inside the tail beam, with its axis parallel to the tail beam axis, and is used to collect the extended length of the plug plate;

[0010] The multi-source information acquisition device sends the information to the data conversion module. The data conversion module converts the multi-source heterogeneous information into digital signals and sends them to the controller. The controller is installed on the front column of the hydraulic support, above the industrial camera; the Internet of Things module esp8266 is installed on the front column of the hydraulic support and connected to the controller through the serial port. The Internet of Things module esp8266 is connected to the MQTT message queue module via wireless. The MQTT message queue module extracts data from the data packets pushed by the Internet of Things module esp8266 and sends them to the cloud platform. The cloud platform is connected to the computer. The computer is equipped with a coal gangue recognition module, which is executed in the computer. The coal gangue recognition module includes a convolutional neural network mechanism and a DS decision module. The coal gangue recognition module feeds back the control decision information to the controller based on the recognition results, thereby realizing the control of the hydraulic support posture.

[0011] Preferably, three vibration sensors are arranged transversely at the bottom of the tail beam via a magnetic base. The vibration acceleration sensor is arranged at the bottom of the tail beam of the hydraulic support, a location with high sensitivity to vibration signals. The transverse arrangement of three acceleration sensors can avoid misjudgments caused by abnormal coal gangue particles at a single location.

[0012] Preferably, there are three sound sensors, which are wireless wall-mounted and arranged longitudinally from near to far at the bottom of the tail beam and the cover beam. They are fixed by welding wires on the flanges at the bottom of the tail beam and the cover beam, so as to receive the sound signals of top coal caving in a more three-dimensional manner.

[0013] Preferably, the displacement sensor is located in the spacing groove between the starting end of the plug-in plate and the inner wall of the tail beam.

[0014] Preferably, the data conversion module includes an ADC module and a CMOS module. The ADC module is used to convert vibration, sound, tilt, and displacement analog voltage signals into digital signals, and the CMOS module is used to convert image information into digital quantities.

[0015] Preferably, the cloud platform includes a MYSQL database and a cloud monitoring and display platform. The cloud monitoring and display platform monitors the hydraulic support top coal caving status in real time. The computer reads multi-source heterogeneous data from the MYSQL database to identify coal gangue and feeds back the identification results to the cloud monitoring and display platform.

[0016] All instruments and equipment are of mining intrinsically safe type or installed in mining intrinsically safe housing.

[0017] A control method using the above-mentioned single-rack coal caving IoT control system based on coal gangue multi-source information fusion and identification includes the following steps:

[0018] ①The controller controls the hydraulic support to retract the plug plate, open the tail beam and start coal discharge;

[0019] ② The vibration sensor collects the tail beam vibration acceleration signal within 0.1s, the sound sensor collects the sound signal generated by the falling coal gangue within 0.1s, the industrial camera collects the image information of the coal flow and the scraper conveyor surface at the end of 0.1s, the inclination sensor collects the current angle between the tail beam and the horizontal plane, and the displacement sensor collects the current extension length of the plug-in board. The analog voltage signal and image information are converted into digital signals by the ADC module and CMOS module respectively, and sent synchronously to the controller via the I2C bus; Figure 11 As shown, each 0.1s interval is defined as a set of collected data: within this 0.1s, the three vibration acceleration sensors continuously collect vibration signals, and the three sound sensors continuously collect sound signals. At the end of the 0.1s, the industrial camera collects image information of the scraper conveyor and coal flow, the inclination sensor collects the angle between the current tail beam and the horizontal plane, and the displacement sensor collects the current extended length of the plugboard. These five types of data are uploaded synchronously.

[0020] ③The controller packages this data into JSON format and sends it to the IoT module ESP8266 via the serial port. The IoT module ESP8266 pushes the entire data packet to the MQTT message queue via Wi-Fi for data extraction. The MQTT rule engine extracts all the data, packages it, and sends it to the MYSQL database for storage.

[0021] ④ The gangue identification module reads the latest uploaded vibration, sound, and image information from the MYSQL database every 0.1s, performs identification, and obtains the identification results. The obtained gangue content range is synchronously uploaded to the cloud monitoring and display platform along with the tail beam inclination angle and the plugboard displacement information. At the same time, the control decision corresponding to the gangue content range is fed back to the controller, and the control decision includes the tail beam angle and the plugboard length.

[0022] ⑤ The controller controls the overflow valve through the solenoid valve, changes the oil supply from the hydraulic tank to the hydraulic cylinder to adjust the pressure, thereby achieving the purpose of controlling the tail beam angle and the length of the plug plate.

[0023] Preferably, in step ④, before performing gangue identification, the convolutional neural network multi-channel gangue identification model is first trained to divide the gangue content into A1, A1-A2, ..., A n-1 -A n , greater than A n , a total of n+1 intervals;

[0024] A coal-gangue mixture with a known gangue content range was placed on the top beam of a hydraulic support. The top coal was manually lowered, causing the mixture to slide down the tail beam. A vibration sensor and an acoustic sensor collected vibration and acoustic signals during the top coal caving test, respectively. The sampling frequency was set to 30,000 Hz, and the time interval was 0.1 s. The coal-gangue mixture was then dropped onto a rear scraper conveyor. An industrial camera was used to capture image information of the coal flow and the scraper conveyor surface. The same label was defined for these three types of multi-source heterogeneous information, and the information was treated as a single sample. M groups of tests were conducted on the coal-gangue mixture for each gangue content range.

[0025] 1) Model training

[0026] The image information, the vibration acceleration information matrix composed of three vibration signals, and the sound information matrix composed of three sound signals are respectively input into three convolutional neural networks. In each gangue content interval, the first m / 2 groups of samples are taken as training sets, and the last m / 2 groups of sample data are taken as test sets. The initial training times are set to 30. If the accuracy is low (the error is higher than the set value), the training times are increased by 10 times until the test set errors of the three convolutional neural networks for n+1 intervals are all lower than the set value σ. Then, the DS decision is used to fuse the convolutional neural network recognition results of the three types of information, image, vibration, and sound, under the same sample to obtain the final gangue content interval, and compare it with the actual gangue content interval of the sample. If the test set error of the multi-source heterogeneous fusion coal gangue recognition model for each gangue content interval is also lower than σ, the model training is terminated. If the final training times reach 150 times or the network is overfitted and still does not meet the accuracy requirements, the highest accuracy is selected as the convolutional neural network training times.

[0027] Further preferably, the three convolutional neural network structures include a single convolution layer, a single pooling layer, and a fully connected layer, wherein zero padding is performed around the vibration signal matrix and the sound signal matrix, the convolution kernel size is 3×3, the pooling size is 2×2, and the fully connected layer uses "Softmax" as the activation function for multi-classification;

[0028] The Softmax function is:

[0029] Among them, i represents the category index, V i is the output of the previous unit of the classifier, C is the total number of categories, Si is the relative probability that the predicted sample is of this class;

[0030] The other key parameters of the convolutional neural network model will be adjusted appropriately based on the test set accuracy. For example, the learning rate is initially set to 0.1 and then divided by 0.5 to improve the accuracy. The batch size is initially set to 10 and then increased by 10 to improve the accuracy. Other parameters use the default values ​​of the convolutional neural network system. The parameters of the three convolutional neural networks may vary.

[0031] 2) Coal gangue identification

[0032] The trained model is then used to identify gangue. The vibration sensor collects the tail beam vibration acceleration signal within 0.1s, the sound sensor collects the sound signal generated by the falling gangue within 0.1s, and the industrial camera collects image information of the coal flow and the scraper conveyor surface at the end of 0.1s. The images are packaged and sent to the MYSQL database through the controller and the Internet of Things module. The trained gangue recognition model reads the latest stored data in the database and inputs them into three convolutional neural networks to obtain three recognition results. The possible output results of each convolutional neural network are θ1, θ2, ..., θ n+1 And uncertainty, etc., a total of n+2 events; the set of these n+2 events is the identification framework Θ under this identification problem:

[0033] Θ={θ1,θ2,…,θ n+1 ,θ'}(2)

[0034] Among them, θ1 is the interval of the gangue content less than A1, θ2 is the interval of the gangue content within the interval A1-A2, ..., θ n+1 The gangue content interval is greater than A n , θ' is uncertainty; each element is mutually exclusive; the set of all subsets of the identification framework is denoted as power set 2 Θ :

[0035]

[0036] Among them is the empty set;

[0037] For each subset in Θ, a basic probability assignment (BPA) is performed according to a specific probability, where the assignment function m is a function from 2 Θ →[0,1], that is, for any subset θ, it must satisfy the following two conditions at the same time:

[0038]

[0039]

[0040] The BPA calculation method for the recognition results of three information sources: image, vibration, and sound is as follows:

[0041] m i (θ j )=α i u j , i=1,2,3; j=1,2,3,…,n+1(6)

[0042] m i (θ')=1-α i (7)

[0043] Where m i (θ j ) is the information source i for the gangue content in θ j Probability distribution function in the interval; α i is the reliability coefficient of the i-th information source (in this paper, the test accuracy of the i-th information source alone under CNN is used instead of the reliability coefficient); u j is the output gangue content of the i-th information source at θ j Membership degree within the interval; m i (θ') is the BPA value of the uncertainty of the i-th information source;

[0044] According to the synthesis rules of DS evidence theory, the BPA of the three information sources are synthesized to obtain the final coal gangue identification result:

[0045]

[0046] in,

[0047] Finally, according to the probability of each gangue content interval determined by DS, the maximum probability interval is selected as the final recognition result.

[0048] Preferably, in step ④, the control decision specifically includes:

[0049] Each gangue content range defines its tail beam angle and plug plate extension length accordingly, as shown in Table 1 below. (Specific values ​​are determined based on actual mining conditions and hydraulic support model)

[0050] Table 1 Correspondence between tail beam angle and plugboard extension length

[0051]

[0052] The gangue content range is divided into less than A1, A1-A2, ... A n-1 -A n , greater than A n The tail boom angle is defined as fully open, θ1, ..., θ n-1, fully closed, the extended length of the plug plate is defined as fully retracted, x1, ..., x n-1 , fully extended;

[0053] The controller initiates automatic top coal caving, opening the tail beam and retracting the flapper, allowing the top coal to slide onto the rear scraper conveyor via the tail beam. The convolutional neural network's multi-source heterogeneous gangue recognition model identifies the coal every 0.1 seconds and provides the corresponding gangue content interval. The tail beam angle and flapper length corresponding to the gangue content obtained from the gangue recognition results are fed back to the controller, which then generates a signal to adjust the oil supply to the coal caving control hydraulic system by controlling the solenoid valve to adjust the overflow valve, thereby adjusting the tail beam angle and flapper length to the set values. During the control process, the inclination sensor and displacement sensor continuously provide feedback on the current tail beam angle and flapper displacement until the set values ​​are reached.

[0054] When the gangue content is greater than A1, the control decision corresponding to the gangue content range is sent to the hydraulic support controller, the tail beam swings to the corresponding angle, and the plug extends to the corresponding length; continue to caving coal, if the gangue content drops below A1, the tail beam is fully opened and the plug is fully retracted; if the gangue content continues to increase to greater than A1, the tail beam is fully opened and the plug is fully retracted. n , the tail beam is controlled to be closed, the insert plate is extended, and coal discharge is stopped; in each gangue content interval, the tail beam and the insert plate are controlled to the corresponding angle and length, thereby realizing a progressive control decision with a transition effect and reducing the hysteresis effect caused by slow control response.

[0055] The beneficial effects of the present invention are:

[0056] 1. The present invention uses a multi-source information acquisition device (vibration sensor, sound sensor, industrial camera, tilt sensor, displacement sensor) to collect multi-source heterogeneous information.

[0057] 2. This invention establishes a convolutional neural network multi-source heterogeneous fusion coal gangue recognition model. The gangue content results derived from multiple sources of information, including the tail boom vibration acceleration signal, the sound signal generated during coal caving, and the surface image information of the scraper conveyor coal flow, are fused at the decision level to achieve the final recognition accuracy.

[0058] 3. This invention clarifies the control method for the tail beam inclination angle and the insert plate length. Based on the multi-threshold classification results, different intervals of gangue content are formulated with corresponding tail beam inclination angles and insert plate lengths, thus forming a progressive hydraulic support top coal caving control strategy.

[0059] 4. The present invention uses the Internet of Things system to achieve real-time monitoring of the gangue content in the coal flow, the angle of the hydraulic support tail beam and the length of the insert plate.

[0060] 5. Regarding sensor arrangement, the present invention has three vibration sensors arranged horizontally at the sensitive protruding end of the tail beam to reduce the probability of misjudgment caused by abnormal coal gangue particles at a single location; three sound sensors are arranged from near to far on the tail beam and shield beam to collect stereo sound information; and the industrial camera is installed on the front column so that it will not be affected by the top coal caving operation.

[0061] 6. The present invention adopts wireless sensors and uses mine-used wireless wifi for data transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 For IoT systems;

[0063] Figure 2 This is the structural diagram of the hydraulic support;

[0064] Figure 3 This is a schematic diagram of the hydraulic system for coal discharge control;

[0065] Figure 4 This is a schematic diagram of the tail boom bottom structure;

[0066] Figure 5 This is a schematic diagram of the bottom structure of the protective beam;

[0067] Figure 6 This is a structural diagram of the hydraulic support and scraper conveyor;

[0068] Figure 7 This is a schematic diagram of multi-source information transmission;

[0069] Figure 8 This is a schematic diagram of the transmission between the gangue identification module and the cloud platform;

[0070] Figure 9 This is a schematic diagram of the transmission of multi-source information and coal gangue identification module;

[0071] Figure 10 is a control flow chart;

[0072] Figure 11 Schematic diagram of the collection process;

[0073] Figure 12 Schematic diagram of DS discrimination;

[0074] Among them: 1- hydraulic support, 2- scraper conveyor, 3- multi-source information acquisition device, 4- data conversion module, 5- controller, 6- Internet of Things module esp8266, 7- MQTT message queue module, 8- computer, 9- cloud platform, 10- coal gangue identification module;

[0075] 1-Hydraulic support mainly includes 11-top beam, 12-shielding beam, 13-tail beam, 14-board, 15-front column, 16-coal discharge control hydraulic system, 161-solenoid valve, 162-relief valve, 163-hydraulic oil tank, 164-hydraulic cylinder, 165-connector oil pipe;

[0076] 3-The multi-source information acquisition device mainly includes 17-industrial camera, 18-vibration sensor, 19-tilt sensor, 20-sound sensor, 21-displacement sensor, 22-ADC module, 23-CMOS module, 24-MYSQL database, and 25-cloud monitoring and display platform. DETAILED DESCRIPTION

[0077] The present invention will be further described below with reference to embodiments and accompanying drawings, but is not limited thereto.

[0078] Example 1:

[0079] A single-rack coal placement IoT control system based on coal gangue multi-source information fusion and recognition, such as Figure 1 、 Figure 2 The structure shown mainly includes a hydraulic support 1, a scraper conveyor 2, and a multi-source information acquisition device 3.

[0080] The hydraulic support is used to support the coal mining face and control the top coal falling. The hydraulic support includes a top beam 11, a shield beam 12, and a tail beam 13 that are connected to each other. A plug-in plate 14 is provided at the end of the tail beam. The plug-in plate is used to adjust the size of the coal discharge port to control the coal discharge speed by retracting and extending the plug-in plate. A front column 15 is provided under the tail beam. A coal discharge control hydraulic system 16 is provided on the hydraulic support. Figure 3 As shown, the coal discharge control hydraulic system includes a solenoid valve 161, a relief valve 162, a hydraulic oil tank 163, a hydraulic cylinder 164, and a joint oil pipe 165. The hydraulic cylinder includes two groups, which are respectively arranged at the bottom of the tail beam and inside the tail beam. The hydraulic cylinder is connected to the hydraulic oil tank through a joint oil pipe. The solenoid valve and the relief valve are connected and installed on the hydraulic oil tank to control the oil supply from the hydraulic oil tank to the hydraulic cylinder.

[0081] The scraper conveyor is installed under the hydraulic support to receive the dropped coal gangue mixture.

[0082] The multi-source information acquisition device includes an industrial camera 17, a vibration sensor 18, an inclination sensor 19, a sound sensor 20, and a displacement sensor 21; Figure 4-Figure 6As shown, the industrial camera is installed on the front column of the hydraulic support using a camera fixing bracket. It is not affected by the top coal caving working conditions, but can obtain the latest top coal caving flow image information in a timely manner. The lens is facing the scraper conveyor to collect image information of the coal flow and the scraper conveyor surface; the vibration sensor is located at the bottom of the tail beam end, where the vibration sensitivity is the highest, and is used to collect vibration signals generated by the coal flow impacting the tail beam; the inclination sensor is set at the bottom of the tail beam, with its axis parallel to the tail beam axis, and is used to collect the angle between the tail beam and the horizontal plane at any time; the sound sensor is placed at the bottom of the tail beam or the shield beam, and is used to collect sound signals generated by the coal gangue sliding during the top coal caving process; the displacement sensor is set inside the tail beam, with its axis parallel to the tail beam axis, and is used to collect the extended length of the plug plate;

[0083] like Figure 7-10 As shown, the multi-source information acquisition device sends the information to the data conversion module 4. The data conversion module converts the multi-source heterogeneous information into digital signal form and sends it to the controller 5. The controller is installed on the front column of the hydraulic support, above the industrial camera; the Internet of Things module esp8266 6 is installed on the front column of the hydraulic support and connected to the controller through the serial port. The Internet of Things module esp8266 6 is connected to the MQTT message queue 7 module via wireless. The MQTT message queue module extracts data from the data packet pushed by the Internet of Things module esp8266 and sends it to the cloud platform 9. The cloud platform is connected to the computer 8. The computer is equipped with a coal gangue recognition module 10. The coal gangue recognition module is executed in the computer. The coal gangue recognition module includes a convolutional neural network mechanism and a DS decision module. The coal gangue recognition module feeds back control decision information to the controller based on the recognition result, thereby realizing the control of the hydraulic support posture.

[0084] The data conversion module includes an ADC module and a CMOS module. The ADC module converts analog voltage signals such as vibration, sound, inclination, and displacement into digital signals, while the CMOS module converts image information into digital quantities. The cloud platform includes a MYSQL database and a cloud monitoring and display platform. The cloud monitoring and display platform provides real-time monitoring of the hydraulic support top coal caving status. The computer accesses the multi-source heterogeneous data in the MYSQL database to identify coal gangue and feeds the identification results back to the cloud monitoring and display platform. All instruments and equipment are either intrinsically safe for mining or installed in mining intrinsically safe enclosures.

[0085] Example 2:

[0086] A single-rack coal caving IoT control system based on multi-source coal gangue information fusion and identification has the same structure as described in Example 1, except that three vibration sensors are arranged transversely at the bottom of the tail beam via magnetic bases. A vibration acceleration sensor is arranged at the bottom of the hydraulic support tail beam, a location highly sensitive to vibration signals. The transverse arrangement of three acceleration sensors can avoid misjudgments caused by abnormal coal gangue particles at a single location.

[0087] There are three wireless wall-mounted sound sensors, which are arranged longitudinally from near to far at the bottom of the tail beam and the cover beam. They are fixed by welding wires on the flanges at the bottom of the tail beam and the cover beam, so as to receive the sound signals of top coal caving in a more three-dimensional way.

[0088] The displacement sensor is located in the spacing groove between the starting end of the plug-in plate and the inner wall of the tail beam.

[0089] Example 3:

[0090] A control method for a single-rack coal caving IoT control system based on coal gangue multi-source information fusion and identification as described in Example 2 includes the following steps:

[0091] ①The controller controls the hydraulic support to retract the plug plate, open the tail beam and start coal discharge;

[0092] ② The vibration sensor collects the tail beam vibration acceleration signal within 0.1s, the sound sensor collects the sound signal generated by the falling coal gangue within 0.1s, the industrial camera collects the image information of the coal flow and the scraper conveyor surface at the end of 0.1s, the inclination sensor collects the current angle between the tail beam and the horizontal plane, and the displacement sensor collects the current extension length of the plug-in board. The analog voltage signal and image information are converted into digital signals by the ADC module and CMOS module respectively, and sent synchronously to the controller via the I2C bus; Figure 11 As shown, each 0.1s interval is defined as a set of collected data: within this 0.1s, the three vibration acceleration sensors continuously collect vibration signals, and the three sound sensors continuously collect sound signals. At the end of the 0.1s, the industrial camera collects image information of the scraper conveyor and coal flow, the inclination sensor collects the angle between the current tail beam and the horizontal plane, and the displacement sensor collects the current extended length of the plugboard. These five types of data are uploaded synchronously.

[0093] ③The controller packages this data into JSON format and sends it to the IoT module ESP8266 via the serial port. The IoT module ESP8266 pushes the entire data packet to the MQTT message queue via Wi-Fi for data extraction. The MQTT rule engine extracts all the data, packages it, and sends it to the MYSQL database for storage.

[0094] ④ The gangue identification module reads the latest uploaded vibration, sound, and image information from the MYSQL database every 0.1s, performs identification, and obtains the identification results. The obtained gangue content range is synchronously uploaded to the cloud monitoring and display platform along with the tail beam inclination angle and the plugboard displacement information. At the same time, the control decision corresponding to the gangue content range is fed back to the controller, and the control decision includes the tail beam angle and the plugboard length.

[0095] In step ④, before gangue identification, the convolutional neural network multi-channel gangue identification model must be trained first, and the gangue content is divided into A1, A1-A2, ..., A n-1 -A n , greater than A n , a total of n+1 intervals;

[0096] A coal-gangue mixture with a known gangue content range was placed on the top beam of a hydraulic support. The top coal was manually lowered, causing the mixture to slide down the tail beam. A vibration sensor and an acoustic sensor collected vibration and acoustic signals during the top coal caving test, respectively. The sampling frequency was set to 30,000 Hz, and the time interval was 0.1 s. The coal-gangue mixture was then dropped onto a rear scraper conveyor. An industrial camera was used to capture image information of the coal flow and the scraper conveyor surface. The same label was defined for these three types of multi-source heterogeneous information, and the information was treated as a single sample. M groups of tests were conducted on the coal-gangue mixture for each gangue content range.

[0097] 1) Model training

[0098] The image information, the vibration acceleration information matrix composed of three vibration signals, and the sound information matrix composed of three sound signals are respectively input into three convolutional neural networks. In each gangue content interval, the first m / 2 groups of samples are taken as training sets, and the last m / 2 groups of sample data are taken as test sets. The initial training times are set to 30. If the accuracy is low (the error is higher than the set value), the training times are increased by 10 times until the test set errors of the three convolutional neural networks for n+1 intervals are all lower than the set value σ. Then, the DS decision is used to fuse the convolutional neural network recognition results of the three types of information, image, vibration, and sound, under the same sample to obtain the final gangue content interval, and compare it with the actual gangue content interval of the sample. If the test set error of the multi-source heterogeneous fusion coal gangue recognition model for each gangue content interval is also lower than σ, the model training is terminated. If the final training times reach 150 times or the network is overfitted and still does not meet the accuracy requirements, the highest accuracy is selected as the convolutional neural network training times.

[0099] The three convolutional neural network structures are as follows: a single convolutional layer, a single pooling layer, and a fully connected layer. Zero padding is performed around the vibration signal matrix and the sound signal matrix. The convolution kernel size is 3×3, the pooling size is 2×2, and the fully connected layer uses "Softmax" as the activation function for multi-classification.

[0100] The Softmax function is:

[0101] Among them, i represents the category index, V i is the output of the previous unit of the classifier, C is the total number of categories, S i is the relative probability that the predicted sample is of this class;

[0102] The other key parameters of the convolutional neural network model were adjusted appropriately based on the test set accuracy: the learning rate was initially set to 0.1 and was divided by 0.5 to improve it; the sample batch size was initially set to 10 and was increased by 10 to improve it; all other parameters used the default values ​​of the convolutional neural network system. The parameters of the three convolutional neural networks may vary.

[0103] 2) Coal gangue identification

[0104] The trained model is then used to identify gangue. The vibration sensor collects the tail beam vibration acceleration signal within 0.1s, the sound sensor collects the sound signal generated by the falling gangue within 0.1s, and the industrial camera collects image information of the coal flow and the scraper conveyor surface at the end of 0.1s. The images are packaged and sent to the MYSQL database through the controller and the Internet of Things module. The trained gangue recognition model reads the latest stored data in the database and inputs them into three convolutional neural networks to obtain three recognition results. The possible output results of each convolutional neural network are θ1, θ2, ..., θ n+1 And uncertainty, etc., a total of n+2 events; the set of these n+2 events is the identification framework Θ under this identification problem:

[0105] Θ={θ1,θ2,…,θ n+1 ,θ'}(2)

[0106] Among them, θ1 is the interval of the gangue content less than A1, θ2 is the interval of the gangue content within the interval A1-A2, ..., θ n+1 The gangue content interval is greater than A n , θ' is uncertainty; each element is mutually exclusive; the set of all subsets of the identification framework is denoted as power set 2 Θ :

[0107]

[0108] Among them is the empty set;

[0109] For each subset in Θ, a basic probability assignment (BPA) is performed according to a specific probability, where the assignment function m is a function from 2 Θ →[0,1], that is, for any subset θ, it must satisfy the following two conditions at the same time:

[0110]

[0111]

[0112] The BPA calculation method for the recognition results of three information sources: image, vibration, and sound is as follows:

[0113] m i(θ j )=α i u j , i=1,2,3; j=1,2,3,…,n+1(6)

[0114] m i (θ')=1-α i (7)

[0115] Where m i (θ j ) is the information source i for the gangue content in θ j Probability distribution function in the interval; α i is the reliability coefficient of the i-th information source (in this paper, the test accuracy of the i-th information source alone under CNN is used instead of the reliability coefficient); u j is the output gangue content of the i-th information source at θ j Membership degree within the interval; m i (θ') is the BPA value of the uncertainty of the i-th information source;

[0116] According to the synthesis rules of DS evidence theory, the BPA of the three information sources are synthesized to obtain the final coal gangue identification result:

[0117]

[0118] in,

[0119] Finally, according to the probability of each gangue content interval determined by DS, the maximum probability interval is selected as the final recognition result.

[0120] In step ④, the control decision specifically includes:

[0121] Each gangue content range defines its tail beam angle and plug plate extension length accordingly, as shown in Table 1 below. (Specific values ​​are determined based on actual mining conditions and hydraulic support model)

[0122] Table 1 Correspondence between tail beam angle and plugboard extension length

[0123]

[0124] The controller initiates automatic top coal caving, opening the tail beam and retracting the flapper, allowing the top coal to slide onto the rear scraper conveyor via the tail beam. The convolutional neural network's multi-source heterogeneous gangue recognition model identifies the coal every 0.1 seconds and provides the corresponding gangue content interval. The tail beam angle and flapper length corresponding to the gangue content obtained from the gangue recognition results are fed back to the controller, which then generates a signal to adjust the oil supply to the coal caving control hydraulic system by controlling the solenoid valve to adjust the overflow valve, thereby adjusting the tail beam angle and flapper length to the set values. During the control process, the inclination sensor and displacement sensor continuously provide feedback on the current tail beam angle and flapper displacement until the set values ​​are reached.

[0125] When the gangue content is greater than A1, the control decision corresponding to the gangue content range is sent to the hydraulic support controller, the tail beam swings to the corresponding angle, and the plug extends to the corresponding length; continue to caving coal, if the gangue content drops below A1, the tail beam is fully opened and the plug is fully retracted; if the gangue content continues to increase to greater than A1, the tail beam is fully opened and the plug is fully retracted. n , the tail beam is controlled to be closed, the insert plate is extended, and coal discharge is stopped; in each gangue content interval, the tail beam and the insert plate are controlled to the corresponding angle and length, thereby realizing a progressive control decision with a transition effect and reducing the hysteresis effect caused by slow control response.

[0126] ⑤ The controller controls the overflow valve through the solenoid valve, changes the oil supply from the hydraulic tank to the hydraulic cylinder to adjust the pressure, thereby achieving the purpose of controlling the tail beam angle and the length of the plug plate.

Claims

1. A control method for a single-rack coal placement IoT control system based on coal gangue multi-source information fusion and identification, characterized in that: The control system includes hydraulic supports, scraper conveyors, and multi-source information acquisition devices; The hydraulic support is used to support the coal mining face and control the top coal caving. The hydraulic support includes a top beam, a shield beam, and a tail beam that are interconnected. A plug-in plate is provided at the end of the tail beam. The plug-in plate is used to adjust the size of the coal caving port to control the coal caving speed. A front column is provided below the tail beam. The hydraulic support is equipped with a coal caving control hydraulic system, which includes a solenoid valve, a relief valve, a hydraulic oil tank, a hydraulic cylinder, and a joint oil pipe. The hydraulic cylinder includes two groups, which are respectively located at the bottom of the tail beam and inside the tail beam. The hydraulic cylinder is connected to the hydraulic oil tank through a joint oil pipe. The solenoid valve and the relief valve are connected and installed on the hydraulic oil tank to control the oil supply from the hydraulic oil tank to the hydraulic cylinder. The scraper conveyor is installed under the hydraulic support to receive the dropped coal gangue mixture; The multi-source information acquisition device includes an industrial camera, a vibration sensor, an inclination sensor, an acoustic sensor, and a displacement sensor. The industrial camera is installed on the front column of the hydraulic support, with the lens facing the scraper conveyor. The vibration sensor is located at the bottom end of the tail beam. The inclination sensor is set at the bottom of the tail beam, with its axis parallel to the axis of the tail beam. The acoustic sensor is placed at the bottom of the tail beam or the shield beam. The displacement sensor is set inside the tail beam, with its axis parallel to the axis of the tail beam. The multi-source information acquisition device sends information to the data conversion module, which converts the multi-source heterogeneous information into digital signals and sends them to the controller. The controller is installed on the front column of the hydraulic support. The Internet of Things module esp8266 is installed on the front column of the hydraulic support and connected to the controller through the serial port. The Internet of Things module esp8266 is connected to the MQTT message queue module via wireless. The MQTT message queue module extracts data from the data packets pushed by the Internet of Things module esp8266 and sends them to the cloud platform. The cloud platform is connected to the computer, which has a coal gangue identification module. The data conversion module includes an ADC module and a CMOS module. The ADC module is used to convert vibration, sound, inclination, and displacement analog voltage signals into digital signals. The CMOS module is used to convert image information into digital quantities. The cloud platform includes a MYSQL database and a cloud monitoring and display platform. The cloud monitoring and display platform monitors the top coal caving status of the hydraulic support in real time. The computer reads the multi-source heterogeneous data in the MYSQL database to identify coal gangue and feeds back the identification results to the cloud monitoring and display platform. The method comprises the following steps: ①The controller controls the hydraulic support to retract the plug plate, open the tail beam and start coal discharge; ② The vibration sensor collects the tail beam vibration acceleration signal within 0.1s, the sound sensor collects the sound signal generated by the falling coal gangue within 0.1s, the industrial camera collects image information of the coal flow and the scraper conveyor surface at the end of 0.1s, the inclination sensor collects the current angle between the tail beam and the horizontal plane, and the displacement sensor collects the current extension length of the plugboard. The analog voltage signal and image information are converted into digital signals by the ADC module and CMOS module respectively, and are synchronously sent to the controller via the I2C bus. ③The controller packages this data into JSON format and sends it to the IoT module ESP8266 via the serial port. The IoT module ESP8266 pushes the entire data packet to the MQTT message queue via Wi-Fi for data extraction. The MQTT rule engine extracts all the data, packages it, and sends it to the MYSQL database for storage. ④ The gangue identification module reads the latest uploaded vibration, sound, and image information from the MYSQL database every 0.1s, performs identification, and obtains the identification results. The obtained gangue content range is synchronously uploaded to the cloud monitoring and display platform along with the tail beam inclination angle and the plugboard displacement information. At the same time, the control decision corresponding to the gangue content range is fed back to the controller, and the control decision includes the tail beam angle and the plugboard length. ⑤ The controller controls the overflow valve through the solenoid valve, changes the oil supply from the hydraulic tank to the hydraulic cylinder to adjust the pressure, thereby controlling the tail beam angle and the length of the plug plate.

2. The control method of a single-rack coal placement IoT control system based on coal gangue multi-source information fusion and identification according to claim 1 is characterized in that: There are three vibration sensors, which are arranged horizontally at the bottom of the tail beam through magnetic bases; There are three sound sensors, which are arranged longitudinally from near to far at the bottom of the tail beam and the shield beam.

3. The control method of a single-rack coal caving IoT control system based on coal gangue multi-source information fusion and identification according to claim 1 is characterized in that: The displacement sensor is located in the spacing groove between the starting end of the plug-in plate and the inner wall of the tail beam.

4. The control method of a single-rack coal caving IoT control system based on coal gangue multi-source information fusion and identification according to claim 1 is characterized in that: In step ④, before gangue identification, the convolutional neural network multi-channel gangue identification model must be trained first, and the gangue content is divided into A1, A1-A2, ..., A n-1 -A n , greater than A n , a total of n+1 intervals; A coal-gangue mixture with a known gangue content range was placed on the top beam of a hydraulic support. The top coal was manually lowered, causing the mixture to slide down the tail beam. A vibration sensor and an acoustic sensor collected vibration and acoustic signals during the top coal caving test, respectively. The sampling frequency was set to 30,000 Hz, and the time interval was 0.1 s. The coal-gangue mixture was then dropped onto a rear scraper conveyor. An industrial camera was used to capture image information of the coal flow and the scraper conveyor surface. The same label was defined for these three types of multi-source heterogeneous information, and the information was treated as a single sample. M groups of tests were conducted on the coal-gangue mixture for each gangue content range. 1) Model training The image information, the vibration acceleration information matrix composed of three vibration signals, and the sound information matrix composed of three sound signals are respectively input into three convolutional neural networks. In each gangue content interval, the first m / 2 groups of samples are taken as training sets, and the last m / 2 groups of sample data are taken as test sets. The initial training times are set to 30. If the accuracy is low, the training times are increased by 10 times until the test set errors of the three convolutional neural networks for n+1 intervals are all lower than the set value σ. Then, the DS decision is used to fuse the convolutional neural network recognition results of the three types of information, image, vibration, and sound, under the same sample to obtain the final gangue content interval, and compare it with the actual gangue content interval of the sample. If the test set error of the multi-source heterogeneous fusion coal gangue recognition model for each gangue content interval is also lower than σ, the model training is terminated. If the final training times reach 150 times or the network is overfitted and still does not meet the accuracy requirements, the highest accuracy is selected as the convolutional neural network training times. 2) Coal gangue identification The trained model is then used to identify gangue. The vibration sensor collects the tail beam vibration acceleration signal within 0.1s, the sound sensor collects the sound signal generated by the falling gangue within 0.1s, and the industrial camera collects image information of the coal flow and the scraper conveyor surface at the end of 0.1s. The images are packaged and sent to the MYSQL database through the controller and the Internet of Things module. The trained gangue recognition model reads the latest stored data in the database and inputs them into three convolutional neural networks respectively to obtain three recognition results. The output of each convolutional neural network is θ1, θ2, ..., θ n+1 And there are n+2 events with uncertainty; the set of these n+2 events is the identification framework Θ under this identification problem: Θ={θ1,θ2,…,θ n+1 ,θ'} (2) Among them, θ1 is the interval of the gangue content less than A1, θ2 is the interval of the gangue content within the interval A1-A2, ..., θ n+1 The gangue content interval is greater than A n , θ' is uncertainty; each element is mutually exclusive; the set of all subsets of the identification framework is denoted as power set 2 Θ : Among them is the empty set; For each subset in Θ, a basic probability assignment (BPA) is performed according to a specific probability, where the assignment function m is a function from 2 Θ →[0,1], that is, for any subset θ, it must satisfy the following two conditions at the same time: The BPA calculation method for the recognition results of three information sources: image, vibration, and sound is as follows: m i (i j )=a i you j ,i=1,2,3;j=1,2,3,…,n+1 (6) m i (θ')=1-α i (7) Where m i (θ j ) is the information source i for the gangue content in θ j Probability distribution function in the interval; α i is the reliability coefficient of the i-th type of information source; u j is the output gangue content of the i-th information source at θ j Membership degree within the interval; m i (θ') is the BPA value of the uncertainty of the i-th information source; According to the synthesis rules of DS evidence theory, the BPA of the three information sources are synthesized to obtain the final coal gangue identification result: in, Finally, according to the probability of each gangue content interval determined by DS, the maximum probability interval is selected as the final recognition result.

5. The control method of the single-rack coal caving IoT control system based on coal gangue multi-source information fusion and identification according to claim 4 is characterized in that: In step 1), the three convolutional neural network structures are: including a single convolutional layer, a single pooling layer, and a fully connected layer. Zero padding is performed around the vibration signal matrix and the sound signal matrix. The convolution kernel size is 3×3, the pooling size is 2×2, and the fully connected layer uses "Softmax" as the activation function for multi-classification. The Softmax function is: Among them, i represents the category index, V i is the output of the previous unit of the classifier, C is the total number of categories, S i is the relative probability that the predicted sample belongs to this class.

6. The control method of the single-rack coal caving IoT control system based on coal gangue multi-source information fusion and identification according to claim 5 is characterized in that: In step 1), other main parameters in the convolutional neural network model will be adjusted appropriately according to the test set accuracy: the learning rate is initially set to 0.1 and is divided by 0.5 each time to improve it; the sample batch capacity is initially set to 10 and is increased by 10 each time to improve it; other parameters use the default values ​​of the convolutional neural network system.

7. The control method of a single-rack coal caving IoT control system based on coal gangue multi-source information fusion and identification according to claim 1 is characterized in that: In step ④, the control decision specifically includes: Each gangue content interval defines its tail beam angle and plug plate extension length accordingly. The gangue content interval is divided into less than A1, A1-A2, ... A n-1 -A n , greater than A n The tail boom angle is defined as fully open, θ1, ..., θ n-1 , fully closed, the extended length of the plug plate is defined as fully retracted, x1, ..., x n-1 , fully extended; The controller initiates automatic top coal caving, opening the tail beam and retracting the flapper, allowing the top coal to slide onto the rear scraper conveyor via the tail beam. The convolutional neural network's multi-source heterogeneous gangue recognition model identifies the coal every 0.1 seconds and provides the corresponding gangue content interval. The tail beam angle and flapper length corresponding to the gangue content obtained from the gangue recognition results are fed back to the controller, which then generates a signal to adjust the oil supply to the coal caving control hydraulic system by controlling the solenoid valve to adjust the overflow valve, thereby adjusting the tail beam angle and flapper length to the set values. During the control process, the inclination sensor and displacement sensor continuously provide feedback on the current tail beam angle and flapper displacement until the set values ​​are reached. When the gangue content is greater than A1, the control decision corresponding to the gangue content range is sent to the hydraulic support controller, the tail beam swings to the corresponding angle, and the plug extends to the corresponding length; continue to caving coal, if the gangue content drops below A1, the tail beam is fully opened and the plug is fully retracted; if the gangue content continues to increase to greater than A1, the tail beam is fully opened and the plug is fully retracted. n , then the tail beam is controlled to close, the plug plate is extended, and coal discharge is stopped.

Images