A Vision-Based Adaptive Control Method and System for Scraper Chain Deflection

Abstract

This invention discloses a visual detection-based adaptive control method and system for scraper conveyor chain deflection, belonging to the field of intelligent control technology for mining machinery. The method includes: deploying a visual detection device on the scraper conveyor trough to acquire images of the scraper chain's operation and, through sequence deflection recognition, obtaining the lateral deflection angle and the rate of change of the deflection angle; constructing a scraper chain deflection state discriminator based on the rotational structure characteristics of the deflection chain loop, performing anomaly detection, and outputting abnormal deflection state parameters; calling the deflection strategy control channel to parse the target deflection control parameters, driving the deflection control module to execute scraper chain deflection control and monitor and provide feedback, thus achieving adaptive closed-loop control. This invention solves the technical problems of traditional detection methods being unable to accurately capture deflection-related states in real time, accurately identify deflection anomalies, and dynamically match and adapt control strategies, achieving adaptive control of the scraper conveyor chain deflection, and improving the technical effect of improving equipment operation stability and control accuracy.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for mining machinery, and in particular to an adaptive control method and system for scraper chain deflection based on vision detection. Background Technology

[0002] As a core transportation device in mines, the accurate detection of scraper chain deflection is crucial for stable equipment operation and safe production. Current technologies primarily rely on contact sensors and manual inspections to determine the scraper chain's operating status. While these methods have played a significant role in routine, static equipment maintenance, their limitations become apparent when applied to the dynamic operation of scraper conveyors due to the increasing demands for intelligent mine management. The complex operating conditions and dynamic movement of the chain links in scraper conveyors prevent traditional detection methods from accurately acquiring scraper chain deflection parameters in real time. This results in data lacking timeliness and accuracy, failing to meet the requirements for precise identification and adaptive control of scraper chain deflection. Summary of the Invention

[0003] This application provides a visual detection-based adaptive control method and system for scraper chain deflection, which solves the technical problems that traditional detection methods cannot accurately capture deflection-related states in real time, accurately identify deflection anomalies, and are difficult to dynamically match and adapt control strategies.

[0004] The first aspect of this application provides a visual detection-based adaptive control method for scraper chain deflection. The method includes: deploying a visual detection device on the conveyor trough of a scraper conveyor; acquiring a set of scraper chain running images in real time using the visual detection device; performing sequence deflection recognition on the scraper chain running image set to obtain the lateral deflection angle and deflection angle change rate of the deflection chain links; constructing a scraper chain deflection state discriminator based on the rotational structural characteristics of the deflection chain links; performing anomaly detection on the lateral deflection angle and deflection angle change rate of the deflection chain links based on the scraper chain deflection state discriminator; outputting abnormal deflection state parameters of the scraper chain; calling a deflection strategy control channel through a deflection control module; using the deflection strategy control channel to analyze the control parameters of the abnormal deflection state parameters of the scraper chain to determine target deflection control parameters; driving the deflection control module to execute scraper chain deflection control and monitoring based on the target deflection control parameters to obtain scraper chain deflection feedback parameters; and performing adaptive closed-loop control of deflection based on the scraper chain deflection feedback parameters.

[0005] A second aspect of this application provides a vision-based adaptive control system for scraper chain deflection. The system includes: a deflection chain link parameter acquisition module, used to deploy a vision detection device on the conveyor trough of the scraper conveyor, acquire a set of scraper chain running images in real time through the vision detection device, perform sequential deflection recognition on the scraper chain running image set, and obtain the lateral deflection angle and the rate of change of the deflection angle of the deflection chain link; and a scraper chain abnormal parameter acquisition module, used to construct a scraper chain deflection state discriminator based on the rotational structural characteristics of the deflection chain link, and perform deflection state discriminator on the deflection chain link. The system performs anomaly detection based on the lateral deflection angle and the rate of change of the deflection angle, and outputs abnormal deflection state parameters of the scraper chain. A target deflection control parameter acquisition module is used to call the deflection strategy control channel through the deflection control module, and uses the deflection strategy control channel to analyze the abnormal deflection state parameters of the scraper chain to determine the target deflection control parameters. A closed-loop control execution module drives the deflection control module to execute scraper chain deflection control and monitor it based on the target deflection control parameters, obtaining scraper chain deflection feedback parameters, and performing adaptive closed-loop deflection control through these parameters.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] This application acquires a set of images of the scraper chain running by deploying a visual inspection device in the transport trough of the scraper conveyor. After processing such as noise reduction, feature extraction, and angle and rate of change calculation, deflection-related data is obtained. Combined with the structural characteristics of the deflection chain link and the pre-built discriminator and strategy control channel, control parameters are analyzed and closed-loop regulation is performed to achieve adaptive and precise control of the scraper chain deflection. This makes the scraper conveyor operation more stable and the deflection control more efficient and reliable, achieving the technical effect of adaptive regulation of the scraper conveyor chain deflection, improving the stability of equipment operation and the accuracy of control. Attached Figure Description

[0008] 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.

[0009] Figure 1 This is a flowchart illustrating the visual detection-based adaptive control method for scraper chain deflection provided in an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of the structure of the visual detection-based adaptive control system for scraper chain deflection provided in the embodiments of this application.

[0011] Figure labeling: Deflection chain link parameter acquisition module 1, scraper chain abnormal parameter acquisition module 2, target deflection control parameter acquisition module 3, closed-loop control execution module 4. Detailed Implementation

[0012] This application provides a visual detection-based adaptive control method and system for scraper chain deflection, which solves the technical problems that traditional detection methods cannot accurately capture deflection-related states in real time, accurately identify deflection anomalies, and are difficult to dynamically match and adapt control strategies.

[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 a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] Example 1, as Figure 1 As shown, a visual detection-based adaptive control method for scraper chain deflection is described, wherein the method includes:

[0016] A visual inspection device is installed on the conveyor trough of the scraper conveyor. The visual inspection device collects a set of images of the scraper chain running in real time. The sequence deflection of the scraper chain running image set is identified to obtain the lateral deflection angle and the rate of change of the deflection angle of the deflection chain link.

[0017] In this embodiment, the scraper conveyor is a continuous transport equipment used to transport bulk materials in mining and other scenarios, and mainly includes core components such as scraper chain, transport trough, drive device, tensioning device and frame.

[0018] Specifically, firstly, industrial CMOS vision inspection devices are installed at key monitoring positions in the middle and both ends of the scraper conveyor trough. These devices are fixed to frames on both sides of the trough using adjustable brackets, with the lenses facing the scraper chain's running plane to ensure the imaging range completely covers the scraper chain's trajectory. Based on the scraper chain's running speed, the frame rate of the vision inspection devices is set to 30-60 frames per second, and the resolution is adjusted to 1920×1080 pixels. Real-time communication is established with the back-end processing module via a data transmission line. After the device is started, it continuously acquires continuous images of the scraper chain during its operation, forming a scraper chain running image set.

[0019] Next, after Gaussian filtering and contrast enhancement preprocessing of the scraper chain running image set, the ROI (Region of Interest) is extracted through the constructed attention mechanism module to obtain the scraper chain image region set. The chain link features are extracted according to the time series to obtain the chain link running sequence feature set. Then, the deflection angle is identified to finally obtain the lateral deflection angle and the rate of change of deflection angle of the deflected chain link. This step will be explained in detail in the following content.

[0020] A scraper chain deflection state discriminator is constructed by combining the rotational structure characteristics of the deflection chain links. Based on the scraper chain deflection state discriminator, anomalies are detected in the lateral deflection angle and the rate of change of the deflection angle of the deflection chain links, and the abnormal deflection state parameters of the scraper chain are output.

[0021] Optionally, firstly, a working condition parameter table is constructed based on the working application scenario of the scraper chain. The scraper conveyor is then tested for deflection according to the table, and the deflection state dataset is recorded. The dataset is then labeled with deflection state discrimination based on the characteristics of the deflection chain loop rotation structure to obtain a multi-condition scraper chain deflection state sample set. Based on this sample set, deflection state recognition training and merging are performed to construct a scraper chain deflection state discriminator. This step will be explained in detail later.

[0022] Next, the real-time acquired lateral deflection angle and rate of change of the deflection chain links are preprocessed using standardization. The Z-score standardization method is used to process the lateral deflection angle and rate of change of the deflection angle, calculating the mean and standard deviation of the two parameters under corresponding operating conditions. The parameters are then standardized according to the formula to ensure that the dimensions of the input parameters are consistent with the dimensions of the sample parameters used during the training of the scraper chain deflection state discriminator. Real-time operating parameters of the scraper conveyor are simultaneously collected, including conveying capacity, operating speed, material characteristics, ambient dust concentration, and scraper chain tension. Based on preset operating condition classification standards, real-time operating condition categories are determined, and one-hot encoding is performed on these categories to obtain operating condition feature vectors that match the discriminator's input requirements.

[0023] Then, the preprocessed standardized values ​​of the lateral deflection angle, the standardized value of the deflection angle change rate, and the working condition feature vector are input into the scraper chain deflection state discriminator. The parameters first enter the discriminator's feature input layer to receive and integrate the feature parameters. The integrated feature parameters are then passed to the decision tree ensemble layer. Through parallel computation of 100 CART decision trees, each decision tree outputs an independent deflection state prediction result based on its own trained discrimination rules. The output layer performs a majority voting operation on the prediction results of all decision trees, counts the number of votes for each state, and determines the state with the most votes as the final deflection state, outputting the corresponding state label. The state labels include four categories: normal, slightly abnormal, severely abnormal, and abrupt deflection.

[0024] Finally, anomaly filtering and parameter integration are performed based on the status labels output by the discriminator. When the output status label is normal, the current scraper chain operation is determined to be normal, and no abnormal parameters are output. When the output status label is slightly abnormal, severely abnormal, or rapidly deflecting, the current scraper chain operation is determined to be abnormal, and the real-time lateral deflection angle value, deflection angle change rate value, working condition category information, and corresponding status label are extracted. The above information is integrated according to a fixed format to form the scraper chain abnormal deflection status parameters. These parameters can be directly transmitted to the subsequent deflection control module, providing data support for the scraper chain deflection adaptive control.

[0025] The deflection control module calls the deflection strategy control channel, and uses the deflection strategy control channel to analyze the control parameters of the abnormal deflection state of the scraper chain to determine the target deflection control parameters.

[0026] In this embodiment, the deflection control module is a hardware and software combination that integrates a controller, an actuator drive interface, and a data interaction unit, and is used to drive actuators such as hydraulic tensioning devices and electric guide mechanisms according to target deflection control parameters.

[0027] Specifically, when the deflection control module receives abnormal deflection status parameters of the scraper chain, it initiates the call process of the deflection strategy control channel. First, the abnormal deflection status parameters of the scraper chain are standardized and preprocessed. The operating condition category is converted into a 5-dimensional binary feature vector using one-hot encoding, and the lateral deflection angle and deflection angle change rate are standardized using Z-score. The mean and standard deviation of the two parameters under the corresponding operating condition are calculated, and the numerical conversion is completed according to the formula, ensuring that the parameter format is completely consistent with the storage format of the deflection strategy control channel.

[0028] The deflection control module calls the deflection strategy control channel pre-stored in the module's storage unit via a standard API interface. The pre-processed abnormal deflection status parameters of the scraper chain are input to the channel's input layer. The input layer performs parameter verification and integration, including verifying whether the parameter dimensions match and whether the values ​​are within the valid range. After successful verification, the integrated parameters are transmitted to the channel's matching layer.

[0029] The matching layer employs key-value pair query logic, using the input abnormal state feature vector as the key to retrieve the pre-built mapping table of abnormal state parameters and control strategy parameters within the channel. When multiple abnormal types are superimposed on the input parameters, the control strategy parameter corresponding to the highest priority is selected according to the priority rule of sharp deflection taking precedence over severe abnormalities, and severe abnormalities taking precedence over mild abnormalities, ensuring the relevance and effectiveness of the control commands.

[0030] After the matching layer outputs the corresponding control strategy parameter thresholds, the output layer of the deflection strategy control channel performs fine-tuning of these thresholds. Combining this with the real-time operating parameters of the scraper conveyor, such as the current tension value, the actual position of the guide wheel, and the real-time speed of the conveyor, the parameter thresholds are fine-tuned by 5%. After fine-tuning, the final target deflection control parameters are determined. These parameters specifically include the target tension value, the guide wheel adjustment step size, and the target running speed value. The accuracy standards for each parameter are clearly defined: the target tension value accuracy is ±0.1 kN, the guide wheel adjustment step size accuracy is ±0.5 mm, and the target running speed accuracy is ±0.01 m / s.

[0031] By standardizing the preprocessing of abnormal deflection state parameters, accurately calling the deflection strategy control channel, and performing multi-priority matching and parsing, target deflection control parameters that fit the actual working conditions are obtained, providing a precise execution basis for the adaptive control of scraper chain deflection.

[0032] Based on the target deflection control parameters, the deflection control module is driven to perform scraper chain deflection control and monitoring, thereby obtaining scraper chain deflection feedback parameters. Deflection adaptive closed-loop regulation is then performed using these scraper chain deflection feedback parameters.

[0033] Specifically, after receiving the target deflection control parameters, the deflection control module first analyzes and allocates these parameters. The target tension force value is allocated to the hydraulic tensioning device control unit, the guide wheel adjustment step size is allocated to the electric guiding mechanism control unit, and the target running speed value is allocated to the frequency converter control unit, ensuring that each execution unit receives the corresponding precise control command. Each execution unit synchronously initiates control operations. The hydraulic tensioning device adjusts the hydraulic system pressure according to the target tension force value, driving the tensioning cylinder to extend and retract to change the tension of the scraper chain. The electric guiding mechanism drives the guide wheel to move laterally according to the adjustment step size, correcting the contact position between the conveyor trough and the scraper chain. The frequency converter adjusts the output frequency, changing the speed of the conveyor drive motor to achieve precise control of the running speed.

[0034] Next, a real-time monitoring process is initiated simultaneously with the control operations. The visual inspection device continuously acquires a set of images of the scraper chain in operation, and calculates the real-time lateral deflection angle and the rate of change of the deflection angle according to the aforementioned sequence deflection recognition method. The hydraulic sensor, position sensor, and speed sensor respectively collect the real-time tension of the scraper chain, the actual position of the guide wheel, and the real-time operating speed of the conveyor. The above visual inspection data and sensor data are integrated to form scraper chain deflection feedback parameters.

[0035] Finally, the deviation between the scraper chain deflection feedback parameters and the target deflection control parameters is calculated. The lateral deflection angle deviation threshold is set to 1°, and the deflection angle change rate deviation threshold is set to 0.5° per second. If the deviations of the feedback parameters and the target parameters are both within the threshold range, the deflection control is considered to have achieved the expected effect, and the control process is stopped. If the deviation exceeds the threshold range, the deflection feedback parameters are re-input into the deflection strategy control channel, new target deflection control parameters are generated, and the control, monitoring, and deviation calculation process is repeated until the deviation meets the threshold requirements, ultimately completing the adaptive closed-loop control of the scraper chain deflection.

[0036] By controlling command parsing and allocation, coordinating operation of execution elements, real-time data monitoring and feedback, and deviation threshold comparison and adjustment, adaptive closed-loop control of scraper chain deflection is achieved, ensuring stable operation of the scraper conveyor.

[0037] Furthermore, the method provided in this application embodiment includes:

[0038] The scraper chain running image set is subjected to Gaussian filtering for noise reduction and contrast enhancement to obtain a usable scraper chain running image set. An attention mechanism module is constructed and used to extract and locate the ROI of the usable scraper chain running image set to obtain a scraper chain image region set. The scraper chain image region set is then sequentially processed according to time series to extract chain link features, resulting in a chain link running sequence feature set. The chain link running sequence feature set is then used to identify deflection angles to obtain the lateral deflection angles and deflection angle change rates of the deflected chain links.

[0039] Specifically, the scraper chain operation image set acquired in the aforementioned steps is first preprocessed. Gaussian filtering with a 3×3 or 5×5 Gaussian kernel is used for noise reduction. The standard deviation of the Gaussian kernel is selected from 0.5 to 1.0 based on the image noise intensity. Random noise caused by dust and light fluctuations in the industrial environment is filtered out by calculating a weighted average of each pixel and its neighboring pixels. Subsequently, a global histogram equalization method is used to enhance image contrast. First, the gray-level histogram of the image is statistically analyzed, and the cumulative distribution probability of each gray level is calculated. Then, a mapping function is used to convert the original gray-level values ​​into uniformly distributed gray-level values, increasing the gray-level difference between the scraper chain and the background environment, resulting in a usable scraper chain operation image set with clear outlines and suppressed noise.

[0040] Next, a spatial attention mechanism module is constructed for ROI extraction and localization. The specific construction process is as follows: The module input is a single-frame RGB image of an image set that can be processed by a scraper chain, and the output is an ROI image containing the scraper chain. This module consists of a feature extraction layer, an attention weight calculation layer, and a feature weighting layer, and requires no additional training. The feature extraction layer uses a 3×3 convolution kernel with two strides of 1 and the same padding method to perform convolution operations on the input image to extract shallow texture and edge features, resulting in a multi-channel feature map. The attention weight calculation layer first performs global average pooling on the multi-channel feature map to obtain a 1×1×C feature vector, where C is the number of channels in the feature map. The input is a fully connected layer with 64 neurons and the activation function is ReLU. Feature mapping is performed, and then the image is transformed into a 1×H×W two-dimensional attention weight map through a reshape operation, where H and W are the height and width of the input image, normalized to between 0 and 1 by the Sigmoid function. The feature weighting layer multiplies the original input image with the two-dimensional attention weight map pixel by pixel, highlighting the high-weight region where the scraper chain is located and suppressing the background. Then, the Otsu method is used for adaptive threshold segmentation to obtain a binary image. Morphological closing operation is performed using 5×5 rectangular structuring elements to eliminate noise regions smaller than 50 pixels. After hole filling is performed on the foreground region in the binary image, the boundary coordinates of the minimum bounding rectangle are extracted by contour detection. The original image is then cropped based on these coordinates to obtain the scraper chain image region set.

[0041] Then, the chain features of the scraper chain image region set are extracted sequentially according to the time series. First, each scraper chain image region is converted into an 8-bit grayscale image. The grayscale image is divided into 16×16 pixel sub-blocks using the sliding window method. The grayscale mean, grayscale variance, and grayscale entropy of each sub-block are calculated. Then, the Sobel operator is used for edge feature extraction. 3×3 Sobel convolution kernels in the horizontal and vertical directions are convolved with the grayscale image to obtain horizontal and vertical edge images. The edge intensity and edge direction of each pixel in the two edge images are calculated. The maximum, minimum, and average direction of the edge intensity within each sub-block are counted. The grayscale statistical features of each sub-block are concatenated with the edge features to form a feature vector of length 6. The combination of the feature vectors of all sub-blocks is the chain feature corresponding to the current frame image. All image regions are processed sequentially according to the time series to obtain the chain operation sequence feature set.

[0042] Finally, the Canny operator is used to perform edge detection on the feature set of the chain loop running sequence to extract the sequence edge lines on both sides of the chain loop. The lateral deflection angle is obtained by Hough transform and deflection angle calculation. The angle is then filtered by moving average according to the timestamp and fitted to generate a deflection angle-time curve. Based on this curve, the rate of change of the deflection angle is calculated by finite difference. This step will be explained in detail in the following content.

[0043] By employing image preprocessing methods, spatial attention mechanism modules, and statistical and edge feature extraction methods, image acquisition, noise reduction and enhancement, target region localization, and feature extraction were gradually completed, accurately obtaining the running sequence features of the scraper chain, providing reliable data support for the subsequent identification of deflection angle and change rate.

[0044] Furthermore, the method provided in this application embodiment includes:

[0045] The Canny operator is used to perform edge detection and extraction on the feature set of the chain loop running sequence to obtain the sequence edge lines on both sides of the chain loop; Hough transform and deflection angle calculation are performed on the sequence edge lines on both sides of the chain loop to obtain the lateral deflection angle of the deflected chain loop; the lateral deflection angle is filtered and fitted according to the timestamp to generate a deflection angle-time curve; the finite difference rate of change is calculated based on the deflection angle-time curve to obtain the rate of change of the deflection angle.

[0046] Optionally, firstly, edge detection extraction based on the Canny operator is performed on the feature set of the chain loop sequence. First, a 5×5 Gaussian kernel is used to smooth the feature set image, with the standard deviation of the Gaussian kernel set to 1.0 to reduce the interference of image noise on edge detection. Then, the Sobel operator is used to calculate the gradient magnitude and gradient direction in the horizontal and vertical directions of the image, obtaining a complete gradient image. Next, non-maximum suppression is implemented. Each pixel in the gradient image is traversed, and the gradient magnitude of that pixel is compared with the magnitudes of the two adjacent pixels along the gradient direction. Only local maxima are retained, and non-edge points are discarded. Finally, a dual threshold is set: a high threshold of 150 and a low threshold of 50. Points with gradient magnitudes higher than the high threshold are marked as strong edges, points between the high and low thresholds and connected to strong edges are marked as weak edges, and points that do not meet the above conditions are judged as background. Finally, strong and weak edges are integrated to obtain the sequence edge lines on both sides of the chain loop.

[0047] Next, a threshold for the length of the straight line is preset. Hough transform is used to detect and filter the straight lines of the sequence edge lines on both sides of the chain to obtain the main edge lines of the chain sequence. The equations of the straight lines of the edge sequences on both sides are obtained by least squares straight line fitting. The angle between each edge line and the horizontal baseline is calculated and the average value is taken as the lateral deflection angle of the deflection chain of the current sequence frame. This step will be explained in detail in the following content.

[0048] Then, the lateral deflection angles of each sequence frame are associated with their corresponding timestamps and arranged chronologically to form the raw angle-time data. A moving average filter with a sliding window size of 5 frames is used. Centered on the current frame, the average of the lateral deflection angles of the two frames before and after it (a total of 5 frames) is calculated. This average is used as the filtered lateral deflection angle of the current frame. This process is repeated for all sequence frames to obtain the filtered angle-time data. Subsequently, a first-order linear polynomial fitting method is used to fit the filtered angle-time data using the least squares method, establishing a linear relationship model between angle and time, and generating a continuous and smooth deflection angle-time curve.

[0049] Finally, based on the deflection angle-time curve, the forward finite difference method is used to calculate the rate of change of the deflection angle. The time interval is determined by the acquisition frame rate of the visual inspection device. If the acquisition frame rate is 30 frames / second, then the time interval is 1 / 30 second. For any angle value θ(t) corresponding to any time t on the curve, the rate of change of the deflection angle v(t) is calculated as v(t)=[θ(t)-θ(t-1)] / [t-(t-1)], where θ(t-1) is the angle value corresponding to the previous time t. The rate of change is calculated for each time moment in turn, and finally the rate of change of the deflection angle is obtained.

[0050] By employing edge detection, line fitting, filtering fitting, and finite difference methods, the lateral deflection angle and the rate of change of the deflection angle of the deflection chain link were accurately obtained, providing reliable quantitative data support for subsequent abnormal judgment of the deflection state of the scraper chain.

[0051] Furthermore, the method provided in this application embodiment includes:

[0052] A preset straight line length threshold is used to perform Hough transform line detection and filtering on the edge lines of the chain loop on both sides according to the preset straight line length threshold to obtain the main edge line of the chain loop sequence. The main edge line of the chain loop sequence is fitted with a least squares method to obtain the straight line equations of the edge sequences on both sides. Based on the straight line equations of the edge sequences on both sides, the first single-side sequence angle and the second single-side sequence angle between the single-side edge line and the horizontal baseline are calculated respectively. The average value of the first single-side sequence angle and the second single-side sequence angle is obtained, and the average value is used as the lateral deflection angle of the deflection chain loop in the current sequence frame.

[0053] Specifically, a line length threshold of 50 pixels is preset. Based on this threshold, Hough transform is used to detect and filter lines on both sides of the chain loop sequence edge lines. The parameters of the Hough transform are set as follows: the step size of the polar angle theta is 1 degree, the step size of the polar radius rho is 1 pixel, and the accumulator threshold is set to 100. The edge points in the image space are mapped to the parameter space through the Hough transform. The number of edge points corresponding to each parameter combination is counted. When the accumulator value reaches the threshold, it is determined that a line has been detected. Then, the pixel length of each detected line is calculated, and lines with a length greater than or equal to 50 pixels are filtered out. Short-distance stray lines are removed to obtain the main edge lines of the chain loop sequence.

[0054] Next, least squares line fitting is performed on the main edge lines of the chain loop sequence. First, each main edge line is traversed, and the coordinate data of all edge pixels are collected. Where i is the pixel index, and each main edge line must contain at least 3 pixels to ensure fitting validity. Coordinate data is rounded to 4 decimal places. The 3σ criterion is used to remove outliers from the collected coordinate data, and all... and The mean and standard deviation of the coordinates are used to identify outliers and remove pixels that deviate from the mean by more than three times the standard deviation, retaining only valid coordinate data. Then, a fitting objective function is constructed, which minimizes the sum of the squared vertical distances from all valid pixels to the fitted line. Let the equation of the fitted line be y = kx + b, where k is the slope and b is the intercept. Key parameters are calculated using the following formula: First, calculate the mean of the coordinates. , Where n is the number of valid pixels, and then the covariance correlation statistic is calculated. and Then solve for the slope. ,intercept After fitting, the root mean square error of all valid pixels to the fitted line is calculated. If the root mean square error is ≤1.0 pixel, the line equation is retained. If it exceeds the threshold, the edge line acquisition quality is rechecked, pixels are added, and fitting is performed again. Finally, the line equations of the edge sequences on both sides of the main edge line on both sides of the chain are obtained.

[0055] Finally, an image coordinate system is constructed with the top left corner of the image as the origin, the x-axis pointing horizontally to the right as the horizontal baseline, and the y-axis pointing vertically downwards. The angle between the straight lines of each edge sequence and the horizontal baseline is calculated based on the equations of the lines on both sides. First, the coordinates of two non-overlapping pixels on each edge line are selected. and Calculate the slope The included angle in radians is obtained by using the arctangent function arctan(k), and then calculated using the formula: "Angle value = radian value × (180°)". ")" converts radian values ​​to angle values. The calculated result is then superimposed with 180° to correct the angle range to 0°~180°, which are recorded as the first single-sided sequence angle and the second single-sided sequence angle, respectively. During the calculation, the radian value is retained to 4 decimal places and the angle value is retained to 1 decimal place. A reasonable threshold for the difference between the two angles is set to 5°. If the difference between the two angles is ≤5°, the arithmetic mean of the two angles (first single-sided sequence angle + second single-sided sequence angle) / 2 is calculated as the lateral deflection angle of the current sequence frame deflection chain. If the difference exceeds 5°, edge detection and line fitting are performed again on the frame image, abnormal edge data is removed, the angle is recalculated, and the mean is taken to finally determine the lateral deflection angle of the current sequence frame.

[0056] Furthermore, the method provided in this application embodiment includes:

[0057] Based on the working application scenarios of the scraper conveyor, a scraper conveyor operating condition parameter table is constructed; the scraper conveyor is subjected to operation deflection tests according to the scraper conveyor operating condition parameter table, and the scraper conveyor operating deflection state dataset is recorded; the scraper conveyor operating deflection state dataset is identified and labeled according to the characteristics of the deflection chain loop rotation structure, resulting in a multi-condition scraper conveyor deflection state sample set; based on the multi-condition scraper conveyor deflection state sample set, deflection state recognition training and merging are performed to construct a scraper conveyor deflection state discriminator.

[0058] Specifically, firstly, based on the actual working application scenarios of the scraper chain, the core influencing factors are identified, and a scraper chain operating condition parameter table is constructed. Conveying capacity, operating speed, material characteristics, ambient dust concentration, and scraper chain tension are selected as key operating condition parameters. The conveying capacity ranges from 50-200 t / h, divided into 50 t / h intervals; the operating speed ranges from 0.8-1.6 m / s, divided into 0.2 m / s intervals; material characteristics are divided into two categories: bulk materials with a particle size ≤50 mm and mixed materials with a particle size of 50-100 mm; ambient dust concentration is divided into low (≤50 mm) and medium (≤50 mm) dust. ), medium (50-150) ), high (>150) Level 3; the scraper chain tension ranges from 10-30kN, with gradients divided in 5kN intervals. These parameter gradients are combined to form a scraper chain operating condition parameter table containing 3×4×2×3×4=288 sets of working conditions, clearly defining the specific parameter values ​​for each set of working conditions.

[0059] According to the constructed scraper conveyor operation parameter table, the operating parameters of the scraper conveyor were set sequentially, and operation deflection tests were conducted and data sets were recorded. Before the test, multiple sets of visual inspection devices were deployed on both sides of the scraper conveyor trough according to the aforementioned steps. At the same time, tension sensors with a measurement range of 0-50kN and an accuracy of ±0.5kN were installed at both ends of the scraper chain to collect data in real time. For each operating condition, the scraper conveyor was first allowed to run stably for 5 minutes. Then, different degrees of deflection were artificially created by adjusting the position of the guide wheels on the side of the trough. The test was conducted continuously for 10 minutes. The lateral deflection angle, the rate of change of the deflection angle, and key operating parameters such as tension force, running speed, and conveying capacity corresponding to each frame of the image were recorded. 30 sets of valid samples were recorded for each operating condition, and finally, a scraper conveyor operation deflection status dataset containing 288×30=8640 data points was formed. The data format was uniformly as follows: conveying capacity, running speed, material characteristics, ambient dust concentration, tension force, lateral deflection angle, rate of change of deflection angle, and test timestamp.

[0060] Next, the deflection state dataset of the scraper chain is subjected to working condition clustering analysis and integration to obtain a multi-working condition deflection state dataset. Combining the characteristics of the deflection chain loop rotation structure, deflection state discrimination rules are mined, verified, evaluated and corrected to obtain multi-working condition deflection state discrimination rules. According to the rules, the multi-working condition deflection state dataset is labeled with deflection state discrimination to obtain a multi-working condition scraper chain deflection state sample set. This step will be explained in detail in the following content.

[0061] Subsequently, a deflection state discriminator was constructed by training and merging deflection state samples from a multi-condition scraper chain deflection state sample set. This discriminator employs a random forest model, consisting of a feature input layer, a decision tree ensemble layer, and an output layer. The feature input layer receives three feature parameters: the condition category (5 condition categories encoded as 5-dimensional binary vectors), the lateral deflection angle, and the rate of change of the deflection angle, all processed by one-hot encoding. The decision tree ensemble layer contains 100 CART decision trees, each with a maximum depth of 10, a minimum number of sample splits of 5, and a minimum number of leaf nodes of 3. Bootstrap sampling with replacement is used, retaining out-of-bag samples for evaluating model generalization ability. Samples are drawn from the training set to train individual decision trees. Feature selection uses random subset selection, randomly selecting two features from each tree. The output layer uses majority voting to integrate the prediction results of all decision trees and outputs the final deflection state label: 0 for normal, 1 for mild abnormality, 2 for severe abnormality, and 3 for abrupt deflection.

[0062] The training process is as follows: First, the feature parameters of the sample set are preprocessed. The working condition category is converted into numerical features using one-hot encoding. The lateral deflection angle and the rate of change of deflection angle are standardized using Z-score, and the formula is as follows: ,in The characteristic mean, The standard deviation of the features is used. The sample set is stratified into training, validation, and test sets in a 7:1:2 ratio to ensure that the proportion of samples in each of the four bias states is consistent. Next, the random forest model parameters are initialized, and the training epochs are set to 50. After each epoch, the model accuracy is evaluated using the validation set. If the accuracy improvement is ≤0.1% for 5 consecutive epochs, training is stopped. Finally, the model performance is tested using the test set, requiring a test set accuracy ≥90% and a macro-recall ≥90%, where macro-recall = arithmetic mean of the recall rates of each class, and single-class recall = the number of true positives. (Number of true positives + number of false negatives). If the target is not met, the number of decision trees is increased to 150 and retrained. If the target is still not met, the decision tree parameters are adjusted and retrained. For example, the maximum depth is set to 12 and the minimum number of sample splits is set to 8, until the performance requirements are met, and finally the construction of the scraper chain deflection state discriminator is completed.

[0063] Through the above-mentioned consecutive steps, a robust and accurate scraper chain deflection state discriminator was constructed, providing reliable technical support for the accurate identification of abnormal scraper chain deflection states.

[0064] Furthermore, the method provided in this application embodiment includes:

[0065] The scraper chain deflection state dataset is subjected to working condition clustering analysis and integration to obtain a multi-working condition deflection state dataset. Deflection state discrimination rules are mined, verified, evaluated, and corrected based on the characteristics of the deflection chain loop rotation structure to obtain multi-working condition deflection state discrimination rules. Deflection state discrimination labels are then applied to the multi-working condition deflection state dataset according to these rules to obtain a multi-working condition scraper chain deflection state sample set.

[0066] Specifically, firstly, the scraper chain deflection state dataset is subjected to working condition clustering analysis and integration to obtain a multi-working condition deflection state dataset. The K-means clustering algorithm is used, selecting conveying capacity, operating speed, material characteristics (quantized to 1 and 2), environmental dust concentration (quantized to 1, 2, and 3), and scraper chain tension (quantized to 1-5) as clustering feature variables. The elbow rule is used to determine the number of clusters K=5, the number of iterations is set to 100, the convergence threshold is 0.001, and the initial cluster centers are randomly selected. The dataset is then substituted into the algorithm for clustering, grouping working conditions with high similarity into one category, resulting in 5 typical working conditions. Outliers in each category that deviate from the cluster center by more than 2 standard deviations are removed. The remaining data are then organized according to working condition categories, forming 5 independent working condition deflection state datasets, i.e., the multi-working condition deflection state dataset.

[0067] Next, the deflection state discrimination rules under multiple working conditions are obtained. Specifically: First, the characteristic parameters of the deflection chain link rotation structure are defined, including the chain link rotation radius and hinge gap. Based on these characteristics, the safe deflection limit of the scraper chain under different working conditions is determined: when the rate of change of deflection angle is ≤1... At this time, a lateral deflection angle ≤3° is considered normal, 3°-5° is considered slightly abnormal, and >5° is considered severely abnormal; when the rate of change of the deflection angle is >1... The state is characterized by a sharp deflection. The C4.5 decision tree algorithm is used for rule mining, with the lateral deflection angle, the rate of change of the deflection angle, and the condition category after one-hot encoding as input features. The output labels are normal, slightly abnormal, severely abnormal, and sharply deflected. A decision tree model is constructed, with a minimum sample size of 10, a confidence threshold of 0.9, and a maximum decision tree depth of 8. Information gain ratio is used as the feature splitting criterion, and a pre-pruning strategy is employed to avoid overfitting.

[0068] Then, the multi-condition deflection state dataset is divided into layers in an 8:2 ratio to ensure that the proportion of each type of deflection state sample is consistent between the training set and the validation set. The model is trained using the training set to obtain the initial discrimination rules, and the overall prediction accuracy of a single condition is calculated using the validation set. If the rule accuracy of a certain condition is <90%, the deflection limit threshold under that condition is adjusted in steps of 0.5°. The adjustment direction is to expand the threshold range of normal state and shrink the threshold range of abnormal state when the accuracy is low. The model is then retrained, and the training iteration limit is set to 50 times. If the number of iterations reaches the limit and the target is still not met, the feature parameters are optimized and the model is trained again until the rule accuracy of all conditions is ≥90%. Finally, a multi-condition deflection state discrimination rule containing discrimination rules for 5 types of conditions is formed.

[0069] Finally, for each type of operating condition's deflection state data, the lateral deflection angle and deflection angle change rate of each data point are extracted and substituted into the corresponding operating condition's discrimination rules to determine its deflection state and label it: 0 for normal, 1 for slight abnormality, 2 for severe abnormality, and 3 for rapid deflection. After labeling, the labeled data for all operating conditions are integrated and divided into training, validation, and test sets in a 7:1:2 ratio. The training set is used for model training, the validation set for parameter adjustment, and the test set for final performance evaluation. The data format is unified as: operating condition category, lateral deflection angle, deflection angle change rate, and deflection state label, thus forming a multi-operating-condition scraper chain deflection state sample set.

[0070] Furthermore, the method provided in this application embodiment includes:

[0071] A set of abnormal deflection cases of the scraper chain is collected. Based on the scraper chain deflection state discriminator, the abnormal deflection cases are identified to obtain an abnormal deflection state parameter set. The abnormal deflection state parameter set is analyzed and optimized to obtain an abnormal deflection control strategy parameter set. The abnormal deflection state parameter set and the abnormal deflection control strategy parameter set are correlated and fitted to construct a deflection strategy control channel, and the deflection strategy control channel is stored in the deflection control module.

[0072] In one embodiment, a case set of abnormal deflection of the scraper conveyor is first collected. The case set includes historical abnormal deflection records from the scraper conveyor and simulated abnormal deflection test data under different operating conditions. Each abnormal state contains at least 50 valid samples. The data content covers the operating condition category, lateral deflection angle, deflection angle change rate, and corresponding equipment operating status information. The case set data is preprocessed: the operating condition category is converted into numerical features using one-hot encoding, and the lateral deflection angle and deflection angle change rate are standardized using Z-score to ensure the data format is consistent with the input format used during the training of the scraper conveyor deflection state discriminator. The preprocessed abnormal deflection case set is input into the constructed scraper conveyor deflection state discriminator. The discriminator receives the case data through its feature input layer, and through parallel computation in the decision tree ensemble layer and majority voting in the output layer, each case is judged as an abnormal state, labeled with the corresponding abnormal type label, and the operating condition parameters, deflection parameters, and abnormal type labels of all cases are integrated to obtain the abnormal deflection state parameter set.

[0073] Next, a scraper chain control strategy library is constructed. This strategy library is used to parse the abnormal deflection state parameter set to obtain the deflection control strategy parameter threshold set. Then, global iterative optimization is carried out within this parameter threshold set to obtain the abnormal deflection control strategy parameter set. This step will be explained in detail in the following content.

[0074] Subsequently, a rule-based mapping method was used to correlate and fit the abnormal deflection state parameter set with the abnormal deflection control strategy parameter set. A basic key-value pair mapping table was configured based on the correspondence between abnormal state parameters and optimal control parameters. The mapping relationship was iteratively optimized and updated by adding new abnormal deflection cases to establish a one-to-one correspondence between abnormal state parameters and optimal control parameters, forming a structured deflection strategy control channel. This channel includes an input layer, a matching layer, and an output layer. The input layer receives parameters in the format of working condition feature vector, standardized value of lateral deflection angle, and standardized value of deflection angle change rate. The matching layer calls the mapping relationship through key-value pair query logic and embeds priority rules when multiple abnormal types are superimposed, i.e., sharp deflection takes precedence over severe abnormality, and severe abnormality takes precedence over mild abnormality. The output layer outputs parameters in the format of tension target value, guide wheel adjustment step size, and running speed target value.

[0075] Finally, the constructed deflection strategy control channel is stored in the storage unit of the deflection control module as a relational database table. The database call interface is set to a standard API interface to ensure that the control module can call the channel in real time to complete the control parameter parsing during operation. For scenarios with multiple superimposed anomaly types, the control strategy priority is set as follows: sharp deflection takes precedence over severe anomalies, and severe anomalies take precedence over mild anomalies. The corresponding control parameters are adjusted sequentially according to priority. After the channel is constructed, the control effect is verified by actual machine testing on a scraper conveyor. If the deviation between the actual machine test results and the simulation results exceeds 10%, the weight coefficients of the multi-objective function and the optimization parameters are re-optimized until the control requirements are met.

[0076] Through the above-mentioned sequential steps, the system constructs a precise, reliable, and implementable deflection strategy control channel, providing core strategy support for the adaptive control of abnormal deflection of the scraper chain.

[0077] Furthermore, the method provided in this application embodiment includes:

[0078] A scraper chain control strategy library is constructed. The scraper chain control strategy library is used to parse the abnormal deflection state parameter set to obtain the deflection control strategy parameter threshold set. Global iterative optimization is performed within the deflection control strategy parameter threshold set to obtain the abnormal deflection control strategy parameter set.

[0079] Optionally, a scraper chain control strategy library is first constructed. This library contains control strategies for different anomaly types, specifically including three core strategies: scraper chain tension adjustment strategy based on hydraulic tensioning device, guide wheel position adjustment strategy based on electric guiding mechanism, and conveyor running speed control strategy based on frequency converter. Different anomaly types are matched with differentiated control parameter threshold ranges. For example, mild anomalies correspond to a tension adjustment range of 15 to 20 kN, a guide wheel adjustment step of 2 mm, and a running speed control range of 1.2 to 1.4 m / s; severe anomalies correspond to a tension adjustment range of 10 to 15 kN or 20 to 25 kN, a guide wheel adjustment step of 5 mm, and a running speed control range of 0.8 to 1.2 m / s; and sharp deflection corresponds to a tension adjustment range of 25 to 30 kN, a guide wheel adjustment step of 8 mm, and a running speed control range of 1.4 to 1.6 m / s. The abnormal deflection state parameter set is input into the scraper chain control strategy library. The corresponding control strategy is matched according to the abnormality type label in the parameter set, and the corresponding control parameter range is obtained by parsing. The control parameter ranges corresponding to all abnormal types are integrated to form the deflection control strategy parameter threshold set.

[0080] Finally, a multi-objective function for deflection control is constructed. This function is used to randomly select parameters within the deflection control strategy parameter threshold set and perform simulation evaluation to obtain the fitness of multiple deflection strategy parameters. Based on these fitness values, the parameter threshold set is globally iteratively optimized to obtain the abnormal deflection control strategy parameter set. This step will be explained in detail in the following sections.

[0081] Furthermore, the method provided in this application embodiment includes:

[0082] A deflection control multi-objective function is constructed. The deflection control multi-objective function is used to randomly select and simulate within the deflection control strategy parameter threshold set to obtain multiple deflection strategy parameter fitnesss. Based on the fitnesss of the multiple deflection strategy parameter thresholds, a global iterative optimization is performed on the deflection control strategy parameter threshold set to obtain an abnormal deflection control strategy parameter set.

[0083] In one embodiment, a multi-objective function for deflection control is first constructed. The optimization objectives of this function are set as three core indicators: shortest control response time, smallest deflection correction amplitude, and lowest equipment operating energy consumption. The weight coefficients of the three indicators are determined using orthogonal experimental design. The three optimization indicators are selected as experimental factors, and five levels are set for each to conduct orthogonal experiments. Range analysis determines the weight coefficients of the three indicators to be 0.4, 0.4, and 0.2, respectively. A dynamic simulation model of the scraper conveyor was built using MATLAB / Simulink. The core parameters of the model were defined as scraper chain stiffness, chain link hinge gap, and friction coefficient between the conveyor trough and the scraper chain. Fifty sets of control parameter combinations were randomly selected from the threshold set of deflection control strategy parameters. Each set of parameter combinations was substituted into the multi-objective function and simulation model for simulation evaluation. The fitness value of each set of parameters was calculated using the formula: Fitness value = 0.4 × (1 / normalized response time) + 0.4 × (1 / normalized correction amplitude) + 0.2 × (1 / normalized energy consumption). The higher the fitness value, the better the control effect of the parameter combination.

[0084] Next, a genetic algorithm is used for global iterative optimization. The population size is set to 50, the number of iterations to 100, the crossover probability to 0.8, and the mutation probability to 0.1. Single-point crossover and basic position mutation are used for crossover and mutation. In each iteration, the top 20% of parameter combinations with fitness values ​​are retained, and new parameter combinations are generated through crossover and mutation. This iterative process is repeated until the fitness value fluctuation is ≤0.5% for 10 consecutive iterations or the number of iterations reaches 100. The parameter combination with the highest fitness value is selected as the optimal control parameter. The optimal control parameters corresponding to all anomaly types are integrated to obtain the anomaly deflection control strategy parameter set. This parameter set is then tested on a real machine. The verification criteria are set as follows: the corrected scraper chain lateral deflection angle ≤3° and the control response time ≤5 seconds. If the deviation between the real machine test results and the simulation results exceeds 10%, the weight coefficients of the multi-objective function are readjusted and optimization is performed again until the verification criteria are met.

[0085] In summary, the visual detection-based adaptive control method for scraper chain deflection provided in this application has the following technical effects:

[0086] This application utilizes a visual inspection device installed in the transport trough of a scraper conveyor to collect a set of images of the scraper chain in operation. The deflection angle and rate of change are obtained through sequence deflection recognition. A deflection state discriminator is constructed by combining the chain link rotation characteristics. After anomaly detection, the control channel is invoked to execute closed-loop control, thereby achieving adaptive and precise control of the scraper chain deflection. This achieves the technical effect of adaptive regulation of the scraper conveyor chain deflection, improving the stability of equipment operation and the accuracy of control.

[0087] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a visual detection-based adaptive control system for scraper chain deflection, the system comprising:

[0088] The deflection chain link parameter acquisition module 1 is used to install a visual inspection device on the conveying trough of the scraper conveyor. The visual inspection device collects a set of scraper chain running images in real time, performs sequence deflection recognition on the set of scraper chain running images, and obtains the lateral deflection angle and the rate of change of the deflection angle of the deflection chain link.

[0089] The scraper chain abnormal parameter acquisition module 2 is used to construct a scraper chain deflection state discriminator by combining the rotation structure characteristics of the deflection chain link. Based on the scraper chain deflection state discriminator, the module performs abnormal discrimination on the lateral deflection angle and the rate of change of the deflection angle of the deflection chain link, and outputs the abnormal deflection state parameters of the scraper chain.

[0090] The target deflection control parameter acquisition module 3 is used to call the deflection strategy control channel through the deflection control module, and use the deflection strategy control channel to parse the control parameters of the abnormal deflection state of the scraper chain to determine the target deflection control parameters.

[0091] The closed-loop control execution module 4 drives the deflection control module to perform scraper chain deflection control and monitor based on the target deflection control parameters, obtains scraper chain deflection feedback parameters, and performs adaptive closed-loop control of deflection through the scraper chain deflection feedback parameters.

[0092] Furthermore, the deflection chain parameter acquisition module 1 is used to perform the following steps:

[0093] The scraper chain running image set is subjected to Gaussian filtering for noise reduction and contrast enhancement to obtain a usable scraper chain running image set. An attention mechanism module is constructed and used to extract and locate the ROI of the usable scraper chain running image set to obtain a scraper chain image region set. The scraper chain image region set is then sequentially processed according to time series to extract chain link features, resulting in a chain link running sequence feature set. The chain link running sequence feature set is then used to identify deflection angles to obtain the lateral deflection angles and deflection angle change rates of the deflected chain links.

[0094] Furthermore, the deflection chain parameter acquisition module 1 is used to perform the following steps:

[0095] The Canny operator is used to perform edge detection and extraction on the feature set of the chain loop running sequence to obtain the sequence edge lines on both sides of the chain loop; Hough transform and deflection angle calculation are performed on the sequence edge lines on both sides of the chain loop to obtain the lateral deflection angle of the deflected chain loop; the lateral deflection angle is filtered and fitted according to the timestamp to generate a deflection angle-time curve; the finite difference rate of change is calculated based on the deflection angle-time curve to obtain the rate of change of the deflection angle.

[0096] Furthermore, the deflection chain parameter acquisition module 1 is used to perform the following steps:

[0097] A preset straight line length threshold is used to perform Hough transform line detection and filtering on the edge lines of the chain loop on both sides according to the preset straight line length threshold to obtain the main edge line of the chain loop sequence. The main edge line of the chain loop sequence is fitted with a least squares method to obtain the straight line equations of the edge sequences on both sides. Based on the straight line equations of the edge sequences on both sides, the first single-side sequence angle and the second single-side sequence angle between the single-side edge line and the horizontal baseline are calculated respectively. The average value of the first single-side sequence angle and the second single-side sequence angle is obtained, and the average value is used as the lateral deflection angle of the deflection chain loop in the current sequence frame.

[0098] Furthermore, the scraper chain abnormal parameter acquisition module 2 is used to perform the following steps:

[0099] Based on the working application scenarios of the scraper conveyor, a scraper conveyor operating condition parameter table is constructed; the scraper conveyor is subjected to operation deflection tests according to the scraper conveyor operating condition parameter table, and the scraper conveyor operating deflection state dataset is recorded; the scraper conveyor operating deflection state dataset is identified and labeled according to the characteristics of the deflection chain loop rotation structure, resulting in a multi-condition scraper conveyor deflection state sample set; based on the multi-condition scraper conveyor deflection state sample set, deflection state recognition training and merging are performed to construct a scraper conveyor deflection state discriminator.

[0100] Furthermore, the scraper chain abnormal parameter acquisition module 2 is used to perform the following steps:

[0101] The scraper chain deflection state dataset is subjected to working condition clustering analysis and integration to obtain a multi-working condition deflection state dataset. Deflection state discrimination rules are mined, verified, evaluated, and corrected based on the characteristics of the deflection chain loop rotation structure to obtain multi-working condition deflection state discrimination rules. Deflection state discrimination labels are then applied to the multi-working condition deflection state dataset according to these rules to obtain a multi-working condition scraper chain deflection state sample set.

[0102] Furthermore, the target deflection control parameter acquisition module 3 is used to perform the following steps:

[0103] A set of abnormal deflection cases of the scraper chain is collected. Based on the scraper chain deflection state discriminator, the abnormal deflection cases are identified to obtain an abnormal deflection state parameter set. The abnormal deflection state parameter set is analyzed and optimized to obtain an abnormal deflection control strategy parameter set. The abnormal deflection state parameter set and the abnormal deflection control strategy parameter set are correlated and fitted to construct a deflection strategy control channel, and the deflection strategy control channel is stored in the deflection control module.

[0104] Furthermore, the target deflection control parameter acquisition module 3 is used to perform the following steps:

[0105] A scraper chain control strategy library is constructed. The scraper chain control strategy library is used to parse the abnormal deflection state parameter set to obtain the deflection control strategy parameter threshold set. Global iterative optimization is performed within the deflection control strategy parameter threshold set to obtain the abnormal deflection control strategy parameter set.

[0106] Furthermore, the target deflection control parameter acquisition module 3 is used to perform the following steps:

[0107] A deflection control multi-objective function is constructed. The deflection control multi-objective function is used to randomly select and simulate within the deflection control strategy parameter threshold set to obtain multiple deflection strategy parameter fitnesss. Based on the fitnesss of the multiple deflection strategy parameter thresholds, a global iterative optimization is performed on the deflection control strategy parameter threshold set to obtain an abnormal deflection control strategy parameter set.

[0108] The visual detection-based adaptive control system for scraper chain deflection provided in this embodiment of the invention can execute the visual detection-based adaptive control method for scraper chain deflection provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0109] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0110] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A visual detection-based adaptive control method for scraper chain deflection, characterized in that, The method includes: A visual inspection device is installed on the conveying trough of the scraper conveyor. The visual inspection device collects a set of images of the scraper chain running in real time. The sequence deflection of the scraper chain running image set is identified to obtain the lateral deflection angle and the rate of change of the deflection angle of the deflection chain link. A scraper chain deflection state discriminator is constructed by combining the characteristics of the deflection chain ring rotation structure. Based on the scraper chain deflection state discriminator, anomalies are detected in the lateral deflection angle and the rate of change of the deflection angle of the deflection chain ring, and the abnormal deflection state parameters of the scraper chain are output. The deflection control module calls the deflection strategy control channel, and uses the deflection strategy control channel to analyze the control parameters of the abnormal deflection state of the scraper chain to determine the target deflection control parameters. Based on the target deflection control parameters, the deflection control module is driven to perform scraper chain deflection control and monitoring, thereby obtaining scraper chain deflection feedback parameters. Deflection adaptive closed-loop regulation is then performed using the scraper chain deflection feedback parameters. Among them, a scraper chain deflection state discriminator is constructed by combining the characteristics of the deflection chain loop rotation structure, including: Based on the working application scenarios of the scraper chain, construct a table of scraper chain operating condition parameters; According to the scraper chain operating condition parameter table, the scraper conveyor was tested for deflection, and the scraper chain deflection status data set was recorded. By combining the characteristics of the deflection chain loop rotation structure, the deflection state dataset of the scraper chain operation is identified and labeled to obtain a multi-condition scraper chain deflection state sample set. Based on the multi-condition scraper chain deflection state sample set, deflection state recognition training and merging are performed to construct a scraper chain deflection state discriminator. The sample set of deflection states of the scraper chain under multiple working conditions was obtained, including: The scraper chain deflection state dataset is subjected to working condition clustering analysis and integration to obtain a multi-working condition deflection state dataset. Combining the characteristics of the deflection chain loop rotation structure, the deflection state discrimination rules of the multi-condition operation deflection state dataset are mined, verified, evaluated and corrected to obtain multi-condition deflection state discrimination rules; According to the multi-condition deflection state discrimination rule, the multi-condition operation deflection state dataset is labeled with deflection state discrimination to obtain a multi-condition scraper chain deflection state sample set.

2. The visual detection-based adaptive control method for scraper chain deflection as described in claim 1, characterized in that, The lateral deflection angle and the rate of change of the deflection angle of the deflection chain link are obtained, including: Gaussian filtering and contrast enhancement are applied to the scraper chain running image set to obtain a usable scraper chain running image set; An attention mechanism module is constructed, and the attention mechanism module is used to extract and locate the ROI of the available scraper chain running image set to obtain the scraper chain image region set. The scraper chain image region set is sequentially processed according to the time series to extract chain loop features, thus obtaining the chain loop operation sequence feature set; The deflection angle of the chain loop running sequence feature set is identified to obtain the lateral deflection angle and the rate of change of the deflection angle of the chain loop.

3. The visual detection-based adaptive control method for scraper chain deflection as described in claim 2, characterized in that, The deflection angle of the chain loop running sequence feature set is identified to obtain the lateral deflection angle and the rate of change of the deflection angle of the chain loop, including: The Canny operator is used to perform edge detection and extraction on the feature set of the chain loop running sequence to obtain the sequence edge lines on both sides of the chain loop; The Hough transform and deflection angle calculation are performed on the edge lines of the sequences on both sides of the chain link to obtain the lateral deflection angle of the deflected chain link; The lateral deflection angle is filtered and fitted using a moving average based on the timestamp to generate a deflection angle-time curve; The rate of change of the deflection angle is obtained by calculating the finite difference rate of change based on the deflection angle-time curve.

4. The visual detection-based adaptive control method for scraper chain deflection as described in claim 3, characterized in that, Obtain the lateral deflection angle of the deflection chain links, including: A preset straight line length threshold is used to perform Hough transform line detection and filtering on the sequence edge lines on both sides of the chain link according to the preset straight line length threshold, so as to obtain the main edge line of the chain link sequence. The least squares method is used to fit the main edge line of the chain loop sequence to obtain the straight line equations of the two side edge sequences. Based on the equations of the straight lines of the two side edge sequences, calculate the first and second single-side sequence angles between the single-side edge straight lines and the horizontal baseline; The average value of the first single-sided sequence angle and the second single-sided sequence angle is obtained, and the average value is used as the lateral deflection angle of the current sequence frame deflection chain loop.

5. The visual detection-based adaptive control method for scraper chain deflection as described in claim 1, characterized in that, The deflection control module invokes the deflection strategy control channel, including: Collect a set of abnormal deflection cases of the scraper chain, and perform anomaly detection on the set of abnormal deflection cases of the scraper chain based on the scraper chain deflection state discriminator to obtain a set of abnormal deflection state parameters; The abnormal deflection state parameter set is analyzed and optimized to obtain the abnormal deflection control strategy parameter set; The abnormal deflection state parameter set is correlated and fitted with the abnormal deflection control strategy parameter set to construct a deflection strategy control channel, and the deflection strategy control channel is stored in the deflection control module.

6. The visual detection-based adaptive control method for scraper chain deflection as described in claim 5, characterized in that, The parameter set for the abnormal deflection control strategy is obtained, including: A scraper chain control strategy library is constructed, and the scraper chain control strategy library is used to parse the abnormal deflection state parameter set to obtain the deflection control strategy parameter threshold set. Global iterative optimization is performed within the threshold set of the deflection control strategy parameters to obtain the abnormal deflection control strategy parameter set.

7. The visual detection-based adaptive control method for scraper chain deflection as described in claim 6, characterized in that, Global iterative optimization is performed within the threshold set of the deflection control strategy parameters to obtain the abnormal deflection control strategy parameter set, including: A deflection control multi-objective function is constructed, and the deflection control multi-objective function is used to randomly select and simulate the parameters within the threshold set of the deflection control strategy parameters to obtain the fitness of multiple deflection strategy parameters. Based on the fitness of the multiple deflection strategy parameters, a global iterative optimization is performed on the threshold set of the deflection control strategy parameters to obtain the abnormal deflection control strategy parameter set.

8. A visual detection-based adaptive control system for scraper chain deflection, characterized in that, For implementing the vision-based adaptive control method for scraper chain deflection according to any one of claims 1-7, the system comprises: The deflection chain link parameter acquisition module is used to deploy a visual inspection device on the conveying trough of the scraper conveyor. The visual inspection device collects a set of scraper chain running images in real time, performs sequence deflection recognition on the set of scraper chain running images, and obtains the lateral deflection angle and the rate of change of the deflection angle of the deflection chain link. The scraper chain abnormal parameter acquisition module is used to construct a scraper chain deflection state discriminator by combining the characteristics of the deflection chain ring rotation structure. Based on the scraper chain deflection state discriminator, the module performs abnormal discrimination on the lateral deflection angle and the rate of change of the deflection angle of the deflection chain ring and outputs the abnormal deflection state parameters of the scraper chain. The target deflection control parameter acquisition module is used to call the deflection strategy control channel through the deflection control module, and use the deflection strategy control channel to parse the control parameters of the abnormal deflection state of the scraper chain to determine the target deflection control parameters. The closed-loop control execution module drives the deflection control module to perform scraper chain deflection control and monitor based on the target deflection control parameters, obtains scraper chain deflection feedback parameters, and performs adaptive closed-loop control of deflection through the scraper chain deflection feedback parameters.

Images