A closed-loop control method for real-time mapping and autonomous decision-making in coal mining faces.

By implementing millisecond-level clock synchronization and graph optimization model through hardware triggering, the problems of spatiotemporal registration and cumulative drift error of multi-source sensors in coal mining faces are solved. Semantic point cloud maps are generated and cutting curves are fitted, realizing closed-loop control of real-time 3D mapping and autonomous decision-making for planning and cutting of coal mining faces, thus promoting intelligent unmanned mining.

CN123064477APending Publication Date: 2026-11-10PEKING UNIV +3
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611587895.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-14
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies for coal mining faces suffer from problems such as difficulties in spatiotemporal registration of multi-source sensors, insufficient suppression of cumulative drift errors, difficulties in closed-loop decision-making for mapping feature extraction and truncation, and lack of semantic information. These issues result in low accuracy and poor real-time performance of 3D mapping, making it difficult to achieve intelligent and unmanned mining.

Method used

Achieving millisecond-level clock hard synchronization through hardware triggering establishes a unified timestamp and unified spatiotemporal benchmark for multi-source data. A graph optimization model is constructed for joint optimization and local feature relocation, generating a semantic point cloud map carrying mining environment and equipment labels. The coordinates of the top and bottom plate boundaries are automatically extracted and the truncation curve is fitted to form a fully closed-loop control.

Benefits of technology

It has achieved autonomous decision-making closed-loop control for real-time 3D mapping and planning of coal mining faces, breaking through the limitations of spatiotemporal registration, error suppression and semantic mapping, promoting intelligent unmanned mining and reducing human intervention.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123064477A_ABST
    Figure CN123064477A_ABST
Patent Text Reader

Abstract

This application provides a closed-loop control method for real-time mapping and autonomous decision-making in coal mining faces, relating to the field of intelligent mining technology. It includes: achieving millisecond-level clock hard synchronization through hardware triggering to complete spatiotemporal registration of multi-source data; constructing a graph optimization model with keyframe poses as vertices and inter-frame and loop closure constraints as edges, jointly optimizing and outputting a high-precision continuous 3D map of the coal mining area; inputting the 3D map, coal mining process parameters, geological model, and equipment operating parameters into a semantic segmentation network to generate a semantic point cloud map carrying mining environment and equipment labels; automatically extracting the roof and floor boundary coordinate sequences and fitting the current cutter cutting curve to generate the next cutter safety cutting boundary and the coal mining machine height adjustment control curve, and issuing an execution order for autonomous cutting; acquiring new multi-source data during cutting and returning it for registration, forming a complete closed-loop control. This application overcomes the limitations of spatiotemporal registration, error suppression, feature extraction, and autonomous cutting closed-loop control in real-time mapping of coal mining faces.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of intelligent and unmanned mining technology, specifically to a closed-loop control method for real-time mapping and autonomous decision-making in coal mining face planning and cutting. Background Technology

[0002] Coal is a crucial component of my country's energy security. Currently, there is a clear requirement to promote intelligent upgrading and transformation, strengthening adaptive cutting and high-order intelligent mining of coal mining equipment, and achieving breakthroughs in technologies such as high-precision perception, precise control, and intelligent decision-making. Intelligent decision-making and adaptive cutting of coal mining machines rely on their high-precision perception of the spatial position and geometry of the roof and floor of the working face. However, extreme working conditions such as high dust levels, low illumination, and strong vibrations at the working face result in low accuracy, poor real-time performance, and difficulties in feature identification and extraction in three-dimensional spatial mapping, severely restricting intelligent and unmanned mining capabilities. Existing technologies mainly suffer from the following shortcomings: (1) Difficulty in spatiotemporal registration of multi-source sensors. Sensors such as lidar, vision cameras, and inertial measurement units have different sampling frequencies, time bases, and spatial coordinate systems, making it difficult to achieve precise clock synchronization and coordinate unification in dynamic mining scenarios. Existing methods mostly use software time synchronization or coarse alignment, which has insufficient synchronization accuracy, resulting in spatiotemporal misalignment of point cloud and image data during mapping, directly affecting the mapping accuracy.

[0003] (2) Insufficient suppression of cumulative drift error. Due to vibration interference caused by coal cutting and frame shifting operations at the coal face, as well as the cumulative drift caused by long-term pose recursion, the accuracy of SLAM mapping decreases significantly with the increase of working distance. Although existing SLAM algorithms based on laser-inertial fusion (such as FAST-LIO2, LIO-SAM, etc.) have improved positioning accuracy to some extent, they are still difficult to effectively suppress cumulative error in long-distance underground operation scenarios where a global absolute reference is lacking. Some solutions use three-dimensional laser scanning stitching methods for post-processing correction, but they cannot meet the needs of real-time mapping and collaborative control.

[0004] (3) Difficulties in feature extraction and cut-off decision-making. Although some studies have explored adaptive cut-off technology based on coal and rock identification and planning cut-off system based on transparent geology, the feature extraction methods of the above methods mainly rely on pure vision, air-coupled radar, equipment operating parameters and other technologies. They do not integrate spatiotemporally registered three-dimensional laser scanning point cloud data, making it difficult to achieve automatic identification and accurate extraction of roof and floor boundary coordinate sequences based on deep learning. As a result, the existing coal mining machine cut-off control mostly adopts memory cut-off or manual preset trajectory methods, making it difficult to adjust the cut-off parameters in real time according to the dynamic changes of the roof and floor of the working face.

[0005] (4) Lack of semantic information leads to insufficient environmental understanding. Existing 3D mapping methods for working faces mostly focus on offline reconstruction of geometric information, lacking semantic recognition and annotation of operational elements such as the exposed roof and floor, geological information, hydraulic supports, coal mining machines, and scraper conveyors. Autonomous cutting by coal mining machines requires accurate identification of semantic information such as the roof and floor boundaries and equipment locations in the mining environment. Relying solely on geometric maps is insufficient to meet the needs of autonomous decision-making and planning.

[0006] In summary, there is an urgent need for a closed-loop control method that can realize real-time 3D mapping and autonomous decision-making for planning and cutting of coal mining faces, in order to overcome the limitations of existing technologies in spatiotemporal registration, error suppression, semantic mapping and autonomous decision-making, and provide technical support for intelligent and unmanned mining of coal mining faces. Summary of the Invention

[0007] In view of the above problems, this application proposes a closed-loop control method for real-time mapping and autonomous decision-making of coal mining face to overcome the shortcomings of the prior art.

[0008] This application provides a closed-loop control method for real-time mapping and autonomous decision-making in coal mining face planning and cutting, including: Millisecond-level clock hard synchronization is achieved through hardware triggering, establishing a unified timestamp and a unified spatiotemporal reference for the initial multi-source data, and completing the spatiotemporal registration of the initial multi-source data. Based on the initial multi-source data after spatiotemporal registration, a graph optimization model is constructed with keyframe poses as vertices and inter-frame constraints and loop closure constraints as edges. Joint optimization and local feature relocalization are then performed to output a high-precision continuous 3D map of the coal mining area. Based on the high-precision continuous three-dimensional map of the coal mining area, coal mining process parameters, geological model, and equipment operating parameters, a semantic point cloud map carrying mining environment tags and equipment tags is generated online. Based on the semantic point cloud map, the coordinate sequence of the top and bottom plate boundaries of the coal mining machine's operating area is automatically extracted and fitted to form the current cutting curve; The next safe cutting boundary and the coal mining machine height adjustment control curve are generated based on the current cutting curve and sent to the coal mining machine to perform autonomous cutting. During the autonomous truncation process, new multi-source data is obtained, and the following steps are returned for execution: establishing a unified timestamp and spatial coordinate system for the new multi-source data, and completing the spatiotemporal registration of the new multi-source data.

[0009] Optionally, millisecond-level clock hard synchronization is achieved through hardware triggering to establish a unified timestamp and a unified spatiotemporal reference for the initial multi-source data, thereby completing the spatiotemporal registration of the initial multi-source data, including: The initial multi-source data is collected using lidar, vision cameras, inertial measurement units, and external navigation and positioning systems. Millisecond-level clock hard synchronization is achieved through hardware trigger signals from the LiDAR, vision camera, inertial measurement unit, and external navigation and positioning system, and navigation coordinates measured by the external navigation and positioning system are received synchronously. Based on the initial multi-source data, and using the millisecond-level clock hard synchronization and the navigation coordinates, a unified timestamp and a unified spatiotemporal reference are established for the initial multi-source data to create a 3D real-time map and an offline map, thus completing the spatiotemporal registration of the initial multi-source data. The initial multi-source data consists of various types of data required for real-time and offline 3D mapping, including: laser point clouds, visual images, inertial data, and navigation coordinate data.

[0010] Optionally, based on the initial multi-source data after spatiotemporal registration, a graph optimization model is constructed with keyframe poses as vertices and inter-frame constraints and loop closure constraints as edges. Joint optimization and local feature relocalization are then performed to output a high-precision continuous 3D map of the coal mining area, including: The initial multi-source data after spatiotemporal registration is input into the SLAM front end for high-frequency odometer calculation. Construct a graph optimization model with keyframe poses as vertices and inter-frame constraints and loop closure constraints as edges; By combining the global navigation and positioning coordinates provided by the external navigation and positioning system, the graph optimization model is jointly optimized and local feature relocation is performed to suppress the cumulative drift error of the coal mining machine operation process in the high-frequency mileage calculation results, and a high-precision continuous three-dimensional map of the coal mining operation area is output.

[0011] Optionally, by combining the global navigation and positioning coordinates provided by the external navigation and positioning system, joint optimization and local feature relocation are performed on the graph optimization model to suppress the cumulative drift error of the coal mining machine operation process in the high-frequency mileage calculation results, and a high-precision continuous three-dimensional map of the coal mining area is output, including: The high-frequency mileage calculation results are corrected using the global navigation and positioning coordinates. This correction includes: global position reference, trajectory optimization, error assessment, and post-processing correction. The graph optimization model uses keyframe poses as vertices and inter-frame constraints and loop closure detection constraints as edges. Through global pose joint optimization and local feature relocalization, it suppresses the cumulative drift error of the corrected high-frequency odometry calculation results. The lidar, the inertial measurement unit, and the vision camera perform high-frequency mileage measurement and mapping based on the high-frequency mileage calculation results after cumulative drift error suppression, and output a high-precision continuous three-dimensional map of the coal mining area.

[0012] Optionally, based on the high-precision continuous 3D map of the coal mining area, coal mining process parameters, geological model, and equipment operating parameters, a semantic point cloud map carrying mining environment and equipment tags is generated online, including: The high-precision continuous 3D map of the coal mining area, the coal mining process parameters, the geological model, and the equipment operating parameters are input into a real-time semantic segmentation network. At the same time, the geometric information of tightly coupled SLAM is fused to generate the semantic point cloud map online. The semantic point cloud map carries mining environment tags and equipment tags. The mining environment tags represent the mining environment category, and the corresponding mining environment includes: the exposed working face roof, floor, goaf, and geological information. The equipment tags represent the equipment category, and the corresponding equipment includes: coal mining machine, hydraulic support, and scraper conveyor.

[0013] Optionally, based on the semantic point cloud map, the coordinate sequence of the roof and floor boundaries of the coal mining machine's operating area is automatically extracted and fitted to form the current cutter cutting curve, including: Based on the semantic point cloud map, the three-dimensional coordinate sequence of the roof boundary and the three-dimensional coordinate sequence of the floor boundary are automatically identified and extracted for the coal mining machine operation area. The current cutting curve is formed by fitting the three-dimensional coordinate sequence of the top plate boundary and the three-dimensional coordinate sequence of the bottom plate boundary.

[0014] Optionally, the next safe cutting boundary and the coal mining machine height adjustment control curve are generated based on the current cutting curve, and sent to the coal mining machine to perform autonomous cutting, including: Based on the current cutting curve, combined with the preset coal mining process, equipment operation, and geological model parameters iteratively calculated, the next cut safety boundary and the coal mining machine height control curve are generated. The next safe cutting boundary and the height control curve of the coal mining machine are sent to the coal mining machine in the form of control commands to execute the autonomous cutting. The coal mining machine height control curve is a sequence of lifting and lowering instructions for the coal mining machine drum at different cutting positions. The next safe cutting boundary is determined based on the coordinates of the top and bottom plate boundaries, the preset safety margin, and the geological model parameters.

[0015] Optionally, during the autonomous truncation process, new multi-source data is obtained, and the following steps are returned for execution: establishing a unified timestamp and spatial coordinate system for the new multi-source data, and completing the spatiotemporal registration of the new multi-source data, including: During the autonomous cutting process of the coal mining machine, various devices that collect multi-source data sense new operating conditions and collect the new multi-source data. Replace the initial multi-source data with the new multi-source data, and return to the execution steps: establish a unified timestamp and spatial coordinate system for the new multi-source data, complete the spatiotemporal registration of the new multi-source data; and execute subsequent steps to form a fully closed-loop control. The coal mining machine is controlled in a fully closed-loop control mode until all autonomous cutting tasks are completed.

[0016] Optionally, the external navigation and positioning system includes: a measuring robot, or a combination of a fully automatic gyro total station, prism, inertial navigation system and encoder, wherein the external navigation and positioning system is used to provide absolute navigation coordinate reference for the coal mining face; The unified spatiotemporal reference includes: a unified time reference and a unified spatial coordinate system reference.

[0017] Optionally, the real-time semantic segmentation network adopts a real-time deep learning semantic segmentation model to perform point-by-point semantic category annotation on the point cloud, image, text, and equipment time-series data of the high-precision continuous three-dimensional map of the coal mining area. The annotation categories include: roof, floor, coal wall, goaf, fault, interbedded rock, fold, collapse column, scour zone, roof fall, spalling, lithology, coal mining machine, hydraulic support, scraper conveyor, and personnel.

[0018] The closed-loop control method for real-time mapping and autonomous decision-making of coal mining faces proposed in this application is particularly suitable for the coordinated control of real-time three-dimensional mapping and autonomous planning and cutting of coal mining machines in complex underground environments. It achieves millisecond-level clock hard synchronization through hardware triggering, establishes a unified spatiotemporal benchmark, and completes spatiotemporal registration of multi-source data; it constructs a graph optimization model, combines it with the global reference of the measurement and control system for joint optimization and local feature relocation, and suppresses the cumulative drift error of continuous 3D maps; it generates semantic point cloud maps online carrying mining environment labels and equipment labels, automatically extracts the boundary coordinates of the roof and floor plates and fits the cutting curve; finally, it generates the next safe cutting boundary and the coal mining machine height adjustment control curve, and senses new working conditions in real time during execution and updates the mapping data, forming a closed-loop control of "mapping-identification-decision-execution-re-sensing". It realizes closed-loop control of real-time 3D mapping and autonomous decision-making for planning and cutting of coal mining faces, breaks through the limitations of existing technologies in spatiotemporal registration, error suppression, semantic mapping and autonomous decision-making, reduces human intervention, promotes intelligent unmanned mining, and provides technical support for intelligent and unmanned mining of coal mining faces. It has broad application prospects and high practicality. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1This is a flowchart of a closed-loop control method for real-time mapping and autonomous decision-making of coal mining face cutting proposed in an embodiment of this application; Figure 2 This is a preferred closed-loop control block diagram illustrated in the embodiments of this application. Detailed Implementation

[0020] The embodiments of this application will now be described in detail. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] The closed-loop control method for real-time mapping and autonomous decision-making of coal mining faces proposed in this application refers to... Figure 1 The flowchart shown includes the following steps: Step 101: Perform millisecond-level clock hard synchronization via hardware triggering to establish a unified timestamp and unified spatiotemporal reference for the initial multi-source data, and complete the spatiotemporal registration of the initial multi-source data.

[0022] The closed-loop control method proposed in this application first requires the acquisition of initial multi-source data, and then the millisecond-level clock hard synchronization is performed through hardware triggering to establish a unified timestamp and a unified spatiotemporal reference for the initial multi-source data, thereby completing the spatiotemporal registration of the initial multi-source data.

[0023] In one embodiment of this application, a preferred method for completing the spatiotemporal registration of initial multi-source data includes: firstly acquiring initial multi-source data through a lidar, a vision camera, an inertial measurement unit, and an external navigation and positioning system; simultaneously performing millisecond-level clock hard synchronization through the hardware trigger signals of the lidar, the vision camera, the inertial measurement unit, and the external navigation and positioning system, and synchronously receiving navigation coordinates measured by the external navigation and positioning system.

[0024] Finally, based on the initial multi-source data, and using millisecond-level clock hard synchronization and navigation coordinates, a unified timestamp and a unified spatiotemporal reference were established for 3D real-time and offline mapping of the initial multi-source data. This completed both spatiotemporal registration of the initial multi-source data and mapping. The initial multi-source data comprised various types of data required for both real-time and offline 3D mapping, including: laser point clouds, visual images, inertial data, and navigation coordinate data.

[0025] The external navigation and positioning system in this embodiment includes: a surveying robot, or a combination of a fully automatic gyro total station, prism, inertial navigation system, and encoder; that is, the surveying robot itself can perform the functions of the external navigation and positioning system; or a combination of a fully automatic gyro total station, prism, inertial navigation system, and encoder can perform the functions of the external navigation and positioning system. The external navigation and positioning system is used to provide absolute navigation coordinate reference for the coal mining face. The unified spatiotemporal reference in this embodiment includes: a unified time reference and a unified spatial coordinate system reference.

[0026] Step 102: Based on the initial multi-source data after spatiotemporal registration, construct a graph optimization model with keyframe poses as vertices and inter-frame constraints and loop closure constraints as edges, and perform joint optimization and local feature relocalization to output a high-precision continuous 3D map of the coal mining area.

[0027] After the initial spatiotemporal registration of the multi-source data is completed, a graph optimization model is constructed based on the spatiotemporally registered initial multi-source data, with keyframe poses as vertices and inter-frame constraints and loop closure constraints as edges. Joint optimization and local feature relocation are then performed to output a high-precision continuous 3D map of the coal mining area.

[0028] In one embodiment of this application, a preferred method for performing joint optimization and local feature relocation to output a high-precision continuous 3D map of a coal mining area includes: First, the initial multi-source data after spatiotemporal registration is input into the SLAM (Simultaneous localization and mapping) front end for high-frequency odometry calculation; then, a graph optimization model is constructed with keyframe poses as vertices and inter-frame constraints and loop closure constraints as edges.

[0029] By combining the global navigation and positioning coordinates provided by the external navigation and positioning system, joint optimization and local feature relocation are performed on the graph optimization model to suppress the cumulative drift error of the coal mining machine during operation in the high-frequency mileage calculation results, and output a high-precision continuous 3D map of the coal mining area. Preferably, the global navigation and positioning coordinates provided by the external navigation and positioning system can be used to correct the high-frequency mileage calculation results. This correction includes: global position reference, trajectory optimization, error assessment, and post-processing correction.

[0030] The graph optimization model uses keyframe poses as vertices and inter-frame constraints and loop closure detection constraints as edges. Through global pose joint optimization and local feature relocalization, it suppresses the cumulative drift error of the corrected high-frequency mileage calculation results. Finally, based on the high-frequency mileage calculation results after the cumulative drift error suppression, the lidar, inertial measurement unit, and vision camera perform high-frequency mileage measurement and mapping, and output a high-precision continuous three-dimensional map of the coal mining area.

[0031] Step 103: Based on the high-precision continuous 3D map of the coal mining area, coal mining process parameters, geological model, and equipment operating parameters, generate a semantic point cloud map online carrying mining environment tags and equipment tags.

[0032] After obtaining a high-precision continuous 3D map of the coal mining area, a semantic point cloud map carrying mining environment tags and equipment tags is generated online based on the high-precision continuous 3D map of the coal mining area, coal mining process parameters, geological model, and equipment operating parameters.

[0033] In one embodiment of this application, a preferred method for generating a semantic point cloud map online carrying mining environment tags and equipment tags includes: A high-precision continuous 3D map of the coal mining area, coal mining process parameters, geological models, and equipment operating parameters are input into a real-time semantic segmentation network. Simultaneously, geometric information from tightly coupled SLAM (deep fusion of real-time semantic segmentation network and tightly coupled SLAM) is integrated to generate a semantic point cloud map online. The semantic point cloud map carries mining environment tags and equipment tags. The mining environment tags represent the mining environment category, which includes: the exposed roof and floor of the working face, the goaf, and geological information. The equipment tags represent the equipment category, which includes: coal mining machines, hydraulic supports, and scraper conveyors.

[0034] In this embodiment, the real-time semantic segmentation network uses a real-time deep learning semantic segmentation model to perform point-by-point semantic category annotation on the point cloud, images, text, and equipment time-series data of the high-precision continuous three-dimensional map of the coal mining area. The annotation categories include: roof, floor, coal wall, goaf, fault, interbedded rock, fold, collapse column, scour zone, roof fall, spalling, lithology, coal mining machine, hydraulic support, scraper conveyor, and personnel.

[0035] Step 104: Based on the semantic point cloud map, automatically extract the coordinate sequence of the top and bottom plate boundaries of the coal mining machine's operating area, and fit it to form the current cutting curve.

[0036] After obtaining the semantic point cloud map carrying mining environment and equipment labels, the coordinate sequence of the top and bottom plate boundaries of the coal mining machine's operating area is automatically extracted based on the semantic point cloud map carrying mining environment and equipment labels, and fitted to form the current cutting curve.

[0037] In one embodiment of this application, a preferred method for automatically extracting the coordinate sequence of the roof and floor boundaries of the coal mining machine's operating area and fitting it to form the current cutter cutting curve includes: Based on the semantic point cloud map carrying mining environment and equipment tags, the three-dimensional coordinate sequence of the roof boundary and the three-dimensional coordinate sequence of the floor boundary are automatically identified and extracted for the coal mining machine operation area; then, based on the three-dimensional coordinate sequence of the roof boundary and the three-dimensional coordinate sequence of the floor boundary, the current cutting curve is fitted.

[0038] Step 105: Generate the next cut safety cutting boundary and the coal mining machine height adjustment control curve based on the current cut curve, and send them to the coal mining machine to perform autonomous cutting.

[0039] After fitting and forming the current cutter cutting curve, the safe cutting boundary of the next cutter and the height adjustment control curve of the coal mining machine are generated based on the current cutter cutting curve, and then sent to the coal mining machine to perform autonomous cutting.

[0040] In one embodiment of this application, a preferred method for generating the next safe cutting boundary and the coal mining machine height adjustment control curve, and issuing it to the coal mining machine to perform autonomous cutting, includes: Based on the current cutting curve, combined with the preset coal mining process, equipment operation, and iterative calculation of geological model parameters, the safe cutting boundary and the height adjustment control curve of the next cut are generated. The safe cutting boundary and the height adjustment control curve of the next cut are then sent to the coal mining machine in the form of control commands so that the coal mining machine can perform autonomous cutting.

[0041] Among these, the coal mining machine height control curve is the sequence of lifting and lowering instructions for the coal mining machine drum at different cutting positions; the next safe cutting boundary is determined based on the coordinates of the top and bottom plate boundaries, the preset safety margin, and the geological model parameters.

[0042] Step 106: During the autonomous truncation process, new multi-source data is obtained, and the execution steps are returned: establish a unified timestamp and spatial coordinate system for the new multi-source data, and complete the spatiotemporal registration of the new multi-source data.

[0043] During the autonomous cutting process of the coal mining machine, new multi-source data is obtained as the process progresses. This new multi-source data differs from the initial multi-source data. Therefore, after obtaining this new multi-source data, the process returns to the following steps: establishing a unified timestamp and spatial coordinate system for the new multi-source data, and completing the spatiotemporal registration of the new multi-source data. That is, steps 101 to 106 are repeated iteratively until all autonomous cutting tasks are completed.

[0044] Specifically, during the autonomous cutting process of the coal mining machine, various devices that collect multi-source data (such as the aforementioned lidar, vision camera, inertial measurement unit, and external navigation and positioning system) will sense new working conditions and collect new multi-source data. The new multi-source data replaces the initial multi-source data in step 101, and the execution steps are returned: a unified timestamp and spatial coordinate system for the new multi-source data are established, and the spatiotemporal registration of the new multi-source data is completed; and subsequent steps are executed, that is, steps 101 to 106 are repeated iteratively to form a fully closed-loop cyclic control ("sensing new multi-source data - building a new high-precision continuous three-dimensional map of the coal mining area - identifying a new semantic point cloud map - deciding on the new safe cutting boundary for the next cut and the coal mining machine height adjustment control curve - executing new autonomous cutting - sensing new multi-source data again"). The coal mining machine is controlled in a fully closed-loop control mode until all autonomous cutting tasks are completed.

[0045] Combination Figure 2 The diagram shows a preferred closed-loop control block diagram. The preferred implementation flow of the closed-loop control method proposed in this application is as follows: After the process starts, step S1: multi-source data acquisition and spatiotemporal configuration, corresponding to the aforementioned step 101. Preferably, visual camera image data can be acquired by a visual camera, lidar point cloud data can be acquired by a lidar, and IMU inertial data + navigation coordinates from an external telemetry and control system can be acquired by an IMU inertial navigation device and an external telemetry and control system. A unified spatiotemporal reference is achieved based on hardware-triggered millisecond-level hard synchronization, laying the foundation for real-time 3D mapping with a unified coordinate system and timestamps.

[0046] Step S2: High-precision mapping and cumulative error suppression, which corresponds to the aforementioned step 102. Preferably, multi-source data is used as input to construct a keyframe pose map, which is then combined with global navigation and positioning coordinates for joint optimization and local feature relocalization to output a globally consistent pose and a high-precision continuous 3D map (i.e., a high-precision continuous 3D map of the coal mining area).

[0047] Step S3: Automatic extraction of AI semantic features of the mining environment, corresponding to steps 103 and 104 above. Preferably, a high-precision continuous 3D map, coal mining technology (i.e., coal mining technology parameters), geological model, and equipment operating parameters are input into a real-time semantic segmentation network and deeply fused with tightly coupled SLAM (i.e., inputting into a real-time semantic segmentation network while simultaneously fusing geometric information from tightly coupled SLAM), generating an online semantic point cloud map carrying roof and floor information, geological information, and equipment labels. Then, the roof and floor boundary coordinate sequences of the coal mining machine's operating area are automatically extracted and fitted to form the current cutter cutting curve.

[0048] Step S4: Planning and cutting autonomous decision-making and closed-loop control, corresponding to steps 105 and 106 above. Preferably, based on the current cutter cutting curve (equivalent to the roof and floor boundary coordinate sequence), combined with coal mining process parameters, equipment operating parameters, and geological model (i.e., preset coal mining process, equipment operating, and geological model parameters), iterative calculations are performed to generate the next cutter safety boundary and the coal mining machine height adjustment control curve. This is then sent to the coal mining machine for autonomous cutting execution. During execution, the coal mining machine feeds back its real-time status and updated sensing data (i.e., new multi-source data) to step S1, forming a continuous loop (i.e., forming a full closed-loop control). The coal mining machine is controlled in a full closed-loop control manner until all autonomous cutting tasks are completed, ending the entire process.

[0049] In summary, the closed-loop control method for real-time mapping and autonomous decision-making of coal mining faces proposed in this application is particularly suitable for the collaborative control of real-time three-dimensional mapping and autonomous planning and cutting of coal mining machines in complex underground environments. It achieves millisecond-level clock hard synchronization through hardware triggering, establishes a unified spatiotemporal benchmark, and completes spatiotemporal registration of multi-source data; it constructs a graph optimization model, combines it with the global reference of the measurement and control system for joint optimization and local feature relocation, and suppresses the cumulative drift error of continuous 3D maps; it generates semantic point cloud maps online carrying mining environment labels and equipment labels, automatically extracts the boundary coordinates of the roof and floor plates and fits the cutting curve; finally, it generates the next safe cutting boundary and the coal mining machine height adjustment control curve, and senses new working conditions in real time during execution and updates the mapping data, forming a closed-loop control of "mapping-identification-decision-execution-re-sensing". It realizes closed-loop control of real-time 3D mapping and autonomous decision-making for planning and cutting of coal mining faces, breaks through the limitations of existing technologies in spatiotemporal registration, error suppression, semantic mapping and autonomous decision-making, reduces human intervention, promotes intelligent unmanned mining, and provides technical support for intelligent and unmanned mining of coal mining faces. It has broad application prospects and high practicality.

[0050] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0051] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0052] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims. All of these forms are within the protection scope of this application.

Claims

1. A closed-loop control method for real-time mapping and autonomous decision-making in coal mining face planning and cutting, characterized in that, include: Millisecond-level clock hard synchronization is achieved through hardware triggering, establishing a unified timestamp and a unified spatiotemporal reference for the initial multi-source data, and completing the spatiotemporal registration of the initial multi-source data. Based on the initial multi-source data after spatiotemporal registration, a graph optimization model is constructed with keyframe poses as vertices and inter-frame constraints and loop closure constraints as edges. Joint optimization and local feature relocalization are then performed to output a high-precision continuous 3D map of the coal mining area. Based on the high-precision continuous three-dimensional map of the coal mining area, coal mining process parameters, geological model, and equipment operating parameters, a semantic point cloud map carrying mining environment tags and equipment tags is generated online. Based on the semantic point cloud map, the coordinate sequence of the top and bottom plate boundaries of the coal mining machine's operating area is automatically extracted and fitted to form the current cutting curve; The next safe cutting boundary and the coal mining machine height adjustment control curve are generated based on the current cutting curve and sent to the coal mining machine to perform autonomous cutting. During the autonomous truncation process, new multi-source data is obtained, and the following steps are returned for execution: establishing a unified timestamp and spatial coordinate system for the new multi-source data, and completing the spatiotemporal registration of the new multi-source data.

2. The closed-loop control method according to claim 1, characterized in that, Millisecond-level clock hard synchronization is achieved through hardware triggering, establishing a unified timestamp and unified spatiotemporal reference for the initial multi-source data, and completing the spatiotemporal registration of the initial multi-source data, including: The initial multi-source data is collected using lidar, vision cameras, inertial measurement units, and external navigation and positioning systems. Millisecond-level clock hard synchronization is achieved through hardware trigger signals from the LiDAR, vision camera, inertial measurement unit, and external navigation and positioning system, and navigation coordinates measured by the external navigation and positioning system are received synchronously. Based on the initial multi-source data, and using the millisecond-level clock hard synchronization and the navigation coordinates, a unified timestamp and a unified spatiotemporal reference are established for the initial multi-source data to create a 3D real-time map and an offline map, thus completing the spatiotemporal registration of the initial multi-source data. The initial multi-source data consists of various types of data required for real-time and offline 3D mapping, including: laser point clouds, visual images, inertial data, and navigation coordinate data.

3. The closed-loop control method according to claim 2, characterized in that, Based on the initial multi-source data after spatiotemporal registration, a graph optimization model is constructed with keyframe poses as vertices and inter-frame constraints and loop closure constraints as edges. Joint optimization and local feature relocalization are then performed to output a high-precision continuous 3D map of the coal mining area, including: The initial multi-source data after spatiotemporal registration is input into the SLAM front end for high-frequency odometer calculation. Construct a graph optimization model with keyframe poses as vertices and inter-frame constraints and loop closure constraints as edges; By combining the global navigation and positioning coordinates provided by the external navigation and positioning system, the graph optimization model is jointly optimized and local feature relocation is performed to suppress the cumulative drift error of the coal mining machine operation process in the high-frequency mileage calculation results, and a high-precision continuous three-dimensional map of the coal mining operation area is output.

4. The closed-loop control method according to claim 3, characterized in that, Combining the global navigation and positioning coordinates provided by the external navigation and positioning system, the graph optimization model is jointly optimized and local feature relocation is performed to suppress the cumulative drift error of the coal mining machine operation process in the high-frequency mileage calculation results, and a high-precision continuous three-dimensional map of the coal mining area is output, including: The high-frequency mileage calculation results are corrected using the global navigation and positioning coordinates. This correction includes: global position reference, trajectory optimization, error assessment, and post-processing correction. The graph optimization model uses keyframe poses as vertices and inter-frame constraints and loop closure detection constraints as edges. Through global pose joint optimization and local feature relocalization, it suppresses the cumulative drift error of the corrected high-frequency odometry calculation results. The lidar, the inertial measurement unit, and the vision camera perform high-frequency mileage measurement and mapping based on the high-frequency mileage calculation results after cumulative drift error suppression, and output a high-precision continuous three-dimensional map of the coal mining area.

5. The closed-loop control method according to claim 1, characterized in that, Based on the high-precision continuous 3D map of the coal mining area, coal mining process parameters, geological model, and equipment operating parameters, a semantic point cloud map carrying mining environment and equipment tags is generated online, including: The high-precision continuous 3D map of the coal mining area, the coal mining process parameters, the geological model, and the equipment operating parameters are input into a real-time semantic segmentation network. At the same time, the geometric information of tightly coupled SLAM is fused to generate the semantic point cloud map online. The semantic point cloud map carries mining environment tags and equipment tags. The mining environment tags represent the mining environment category, and the corresponding mining environment includes: the exposed working face roof, floor, goaf, and geological information. The equipment tags represent the equipment category, and the corresponding equipment includes: coal mining machine, hydraulic support, and scraper conveyor.

6. The closed-loop control method according to claim 1, characterized in that, Based on the semantic point cloud map, the coordinate sequence of the roof and floor boundaries of the coal mining machine's operating area is automatically extracted and fitted to form the current cutting curve, including: Based on the semantic point cloud map, the three-dimensional coordinate sequence of the roof boundary and the three-dimensional coordinate sequence of the floor boundary are automatically identified and extracted for the coal mining machine operation area. The current cutting curve is formed by fitting the three-dimensional coordinate sequence of the top plate boundary and the three-dimensional coordinate sequence of the bottom plate boundary.

7. The closed-loop control method according to claim 1, characterized in that, Based on the current cutting curve, the next safe cutting boundary and the coal mining machine height adjustment control curve are generated and sent to the coal mining machine for autonomous cutting, including: Based on the current cutting curve, combined with the preset coal mining process, equipment operation, and geological model parameters iteratively calculated, the next cut safety boundary and the coal mining machine height control curve are generated. The next safe cutting boundary and the height control curve of the coal mining machine are sent to the coal mining machine in the form of control commands to execute the autonomous cutting. The coal mining machine height control curve is a sequence of lifting and lowering instructions for the coal mining machine drum at different cutting positions. The next safe cutting boundary is determined based on the coordinates of the top and bottom plate boundaries, the preset safety margin, and the geological model parameters.

8. The closed-loop control method according to claim 1, characterized in that, During the autonomous truncation process, new multi-source data is obtained, and the following steps are returned for execution: establishing a unified timestamp and spatial coordinate system for the new multi-source data, and completing the spatiotemporal registration of the new multi-source data, including: During the autonomous cutting process of the coal mining machine, various devices that collect multi-source data sense new operating conditions and collect the new multi-source data. Replace the initial multi-source data with the new multi-source data, and return to the execution steps: establish a unified timestamp and spatial coordinate system for the new multi-source data, complete the spatiotemporal registration of the new multi-source data; and execute subsequent steps to form a fully closed-loop control. The coal mining machine is controlled in a fully closed-loop control mode until all autonomous cutting tasks are completed.

9. The closed-loop control method according to claim 2, characterized in that, The external navigation and positioning system includes: a surveying robot, or a combination of a fully automatic gyro total station, prism, inertial navigation and encoder, and the external navigation and positioning system is used to provide absolute navigation coordinate reference for the coal mining face; The unified spatiotemporal reference includes: a unified time reference and a unified spatial coordinate system reference.

10. The closed-loop control method according to claim 5, characterized in that, The real-time semantic segmentation network uses a real-time deep learning semantic segmentation model to perform point-by-point semantic category annotation on the point cloud, images, text, and equipment time-series data of the high-precision continuous three-dimensional map of the coal mining area. The annotation categories include: roof, floor, coal wall, goaf, fault, interbedded rock, fold, collapse column, scour zone, roof fall, spalling, lithology, coal mining machine, hydraulic support, scraper conveyor, and personnel.