Coal mine belt conveying monitoring method and system based on AI intelligent control

Abstract

The invention provides a coal mine belt conveying monitoring method and system based on AI intelligent control, and relates to the technical field of intelligent control, and the method comprises the steps that monitoring nodes are arranged at key positions of a belt conveying system, and the monitoring nodes are located at a driving roller bearing seat, a tension sensor installation point and a discharge port chute; constructing a space geometric area based on the monitoring nodes, performing grid division on the space geometric area, forming adaptive adjustment parameters according to the unit distribution density of the divided grids, and dynamically adjusting the acquisition frequency of the operation data based on the adaptive adjustment parameters; and uploading the operation data acquired after frequency regulation to an edge computing gateway, and performing real-time reasoning analysis on the operation data based on a preset AI model, namely preprocessing the data to remove invalid data, and identifying the belt operation fault type. The method can improve the dual recognition precision of the fault type and position, and guarantees the safe and efficient operation of the coal mine transportation link.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a monitoring method and system for coal mine belt conveyor based on AI intelligent control. Background Technology

[0002] In modern coal mining systems, belt conveyors serve as the core equipment of the main transportation system, and their operational stability directly determines mine production efficiency and operational safety. As belt conveyors develop towards higher speeds, longer distances, and heavier loads, traditional monitoring technologies are no longer sufficient to meet the safety management requirements under complex working conditions. Downtime caused by malfunctions not only results in direct economic losses but may also lead to equipment damage or even personal injury accidents.

[0003] Current monitoring technology for coal mine belt conveyors still faces various bottlenecks: First, the data acquisition mode is rigid. For example, existing sensors mostly use fixed-frequency acquisition, which may lead to missed hazards in critical areas of the equipment due to insufficient sampling, while causing data redundancy in stable areas. This not only fails to accurately capture early signs of faults but also increases the burden of storage and transmission. Second, the accuracy of fault identification is limited. For example, traditional AI visual inspection mostly focuses on 2D surface defects, lacks spatial correlation analysis with equipment operating parameters, and is sensitive to data gaps, making it difficult to achieve dual accurate positioning of fault type and location. The accuracy rate of identifying key faults such as longitudinal tearing is often below 85%. Third, the control strategy lacks linkage. For example, existing systems mostly adopt a single-point alarm plus unified shutdown mode, without implementing graded handling based on the fault's impact range. Unnecessary shutdowns caused by minor faults occur frequently, seriously affecting production continuity. Fourth, the model iteration capability is lacking. The monitoring system and the handling results are disconnected, making it impossible to transform fault handling experience into motivation for model optimization, leading to the recurrence of similar faults. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a coal mine belt conveyor monitoring method and system based on AI intelligent control, which can realize early warning, accurate location, graded handling and continuous optimization of belt conveyor system failures, and ensure the safe and efficient operation of coal mine transportation links.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for monitoring coal mine belt conveyor transport based on AI intelligent control, the method comprising: Monitoring nodes are set at key locations in the belt conveyor system, including the drive roller bearing housing, tension sensor installation point, and unloading chute. A spatial geometric region is constructed based on the monitoring nodes, and the spatial geometric region is divided into grids. Adaptive adjustment parameters are formed based on the unit distribution density of the grid after division, and the acquisition frequency of the operating data is dynamically adjusted based on the adaptive adjustment parameters. The operational data collected after frequency adjustment is uploaded to the edge computing gateway, and the operational data is used for real-time reasoning and analysis based on the pre-built AI model; that is, the data is pre-processed to remove invalid data and identify the types of belt operation failures. Based on the identified fault types and their corresponding spatial locations, spatial partitions centered on the fault locations are constructed by calculating the spatial influence range of each fault point. Each partition is defined as the influence area of ​​a fault point. Based on the constructed spatial partitioning results, the corresponding hierarchical control strategy is triggered; if a minor fault is identified, an operation command is automatically issued and a warning message is pushed; if a serious fault is identified, an emergency shutdown procedure is triggered, and the upstream and downstream equipment in the belt conveyor system are controlled in conjunction with the affected area to obtain execution feedback. Based on the execution feedback, obtain the control execution results; based on the control execution results, generate a fault handling report that includes the fault type, specific handling plan and system recovery time; Based on the fault handling report, the parameters of the AI ​​model are fine-tuned regularly to enable continuous iteration and optimization of the model.

[0006] Furthermore, three monitoring nodes are set at key locations in the belt conveyor system: the drive roller bearing housing, the tension sensor mounting point, and the unloading chute. A spatial geometric region is constructed based on these three monitoring nodes, and this region is then meshed. An adaptive adjustment parameter is formed based on the unit distribution density of the mesh, and the data acquisition frequency is dynamically adjusted based on this parameter. A first monitoring node is set at the drive roller bearing housing to collect bearing vibration and temperature data; a second monitoring node is set at the tension sensor installation point to collect belt tension data; and a third monitoring node is set at the discharge port chute to collect material flow and blockage status data. Based on the physical locations of the three monitoring nodes, a triangular spatial geometric region is constructed; the geometric center point of the triangular spatial geometric region is calculated, that is, a polar coordinate system is established with the geometric center point as the origin; Based on the polar coordinate system, the triangular spatial geometric region is divided into multiple sector-shaped annular grid units; The distribution density of the sector-shaped ring grid cells is statistically analyzed, and adaptive adjustment parameters are formed based on the distribution density. The data acquisition frequency of the three monitoring nodes is dynamically adjusted based on the adaptive adjustment parameters.

[0007] Furthermore, the operational data collected after frequency adjustment is uploaded to the edge computing gateway, where real-time reasoning and analysis are performed based on a pre-built AI model; that is, the data is preprocessed to remove invalid data and to identify belt operation fault types, including: Receive the frequency-adjusted operating data, and perform format unification and outlier verification on the operating data to complete the data standardization process; The standardized operational data is mapped to the spatial region corresponding to the sector ring grid cell to establish the correspondence between data and spatial location; Based on the established spatial location correspondence, missing data is filled in by spatial interpolation. Based on the data values ​​of adjacent grid cells, the estimated value of the current grid cell is obtained by distance weighting. A complete spatial dataset is formed based on the estimated value. The complete spatial dataset is input into a pre-built AI model. The spatial features in the data are extracted through the multi-layer convolutional neural network of the AI ​​model. The feature representation of key areas is enhanced according to the attention mechanism. Finally, the fault type is effectively identified and classified through a fully connected layer. The fault type identification results are output, including the fault type and its corresponding spatial location information.

[0008] Furthermore, based on the identified fault types and their corresponding spatial locations, spatial partitions centered on the fault locations are constructed by calculating the spatial influence range of each fault point. Each partition is defined as the influence area of ​​a fault point, including: Obtain the fault type and its corresponding spatial location information; Based on spatial location information, the geometric distance between each fault point and the three monitoring nodes is calculated; Based on the severity level of the fault type, the corresponding distance correction parameter is determined, and the geometric distance and the correction parameter are weighted and fused to obtain the optimized influence radius; Centered on each fault point and based on the optimized radius of influence, multiple circular influence areas are formed within the triangular spatial geometric region. Spatial Boolean operations are performed on multiple circular influence areas to obtain precise spatial partitions, with each partition defined as the influence area of ​​a fault point.

[0009] Furthermore, based on the constructed spatial partitioning results, corresponding hierarchical control strategies are triggered; if a minor fault is identified, operation instructions are automatically issued and warning information is pushed; if a serious fault is identified, an emergency shutdown procedure is triggered, and upstream and downstream equipment in the belt conveyor system are controlled in conjunction with the affected area to obtain execution feedback, including: Based on the constructed spatial partitioning results, the spatial coverage characteristics of the affected area of ​​each fault point are analyzed. Based on the fault type identification results and the spatial coverage characteristics of the affected area, the fault level is classified according to the preset decision rules to form a specific fault level judgment result. Based on the fault level determination results, when a minor fault is determined, an equipment adjustment instruction is automatically generated based on the affected area range and sent to the corresponding conveyor belt actuator. At the same time, an early warning information is pushed to the monitoring center. When the fault level is determined to be a serious fault, the emergency shutdown procedure is immediately triggered, and the upstream and downstream equipment in the affected area are controlled in conjunction with the spatial distribution characteristics of the affected area. Record the execution status of all control commands, collect device response data, and form complete execution feedback information.

[0010] Furthermore, based on the execution feedback, the control execution results are obtained; based on the control execution results, a fault handling report is generated, including the fault type, specific handling plan, and system recovery time, including: The execution feedback information is analyzed and processed to parse the control command execution status and device response data to obtain the parsed execution feedback information; Based on the parsed execution feedback information, the effectiveness of the control strategy is evaluated to obtain the results of the effectiveness assessment. Calculate the system recovery time based on the time of failure occurrence and the time of completion of control command execution; The fault type, handling plan, assessment results of handling effectiveness, and system recovery time are integrated to form a structured fault handling report; The fault handling report is stored in the database and simultaneously pushed to the monitoring center for display.

[0011] Furthermore, based on the fault handling reports, the AI ​​model undergoes regular parameter fine-tuning to enable continuous iteration and optimization, including: Extract key data from the stored fault handling reports, including fault type, handling effectiveness assessment results, and system recovery time; A model training sample set is constructed based on the extracted key data, and the training sample set is divided into a training dataset and a validation dataset. The training dataset is then normalized. The normalized training dataset is input into the AI ​​model for retraining. The model parameters are adjusted by using gradient descent to obtain the preliminary optimized model parameters. The performance of an AI model with preliminarily optimized model parameters is evaluated based on a validation dataset, and the improvement in model recognition accuracy is confirmed based on the evaluation results. The optimized model parameters, which have undergone performance verification and evaluation and met the preset accuracy requirements, are updated to the edge computing gateway to complete the iterative deployment of the model.

[0012] Secondly, a coal mine belt conveyor monitoring system based on AI intelligent control includes: The module is used to set three monitoring nodes at key locations in the belt conveyor system. The three monitoring nodes are located at the drive roller bearing housing, the tension sensor mounting point, and the unloading chute. The module is used to construct a spatial geometric region based on three monitoring nodes, divide the spatial geometric region into grids, form an adaptive adjustment parameter based on the cell distribution density of the divided grid, and dynamically adjust the acquisition frequency of the running data based on the adaptive adjustment parameter. The processing module is used to upload the collected operating data after frequency adjustment to the edge computing gateway, and perform real-time reasoning and analysis on the operating data based on the pre-built AI model; that is, to preprocess the data to remove invalid data and identify the type of belt operation failure. The definition module is used to analyze the spatial impact range of the fault point based on the identified fault type results, and to construct spatial partitions according to the fault location and monitoring nodes. Each partition is defined as the impact area of ​​a fault point. The control module is used to trigger corresponding hierarchical control strategies based on the spatial partitioning results; if a minor fault is identified, it will automatically issue operation instructions and push early warning information; if a serious fault is identified, it will trigger an emergency shutdown procedure and coordinate the control of upstream and downstream equipment in the belt conveyor system according to the affected area to obtain execution feedback. The execution module is used to obtain control execution results based on execution feedback; based on the control execution results, it generates a fault handling report that includes the fault type, specific handling plan and system recovery time; based on the fault handling report, it performs regular parameter fine-tuning of the AI ​​model to continuously iterate and optimize the model.

[0013] Thirdly, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0014] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0015] The above-described solution of the present invention has at least the following beneficial effects: This invention employs multiple technical methods to achieve comprehensive optimization: Monitoring nodes are set up at key locations in the belt conveyor system; a triangular spatial geometric region is first constructed, then a fan-shaped ring grid is divided; adaptive adjustment parameters are formed based on the grid cell distribution density to dynamically adjust the data acquisition frequency; standardized operational data is mapped to the grid to establish a correspondence between data and spatial location, while spatial interpolation fills in missing data; multi-layer convolutional neural networks are used to extract spatial features of the data, and an attention mechanism is combined to enhance the feature representation of key areas; the impact radius is calculated based on fault type and location, spatial partitioning is constructed through spatial Boolean operations, and differentiated handling is triggered according to fault level; for minor faults, equipment adjustment instructions are issued and early warning information is pushed; for severe faults, an emergency shutdown procedure is triggered and upstream and downstream equipment is linked; in addition, key data is extracted from fault handling reports to construct training samples. The data is collected, and the model parameters are adjusted using the gradient descent algorithm. After verification, the AI ​​model is updated. Through the above technical means, this invention overcomes four major problems of existing technologies in turn: First, it solves the problem of missed data collection in key areas and data redundancy in stable areas caused by rigid data collection modes; second, it solves the problem of inaccurate positioning caused by the lack of spatial correlation analysis in fault identification and sensitivity to missing data; third, it solves the problem of unnecessary downtime caused by the lack of linkage in control strategies; and fourth, it solves the problem of repeated occurrence of similar faults caused by the disconnect between model and disposal results data and the inability to iterate. In turn, it achieves multiple technical effects: the data collection accuracy in key areas is significantly improved, while redundant data is greatly reduced; the accuracy of fault identification and location positioning is significantly improved; production continuity is enhanced, and fault recovery time is effectively shortened; the AI ​​model can continuously adapt to dynamic working conditions and always maintain high recognition performance. Ultimately, it comprehensively ensures the safe and efficient operation of the coal mine belt conveyor. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a coal mine belt conveyor monitoring method based on AI intelligent control, as provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of a coal mine belt conveyor monitoring system based on AI intelligent control, provided by an embodiment of the present invention. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0019] like Figure 1As shown in the figure, an embodiment of the present invention proposes a monitoring method for coal mine belt conveyor based on AI intelligent control, the method comprising the following steps: Step 1: Set up monitoring nodes at key locations in the belt conveyor system. The monitoring nodes are located at the drive roller bearing housing, tension sensor installation point, and unloading chute. Construct a spatial geometric region based on the monitoring nodes, divide the spatial geometric region into grids, and form adaptive adjustment parameters based on the unit distribution density of the divided grid. Dynamically adjust the acquisition frequency of the operating data based on the adaptive adjustment parameters. Step 2: Upload the collected operating data after frequency adjustment to the edge computing gateway, and perform real-time reasoning and analysis on the operating data based on the pre-built AI model; that is, preprocess the data to remove invalid data and identify the type of belt operation failure. Step 3: Based on the identified fault type and its corresponding spatial location, calculate the spatial influence range of each fault point and construct a spatial partition centered on the fault location. Each partition is defined as the influence area of ​​a fault point. Step 4: Based on the constructed spatial partitioning results, trigger the corresponding hierarchical control strategy; if a minor fault is identified, automatically issue operation instructions and push early warning information; if a serious fault is identified, trigger the emergency shutdown procedure, and coordinate control of upstream and downstream equipment in the belt conveyor system according to the affected area to obtain execution feedback. Step 5: Obtain the control execution results based on the execution feedback; based on the control execution results, generate a fault handling report that includes the fault type, specific handling plan, and system recovery time; Step 6: Based on the fault handling report, perform regular parameter fine-tuning of the AI ​​model to continuously iterate and optimize the model.

[0020] In this embodiment of the invention, by setting monitoring nodes at key locations in the belt conveyor system, constructing spatial geometric regions and dividing them into grids, and dynamically adjusting the data acquisition frequency according to the grid cell distribution density, the data acquisition accuracy in key areas is improved while reducing data redundancy in stable areas. After preprocessing the frequency-adjusted operating data, it is analyzed in real time by a pre-set AI model to effectively eliminate invalid data and accurately identify fault types. The impact range is calculated based on the fault type and location, spatial partitioning is constructed, and hierarchical control is implemented. Minor faults automatically issue operation instructions and push early warnings, while serious faults trigger emergency shutdowns and link upstream and downstream equipment, significantly reducing unnecessary shutdowns and shortening fault recovery time. At the same time, the AI ​​model parameters are fine-tuned periodically in conjunction with fault handling reports to achieve continuous iterative optimization of the model to adapt to dynamic working conditions, ultimately ensuring the safe and efficient operation of the coal mine belt conveyor system.

[0021] In a preferred embodiment of the present invention, step 1 above may include: Step 1.1: Set up a first monitoring node at the drive roller bearing housing to collect bearing vibration and temperature data; set up a second monitoring node at the tension sensor installation point to collect belt tension data; set up a third monitoring node at the discharge chute to collect material flow and blockage status data. Specifically, this includes: prioritizing the installation of monitoring nodes at locations in the coal mine belt conveyor system where failures are frequent and have a critical impact on operational stability; installing a first monitoring node at the drive roller bearing housing at the belt head, which is equipped with a vibration sensor and a temperature sensor to specifically collect data on the vibration amplitude and frequency of the bearing during operation, as well as the real-time temperature of the bearing surface. The system collects real-time tension data, which directly reflects potential bearing malfunctions such as wear and lack of lubrication. A second monitoring node is installed at the tension sensor installation point in the middle of the belt. This node is linked to the belt tension sensor and continuously collects the real-time tension value during belt operation to determine whether the belt is too loose, causing slippage, or too tight, causing tearing. A third monitoring node is installed inside the discharge chute at the tail of the belt. This node is equipped with a flow sensor and an infrared detection device to collect the instantaneous flow rate and cumulative flow rate data of the material at the discharge port, as well as data on whether there is material accumulation or blockage inside the chute, to avoid belt overload caused by poor discharge.

[0022] Step 1.2: Based on the physical locations of the three monitoring nodes, a triangular spatial geometric region is constructed. The geometric center point of the triangular spatial geometric region is calculated, i.e., a polar coordinate system is established with the geometric center point as the origin. Specifically, based on the determined physical locations of the three monitoring nodes, the three-dimensional coordinate data of each monitoring node is obtained through on-site measurement. The coordinates of these three nodes are used as the three vertices of the triangle to construct a triangular spatial geometric region that can cover the key operating section of the belt conveyor. This region precisely encompasses the three core components: the drive roller, the tension detection section in the middle of the belt, and the unloading chute. Then, the coordinates of the geometric center point of this region are calculated according to the geometric characteristics of the triangle. With this geometric center point as the origin and the line connecting the origin to the three monitoring nodes as the reference direction, a polar coordinate system is established. The polar radius represents the distance from the origin to the target position, and the polar angle represents the angle between the target position and the reference direction. Through the polar coordinate system, the spatial relationship between any position within the triangular region and the three monitoring nodes can be accurately quantified.

[0023] Step 1.3, based on the polar coordinate system, divide the triangular spatial geometric region into multiple sector-shaped ring grid units. Specifically, this includes: based on the established polar coordinate system, subdividing the triangular spatial geometric region through grid division calculation; firstly, taking the origin of the polar coordinate system as the center, dividing the triangular region into multiple rings from the inside out according to a preset radius interval, such as every 0.5 meters; then, taking the line connecting the origin to the first monitoring node as the starting direction, dividing each ring evenly into multiple sector areas according to a preset angle interval, such as every 30 degrees, ultimately forming multiple sector-shaped ring grid units; each sector-shaped ring grid unit has a unique polar coordinate range identifier, which can accurately correspond to the specific spatial position of the belt conveyor system. For example, one sector-shaped ring grid unit corresponds exactly to the area directly above the drive roller bearing seat, and another grid unit corresponds to the left inner wall area of ​​the unloading port chute.

[0024] Step 1.4: Calculate the distribution density of the fan-shaped ring belt grid cells. Based on the distribution density, form adaptive adjustment parameters. Dynamically adjust the data acquisition frequency of the three monitoring nodes based on these adaptive adjustment parameters. Specifically, this includes: First, calculating the distribution density of each divided fan-shaped ring belt grid cell. The calculation method is to calculate the number of fan-shaped ring belt grid cells per unit area in different sub-regions within the triangular spatial geometric region. The sub-regions around the drive roller bearing housing and the discharge chute have denser grid cells and a higher risk of failure due to their complex equipment structure. Conversely, the stable operating sub-region in the middle of the belt, far from critical components, has relatively sparser grid cells and a lower number of cells per unit area. Based on this difference in grid distribution density, an adaptive adjustment parameter is formed. Parameters should be adjusted: higher density sub-regions correspond to larger adaptive adjustment parameter values, and lower density sub-regions correspond to smaller parameter values. These adaptive adjustment parameters are then transmitted to the control modules of the three monitoring nodes. These control modules dynamically adjust the data acquisition frequency of each node: the first monitoring node, due to its high surrounding grid density and large parameter values, has its acquisition frequency adjusted to 15-20 Hz to capture subtle vibrations and temperature changes in the bearing; the third monitoring node, corresponding to the high-density grid around the discharge chute, also has its acquisition frequency increased to 12-18 Hz to ensure timely detection of early blockages; while the second monitoring node, corresponding to the low-density grid in the stable central area of ​​the conveyor belt, has its acquisition frequency adjusted to 5-8 Hz to avoid redundant data. In this embodiment of the invention, by setting monitoring nodes at the drive roller bearing housing, tension sensor installation point, and discharge chute, core operating data such as bearing vibration and temperature, belt tension, material flow rate, and blockage status can be collected in a targeted manner, ensuring that no key operating parameters of the equipment are missed. Based on the physical locations of the three nodes, a triangular spatial geometric region is constructed, and a polar coordinate system is established with the center point of the region. This allows the collected data to be precisely bound to the specific spatial location of the equipment, providing a clear dimension for spatial correlation analysis of the data. Furthermore, based on the polar coordinate system, a fan-shaped annular grid unit is divided, and adaptive adjustment parameters are formed according to the grid distribution density to dynamically adjust the collection frequency. The sampling frequency can be increased in the key areas of the equipment with dense grids to ensure data accuracy, while the frequency can be reduced in the stable areas with sparse grids to reduce data redundancy. Ultimately, this not only improves the targeting and accuracy of the operating data collection but also optimizes the data collection efficiency, laying a high-quality data foundation for the real-time inference analysis of the AI ​​model.

[0025] In a preferred embodiment of the present invention, step 2 above may include: Step 2.1: Receive the frequency-adjusted operating data and perform format standardization and outlier verification on the operating data to complete data standardization processing. Specifically, the edge computing gateway receives frequency-adjusted operating data transmitted from three monitoring nodes. This data includes bearing vibration and temperature data from the first monitoring node, belt tension data from the second monitoring node, and material flow and blockage status data from the third monitoring node. Due to the different sensor types of different monitoring nodes, the output data formats differ. For example, vibration data is transmitted as voltage signals, and temperature data is transmitted as numerical values. Through preset format conversion rules, all data are uniformly converted into standardized JSON format data to ensure data structure consistency. Subsequently, outlier verification is performed. Based on the normal operating parameter range of the coal mine belt conveyor equipment, such as the normal bearing temperature range of 25 to 65℃ and the normal belt tension range of 15 to 30kN, each standardized data point is compared. If a data point exceeds the normal range, it is determined to be an outlier and removed, thus completing the data standardization processing.

[0026] Step 2.2 maps the standardized operational data to the spatial regions corresponding to the fan-shaped ring grid units to establish a correspondence between data and spatial location. Specifically, this includes: based on the standardized data, calling the polar coordinate information of the pre-divided fan-shaped ring grid units to establish a mapping relationship between data and grid units; for the bearing vibration and temperature data collected by the first monitoring node, the data is mapped to the adjacent fan-shaped ring grid units according to the physical polar coordinates of the node. For example, if the first monitoring node is located at polar coordinates (5m, 30°), its data is mapped to fan-shaped ring grid units with a polar diameter of 4.5 to 5.5m and a polar angle of 25° to 35°; the belt tension data of the second monitoring node and the material flow and blockage status data of the third monitoring node are also mapped to their respective adjacent fan-shaped ring grid units in the same way; through this mapping, each fan-shaped ring grid unit can be associated with the corresponding operational data, thereby establishing a direct correspondence between operational data and the specific spatial location of the equipment.

[0027] Step 2.3: Based on the established spatial location correspondence, spatial interpolation is used to fill in the missing data. The estimated value of the current grid cell is obtained by calculating distance weights based on the data values ​​of adjacent grid cells. A complete spatial dataset is then formed based on these estimated values. Specifically, after establishing the data-spatial location correspondence, some sector-shaped ring grid cells may lack associated operational data due to temporary sensor malfunctions, signal interference, or other issues, resulting in data loss. In this case, based on the established spatial location correspondence, the grid cell with missing data (hereinafter referred to as the target grid cell) and its adjacent grid cells with complete data (hereinafter referred to as the reference grid cells) are identified. Then, a spatial interpolation method is used to first measure the distance between the target grid cell and each reference grid cell. The polar coordinate distance between grid cells is used to assign a higher weight to the reference grid cell that is closer to it. Then, the data value of each reference grid cell is multiplied by its corresponding weight value, the sum is divided by the total weight value, and the estimated value of the target grid cell is obtained. For example, if the target grid cell is 0.5m away from reference grid cell A, the weight is 0.6; if it is 1m away from reference grid cell B, the weight is 0.4. If the temperature data of reference grid cell A is 50℃ and the temperature data of reference grid cell B is 48℃, then the estimated temperature of the target grid cell is (50×0.6+48×0.4)÷(0.6+0.4)=49.2℃. By filling in all the missing data in this way, a complete spatial dataset covering all the sector ring grid cells is finally formed.

[0028] Step 2.4 involves inputting the complete spatial dataset into a pre-built AI model. The AI ​​model uses a multi-layer convolutional neural network to extract spatial features from the data and strengthens the feature representation of key regions using an attention mechanism. Finally, a fully connected layer is used to effectively identify and classify fault types. Specifically, this includes: inputting the complete spatial dataset into a pre-built AI model; the AI ​​model first processes the data using a multi-layer convolutional neural network, with each convolutional layer extracting spatial features from the data. For example, the first convolutional layer extracts the basic operational features of a single sector-shaped ring grid cell, such as temperature and tension values. Subsequent convolutional layers combine the features of adjacent grid cells to extract more complex spatial correlation features, such as the spatial characteristics of a certain area. The model extracts features such as synchronized temperature increases across multiple grid cells within the domain, and the correlation between abnormal tension and flow blockage. During feature extraction, the model uses an attention mechanism to assign higher attention weights to features corresponding to the fan-shaped ring grid cells around the drive roller bearing housing and the unloading port chute, strengthening the feature representation of these key areas and allowing the model to focus more on the abnormal features of high-risk areas. Finally, the model inputs the extracted features into a fully connected layer, which classifies and identifies fault types based on preset fault classification criteria, such as determining bearing overheating faults if the bearing temperature exceeds 70°C and unloading port flow drops by 90% as unloading port blockage faults, thus accurately determining the possible fault types in the current belt conveyor system.

[0029] The process of building and training the AI ​​model is as follows: The AI ​​model is built with the core objective of accurate identification of spatially correlated faults. Its architecture is designed by combining spatial grid partitioning and multi-dimensional data acquisition. The main body of the model employs a fusion structure of multi-layer convolutional neural networks and an attention mechanism. The parameters of the convolutional layers are strictly matched to the features of the partitioned fan-shaped annular grid cells. The kernel size is set according to the polar coordinate interval of the grid cells, and the stride is consistent with the radius interval of the annular zone, ensuring accurate extraction of local operational features of individual grid cells, such as bearing vibration frequency, material flow rate changes, and spatial correlation features of adjacent grid cells, such as the coordinated feature of synchronous temperature rise in multiple grids around the drive drum. The attention mechanism module, based on identified key areas, such as the high-density grid areas around the drive drum bearing housing and the discharge chute, pre-sets weight allocation rules to automatically strengthen the feature representation of high-fault-risk areas during feature extraction, weaken the interference of redundant features in stable areas, and improve the model's sensitivity to key faults. The model's output layer employs a dual-branch structure. One branch corresponds to the pre-defined fault types of this invention, such as overheating of the drive roller bearing, blockage of the discharge port, and longitudinal tearing of the belt, outputting the probability values ​​of each fault. The other branch outputs the polar coordinate range of the fan-shaped ring grid cell where the fault is located, directly matching the fault location output requirements. Simultaneously, the model incorporates a built-in data preprocessing module, integrating format unification, outlier verification logic, and spatial interpolation calculation functions. This automatically handles format differences and missing data in the collected data, ensuring that the spatial dataset input to the model remains complete and consistent, providing stable support for training and inference.

[0030] The training process of the AI ​​model is divided into two stages: initial training and iterative optimization. The entire process relies on the closed-loop system of data acquisition, fault handling, and reporting feedback as described in this invention. The initial training stage aims to build basic recognition capabilities. Based on historical and simulated fault data from the coal mine belt conveyor system, technicians associate each fault data point with a corresponding sector-shaped ring grid cell according to mapping rules, constructing an initial sample set containing fault type, operating parameters, and spatial location. This initial sample set is divided into a training set and a validation set in a 7:3 ratio. Numerical parameters in the training set, such as temperature, tension, and flow rate, are normalized to eliminate the impact of parameter magnitude differences on training. The training set is then input into the model. Using fault type recognition accuracy and location deviation as the joint loss function, a gradient descent algorithm is used to iteratively adjust the weights of the convolutional layers and the attention mechanism until the model's fault recognition accuracy on the validation set reaches 85%, which is the initial threshold. This completes the initial training and deployment to the edge computing gateway for real-time fault inference.

[0031] The iterative optimization phase relies entirely on the feedback process of this invention to continuously improve model performance. First, key data is extracted from the fault handling report. This key data includes the complete spatial dataset at the time of the fault, which is associated with the operating parameters of the fan-shaped ring grid cells; fault type labels verified by the handling process, used to correct initial identification biases; handling effectiveness evaluation results, used to label sample credibility; and system recovery time, where "system" is an abbreviation for belt conveyor system, and the system recovery time is used to quantify the impact of fault severity on features. These real-world operating condition data are added to the initial sample set to construct a dynamically updated iterative sample set. Next, the training and validation sets are re-divided, and the newly added samples undergo the same normalization process as the initial training to ensure the continuity of data distribution. Then, the training set is input into the deployed model for retraining. At this point, the gradient descent algorithm prioritizes adjusting parameters for samples with completely effective handling, strengthening the model's learning of these high-credibility fault features. Simultaneously, for ineffective handling samples, the reasons for misjudgment are analyzed, such as missing spatial correlation features. By expanding the coverage of key areas of the attention mechanism, the model's ability to identify complex faults is improved. After retraining, the model performance is evaluated using a validation set, focusing on verifying whether the recognition accuracy of key faults, such as longitudinal belt tears, has improved and whether the location deviation has been reduced to within 0.5 meters. If the evaluation results meet the preset accuracy requirements, such as an overall recognition accuracy of ≥95%, the optimized model parameters are incrementally updated to the edge computing gateway to complete the iterative deployment. If the requirements are not met, more data on handling similar faults are added, and training is restarted. Through this iterative training, which is deeply integrated with the monitoring closed loop of this invention, the model can continuously absorb actual fault handling experience, gradually adapt to the dynamic changes in the working conditions of coal mine belt transportation, and completely solve the problem of traditional model training and application being disconnected and unable to be continuously optimized, ensuring that the fault recognition accuracy continuously improves with the system's running time.

[0032] Step 2.5: Output the fault type identification results, which include the fault type and corresponding spatial location information. Specifically, after completing the fault type identification, the AI ​​model will output the fault identification results. The output includes not only the specific fault type, such as overheating of the drive roller bearing, blockage of the unloading chute, and excessive belt tension, but also the spatial location information corresponding to the fault. The spatial location information will be combined with the established polar coordinate system to clarify the polar coordinate range of the sector ring belt grid unit where the fault is located, such as a polar diameter of 5 to 6 meters and a polar angle of 20° to 40°. At the same time, through the preset coordinate transformation rules, the polar coordinate range will be converted into a description of the actual physical location of the belt conveyor system, such as 1 to 2 meters to the left of the drive roller bearing seat or the middle area of ​​the unloading chute, so that the staff in the monitoring center can clearly and intuitively grasp the specific type and location of the fault.

[0033] In this embodiment of the invention, by standardizing the format and verifying outliers of the frequency-adjusted operating data, messy and invalid interference data can be eliminated, ensuring the consistency and reliability of the data input for analysis. Mapping standardized data to the spatial region corresponding to the fan-shaped ring grid unit establishes a direct correlation between operating data and the specific spatial location of the equipment, breaking the limitation of traditional data being disconnected from location. Based on the spatial location correspondence, missing data is filled through spatial interpolation calculation, and reasonable estimates can be obtained by relying on the distance weights of adjacent grid unit data, avoiding analysis interruption or deviation due to missing data. Inputting the complete spatial dataset into the AI ​​model, extracting spatial features with the help of multi-layer convolutional neural networks, and strengthening the expression of key areas through attention mechanisms can accurately capture fault correlation features in the data, improving the ability to identify complex faults. The output includes identification results containing fault type and corresponding spatial location information, allowing staff to clearly grasp the core information of the fault. These progressive steps not only ensure data quality and integrity but also achieve accurate fault identification and location.

[0034] In a preferred embodiment of the present invention, step 3 above may include: Step 3.1, obtaining the fault type and corresponding spatial location information, specifically includes: receiving the fault identification results output by the AI ​​model, and extracting the fault type and corresponding spatial location information from them; the fault type includes specific categories such as overheating of the drive roller bearing, excessive belt tension, and blockage of the unloading chute, etc., and the corresponding spatial location information is the polar coordinate range of the sector ring grid cell where the fault is located, such as a polar diameter of 5 to 6 m and a polar angle of 20° to 40°. This information will be temporarily stored in the gateway's temporary database to provide an initial basis for calculating the fault's impact range, and the accuracy of this information directly depends on the accuracy of the fault identification results to ensure the data continuity of the preceding and following steps.

[0035] Step 3.2: Based on spatial location information, calculate the geometric distance between each fault point and the three monitoring nodes. Specifically, this includes: retrieving the acquired spatial location information of the fault points, determining the polar coordinate center point of each fault point, and simultaneously retrieving the polar coordinate data of the three recorded monitoring nodes, where the first monitoring node corresponds to the polar coordinates of the drive roller bearing seat, the second corresponds to the polar coordinates of the tension sensor installation point, and the third corresponds to the polar coordinates of the unloading chute. Then, according to the polar coordinate distance calculation formula, calculate the straight-line distance between the center point of each fault point and the three monitoring nodes. For example, if the polar coordinates of a fault point are (5.5m, 30°) and the polar coordinates of the first monitoring node are (5m, 30°), then the geometric distance between them is 0.5m. In this way, three sets of geometric distance data between each fault point and the three monitoring nodes are obtained. These data will serve as the basic parameters for determining the radius of influence of the fault.

[0036] Step 3.3: Based on the severity level of the fault type, determine the corresponding distance correction parameter. Weightedly fuse the geometric distance and the correction parameter to obtain the optimized radius of influence. Specifically, this involves: first, pre-setting the fault severity level and corresponding distance correction parameter based on the fault severity of the coal mine belt conveyor system. For example, a minor fault, such as belt tension slightly above the normal range but not reaching a dangerous value, corresponds to a correction parameter of 1.0; a moderate fault, such as bearing temperature slightly exceeding the safety threshold, corresponds to a correction parameter of 1.2; and a severe fault, such as complete blockage of the discharge port and bearing temperature far exceeding the safety threshold, corresponds to a correction parameter of 1.5. Then, combining the obtained fault type, determine the severity level of the fault to which it belongs, thereby determining the corresponding distance correction parameter. Next, weightedly fuse the calculated geometric distance with the correction parameter, specifically by taking the average of the three geometric distances and multiplying it by the distance correction parameter to obtain the optimized radius of influence. For example, if a fault is at the severe level with a correction parameter of 1.5, and the average geometric distance to the three monitoring nodes is 1.2m, then the optimized radius of influence is 1.2 multiplied by 1.5, which equals 1.8m.

[0037] Step 3.4: Using each fault point as the center and the optimized influence radius as the benchmark, multiple circular influence areas are formed within the triangular spatial geometric region. Specifically, this includes: using the determined polar coordinate center point of the fault point as the center and the obtained optimized influence radius as the radius, drawing a circle within the constructed triangular spatial geometric region to form the circular influence area of ​​the fault point. During the drawing process, polar coordinate boundary verification is used to ensure that the circular influence area does not exceed the range of the triangular spatial geometric region. For example, when the fault point is close to the edge of the triangular region, the circle will be truncated by the triangular boundary, retaining only the part inside the triangle. For example, if a fault point is located near the edge of the unloading chute in the triangular region, and the optimized influence radius is 1.8m, the drawn circular part exceeds the triangular region. In this case, the excess part will be automatically removed, and only the fan-shaped area inside the triangle will be retained as the initial influence range of the fault point. This processing method ensures that the influence area is consistent with the actual spatial range of the belt conveyor system.

[0038] Step 3.5 involves performing spatial Boolean operations on multiple circular influence areas to obtain precise spatial partitions. Each partition is defined as the influence area of ​​a fault point. Specifically, when multiple fault points exist simultaneously in a belt conveyor system, the resulting circular influence areas may overlap. In this case, spatial Boolean operations are performed on these circular influence areas. First, the intersection of any two circular influence areas is calculated to determine whether the overlapping area belongs to the influence range of both fault points. If the area of ​​the overlapping area is more than 30% of the area of ​​a certain circular area, then the overlapping area is classified as the influence area of ​​the fault point with a greater impact. By comparing the severity of the two faults, the fault point with the higher severity level is given priority. If the overlapping area is small, the overlapping area is assigned to the influence areas of the two fault points respectively. For independent circular areas without overlap, they are directly used as the influence area of ​​the corresponding fault point. Through such spatial Boolean operations, multiple non-overlapping precise spatial partitions are finally obtained, and each partition is clearly defined as the independent influence area of ​​a fault point.

[0039] In this embodiment of the invention, acquiring fault type and corresponding spatial location information provides a core foundation for accurately analyzing the fault's impact range, avoiding the biases of traditional judgments. Calculating the geometric distance between the fault point and three monitoring nodes quantifies the spatial correlation between the fault point and key monitoring locations, providing quantifiable evidence. Combining the fault severity to determine correction parameters, and weighted fusion with geometric distances, yields an optimized impact radius, making the impact range calculation more closely aligned with complex operating conditions. A circular impact area is formed using the optimized radius, clearly defining the initial boundary of a single fault, making the range more intuitive. Spatial Boolean operations are used to process multiple circular areas, resolving issues of overlapping or omissions, ensuring that each fault's impact area is clearly defined. These progressive steps accurately quantify and clearly delineate the fault's impact range, laying the foundation for implementing targeted hierarchical control and improving upon the problem of inappropriate control strategies in traditional monitoring.

[0040] In a preferred embodiment of the present invention, step 4 above may include: Step 4.1, based on the constructed spatial partitioning results, analyze the spatial coverage characteristics of the affected area of ​​each fault point. Specifically, this includes: retrieving the precise spatial partitioning results, which clearly define the independent affected area range corresponding to each fault point; combining the actual equipment layout diagram of the coal mine belt conveyor system, dissecting the spatial coverage characteristics of each affected area one by one: first, confirming the equipment components covered by the affected area, such as whether it only covers the middle section of the belt body, or also includes key components such as the drive roller bearing housing and the unloading chute; then, calculating the actual area of ​​the affected area, for example, the affected area of ​​a certain fault only covers 8 square meters of the non-core section of the belt, while the affected area of ​​another fault covers 22 square meters and includes the area around the tension sensor installation point; finally, determining whether the affected area involves the critical links of equipment operation, such as whether it covers the power transmission section of the belt, the material transfer connection section, etc.; through this analysis, the actual impact range and potential harm of each fault on the belt conveyor system can be clearly understood.

[0041] Step 4.2: Based on the fault type identification results and the spatial coverage characteristics of the affected area, fault level classification is performed using preset decision rules to form specific fault level judgment results. This includes: first, obtaining the fault type identification results, such as whether it is a slightly high temperature of the drive drum bearing, minor blockage of the unloading chute, or longitudinal tear of the belt; then, combining the spatial coverage characteristics of the affected area, fault level classification is carried out according to the preset decision rules. The preset decision rules are formulated in advance based on the actual coal mine production and the severity of the fault. For example, the rule specifies: if the fault type is belt tension exceeding the normal range by less than 5% and the affected area only covers non-critical sections of the belt and is less than 12 square meters, it is judged as a minor fault; if the fault type is a drive drum bearing... If the bearing temperature exceeds the safety threshold by less than 8°C and the affected area covers the bearing housing and the surrounding area is between 12 and 25 square meters, it is judged as a moderate fault. If the fault type is a longitudinal tear of the belt, a complete blockage of the discharge port chute, or a bearing temperature exceeding the safety threshold by more than 12°C and the affected area covers key components and the area exceeds 25 square meters, it is judged as a severe fault. The specific fault level judgment result is generated by substituting the type and spatial coverage characteristics of each fault into the rule and comparing them one by one. For example, a minor blockage of the discharge port chute, with an affected area of ​​9 square meters and not involving key components, is judged as a minor fault. If the drive roller bearing temperature exceeds the safety threshold by 10°C and the affected area covers the bearing housing and the surrounding area of ​​18 square meters, it is judged as a moderate fault.

[0042] Step 4.3: Based on the fault level determination result, when a minor fault is determined, an equipment adjustment command is automatically generated based on the affected area range and sent to the corresponding conveyor belt actuator. Simultaneously, an early warning information is pushed to the monitoring center. Specifically, when a fault is determined to be minor, the affected area range is automatically analyzed, and a targeted equipment adjustment command is generated. For example, if the fault is that the tension in the middle section of the belt exceeds the normal range by 4% and the affected area covers a 30-meter section of the belt, a command is generated to adjust the tension in the middle section of the belt to the lower limit of the normal range, such as 16kN, with an adjustment rate controlled at 0.5kN per minute to avoid a sudden drop in tension causing belt slippage. The fault was a minor blockage in the unloading chute, affecting an area only 5 meters from the chute inlet. This triggered an instruction to activate the side vibration unblocking device of the unloading chute, adjusting the vibration frequency to 25Hz and running it continuously for 5 minutes. After the instruction was generated, the adjustment instructions were sent to the corresponding conveyor belt actuators, such as the tension regulating motor and the controller of the chute vibration device. At the same time, an early warning message was automatically generated, including the fault type, the specific physical location of the affected area (e.g., the section from K2+150 to K2+180 in the middle of the belt), the content of the issued adjustment instructions, and the suggested time for manual review. The early warning message was pushed through two channels: the large display screen in the monitoring center and the mobile terminals of the staff.

[0043] Step 4.4: When the fault level is determined to be a severe fault, an emergency shutdown procedure is immediately triggered. Based on the spatial distribution characteristics of the affected area, upstream and downstream equipment within the affected area are controlled in a coordinated manner. Specifically, when a fault is determined to be a severe fault, an emergency shutdown procedure is immediately triggered, sending a shutdown signal to the control cabinet of the main drive motor of the belt to ensure that the main drive motor cuts off power and stops operating within 2 seconds to prevent the fault from escalating further. Simultaneously, the spatial distribution characteristics of the affected area are retrieved and analyzed to determine the affected area and related upstream and downstream equipment; for example, if the fault is a longitudinal tear of the belt and the affected area covers the section from K3+200 to K3+250 of the belt. The system connects upstream to Coal Feeder No. 2 and downstream to Transfer Conveyor No. 3. It sends an instruction to Coal Feeder No. 2 to immediately stop feeding and close its feed gate to prevent material accumulation at the tear. It also sends an instruction to Transfer Conveyor No. 3 to activate its coal cleaning mode, increasing the conveyor belt speed to 1.2 m / s to clear any remaining material. Finally, it sends an instruction to the belt brake devices on both sides of the tear area to activate mechanical brakes and lock the conveyor belts to prevent slippage due to material gravity after the belts stop. Through this coordinated control of upstream and downstream equipment, the system quickly cuts off the material and power transmission links of the fault, reducing secondary damage to surrounding equipment.

[0044] Step 4.5: Record the execution status of all control commands, collect equipment response data, and form complete execution feedback information. Specifically, this includes: after issuing adjustment commands or triggering emergency stop and linkage control, real-time data recording begins, recording the execution status of all control commands throughout the process: including the precise time of command issuance, the command reception status of the actuators (e.g., the tension regulating motor controller has received the command, the time the command started execution, parameter changes during command execution (e.g., the real-time curve of belt tension decreasing from 18kN to 16kN), and the time the command execution was completed); simultaneously, collect equipment response data through sensors distributed in each actuator, such as the actual tension value of the belt after tension adjustment, the change in material flow at the discharge port after the unblocking device operates, the actual shutdown time of the main drive motor after emergency stop, and the completion status of upstream and downstream equipment actions (e.g., the feed gate of No. 2 coal feeder is completely closed, and the material in No. 3 transfer machine has been emptied); integrate these execution status data and equipment response data according to the structure of fault number, command content, execution process, and response result to form complete execution feedback information, which is then stored in the historical database.

[0045] In this embodiment of the invention, the spatial coverage characteristics of the affected area of ​​each fault point are analyzed based on the spatial partitioning results, which can clearly grasp the specific equipment range affected by the fault and provide accurate spatial basis for differentiated control. Combining the fault type and the coverage characteristics of the affected area, the fault level is divided by preset decision rules, avoiding the drawback of using a uniform handling method for all faults regardless of their severity in traditional monitoring. For minor faults, equipment adjustment instructions are automatically generated and early warnings are pushed, which not only realizes timely intervention in faults but also avoids unnecessary downtime and effectively ensures production continuity. For serious faults, shutdown is immediately triggered and upstream and downstream equipment in the affected area are controlled in conjunction, which can quickly cut off the fault propagation path, prevent the accident from escalating, and improve operational safety. Recording the execution status of control instructions and equipment response data to form feedback information not only allows for the traceability of fault handling effects but also provides real data support for the iterative optimization of AI models, thus realizing precise, coordinated, and closed-loop management of fault handling as a whole.

[0046] In a preferred embodiment of the present invention, step 5 above may include: Step 5.1 involves analyzing and processing the execution feedback information to parse the control command execution status and equipment response data, resulting in parsed execution feedback information. Specifically, this includes: retrieving complete execution feedback information, which contains the control command execution status and equipment response data; processing this information by: first extracting key content of the control command execution status, such as command issuance time, whether the actuator successfully received the command, the start and completion times of the command, and whether any interruptions occurred during execution; then disassembling the equipment response data, such as the actual change curve of belt tension after minor fault adjustments, the recovery data of discharge port flow after the unblocking device operates, or the downtime of the main drive motor after a serious fault shutdown, and the completion status of upstream and downstream equipment actions; and finally organizing these extracted contents according to the logic of fault number, command type, execution status, and equipment response, removing redundant raw data codes, and transforming them into structured information that staff can intuitively understand, thus obtaining the parsed execution feedback information.

[0047] Step 5.2: Based on the parsed execution feedback information, evaluate the implementation effect of the control strategy to obtain the effectiveness assessment result. Specifically, this includes: evaluating the implementation effect of the control strategy from multiple dimensions based on the parsed execution feedback information; for minor faults, mainly judging whether the equipment adjustment command restores the abnormal parameters to normal, such as whether the belt tension has dropped from the out-of-range value to the standard range, whether the unloading port blockage has been cleared by the unblocking device and the flow has returned to stability, and checking whether it has caused new equipment abnormalities; for serious faults, focusing on evaluating whether the emergency shutdown timely prevents the fault from expanding, and whether the linkage between upstream and downstream equipment effectively reduces secondary problems, such as whether a large amount of material is avoided after the belt tears, and whether related equipment has not suffered additional damage due to the fault; classifying the effectiveness level of the handling according to the assessment results, such as completely effective (the fault is resolved and there are no secondary problems), basically effective (the fault is alleviated but subsequent fine-tuning is required), and ineffective (the fault is not resolved or has caused new problems), and finally forming the effectiveness assessment result of the handling.

[0048] Step 5.3: Calculate the system recovery time based on the fault occurrence time and the control command execution completion time. This includes: first, determining two key time points: the fault occurrence time, which comes from the time the AI ​​model first identifies and records the fault; and the control command execution completion time, which is extracted from the parsed execution feedback information, i.e., the time when all control commands for the fault, such as adjustment commands, shutdown commands, and linkage commands, are executed and the equipment parameters stabilize. The system recovery time is obtained by calculating the difference between the two time points, for example, the time difference mentioned above is 13 minutes and 30 seconds. If it is a serious fault, the system recovery time will also include the safety check time before restarting the equipment after fault handling, ensuring that the calculation results can truly reflect the entire process from the fault occurrence to the system being able to operate normally.

[0049] Step 5.4 integrates the fault type, handling plan, handling effectiveness assessment results, and system recovery time to form a structured fault handling report. Specifically, this includes: integrating key information from each stage to form a structured fault handling report; the report first clearly identifies the fault type, derived from fault identification results, such as an overheating fault in the drive drum bearing; then it details the handling plan, including specific control commands implemented, such as issuing a command to start the bearing cooling device and coordinating with the upstream coal feeder to reduce load; next, it enters the handling effectiveness assessment results, such as "completely effective," indicating the bearing temperature drops to the normal range within 30 minutes; finally, it records the calculated system recovery time, such as a total recovery time of 45 minutes; the report also supplements the specific spatial location of the fault and the key parameter change curves during the handling process. The overall report uses a structured format divided into chapters to ensure clear and logical information flow, facilitating quick access for staff to review key information throughout the fault handling process, thus forming a structured fault handling report.

[0050] Step 5.5 involves storing the fault handling report in the database and simultaneously pushing it to the monitoring center for display. This includes: In this embodiment of the invention, parsing and processing the execution feedback information transforms fragmented control command execution status and equipment response data into clear, structured information, providing an accurate data foundation for evaluating the effectiveness of handling. Evaluating the implementation effect of control strategies based on the parsed information allows for a direct assessment of whether handling measures have effectively resolved the fault, compensating for the lack of effective assessment in traditional monitoring. Calculating the system recovery time quantifies the efficiency of fault handling, providing a time-based reference for optimizing the handling process. Integrating key information such as fault type and handling plan to form a structured report ensures clear and organized information throughout the fault handling process, facilitating quick understanding of core content by staff. Storing the report in the database and pushing it to the monitoring center achieves long-term retention of fault handling data, accumulating real data for AI model iteration, and allowing monitoring personnel to promptly understand the handling results. These progressively advancing steps construct a complete record and feedback loop for fault handling, further enhancing the standardization and intelligence of the coal mine belt conveyor monitoring system.

[0051] In a preferred embodiment of the present invention, step 6 above may include: Step 6.1: Extract key data from the stored fault handling reports. This key data includes the fault type, the effectiveness assessment results of the handling, and the system recovery time. Specifically, this includes: fault handling reports stored periodically in the database; key data is automatically extracted using a preset data extraction script; the extracted content includes the fault type recorded in the report, such as overheating of the drive roller bearing or blockage of the discharge chute; the effectiveness assessment results of the handling, i.e., whether it is completely effective, basically effective, or ineffective; and the system recovery time, i.e., the total time from the occurrence of the fault to the system returning to normal operation. During the extraction process, the data undergoes preliminary verification, such as checking whether the fault type matches the assessment results and whether there are obvious logical errors in the recovery time, to ensure that the extracted key data is authentic and reliable, providing core materials derived from actual handling scenarios for model training.

[0052] Step 6.2: Construct a model training sample set based on the extracted key data, and divide the training sample set into a training dataset and a validation dataset. Normalize the training dataset, specifically including: constructing the model training sample set based on the logic of fault characteristics and handling feedback according to the extracted key data; each sample contains the operating parameter features at the time of the fault occurrence, where the operating parameter features are associated with the complete spatial dataset, fault type label, handling effectiveness label, and recovery time feature; then, randomly divide the sample set into a training dataset and a validation dataset in a 7:3 ratio. The training dataset is used for model parameter adjustment, and the validation dataset is used for performance evaluation; for numerical features in the training dataset, such as recovery time and vibration amplitude, normalize them using the maximum and minimum value standardization method to compress the data to the range of 0 to 1, avoiding the impact of excessive differences in the numerical range of different features on the model training effect, and ensuring that the data input to the model has consistency and comparability.

[0053] Step 6.3 involves inputting the normalized training dataset into the AI ​​model for retraining. Gradient descent is used to adjust the model parameters to obtain initially optimized model parameters. Specifically, this includes: inputting the normalized training dataset into the currently deployed AI model, i.e., the multi-layer convolutional neural network model used for fault identification, for retraining; during training, the model optimizes for fault type identification accuracy by continuously adjusting the weight parameters of each network layer using the gradient descent algorithm; in each iteration, the error between the model's prediction and the actual fault type is calculated, and parameters are adjusted according to the error direction to gradually reduce the prediction error; for example, when the model's accuracy in identifying longitudinal belt tear faults is low, gradient descent will focus on adjusting the weights of convolutional layers related to this fault feature to enhance the model's sensitivity to tear features; after a preset number of iterations, such as 50 rounds, when the model's prediction error stabilizes, training is stopped, and initially optimized model parameters are obtained.

[0054] Step 6.4: Evaluate the performance of the AI ​​model with preliminarily optimized model parameters based on the validation dataset. Confirm the improvement in model recognition accuracy based on the evaluation results. Specifically, this includes: inputting the partitioned validation dataset into the preliminarily optimized AI model to evaluate the model's actual performance; evaluation metrics include fault recognition accuracy, recall rates for different fault types, such as the recall rate for longitudinal tear faults, and prediction speed; comparing these metrics with the model performance before optimization to confirm whether the model has met the preset improvement standards, such as requiring an overall recognition accuracy improvement of more than 5%, and an improvement in the recognition accuracy of key faults, such as longitudinal tear, from 85% to more than 90%; if the evaluation results do not meet the standards, return to adjust the learning rate of gradient descent or increase the number of iterations for retraining; if the standards are met, generate the final model performance evaluation report.

[0055] Step 6.5 updates the optimized model parameters, which have undergone performance verification and evaluation and met the preset accuracy requirements, to the edge computing gateway to complete the iterative deployment of the model. Specifically, this includes: when the evaluation results show that the optimized model meets the preset accuracy requirements, such as an overall recognition accuracy ≥ 92%, initiating the model update process; sending the verified optimized model parameters to the edge computing gateway via an encrypted data transmission channel to replace the old model parameters in the gateway; during the parameter update process, pausing the gateway's real-time inference task to ensure the integrity of the parameter replacement; automatically restarting the inference service after the update is complete and conducting small-scale data testing to verify whether the new parameters are working properly; simultaneously, recording the model update time, version number, and performance improvement data, and storing them in the model iteration log for future traceability; in this way, the optimized AI model can be quickly applied to actual monitoring scenarios, solving the problem of data disconnect between the model and the handling results in traditional monitoring, and the inability to iterate, enabling the model to continuously adapt to the dynamic working conditions of coal mine belt conveyors.

[0056] In this embodiment of the invention, key data such as fault type, effectiveness assessment results of fault handling, and system recovery time are extracted from fault handling reports to provide real-world case data for AI model iteration, addressing the lack of effective data support for traditional monitoring model optimization. Based on this key data, a training sample set is constructed, training and validation datasets are divided, and the training data is normalized, laying a high-quality data foundation for model retraining. The normalized training data is input into the AI ​​model for retraining, and parameters are adjusted through gradient descent to achieve targeted model optimization, better adapting it to actual fault characteristics. The performance of the optimized model is evaluated using the validation dataset to ensure a substantial improvement in recognition accuracy and avoid substandard accuracy. The qualified parameters are updated to the edge computing gateway for deployment, allowing the optimized model to be quickly implemented. These steps form a complete iterative closed loop of data extraction, sample construction, model training, performance verification, and deployment, effectively solving the shortcomings of traditional monitoring models being disconnected from handling results data and unable to continuously iterate. This allows the AI ​​model to continuously adapt to the dynamic working conditions of coal mines as fault handling experience accumulates, maintaining high fault identification accuracy and providing continuous intelligent support for the long-term reliable operation of belt conveyor monitoring.

[0057] like Figure 2 As shown, embodiments of the present invention also provide a coal mine belt conveyor monitoring system based on AI intelligent control, comprising: The module is used to set three monitoring nodes at key locations in the belt conveyor system. The three monitoring nodes are located at the drive roller bearing housing, the tension sensor mounting point, and the unloading chute. The module is used to construct a spatial geometric region based on three monitoring nodes, divide the spatial geometric region into grids, form an adaptive adjustment parameter based on the cell distribution density of the divided grid, and dynamically adjust the acquisition frequency of the running data based on the adaptive adjustment parameter. The processing module is used to upload the collected operating data after frequency adjustment to the edge computing gateway, and perform real-time reasoning and analysis on the operating data based on the pre-built AI model; that is, to preprocess the data to remove invalid data and identify the type of belt operation failure. The definition module is used to analyze the spatial impact range of the fault point based on the identified fault type results, and to construct spatial partitions according to the fault location and monitoring nodes. Each partition is defined as the impact area of ​​a fault point. The control module is used to trigger corresponding hierarchical control strategies based on the spatial partitioning results; if a minor fault is identified, it will automatically issue operation instructions and push early warning information; if a serious fault is identified, it will trigger an emergency shutdown procedure and coordinate the control of upstream and downstream equipment in the belt conveyor system according to the affected area to obtain execution feedback. The execution module is used to obtain control execution results based on execution feedback; based on the control execution results, it generates a fault handling report that includes the fault type, specific handling plan and system recovery time; based on the fault handling report, it performs regular parameter fine-tuning of the AI ​​model to continuously iterate and optimize the model.

[0058] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An AI intelligent control-based coal mine belt transportation monitoring method, characterized in that, The method comprises: Monitoring nodes are arranged at key positions of the belt conveying system, and the monitoring nodes are located at a driving drum bearing seat, a tension sensor mounting point, and a discharge port chute; a spatial geometric region is constructed based on the monitoring nodes, the spatial geometric region is meshed, an adaptive adjustment parameter is formed according to the unit distribution density of the meshed grid, and the collection frequency of operation data is dynamically adjusted based on the adaptive adjustment parameter; The operation data collected after frequency adjustment is uploaded to an edge computing gateway, real-time inference analysis is performed on the operation data based on a preset AI model, that is, invalid data is removed through preprocessing of the data, and a belt operation fault type is identified; Based on the identified fault type and corresponding spatial position, the spatial influence range of each fault point is calculated, a spatial partition centered on the fault position is constructed, and each partition is defined as the influence area of a fault point; Based on the constructed spatial partition result, a corresponding hierarchical control strategy is triggered; if a slight fault is identified, an operation instruction is automatically issued and warning information is pushed; if a serious fault is identified, an emergency shutdown program is triggered, and the upstream and downstream equipment in the belt conveying system is linked and controlled according to the influence area to obtain an execution feedback; According to the execution feedback, a control execution result is obtained; based on the control execution result, a fault treatment report containing the fault type, the specific treatment scheme, and the system recovery time is formed; Based on the fault treatment report, the AI model is periodically fine-tuned to continuously iterate and optimize the model.

2. The coal mine belt transportation monitoring method based on AI intelligent control according to claim 1, characterized in that, Monitoring nodes are arranged at key positions of the belt conveying system, and the monitoring nodes are located at a driving drum bearing seat, a tension sensor mounting point, and a discharge port chute; a spatial geometric region is constructed based on the monitoring nodes, the spatial geometric region is meshed, an adaptive adjustment parameter is formed according to the unit distribution density of the meshed grid, and the collection frequency of operation data is dynamically adjusted based on the adjustment parameter, including: A first monitoring node is arranged at the driving drum bearing seat to collect bearing vibration and temperature data; a second monitoring node is arranged at the tension sensor mounting point to collect belt tension data; and a third monitoring node is arranged at the discharge port chute to collect material flow and blockage state data; Based on the physical positions of the three monitoring nodes, a triangular spatial geometric region is constructed; the geometric center point of the triangular spatial geometric region is calculated, that is, a polar coordinate system is established with the geometric center point as the origin; Based on the polar coordinate system, the triangular spatial geometric region is divided into a plurality of fan-shaped ring belt grid units; The distribution density of the fan-shaped ring belt grid units is counted, an adaptive adjustment parameter is formed according to the distribution density, and the data collection frequency of the three monitoring nodes is dynamically adjusted based on the adaptive adjustment parameter.

3. The coal mine belt transportation monitoring method based on AI intelligent control according to claim 2, characterized in that, The operation data collected after frequency adjustment is uploaded to an edge computing gateway, and real-time inference and analysis are performed on the operation data based on a preset AI model; That is, invalid data is removed through preprocessing of the data, and a belt operation fault type is identified, including: The frequency-adjusted operation data is received, and the operation data is uniformly formatted and subjected to abnormal value checking to complete data standardization processing; Map the running data after standardization to the spatial area corresponding to the sector ring belt grid unit to establish the correspondence between data and spatial position; Based on the established spatial position correspondence, the missing data is filled by using spatial interpolation calculation, the estimated value of the current grid unit is obtained through distance weight calculation based on the data value of the adjacent grid unit, and the complete spatial data set is formed based on the estimated value; Input the complete spatial data set into the preset AI model, extract the spatial features in the data through the multi-layer convolutional neural network of the AI model, and strengthen the feature expression of the key area according to the attention mechanism, and finally effectively identify and classify the fault type through the full connection layer; Output the fault type identification result, which includes the fault type and the corresponding spatial position information.

4. The coal mine belt transportation monitoring method based on AI intelligent control according to claim 3, characterized in that, Based on the identified fault type and the corresponding spatial position, the spatial influence range of each fault point is calculated, and a spatial partition centered on the fault position is constructed, each partition is defined as the influence area of a fault point, including: Obtain the fault type and the corresponding spatial position information; Based on the spatial position information, calculate the geometric distance between each fault point and the three monitoring nodes; According to the severity level of the fault type, determine the corresponding distance correction parameter, and weight and fuse the geometric distance and the correction parameter to obtain the optimized influence radius; Take each fault point as the center and the optimized influence radius as the reference to form multiple circular influence areas in the triangular spatial geometric area; Perform spatial Boolean operation on the multiple circular influence areas to obtain the accurate spatial partition, and each partition is defined as the influence area of a fault point.

5. The coal mine belt transportation monitoring method based on AI intelligent control according to claim 4, characterized in that, Based on the constructed spatial partition result, trigger the corresponding hierarchical control strategy; if it is identified as a minor fault, automatically issue operation instructions and push warning information; If it is identified as a serious fault, trigger the emergency shutdown program, and control the upstream and downstream equipment in the belt transportation system according to the influence area to obtain the execution feedback, including: Based on the constructed spatial partition result, analyze the spatial coverage characteristics of the influence area of each fault point; Based on the fault type identification result and the spatial coverage characteristics of the influence area, classify the fault level through the preset decision rule to form the specific fault level judgment result; According to the fault level judgment result, when it is judged as a minor fault, automatically form the equipment adjustment instruction based on the influence area range, and issue the adjustment instruction to the corresponding transportation belt execution mechanism, and push the warning information to the monitoring center; When the fault level judgment result is a serious fault, immediately trigger the emergency shutdown program, and control the upstream and downstream equipment in the influence area according to the spatial distribution characteristics of the influence area; Record the execution state of all control instructions, collect device response data, and form complete execution feedback information.

6. The coal mine belt transportation monitoring method based on AI intelligent control according to claim 5, characterized in that, According to the execution feedback, obtain the control execution result; Based on the control execution result, form a fault handling report containing the fault type, specific handling scheme and system recovery time, including: Analyze and process the execution feedback information to analyze the control instruction execution state and device response data in it to obtain the analyzed execution feedback information; Based on the parsed execution feedback information, the implementation effect of the control strategy is evaluated, and a treatment effectiveness evaluation result is obtained; According to the fault occurrence time and the control instruction execution completion time, the system recovery time is calculated; The fault type, treatment scheme, treatment effectiveness evaluation result and system recovery time are integrated to form a structured fault treatment report; The fault treatment report is stored in the database and pushed to the monitoring center for display.

7. The coal mine belt transportation monitoring method based on AI intelligent control according to claim 6, characterized in that, Based on the fault treatment report, the AI model is periodically fine-tuned to continuously iterate and optimize the model, including: Extract key data from the stored fault treatment report, including fault type, treatment effectiveness evaluation result and system recovery time; Based on the extracted key data, a model training sample set is constructed, and the training sample set is divided into a training data set and a validation data set. The training data set is normalized; The normalized training data set is input into the AI model for retraining. The model parameters are adjusted by gradient descent calculation to obtain preliminary optimized model parameters; Based on the validation data set, the performance of the AI model with preliminary optimized model parameters is evaluated, and the improvement effect of model recognition accuracy is confirmed according to the evaluation result; The optimized model parameters that pass the performance verification evaluation and meet the preset precision requirements are updated to the edge computing gateway to complete the iterative deployment of the model.

8. An AI intelligent control-based coal mine belt transportation monitoring system, which implements the method of any one of claims 1 to 7, characterized in that, It includes: A setting module is used to set three monitoring nodes at key positions of the belt transportation system. The three monitoring nodes are located at the drive roller bearing seat, the tension sensor mounting point and the discharge port chute. A construction module is used to construct a spatial geometric region based on the three monitoring nodes. The spatial geometric region is divided into grids, and a self-adaptive adjustment parameter is formed according to the unit distribution density of the divided grids. The collection frequency of the running data is dynamically adjusted based on the self-adaptive adjustment parameter. A processing module is used to upload the running data collected after frequency adjustment to the edge computing gateway, and perform real-time inference and analysis on the running data based on the pre-set AI model. That is, the data is preprocessed to eliminate invalid data and identify the belt running fault type. A definition module is used to analyze the spatial influence range of the fault point based on the identified fault type result, and to construct spatial partitions according to the fault location and monitoring nodes. Each partition is defined as an influence area of a fault point. A control module is used to trigger the corresponding hierarchical control strategy based on the spatial partition result. If it is identified as a minor fault, an operation instruction is automatically issued and a warning information is pushed. If it is identified as a serious fault, an emergency shutdown program is triggered, and the upstream and downstream equipment in the belt transportation system is controlled according to the influence area to obtain an execution feedback. An execution module is used to obtain a control execution result according to the execution feedback. Based on the control execution result, a fault treatment report containing fault type, specific treatment scheme and system recovery time is formed. Based on the fault treatment report, the AI model is periodically fine-tuned to continuously iterate and optimize the model.

9. A computing device, comprising: It includes: One or more processors; A storage device for storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a program, which when executed by a processor, implements the method of any one of claims 1-7.

Images