Self-adaptive correction method and system for coal mine scraper conveyor based on visual identification

Abstract

The invention relates to a coal mine scraper conveyor self-adaptive deviation correction method and system based on visual recognition, and relates to the technical field of coal mine conveying deviation correction, and the method comprises the steps that original time sequence image streams of a scraper chain and a ledge area of a target scraper conveyor are continuously obtained through an anti-explosion visual collection device; constructing an image cleanliness evaluation model, carrying out frame-by-frame analysis on the original time sequence image flow, and calculating and outputting a corresponding image cleanliness coefficient; dynamically configuring a grading decision-making mechanism by combining real-time working condition parameters of the target scraper conveyor based on the image cleanliness coefficient; and based on the corrected decision threshold and the image cleanliness coefficient, activating a corresponding decision mode, generating a deviation correction control quantity adaptive to the current working condition, and driving to execute adaptive deviation correction. The method solves the problems that a traditional correction method is difficult to deal with a coal mine high-dust environment, and false alarm, missing alarm and control disorder are likely to occur due to image quality degradation.

Description

Technical Field

[0001] This application relates to the field of coal mine conveyor correction, and in particular to an adaptive correction method and system for coal mine scraper conveyors based on visual recognition. Background Technology

[0002] In coal mining operations, scraper conveyors are the core conveying equipment. The smooth operation of the scraper chain is directly related to mining efficiency and operational safety. Precise correction of the scraper conveyor has become a key technical requirement to ensure continuous and stable underground production.

[0003] Currently, traditional visual correction methods do not fully consider the harsh environmental characteristics of high dust levels in underground coal mines. They cannot effectively solve the problems of image quality degradation and loss of key visual features caused by coal dust adhering to the lens. This can easily lead to false alarms, missed alarms, or control malfunctions in the correction system, affecting the reliability of correction and the safety of equipment operation. Summary of the Invention

[0004] This application provides a visual recognition-based adaptive correction method and system for coal mine scraper conveyors, which improves the problems of traditional correction methods being unable to cope with the high dust environment of coal mines and being prone to false alarms, missed alarms and control disorder due to image quality degradation.

[0005] The embodiments of this application disclose the following technical solutions: In a first aspect, embodiments of this application provide an adaptive correction method for a coal mine scraper conveyor based on visual recognition, the method comprising: The original time-series image stream of the scraper chain and trough area of ​​the target scraper conveyor is continuously acquired through an explosion-proof visual acquisition device. An image cleanliness assessment model is constructed, and the original time-series image stream is analyzed frame by frame to calculate and output the corresponding image cleanliness coefficient. Based on the image cleanliness coefficient and combined with the real-time operating parameters of the target scraper conveyor, a hierarchical decision-making mechanism is dynamically configured, wherein the hierarchical decision-making mechanism includes a visual correction mode, a prediction compensation mode, and a safety fault tolerance mode. Based on the corrected decision threshold and the image cleanliness coefficient, the corresponding decision mode is activated, a correction control quantity adapted to the current working condition is generated, and adaptive correction is driven to be executed.

[0006] Secondly, embodiments of this application provide an adaptive correction system for coal mine scraper conveyors based on visual recognition, the system comprising: The image acquisition module is used to continuously acquire the original time-series image stream of the scraper chain and trough side area of ​​the target scraper conveyor through an explosion-proof vision acquisition device; The quality assessment module is used to construct an image cleanliness assessment model, perform frame-by-frame analysis on the original time-series image stream, and calculate and output the corresponding image cleanliness coefficient. The decision configuration module is used to dynamically configure a hierarchical decision mechanism based on the image cleanliness coefficient and the real-time operating parameters of the target scraper conveyor. The hierarchical decision mechanism includes a visual correction mode, a prediction compensation mode, and a safety fault tolerance mode. The correction execution module is used to activate the corresponding decision mode based on the corrected decision threshold and the image cleanliness coefficient, generate a correction control quantity adapted to the current working condition, and drive the execution of adaptive correction.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application proposes a visual recognition-based adaptive correction method and system for scraper conveyors in coal mines. By acquiring images of the scraper conveyor in operation step-by-step, constructing an image cleanliness assessment model, dynamically configuring a hierarchical decision-making mechanism, generating adaptive correction control quantities, and driving execution, precise adaptive correction of the scraper conveyor in the complex environment of underground coal mines is achieved. First, raw time-series image streams of the scraper chain and trough side areas are continuously acquired using an explosion-proof visual acquisition device. Then, an image cleanliness assessment model is constructed to analyze the image stream frame by frame, outputting image cleanliness coefficients. Next, the basic decision threshold is corrected based on real-time operating parameters, and a hierarchical decision-making mechanism including visual correction, predictive compensation, and safety tolerance is dynamically configured. Then, based on the comparison between the image cleanliness coefficient and the corrected threshold, the corresponding decision mode is activated, and correction control quantities are generated through visual recognition, state prediction, data fusion, or non-visual sensor monitoring. Finally, the adaptive correction action is driven by the actuator, forming a closed-loop process of "image acquisition - quality assessment - decision adaptation - correction execution".

[0008] The technical solution of this application solves the problem that the single mode of traditional correction methods is difficult to adapt to the ever-changing dust interference and real-time fluctuations in working conditions in underground mines. It avoids correction interruption or malfunction when visual recognition fails. At the same time, through multi-dimensional compensation, data fusion and safety redundancy mechanisms, it improves the accuracy, stability and safety of scraper conveyor correction, and provides technical support for the efficient and stable operation of underground conveying equipment in coal mines. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic flowchart of the adaptive correction method for coal mine scraper conveyors based on visual recognition provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the adaptive correction system for a coal mine scraper conveyor based on visual recognition, provided in an embodiment of this application.

[0011] The components represented by each number in the attached diagram are explained below: Image acquisition module 01, quality assessment module 02, decision configuration module 03, and correction execution module 04. Detailed Implementation

[0012] This application provides a visual recognition-based adaptive correction method and system for coal mine scraper conveyors, which solves the technical problems in the prior art that make it difficult to cope with the high dust environment of coal mines and that false alarms, missed alarms and control disorders are easily caused by image quality degradation.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] Example 1, as shown in the appendix Figure 1 As shown, this application provides an adaptive correction method for coal mine scraper conveyors based on visual recognition. The method includes the following steps: S110: Continuously acquire the original time-series image stream of the scraper chain and trough side area of ​​the target scraper conveyor through an explosion-proof visual acquisition device; In this embodiment of the application, under the harsh working conditions of high dust and strong vibration in underground coal mines, the scraper conveyor is prone to generating continuous dust when transporting coal at high speed. In order to capture the dynamic relative position of the scraper chain and the trough side, and to provide continuous raw data support for subsequent image quality assessment and graded correction decision-making, it is necessary to use an explosion-proof visual acquisition device adapted to the special underground environment to acquire images, and to ensure the stability of the acquisition process and the integrity of the image information.

[0015] First, the explosion-proof visual acquisition device must meet the explosion-proof safety standards for underground coal mines, such as having ExdIMb-level explosion-proof certification and an anti-adhesion and anti-fogging coating sprayed on the lens surface to reduce the adhesion speed of coal dust to the lens surface and reduce the risk of image quality degradation. Simultaneously, the device must have strong vibration resistance to adapt to the high-frequency vibration environment of the scraper conveyor during operation, in order to avoid image blurring or acquisition interruption due to vibration.

[0016] Specifically, during deployment, the explosion-proof visual acquisition device must be aimed at the core monitoring area of ​​the scraper conveyor, namely the connection between the scraper chain and the trough side, to ensure that the image can completely cover the running trajectory of the scraper chain and the edge contour of the trough side. The installation location needs to be determined based on the structural characteristics of the conveyor, such as near the connection between the head, tail, and middle troughs, as these areas are prone to scraper chain deviation. The installation distance should be reasonably set according to the lens focal length. For example, when using a 20mm lens, the installation distance should be controlled at 1.2 meters to ensure complete coverage of the monitoring area while clearly presenting the detailed features of the scraper chain.

[0017] Furthermore, the acquisition of the original time-series image stream needs to balance continuity and timeliness, with an appropriate sampling frame rate set according to the operating speed of the scraper conveyor. For example, when the conveyor's operating speed is 2 m / s, the sampling frame rate is set to 25 fps to ensure real-time capture of the dynamic displacement of the scraper chain and avoid missing key operating states due to an excessively low frame rate. Simultaneously, a stable image transmission link is configured, employing an interference-resistant wired transmission method to ensure that the original time-series image stream can be transmitted to the subsequent processing modules without delay or loss.

[0018] For example, on a 280-meter-long scraper conveyor in a coal mine, six explosion-proof visual acquisition devices were deployed every 60 meters at the head, tail, and middle troughs. All devices were fixed with shock-absorbing brackets, with the lenses facing the connection area between the scraper chain and the trough wall, installed at a distance of 1.2 meters, and a sampling frame rate of 25fps. Under the condition that the conveyor was running at a speed of 2m / s and the load current was 160A, each device synchronously and continuously acquired raw time-series image streams. Each image stream generated 25 clear images per second, completely recording the running posture of the scraper chain and its relative position change with the trough wall, providing data support for subsequent calculation of image cleanliness coefficients.

[0019] S120: Construct an image cleanliness evaluation model, perform frame-by-frame analysis on the original time-series image stream, calculate and output the corresponding image cleanliness coefficient; In this embodiment of the application, in order to quantify the cleanliness of each frame of the image and distinguish the degree of influence of different dust interference on visual recognition, it is necessary to construct an image cleanliness evaluation model adapted to the special imaging environment downhole, and to conduct frame-by-frame analysis on the original time-series image stream to output an image cleanliness coefficient that can reflect the actual usability of the image, so as to provide a basis for the dynamic configuration of the subsequent hierarchical decision-making mechanism.

[0020] The method provided in this application includes the following steps in constructing the image cleanliness assessment model: Based on the operating environment of the scraper conveyor in the coal mine, a set of sample time-series images of the scraper chain and trough side area under different dust interference conditions was collected, and the cleanliness of each frame in the sample time-series image set was labeled to obtain the corresponding sample image cleanliness set. A network architecture for an image cleanliness assessment model is constructed based on deep learning. The image cleanliness assessment model is trained under supervision using the sample time-series image set and the sample image cleanliness set until the model validation accuracy converges, thus completing the model construction.

[0021] First, considering the unique operating environment of high dust and strong vibration in underground coal mines, time-series images of the scraper chain and trough sides were specifically collected. Specifically, the acquisition process needed to comprehensively cover different dust interference scenarios, including shots with slight dust floating, moderate dust adhesion, and situations where key visual features were obscured by heavy dust. Simultaneously, images under different operating speeds and load currents of the scraper conveyor were taken into account to ensure that the collection of time-series images could fully reflect the diversity of actual working conditions.

[0022] Meanwhile, the image resolution must be consistent with the actual acquisition resolution of the explosion-proof visual acquisition device, for example, both being 1920×1080 pixels, to ensure the compatibility of sample data with actual application data and avoid deviations in model evaluation accuracy due to resolution differences.

[0023] Furthermore, each sample image frame is labeled with its cleanliness level. The core criteria for labeling are the sharpness of the scraper chain edge, the integrity of the groove contour, and the degree to which dust interference obscures image details. A continuous numerical value of 0-1 is used for quantification. For example, if the scraper chain edge is sharp, the groove contour is undistorted, and there is no dust obscuration or only very slight dust traces, the label is 0.95; if key features are slightly blurred but details are still discernible, the label is 0.6; if dust covers a large area of ​​the monitoring area and key features are unclear, the label is 0.2.

[0024] Furthermore, after completing sample collection and annotation, a network architecture for an image cleanliness assessment model is constructed based on deep learning. Considering the core requirements of downhole image cleanliness assessment, a lightweight convolutional neural network, such as MobileNet, is selected as the basic architecture. This network possesses efficient feature extraction capabilities and low computational consumption, making it suitable for the limited hardware resources of downhole equipment.

[0025] During the architecture construction process, network parameters need to be set reasonably according to the scale of the sample data. For example, if the sample image scale is 50,000 frames, a network structure containing 11 convolutional layers and 3 fully connected layers can be constructed. The convolutional layers use 3×3 convolutional kernels (stride 1, same padding). Batch normalization layers and ReLU activation functions are inserted after the convolutional layers to enhance the model's ability to learn the correlation between dust interference features and image cleanliness, while reducing the risk of overfitting. If the sample data scale is small (e.g., less than 20,000 frames), the network can be simplified to a 9-layer architecture, reducing 2 convolutional layers, and introducing a transfer learning strategy. Based on the model pre-trained on a general image cleanliness dataset, fine-tuning is performed to reduce training difficulty and improve the model's generalization ability.

[0026] Furthermore, a sample time-series image set and a sample image cleanliness set are used to supervise the training of the constructed image cleanliness assessment model. During training, sample images are input into the model, and the model outputs predicted image cleanliness values. The predicted results are then compared with the true labels in the sample image cleanliness set. The prediction error is quantified by calculating the mean squared error loss function, and the network parameters are continuously adjusted based on the gradient descent algorithm to optimize the model's evaluation accuracy.

[0027] Simultaneously, the sample data needs to be divided into training, validation, and test sets in a reasonable ratio, such as a 7:2:1 ratio. The training set is used for iterative updates of model parameters, while the validation set is used for real-time monitoring of model performance during training to avoid overfitting. When the model validation accuracy on the validation set fluctuates by no more than 0.3% for 15 consecutive rounds, the model training is considered converged, and training is stopped. At this point, the model has a stable ability to evaluate image cleanliness.

[0028] Furthermore, during supervised training, the training effect needs to be periodically verified and adjusted. For example, after every 60 rounds of training, the model accuracy is tested using validation set data. If the accuracy growth is found to be slowing down, the learning rate can be adjusted from the initial 0.001 to 0.0006. If the model's cleanliness prediction error for images with moderate dust interference is found to be large, 800 additional sample images from moderate dust scenes are added and re-integrated into the sample set for training. If the validation set accuracy decreases, the batch size can be appropriately reduced (e.g., from 64 images / batch to 32 images / batch) to enhance the model's learning of complex features. This continues until the image cleanliness prediction accuracy for all dust scenes on the validation set is stable above 92% and shows no significant decrease for 10 consecutive rounds, ensuring that the model has a balanced and reliable evaluation capability.

[0029] Finally, the image cleanliness assessment model constructed through the above steps can quickly and accurately quantify the cleanliness of the input scraper conveyor image, output the image cleanliness coefficient in the range of 0-1, effectively distinguish the image quality differences under different dust interference, and provide solid technical support for the subsequent dynamic configuration of the hierarchical decision-making mechanism combined with real-time operating parameters.

[0030] Furthermore, after the image cleanliness assessment model is deployed, trained, and initialized, the original time-series image stream transmitted in real time by the explosion-proof visual acquisition device is input into the model frame by frame in the order of acquisition time. The model first performs preprocessing such as edge invalid region cropping and brightness standardization on each frame of the image, and then automatically extracts the feature information of the scraper chain and groove side, and performs matching calculation with different cleanliness feature templates learned during the training process to output the corresponding image cleanliness coefficient, so as to quantify the degree of dust interference and actual usable quality of each frame of the image.

[0031] This step involves real-time frame-by-frame analysis and quantification using the model to ensure that the output of the image cleanliness coefficient is synchronized with the operating status of the scraper conveyor. This avoids the lag in manual evaluation and ensures the consistency of quality judgment, providing reliable data support for the subsequent dynamic configuration of the hierarchical decision-making mechanism based on real-time operating parameters.

[0032] S130: Based on the image cleanliness coefficient and combined with the real-time operating parameters of the target scraper conveyor, a hierarchical decision-making mechanism is dynamically configured, wherein the hierarchical decision-making mechanism includes a visual correction mode, a prediction compensation mode, and a safety fault tolerance mode. In this embodiment of the application, in order to adapt the correction decision to different image quality and working condition combinations and avoid insufficient correction accuracy caused by fixed decision logic, it is necessary to dynamically configure a hierarchical decision mechanism by combining the image cleanliness coefficient and real-time working condition parameters to achieve correction and safety assurance in different scenarios.

[0033] Specifically, a basic first decision threshold and a basic second decision threshold corresponding to the hierarchical decision-making mechanism are first preset, and the basic first decision threshold is greater than the basic second decision threshold, so as to provide a benchmark reference for subsequent threshold correction.

[0034] Subsequently, the real-time operating parameters of the target scraper conveyor and the current image cleanliness coefficient are acquired simultaneously. The real-time operating parameters focus on the operating speed and load current to comprehensively capture the dynamic changes in the equipment's operating status.

[0035] Furthermore, the basic first and second decision thresholds are corrected based on the acquired real-time operating parameters to ensure that the corrected thresholds match the current equipment operating status and prevent erroneous activation of decision modes due to changes in operating conditions. Then, the current image cleanliness coefficient is compared with the two corrected decision thresholds, and the corresponding correction mode is activated based on the comparison result.

[0036] If the image cleanliness coefficient is greater than or equal to the corrected first decision threshold, it indicates that the image quality meets the requirements for direct recognition. The visual correction mode is activated to identify and solve the scraper chain position features of the image frames in the original time-series image stream.

[0037] In addition, if the image cleanliness coefficient is between the two corrected decision thresholds, it indicates that the image quality is slightly disturbed, and the prediction compensation mode is activated to perform visual enhancement processing, state prediction and data fusion on the image frame.

[0038] In addition, if the image cleanliness coefficient is less than the corrected second decision threshold, it indicates that the image quality cannot support visual recognition, and the safety fault-tolerant mode is activated to monitor the scraper chain for correction through non-visual sensors.

[0039] This step dynamically adjusts the decision threshold and matches the correction mode, enabling the correction decision to adapt to the complex downhole environment and equipment operating status in real time. This ensures the correction accuracy when the image quality is good, while also solving the problem of correction reliability when the image quality is poor.

[0040] Step S130 in the method provided in this application embodiment includes: A basic first decision threshold and a basic second decision threshold are preset to correspond to the hierarchical decision-making mechanism, wherein the basic first decision threshold is greater than the basic second decision threshold; The real-time operating parameters of the target scraper conveyor are obtained, and the current image cleanliness coefficient is obtained. The real-time operating parameters include the operating speed and load current. Based on the real-time operating parameters of the target scraper conveyor, the basic first decision threshold and the basic second decision threshold are corrected. The image cleanliness coefficient is compared with the corrected decision threshold: If the image cleanliness coefficient is greater than or equal to the corrected first decision threshold, then the visual correction mode is activated to perform scraper chain position feature recognition and calculation on the image frames in the original time-series image stream. If the image cleanliness coefficient is less than the corrected first decision threshold and greater than or equal to the corrected second decision threshold, then the prediction compensation mode is activated to perform visual enhancement processing, state prediction and data fusion on the image frames in the original time-series image stream. If the image cleanliness coefficient is less than the corrected second decision threshold, the safety fault-tolerant mode is activated to perform correction monitoring on the scraper chain based on non-visual sensors.

[0041] In this embodiment of the application, in order to enable the correction decision to adapt in real time to the changes in dust interference and the fluctuations in the operating status of the scraper conveyor in the coal mine, it is necessary to dynamically configure a hierarchical decision mechanism through a process of preset basic threshold, acquiring real-time data, correcting threshold and pattern matching, so as to achieve correction and safety redundancy under different working conditions.

[0042] Specifically, the first step is to pre-determine the basic first decision threshold and the basic second decision threshold corresponding to the hierarchical decision-making mechanism. The determination of the basic threshold requires offline analysis of a large amount of historical underground operation data to identify the correlation between the image cleanliness coefficient and the actual correction effect and equipment operation safety. At the same time, it is combined with the field experience of coal mine equipment operation and maintenance experts to conduct multiple rounds of calibration and verification to ensure that the basic threshold can provide a scientific benchmark for subsequent dynamic correction.

[0043] For example, by analyzing 100,000 frames of operational data under different dust interference and operating conditions, and combining the experts' requirements for correction accuracy and safety, the basic first decision threshold was finally calibrated to be 0.85 and the basic second decision threshold to be 0.45. The basic first decision threshold was always guaranteed to be greater than the basic second decision threshold, forming a clear mode switching gradient, which provides a premise for adjusting the decision logic according to image quality and operating conditions.

[0044] Furthermore, the real-time operating parameters of the target scraper conveyor and the current image cleanliness coefficient are acquired simultaneously. The real-time operating parameters focus on operating speed and load current. The operating speed is collected in real-time by the incremental encoder integrated into the scraper conveyor, with a sampling frequency set to 50Hz to ensure dynamic speed fluctuations are captured. In addition, the load current is collected by a current transformer in the motor control cabinet, with a data update cycle of 10ms to accurately reflect the load status of the drive motor.

[0045] Meanwhile, the current image cleanliness coefficient is output frame by frame by the image cleanliness assessment model, with a value range of 0-1, to quantify the degree of dust interference in the current image.

[0046] Furthermore, the basic first decision threshold and basic second decision threshold are dynamically corrected by combining real-time operating parameters, so that the corrected thresholds can match the current operating status of the equipment and avoid the decision mode being mistakenly activated due to changes in operating conditions.

[0047] The method provided in this application embodiment, which combines the real-time operating parameters of the target scraper conveyor to correct the basic first decision threshold and the basic second decision threshold, includes: Based on the operating speed of the target scraper conveyor, a speed correction factor is calculated, wherein the speed correction factor is positively correlated with the operating speed; Based on the load current of the target scraper conveyor, a load correction factor is calculated, wherein the load correction factor is positively correlated with the load current; Calculate the comprehensive operating condition influence coefficient based on the speed correction factor and the load correction factor; The comprehensive operating condition influence coefficient is multiplied by the benchmark first decision threshold and the benchmark second decision threshold respectively to obtain the corrected first decision threshold and the corrected second decision threshold.

[0048] Specifically, the speed correction factor is first calculated based on the operating speed of the target scraper conveyor. This speed correction factor reflects the deviation between the current operating speed and the standard state, and its value is positively correlated with the operating speed. The calculation formula can be expressed as "speed correction factor = operating speed / rated speed".

[0049] The rated speed is the standard operating speed calibrated at the factory when the scraper conveyor leaves the factory. It is a known fixed constant value. For example, if the rated speed of a certain model of scraper conveyor is set to 2m / s, and the current real-time operating speed is detected to be 1.6m / s, then the corresponding speed correction factor is 1.6 / 2=0.8; if the operating speed is increased to 2.2m / s, the speed correction factor is 2.2 / 2=1.1, so as to reflect the trend of the influence of speed change on threshold correction.

[0050] Furthermore, a load correction factor is calculated based on the load current of the target scraper conveyor. This load correction factor is used to quantify the difference between the current load and the rated load, and is positively correlated with the load current. The calculation formula can be expressed as "load correction factor = load current / rated current".

[0051] The rated current is the calibrated current value of the scraper conveyor drive motor under rated operating conditions. It is also a known fixed constant value. For example, if the rated current of the drive motor is 200A, and the current real-time load current is 180A, then the load correction factor is 180 / 200=0.9; if the load current increases to 210A, the load correction factor is 210 / 200=1.05, in order to capture the impact of load fluctuations on the correction decision.

[0052] Furthermore, based on the calculated speed correction factor and load correction factor, a comprehensive operating condition influence coefficient is calculated to integrate the dual influence of operating speed and load current on the decision threshold.

[0053] The method provided in this application embodiment calculates the comprehensive operating condition influence coefficient based on the speed correction factor and the load correction factor, including: Based on the current image cleanliness coefficient, the weight enhancement coefficient of the velocity correction factor is calculated, wherein the weight enhancement coefficient is negatively correlated with the current image cleanliness coefficient; Multiply the weight enhancement coefficient by the load correction factor to obtain the adjusted load correction factor; Calculate the weighted average of the speed correction factor and the adjusted load correction factor as the initial comprehensive operating condition influence coefficient; The initial comprehensive operating condition influence coefficient is normalized so that its value falls within a preset coefficient range, and the normalized value is output as the final comprehensive operating condition influence coefficient.

[0054] Specifically, the weight enhancement coefficient of the speed correction factor is first calculated based on the current image cleanliness coefficient. The role of the weight enhancement coefficient is to dynamically adjust the influence weight of the load factor according to the image quality, and it is negatively correlated with the current image cleanliness coefficient. That is, the more blurred the image, the larger the weight enhancement coefficient, thereby strengthening the decision-making impact of load fluctuations in scenarios where the reliability of visual recognition decreases.

[0055] For example, the current image cleanliness coefficient is 0.8, and the corresponding weight enhancement coefficient is 0.3; if dust interference intensifies and causes the cleanliness coefficient to drop to 0.3, the weight enhancement coefficient will be increased to 0.8 to ensure that the key role of load factors is fully reflected under poor image quality.

[0056] Furthermore, the calculated weight enhancement coefficient is multiplied by the load correction factor to obtain the adjusted load correction factor, thereby achieving the correlation between load impact and image quality. For example, under a certain working condition, the load correction factor is 0.9. Combined with the weight enhancement coefficient of 0.3 corresponding to the cleanliness coefficient of 0.8, the adjusted load correction factor = 0.9 × 0.3 = 0.27. If the cleanliness coefficient is reduced to 0.3 and the weight enhancement coefficient is 0.8, the adjusted value corresponding to the same load correction factor is 0.9 × 0.8 = 0.72, thus reflecting the dynamic adjustment effect of image quality on the weight of load impact.

[0057] Furthermore, the weighted average of the speed correction factor and the adjusted load correction factor is calculated as the initial comprehensive operating condition influence coefficient. To balance the influence of the two types of parameters, the weight of both the speed correction factor and the adjusted load correction factor is set to 0.5, and the multi-dimensional influence is integrated through an arithmetic mean.

[0058] For example, if the current speed correction factor is 1.1 and the adjusted load correction factor is 0.27, the initial comprehensive operating condition influence coefficient is (1.1+0.27) / 2=0.685; if the speed correction factor is 0.9 and the adjusted load correction factor is 0.72, the initial comprehensive operating condition influence coefficient is (0.9+0.72) / 2=0.81, in order to further integrate the triple influence of speed, load and image quality.

[0059] Furthermore, the initial comprehensive working condition influence coefficient is normalized to avoid the coefficient exceeding a reasonable range under extreme working conditions, which could lead to over- or under-correction of the decision threshold. The preset coefficient range needs to be determined based on the actual working conditions in the coal mine, for example, set to 0.5-1.5, to ensure that the coefficient always remains within an effective adaptation range.

[0060] For example, if the initial comprehensive operating condition influence coefficient is 1.6, which exceeds the upper limit of the interval, it is adjusted to 1.5 after normalization; if the initial coefficient is 0.4, which is lower than the lower limit of the interval, it is corrected to 0.5 after normalization; if the initial coefficient is 0.81, which is within the interval, the original value is kept unchanged, and a stable comprehensive operating condition influence coefficient is finally output.

[0061] Furthermore, after obtaining the comprehensive operating condition influence coefficient, the coefficient is multiplied by the preset first and second benchmark decision thresholds respectively. Through this quantitative adaptation method, the corrected first and second decision thresholds that are accurately matched with the current operating conditions are obtained, so that the decision thresholds can dynamically respond to changes in the equipment operating status and image quality.

[0062] For example, if the preset first decision threshold is 0.85, the second decision threshold is 0.45, and the comprehensive operating condition influence coefficient is 0.81, multiplying them by the benchmark thresholds respectively, the corrected first decision threshold = 0.85 × 0.81 = 0.6885, and the corrected second decision threshold = 0.45 × 0.81 = 0.3645. At this time, the threshold is appropriately lowered according to the operating conditions to adapt to the relatively mild operating state.

[0063] Furthermore, if the comprehensive working condition influence coefficient is 1.5, then the corrected first decision threshold = 0.85 × 1.5 = 1.275, and the corrected second decision threshold = 0.45 × 1.5 = 0.675. The thresholds are adjusted accordingly to cope with complex working conditions such as high speed and heavy load.

[0064] Conversely, if the comprehensive operating condition influence coefficient is 0.5, then the corrected first decision threshold = 0.85 × 0.5 = 0.425, and the corrected second decision threshold = 0.45 × 0.5 = 0.225, to ensure the rationality of decision mode switching under low-speed and light-load operating conditions.

[0065] Furthermore, after obtaining the corrected first decision threshold and the corrected second decision threshold, the current image cleanliness coefficient is compared with these two corrected thresholds one by one. By judging the image quality and the corrective ability to adapt to the current working conditions, the corresponding decision mode is activated to ensure that the corrective strategy is both consistent with the reliability of image recognition and can match the operating status of the equipment.

[0066] If the current image cleanliness coefficient is greater than or equal to the corrected first decision threshold, it indicates that the image quality is at a good level, key visual features such as the scraper chain edge and groove contour are clearly distinguishable, and the reliability of visual recognition is fully adapted to the current working conditions. At this time, the visual correction mode is activated.

[0067] Specifically, when the visual correction mode is activated, the current frame image in the original time-series image stream is first processed. The edge of the scraper chain and the preset groove baseline are identified through edge detection and feature extraction algorithms. Then, the lateral offset and deflection angle of the scraper chain are calculated based on the geometric relationship to generate a correction command that adapts to the current offset state, driving the actuator to respond and adjust quickly.

[0068] Among them, the edge detection and feature extraction algorithms can adopt the Canny edge detection algorithm and the SIFT feature extraction algorithm to effectively suppress the interference of downhole dust and noise and capture the contour features of the scraper chain and the groove side.

[0069] In addition, the preset trough side baseline is determined by fitting the average position of the trough side edge by collecting multiple frames of standard images without dust interference during the no-load operation phase after equipment installation, so as to ensure the consistency between the baseline and the actual running trajectory of the scraper conveyor.

[0070] For example, when the corrected first decision threshold is 0.6885 and the current image cleanliness coefficient is 0.75, the lateral offset is calculated to be 5mm and the deflection angle is 1.2°. Then, based on the calculation result, a targeted correction command is generated and sent to the hydraulic correction actuator through the industrial bus. The actuator is driven to respond quickly and adjust the scraper chain to the preset running trajectory to ensure the accuracy and timeliness of the correction.

[0071] Furthermore, if the current image cleanliness coefficient is less than the corrected first decision threshold and greater than or equal to the corrected second decision threshold, it indicates that the image is slightly affected by dust and the key features are slightly blurred, but it still has certain extraction value. The accuracy of simple visual recognition is insufficient, and it is necessary to combine it with working condition prediction for compensation. At this time, the prediction compensation mode is activated to perform visual enhancement processing, state prediction and data fusion on the image frames in the obtained original time-series image stream to obtain the fusion position of the scraper chain.

[0072] The method provided in this application embodiment activates a prediction compensation mode to perform visual enhancement processing, state prediction, and data fusion on image frames in the original temporal image stream, including: The image frame is subjected to visual enhancement processing based on the image cleanliness coefficient to extract the visual observation position of the scraper chain; Based on the real-time operating parameters of the target scraper conveyor, the state and position of the scraper chain are predicted. Based on the image cleanliness coefficient, the visual fusion weight is calculated within the interval formed by the modified first decision threshold and the modified second decision threshold; Based on the visual fusion weights, the visual observation position and the state prediction position are weighted and fused to obtain the fused position.

[0073] Specifically, the image frames in the original time-series image stream are first subjected to visual enhancement processing based on the image cleanliness coefficient. The core is to dynamically configure the enhancement algorithm parameters according to the degree of dust interference in the image to ensure that the enhancement effect is adapted to the image quality.

[0074] For example, when the image cleanliness coefficient is 0.5, it indicates that the image is slightly blurry. In this case, an adaptive median filtering algorithm is used to suppress dust noise, while histogram equalization is used to improve the contrast between the scraper chain and the background. If the cleanliness coefficient is 0.65, the dust interference is relatively light, so only light Gaussian filtering and edge sharpening are enabled to avoid feature distortion caused by over-enhancement.

[0075] After enhancement processing, the edges of the scraper chain are identified using the Canny edge detection algorithm. Combined with a preset groove baseline, the lateral offset and deflection angle of the scraper chain are calculated based on geometric relationships, and this is used as the visual observation position. For example, in the enhanced image, the edges of the scraper chain are clearly distinguishable, and the calculated lateral offset is 4.2mm and the deflection angle is 0.8°. This set of data is the visual observation position of the current frame.

[0076] Furthermore, based on the real-time operating parameters of the target scraper conveyor, the state prediction position of the scraper chain is predicted to provide a reliable position reference that is not affected by image quality, thus complementing the visual observation position.

[0077] The method provided in this application embodiment predicts the state prediction position of the scraper chain based on the real-time operating parameters of the target scraper conveyor, including: Based on the operating speed of the target scraper conveyor, calculate the theoretical basic displacement of the scraper chain within the current sampling period; Based on the load current of the target scraper conveyor and combined with the load displacement influence coefficient, the theoretical basic displacement is corrected by load to obtain the load-corrected displacement. Based on the current image cleanliness coefficient, the displacement compensation amount is determined, and the displacement compensation amount is superimposed with the load correction displacement amount to obtain the final predicted displacement amount; The final predicted displacement is superimposed with the merged position of the scraper chain obtained in the previous sampling period to calculate the predicted state position of the scraper chain.

[0078] Specifically, firstly, based on the operating speed of the target scraper conveyor, the theoretical basic displacement of the scraper chain within the current sampling period is calculated. This displacement is a basic reference value based on the motion characteristics of the equipment, and the calculation formula can be expressed as "theoretical basic displacement = operating speed × sampling period duration".

[0079] The sampling period is set according to the operating response requirements of the scraper conveyor, for example, 0.02s. If the current real-time operating speed is 1.8m / s, then the theoretical basic displacement is 1.8m / s × 0.02s = 0.036m, which reflects the basic motion displacement of the scraper chain under no load interference.

[0080] Furthermore, based on the load current of the target scraper conveyor, the theoretical displacement is corrected by load displacement influence coefficient to obtain the corrected displacement.

[0081] The load displacement influence coefficient is a dimensionless constant calibrated using a large amount of downhole experimental data. It is used to quantify the influence of load changes on the scraper chain displacement. For example, the value is 0.12 mm / A. The calculation formula can be expressed as "load correction displacement = [(load current × load displacement influence coefficient) + 1] × theoretical basic displacement".

[0082] Furthermore, a displacement compensation amount is determined based on the current image cleanliness coefficient. This displacement compensation amount is used to correct prediction deviations when the image quality is poor, so as to ensure that the position prediction is consistent with the actual state.

[0083] The method provided in this application embodiment, which determines the displacement compensation amount based on the current image cleanliness coefficient, includes: Based on the current image cleanliness coefficient, a first displacement compensation component is calculated, wherein the first displacement compensation component is negatively correlated with the current image cleanliness coefficient; The second displacement compensation component is calculated based on the operating speed, and the third displacement compensation component is calculated based on the load current. Based on the current image cleanliness coefficient, the second displacement compensation component and the third displacement compensation component are weighted and fused to obtain the fourth displacement compensation component. The first displacement compensation component is superimposed with the fourth displacement compensation component to obtain the total displacement compensation amount. Based on the difference between the total displacement compensation and the displacement compensation calculated in the previous sampling period, a first-order low-pass filter is applied for smoothing to output the final displacement compensation.

[0084] First, the first displacement compensation component is calculated based on the current image cleanliness coefficient. This first displacement compensation component compensates for the impact of poor image quality on displacement prediction and is negatively correlated with the current image cleanliness coefficient; the more blurry the image, the larger the compensation component.

[0085] Specifically, the calculation formula for the first displacement compensation component can be expressed as "first displacement compensation component = (corrected first decision threshold - current image cleanliness coefficient) × cleanliness compensation coefficient". The cleanliness compensation coefficient is determined through extensive downhole experimental data, and can be, for example, taken as 0.5 mm / unit cleanliness.

[0086] For example, if the corrected first decision threshold is 0.6885 and the current image cleanliness coefficient is 0.5, substituting into the formula, we can obtain the first displacement compensation component = (0.6885-0.5)×0.5mm = 0.09425mm, which reflects the degree of influence of image quality on displacement prediction.

[0087] Furthermore, the second and third displacement compensation components are calculated based on the operating speed and load current, respectively, to cover the prediction deviation caused by operating condition fluctuations. Specifically, the calculation formula for the second displacement compensation component can be expressed as "Second displacement compensation component = Operating speed × Speed ​​compensation coefficient". The speed compensation coefficient is a constant obtained by analyzing the correlation between speed fluctuations and displacement prediction errors in historical operating data and calibrating it through linear fitting; for example, it can be taken as 0.005s.

[0088] For example, if the current operating speed of the target scraper conveyor is 1.8 m / s, then the second displacement compensation component = 1.8 m / s × 0.005 s = 0.009 m = 9 mm.

[0089] Furthermore, the calculation formula for the third displacement compensation component can be expressed as "Third displacement compensation component = Load current × Current compensation coefficient". The current compensation coefficient is obtained through offline calibration experiments; for example, it can be taken as 0.08 mm / A. If the current load current is 160A, then the third displacement compensation component = 160A × 0.08 mm / A = 12.8 mm, quantifying the influence of speed and load on displacement prediction.

[0090] Furthermore, the second and third displacement compensation components are weighted and fused based on the current image cleanliness coefficient to obtain the fourth displacement compensation component, thereby achieving dynamic adaptation between working condition influence and image quality.

[0091] Specifically, the core of the fusion process is calculating the weighting coefficient, which is expressed by the formula: "Weighting coefficient = (corrected first decision threshold - current image cleanliness coefficient) / (corrected first decision threshold - corrected second decision threshold)". This weighting coefficient varies with the image cleanliness coefficient; the more blurred the image, the larger the weighting coefficient, and the higher the influence of the third displacement compensation component corresponding to the load current.

[0092] For example, if the corrected second decision threshold is 0.3645 and the current image cleanliness coefficient is 0.5, then the weight coefficient = (0.6885-0.5) / (0.6885-0.3645)≈0.5818.

[0093] Furthermore, the calculation formula for the fourth displacement compensation component can be expressed as "fourth displacement compensation component = weighting coefficient × third displacement compensation component + (1 - weighting coefficient) × second displacement compensation component". Substituting the values, we can obtain: fourth displacement compensation component = 0.5818 × 12.8 mm + 0.4182 × 9 mm ≈ 11.21 mm, which further realizes the dynamic balance between speed and load influence.

[0094] Furthermore, the first displacement compensation component and the fourth displacement compensation component are superimposed to obtain the total displacement compensation amount, thus integrating the dual compensation for image quality and operating condition fluctuations. For example, if the first displacement compensation component is 0.09425mm and the fourth displacement compensation component is 11.21mm, then the total displacement compensation amount = 0.09425mm + 11.21mm ≈ 11.30mm, which covers the compensation requirements of all key influencing factors.

[0095] Finally, based on the difference between the total displacement compensation and the displacement compensation calculated in the previous sampling period, a first-order low-pass filter is applied for smoothing to avoid fluctuations in the predicted position caused by sudden changes in the compensation amount.

[0096] Among them, the first-order low-pass filter is a digital domain first-order IIR (infinite impulse response) low-pass filter. Its core advantages are simple structure, low computational load, adaptability to the real-time processing requirements of downhole control terminals, and effective suppression of high-frequency fluctuation noise in displacement compensation.

[0097] For example, if the displacement compensation amount in the previous sampling period is 10.921 mm and the total displacement compensation amount is 11.3 mm, the difference between the two is 0.379 mm. After processing by a first-order low-pass filter, the final output displacement compensation amount is 11.0 mm, so as to ensure the continuity and stability of the compensation amount.

[0098] Furthermore, after obtaining the final displacement compensation amount, this displacement compensation amount is superimposed with the previously calculated load correction displacement amount to obtain the final predicted displacement amount, so as to realize the fusion of load correction and image quality and working condition fluctuation compensation.

[0099] Specifically, the load-corrected displacement is a displacement reference calculated based on the operating speed and corrected for the load current. For example, if the load-corrected displacement is 36.7 mm, the final displacement compensation is 11.0 mm. When the two are added together, the final predicted displacement is 36.7 mm + 11.0 mm = 47.7 mm, which further integrates the effects of multi-dimensional correction and compensation.

[0100] Furthermore, the final predicted displacement is superimposed with the fusion position of the scraper chain obtained in the previous sampling period to calculate the predicted state position of the scraper chain, so as to ensure that the predicted state position can continuously reflect the running trajectory of the scraper chain and seamlessly connect with the state of the previous period.

[0101] For example, if the fusion position of the previous sampling period is 10523mm and the final predicted displacement is 47.7272mm, then the state prediction position = 10523mm + 47.7mm = 10570.7mm. This result covers the influence of multiple factors such as running speed, load current, and image quality, and provides a position reference for visual fusion in subsequent prediction compensation mode.

[0102] Furthermore, based on the current image cleanliness coefficient, a visual fusion weight is calculated within the interval formed by the modified first decision threshold and the modified second decision threshold. This weight is used to dynamically allocate the influence ratio of the visual observation position and the state prediction position, so that the fusion result is adapted to the current image quality.

[0103] Specifically, the calculation formula can be expressed as "Visual fusion weight = (current image cleanliness coefficient - corrected second decision threshold) / (corrected first decision threshold - corrected second decision threshold)". The weight value changes linearly and continuously with the image cleanliness coefficient. The clearer the image, the greater the weight, and the higher the proportion of the influence of the visual observation position.

[0104] For example, if the corrected first decision threshold is 0.6885, the second decision threshold is 0.3645, and the current image cleanliness coefficient is 0.5, substituting into the formula, we can get the visual fusion weight = (0.5-0.3645) / (0.6885-0.3645)≈0.412, that is, the influence of the visual observation position accounts for 41.2%, and the influence of the state prediction position accounts for 58.8%.

[0105] Furthermore, based on the calculated visual fusion weights, the visual observation position and the state prediction position are weighted and fused to obtain the fused position. This integrates the advantages of both types of position data, improving the accuracy and reliability of the scraper chain position determination. Specifically, the fusion calculation formula is "fused position = visual fusion weight × visual observation position + (1 - visual fusion weight) × state prediction position".

[0106] For example, if the previously extracted visual observation position is 10598mm, the state prediction position is 10570.7mm, and the visual fusion weight is 0.412, substituting the values, we get: fusion position = 0.412 × 10598mm + 0.588 × 10570.7mm ≈ 10581.5mm. This fusion position retains the accurate capture of the current frame position by visual recognition, and also compensates for the errors caused by image interference through condition prediction.

[0107] In addition, if the current image cleanliness coefficient is less than the corrected second decision threshold, it means that the image is heavily obscured by dust, and key visual features cannot be identified at all. Visual correction and prediction compensation cannot guarantee reliability. At this time, the safety fault tolerance mode is activated, and correction is achieved through independent monitoring by non-visual sensors to ensure the safe operation of the scraper conveyor.

[0108] The method provided in this application embodiment activates a safety fault-tolerant mode and performs correction monitoring of the scraper chain based on non-visual sensors, including: The deviation correction decision based on the original time-series image stream is interrupted, and the deviation status is judged based on the monitoring signal of the independent monitoring device. The monitoring signal output by the independent monitoring device is continuously acquired, wherein the independent monitoring device includes a laser beam sensor and a mechanical deviation switch; Based on the monitoring signal, determine whether the scraper chain has triggered the deviation alarm condition; If the deviation alarm condition is triggered, a preset safety correction instruction is generated and executed, wherein the preset safety correction instruction includes periodic correction actions with a lower amplitude than normal. A mode degradation alarm is sent synchronously to the monitoring center. The mode degradation alarm includes the current image cleanliness coefficient and the type of independent monitoring device that triggered it.

[0109] Specifically, all deviation correction decision logic based on the original time-series image stream is first interrupted. At this point, the image is so heavily obscured by dust that it is impossible to extract effective position features. Continuing to rely on visual data will lead to deviation correction failure or malfunction. The system quickly switches to using the monitoring signal from an independent monitoring device to determine the deviation status. The independent monitoring device has no hardware connection with the visual acquisition system and operates completely independently to ensure the reliability of the monitoring data and the timeliness of the decision switching.

[0110] The independent monitoring device includes a laser beam sensor and a mechanical deviation switch. The laser beam sensor is evenly deployed along both sides of the scraper conveyor body, with a spacing of 2 meters according to the equipment width. The distance between each sensor and the scraper chain running trajectory is set to 5 mm. It transmits and receives laser signals in real time, and calculates the lateral deviation of the scraper chain by the degree of signal obstruction.

[0111] In addition, the mechanical deviation switch is installed at the extreme edge of the scraper chain's running trajectory. It adopts a spring-triggered structure, and the triggering force has been experimentally calibrated to 5N. As a physical redundancy protection for the laser sensor, the dual devices work together to cover the entire scenario of deviation monitoring.

[0112] Furthermore, the monitoring signals output by the independent monitoring device are continuously acquired. The sampling frequency of the laser beam sensor is set to 100Hz. The monitoring data is transmitted to the control module in real time via the CAN bus in the form of digital signals. The data update cycle does not exceed 10ms to ensure that the slight offset trend of the scraper chain can be captured quickly.

[0113] In addition, the trigger signal of the mechanical misalignment switch is a passive switching quantity. Once the scraper chain touches the switch contact, a high-level trigger signal is immediately generated and uploaded, without the need for complex data processing, so as to ensure the response time in extreme misalignment scenarios.

[0114] Simultaneously, the acquired monitoring signals are validated to remove abnormal data caused by downhole electromagnetic interference or dust adhering to the sensor surface. For example, if the offset detected by the laser sensor for three consecutive frames shows a sudden change in amplitude exceeding 5mm without a continuous trend, it is determined to be an invalid signal, and the offset data from the previous valid cycle is automatically used as a replacement. The trigger signal of the mechanical deviation switch is subjected to anti-jitter processing, with a signal holding time of 20ms set to avoid false triggering caused by scraper chain vibration, thus ensuring the accuracy of signal analysis.

[0115] Furthermore, based on the verified monitoring signals, it is determined whether the scraper chain has triggered a deviation alarm condition. The alarm conditions are preset to two scenarios, and meeting either one is considered an alarm state. Specifically, one is that the lateral deviation of the scraper chain detected by the laser beam sensor reaches or exceeds 10mm for 50ms. This threshold has been calibrated through a large number of downhole experiments to avoid frequent alarms caused by slight deviations, while also being able to respond promptly to serious deviations that may damage the equipment.

[0116] Secondly, the mechanical misalignment switch is triggered, meaning the scraper chain has touched the physical switch at its limit edge, indicating that the misalignment has endangered equipment safety. For example, if the laser beam sensor detects a lateral offset of 11mm for 60ms continuously, or if the scraper chain touches the mechanical misalignment switch due to severe misalignment, the misalignment alarm condition is immediately triggered.

[0117] Specifically, if the deviation alarm condition is triggered, a preset safety correction command is generated and executed. This command is a conservative correction strategy adapted to extreme working conditions. The core principle is small-amplitude, periodic adjustments to avoid secondary risks such as collision between the scraper chain and the groove side, chain stretching and deformation, or material spillage caused by large-amplitude and rapid correction.

[0118] Among them, in the preset safety correction command, the correction amplitude is set to 2mm / time, which is lower than the adjustment amplitude of 5-8mm in the conventional visual correction mode. The correction cycle is set to 0.5s, and the scraper chain is slowly pushed to reset by the hydraulic actuator.

[0119] For example, the generated correction command is "control the hydraulic valve group to output pressure of 0.8MPa, push the correction roller to deviate by 2mm, maintain for 0.5s and then detect the deviation. If the deviation does not meet the standard, repeat the execution" until the lateral deviation detected by the laser beam sensor drops below 5mm, and then stop the periodic correction action.

[0120] Simultaneously, a mode degradation alarm is sent to the ground monitoring center. The alarm information is uploaded in real time via industrial Ethernet and includes key data such as the current image cleanliness coefficient, the type of independent monitoring device triggered (laser beam sensor or mechanical deviation switch), and the current operating speed and load current of the scraper conveyor, which facilitates maintenance personnel to quickly determine the cause of the fault.

[0121] Ultimately, through independent monitoring, accurate alarms, conservative correction, and real-time reporting, a reliable safety fallback mechanism was built in the extreme scenario where visual recognition completely fails. This ensures the continuous operation of the scraper conveyor while minimizing the risk of equipment damage caused by deviation, providing comprehensive protection for safe production underground.

[0122] S140: Based on the corrected decision threshold and the image cleanliness coefficient, activate the corresponding decision mode, generate a correction control quantity adapted to the current working condition, and drive the execution of adaptive correction.

[0123] In this embodiment of the application, in order to avoid the scraper chain running off track and equipment wear caused by insufficient adaptation of a single correction mode, it is necessary to activate the adaptation decision mode through threshold comparison, generate the correction control quantity that fits the working condition and drive its execution, so as to achieve adaptive correction and ensure the continuous and stable operation of the equipment.

[0124] First, the activation determination of the decision mode is performed, which is a prerequisite for generating the adaptive control quantity. The corrected first and second decision thresholds have incorporated the influence of real-time operating conditions, while the current image cleanliness coefficient quantifies the reliability of visual recognition. By directly comparing these three, the most suitable correction mode is determined.

[0125] Specifically, the comparison process employs real-time parallel computing logic. The control module completes the judgment within 10ms of obtaining the corrected threshold and the current cleanliness coefficient, ensuring the timeliness of mode switching and avoiding the exacerbation of deviation due to decision delay.

[0126] For example, the corrected first decision threshold is 0.6885 and the second decision threshold is 0.3645. When the current image cleanliness coefficient is 0.75, it is determined that the visual recognition is reliable and the visual correction mode is activated; when the current image cleanliness coefficient is 0.5, it is determined that there is slight interference and the prediction compensation mode is activated; when the current image cleanliness coefficient is 0.3, it is determined that there is visual failure and the safety fault tolerance mode is activated.

[0127] Furthermore, for different activation modes, correction control quantities are generated that are deeply adapted to the working conditions and image quality. Specifically, in visual correction mode, the control quantity is generated based on the position feature recognition results of the scraper chain. The core is to convert the identified lateral offset and deflection angle into the action parameters of the actuator. For example, if the identified lateral offset is 5mm and the deflection angle is 1.2°, combined with the width of the scraper conveyor body and the response characteristics of the actuator, it is converted into a push rod extension of 3mm and an action time of 0.3s for the hydraulic correction actuator. This control quantity is output in the form of a digital signal to ensure that the correction action accurately corresponds to the offset state.

[0128] Furthermore, in predictive compensation mode, the control quantity is generated based on the deviation between the fused position and the preset operating trajectory. The fused position integrates the advantages of both visual observation and operational condition prediction, and the deviation calculation adopts real-time comparison logic. For example, if the fused position is 10581.5mm and the preset trajectory position is 10580mm, the deviation is 1.5mm, which corresponds to generating a control quantity with a push rod extension of 1.5mm and an action time of 0.3s, in order to balance the correction accuracy and the smoothness of the action.

[0129] Furthermore, in the safety-tolerant mode, the control parameters are preset conservative parameters that are dynamically adjusted based on monitoring signals from non-visual sensors. For example, when the laser beam sensor detects a lateral offset of 11mm, it generates a periodic control quantity with a push rod extension of 2mm and an action cycle of 0.5s, avoiding equipment impact caused by large movements and ensuring that the correction process is safe and controllable.

[0130] Finally, the drive actuator completes the adaptive correction action. The actuator is a hydraulic correction actuator, which has the characteristics of large output force, fast response speed and strong resistance to dust pollution, and is suitable for the harsh environment of underground coal mines.

[0131] The control signals are transmitted to the actuator controller via a CAN bus in differential signal form, with the transmission baud rate set to 500kbps to ensure interference resistance and real-time performance. After receiving the control signals, the actuator adjusts the hydraulic pressure through a proportional relief valve to drive the correction rollers along a preset trajectory, with the movement error controlled within ±0.1mm.

[0132] Meanwhile, the actuator has a built-in displacement sensor that collects the actual position of the idler roller in real time to form a closed-loop feedback. If the deviation between the feedback value and the target position corresponding to the control quantity exceeds 0.3mm, the controller will automatically adjust the output signal until the deviation meets the requirements.

[0133] This step, through the process of "precise mode activation - targeted generation of control quantities - closed-loop action of actuators," achieves dynamic adaptation of the correction strategy to the environment and working conditions. It not only ensures high-precision correction when visual recognition is reliable, but also covers complex interference and extreme failure scenarios through predictive compensation and safety fault tolerance modes, thus constructing an adaptive correction closed loop under all working conditions and providing a guarantee for the stable operation of the scraper conveyor.

[0134] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects: This application proposes an adaptive correction method for coal mine scraper conveyors based on visual recognition. First, an explosion-proof visual acquisition device continuously acquires raw time-series image streams of the scraper chain and trough side areas, ensuring the acquired data is adapted to the harsh underground environment of high dust and strong vibration, providing continuous and complete raw data for subsequent analysis. Next, an image cleanliness assessment model is constructed, analyzing the raw time-series image stream frame by frame and outputting image cleanliness coefficients to quantify the impact of dust interference on image quality. Then, based on the image cleanliness coefficients and the real-time operating parameters of the scraper conveyor, a basic decision threshold is corrected by calculating speed correction factors, load correction factors, and a comprehensive operating condition influence coefficient, dynamically configuring a hierarchical decision mechanism including visual correction, predictive compensation, and safety fault tolerance. Next, the corresponding decision mode is activated based on the comparison between the image cleanliness coefficients and the corrected thresholds. Finally, a correction control quantity adapted to the current operating conditions is generated, driven by a hydraulic actuator to complete the adaptive correction, forming a closed-loop control process.

[0135] The method provided in this application, through the technical solution of "image acquisition - quality assessment - threshold correction - mode adaptation - correction execution", solves the problem that traditional correction methods are difficult to adapt to the variable dust interference and real-time fluctuations in working conditions in underground mines. It avoids correction interruption or malfunction when visual recognition fails, improves the stability and safety of scraper conveyor correction, and provides technical support for the stable operation of underground conveying equipment in coal mines.

[0136] Example 2, as shown in the appendix Figure 2 As shown, based on the inventive concept of the visual recognition-based adaptive correction method for coal mine scraper conveyors provided in Embodiment 1, this application also provides a visual recognition-based adaptive correction system for coal mine scraper conveyors, specifically including: Image acquisition module 01 is used to continuously acquire the original time-series image stream of the scraper chain and trough side area of ​​the target scraper conveyor through an explosion-proof vision acquisition device; Quality assessment module 02 is used to construct an image cleanliness assessment model, perform frame-by-frame analysis on the original time-series image stream, calculate and output the corresponding image cleanliness coefficient; The decision configuration module 03 is used to dynamically configure a hierarchical decision mechanism based on the image cleanliness coefficient and the real-time operating parameters of the target scraper conveyor. The hierarchical decision mechanism includes a visual correction mode, a prediction compensation mode, and a safety fault tolerance mode. The correction execution module 04 is used to activate the corresponding decision mode based on the corrected decision threshold and the image cleanliness coefficient, generate a correction control quantity adapted to the current working condition, and drive the execution of adaptive correction.

[0137] In one embodiment, the quality assessment module 02 is further used for: Based on the operating environment of the scraper conveyor in the coal mine, a set of sample time-series images of the scraper chain and sidewall area under different dust interference conditions is collected. Each frame in the sample time-series image set is labeled with image cleanliness to obtain the corresponding sample image cleanliness set. A network architecture for an image cleanliness evaluation model is constructed based on deep learning. The image cleanliness evaluation model is then trained under supervised supervision using the sample time-series image set and the sample image cleanliness set until the model validation accuracy converges, thus completing the model construction.

[0138] In one embodiment, the decision configuration module 03 is further configured to: A first and a second basic decision threshold are preset to correspond to the hierarchical decision-making mechanism, wherein the first basic decision threshold is greater than the second basic decision threshold. Real-time operating parameters of the target scraper conveyor are obtained, and the current image cleanliness coefficient is acquired, wherein the real-time operating parameters include operating speed and load current. The first and second basic decision thresholds are corrected based on the real-time operating parameters of the target scraper conveyor. The image cleanliness coefficient is compared with the corrected decision thresholds: if the image cleanliness coefficient is greater than or equal to the corrected first decision threshold, a visual correction mode is activated to identify and calculate the scraper chain position features of the image frames in the original time-series image stream; if the image cleanliness coefficient is less than the corrected first decision threshold but greater than or equal to the corrected second decision threshold, a prediction compensation mode is activated to perform visual enhancement processing, state prediction, and data fusion on the image frames in the original time-series image stream; if the image cleanliness coefficient is less than the corrected second decision threshold, a safety fault-tolerant mode is activated to perform correction monitoring of the scraper chain based on non-visual sensors.

[0139] Furthermore, the decision configuration module 03 also includes: Based on the operating speed of the target scraper conveyor, a speed correction factor is calculated, wherein the speed correction factor is positively correlated with the operating speed; based on the load current of the target scraper conveyor, a load correction factor is calculated, wherein the load correction factor is positively correlated with the load current; based on the speed correction factor and the load correction factor, a comprehensive operating condition influence coefficient is calculated; the comprehensive operating condition influence coefficient is multiplied by a benchmark first decision threshold and a benchmark second decision threshold respectively to obtain a corrected first decision threshold and a corrected second decision threshold.

[0140] Furthermore, the decision configuration module 03 also includes: Based on the current image cleanliness coefficient, the weight enhancement coefficient of the speed correction factor is calculated, wherein the weight enhancement coefficient is negatively correlated with the current image cleanliness coefficient; the weight enhancement coefficient is multiplied by the load correction factor to obtain the adjusted load correction factor; the weighted average of the speed correction factor and the adjusted load correction factor is calculated as the initial comprehensive operating condition influence coefficient; the initial comprehensive operating condition influence coefficient is normalized so that its value falls within a preset coefficient range, and the normalized value is output as the final comprehensive operating condition influence coefficient.

[0141] Furthermore, the decision configuration module 03 also includes: The image frame is subjected to visual enhancement processing based on the image cleanliness coefficient to extract the visual observation position of the scraper chain; the state prediction position of the scraper chain is predicted based on the real-time operating parameters of the target scraper conveyor; a visual fusion weight is calculated within the interval formed by the modified first decision threshold and the modified second decision threshold according to the image cleanliness coefficient; and the visual observation position and the state prediction position are weighted and fused based on the visual fusion weight to obtain the fused position.

[0142] Furthermore, the decision configuration module 03 also includes: Based on the operating speed of the target scraper conveyor, the theoretical basic displacement of the scraper chain within the current sampling period is calculated. Based on the load current of the target scraper conveyor and combined with the load displacement influence coefficient, the theoretical basic displacement is corrected by load to obtain the load-corrected displacement. Based on the current image cleanliness coefficient, the displacement compensation amount is determined, and the displacement compensation amount is superimposed with the load-corrected displacement amount to obtain the final predicted displacement. The final predicted displacement is superimposed with the fusion position of the scraper chain obtained in the previous sampling period to calculate the predicted state position of the scraper chain.

[0143] Furthermore, the decision configuration module 03 also includes: Based on the current image cleanliness coefficient, a first displacement compensation component is calculated, wherein the first displacement compensation component is negatively correlated with the current image cleanliness coefficient; a second displacement compensation component is calculated based on the running speed, and a third displacement compensation component is calculated based on the load current; based on the current image cleanliness coefficient, the second displacement compensation component and the third displacement compensation component are weighted and fused to obtain a fourth displacement compensation component; the first displacement compensation component and the fourth displacement compensation component are superimposed to obtain the total displacement compensation amount; based on the difference between the total displacement compensation amount and the displacement compensation amount calculated in the previous sampling period, a first-order low-pass filter is applied for smoothing to output the final displacement compensation amount.

[0144] Furthermore, the decision configuration module 03 also includes: The system interrupts the deviation correction decision based on the original time-series image stream and switches to deviation status judgment based on the monitoring signals of an independent monitoring device; it continuously acquires the monitoring signals output by the independent monitoring device, which includes a laser beam sensor and a mechanical deviation switch; based on the monitoring signals, it determines whether the scraper chain has triggered a deviation alarm condition; if the deviation alarm condition is triggered, it generates and executes a preset safety correction command, which includes periodic correction actions with a lower amplitude than normal; and it simultaneously sends a mode degradation alarm to the monitoring center, which includes the current image cleanliness coefficient and the type of independent monitoring device that triggered the error.

[0145] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A visual recognition-based adaptive correction method for coal mine scraper conveyors, characterized in that, The method includes: The original time-series image stream of the scraper chain and trough area of ​​the target scraper conveyor is continuously acquired through an explosion-proof visual acquisition device. An image cleanliness assessment model is constructed, and the original time-series image stream is analyzed frame by frame to calculate and output the corresponding image cleanliness coefficient. Based on the image cleanliness coefficient and combined with the real-time operating parameters of the target scraper conveyor, a hierarchical decision-making mechanism is dynamically configured, wherein the hierarchical decision-making mechanism includes a visual correction mode, a prediction compensation mode, and a safety fault tolerance mode. Based on the corrected decision threshold and the image cleanliness coefficient, the corresponding decision mode is activated, a correction control quantity adapted to the current working condition is generated, and adaptive correction is driven to be executed.

2. The adaptive correction method for coal mine scraper conveyors based on visual recognition according to claim 1, characterized in that, The steps for constructing the image cleanliness assessment model include: Based on the operating environment of the scraper conveyor in the coal mine, a set of sample time-series images of the scraper chain and trough side area under different dust interference conditions was collected, and the cleanliness of each frame in the sample time-series image set was labeled to obtain the corresponding sample image cleanliness set. A network architecture for an image cleanliness assessment model is constructed based on deep learning. The image cleanliness assessment model is trained under supervision using the sample time-series image set and the sample image cleanliness set until the model validation accuracy converges, thus completing the model construction.

3. The adaptive correction method for coal mine scraper conveyors based on visual recognition according to claim 1, characterized in that, Based on the image cleanliness coefficient and combined with the real-time operating parameters of the target scraper conveyor, a hierarchical decision-making mechanism is dynamically configured, including: A basic first decision threshold and a basic second decision threshold are preset to correspond to the hierarchical decision-making mechanism, wherein the basic first decision threshold is greater than the basic second decision threshold; The real-time operating parameters of the target scraper conveyor are obtained, and the current image cleanliness coefficient is obtained. The real-time operating parameters include the operating speed and load current. Based on the real-time operating parameters of the target scraper conveyor, the basic first decision threshold and the basic second decision threshold are corrected. The image cleanliness coefficient is compared with the corrected decision threshold: If the image cleanliness coefficient is greater than or equal to the corrected first decision threshold, then the visual correction mode is activated to perform scraper chain position feature recognition and calculation on the image frames in the original time-series image stream. If the image cleanliness coefficient is less than the corrected first decision threshold and greater than or equal to the corrected second decision threshold, then the prediction compensation mode is activated to perform visual enhancement processing, state prediction and data fusion on the image frames in the original time-series image stream. If the image cleanliness coefficient is less than the corrected second decision threshold, the safety fault-tolerant mode is activated to perform correction monitoring on the scraper chain based on non-visual sensors.

4. The adaptive correction method for coal mine scraper conveyors based on visual recognition according to claim 3, characterized in that, Based on the real-time operating parameters of the target scraper conveyor, the basic first decision threshold and the basic second decision threshold are corrected, including: Based on the operating speed of the target scraper conveyor, a speed correction factor is calculated, wherein the speed correction factor is positively correlated with the operating speed; Based on the load current of the target scraper conveyor, a load correction factor is calculated, wherein the load correction factor is positively correlated with the load current; Calculate the comprehensive operating condition influence coefficient based on the speed correction factor and the load correction factor; The comprehensive operating condition influence coefficient is multiplied by the benchmark first decision threshold and the benchmark second decision threshold respectively to obtain the corrected first decision threshold and the corrected second decision threshold.

5. The adaptive correction method for coal mine scraper conveyors based on visual recognition according to claim 4, characterized in that, Based on the speed correction factor and the load correction factor, calculate the comprehensive operating condition influence coefficient, including: Based on the current image cleanliness coefficient, the weight enhancement coefficient of the velocity correction factor is calculated, wherein the weight enhancement coefficient is negatively correlated with the current image cleanliness coefficient; Multiply the weight enhancement coefficient by the load correction factor to obtain the adjusted load correction factor; Calculate the weighted average of the speed correction factor and the adjusted load correction factor as the initial comprehensive operating condition influence coefficient; The initial comprehensive operating condition influence coefficient is normalized so that its value falls within a preset coefficient range, and the normalized value is output as the final comprehensive operating condition influence coefficient.

6. The adaptive correction method for coal mine scraper conveyors based on visual recognition according to claim 3, characterized in that, Activate the prediction compensation mode to perform visual enhancement processing, state prediction, and data fusion on the image frames in the original time-series image stream, including: The image frame is subjected to visual enhancement processing based on the image cleanliness coefficient to extract the visual observation position of the scraper chain; Based on the real-time operating parameters of the target scraper conveyor, the state and position of the scraper chain are predicted. Based on the image cleanliness coefficient, the visual fusion weight is calculated within the interval formed by the modified first decision threshold and the modified second decision threshold; Based on the visual fusion weights, the visual observation position and the state prediction position are weighted and fused to obtain the fused position.

7. The adaptive correction method for coal mine scraper conveyors based on visual recognition according to claim 6, characterized in that, Based on the real-time operating parameters of the target scraper conveyor, the state and position of the scraper chain are predicted, including: Based on the operating speed of the target scraper conveyor, calculate the theoretical basic displacement of the scraper chain within the current sampling period; Based on the load current of the target scraper conveyor and combined with the load displacement influence coefficient, the theoretical basic displacement is corrected by load to obtain the load-corrected displacement. Based on the current image cleanliness coefficient, the displacement compensation amount is determined, and the displacement compensation amount is superimposed with the load correction displacement amount to obtain the final predicted displacement amount; The final predicted displacement is superimposed with the merged position of the scraper chain obtained in the previous sampling period to calculate the predicted state position of the scraper chain.

8. The adaptive correction method for coal mine scraper conveyors based on visual recognition according to claim 7, characterized in that, Based on the current image cleanliness coefficient, determine the displacement compensation amount, including: Based on the current image cleanliness coefficient, a first displacement compensation component is calculated, wherein the first displacement compensation component is negatively correlated with the current image cleanliness coefficient; The second displacement compensation component is calculated based on the operating speed, and the third displacement compensation component is calculated based on the load current. Based on the current image cleanliness coefficient, the second displacement compensation component and the third displacement compensation component are weighted and fused to obtain the fourth displacement compensation component. The first displacement compensation component is superimposed with the fourth displacement compensation component to obtain the total displacement compensation amount. Based on the difference between the total displacement compensation and the displacement compensation calculated in the previous sampling period, a first-order low-pass filter is applied for smoothing to output the final displacement compensation.

9. The adaptive correction method for coal mine scraper conveyors based on visual recognition according to claim 3, characterized in that, Activate the safety fault-tolerant mode and perform correction monitoring of the scraper chain based on non-visual sensors, including: The deviation correction decision based on the original time-series image stream is interrupted, and the deviation status is judged based on the monitoring signal of the independent monitoring device. The monitoring signal output by the independent monitoring device is continuously acquired, wherein the independent monitoring device includes a laser beam sensor and a mechanical deviation switch; Based on the monitoring signal, determine whether the scraper chain has triggered the deviation alarm condition; If the deviation alarm condition is triggered, a preset safety correction instruction is generated and executed, wherein the preset safety correction instruction includes periodic correction actions with a lower amplitude than normal. A mode degradation alarm is sent synchronously to the monitoring center. The mode degradation alarm includes the current image cleanliness coefficient and the type of independent monitoring device that triggered it.

10. A visual recognition-based adaptive correction system for coal mine scraper conveyors, characterized in that, The system is used to execute the adaptive correction method for coal mine scraper conveyors based on vision recognition as described in any one of claims 1-9, and the system comprises: The image acquisition module is used to continuously acquire the original time-series image stream of the scraper chain and trough side area of ​​the target scraper conveyor through an explosion-proof vision acquisition device; The quality assessment module is used to construct an image cleanliness assessment model, perform frame-by-frame analysis on the original time-series image stream, and calculate and output the corresponding image cleanliness coefficient. The decision configuration module is used to dynamically configure a hierarchical decision mechanism based on the image cleanliness coefficient and the real-time operating parameters of the target scraper conveyor. The hierarchical decision mechanism includes a visual correction mode, a prediction compensation mode, and a safety fault tolerance mode. The correction execution module is used to activate the corresponding decision mode based on the corrected decision threshold and the image cleanliness coefficient, generate a correction control quantity adapted to the current working condition, and drive the execution of adaptive correction.

Images