Scraper fault detection device and method based on machine vision

Abstract

The invention discloses a scraper fault detection device and method based on machine vision. The scraper fault detection device comprises a camera, an AI edge reasoning host, a display and an audible and visual alarm; the scraper conveyor fault detection method comprises the following steps that S1, a camera collects an operation image of a scraper conveyor in real time; s2, the camera transmits a collected scraper operation image to an AI edge reasoning host through a mining network cable; s3, the AI edge reasoning host processes the scraper operation image through a built-in target detection method and transmits the processed scraper operation image to a display through a mining network cable for display; s4, the AI edge reasoning host analyzes and judges the processed operation image of the scraper through a built-in state analysis method, and when it is judged that chain breakage exists in the scraper, the next step of operation is executed; and S5, the displayer, the audible and visual alarm and the scraper controller conduct alarming and shutdown operation according to the judgment result.

Description

Technical Field

[0001] The present invention relates to the field of coal mine equipment detection, and in particular to a scraper fault detection device and method based on machine vision. Background Art

[0002] In modern industrial production systems, scrapers, as core equipment for material transportation, are widely used in various industries, including coal mining, chemical industry, and metallurgy. In coal mining, for example, scrapers undertake the critical task of transporting mined coal from the working face to the haulage tunnels, and their operating status directly affects the efficiency and continuity of coal mining. In chemical production, scrapers are responsible for transporting various chemical raw materials and finished products, ensuring the stable operation of the production process. Therefore, the stable operation of scrapers plays a vital role in the smooth operation of the entire industrial production process.

[0003] However, chain breakage is a common and potentially devastating problem in scraper operations. Once a chain breaks, production is forced to halt. According to statistics, each interruption caused by a chain break in a scraper can cost a company tens or even hundreds of thousands of yuan in economic losses. Chain breakage can also cause cascading damage to the equipment, such as scraper deformation and increased sprocket wear, increasing repair costs and maintenance difficulties. More seriously, chain breakage can lead to safety incidents, posing a serious threat to the health and safety of on-site workers. Broken chains can eject, causing casualties.

[0004] Currently, scraper chain break detection methods are mainly divided into two categories: contact and non-contact, but both have obvious drawbacks. Contact detection methods, such as chain tension detection, tongue plate detection, and sprocket stress detection, are prone to wear and tear due to the harsh working environment of scraper conveyors, which are constantly affected by factors such as impact and friction from coal gangue. These detection sensors have poor resistance to harsh environments, making it difficult to accurately detect chain breakage under complex working conditions, and reliability cannot be guaranteed. Among non-contact detection methods, the sprocket speed detection method determines the chain status by measuring the difference in speed between the two sprockets in real time. However, its sensitivity is low and it cannot detect minor chain faults in a timely manner. Moreover, this method lacks applicability for front and rear dual-drive scraper conveyors. The Hall sensor detection method has problems such as inconvenient sensor installation and complex structure. In actual application, it is easily damaged or worn, affecting the detection effect.

[0005] Therefore, a new and more reliable detection technology is urgently needed to overcome the defects of existing methods, achieve accurate and efficient detection of scraper chain breakage failures, and ensure the safety and stability of industrial production. Summary of the Invention

[0006] The purpose of the present invention is to provide a scraper fault detection device and method based on machine vision to improve the reliability and practicality of scraper chain break detection in order to solve the above problems.

[0007] In order to achieve the above object, the technical solution of the present invention is:

[0008] A machine vision-based scraper fault detection device includes a camera for image acquisition, an AI edge inference host for inference based on the image, a display for displaying the image, and an audible and visual alarm for performing an alarm operation; the camera is fixedly connected to the tail position of the scraper through a bracket, the camera is connected to the AI ​​edge inference host through a mining network cable, and the AI ​​edge inference host is connected to the display, the audible and visual alarm, and the scraper controller through the mining network cable.

[0009] Furthermore, the scraper fault detection device also includes a power supply module for power supply. The power supply module is built into the AI ​​edge inference host. The power supply module is connected to the camera, display, and sound and light alarm through a mining cable and supplies power to them.

[0010] A method for detecting scraper failure based on machine vision comprises the following steps:

[0011] S1, the camera collects the running image of the scraper in real time;

[0012] S2. The camera transmits the collected scraper operation images to the AI ​​edge inference host via the mining network cable;

[0013] S3, the AI ​​edge inference host processes the scraper operation image through the built-in target detection method and transmits the processed scraper operation image to the monitor through the mining network cable for display;

[0014] S4: The AI ​​edge inference host analyzes and judges the processed scraper operation image through the built-in state analysis method to determine whether the scraper has a chain break. If it is determined that the scraper has no chain break, steps S1 to S4 are repeated to achieve cyclic monitoring of the scraper. If it is determined that the scraper has a chain break, the next step is entered.

[0015] S5. The AI ​​edge inference host sends a disconnection message to the display through the mining network cable. The display pops up an alarm and displays the scraper failure information. The AI ​​edge inference host triggers the sound and light alarm through the mining network cable to start and perform the sound and light alarm operation. The AI ​​edge inference host sends a shutdown command to the scraper controller through the mining network cable, and the scraper controller performs the shutdown protection operation.

[0016] Furthermore, in step S3, the target detection method built into the AI ​​edge inference host is based on the YOLOv11 method, the Conv module in the YOLOv11 method is replaced by the ARConv module, the C2PSA module in the YOLOv11 method is replaced by the FSTA module, and the SE attention mechanism module is added between the Backbone module and the Neck module in the YOLOv11 method.

[0017] Furthermore, the step S4 specifically includes the following steps:

[0018] S41, perform a rectangular frame extraction operation on each scraper in the processed scraper machine running image, and obtain a rectangular frame set B={b i} n i＝1 , in the rectangular box set, each rectangular box b i =(x i ,y i ,w i ,h i ,θ i ); where (x i ,y i ) is the center coordinate of the rectangular frame; w i is the width of the rectangular frame; h i is the height of the rectangular frame; θ i is the original rotation angle;

[0019] S42, define and calculate the angle α corresponding to the long side direction of each rectangular frame i ; Its calculation formula is:

[0020]

[0021] S43, angle α i After normalization, the calculation formula is:

[0022]

[0023] Among them, α i ′∈[0,180°);

[0024] S44, filtering out invalid rectangular frames in the processed scraper running image to obtain a valid rectangular frame set B valid , the valid rectangular box set B valid The corresponding valid angle set is A={α i ′} bi∈Bvalid If |A|<2, the AI ​​edge inference host determines that the scraper does not have a broken link; if |A|≥2, proceed to the next step;

[0025] S45. Calculate the reference angle μ using the following formula:

[0026]

[0027] S46. Calculate the absolute deviation between each angle in the valid angle set A and the reference angle μ. The calculation formula is:

[0028] d k =|α k ′-μ∣,k=1,2,…,∣A∣;

[0029] S47. Set the angle deviation threshold T. If the maximum value of the absolute deviation calculated in step S46 exceeds the angle deviation threshold T, the AI ​​edge inference host determines that the scraper has a broken chain; otherwise, the AI ​​edge inference host determines that the scraper does not have a broken chain.

[0030] Compared with the prior art, the present invention has the following advantages and positive effects:

[0031] 1. This invention uses a camera for image acquisition, eliminating the need for contact sensors installed on chains or sprockets. This completely avoids sensor wear caused by coal gangue impact and friction, addressing the core drawbacks of traditional contact detection methods, such as rapid sensitivity loss and low reliability. The entire device is adaptable to complex industrial environments such as coal mining and chemical industries, characterized by high dust, strong vibration, and humidity, significantly extending its service life.

[0032] 2. The AI ​​edge inference host in the present invention integrates a power supply module and a high-speed data processing unit to realize localized closed-loop processing of image acquisition, method reasoning, and alarm control. Compared with cloud-based detection solutions, it eliminates network delays and data transmission risks, shortens fault response time to milliseconds, and meets the strict requirements of industrial control for data security and privacy protection.

[0033] 3. The present invention uses standardized mining cables for connection, and the camera bracket can be flexibly installed at the tail of the scraper. It only needs to ensure that the field of view covers at least two scrapers, and there is no need for structural modification of the existing equipment. Compared with traditional non-contact detection methods such as Hall sensors, it avoids complex mechanical installation and calibration processes, significantly reducing the difficulty of project implementation and maintenance costs.

[0034] 4. This invention performs image processing based on the improved target detection model of YOLOv11. It dynamically matches the slender rotating target characteristics of the scraper chain and scraper through the ARConv adaptive rectangular convolution kernel. Compared with the traditional horizontal rectangular frame detection method, it can accurately capture the edge texture characteristics of inclined targets and reduce the missed detection rate. At the same time, combined with the SE channel attention module, it effectively suppresses complex background noise and improves the target differentiation of low-quality images. The detection accuracy can maintain a high level under typical lighting conditions in coal mines.

[0035] 5. The present invention introduces the FSTA frequency-domain spatiotemporal attention module to replace the traditional spatial attention mechanism, and utilizes the rotational invariance and spatiotemporal noise suppression capabilities of frequency-domain features to solve the problem of early chain break detection caused by the small difference in sprocket speeds in dual-drive scrapers.

[0036] 6. The state analysis method of the present invention converts the chain running state into a quantifiable angle deviation index through angle standardization and deviation calculation, thereby improving the accuracy of its detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 It is a system connection framework diagram of the present invention;

[0039] Figure 2 Flow chart of the detection method of the present invention;

[0040] Figure 3 This is the improved target detection model diagram of YOLOv11;

[0041] Figure 4 It is a structural diagram of the ARConv module;

[0042] Figure 5 It is a structural diagram of the FSTA module;

[0043] Figure 6 This is a structural diagram of the SE attention mechanism module;

[0044] Figure 7 It is the upper half of the logic flow chart of the state analysis method;

[0045] Figure 8 It is the lower half of the logic flow chart of the state analysis method;

[0046] Figure 9 It is a detection effect diagram of the present invention. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts, any modifications, equivalent replacements, improvements, etc., shall be included in the scope of protection of the present invention.

[0048] The present invention discloses a scraper conveyor fault detection device and method based on machine vision, which makes targeted breakthroughs around the defects of existing detection methods, aiming to solve problems such as easy wear of contact detection and low sensitivity of non-contact detection, and comprehensively improve the reliability and practicality of scraper conveyor chain break detection.

[0049] like Figure 1 As shown, the scraper fault detection device based on machine vision includes but is not limited to a mine intrinsically safe camera for image acquisition, a mine intrinsically safe sound and light alarm, a mine explosion-proof and intrinsically safe AI edge inference host, a mine intrinsically safe display, a bracket for installing a mine intrinsically safe camera, and cables for connecting various devices.

[0050] The bracket for installing the intrinsically safe camera for mining is installed at the tail position of the machine, and the intrinsically safe camera for mining is installed on the bracket, which should ensure that the camera's field of view can see at least two scrapers.

[0051] The mine-use flameproof and intrinsically safe AI edge inference host is internally equipped with an intrinsically safe power supply module, which supplies power to the mine-use intrinsically safe camera, mine-use intrinsically safe sound and light alarm, and mine-use intrinsically safe display through a mine cable.

[0052] The mine-used intrinsically safe camera communicates signals with the mine-used flameproof and intrinsically safe AI edge inference host through the mine network cable, transmits the original image to the mine-used flameproof and intrinsically safe AI edge inference host for inference, and the mine-used flameproof and intrinsically safe AI edge inference host transmits the image after inference and the image before inference to the mine-used intrinsically safe display through the mine network cable for display.

[0053] The mining cable used to connect the mining explosion-proof and intrinsically safe AI edge inference host and the mining intrinsically safe camera and mining intrinsically safe display should include at least two cores for positive and negative power supply.

[0054] The mining cable 2 used to connect the mining explosion-proof and intrinsically safe AI edge inference host and the mining intrinsically safe sound and light alarm should contain at least four cores, two of which are used for positive and negative power supply, and the other two cores are used for 485 alarm signal communication.

[0055] like Figure 2As shown, the workflow of the scraper fault detection device should include the following steps:

[0056] Step 1: Image acquisition and data transmission

[0057] 1. The mining intrinsically safe camera collects real-time images of the scraper chain operation (covering at least two scrapers);

[0058] 2. The collected original images are transmitted to the mine-used explosion-proof and intrinsically safe AI edge inference host via a mining network cable.

[0059] Step 2: Image processing and result output

[0060] 1. Process the image using the built-in method of the AI ​​edge inference host;

[0061] 2. The original image before processing and the annotated image after processing are transmitted to the mine intrinsically safe display for display.

[0062] Step 3: Determine if a link is broken

[0063] The AI ​​edge inference host determines whether there is a link break based on the detection results:

[0064] If normal, continue cyclic monitoring;

[0065] If abnormal (determined to be a broken link), proceed to the next step.

[0066] Step 4: Alarm and control response

[0067] 1. The display pops up an alarm and displays fault information;

[0068] 2. The AI ​​edge inference host triggers the mine-used intrinsically safe sound and light alarm through the 485 signal, which sounds an alarm;

[0069] 3. The AI ​​edge inference host simultaneously sends a shutdown command to the scraper equipment controller to perform shutdown protection operations.

[0070] The advantages of this detection device are as follows:

[0071] 1. Non-contact detection design, strong resistance to harsh environments; the detection device uses a mining intrinsically safe camera for image acquisition, eliminating the need to install contact sensors on chains or sprockets, completely avoiding the wear and tear of sensors caused by coal gangue collisions, friction, and other working conditions, and solving the core defects of traditional contact detection methods such as rapid sensitivity decay and low reliability; the entire equipment complies with mining intrinsic safety / explosion-proof standards, and can adapt to complex industrial environments such as coal mines and chemical industries with high dust, strong vibration, and humidity, significantly extending its service life.

[0072] 2. Edge computing architecture, combining real-time and security. The mining-grade explosion-proof and intrinsically safe AI edge inference host in this detection device integrates an intrinsically safe power supply module and a high-speed data processing unit to achieve localized closed-loop processing of image acquisition, method reasoning, and alarm control. Compared with cloud-based detection solutions, it eliminates network delays and data transmission risks, shortens fault response time to milliseconds, and meets the strict requirements of industrial control for data security and privacy protection.

[0073] 3. Modular design and easy deployment: The detection device uses standardized mining cables for connection (integrated power supply and communication), and the camera bracket can be flexibly installed at the tail of the scraper. It only needs to ensure that the field of view covers at least two scrapers, without the need for structural modification of existing equipment. Compared with traditional non-contact detection methods such as Hall sensors, it avoids complex mechanical installation and calibration processes, significantly reducing the difficulty of project implementation and maintenance costs.

[0074] The mining flameproof and intrinsically safe AI edge inference host has a built-in intelligent detection method for scraper chain breakage based on rotating target detection; the intelligent detection method for scraper chain breakage based on rotating target detection includes a target detection method and a state analysis method;

[0075] like Figure 3 As shown in the figure, the target detection method is improved based on YOLOv11, which replaces the Conv module with the ARConv module. In terms of feature extraction, the ARConv module can adaptively adjust the convolution kernel to a suitable rectangle to fit the slender and non-horizontal shape characteristics of the scraper conveyor. It can also dynamically adapt to targets of different scales and accurately capture features such as edges and textures. In target positioning, it can better align rotating targets, retain direction information, and reduce false detections and missed detections. The model generalization ability is also enhanced and can adapt to different working scenarios and data distribution changes. In addition, it can also optimize model performance and reduce computational redundancy, such as Figure 4 As shown in the figure, an SE attention mechanism module is added between the Backbone module and the Neck module. The channel attention mechanism is used to optimize the feature expression, making it more adaptable to the challenges of complex background, rotation posture changes, low-quality images, etc. in the detection of coal mine scraper conveyors. Under the premise of almost no increase in computing cost, the feature discrimination of the target, multi-scale fusion efficiency and robustness are improved. Figure 6 As shown in the figure, the C2PSA module is replaced by the FSTA module, which can improve the accuracy, robustness and edge deployment efficiency of target detection in complex environments through the rotation invariance of frequency domain features, spatiotemporal noise suppression and multi-scale adaptation, as shown in the figure. Figure 5 shown.

[0076] like Figure 7 、 Figure 8 As shown, the state analysis method is implemented by the following steps:

[0077] Step 1: Input data and parameter extraction

[0078] 1.1 Input Definition:

[0079] Perform a rectangular frame extraction operation on each scraper in the processed scraper machine running image (such as Figure 9 As shown), the rectangular frame set is B={b i} n i＝1 , where each rectangular box b i =(x i ,y i ,w i ,h i ,θ i );

[0080] Among them, (x i ,y i ) are the center coordinates;

[0081] w i ,h i For width and height (w i ,h i >0);

[0082] θ i The original rotation angle (clockwise relative to the x-axis, unit: degree).

[0083] 1.2 Parameter extraction:

[0084] Extract w for each rectangular box i ,h i ,θ i , forming a triple set (w i ,h i ,θ i ).

[0085] Step 2: Long side determination and angle standardization

[0086] 2.1 Long side determination and angle correction

[0087] Define the angle α corresponding to the long side of the rectangular box i :

[0088]

[0089] 2.2 Angle normalization to [0,180°)

[0090] Modulo operation is used to eliminate angle periodicity and ensure angle comparability:

[0091]

[0092] in is the floor function, ensuring that α i ′∈[0,180°).

[0093] Step 3: Invalid data filtering and sample size check

[0094] 3.1 Filter invalid boxes

[0095] The set B of valid rectangles valid Need to meet: B valid ={b i ∣w i >0,h i >0,θ i efficient}

[0096] The valid angle set corresponding to the valid rectangular frame is A={α i ′} bi∈Bvalid .

[0097] 3.2 Sample size check

[0098] If |A|<2 (the number of valid angles is less than 2), the method terminates and returns, and the broken link status = False.

[0099] Step 4: Calculate the angle deviation

[0100] 4.1 Reference Angle Calculation

[0101] Select the reference angle μ (usually the average value, but the median can also be used as required):

[0102] (average value);

[0103] 4.2 Absolute deviation calculation

[0104] The absolute deviation of each angle from the reference angle is: d k =|α k ′-μ∣,k=1,2,…,∣A∣;

[0105] Step 5: Broken link determination

[0106] Set the angle deviation threshold T (e.g. 15°≤T≤30°). If the maximum deviation exceeds the threshold: Otherwise, return disconnected status = False.

[0107] The technical advantages of this detection method are as follows:

[0108] 1. Targeted optimization of rotating target detection technology; this detection method is based on the improved target detection model of YOLOv11. It dynamically matches the slender rotating target characteristics of scraper chains and scrapers through the ARConv adaptive rectangular convolution kernel. Compared with the traditional horizontal rectangular box (HBB) detection method, it can accurately capture the edge texture characteristics of inclined targets and reduce the missed detection rate. At the same time, combined with the SE channel attention module, it effectively suppresses complex background noise (such as coal dust and equipment reflections) and improves target differentiation in low-quality images (low illumination and motion blur). The detection accuracy can maintain a high level under typical lighting conditions (10-50 lux) in coal mines.

[0109] 2. Multi-dimensional feature fusion and robustness enhancement; this detection method introduces the FSTA frequency-domain spatiotemporal attention module to replace the traditional spatial attention mechanism, and utilizes the rotational invariance and spatiotemporal noise suppression capabilities of frequency-domain features to solve the problem of early chain break detection caused by the small difference in sprocket speeds in dual-drive scrapers.

[0110] 3. Quantify the chain break determination model to eliminate subjective misjudgment; the state analysis method in this detection method converts the chain operation status into a quantifiable angle deviation indicator (threshold range is adjustable from 15° to 30°) through angle standardization and deviation calculation, thereby improving the accuracy of its detection results.

Claims

1. A scraper fault detection device based on machine vision, characterized by: The scraper fault detection device includes a camera for image acquisition, an AI edge inference host for inference based on the image, a display for displaying the image, and an audible and visual alarm for performing alarm operations; the camera is fixedly connected to the tail position of the scraper through a bracket, the camera is connected to the AI ​​edge inference host through a mining network cable, and the AI ​​edge inference host is connected to the display, the audible and visual alarm and the scraper controller through the mining network cable.

2. The machine vision-based scraper fault detection device according to claim 1, characterized in that: The scraper fault detection device also includes a power supply module for power supply. The power supply module is built into the AI ​​edge inference host. The power supply module is connected to the camera, display, and sound and light alarm through a mining cable and supplies power to them.

3. A method for detecting a scraper failure based on machine vision, which is implemented by the scraper failure detection device based on machine vision according to claim 1, characterized in that: The following steps are involved: S1, the camera collects the running image of the scraper in real time; S2. The camera transmits the collected scraper operation images to the AI ​​edge inference host via the mining network cable; S3, the AI ​​edge inference host processes the scraper operation image through the built-in target detection method and transmits the processed scraper operation image to the monitor through the mining network cable for display; S4: The AI ​​edge inference host analyzes and judges the processed scraper operation image through the built-in state analysis method to determine whether the scraper has a chain break. If it is determined that the scraper has no chain break, steps S1 to S4 are repeated to achieve cyclic monitoring of the scraper. If it is determined that the scraper has a chain break, the next step is entered. S5. The AI ​​edge inference host sends a disconnection message to the display through the mining network cable. The display pops up an alarm and displays the scraper failure information. The AI ​​edge inference host triggers the sound and light alarm through the mining network cable to start and perform the sound and light alarm operation. The AI ​​edge inference host sends a shutdown command to the scraper controller through the mining network cable, and the scraper controller performs the shutdown protection operation.

4. The method for detecting scraper failure based on machine vision according to claim 3, wherein: In step S3, the target detection method built into the AI ​​edge inference host is based on the YOLOv11 method, the Conv module in the YOLOv11 method is replaced by the ARConv module, the C2PSA module in the YOLOv11 method is replaced by the FSTA module, and the SE attention mechanism module is added between the Backbone module and the Neck module in the YOLOv11 method.

5. The method for detecting scraper failure based on machine vision according to claim 4, characterized in that: The step S4 specifically includes the following steps: S41, perform a rectangular frame extraction operation on each scraper in the processed scraper machine running image, and obtain a rectangular frame set B={b i } n i＝1 , in the rectangular box set, each rectangular box b i =(x i ,y i ,w i ,h i ,θ i ); where (x i ,y i ) is the center coordinate of the rectangular frame; w i is the width of the rectangular frame; h i is the height of the rectangular frame; θ i is the original rotation angle; S42, define and calculate the angle α corresponding to the long side direction of each rectangular frame i ; Its calculation formula is: S43, angle α i After normalization, the calculation formula is: Among them, α i ′∈[0,180°); S44, filtering out invalid rectangular frames in the processed scraper running image to obtain a valid rectangular frame set B valid , the valid rectangular box set B valid The corresponding valid angle set is A={α i ′} bi∈Bvalid If |A|<2, the AI ​​edge inference host determines that the scraper does not have a broken link; if |A|≥2, proceed to the next step; S45. Calculate the reference angle μ using the following formula: S46. Calculate the absolute deviation between each angle in the valid angle set A and the reference angle μ. The calculation formula is: d k =∣α k ′-μ∣,k=1,2,…,∣A∣; S47. Set the angle deviation threshold T. If the maximum value of the absolute deviation calculated in step S46 exceeds the angle deviation threshold T, the AI ​​edge inference host determines that the scraper has a broken chain; otherwise, the AI ​​edge inference host determines that the scraper does not have a broken chain.

Images