Chain monitoring method of scraper conveyor, medium and equipment

Abstract

The invention relates to the technical field of scraper conveyors, in particular to a chain monitoring method, medium and equipment for a scraper conveyor, and the method comprises the steps: transmitting a real-time image of a chain of the scraper conveyor to an edge calculation node; preprocessing a real-time image, and inputting the preprocessed real-time image into an improved YOLOv11 model for feature extraction, wherein the improved YOLOv11 model comprises a dynamic alignment detection head, a BRA attention mechanism and a bidirectional feature pyramid network; the state of the chain of the scraper conveyor is analyzed through the edge calculation nodes, and the edge calculation nodes analyze the state of the chain of the scraper conveyor based on the chain characteristics; and when the edge computing node recognizes that the state of the chain is abnormal, an alarm is triggered through the edge computing node, and a shutdown instruction is sent to a controller of the scraper conveyor. Therefore, the real-time performance, the reliability and the accuracy of monitoring the chain of the scraper conveyor are improved.

Description

Technical Field

[0001] This invention relates to the field of scraper conveyor technology, specifically to a method, medium, and equipment for monitoring the chain of a scraper conveyor. Background Technology

[0002] As a transportation device in modern fully mechanized coal mining faces, the reliability of scraper conveyors directly affects the production efficiency and safety level of the entire coal mining system. During continuous operation under long-term high loads and harsh environments, chain breakage, chain jamming, and chain skipping can easily lead to production interruptions, equipment damage, and even safety accidents. In particular, chain breakage can trigger a chain reaction such as sprocket jamming and motor overload, posing a significant threat to the safety of underground workers and causing huge economic losses.

[0003] In some scenarios, chain monitoring of scraper conveyors primarily relies on traditional manual inspections or contact sensor technology. Manual inspection methods depend on operators periodically checking or identifying locations by sound, making 24-hour continuous monitoring impossible. This introduces risks of missed or false detections, and workers approaching operating equipment are at risk of mechanical injury. While contact sensors, such as tension sensors and proximity switches, improve automation, long-term friction with the chain leads to wear, deformation, and decreased reliability. Furthermore, they can only detect severe faults like complete breakage, failing to identify early damage such as chain link cracks or tensile deformation. Therefore, existing methods for monitoring scraper conveyor chains offer low real-time performance, reliability, and accuracy. Summary of the Invention

[0004] To address the technical problems of low real-time performance, reliability, and accuracy in monitoring the chain of a scraper conveyor, the present invention aims to provide a method, medium, and equipment for monitoring the chain of a scraper conveyor. The specific technical solution adopted is as follows:

[0005] In a first aspect, embodiments of the present invention disclose a chain monitoring method for a scraper conveyor, comprising: transmitting a real-time image of the scraper conveyor chain to an edge computing node, wherein an improved YOLOv11 model is deployed in the edge computing node; inputting the preprocessed real-time image into the improved YOLOv11 model for feature extraction, wherein the improved YOLOv11 model includes a dynamic alignment detection head, a BRA attention mechanism, and a bidirectional feature pyramid network; performing hierarchical routing calculation on the preprocessed real-time image in the backbone network through the BRA attention mechanism to obtain multi-scale features; enhancing the multi-scale features in the neck network through the bidirectional feature pyramid network to obtain enhanced features; and performing task alignment on the enhanced features in the head network through the dynamic alignment detection head to obtain the chain features of the real-time image; analyzing the state of the scraper conveyor chain using the edge computing node, wherein the edge computing node analyzes the state of the scraper conveyor chain based on the chain features; and triggering an alarm and sending a stop command to the controller of the scraper conveyor when the edge computing node detects an abnormal chain state.

[0006] In a second aspect, embodiments of the present invention disclose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the chain monitoring method for a scraper conveyor as described in the first aspect.

[0007] Thirdly, embodiments of the present invention disclose an electronic device, including: a processor and a memory; wherein the memory is used to store a computer program that can run on the processor; the processor is used to execute the program stored in the memory to implement the steps of the chain monitoring method for the scraper conveyor mentioned in the first aspect above.

[0008] The technical solution disclosed in this invention transmits real-time images of the scraper conveyor chain to an edge computing node and deploys an improved YOLOv11 model on the edge computing node. This enables local data processing, avoiding the latency of remote data transmission to the cloud platform and allowing for millisecond-level identification and response to chain anomalies. When an anomaly is detected, the edge computing node directly triggers an audible and visual alarm and sends a shutdown command to the scraper conveyor controller, significantly shortening the time window from fault occurrence to protective measures. This achieves 24-hour continuous real-time monitoring, ensuring the real-time monitoring of the scraper conveyor chain and effectively preventing equipment damage or safety accidents caused by response delays. The improved YOLOv11 model, through the collaborative work of a dynamically aligned detection head, a BRA attention mechanism, and a bidirectional feature pyramid network, significantly improves the accuracy and robustness of feature extraction, thereby enhancing the reliability and accuracy of scraper conveyor chain monitoring. The BRA attention mechanism performs region segmentation and token-to-token attention operations on the preprocessed real-time image, enhancing the features of the target area of ​​the chain, suppressing interference from complex downhole backgrounds, and accurately capturing local details and global semantic information of the chain through hierarchical routing, reducing missed detections and false detections, and improving the reliability and accuracy of current chain monitoring for scraper conveyors, which is currently low. The bidirectional feature pyramid network performs bidirectional fusion and dynamic weight adaptation on the multi-scale features output by BRA, taking into account both high-level semantic information and low-level detailed features. This improves the ability to identify early damage, further enhancing the reliability and accuracy of chain monitoring for scraper conveyors. The dynamic alignment detection head performs adaptive alignment of enhanced features for classification and regression tasks, using deformable convolution and spatial saliency masks to ensure that the predicted bounding box accurately fits the chain boundary, further reducing the false alarm rate. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a chain monitoring method for a scraper conveyor provided in an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of the structure of a monitoring device provided in an embodiment of the present invention;

[0011] Figure 3 A schematic diagram of the structure of an improved YOLOv11 model provided in an embodiment of the present invention;

[0012] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0013] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a chain monitoring method, medium, and equipment for a scraper conveyor according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0014] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The specific implementation of a chain monitoring method for a scraper conveyor provided by this invention is described below in conjunction with the accompanying drawings.

[0015] like Figure 1 As shown, Figure 1 This is a flowchart illustrating a chain monitoring method for a scraper conveyor provided in an embodiment of the present invention. The method includes:

[0016] Step S101: Transmit the real-time image of the scraper conveyor chain to the edge computing node, where an improved YOLOv11 model is deployed.

[0017] Specifically, such as Figure 2 As shown, in this embodiment of the invention, a monitoring device is installed at a key location (such as the head or tail) of the scraper conveyor. Figure 2 This is a schematic diagram of a monitoring device provided in an embodiment of the present invention. The monitoring device includes a mounting base 1, a mounting beam arm 2, sheet metal parts 3, a shock absorber 4, a universal joint 5, a camera mounting plate 6, sheet metal parts 7, a light source 8 and 9, and an industrial camera 10. The shock absorber and universal joint are used to buffer vibrations and adjust the camera angle to ensure image stability; the light source is an LED supplementary lighting system, which automatically adjusts its intensity according to the ambient brightness and dust concentration to avoid uneven lighting or dust interference. The industrial camera is a high-definition explosion-proof model (e.g., resolution 1920×1080) that supports low-light acquisition. The camera, assisted by the supplementary lighting system, continuously captures images of the chain running at a rate of 25 frames per second, ensuring high image clarity and rich features. The acquisition time is fixed during the peak period of chain operation to cover various working conditions. The supplementary lighting system has a built-in photosensitive sensor and a dust detection module to monitor the underground environment in real time. When the brightness is below 50 lux, the supplementary lighting is automatically enhanced; when the dust concentration is high, infrared wavelengths are used to reduce scattering. The power supply module is intrinsically safe, ensuring safety during downhole operations.

[0018] Furthermore, industrial cameras transmit real-time images of the chain to a data switch in the underground machine room via shielded mining network cables. The switch then uploads these real-time images to the surface data processing layer via a mining industrial ring network. The ring network employs a redundant design to ensure automatic switching in case of transmission link interruption, preventing data loss. Image data is encapsulated in TCP / IP protocol packets, transmitted in H.264 compressed video stream or JPEG sequence frames to balance bandwidth and real-time performance. Transmission latency is controlled within 100 milliseconds to meet the real-time requirements of edge computing. During transmission, the underground switch can perform preliminary data filtering, such as discarding duplicate frames or low-quality images, reducing the load on edge nodes.

[0019] Furthermore, edge computing nodes are deployed in the ground monitoring room, employing dedicated servers such as Intel Xeon processors, 32GB RAM, and NVIDIA Tesla T4 GPUs to support high-performance inference. The servers are connected to the underground network via a mining ring network access point and are configured with firewalls to ensure data security.

[0020] Furthermore, a Linux operating system is installed on the server, and a Docker container environment is deployed. The improved YOLOv11 model and its dependencies are encapsulated within the container. The improved YOLOv11 model includes a Dynamic Alignment Detection Head (DA-Head), a BRA attention mechanism, and a Bidirectional Feature Pyramid Network (BiFPN).

[0021] Step S102: The preprocessed real-time image is input into the improved YOLOv11 model for feature extraction.

[0022] The improved YOLOv11 model includes a dynamic alignment detection head, a BRA attention mechanism, and a bidirectional feature pyramid network. The BRA attention mechanism performs hierarchical routing calculations on the preprocessed real-time image in the backbone network to obtain multi-scale features. The bidirectional feature pyramid network enhances the multi-scale features in the neck network to obtain enhanced features. The dynamic alignment detection head aligns the enhanced features in the head network to obtain the chain features of the real-time image.

[0023] Specifically, after receiving real-time images, the edge computing nodes first call libraries such as OpenCV to perform preprocessing operations: First, a nonlocal means denoising algorithm is used to reduce image noise caused by underground dust and vibration. Then, a histogram equalization algorithm is used to enhance contrast to highlight chain edges and texture features. Finally, the Canny edge detector is applied to eliminate background interference and focus on the chain outline. The image resolution is maintained at 1920×1080 after preprocessing to ensure data standardization before input to the model.

[0024] Furthermore, the preprocessed image is input into the backbone network of the YOLOv11 model, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of an improved YOLOv11 model provided in an embodiment of the present invention. Specifically, a dynamic alignment detection head is introduced into the head network of the original YOLOv11 model to replace the original detection head. The dynamic alignment detection head performs adaptive alignment for classification and regression tasks through deformable convolution and spatial saliency masking. A BRA attention mechanism is integrated into the backbone network of the original YOLOv11 model. The BRA attention mechanism captures local and global features of the chain in real-time images through region-level routing and token-to-token attention computation, obtaining multi-scale features. A bidirectional feature pyramid network is introduced into the neck network of the original YOLOv11 model, and feature enhancement is performed through multi-scale feature enhancement and dynamic weight adaptive mechanisms to improve the semantic expression and detail preservation of the scraper conveyor chain, resulting in the improved YOLOv11 model.

[0025] Furthermore, when performing feature extraction in the improved YOLOv11 model, the BRA attention mechanism first performs hierarchical routing calculations on the preprocessed real-time image in the backbone network. As an optional embodiment of this invention, the multi-scale features obtained by performing hierarchical routing calculations on the preprocessed real-time image in the backbone network using the BRA attention mechanism include: inputting the preprocessed real-time image into the backbone network; performing preliminary feature transformation through the convolutional layers of the backbone network to obtain a basic feature map; dividing the basic feature map into multiple non-overlapping regions and extracting the regional features of the non-overlapping regions; performing linear mapping on the regional features to obtain query tensors, key tensors, and value tensors; aggregating the query tensors, key tensors, and value tensors through regional-level routing to obtain aggregated key-value pairs; and performing token-to-token attention calculations on the aggregated key-value pairs to obtain multi-scale features.

[0026] Specifically, in this embodiment of the invention, the preprocessed real-time image is input into the backbone network, and preliminary feature transformation is performed through the convolutional layers of the backbone network to obtain a basic feature map. Then, the basic feature map is divided into S×S non-overlapping regions, where S can be 4. The size of each non-overlapping region is H×W / S², where H represents the height of the original image and W represents the width of the original image. Next, region features are extracted from each non-overlapping region. Then, regional features By performing a linear mapping, we obtain the key tensor K and the value tensor V, with the specific mapping relationship as follows: ;

[0027] In the above formula, . To query the weight projection of tensor Q, The weighted projection of the key tensor K, The weighted projection of the value tensor V.

[0028] Furthermore, in this embodiment of the invention, the query tensor, key tensor, and value tensor are aggregated through region-level routing to obtain aggregated key-value pairs. Specifically, the query tensor Q and key tensor K are first subjected to average pooling to obtain region-level features. and Then calculate the regional features. and The relevance adjacency matrix is ​​specifically represented by the following formula: ;

[0029] In the above formula, This is the adjacency matrix of the correlations; For regional query tensors; It is a regional key tensor.

[0030] Furthermore, in coarse-grained regional-level routing calculations, the routing index matrix is ​​utilized. The indexes of the first k connections are recorded row by row so that only the first k connections of each region are retained when pruning the correlation graph. Tensor aggregation is performed on the key tensor and value tensor, as shown in the following equation: ;

[0031] In the above formula, The bond tensor after aggregation; This is the aggregated value tensor.

[0032] Finally, for the scattered key-value pairs, attention operations need to be performed on the aggregated K and V pairs to perform fine-grained token-to-token attention computation, as shown in the following formula: ;

[0033] In the above formula, The multi-scale features are the final output of the BRA attention mechanism; Enhance local context.

[0034] Thus, LCE(V) is a local context enhancement that strengthens details through convolutional layers. It outputs multi-scale features O, capturing local and global information of the chain and improving robustness to occlusion and illumination changes.

[0035] Furthermore, as an optional embodiment of the present invention, feature enhancement of multi-scale features in the neck network through a bidirectional feature pyramid network to obtain enhanced features includes: inputting multi-scale features into the bidirectional feature pyramid network, fusing multi-scale features through the bidirectional paths of the bidirectional feature pyramid network to obtain fused features; and enhancing the fused features to obtain enhanced features.

[0036] Specifically, multi-scale features, output from the backbone network, are input into the BiFPN module of the neck network for feature enhancement. The Bidirectional Feature Pyramid Network (BiFPN) employs both top-down and bottom-up paths. The top-down path fuses high-level features (semantically rich) with low-level features (clear details) through upsampling. Specifically, in the feature fusion stage, an FEM feature enhancement module is introduced, extracting multi-scale features in parallel through multi-branch and dilated convolutions to enhance the semantics and local contextual information of the chain targets, thereby improving the discriminative power of the feature maps. The BiFPN structure uses a bidirectional fusion approach, allowing the feature maps to capture both high-level semantic information and retain low-level detailed features. Simultaneously, a dynamic feature weight adaptive mechanism is introduced to automatically adjust the dynamic weights of high-level and low-level features in each layer, achieving intelligent feature selection and fusion: high-level features enhance semantic expression, while low-level features preserve spatial details. The bottom-up path preserves spatial information through downsampling. The fusion formula for each layer is: ;

[0037] In the above formula, This indicates the fusion feature. Represents the dynamic weights of high-level features. Indicates characteristics of high-level personnel. Represents underlying features. This represents the dynamic weights of the underlying features.

[0038] Furthermore, in the fusion process, this embodiment of the invention introduces an FEM module, which extracts multi-scale context in parallel through multi-branch dilated convolutions (dilation rates=1,3,5) to enhance the semantic representation of the chain target. The output enhanced features significantly improve the discriminative power for subtle anomalies such as chain cracks and stretching.

[0039] Furthermore, as an optional embodiment of the present invention, the chain features of the real-time image are obtained by performing task alignment of the enhanced features in the head network through the dynamic alignment detection head: inputting the enhanced features into the dynamic alignment detection head, extracting the interaction features of the enhanced features in the classification and regression tasks through the shared convolutional layer of the dynamic alignment detection head; performing average pooling on the interaction features through the dynamic alignment detection head to obtain global context information, and calculating the task attention weight of the global context information based on the global context information; obtaining the classification-specific features and regression-specific features of the classification and regression tasks through the dynamic alignment detection head based on the task attention weight; and performing task alignment on the classification-specific features and regression-specific features through the deformable convolution and spatial saliency mask of the dynamic alignment detection head to obtain the chain features of the real-time image.

[0040] Specifically, in this embodiment of the invention, the enhanced features are input to the dynamic alignment detection head (DA-Head module) of the head network to perform task alignment and output chain features. The DA-Head module first extracts the interaction features between classification and regression tasks through shared convolutional layers, specifically expressed by the following formula: ;

[0041] In the above formula, This indicates that interactive features for classification and regression tasks are extracted by sharing convolutional layers. express Layer feature map. This represents a set of shared convolutions with a kernel size of 3×3, consisting of convolutional layers, group normalization, and SiLU activation functions. This indicates a feature concatenation operation along the channel dimension.

[0042] Furthermore, the interaction features are averaged using a dynamic alignment detection head to obtain global context information. Then based on global context information The task attention weights for calculating global context information are expressed as follows: ;

[0043] In the above formula, This represents the task attention weights. σ is the Sigmoid function, and δ is the ReLU activation function.

[0044] Furthermore, according to Dynamically modulated feature representation yields task-specific output, specifically expressed as follows: ;

[0045] In the above formula, Able to be based on Adaptive modulation is performed to enable classification and regression tasks to acquire different feature representations, thereby obtaining classification-specific features and regression-specific features for classification and regression tasks. This indicates the task attention weight. express according to The classification-specific features and regression-specific features obtained after adaptive modulation for classification and regression tasks.

[0046] Furthermore, the regression branch introduces deformable convolution DCNv3 to adaptively learn sampling point offsets, improving the fit between the predicted bounding box and the chain boundary; the classification branch generates a spatial saliency mask to highlight the foreground chain region and suppress background interference. The final output is the chain features of the real-time image, including the object detection box coordinates, class confidence, and spatial feature vector.

[0047] Thus, this embodiment of the invention relies on the collaborative work of three modules of the improved YOLOv11 model: the BRA attention mechanism to extract multi-scale features, BiFPN to enhance feature representation, and DA-Head to achieve accurate task alignment. The entire process runs end-to-end on edge computing nodes, taking less than 50 milliseconds to ensure real-time performance. This guarantees the accuracy and reliability of chain feature extraction.

[0048] Step S103: The state of the scraper conveyor chain is analyzed using edge computing nodes. The edge computing nodes analyze the state of the scraper conveyor chain based on chain characteristics.

[0049] Specifically, edge computing nodes receive chain features from the improved YOLOv11 model output. These features include object detection box coordinates, class confidence scores, and spatial feature vectors. The model output is a multidimensional tensor, and the edge nodes extract key information through a parsing script. For example, each detection box contains bounding box coordinates (x_min, y_min, x_max, y_max), class probabilities (such as confidence scores for normal chains, broken chains, stuck chains, skipped chains, and relaxed stacked chains), and feature map embedding vectors for fine-grained analysis. During parsing, only detection results with a confidence score higher than 0.5 are retained to reduce false positives.

[0050] To improve reliability, edge computing nodes cache detection results for 5 consecutive frames. Motion trajectories are calculated using a sliding window, and if the chain position changes abruptly beyond a threshold within 3 consecutive frames, it is marked as a potential anomaly (such as chain skipping).

[0051] Furthermore, if the model outputs a "broken chain" category confidence score above 0.8 for three consecutive frames, and the bounding boxes show an interruption in chain continuity (e.g., an abnormally increased distance between boxes), then it is determined to be a broken chain. Simultaneously, considering motion characteristics: a normal chain should have a uniform speed, while a broken chain will have a sudden drop in speed to 0.

[0052] Furthermore, if the "chain jam" confidence score is higher than 0.75 and the bounding box position remains unchanged for 5 consecutive frames (displacement difference < 5 pixels), it is determined to be a chain jam. The node calculates the relative position of the chain and sprocket, and if interference occurs (such as the overlap area of ​​the bounding box exceeds 30%), the judgment is strengthened.

[0053] Furthermore, based on motion trajectory analysis: if the chain bounding box abruptly changes in the vertical direction (e.g., offset > 50 pixels), and the "jump chain" confidence score is higher than 0.7, it is determined to be a jump chain. Nodes are further verified for trajectory anomalies using optical flow.

[0054] Furthermore, when the confidence level of "relaxed chain stacking" is higher than 0.7, and the bounding box morphology shows that the chains are drooping or stacking (such as an abnormal decrease in the aspect ratio of the box), the relaxation is judged as relaxed when the relaxation changes are compared with historical data.

[0055] Furthermore, the edge computing nodes employ a weighted voting mechanism, combining category confidence, motion characteristics, and spatial relationships such as the relative positions of the chain and the central slot. The weights are obtained through training with historical data; for example, confidence weight is 0.6, motion characteristic weight is 0.3, and spatial relationship weight is 0.1.

[0056] Furthermore, as an optional embodiment of the present invention, abnormal conditions of the scraper conveyor chain include: chain breakage, chain jamming, chain skipping, or loose chain stacking.

[0057] Step S104: When the edge computing node detects that the chain is in an abnormal state, it triggers an alarm and sends a shutdown command to the scraper conveyor controller.

[0058] Specifically, if all anomaly confidence levels are below the threshold, the node records operational data and repeats the image acquisition and processing flow. Simultaneously, normal state data is stored in the database for model optimization. When any anomaly is detected, the edge computing node sends an alarm signal to the downhole control center via the ring network. A pop-up window on the display shows the anomaly type (e.g., "Chain Break Alarm"), and the audible and visual alarms are activated. The edge computing node synchronously sends a shutdown command to the scraper conveyor controller, with a command delay of <100 milliseconds. The controller executes an emergency shutdown to prevent the fault from escalating. The edge computing node also generates a detailed report, including the anomaly time, location, and confidence level, to help maintenance personnel quickly locate the problem.

[0059] Thus, through the technical solution provided in this invention, intelligent analysis of the scraper conveyor chain status is achieved by parsing chain features, determining multi-dimensional states, optimizing thresholds, and automating processing. The entire process runs on an edge computing node, taking less than 50 milliseconds, meeting the real-time requirements of downhole operation. This invention effectively improves the accuracy and efficiency of chain monitoring and reduces reliance on manual intervention.

[0060] Furthermore, as an optional embodiment of the present invention, the chain monitoring method further includes: continuously acquiring the running data of the scraper conveyor chain, and using the running data of the scraper conveyor chain to optimize the model parameters of the improved YOLOv11 model.

[0061] Furthermore, as an optional embodiment of the present invention, before transmitting the real-time image of the scraper conveyor chain to the edge computing node, the method further includes: dynamically adjusting the supplementary light intensity of the supplementary light system according to the brightness and dust concentration of the environment in which the scraper conveyor chain is located, so as to acquire the real-time image of the scraper conveyor chain through an industrial camera, wherein the supplementary light system is fixed near the scraper conveyor by a mounting bracket.

[0062] The technical solution disclosed in this invention transmits real-time images of the scraper conveyor chain to an edge computing node and deploys an improved YOLOv11 model on the edge computing node. This enables local data processing, avoiding the latency of remote data transmission to the cloud platform and allowing for millisecond-level identification and response to chain anomalies. When an anomaly is detected, the edge computing node directly triggers an audible and visual alarm and sends a shutdown command to the scraper conveyor controller, significantly shortening the time window from fault occurrence to protective measures. This achieves 24-hour continuous real-time monitoring, ensuring the real-time monitoring of the scraper conveyor chain and effectively preventing equipment damage or safety accidents caused by response delays. The improved YOLOv11 model, through the collaborative work of a dynamically aligned detection head, a BRA attention mechanism, and a bidirectional feature pyramid network, significantly improves the accuracy and robustness of feature extraction, thereby enhancing the reliability and accuracy of scraper conveyor chain monitoring. The BRA attention mechanism performs region segmentation and token-to-token attention operations on the preprocessed real-time image, enhancing the features of the target area of ​​the chain, suppressing interference from complex downhole backgrounds, and accurately capturing local details and global semantic information of the chain through hierarchical routing, reducing missed detections and false detections, and improving the reliability and accuracy of current chain monitoring for scraper conveyors, which is currently low. The bidirectional feature pyramid network performs bidirectional fusion and dynamic weight adaptation on the multi-scale features output by BRA, taking into account both high-level semantic information and low-level detailed features. This improves the ability to identify early damage, further enhancing the reliability and accuracy of chain monitoring for scraper conveyors. The dynamic alignment detection head performs adaptive alignment of enhanced features for classification and regression tasks, using deformable convolution and spatial saliency masks to ensure that the predicted bounding box accurately fits the chain boundary, further reducing the false alarm rate.

[0063] Corresponding to the chain monitoring method for scraper conveyors provided in the above embodiments, based on the same technical concept, this embodiment of the invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the chain monitoring method for scraper conveyors mentioned in the above embodiments.

[0064] Corresponding to the chain monitoring method for scraper conveyors provided in the above embodiments, based on the same technical concept, this embodiment of the invention also provides an electronic device for executing the above-described chain monitoring method for scraper conveyors. Figure 4 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present invention, as shown below. Figure 4 As shown. Electronic devices can vary considerably due to differences in configuration or performance, and may include one or more processors 401 and memories 402. The memory 402 stores computer programs that can run on the processor 401, and the processor 401 executes the programs stored in the memory 402 to achieve the above. Figure 1 The various steps in the method embodiment are described. The memory 402 can be temporary or persistent storage. The application stored in the memory 402 may include one or more modules (not shown), each module may include a series of computer-executable instructions for the electronic device.

[0065] Furthermore, the processor 401 may be configured to communicate with the memory 402 and execute a series of computer-executable instructions stored in the memory 402 on the electronic device. The electronic device may also include one or more power supplies 403, one or more wired or wireless network interfaces 404, one or more input / output interfaces 405, and one or more keyboards 406.

[0066] Specifically, in this embodiment, the electronic device includes a processor, a communication interface, a memory, and a communication bus; wherein, the processor, the communication interface, and the memory communicate with each other via the bus; the memory is used to store computer programs; and the processor is used to execute the programs stored in the memory to achieve the above. Figure 1 The various steps in the method embodiments are the same as those in the above method embodiments, and have the same beneficial effects. To avoid repetition, the embodiments of the present invention will not be described again here.

[0067] It should be noted that the electronic device provided in this embodiment of the invention and the chain monitoring method for scraper conveyors provided in this embodiment of the invention are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned chain monitoring method for scraper conveyors, and has the same or similar beneficial effects. Repeated parts will not be described again.

[0068] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0069] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method of chain monitoring for an apron conveyor, characterized by, The chain monitoring method of the scraper conveyor includes: transmitting a real-time image of a chain of a scraper conveyor to an edge computing node, wherein an improved YOLOv11 model is deployed in the edge computing node; inputting the preprocessed real-time image into the improved YOLOv11 model for feature extraction, wherein the improved YOLOv11 model includes a dynamic alignment detection head, a BRA attention mechanism, and a bidirectional feature pyramid network, the BRA attention mechanism is used for hierarchical routing calculation on the preprocessed real-time image in a backbone network to obtain multi-scale features, the bidirectional feature pyramid network is used for feature enhancement on the multi-scale features in a neck network to obtain enhanced features, and the dynamic alignment detection head is used for task alignment on the enhanced features in a head network to obtain chain features of the real-time image; analyzing a state of the chain of the scraper conveyor by using the edge computing node, wherein the edge computing node analyzes the state of the chain of the scraper conveyor based on the chain features; when the edge computing node identifies that the state of the chain is abnormal, triggering an alarm by the edge computing node and sending a shutdown instruction to a controller of the scraper conveyor.

2. The chain monitoring method of a flight conveyor according to claim 1, characterized in that, Before the step of inputting the preprocessed real-time image into the improved YOLOv11 model for feature extraction, the method further includes: introducing a dynamic alignment detection head into a head network of an original YOLOv11 model to replace an original detection head, wherein the dynamic alignment detection head performs adaptive alignment of classification and regression tasks through deformable convolution and spatial saliency mask; integrating a BRA attention mechanism into a backbone network of the original YOLOv11 model, wherein the BRA attention mechanism captures local and global features of the chain in the real-time image through region-level routing and token-to-token attention calculation to obtain multi-scale features; introducing a bidirectional feature pyramid network into a neck network of the original YOLOv11 model, wherein the bidirectional feature pyramid network performs feature enhancement through multi-scale feature enhancement and a dynamic weight adaptive mechanism to improve semantic expression and detail retention of the chain of the scraper conveyor, and thus the improved YOLOv11 model is obtained.

3. The chain monitoring method of a flight conveyor according to claim 2, characterized in that, The step of obtaining the chain features of the real-time image by using the dynamic alignment detection head to perform task alignment on the enhanced features in the head network includes: inputting the enhanced features into the dynamic alignment detection head, and extracting interaction features of the enhanced features in classification and regression tasks through a shared convolution layer of the dynamic alignment detection head; performing average pooling on the interaction features by the dynamic alignment detection head to obtain global context information, and calculating task attention weights of the global context information based on the global context information; obtaining classification-specific features and regression-specific features of the classification and regression tasks based on the task attention weights by the dynamic alignment detection head; performing task alignment on the classification-specific features and the regression-specific features by deformable convolution and spatial saliency mask of the dynamic alignment detection head to obtain the chain features of the real-time image.

4. The chain monitoring method of a flight conveyor according to claim 2, characterized in that, The multi-scale features are obtained by performing hierarchical routing calculation on the preprocessed real-time image by the backbone network through the BRA attention mechanism. The preprocessed real-time image is input into the backbone network, and preliminary feature transformation is performed on the preprocessed real-time image by a convolutional layer of the backbone network to obtain a basic feature map. The basic feature map is divided into a plurality of non-overlapping regions, and region features of the non-overlapping regions are extracted. Linear mapping is performed on the region features to obtain a query tensor, a key tensor and a value tensor. The query tensor, the key tensor and the value tensor are aggregated by region-level routing to obtain an aggregated key-value pair. Token-to-token attention calculation is performed on the aggregated key-value pair to obtain the multi-scale features.

5. The chain monitoring method of a flight conveyor according to claim 2, characterized by, The multi-scale features are input into the bidirectional feature pyramid network, and multi-scale feature fusion is performed on the multi-scale features by a bidirectional path of the bidirectional feature pyramid network to obtain a fused feature. The fused feature is enhanced to obtain the enhanced feature. The abnormal conditions of the state of the chain of the flight conveyor include chain breakage, chain jamming, chain jumping or slack chain stacking.

6. The chain monitoring method of a flight conveyor according to claim 1, characterized by, Before the real-time image of the chain of the flight conveyor is transmitted to the edge computing node, the method further comprises:

7. The chain monitoring method of a flight conveyor according to claim 1, characterized by, According to the brightness and dust concentration of the environment in which the chain of the flight conveyor is located, the light supplement intensity of a light supplement system is dynamically adjusted to acquire the real-time image of the chain of the flight conveyor by an industrial camera, and the light supplement system is fixed to the vicinity of the flight conveyor by a mounting frame. The chain monitoring method further comprises:

8. The chain monitoring method of a flight conveyor according to claim 1, characterized by, The running data of the chain of the flight conveyor is continuously acquired, and the model parameters of the improved YOLOv11 model are optimized by using the running data of the chain of the flight conveyor. The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the chain monitoring method of the flight conveyor.

9. A computer-readable storage medium, characterized in that, It comprises:

10. An electronic device, comprising: a processor and a memory; wherein the memory is used to store a computer program that can be run on the processor; the processor is used to execute the program stored on the memory to implement the steps of the chain monitoring method of the flight conveyor as claimed in any one of claims 1-8. ​

Images