Intelligent Speed ​​Control System and Method for Scraper Conveyors Based on Digital Twin Coal Flow Distribution and Effect Evaluation

Abstract

This application provides an intelligent speed control system and method for scraper conveyors based on digital twin coal flow distribution and effect evaluation, belonging to the field of intelligent speed control for scraper conveyors. The system includes a system control module, a multi-source data acquisition module, a data preprocessing module, a multi-modal feature fusion module, a data-driven prediction module, a digital twin module, a verification module, a variable frequency speed control module, and an effect evaluation module. The multi-source data acquisition module acquires multi-source data; the multi-modal feature fusion module fuses the multi-source data; the data-driven prediction module predicts coal flow load and initial speed control; the verification module determines whether the predicted value is within a reasonable threshold; the variable frequency speed control module performs variable frequency speed control based on the verification result when the verification fails; the effect evaluation module uses energy consumption, efficiency, and wear as multi-objective evaluation indicators to verify and update the data-driven prediction module. This application has significant practical implications for reducing energy consumption, improving coal mining efficiency, and reducing component wear.

Description

Technical Field

[0001] This application relates to the field of intelligent speed regulation technology for scraper conveyors, and in particular to an intelligent speed regulation system and method for scraper conveyors based on digital twin coal flow distribution and effect evaluation. Background Technology

[0002] In modern coal mine fully mechanized longwall faces, scraper conveyors are the core equipment connecting the coal mining machine and the subsequent transportation system. Their operating status directly determines the production efficiency and economic benefits of the entire working face. However, in actual production, the coal cutting process of the coal mining machine is not uniform in speed and quantity. Affected by a variety of complex and dynamically changing factors such as coal seam occurrence conditions, geological structure changes, coal mining machine cutting technology, and driver operating habits, the instantaneous coal output of the coal mining machine on the working face exhibits drastic and nonlinear fluctuations. This drastic fluctuation in coal flow leads to an extremely uneven amount of coal carried by the scraper conveyor, resulting in energy waste and a decrease in coal mining efficiency. At the same time, it exacerbates the ineffective wear of key components such as chains, scrapers, and central troughs, significantly increasing operation and maintenance costs and shortening equipment lifespan. Summary of the Invention

[0003] To address the aforementioned technical issues, this application proposes a trigger intelligent speed control system and method based on digital twin coal flow distribution and effect evaluation.

[0004] The technical solution adopted in this application is: an intelligent speed control system for scraper conveyors based on digital twin coal flow distribution and effect evaluation, comprising:

[0005] Multi-source data acquisition module: used to acquire three-dimensional point cloud data, RGB image data, and multi-source data of the operating load and operating status parameters of the scraper conveyor in the coal flow area.

[0006] Data preprocessing module: used to process multi-source data and extract and classify general features to obtain multimodal features including point cloud geometric features, RGB image texture features and operating sensor features;

[0007] Multimodal feature fusion module: used to fuse multimodal features after they have been processed by the data preprocessing module;

[0008] Data-driven prediction module: used to predict coal flow load based on the multi-dimensional feature vector sequence output by the multi-modal feature fusion module, and output preliminary speed regulation parameters;

[0009] Digital twin module: used to achieve virtual simulation through 3D modeling, data mapping and synchronization strategies;

[0010] The system control module is used to receive various raw data transmitted from the multi-source data acquisition module, schedule the execution of the data preprocessing module, multimodal feature fusion module and data-driven prediction module in the order of preset control logic and timing rules; and during the operation of the digital twin module, it monitors the consistency of the operating status of the virtual model and the corresponding physical device in real time, and immediately starts the variable frequency speed control module when the verification module fails the judgment result.

[0011] Verification module: Used to compare the predicted value of the data-driven prediction module with historical patterns and the safety parameter thresholds specified by the equipment manufacturer. If the verification passes, the speed of the scraper conveyor is controlled according to the predicted value. If the verification fails, the variable frequency speed control module is activated.

[0012] Variable frequency speed control module: used to accurately calculate the frequency adjustment amount based on the deviation between the predicted value and the safety parameter threshold specified by the equipment manufacturer when the verification fails;

[0013] Performance evaluation module: Using energy consumption, efficiency and wear as multi-objective evaluation indicators, the data-driven prediction module is validated and updated.

[0014] Furthermore, the multi-source data acquisition module includes a lidar installed directly above the scraper conveyor chute, a binocular camera coaxially mounted with the lidar, and sensors for acquiring scraper conveyor operating data. The scanning range of the lidar and the image area acquired by the binocular camera completely overlap, both fully covering the entire coal flow conveying area. Point cloud data and RGB image data of the coal flow conveying area are acquired by the lidar and the binocular camera, respectively.

[0015] The sensors include a current sensor connected in series in the motor power supply circuit of the scraper conveyor and a speed sensor installed at the end of the drive drum shaft of the scraper conveyor.

[0016] Furthermore, when processing data, the data preprocessing module first connects multi-source data to the module for time synchronization calibration, unifying the data collected by LiDAR, binocular cameras, and sensors to the same time reference. Then, noise is eliminated through filtering. Next, the data format is converted to a standardized format and dimensionality reduction is performed. Then, based on spatial coordinate matching and timestamp association, a mapping relationship is constructed between point cloud data, RGB image data, and sensor data. Finally, multimodal feature extraction is performed.

[0017] Furthermore, the time synchronization calibration in the data preprocessing module uses a network-based time protocol or a more precise time protocol to stamp the output data of all data sources with a unified high-precision timestamp. At the same time, an adaptive time synchronization mechanism for downhole noise is also set up, which uses the load fluctuation of the current sensor as an anchor point to dynamically calibrate the timestamp deviation between point cloud data and RGB image data caused by downhole vibration anomalies, and achieves millisecond-level synchronization by combining Kalman filtering.

[0018] Furthermore, multimodal feature extraction is performed by classifying and processing different data types, including outlier removal, voxel sampling, point cloud geometric feature extraction, and volume calculation for point cloud data.

[0019] Image quality enhancement, noise filtering, edge detection, texture feature calculation, and contour and foreign object detection are performed on RGB image data.

[0020] The core indicators of the working condition monitoring sensor data are extracted and verified, and abnormal and invalid data are removed.

[0021] Furthermore, the process of the multimodal feature fusion module is as follows:

[0022] First, RGB image features are fused with point cloud geometric features. Based on the volume, curvature, and density features of the point cloud, texture features and foreign object confidence of the RGB image data are fused to construct coal flow features. A weighted fusion algorithm is used to assign weights according to the importance of different features. Then, the coal flow features are fused with operating condition sensor data to realize the correlation between coal flow status and equipment operating status. The correlation between coal flow features and equipment operating parameters is established through a feature mapping matrix to achieve dynamic matching between coal flow status and equipment load.

[0023] Furthermore, based on the spatiotemporal coordinate unification of RGB image features and point cloud geometric features and the independent extraction of multi-source features, the fusion process introduces an adaptive weight function for foreign objects. The confidence score of foreign objects generated by dual-dimensional detection based on the color threshold deviation and contour anomaly of the RGB image side is used as the gating signal. The confidence score of foreign objects is a quantitative indicator for underground coal conveying scenarios, with a value range of 0-1, which is used to accurately characterize the probability of the existence of non-coal foreign objects in the coal flow detection area.

[0024] Furthermore, the verification module pre-stores a priori pattern model of the normal operating range of the equipment based on a large amount of historical data, as well as safety parameter thresholds specified by the equipment manufacturer. After receiving the 5-10 second coal flow load prediction output by the data-driven prediction module, the verification module calculates the matching degree with the priori pattern model and compares it with the safety parameter thresholds to determine whether the predicted value conforms to historical patterns. If the predicted value is within a reasonable range and conforms to historical patterns, it is judged as passing, and the system maintains the current speed. If the predicted value exceeds the reasonable range or deviates significantly from historical patterns, it is judged as failing, and the variable frequency speed control module is triggered.

[0025] Furthermore, the effectiveness evaluation module includes a built-in effectiveness evaluation model, which is represented by a comprehensive evaluation function. The expression of the comprehensive evaluation function is as follows:

[0026] ;

[0027] Where α, β, and γ are weight coefficients set according to production needs; after each speed adjustment action, this module calculates the comprehensive evaluation score Score over a period of time. If the score is lower than the preset threshold, it is determined that the prediction and speed adjustment effect are not up to standard. The reasons are analyzed and the data-driven prediction module is updated. By adding the current data to the training set, the model of the data-driven prediction module is incrementally learned or the parameters are tuned to achieve closed-loop optimization.

[0028] The three weighting coefficients, α, β, and γ, are dynamically adjusted according to production needs using an adaptive optimization algorithm.

[0029] A method for intelligent speed control of scraper conveyors based on digital twin coal flow distribution and effect evaluation, employing the aforementioned intelligent speed control system for scraper conveyors based on digital twin coal flow distribution and effect evaluation, includes the following steps:

[0030] Step 1: Multi-source data acquisition and preprocessing. Based on the multi-source data acquisition module, three-dimensional point cloud data, RGB image data, and operating load and operating status parameters of the scraper conveyor are collected. The multi-source data is preprocessed and feature extracted and classified by the data preprocessing module to obtain multi-modal features including point cloud geometric features, RGB image texture features and operating feature of the working condition sensor.

[0031] Step 2: Perform feature fusion on the multimodal features to obtain a multidimensional feature vector sequence;

[0032] Step 3: Based on the multidimensional feature vector sequence, use Transformer temporal correlation to predict the coal flow load for a duration of 5-10 seconds and output preliminary speed regulation parameters;

[0033] Step 4: Verify the predicted value in Step 3. If the verification is successful, control the speed of the scraper conveyor according to the predicted value, and at the same time input the predicted value into the digital twin module for synchronous simulation and monitoring.

[0034] S5: If the verification fails, the variable frequency speed control module will be activated to adjust the speed of the scraper conveyor. After speed adjustment, its energy consumption, efficiency, and wear data will be evaluated to see if they meet the standards. If they meet the standards, the scraper conveyor can be controlled to operate normally. If the evaluation fails, the data-driven prediction model will be updated and the updated data-driven prediction model will be used to predict the coal flow load.

[0035] The advantages of this application over the prior art are as follows:

[0036] 1. In this application, multi-source data such as lidar point cloud, binocular vision RGB image and traditional working condition sensor are innovatively integrated to realize all-round and three-dimensional perception of coal flow morphology, surface texture and equipment load, providing a comprehensive and accurate data foundation for intelligent speed regulation of scraper conveyor.

[0037] 2. This application proposes a multi-objective effect evaluation module centered on "energy consumption, efficiency, and wear," forming a closed-loop feedback. After each speed adjustment decision, the system automatically evaluates its comprehensive impact on energy consumption, transportation efficiency, and equipment wear. This enables the system to continuously learn and optimize itself, ultimately finding the optimal operating range with the highest energy efficiency and lowest equipment wear while ensuring safe transportation. This achieves a fundamental shift from single-function control to maximizing comprehensive benefits. Attached Figure Description

[0038] The following description, in conjunction with the accompanying drawings, further illustrates this application:

[0039] Figure 1 This application provides a schematic diagram of the system module structure and corresponding data transmission in an embodiment.

[0040] Figure 2 This is a flowchart of multi-source data processing provided in an embodiment of this application;

[0041] Figure 3 A flowchart illustrating the intelligent speed control closed-loop process of a scraper conveyor provided in an embodiment of this application. Detailed Implementation

[0042] like Figures 1 to 3 As shown, this application provides an intelligent speed control system for scraper conveyors based on digital twin coal flow distribution and effect evaluation, including the following modules:

[0043] Multi-source data acquisition module: used to acquire multi-source data such as three-dimensional point cloud data, RGB image data, and operating load and operating status parameters of the scraper conveyor in the coal flow area;

[0044] Data preprocessing module: used to perform spatiotemporal alignment, denoising, filtering, format conversion, dimensionality reduction, enhancement and other processing on multi-source data, as well as general feature extraction and classification, to obtain multimodal features including point cloud geometric features, RGB image texture features and operating sensor features;

[0045] Multimodal feature fusion module: used to fuse multimodal features after they have been processed by the data preprocessing module;

[0046] The data-driven prediction module is used to predict coal flow load over a period of 5-10 seconds by using the Transformer to achieve temporal correlation based on the multi-dimensional feature vector sequence output by the multi-modal feature fusion module, and output preliminary speed regulation parameters.

[0047] Digital twin module: Achieves virtual simulation through 3D modeling, data mapping, and synchronization strategies;

[0048] The system control module coordinates the data and command flows between multiple modules. Specifically, it receives various raw data from the multi-source data acquisition module and schedules the execution of the data preprocessing module, multimodal feature fusion module, and data-driven prediction module according to preset control logic and timing rules. During the operation of the digital twin module, it monitors the consistency between the virtual model and the corresponding physical equipment in real time, including the matching degree of key indicators such as coal flow morphology and equipment operating parameters. When the verification module fails the judgment, it immediately activates the priority processing mechanism of the variable frequency speed control module to ensure rapid response of speed control actions. At the same time, the system control module integrates real-time evaluation data fed back by the effect evaluation module and dynamically adjusts the core parameters of the model of the data-driven prediction module according to the evaluation results to optimize the prediction accuracy and adaptability of the model.

[0049] Verification module: Used to compare the predicted value of the data-driven prediction module with historical patterns and the safety parameter thresholds specified by the equipment manufacturer. If the verification passes, the speed of the scraper conveyor is controlled according to the predicted value. If the verification fails, the variable frequency speed control module is activated.

[0050] Variable frequency speed control module: When the verification fails, it is used to accurately calculate the frequency adjustment amount based on the deviation between the predicted value and the safety parameter threshold specified by the equipment manufacturer, using a PID control algorithm and the coordinated action of proportional, integral, and derivative components.

[0051] Performance evaluation module: Using energy consumption, efficiency and wear as multi-objective evaluation indicators, the data-driven prediction module is validated and updated.

[0052] In this embodiment, LiDAR, binocular cameras, and current and speed sensors can be used as data acquisition devices. Specifically, the LiDAR is installed above the scraper conveyor to collect three-dimensional point cloud data of the coal flow and its surroundings. The specific installation position is at a preset height directly above the scraper conveyor chute to ensure that the scanning range completely covers the entire coal flow transportation area. The binocular camera is used to collect three-dimensional RGB image data of the coal flow area to obtain data such as surface texture and foreign objects. The binocular camera is installed correspondingly to the LiDAR. Specifically, the binocular camera and the LiDAR are coaxial and maintain a fixed distance. The installation angle is calibrated to ensure that the acquired image area completely overlaps with the LiDAR scanning area. The current and speed sensors (operating condition monitoring sensors) are installed on the scraper conveyor motor and transmission components. The current sensor is connected in series in the motor power supply circuit, and the speed sensor is installed at the end of the transmission drum shaft to collect the operating load and status parameters of the scraper conveyor in real time, including key data such as motor operating current and transmission component speed. The sampling frequency meets the real-time monitoring requirements.

[0053] In one specific embodiment, an industrial-grade LiDAR, such as a high-beam solid-state or hybrid solid-state LiDAR, is installed above the scraper conveyor chute to acquire high-density 3D point cloud data of the coal flow and its surrounding environment in real time through high-speed scanning. Its sampling frequency needs to reach 10-20Hz to capture the rapid dynamic changes in the coal flow. An industrial-grade binocular stereo camera is installed corresponding to the LiDAR to acquire synchronous stereo images of the coal flow area. Binocular vision not only provides RGB image texture information but also assists in generating depth maps through parallax calculation, complementing the point cloud data. The camera needs to have HDR (High Dynamic Range) to cope with complex lighting changes in the mine, and its frame rate needs to be synchronized with the LiDAR. High-precision current, speed, and vibration sensors are installed on key transmission components such as the scraper conveyor motor and reducer. For example, Hall effect current sensors and photoelectric encoder-type speed sensors are used. These sensors require higher data acquisition frequencies, typically in the kHz range, to capture transient signals such as current surges and speed fluctuations. To ensure real-time performance, the data acquisition hardware platform can be an embedded system based on an ARM architecture processor and equipped with an RTOS (Real-Time Operating System) to ensure deterministic task scheduling and low latency.

[0054] The data preprocessing module first integrates multi-source data for time synchronization calibration, unifying data collected from different devices such as LiDAR, binocular cameras, and sensors to the same time base to ensure data consistency. Then, it filters out noise, employing appropriate filtering algorithms for different data types, such as bilateral filtering for point cloud data and moving average filtering for sensor data. Next, it converts the data format to a standardized format and performs dimensionality reduction, unifying non-standard data outputs from different devices into the system's preset standardized format. Principal component analysis or linear discriminant analysis are used for dimensionality reduction to remove redundant information. The process involves several steps: first, establishing a correspondence between multiple data sources, constructing a mapping relationship between point cloud data, RGB image data, and sensor data based on spatial coordinate matching and timestamp association; second, performing multimodal feature extraction, including outlier removal, voxel sampling, extraction of point cloud geometric features (coal flow morphology features), and volume calculation for point cloud data; third, image quality enhancement, noise filtering, edge detection, texture feature calculation, and contour and foreign object detection for RGB image data; and fourth, extraction and verification of core indicators for sensor data, such as peak current, average rotational speed, and torque fluctuation range, which are then compared with preset effective data thresholds to remove abnormal and invalid data.

[0055] Time synchronization calibration is the primary prerequisite for multi-source data fusion. The system can use a network-based time protocol or a more precise time protocol to stamp all data sources with a unified high-precision timestamp. During data processing, a time window is used to align data frames from different sources, ensuring that subsequent fusion captures scene snapshots from the same moment. Simultaneously, the data preprocessing module incorporates a downhole noise adaptive time synchronization mechanism. Utilizing load fluctuations from current sensors as anchor points, it dynamically calibrates the timestamp discrepancies between point cloud data and RGB image data caused by downhole vibration anomalies. Combined with Kalman filtering, millisecond-level synchronization is achieved. This mechanism significantly improves the alignment accuracy of multi-source data and avoids feature drift caused by traditional fixed synchronization.

[0056] In this embodiment, the point cloud data processing flow mainly includes denoising and sampling, and feature extraction. Denoising and sampling first use either a statistical outlier removal algorithm or a radius outlier removal algorithm to filter out isolated noise points. Specifically, the statistical outlier removal algorithm calculates the distance deviation between each point and its neighboring points, identifying points exceeding a preset confidence interval as noise points and removing them. The radius outlier removal algorithm searches the neighborhood of each point with a specified radius, filtering out points with fewer than a threshold number of points in the neighborhood as noise points. Then, to reduce computational complexity while preserving morphological features, a voxel grid downsampling method is used to reduce the dimensionality of the point cloud. This method constructs a three-dimensional voxel grid, retaining the most representative points within each voxel, thus reducing the amount of data while maintaining the original three-dimensional morphological features of the coal flow to the greatest extent. Point cloud geometric feature extraction is used to calculate the macroscopic morphological features of the coal flow. The volume of the coal flow is estimated by calculating the number of voxels occupied by the point cloud, and combined with preset volume parameters of the voxels, the approximate volume of the actual coal flow is quickly calculated. The feature extraction process is as follows: extract the local geometric features of the point cloud, including the normal vector and curvature of each point, calculate the normal vector by fitting the plane of the neighborhood point cloud using the least squares method, and solve the curvature value based on the rate of change of the normal vector. These features can reflect the smoothness and undulation of the coal flow surface, providing detailed support for subsequent coal flow state analysis.

[0057] In denoising, for any point in the point cloud P First, find its k nearest neighbors, denoted as . .calculate Average distance to these k neighboring points :

[0058] (1);

[0059] in It is the Euclidean distance between two points.

[0060] Then, calculate the mean of the average distances of all points in the entire point cloud. and standard deviation :

[0061] (2);

[0062] (3).

[0063] A point A point is identified as an outlier if its average neighborhood distance exceeds a threshold defined by a global statistic. This threshold is set as the mean plus a multiple of the standard deviation.

[0064] (4);

[0065] in It is the standard deviation multiple, an adjustable parameter. If Then point It is considered noise and removed.

[0066] In point cloud geometric feature extraction, the volume of coal flow in a region is calculated using the voxel raster method. Simultaneously, the volume change rate between adjacent time slices is calculated to reflect the increasing or decreasing trend of coal flow. The volume change rate is obtained by comparing the difference in coal flow volume between two consecutive time points with the time interval, providing a direct visual representation of coal flow changes. The three-dimensional morphological features of the coal flow are obtained by calculating the "normal vector, curvature, and point cloud density" of the point cloud. Point cloud density is obtained by statistically analyzing the number of points per unit volume, aiding in determining the thickness of the coal flow accumulation.

[0067] If the preset volume of the voxel is And point clouds occupy Individual factors determine the estimated volume of the coal flow. for:

[0068] (5);

[0069] By comparing two consecutive timestamps and coal flow volume and The volume change rate is obtained as follows:

[0070] (6);

[0071] in , , These represent the grid step size of a single preset voxel in the three orthogonal directions of the X-axis, Y-axis, and Z-axis of the Cartesian coordinate system.

[0072] The RGB image data processing flow in this embodiment mainly includes: quality enhancement and feature extraction. Quality enhancement primarily addresses complex environmental issues such as uneven lighting and dust interference in mines. Adaptive histogram equalization and other algorithms are used to enhance image contrast. This algorithm dynamically adjusts the grayscale range based on the local grayscale distribution of the image, effectively improving the detail display in dark areas. Median filtering or Gaussian filtering is used to remove noise. Median filtering efficiently removes salt-and-pepper noise, while Gaussian filtering smooths random noise and improves overall image clarity. Feature extraction utilizes the gray-level co-occurrence matrix to calculate texture features such as "energy, entropy, and contrast" of the image. By statistically analyzing the co-occurrence probability of grayscale values ​​at different directions and distances, these texture parameters are quantified and extracted. These features reflect the coal block size distribution; coarser texture indicates larger coal blocks, while more uniform texture corresponds to smaller blocks. The feature extraction process is as follows: coal flow contours are extracted using edge detection algorithms such as Canny. The image is first smoothed using Gaussian filtering, then the gradient intensity and direction are calculated. Clear coal flow edge contours are obtained through non-maximum suppression and double threshold segmentation. Meanwhile, a color model is established based on the color range of coal under standard illumination. The RGB color parameter ranges of different coal types under standard illumination are collected in advance to build a standardized color model library. By setting thresholds or using machine learning classifiers, areas with abnormal colors are identified for foreign object detection. When an area exceeding the color range of coal is detected, it is marked as a suspected foreign object area and triggers the subsequent verification process.

[0073] To quantify the coal block size distribution, the system uses a gray-level co-occurrence matrix (GLCM) to extract texture features. Let the image gray levels be L, and the GLCM be a... matrix , of which elements The value represents the direction. and distance Above, a grayscale value The pixel and its neighboring gray values ​​are The number of times each pixel pair occurs is recorded. This is typically normalized to obtain a probability matrix. Based on this matrix, multiple texture features, including energy, entropy, and contrast, can be calculated.

[0074] The expression for energy is as follows:

[0075] (7);

[0076] The expression for entropy is as follows:

[0077] (8);

[0078] The expression for contrast is as follows:

[0079] (9);

[0080] in The original gray-level co-occurrence moment.

[0081] Based on the data from the operating condition monitoring sensors, the load characteristics of the coal flow operation are obtained by analyzing the mean, peak values, and load fluctuations of current, speed, and torque. The load fluctuations are obtained by calculating the standard deviation of current, speed, or torque over a period of time, reflecting the stability of the load.

[0082] The multimodal feature fusion module first unifies the coordinate system of the preprocessed RGB image features and point cloud data. By converting the pixel coordinates of the RGB image and the point cloud coordinates into a coordinate system with the center of the scraper conveyor chute as the coordinate, a three-dimensional world coordinate system is established with the central axis of the scraper conveyor chute as the reference. Through camera calibration parameters and the lidar extrinsic matrix, the conversion from pixel coordinates and point cloud coordinates to the world coordinate system is completed, realizing the unification of spatial position. Finally, multimodal feature fusion is performed to obtain fused feature values.

[0083] The RGB image features include texture features and foreign object confidence features, while the point cloud geometric features include the volume, curvature, and density features of the point cloud.

[0084] Specifically, firstly, RGB image features are fused with point cloud geometric features. Based on the volume, curvature, and density features of the point cloud, texture features and foreign object confidence of the RGB image data are fused to construct coal flow features. A weighted fusion algorithm is used to assign weights according to the importance of different features, organically combining point cloud geometric features with RGB image features to form a more comprehensive description of coal flow features. Then, the coal flow features are fused with the physical parameters of the working condition monitoring to associate the coal flow status with the equipment operating status. The correlation between the coal flow features and the equipment operating parameters is established through a feature mapping matrix to achieve dynamic matching between the coal flow status and the equipment load.

[0085] The fusion process, based on the spatiotemporal coordinate unification of RGB image features and point cloud geometric features and the independent extraction of multi-source features, introduces an adaptive weighting function for foreign objects. It uses the foreign object confidence score, generated from dual-dimensional detection of RGB image-side color threshold deviation and contour anomaly, as a gating signal. The foreign object confidence score is a quantitative indicator specifically designed for underground coal conveying scenarios, ranging from 0 to 1. It accurately characterizes the probability of the presence of non-coal foreign objects such as gangue within the coal flow detection area and is the core basis for the dynamic weight allocation in the aforementioned weighted fusion algorithm. This allows for real-time adjustment of the fusion weights. When the foreign object confidence score is high, indicating a risk of foreign objects in the coal flow, the contribution ratio of RGB texture features is increased, while the fusion weights of geometric features such as point cloud volume, curvature, and density are suppressed. When the foreign object confidence score is low, indicating a normal and clean coal flow, a balanced weight allocation is applied to the two types of features. This mechanism effectively filters interference from common underground debris such as gangue, improves the accuracy of coal flow purity identification, and provides stable and reliable feature support for the subsequent fusion and matching of coal flow features with physical parameters of operating conditions.

[0086] The core of this function is a gating mechanism: Preprocessed RGB image features are used to calculate texture features via a gray-level co-occurrence matrix, and combined with a contour detection algorithm to extract coal flow edges. Foreign objects are identified using a standard coal flow color threshold and shape anomalies, and a confidence level is calculated. , where 0 represents pure coal flow and 1 represents high-confidence foreign matter.

[0087] The gating weights are dynamically generated using the sigmoid gating function.

[0088] (10);

[0089] in For the sigmoid function, These are learnable parameters, optimized through backpropagation during training. For dynamic gating weights of RGB image features, To and Completely complementary gating weight coefficients.

[0090] The fusion formula is:

[0091] (11);

[0092] in As a feature of fusion, This is a confidence vector for texture and foreign object. For volume / morphological characteristics, For load parameters, weights Fixed. This gating mechanism, which draws on attention gating, ensures that reliable modal information is prioritized during fusion, avoiding misjudgments caused by volume expansion due to foreign objects.

[0093] The data-driven prediction process of the data-driven prediction module is as follows: First, it receives the multi-dimensional feature vector sequence output by the multi-modal feature fusion module, performs feature dimensionality reduction and weight allocation on the feature vectors, and uses principal component analysis or autoencoder to remove feature redundancy. At the same time, it uses an attention mechanism to assign adaptive weights to features of different dimensions to highlight the influence of key features. Then, it is input into the Transformer encoder. Based on the Transformer architecture, it captures the long-term dependencies between features at different time steps through multi-head self-attention. The decoder outputs the coal flow load prediction value for the next 5-10 seconds as the initial speed adjustment parameter. The multi-head self-attention mechanism uses multiple attention heads to compute in parallel, capturing time dependencies at different scales to improve the accuracy of prediction. Then, it trains the model using historical normal operation data, and balances prediction accuracy and stability through weighted loss. The weighted loss function assigns different weights to the prediction results in different error ranges, imposing higher penalties on prediction results with large deviations, while also taking into account the stability of the model output. Finally, it combines the feedback from the effect evaluation model built into the effect evaluation module to continuously iterate the parameters to adapt to the coal flow change trend. Based on the error information from the evaluation feedback, it dynamically adjusts the model's training parameters to improve the model's adaptability to complex coal flow changes.

[0094] The digital twin module realizes the digital twin process of coal flow in a scraper conveyor through three steps: 3D twin construction, data mapping, and real-time virtual-real synchronization. The 3D twin construction involves building a 3D model twin in Unity3D based on the scraper conveyor parameters. Three-dimensional features of the coal flow are generated using point cloud reconstruction algorithms, and then the surface texture of the coal flow from RGB images is overlaid to construct a virtual coal flow. Data mapping is achieved by converting pre-processed physical quantities from monitoring sensors into digital signals for the twin. Real-time virtual-real synchronization uses point cloud frame difference analysis to capture dynamic changes in the coal flow based on pre-processed point cloud data, driving real-time updates of the twin. The digital twin module also includes a human-machine interface for visually displaying the 3D distribution of the coal flow, equipment operating parameters, and performance evaluation indicators. The human-machine interface allows users to manually adjust speed parameters and records operation logs for model optimization.

[0095] The digital twin module provides a human-machine interface, allowing operators to observe the three-dimensional distribution, thickness, and presence of foreign objects in the coal flow from any angle, and to view the operating parameters and evaluation indicators such as energy consumption and efficiency of each component in real time. In simulation mode, users can manually adjust the speed control parameters, and the twin system will simulate the operating status and changes in evaluation indicators of the scraper conveyor under those parameters based on the built-in physical model. This provides a zero-risk experimental platform for training operators and validating new control strategies.

[0096] The verification module pre-stores a priori model of the equipment's normal operating range based on a large amount of historical data, as well as safety parameter thresholds specified by the equipment manufacturer. After receiving a 5-10 second coal flow load prediction from the data-driven prediction module, the verification module calculates the matching degree with the priori model and compares it with the safety parameter thresholds to determine if the predicted value conforms to historical patterns. If the predicted value is within a reasonable range and conforms to historical patterns, it is considered passed, and the system maintains the current speed. If the predicted value exceeds the reasonable range or deviates significantly from historical patterns, it is considered failed, and the variable frequency speed control module is triggered. Furthermore, the verification module constructs a normal operating range model of the equipment through statistical analysis of massive amounts of historical normal operating data. By employing statistical process control methods, it calculates control charts for various key parameters, such as current and power factor, under different load levels, defining upper and lower control limits. Moreover, the safety parameter thresholds are absolute red lines specified by the equipment manufacturer or set according to industry safety regulations, such as the maximum allowable current of the motor and the maximum allowable temperature of the reducer; these are hard constraints that must be unconditionally followed.

[0097] When the variable frequency speed control module receives a "fail" command from the verification module, it uses a PID control algorithm to generate a corresponding frequency adjustment command based on the deviation between the predicted value and the threshold. The variable frequency speed control module then adjusts the speed of the scraper conveyor motor through the frequency converter to achieve on-demand speed regulation.

[0098] In the PID control algorithm, the sampling period is set to be... In the Each sampling time has an error of [number] times. Incremental PID algorithm calculates the increment of the control quantity. :

[0099] (12);

[0100] The final control output is:

[0101] (13);

[0102] By adjusting These three parameters enable the scraper conveyor to smoothly and quickly follow changes in coal flow load.

[0103] The built-in effectiveness evaluation model in the effectiveness evaluation module uses a comprehensive evaluation function, the expression of which is as follows:

[0104] (14);

[0105] Where α, β, and γ are weighting coefficients set according to production needs; after each speed adjustment action, this module calculates a comprehensive evaluation score Score over a period of time. If the score is lower than the preset threshold, it is determined that the prediction and speed adjustment effect is not up to standard. The reasons are analyzed and the data-driven prediction module is updated. By adding the current data to the training set, the model of the data-driven prediction module is incrementally learned or the parameters are tuned to achieve closed-loop optimization.

[0106] The three weight coefficients (α, β, γ) are not fixed but can be dynamically adjusted according to production needs, which reflects the intelligence of the evaluation model. The system can employ advanced adaptive optimization algorithms to calibrate α, β, and γ online. An online reinforcement learning framework can be introduced, taking the system's state, speed adjustment decisions, and overall evaluation score as input. The reinforcement learning agent aims to learn an optimal weight adjustment strategy to maximize long-term cumulative rewards. Particle swarm optimization is also a viable option. α, β, and γ can be used as particle positions in the swarm, and the average score over a period of time can be used as the fitness function. Through iterative optimization of the particle swarm, the optimal weight combination under the current production mode can be found. The weight coefficients can be updated based on a sliding data window, such as the operating data of the past hour. The parameter update cycle can be set to a fixed time or triggered when a significant change in the production task is detected.

[0107] Energy efficiency indicators use unit energy consumption, with SEC as the core metric. SEC is defined as the energy consumed to transport a unit mass of material a unit distance:

[0108] (15);

[0109] in It is the total input power (kW) of the scraper conveyor, which can be measured by current, voltage, and power factor sensors; It is the material conveying rate (ton / h), i.e., transportation efficiency; It is the length of the scraper conveyor (km).

[0110] Transportation efficiency is directly measured using the quality delivery rate.

[0111] (16);

[0112] in It is the bulk density of coal. This can be used as an empirical value or estimated by a model; It is the average cross-sectional area of ​​the coal flow. It can be calculated from point clouds or image contours; It is the operating speed of the scraper chain. The speed is measured by a speed sensor.

[0113] The equipment wear index is based on Akkad's law of wear, which correlates wear volume with normal force, sliding distance, and material hardness:

[0114] (17);

[0115] in It represents the volume of material lost due to wear. It is a dimensionless wear coefficient that depends on the contact material pair and the wear mechanism; It is the normal load acting on the contact surface. For scrapers and chutes, this is related to the weight of the coal and the weight of the scraper itself; It is the total sliding distance; It is the Brinell hardness of softer materials.

[0116] This application also proposes an intelligent speed control method for scraper conveyors based on digital twin coal flow distribution and effect evaluation. Based on the hardware of the above system, the method includes the following steps:

[0117] Step 1: Multi-source data acquisition and preprocessing. Based on the multi-source data acquisition module, three-dimensional point cloud data, RGB image data, and operating load and operating status parameters of the scraper conveyor in the coal flow area are collected. The multi-source data is preprocessed and feature extracted and classified by the data preprocessing module to obtain multi-modal features including point cloud geometric features, RGB image texture features and operating sensor features.

[0118] Step 2: Perform feature fusion on the multimodal features to obtain a multidimensional feature vector sequence;

[0119] Step 3: Based on the multidimensional feature vector sequence, use Transformer temporal correlation to predict the coal flow load for a duration of 5-10 seconds and output preliminary speed regulation parameters;

[0120] Step 4: Verify the predicted value in Step 3. If the verification is successful, control the speed of the scraper conveyor according to the predicted value, and at the same time input the predicted value into the digital twin module for synchronous simulation and monitoring.

[0121] S5: If the verification fails, the variable frequency speed control module will be activated to adjust the speed of the scraper conveyor. After speed adjustment, its energy consumption, efficiency, and wear data will be evaluated to see if they meet the standards. If they meet the standards, the scraper conveyor can be controlled to operate normally. If the evaluation fails, the data-driven prediction model will be updated and the updated data-driven prediction model will be used to predict the coal flow load.

[0122] The system and method described in this application have significant practical implications for reducing energy consumption, improving coal mining efficiency, and reducing component wear.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A digital twin-based coal flow distribution and effect evaluation-based intelligent speed regulation system for a scraper, characterized by: include: Multi-source data acquisition module: used to acquire three-dimensional point cloud data, RGB image data, and multi-source data of the operating load and operating status parameters of the scraper conveyor in the coal flow area. Data preprocessing module: used to process multi-source data and extract and classify general features to obtain multimodal features including point cloud geometric features, RGB image texture features and operating sensor features; Multimodal feature fusion module: used to fuse multimodal features after they have been processed by the data preprocessing module; Data-driven prediction module: used to predict coal flow load based on the multi-dimensional feature vector sequence output by the multi-modal feature fusion module, and output preliminary speed regulation parameters; Digital twin module: used to achieve virtual simulation through 3D modeling, data mapping and synchronization strategies; System control module: Used to receive various types of raw data transmitted from the multi-source data acquisition module, and schedule the execution of the data preprocessing module, multimodal feature fusion module and data-driven prediction module in the order of preset control logic and timing rules; During the operation of the digital twin module, the consistency between the virtual model and the corresponding physical device's operating status is monitored in real time, and the variable frequency speed control module is immediately activated when the verification module fails to pass the judgment. Verification module: Used to compare the predicted value of the data-driven prediction module with historical patterns and the safety parameter thresholds specified by the equipment manufacturer. If the verification passes, the speed of the scraper conveyor is controlled according to the predicted value. If the verification fails, the variable frequency speed control module is activated. Variable frequency speed control module: used to accurately calculate the frequency adjustment amount based on the deviation between the predicted value and the safety parameter threshold specified by the equipment manufacturer when the verification fails; Performance evaluation module: Using energy consumption, efficiency and wear as multi-objective evaluation indicators, the data-driven prediction module is validated and updated.

2. The intelligent speed regulation system for a scraper based on digital twin coal flow distribution and effect evaluation according to claim 1, characterized in that: The multi-source data acquisition module includes a lidar installed directly above the scraper conveyor chute, a binocular camera mounted coaxially with the lidar, and sensors for acquiring scraper conveyor operating data. The scanning range of the lidar and the image area acquired by the binocular camera completely overlap, both fully covering the entire coal flow conveying area. Point cloud data and RGB image data of the coal flow conveying area are acquired by the lidar and the binocular camera, respectively. The sensors include a current sensor connected in series in the motor power supply circuit of the scraper conveyor and a speed sensor installed at the end of the drive drum shaft of the scraper conveyor.

3. The intelligent speed regulation system for a scraper based on digital twin coal flow distribution and effect evaluation according to claim 2, characterized in that: When processing data, the data preprocessing module first connects the multi-source data to the data preprocessing module and performs time synchronization calibration to unify the data collected by LiDAR, binocular camera and sensor under the same time reference. Then, noise is eliminated through filtering; the data format is converted to a standardized format and dimensionality is reduced; then, based on spatial coordinate matching and timestamp association, a mapping relationship is constructed between point cloud data, RGB image data and sensor data; finally, multimodal feature extraction is performed.

4. The intelligent speed regulation system for a scraper based on digital twin coal flow distribution and effect evaluation according to claim 3, characterized in that: The time synchronization calibration in the data preprocessing module uses a network-based time protocol or a more precise time protocol to stamp the output data of all data sources with a unified high-precision timestamp. At the same time, an adaptive time synchronization mechanism for downhole noise is also set up, which uses the load fluctuation of the current sensor as an anchor point to dynamically calibrate the timestamp deviation between point cloud data and RGB image data caused by downhole vibration anomalies, and achieves millisecond-level synchronization by combining Kalman filtering.

5. The intelligent speed control system for scraper conveyors based on digital twin coal flow distribution and effect evaluation as described in claim 3, characterized in that: Multimodal feature extraction is a classification process based on different data types, including outlier removal, voxel sampling, point cloud geometric feature extraction, and volume calculation for point cloud data. Image quality enhancement, noise filtering, edge detection, texture feature calculation, and contour and foreign object detection are performed on RGB image data. The core indicators of the working condition monitoring sensor data are extracted and verified, and abnormal and invalid data are removed.

6. The intelligent speed control system for scraper conveyors based on digital twin coal flow distribution and effect evaluation as described in claim 5, characterized in that: The process of the multimodal feature fusion module is as follows: First, RGB image features are fused with point cloud geometric features. Based on the volume, curvature, and density features of the point cloud, texture features and foreign object confidence of the RGB image data are fused to construct coal flow features. A weighted fusion algorithm is used to assign weights according to the importance of different features. Then, the coal flow features are fused with operating condition sensor data to realize the correlation between coal flow status and equipment operating status. The correlation between coal flow features and equipment operating parameters is established through a feature mapping matrix to achieve dynamic matching between coal flow status and equipment load.

7. The intelligent speed control system for scraper conveyors based on digital twin coal flow distribution and effect evaluation according to claim 6, characterized in that: Based on the spatiotemporal coordinate unification of RGB image features and point cloud geometric features and the independent extraction of multi-source features, the fusion process introduces an adaptive weight function for foreign objects, using the confidence score of foreign objects generated based on the dual-dimensional detection of color threshold deviation and contour anomaly degree on the RGB image side as the gating signal. The foreign object confidence score is a quantitative indicator for underground coal conveying scenarios, with a value range of 0-1, used to accurately characterize the probability of the presence of non-coal foreign objects in the coal flow detection area.

8. The intelligent speed control system for scraper conveyors based on digital twin coal flow distribution and effect evaluation according to claim 1, characterized in that: The verification module contains a priori pattern model of the normal operating range of the equipment based on a large amount of historical data, as well as safety parameter thresholds specified by the equipment manufacturer. After receiving the 5-10 second coal flow load prediction output by the data-driven prediction module, the verification module calculates the matching degree with the prior law model and compares it with the safety parameter threshold to determine whether the predicted value conforms to historical patterns. If the predicted value is within a reasonable range and conforms to historical patterns, it is judged as passing, and the system maintains the current speed. If the predicted value exceeds the reasonable range or deviates significantly from historical patterns, it is judged as failing, and the variable frequency speed control module is triggered.

9. The intelligent speed control system for scraper conveyors based on digital twin coal flow distribution and effect evaluation as described in claim 1, characterized in that: The effectiveness evaluation module has a built-in effectiveness evaluation model, which is represented by a comprehensive evaluation function. The expression of the comprehensive evaluation function is as follows: ； Where α, β, and γ are weight coefficients set according to production needs; after each speed adjustment action, this module calculates the comprehensive evaluation score Score over a period of time. If the score is lower than the preset threshold, it is determined that the prediction and speed adjustment effect are not up to standard. The reasons are analyzed and the data-driven prediction module is updated. By adding the current data to the training set, the model of the data-driven prediction module is incrementally learned or the parameters are tuned to achieve closed-loop optimization. The three weighting coefficients, α, β, and γ, are dynamically adjusted according to production needs using an adaptive optimization algorithm.

10. A method for intelligent speed regulation of scraper conveyors based on digital twin coal flow distribution and effect evaluation, characterized in that: The intelligent speed control system for scraper conveyors based on digital twin coal flow distribution and effect evaluation as described in any one of claims 1-9 includes the following steps: Step 1: Multi-source data acquisition and preprocessing. Based on the multi-source data acquisition module, three-dimensional point cloud data, RGB image data, and multi-source data of the operating load and operating status parameters of the scraper conveyor are collected. The multi-source data is preprocessed and feature extracted and classified by the data preprocessing module to obtain multi-modal features including point cloud geometric features, RGB image texture features and operating feature of the working condition sensor. Step 2: Perform feature fusion on the multimodal features to obtain a multidimensional feature vector sequence; Step 3: Based on the multidimensional feature vector sequence, use Transformer temporal correlation to predict the coal flow load for a duration of 5-10 seconds and output preliminary speed regulation parameters; Step 4: Verify the predicted value in Step 3. If the verification is successful, control the speed of the scraper conveyor according to the predicted value, and at the same time input the predicted value into the digital twin module for synchronous simulation and monitoring. S5: If the verification fails, the variable frequency speed control module will be activated to adjust the speed of the scraper conveyor. After speed adjustment, its energy consumption, efficiency, and wear data will be evaluated to see if they meet the standards. If they meet the standards, the scraper conveyor can be controlled to operate normally. If the evaluation fails, the data-driven prediction model will be updated and the updated data-driven prediction model will be used to predict the coal flow load.

Images