Novel chain jumping monitoring and alarming system for side double-chain scraper

Abstract

The invention discloses a novel side double-chain scraper chain skipping monitoring and alarming system, and relates to the technical field of double-chain scraper chain skipping monitoring, the novel side double-chain scraper chain skipping monitoring and alarming system comprises a starlight-level camera, a sensing layer, a data analysis layer, a data processing layer and an application layer, the starlight-level camera continuously shoots video images at a set frame rate and 4 million pixels, and the sensing layer is used for sensing the data analysis layer and the data processing layer; the image acquisition layer is used for acquiring image data and transmitting the image data to the signal processing module in real time, the sensing layer is used for capturing tiny displacement changes during chain operation, the data analysis layer is used for judging fault types through a pre-trained fault classification model, the data processing layer is used for carrying out real-time reasoning analysis by applying an AI model, and the application layer is used for carrying out sound and light alarm. And the starlight-level camera continuously shoots video images at a set frame rate, not less than 25fps and 4 million pixels and transmits image data to the signal processing module in real time, and the system has the characteristics of comprehensive and accurate monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal mine coal preparation equipment monitoring, in particular to a novel side double-chain scraper machine chain jumping monitoring and alarm system. BACKGROUND

[0002] In a coal preparation plant, the side double-chain scraper machine is an important transportation equipment. However, the scraper machine often has faults such as single-sided chain jumping, single-sided tooth jumping, broken connecting ring, and chain breaking. The existing protection device cannot effectively identify these fault types, and sometimes it can cause serious accidents such as chain winding of the machine head wheel, which seriously affects production safety and efficiency.

[0003] The causes of chain jumping of the scraper machine are complex, including improper chain tightness, chain wheel and chain pitch mismatch, double chain length error, scraper jamming, and component wear. In the prior art, domestic and foreign scraper machine protection devices mainly focus on chain breaking protection, and cannot accurately identify single-sided tooth jumping, single-sided chain jumping and other faults, and it is also difficult to monitor the signs of chain jumping in advance, which leads to untimely discovery of faults and further expansion of fault damage.

[0004] Due to the untimely discovery and handling of chain jumping faults, serious consequences often occur. After the chain jumps out of the chain wheel, it will cause severe friction and collision with other components of the scraper machine, resulting in damage to the chain, chain wheel, scraper and other components, shortening the service life of the equipment and increasing maintenance costs. At the same time, chain jumping also causes interruption of material transportation, affects the continuity of the entire production process, reduces production efficiency, and brings great economic losses to the enterprise. Some existing technologies use simple contact type monitoring devices, such as installing limit switches near the chain, which can only trigger an alarm when the chain jumps out of the chain wheel with a large displacement, and cannot provide early warning for chain jumping hazards. Therefore, there is an urgent need for a system that can comprehensively and accurately monitor various chain jumping and chain breaking faults of the side double-chain scraper machine. SUMMARY

[0005] The purpose of the present application is to provide a novel side double-chain scraper machine chain jumping monitoring and alarm system to solve the problems raised in the background.

[0006] In order to solve the above technical problems, the present application provides the following technical scheme: a novel side double-chain scraper machine chain jumping monitoring and alarm system, comprising a starlight level camera, a perception layer, a data analysis layer, a data processing layer, and an application layer. The starlight level camera continuously captures video images at a set frame rate and 4 million pixels, and transmits image data to a signal processing module in real time. The perception layer is used to capture the small displacement changes when the chain is running. The data analysis layer is used to determine the fault type through a pre-trained fault classification model. The data processing layer uses an AI model for real-time inference analysis. The application layer is used for sound and light alarm.

[0007] According to the technical scheme, the starlight level camera continuously shoots video images at a set frame rate of no less than 25 fps and 4 million pixels, and transmits image data to the signal processing module in real time. Since the starlight level camera can provide color images in low light conditions, compared with traditional black and white monitoring pictures, it can provide more detailed information for subsequent image analysis, and help to more accurately determine whether the chain jumps, whether the scraper has abnormal displacement, etc.

[0008] The perception layer is an infrared sensor module, two high-precision laser infrared displacement sensors are selected and installed above the chains on both sides of the scraper machine. The measurement range covers the displacement interval from normal chain operation to chain jump, such as 0-50mm, with a resolution of 0.01mm, ensuring that small displacement changes can be captured. The sensor uses modulated laser technology and has strong anti-environmental light interference ability. Even in the presence of dust and light changes in the working environment of the scraper machine, it can still maintain stable measurement accuracy. The sensor continuously emits a laser beam to the edge of the chain. The laser beam is reflected by the chain and received by the receiving end. The sensor calculates the real-time distance between the chain and the sensor according to the time difference and phase difference between emission and reception. When operating normally, the distance value is stable within a predetermined normal range, such as 10±0.5mm. By detecting the change in the reflected distance of the chain, the sensor outputs on-off signals and high-low level signals to sense chain loosening, deviation and breakage, and signal mutations when blocked.

[0009] The data analysis layer includes an industrial host computer, which includes an image recognition unit, a signal processing unit, and a machine learning classification algorithm model.

[0010] The data processing layer selects an industrial-grade AI board with high integration and strong computing power. The AI board is equipped with a high-performance neural network processor with strong parallel computing power, which can quickly process massive data collected by multiple sensors. At the same time, it has rich interface resources, including multiple GPIO interfaces, UART interfaces, SPI interfaces, and Ethernet interfaces, etc., which can meet the connection needs of sensor modules, signal processing modules, alarm modules and other components, and has good scalability, facilitating the upgrading of subsequent system functions.

[0011] The application layer includes a field alarm module: installed near the scraper, using an audible and visual alarm mode, timely prompting the field staff when a fault occurs; a remote monitoring terminal: set in the duty room, displaying fault information through a man-machine interface, integrating an audible and visual alarm module, used for field alarm, a red LED lamp + buzzer, triggering immediately when a fault is detected; integrating a data storage module, such as an SD card, for audible and visual alarm, realizing remote monitoring and early warning; a data storage and management module: storing the original data of the perception layer, image, sensor signal, fault judgment result of the data analysis layer, fault type, time, position and execution record of the control layer; supporting historical data query and trend analysis, providing data support for fault tracing and equipment maintenance optimization.

[0012] According to the technical scheme, the functions of the industrial host and the machine learning classification algorithm model include:

[0013] Image recognition: based on the YOLO11-0BB rotating target detection model, a large number of normal scraper, chain and broken chain image data are labeled, the model directly outputs the 4 vertex coordinates of the scraper in the image, the inclination angle and the broken chain state;

[0014] Signal analysis: filtering and noise reduction processing is performed on the infrared sensor signal, the time difference and amplitude change of the chain signals on both sides are calculated, and the chain tightness difference and breakage are judged;

[0015] Comprehensive judgment: fusion of image features and sensor signals, determination of fault types through a pre-trained fault classification model, such as a support vector machine, including single-sided tooth skipping / chain skipping / chain breaking, double-sided chain skipping / chain breaking, generation of fault level and processing instructions.

[0016] According to the technical scheme, the AI board is built-in with a trained chain skipping recognition model, which is trained based on a large amount of sensor data of the scraper under normal operation and chain skipping state, including distance data of the infrared displacement sensor, vibration signal of the vibration sensor and image data of the starlight level camera. After receiving various data transmitted by the signal processing module, the AI board uses the model for real-time inference analysis. Through deep learning and pattern recognition of data, the characteristics of chain skipping faults can be accurately identified, effectively distinguishing between small fluctuations in normal operation and real chain skipping abnormalities, and further improving the accuracy of chain skipping judgment.

[0017] According to the technical scheme, the working method of the system includes:

[0018] S1, system installation and debugging, specifically including scraper test model building, perception layer sensor arrangement, adjusting sensor position, deploying wireless transmission unit and Internet of Things gateway, infrared sensor signal sampling, data fusion;

[0019] S2, test process, specifically includes device installation and parameter debugging, adjusting model chain speed, artificially manufacturing faults, adjusting scraper chain speed, simulating fault test;

[0020] S3, field environment test and improvement, specifically includes field installation preparation, field test content, system optimization.

[0021] S4, using algorithm to track and predict the speed of the scraper.

[0022] According to the above technical scheme, the S1 is specifically:

[0023] The scraper test model building stage, according to the proportion, builds the edge double chain scraper test model, simulates the actual running environment, the model needs to restore the core structure of the scraper, including the head sprocket, the tail sprocket, the chain on both sides, the scraper, the middle groove and the driving device, focusing on simulating the meshing relationship of chain and sprocket, the motion trajectory of scraper and the common fault occurrence scene, such as chain tightness difference, chain wheel tooth wear;

[0024] The perception layer sensor is arranged, a 200 million pixel high-definition camera is installed above the head sprocket of the scraper, the frame rate is 30fps, and the meshing area of the chain and the sprocket is shot; an infrared sensor is installed every 5 meters at the tail and the middle groove, the detection distance is 0.5-2 meters, the accuracy is ±1mm, and the camera is connected to the Internet of Things gateway through Ethernet, the transmission rate is 100Mbps, the infrared sensor is transmitted through the LoRa wireless module, the transmission distance is 500 meters, the gateway collects data and then transmits it to the data analysis layer industrial host through optical fiber;

[0025] Adjust the position of the sensor to ensure that it can accurately collect the chain running signal, the distance between the probe and the chain is preferably 0.5-1m, and the support is adjusted and fixed by the iron stand;

[0026] Deploy a wireless transmission unit and an Internet of Things gateway to ensure stable communication between the perception layer and the data processing layer;

[0027] The industrial host uses Intel Core i7 processor, runs an image recognition model based on AI machine vision, and analyzes the chain offset in real time. When the unilateral offset is greater than or equal to 5mm, it is determined that the tooth jumps;

[0028] The signal processing unit samples the infrared sensor signal at a sampling rate of 1kHz, calculates the time difference of the signals on both sides, and determines that the chain tightness is inconsistent when the difference is greater than or equal to 10ms, which has the potential risk of chain jump;

[0029] Fuse two types of data, output the fault type through the trained neural network model, such as unilateral chain jump and bilateral chain break.

[0030] According to the above technical scheme, the S2 is specifically:

[0031] Device installation and parameter adjustment: install the packaged monitoring device on the test model, ensure that the infrared sensor is aligned with the chain movement track, the displacement sensor is attached to the side of the middle groove, connect the power supply and the upper computer, including a notebook computer, install the data acquisition software, adjust the sensor sampling frequency, initially set to 1 kHz, and the wireless transmission baud rate to 9600 bps parameters;

[0032] Adjust the chain speed of the model, including 0.5 m / s, 1.0 m / s, 1.5 m / s, collect the pulse signals of the infrared sensors on both sides under normal operating conditions, record the pulse frequency, period and time difference of the signals on both sides as the standard pulse template;

[0033] Artificially create single-sided tooth skipping, adjust the deviation of the sprocket tooth and the chain pitch, loosen the single-sided chain tension, break the chain, disconnect the single-sided and double-sided chain faults, collect the pulse signals under the corresponding state, such as pulse missing, frequency mutation and displacement difference data, such as single-sided displacement exceeding 5 mm, and establish a fault feature database;

[0034] Adjust the chain speed of the scraper, ranging from 1 to 3 m / s, collect pulse signals and displacement data under normal operating conditions as reference data;

[0035] Simulation fault test phase: single fault test, simulate single-sided tooth skipping, single-sided chain skipping, single-sided chain breaking, double-sided chain skipping and double-sided chain breaking on the model, each fault is tested 30 times, record the alarm response time of the device, target ≤ 1 second and accuracy target ≥ 85%, optimize the program logic for false positives and false negatives, such as adding signal verification for 3 consecutive periods;

[0036] Composite scene test, simulate complex working conditions, such as chain speed fluctuation + slight single-sided tooth skipping, test the stability of the device under multiple interference factors; simulate fault recovery process, such as manual reset after chain skipping, verify whether the device can automatically recover to normal monitoring state.

[0037] According to the above technical scheme, the S3 is specifically:

[0038] On-site installation preparation, select five scraper machines with high fault rates in the lump coal workshop of the No. 1 mine coal preparation plant as test objects, adjust the sensor installation position according to the on-site space layout, such as 30 cm above the head sprocket, use explosion-proof bracket fixing device to adapt to the dust environment of the coal preparation plant, connect the workshop industrial power supply and wireless Internet of Things gateway, and meet the wireless network requirements;

[0039] On-site test content, continuous operation for 30 days, record the running stability of the device under actual working conditions, such as fault-free running time and data transmission success rate, target ≥ 99%;

[0040] The fault recognition accuracy in the actual scene is calculated by comparing the device alarm results with the manual inspection results, and the target is greater than or equal to 90%.

[0041] The two-stage alarm mechanism is tested: whether the on-site sound and light alarm is clear and distinguishable, and whether the man-machine interface of the duty room can accurately display the fault position and type.

[0042] System optimization: according to the test results on site, the device is optimized as follows: enhance the dustproof performance of the sensor, add a dust cover; optimize the wireless communication protocol, such as adding a data retransmission mechanism to solve the signal interference problem; update the data analysis host program based on the field fault data to improve the recognition ability of the slight skip tooth edge fault.

[0043] The S4 is specifically:

[0044] The scraper recognition model outputs the target frame (xyxy), category and confidence of each frame; then enters the polygon region filtering stage, and only retains the target entering the specified ROI (region of interest); ByteTrack realizes data association by combining the track of the last frame with the detection results of this frame, and assigns a stable tracking ID to each target; then the detection results are subjected to lightweight Kalman filter smoothing processing to reduce jitter; the displacement of the center of the same TrackerID target in adjacent two frames is calculated to obtain the pixel speed, and then the square root of the area ratio is used for perspective correction, and finally the speed unit is meter / second; one-dimensional Kalman filter is used for time smoothing processing of the speed, and the speed of each target in the last N seconds is recorded in its historical data; the historical speed is periodically (every N seconds) statistically calculated EMA exponential weighted average to calculate the average speed of the scraper.

[0045] In the S4, the ByteTrack tracking technology uses a two-stage data association method to realize efficient object tracking. First, high-confidence detection results are matched with existing tracks to ensure high-precision tracking. For those tracks that cannot be successfully matched, low-confidence detection results are used for supplementary tracking to reduce missed detection. The intersection over union (IoU) is used as the matching metric, and the formula is: IoU(B1,B2)=|B1∩B2| / |B1∪B2|, where B1 and B2 represent two detection boxes, |B1∩B2| represents the intersection area of the two boxes, and |B1∪B2| represents the union area of the two boxes. The Hungarian algorithm is used for optimal matching assignment, and Kalman filter is introduced for position prediction to further improve the accuracy and stability of the matching. This comprehensive matching method ensures the robustness and efficiency of the tracking process.

[0046] Compared with the prior art, the present application has the beneficial effects that: the present application realizes accurate identification of all types of faults such as unilateral tooth jumping, unilateral chain jumping, unilateral chain breaking, bilateral chain jumping and bilateral chain breaking, filling the technical blank of domestic fine fault monitoring of unilateral tooth jumping and chain jumping of the edge double-chain scraper. BRIEF DESCRIPTION OF DRAWINGS

[0047] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, illustrate the present application and explain the technical scheme of the present application, and do not constitute a limitation on the present application. In the drawings:

[0048] Figure 1 is a schematic diagram of the overall module structure of the present application. DETAILED DESCRIPTION

[0049] The technical scheme in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0050] Please refer to Figure 1 The present application provides a technical scheme: a new edge double-chain scraper chain jumping monitoring and alarm system, which comprises a starlight level camera, a perception layer, a data analysis layer, a data processing layer and an application layer. The starlight level camera continuously shoots video images at a set frame rate and 4 million pixels, and transmits image data to a signal processing module in real time. The perception layer is used to capture small displacement changes when the chain is running. The data analysis layer is used to determine the fault type through a pre-trained fault classification model. The data processing layer uses an AI model for real-time inference analysis. The application layer is used for sound and light alarm.

[0051] The starlight level camera continuously shoots video images at a set frame rate, not less than 25fps, 4 million pixels, and transmits image data to a signal processing module in real time. Since it can provide color images in low light conditions, compared with traditional black and white monitoring pictures, it can provide more detailed information for subsequent image analysis, which helps to more accurately determine whether the chain jumps, whether the scraper has abnormal displacement, etc.

[0052] The perception layer is an infrared sensor module, two high-precision laser infrared displacement sensors are selected and installed above the chains on both sides of the scraper, the measurement range covers the displacement interval from normal operation of the chain to the occurrence of the chain jump, such as 0-50mm, the resolution reaches 0.01mm, which ensures that the small displacement change can be captured, the sensor uses a modulated laser technology, and has strong anti-environmental light interference ability, even in the presence of dust and light changes in the working environment of the scraper, the stable measurement accuracy can still be maintained, the sensor continuously emits a laser beam to the edge of the chain, the laser beam is reflected by the chain and received by the receiving end, the sensor calculates the real-time distance between the chain and the sensor according to the time difference and phase difference between the emission and reception, when the chain is in normal operation, the distance value is stable in the preset normal range, such as 10±0.5mm, the change of the reflected distance of the chain is detected to output on-off signals and high-low level signals, and the loosening, deviation and breakage of the chain are sensed, such as signal mutation when the chain is blocked;

[0053] The data analysis layer includes an industrial host computer, the industrial host computer includes an image recognition unit, a signal processing unit and a machine learning classification algorithm model;

[0054] The data processing layer selects an industrial AI board with high integration and strong computing power, the AI board is equipped with a high-performance neural network processor and has strong parallel computing capability, can quickly process massive data collected by multiple sensors, at the same time, the AI board has rich interface resources, including multiple GPIO interfaces, UART interfaces, SPI interfaces and Ethernet interfaces, etc., can meet the connection requirements of other components such as sensor modules, signal processing modules and alarm modules, and has good expansibility, facilitating the upgrading of subsequent system functions;

[0055] The application layer includes a field alarm module: installed near the scraper, adopts an audible and visual alarm mode, and timely prompts the on-site staff when a fault occurs; a remote monitoring terminal: set in the duty room, displays fault information through a man-machine interface, integrates an audible and visual alarm module, and is used for field alarm, a red LED lamp and a buzzer are triggered immediately when a fault is detected; integrates a data storage module, such as an SD card, performs audible and visual alarm, and realizes remote monitoring and early warning; a data storage and management module: stores the original data of the perception layer, image, sensor signal, fault judgment result of the data analysis layer, fault type, time, position and control layer execution record; supports historical data query and trend analysis, and provides data support for fault tracing and equipment maintenance optimization;

[0056] The functions of the industrial host computer and the machine learning classification algorithm model include:

[0057] Image recognition: a rotating target detection model based on YOLO11-0BB, a large amount of normal scraper, chain and broken chain image data are labeled, the model directly outputs four vertex coordinates of the scraper in the image, an inclination angle and a broken chain state;

[0058] Signal analysis: filter and denoise the infrared sensor signal, calculate the time difference and amplitude change of the signals on both sides of the chain, and judge the tightness difference and breakage of the chain;

[0059] Comprehensive judgment: fuse image features and sensor signals, and determine fault types through a pre-trained fault classification model, such as a support vector machine, including single-sided tooth skipping / link skipping / broken chain, double-sided link skipping / broken chain, and generate fault levels and processing instructions;

[0060] The AI board is built-in with a trained link skipping recognition model. Based on a large amount of sensor data from the scraper under normal operation and link skipping state, including distance data from infrared displacement sensors, vibration signals from vibration sensors, and image data from starlight level cameras, the model is trained. After receiving various data transmitted by the signal processing module, the AI board uses the model for real-time inference analysis. Through deep learning and pattern recognition of data, it can more accurately identify the characteristics of link skipping faults and effectively distinguish between minor fluctuations in normal operation and real link skipping abnormalities, further improving the accuracy of link skipping judgment.

[0061] The working method of the system includes:

[0062] S1, system installation and debugging, specifically including scraper test model building, perception layer sensor arrangement, adjusting sensor position, deploying wireless transmission unit and Internet of Things gateway, infrared sensor signal sampling, and data fusion;

[0063] S2, test process, specifically including device installation and parameter adjustment, adjusting model chain speed, artificially creating faults, adjusting scraper chain speed, and simulating fault testing;

[0064] S3, on-site environment testing and improvement, specifically including on-site installation preparation, on-site testing content, and system optimization;

[0065] S4, using algorithms to track and predict the speed of the scraper.

[0066] S1 specifically includes:

[0067] In the scraper test model building stage, a side-by-side chain scraper test model is built according to the scale, simulating the actual operating environment. The model needs to reproduce the core structure of the scraper, including the head sprocket, tail sprocket, two side chains, scraper, middle slot, and driving device. It focuses on simulating the meshing relationship between the chain and the sprocket, the motion trajectory of the scraper, and common fault scenarios, such as chain tightness difference and chain sprocket tooth wear;

[0068] The perception layer sensor arrangement is installed with a 200-megapixel high-definition camera directly above the chain wheel of the sprocket machine head, with a frame rate of 30 fps, and the camera shoots the meshing area of the chain and the sprocket wheel; an infrared sensor is installed every 5 meters at the tail and the middle groove, with a detection distance of 0.5-2 meters and an accuracy of ±1 mm, and the sensor is symmetrically arranged above the chain on both sides at a distance of 10 cm, and the camera is connected to the Internet of Things gateway through Ethernet with a transmission rate of 100 Mbps, and the infrared sensor is connected to the gateway through the LoRa wireless module with a transmission distance of 500 meters to send signals, and the gateway aggregates data and transmits it to the data analysis layer industrial host through optical fiber;

[0069] Adjust the sensor position to ensure that it can accurately collect chain running signals, and the distance between the probe and the chain is preferably 0.5-1 m, which is adjusted and fixed by a supporting iron stand;

[0070] Deploy a wireless transmission unit and an Internet of Things gateway to ensure stable communication between the perception layer and the data processing layer;

[0071] The industrial host uses an Intel Core i7 processor to run an image recognition model based on AI machine vision to analyze the chain offset in real time, and determines that the chain is jumping when the unilateral offset is greater than or equal to 5 mm;

[0072] The signal processing unit samples the infrared sensor signal at a sampling rate of 1 kHz, calculates the time difference between the two signals, and determines that the chain is tight when the difference is greater than or equal to 10 ms, indicating a potential chain jumping risk;

[0073] Fusion of two types of data, through the trained neural network model output fault type, such as unilateral chain jump, double chain break;

[0074] S2 is specifically:

[0075] Device installation and parameter adjustment: install the packaged monitoring device on the test model, ensure that the infrared sensor is aligned with the chain motion track, the displacement sensor is attached to the side of the middle groove, connect the power supply and the upper computer including a notebook computer, install the data acquisition software, adjust the sensor sampling frequency, and set the initial parameters to 1 kHz and the wireless transmission baud rate to 9600 bps;

[0076] Adjust the chain speed of the model, including 0.5 m / s, 1.0 m / s, and 1.5 m / s, collect the pulse signals of the two infrared sensors under normal operating conditions, record the pulse frequency, period, and time difference between the two signals, and use them as standard pulse templates;

[0077] Artificially create unilateral chain jump, adjust the chain pitch deviation of the sprocket wheel and the chain, unilateral chain jump, loosen the unilateral chain tension, chain break, and disconnect the unilateral and bilateral chain faults, collect the pulse signals under the corresponding conditions, such as pulse loss, frequency mutation, and displacement difference data, such as unilateral displacement exceeding 5 mm, and establish a fault feature database;

[0078] Adjust the speed of the scraper chain, ranging from 1 to 3 m / s, collect pulse signals and displacement data under normal operating conditions as baseline data;

[0079] Simulation fault test phase: single fault test, simulate single-sided tooth skipping, single-sided chain skipping, single-sided chain breaking, double-sided chain skipping, and double-sided chain breaking on the model, each fault is repeated 30 times, record the alarm response time of the device, the target is ≤1 second and the accuracy target is ≥85%, optimize the program logic for false positives and false negatives, such as adding a signal verification for 3 consecutive cycles;

[0080] Composite scene test, simulate complex working conditions such as chain speed fluctuation + slight single-sided tooth skipping, test the stability of the device under multiple interference factors; simulate fault recovery process, such as manual reset after chain skipping, verify whether the device can automatically recover to normal monitoring state;

[0081] S3 is specifically:

[0082] On-site installation preparation, select five scraper machines with high fault rates in the lump coal workshop of the mine coal preparation plant as test objects, adjust the sensor installation position according to the on-site space layout, such as 30 cm above the head sprocket, use explosion-proof support fixtures to adapt to the dust environment in the coal preparation plant, connect to the workshop industrial power supply and wireless Internet of Things gateway, and meet the requirements of wireless network deployment;

[0083] On-site test content, continuous operation for 30 days, record the running stability of the device under actual working conditions, such as fault-free running time and data transmission success rate, target ≥99%;

[0084] Compare the device alarm results with the manual inspection results, combine the actual faults and artificial simulation faults, and calculate the fault identification accuracy rate in the actual scene, target ≥90%;

[0085] Test the two-level alarm mechanism: whether the on-site sound and light alarm is clear and distinguishable, and whether the human-machine interface in the duty room can accurately display the fault position and type;

[0086] System optimization, according to the on-site test results, optimize the device as follows: enhance the dustproof performance of the sensor, add a dust cover; optimize the wireless communication protocol, such as adding a data retransmission mechanism to solve the signal interference problem; update the data analysis host program based on on-site fault data to improve the identification ability of slight tooth skipping edge faults.

[0087] S4 is specifically:

[0088] The scraper recognition model outputs the target frame (xyxy), class and confidence; then enters the polygon region filtering stage, only the target entering the specified ROI (region of interest) is retained; ByteTrack realizes data association by combining the track of the last frame with the detection result of the current frame, and assigns a stable tracking ID to each target; then the detection result is smoothed by light Kalman filtering to reduce jitter; the displacement of the same TrackerID target center in adjacent two frames is calculated to obtain the pixel speed, and then the speed is corrected by the square root of the area ratio to obtain the speed unit in meters per second; one-dimensional Kalman filtering is used to time smooth the speed, and the speed of each target in the last N seconds is recorded in its historical data; the historical speed is periodically (every N seconds) calculated by EMA exponential weighted average to calculate the average speed of the scraper.

[0089] In S4, the ByteTrack tracking technology uses a two-stage data association method to achieve efficient object tracking. First, high-confidence detection results are matched with existing tracks to ensure high-precision tracking. For those tracks that fail to match, low-confidence detection results are used for supplementary tracking to reduce missed detection. The intersection over union (IoU) is used as the matching metric, with the formula: IoU(B1, B2) = |B1∩B2| / |B1∪B2|, where B1 and B2 represent two detection boxes, |B1∩B2| represents the intersection area of the two boxes, and |B1∪B2| represents the union area of the two boxes. The Hungarian algorithm is used for optimal matching assignment, and Kalman filtering is introduced for position prediction to further improve the accuracy and stability of the matching. This comprehensive matching method ensures the robustness and efficiency of the tracking process.

[0090] One-dimensional Kalman filter (speed smoothing) prediction formula: x^ k | k-1 = x^ k-1 , P k | k-1 = P k-1 + Q; where x^ k | k-1 : predicted speed value at time k, x^ k-1 : last speed estimate, P l | k-1 : prediction covariance (uncertainty), P k-1 : last covariance, Q: process noise variance, describing the uncertainty of the model.

[0091] Correction formula: K k = P k | k-1 / (P k |k-1 + R), x^ k = x^ k k-1 + K k (z k - x^ k k-1

[0092] P k = (1 - K k ) P k k-1 where K k : Kalman gain, determines the weight of prediction and observation, z k : observed velocity (directly calculated from pixel displacement), x^ k : corrected velocity estimate, R: measurement noise variance, describes the uncertainty of observation, P k : corrected covariance.

[0093] Velocity estimation and perspective correction formula: pixel velocity: v_pix = d * f, d = sqrt((cx - cx' 2 + (cy - cy') 2 ), d: pixel displacement of the center point of two adjacent frames, f: frame rate (FPS), v_pix: pixel velocity (pixel / s), area and depth relationship: A ∝ 1 / Z 2 → Z / Z0 ≈ sqrt(A0 / A), A: current frame target box area, A0: area of the reference frame (when the target first appears).

[0094] Z: current depth. Z0: reference depth. Meaning: the larger the box area, the closer the target, the smaller the area, the farther the target. Physical velocity: v = (v_pix / κ) * sqrt(A0 / A), where v_pix: pixel velocity. κ: pixel / meter conversion coefficient (pixel_per_meter). sqrt(A0 / A): perspective correction factor. v: physical velocity (m / s).

[0095] N-second velocity statistics method: linear weighting: v - = Σ(t * v_t) / Σt, where n: total number of samples (usually = FPS * N seconds). v_t: velocity at the t-th moment. t: weight (time sequence, the newer the weight, the greater). v - : weighted average velocity, which can better reflect the recent speed trend.

[0096] ​​​​The present application has: 1, fault identification comprehensiveness: can identify single-sided tooth skipping, single-sided chain skipping, single-sided chain breaking, double-sided chain skipping, double-sided chain breaking and other fault types, which makes up for the shortcomings of the prior art. 2, monitoring accuracy: combining vision and infrared sensing technology, covering fault types that cannot be identified by existing technologies such as single-sided tooth skipping, chain skipping and chain breaking, with a monitoring coverage of 100%; 3, timely response: through on-site and remote two-level alarm mechanisms, ensuring that faults are discovered and handled in the shortest time, reducing production downtime. 4, high degree of intelligence: realizing automatic data collection, transmission, analysis and alarm, reducing manual inspection workload and improving intelligent management level of coal preparation plants. Through multi-source data fusion and algorithm analysis, the fault judgment accuracy is greater than or equal to 90%, reducing false positives and false negatives; 5, good economy: resistant to dust, vibration and humid environment in coal preparation plants, suitable for long-term stable operation in industrial sites. Reducing equipment maintenance costs, prolonging the service life of the scraper, improving production efficiency, and having significant economic and social benefits.

[0097] The present application realizes accurate identification of all types of faults such as single-sided tooth skipping, single-sided chain skipping, single-sided chain breaking, double-sided chain skipping and double-sided chain breaking, filling the technical gap in domestic monitoring of subdivided faults such as single-sided tooth skipping and chain skipping of the double-chain scraper.

[0098] It should be noted that, in this text, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" and any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article and device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or elements inherent to such a process, method, article and device.

[0099] Finally, it should be noted that: the above only describes the preferred embodiments of the present application, and is not intended to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, and / or make equivalent replacements to some technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A novel chain skipping monitoring and alarm system for a double-chain scraper conveyor, characterized in that: It includes a starlight-level camera, a perception layer, a data analysis layer, a data processing layer, and an application layer. The starlight-level camera continuously captures video images at a set frame rate and 4 million pixels, and transmits the image data to the signal processing module in real time. The perception layer is used to capture minute displacement changes during chain operation. The data analysis layer is used to determine the fault type through a pre-trained fault classification model. The data processing layer uses an AI model for real-time reasoning and analysis. The application layer is used for audible and visual alarms.

2. The novel chain skipping monitoring and alarm system for a double-chain scraper conveyor according to claim 1, characterized in that: The starlight-level camera continuously captures video images at a set frame rate of no less than 25fps and 4 megapixels, and transmits the image data to the signal processing module in real time. The sensing layer is an infrared sensor module, using two high-precision laser infrared displacement sensors, which are respectively installed above the chains on both sides of the scraper conveyor. Their measurement range covers the displacement range that occurs when the chain jumps during normal operation, such as 0-50mm, with a resolution of 0.01mm, which can capture minute displacement changes. The sensors use modulated laser technology, continuously emitting laser beams towards the edge of the chain. The laser beams are reflected by the chain and received by the receiver. The sensors calculate the real-time distance between the chain and the sensor based on the time difference and phase difference between emission and reception. During normal operation, this distance value is stable within a preset normal range, such as 10±0.5mm. By detecting changes in the chain reflection distance, on / off signals and high / low level signals are output to sense chain looseness, offset, and breakage, such as sudden signal changes when obstructed. The data analysis layer includes an industrial host, which contains an image recognition unit, a signal processing unit, and a machine learning classification algorithm model. The data processing layer uses an industrial-grade AI board, which is equipped with a high-performance neural network processor and has powerful parallel computing capabilities. It can quickly process massive amounts of data collected by multiple source sensors. At the same time, it has rich interface resources, including multiple GPIO interfaces, UART interfaces, SPI interfaces and Ethernet interfaces. The application layer includes an on-site alarm module: installed near the scraper conveyor, using audible and visual alarms to promptly alert on-site personnel when a fault occurs; a remote monitoring terminal: located in the duty room, displaying fault information through a human-machine interface, integrating an audible and visual alarm module for on-site alarm use, with a red LED light and buzzer that triggers immediately when a fault is detected; and an integrated data storage module, such as an SD card, for audible and visual alarms, enabling remote monitoring and early warning. Data storage and management module: Stores raw data from the perception layer, images, sensor signals, fault judgment results from the data analysis layer, fault type, time, location, and control layer execution records; supports historical data query and trend analysis, providing data support for fault tracing and equipment maintenance optimization.

3. The novel chain skipping monitoring and alarm system for a double-chain scraper conveyor according to claim 2, characterized in that: The functions of the industrial host and machine learning classification algorithm model include: Image recognition: Based on the YOLO11-0BB rotating target detection model, a large amount of normal scraper, chain and broken chain image data are labeled. The model directly outputs the coordinates of the four vertices of the scraper in the image, the tilt angle, and whether the chain is broken. Signal analysis: The infrared sensor signal is filtered and noise-reduced, and the time difference and amplitude change of the signals on both sides of the chain are calculated to determine the difference in chain tension and breakage. Comprehensive judgment: By fusing image features and sensor signals, the fault type is determined through a pre-trained fault classification model, such as a support vector machine, including single-sided skipped tooth / chain skipped / chain broken, and double-sided skipped chain / chain broken, and fault level and processing instructions are generated.

4. The novel chain skipping monitoring and alarm system for a double-chain scraper conveyor according to claim 3, characterized in that: The AI ​​board has a built-in trained chain skipping recognition model. This model is trained based on a large amount of sensor data from the scraper conveyor during normal operation and chain skipping, including distance data from infrared displacement sensors, vibration signals from vibration sensors, and image data from starlight-level cameras. When it receives various types of data from the signal processing module, the AI ​​board uses this model to perform real-time reasoning and analysis.

5. A novel chain skipping monitoring and alarm system for a double-chain scraper conveyor according to claim 4, characterized in that: The working method of this system includes: S1. System installation and debugging, specifically including the construction of the scraper conveyor test model, the arrangement of sensors in the sensing layer, the adjustment of sensor positions, the deployment of wireless transmission units and IoT gateways, infrared sensor signal sampling, and data fusion; S2. The test process specifically includes device installation and parameter debugging, adjusting the model chain speed, artificially creating faults, adjusting the scraper conveyor chain speed, and simulating fault tests. S3. On-site environment testing and improvement, specifically including on-site installation preparation, on-site testing content, and system optimization; S4. Use algorithms to track and predict the speed of the scraper conveyor.

6. The novel chain skipping monitoring and alarm system for a double-chain scraper conveyor according to claim 5, characterized in that: Specifically, S1 is: During the scraper conveyor test model building phase, a side double-chain scraper conveyor test model was built to scale to simulate the actual operating environment. The model needs to restore the core structure of the scraper conveyor, including the head sprocket, tail sprocket, side chains, scraper, central groove and drive device. The model focuses on simulating the meshing relationship between the chain and sprocket, the scraper movement trajectory and common fault scenarios, such as chain tension differences and sprocket tooth wear. The sensor arrangement in the perception layer includes a 2-megapixel high-definition camera installed directly above the sprocket at the head of the scraper conveyor, operating at 30fps, to capture images of the chain-sprocket meshing area; and an infrared sensor installed every 5 meters at the tail and middle trough, with a detection distance of 0.5-2 meters and an accuracy of ±1mm, symmetrically arranged 10cm above the chains on both sides. The camera is connected to an IoT gateway via Ethernet with a transmission rate of 100Mbps, and the infrared sensors transmit signals via a LoRa wireless module with a transmission distance of 500 meters. After the gateway aggregates the data, it transmits it to the industrial host in the data analysis layer via optical fiber. Adjust the sensor position to ensure it can accurately collect chain running signals. The distance between the probe and the chain should be 0.5-1m. Fix it with a support and adjustment iron frame. Deploy wireless transmission units and IoT gateways to ensure stable communication between the perception layer and the data processing layer; The industrial host uses an Intel Core i7 processor and runs an AI machine vision-based image recognition model to analyze chain offset in real time. When the offset on one side is ≥5mm, it is determined to be a skipped tooth. The signal processing unit samples the infrared sensor signal at a sampling rate of 1kHz and calculates the time difference between the signals on both sides. When the difference is ≥10ms, it is determined that the chain tension is inconsistent, which poses a potential risk of chain skipping. By fusing the two types of data, a trained neural network model is used to output the fault type, such as single-sided skipped chain or double-sided broken chain.

7. A novel chain skipping monitoring and alarm system for a double-chain scraper conveyor according to claim 6, characterized in that: Specifically, S2 is: Device installation and parameter debugging: Install the packaged monitoring device on the test model, ensure that the infrared sensor is aligned with the chain movement trajectory, the displacement sensor is attached to the side of the middle groove, connect the power supply and host computer, including a laptop, install the data acquisition software, and debug the sensor sampling frequency, initially set to 1kHz and wireless transmission baud rate of 9600bps. Adjust the speed of the model chain to 0.5m / s, 1.0m / s, and 1.5m / s. Under normal operating conditions, collect pulse signals from the infrared sensors on both sides, record the pulse frequency, period, and time difference between the signals on both sides, and use them as a standard pulse template. Artificially induce unilateral tooth skipping, adjust the sprocket tooth and chain pitch deviation, unilateral chain skipping, relax unilateral chain tension, break the chain, disconnect unilateral and double-sided chain faults, collect pulse signals under the corresponding conditions, such as pulse loss, frequency change and displacement difference data, such as unilateral displacement exceeding 5mm, and establish a fault feature database. Adjust the scraper conveyor chain speed, ranging from 1-3 m / s, and collect pulse signals and displacement data under normal operating conditions as reference data; Simulated fault testing phase: Single fault test, simulating 5 faults on the model, including single-sided tooth skipping, single-sided chain skipping, single-sided chain breakage, double-sided chain skipping, and double-sided chain breakage. Each fault is repeated 30 times. The alarm response time of the device is recorded, with a target of ≤1 second and an accuracy target of ≥85%. The program logic is optimized for false alarms and missed alarms, such as adding signal verification for 3 consecutive cycles. Composite scenario testing simulates complex working conditions, such as chain speed fluctuations and slight unilateral tooth skipping, to test the stability of the device under multiple interference factors. Simulate the fault recovery process, such as manual reset after a chain jump, to verify whether the device can automatically return to normal monitoring status.

8. A novel chain skipping monitoring and alarm system for a double-chain scraper conveyor according to claim 7, characterized in that: Specifically, S3 is: For on-site installation preparation, five scraper conveyors with frequent failures in the lump coal workshop of the No. 1 Coal Preparation Plant were selected as test objects. The sensor installation position was adjusted according to the on-site spatial layout, such as 30cm above the machine head sprocket, and an explosion-proof bracket was used for fixing the sensor. The sensor was connected to the workshop's industrial power supply and wireless IoT gateway. The on-site testing involves continuous operation for 30 days, recording the device's operational stability under actual working conditions, such as fault-free operation time and data transmission success rate, with a target of ≥99%. By combining actual faults that occur in the workshop with manually simulated faults, comparing the alarm results of the equipment with the results of manual inspections, the accuracy rate of fault identification in actual scenarios is statistically analyzed, with a target of ≥90%. Test the two-level alarm mechanism: whether the on-site audible and visual alarms are clear and distinguishable, and whether the human-machine interface in the duty room can accurately display the location and type of the fault. System optimization: Based on the field test results, the following optimizations were made to the device: enhance the dustproof performance of the sensors by adding dust covers; optimize the wireless communication protocol, such as adding a data retransmission mechanism to solve the signal interference problem; update the data analysis host program based on field fault data to improve the ability to identify minor skipped tooth edge faults.

9. A novel chain skipping monitoring and alarm system for a double-chain scraper conveyor according to claim 8, characterized in that: Specifically, S4 is: The scraper conveyor recognition model outputs the target bounding box (xyxy), category, and confidence score for each frame. It then enters a polygon region filtering stage, retaining only targets that have entered a specified ROI (Region of Interest). ByteTrack associates the data by combining the trajectory from the previous frame with the detection results of the current frame and assigns a stable tracking ID to each target. Next, a lightweight Kalman filter is applied to the detection results for smoothing to reduce jitter. Pixel velocity is calculated using the displacement of the center of the same TrackerID target in two adjacent frames, and perspective correction is performed using the square root of the area ratio, ultimately yielding the velocity in meters per second. A one-dimensional Kalman filter is used for temporal smoothing of the velocity, and the velocity of each target in the most recent N seconds is recorded in its historical data. The average speed of the scraper conveyor is calculated by periodically (every N seconds) using a weighted average of historical speeds and EMA index.

10. A novel chain skipping monitoring and alarm system for a double-chain scraper conveyor according to claim 8, characterized in that: In S4, the ByteTrack tracking technology employs a two-stage data association method to achieve efficient object tracking. First, it matches high-confidence detection results with existing trajectories to ensure high-precision tracking. For trajectories that fail to match, it further uses low-confidence detection results for supplementary tracking. The Intersection over Union (IoU) ratio is used as the matching metric, and its calculation formula is: IoU(B1,B2)=|B1∩B2| / |B1∪B2|, where B1 and B2 represent two detection boxes, |B1∩B2| represents the intersection area of ​​the two boxes, and |B1∪B2| represents the union area of ​​the two boxes. The Hungarian algorithm is combined for optimal matching allocation, and Kalman filtering is introduced for position prediction.

Images