Intelligent coal unloading control system and method of coal unloading machine

Abstract

The present application relates to the technical field of intelligent control of coal unloading machine, in particular to an intelligent coal unloading control system and method, in the present application, the environmental fusion perception unit fuses the coal pile point cloud data of the three-dimensional laser scanner and the texture information of the RGB-D camera, constructs a real-time digital twin model through hierarchical semantic segmentation, maps the coal pile surface morphology, spatial distribution and residual coal layer thickness in the stage of cleaning, the autonomous decision planning unit adopts a hierarchical optimization strategy, uses a heuristic space segmentation algorithm in the global layer to divide the coal unloading grid and plan the grab bucket, generates a spiral progressive cleaning trajectory of the adaptive cutting point based on the curvature gradient of the coal pile in the local layer, the dynamic posture adjustment unit collects the digging torque data, combines the coal pile thickness to build a coal quality hardness classification model, dynamically corrects the grab bucket cutting angle, digging depth and swing amplitude, and the closed-loop vibration suppression control unit generates a vibration suppression signal using the active damping controller of Lyapunov stability theory to achieve optimal energy consumption of the grab bucket.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for coal unloaders, and more specifically, to an intelligent coal unloading control system and method for a coal unloader. Background Technology

[0002] Intelligent control of coal unloaders is an important technology, specifically applied to the precise control of the grab bucket operation of coal unloaders in coal loading and unloading scenarios. The core is to achieve high efficiency, safety and energy saving of coal unloading operations through dynamic perception and intelligent decision-making, and to meet the high requirements of large-scale coal transportation for unloading efficiency and equipment safety. Currently, coal unloaders mostly rely on preset fixed trajectories or manual operation, which makes it difficult to adapt to the dynamic changes in the shape of the coal pile in the car, resulting in limited operating efficiency and safety.

[0003] Existing coal unloading machine control systems cannot achieve integrated control that accurately perceives the dynamic characteristics of the coal pile, intelligently adapts the grab bucket's movements, and coordinates vibration and energy consumption optimization. This results in low unloading efficiency, high safety risks, and excessive energy consumption. Traditional systems lack real-time and accurate mapping of the coal pile's surface morphology, spatial distribution, and the thickness of residual coal layers during the cleaning stage. They only drive the grab bucket to operate according to a fixed trajectory, which can easily lead to coal pile collapse or missed unloading due to improper entry angle or uncontrolled digging depth. During operation, the vibration of the grab bucket caused by mechanical inertia and the reaction force of the coal pile is not effectively suppressed, which not only causes the grab bucket's posture to deviate, affecting the operation accuracy, but also aggravates the wear of the wire rope and energy consumption. At the same time, it is impossible to dynamically adjust the action parameters of the grab bucket's swing amplitude according to the hardness of the coal, further reducing the adaptability of the operation. These limitations make it difficult for coal unloading machines to cope with complex and ever-changing coal unloading conditions and cannot simultaneously meet the requirements of efficiency, safety, and energy saving. To solve this problem, we provide an intelligent coal unloading machine control system and method. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent coal unloading control system and method to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, an intelligent coal unloading machine unloading control system is provided, comprising:

[0006] The environmental fusion perception unit acquires three-dimensional point cloud data of the coal pile inside the car in real time through a three-dimensional laser scanner, and uses a three-dimensional point cloud reconstruction and semantic segmentation fusion algorithm to construct a real-time digital twin model of the coal unloading operation environment of the car, which maps the surface morphology, spatial distribution and thickness changes of the residual coal layer during the cleaning stage of the coal pile.

[0007] Based on the real-time digital twin model, the autonomous decision-making and planning unit generates a global unloading sequence and a local cleaning path through a hierarchical optimization strategy. The global layer uses a heuristic spatial segmentation algorithm to decompose the car into unloading grid units and plan the grab bucket traversal priority. The local layer generates a spiral progressive cleaning trajectory with an adaptive entry point based on the coal pile curvature gradient analysis.

[0008] During the coal unloading operation, the dynamic posture adjustment unit analyzes the coal pile shape change data in real time and dynamically corrects the grab's entry angle, digging depth and swing amplitude through a mechanical feedback drive strategy.

[0009] The closed-loop vibration suppression control unit combines the grab bucket's posture vibration spectrum collected by sensors, and uses an active damping controller based on Lyapunov stability theory to generate a vibration suppression compensation signal. The acceleration curve of the wire rope winch mechanism is adjusted by a frequency converter driver to achieve smooth tracking of the grab bucket's running trajectory and optimal energy consumption.

[0010] The second objective of this invention is to provide a method for implementing an intelligent coal unloading control system for a coal unloader, comprising the following steps:

[0011] S1. Real-time point cloud data of coal pile in the car is collected by a 3D laser scanner, and texture information from an RGB-D camera is fused with a hierarchical semantic segmentation strategy to construct a real-time digital twin model, which dynamically maps the surface morphology, spatial distribution, and residual coal layer thickness changes during the cleaning stage of the coal pile.

[0012] S2. Based on the real-time digital twin model, a hierarchical optimization strategy is adopted to generate a global coal unloading sequence and a local cleaning path. The global layer decomposes the car into coal unloading grid units through a heuristic spatial segmentation algorithm and plans the traversal priority of the grab bucket. The local layer generates a spiral progressive cleaning trajectory with an adaptive entry point based on the coal pile curvature gradient analysis.

[0013] S3. During the coal unloading operation of the grab bucket, the coal pile shape change data is analyzed in real time, and the grab bucket cutting angle, digging depth and swing amplitude are dynamically corrected through the mechanical feedback drive strategy. The action parameters are adaptively adjusted according to the coal hardness grading model.

[0014] S4. Combining the grab bucket's position and vibration spectrum, an active damping controller based on Lyapunov stability theory is used to generate a vibration suppression compensation signal. The acceleration curve of the wire rope winch mechanism is controlled by a frequency converter to achieve smooth tracking of the grab bucket's running trajectory and optimal energy consumption.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0016] This invention constructs a real-time digital twin model of the coal pile through an environmental fusion sensing unit to accurately map the surface morphology, spatial distribution, and residual coal seam thickness of the coal pile. The autonomous decision-making and planning unit uses a hierarchical optimization strategy to generate a global coal unloading sequence and a local spiral progressive cleaning trajectory. The dynamic posture adjustment unit relies on a mechanical feedback-driven strategy to correct the grab angle, digging depth, and swing amplitude in real time. The closed-loop vibration suppression control unit achieves vibration suppression and energy consumption optimization through the collaboration of an active damping controller based on Lyapunov stability theory and a frequency converter. Overall, it achieves the effects of accurate perception of the dynamic characteristics of the coal pile, adaptive matching of grab actions and working conditions, effective elimination of vibration interference, and optimal energy consumption. This effectively solves the problems of low coal unloading efficiency, high safety risks, and excessive energy consumption caused by the inability of existing coal unloading machine control systems to achieve accurate perception of the dynamic characteristics of the coal pile, intelligent adaptation of grab actions, and integrated control of vibration and energy consumption optimization. Attached Figure Description

[0017] Figure 1 This is an overall block diagram of the present invention;

[0018] Figure 2 This is the overall flowchart of the present invention.

[0019] The meanings of the labels in the diagram are as follows:

[0020] 1. Environmental fusion perception unit; 2. Autonomous decision-making and planning unit; 3. Dynamic posture adjustment unit; 4. Closed-loop vibration suppression control unit. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention provides an intelligent coal unloading machine control system. Please refer to [link / reference]. Figure 1 As shown, it includes:

[0023] The environmental fusion perception unit 1 acquires three-dimensional point cloud data of the coal pile inside the car in real time through a three-dimensional laser scanner, and uses a three-dimensional point cloud reconstruction and semantic segmentation fusion algorithm to construct a real-time digital twin model of the coal unloading operation environment of the car, which maps the surface morphology, spatial distribution and thickness change of the residual coal layer during the cleaning stage of the coal pile.

[0024] Based on the real-time digital twin model, the autonomous decision-making planning unit 2 generates a global unloading coal sequence and a local cleaning path through a hierarchical optimization strategy. The global layer uses a heuristic spatial segmentation algorithm to decompose the car into unloading grid units and plan the grab bucket traversal priority. In the local layer, a spiral progressive cleaning trajectory with adaptive entry points is generated based on the coal pile curvature gradient analysis.

[0025] During the coal unloading operation, the dynamic posture adjustment unit 3 analyzes the coal pile shape change data in real time and dynamically corrects the grab's entry angle, digging depth and swing amplitude through a mechanical feedback drive strategy.

[0026] The closed-loop vibration suppression control unit 4 combines the grab bucket posture vibration spectrum collected by the sensor, and uses an active damping controller based on Lyapunov stability theory to generate a vibration suppression compensation signal. The acceleration curve of the wire rope winch mechanism is adjusted by the frequency converter driver to achieve smooth tracking of the grab bucket's running trajectory and optimal energy consumption.

[0027] The specific method for constructing a real-time digital twin model using the environmental fusion perception unit 1 is as follows:

[0028] Through a multi-source heterogeneous data spatiotemporal fusion mechanism, point cloud data acquired by a 3D laser scanner and texture information acquired by an RGB-D camera are spatiotemporally aligned. Using a hierarchical semantic segmentation strategy, the boundary of the carriage structure is first segmented by a point cloud density clustering algorithm, and then a convolutional neural network is used to identify the semantic regions of the coal pile and residues. Among them, the point cloud topology analysis module detects the surface convex and concave features of the coal pile based on normal vector consistency, and the spatial interpolation algorithm reconstructs the geometry of the occluded area. The thickness inversion model dynamically calculates the thickness of the residual coal seam based on the mapping relationship between point cloud reflection intensity and coal density. This model eliminates point cloud distortion caused by dust during operation through a dust interference compensation filter, and finally generates a real-time digital twin model to provide environmental topology representation for autonomous decision planning unit 2.

[0029] The environmental fusion sensing unit 1 maps the surface morphology, spatial distribution, and thickness changes of the residual coal seam during the cleaning stage of the coal pile as follows:

[0030] For the surface morphology mapping of the coal pile, the moving least squares method is used to fit the curvature field of the point cloud to generate a thermal map of the coal pile undulation characterized by Gaussian curvature. For the spatial distribution mapping, a topological grid of the coal pile volume distribution is constructed based on the Deloni triangulation. The coal pile aggregation degree is quantified by calculating the centroid offset of the point cloud in each grid cell. For the thickness change mapping of the residual coal seam during the cleaning stage, a high-frequency scanning mode is started during the cleaning stage in combination with the thickness inversion model. The elevation difference between the bottom reference surface and the bottom surface of the coal pile is dynamically extracted using a multi-layer point cloud difference algorithm, and the thickness contour cloud map is updated to the real-time digital twin model in real time.

[0031] The specific execution process of the hierarchical optimization strategy in Autonomous Decision Planning Unit 2 is as follows:

[0032] At the global level, based on the thermal map of the spatial distribution of coal piles, an adaptive region growing algorithm is used to decompose the car into several unloading grid units, and a grab bucket traversal priority sequence is generated with the coal pile volume density in the grid as the weight.

[0033] In the local layer, based on the Gaussian curvature heat map, a spiral progressive cleaning trajectory planning is initiated for high curvature areas. This includes generating a spiral path with equal pitch along the coal pile gradient descent direction, using the curvature extreme point as the initial entry point, and adaptively adjusting the spiral step spacing through the curvature radius. The global layer and the local layer collaborate through a dynamic priority arbitrator, triggering global sequence replanning when the coal pile density in a local area changes abruptly.

[0034] The heuristic space partitioning algorithm in Autonomous Decision Planning Unit 2 specifically includes:

[0035] A grid decomposition strategy with spatial structural constraints of the car body is introduced. A three-dimensional bounding box of the car body is established based on the point cloud data of the car body sidewalls. Then, the optimal grid division granularity is calculated based on the entropy value of the coal pile volume distribution. The dynamic weight allocation module assigns a comprehensive weight coefficient to each grid cell. This coefficient is jointly determined by the effective volume of the coal pile in the grid, the Manhattan distance between the grab bucket and the current position, and the coal pile stability factor. Finally, an improved ant colony algorithm is used to generate the shortest path coverage sequence of the grab bucket using the comprehensive weight coefficient as heuristic information.

[0036] The spiral-progressive cleanup trajectory of Autonomous Decision-Making Planning Unit 2 specifically includes:

[0037] Based on the curvature gradient field of the coal pile, a curvature-driven trajectory generator is designed. It proposes to use a small-pitch, high-density spiral in the high curvature region to avoid the risk of collapse, and a large-pitch spiral in the low curvature region to accelerate coverage. The entry angle optimization module dynamically adjusts the initial entry angle of the grab bucket according to the curvature direction vector, so that the grab bucket blade is always perpendicular to the coal pile cutting plane. The trajectory smoothing module uses Bézier curves to fit discrete spiral point sequences to generate a continuous and differentiable grab bucket running trajectory.

[0038] The mechanical feedback driving strategy of the dynamic posture adjustment unit 3 specifically includes:

[0039] A six-dimensional force sensor embedded in the grab bucket hinge collects excavation resistance torque data in real time. Combined with the coal pile thickness distribution map, a coal hardness grading model is constructed. When the resistance torque exceeds the threshold, the angle adjustment module increases the grab bucket tilt angle according to the curvature direction of the current entry point to break the hard coal core. The depth controller dynamically limits the excavation depth to avoid overloading based on the linear relationship between the coal pile thickness inversion result and the resistance torque. When the swing strategy generator detects periodic fluctuations in the resistance torque, it triggers a resonant excavation mode to increase the swing amplitude to loosen the coal seam, and constrains the swing amplitude boundary through the spiral trajectory curvature data.

[0040] The active damping controller of the closed-loop vibration suppression control unit 4 specifically includes:

[0041] Employing a framework that integrates Lyapunov stability theory and online spectrum analysis, the vibration signal acquired by the IMU sensor is first decomposed using wavelet packets to extract the dominant longitudinal vibration frequency and the dominant transverse vibration frequency of the wire rope. An adaptive observer constructs the state-space equation based on a dual-frequency coupled vibration model. The Lyapunov energy function constructor generates a compensating torque that makes the derivative of the energy function negative definite, based on the sum of the vibration kinetic energy and potential energy. This compensating torque is then superimposed onto the torque command of the frequency converter after passing through a bandwidth limiting filter, achieving exponential convergence of vibration energy.

[0042] The optimal energy consumption implementation method for the closed-loop vibration damping control unit 4 is as follows:

[0043] The variable frequency drive uses an acceleration curve smoothing optimization algorithm to decompose the vibration suppression compensation signal into coordinated control commands for the wire rope hoisting motor and the slewing motor. For the hoisting mechanism, the acceleration torque slope is dynamically adjusted according to the change rate of the grab bucket potential energy, and the potential energy recovery mode is activated in the descent phase. For the slewing mechanism, the centripetal force change is predicted based on the spiral trajectory curvature data to compensate for centrifugal vibration in advance. Finally, the power equalizer coordinates the operating points of the two motors.

[0044] Further explanation is needed regarding the specific implementation method of constructing the real-time digital twin model of the environmental fusion perception unit 1. Before the intelligent coal unloader begins unloading operations, the environmental fusion perception unit 1 needs to first transform the physical coal unloading environment of the car into a precise digital model to provide complete environmental information support for subsequent autonomous decision-making. Its core is to eliminate data bias and environmental interference through multi-source data collaborative fusion and refined semantic parsing, and finally generate a real-time digital twin model. The specific implementation method is as follows:

[0045] When constructing the model, the environmental fusion perception unit 1 first processes the basic data through a multi-source heterogeneous data spatiotemporal fusion mechanism. Multi-source heterogeneous data refers to point cloud data acquired by a 3D laser scanner, which focuses on capturing the 3D spatial structure of the coal pile and the carriage, and texture information acquired by an RGB-D camera, which focuses on recording surface color and details. Due to differences in the installation location and operating principles of the acquisition equipment, these two types of data exhibit temporal and spatial discrepancies, requiring spatiotemporal fusion for unification. Temporally, synchronized timestamps are configured for both types of equipment to ensure that each frame of point cloud data and its corresponding texture information are acquired at the same time. Spatially, pre-calibrated device extrinsic parameters are used to map the 3D coordinates of the point cloud data to the image from the RGB-D camera. Like a coordinate system, each spatial point in the point cloud can be accurately matched to the corresponding pixel in the texture image, avoiding structural and textural misalignment issues in subsequent models. After completing the spatiotemporal alignment of the data, a hierarchical semantic segmentation strategy is used to analyze the environmental composition. This strategy follows a logical division based on the overall boundary first, followed by local regions. First, the structural boundaries of the car body are segmented using a point cloud density clustering algorithm. The car body is made of metal plates, and the point cloud collection density on its surface is much higher than that of loose coal piles. The algorithm sets a density threshold and clusters point clouds with densities higher than the threshold into independent regions. Then, combined with the geometric features of the car body's cuboid shape, the boundaries of the side walls, bottom plate, and top plate of the car body are further filtered out to clarify the physical spatial scope of the coal unloading operation. Subsequently, a convolutional neural network is used to identify the semantic regions of the coal pile and residues. This network is pre-trained with labeled data and can subdivide the remaining area after segmenting the car body boundaries into two categories: coal pile residues and residues, based on the structural and texture features of the data. The residues refer to a small amount of coal slag or impurities attached to the car body walls, which need to be distinguished from the main coal pile to provide a basis for subsequent cleaning operations. Building upon semantic segmentation, the model details are further refined through specialized modules, as follows:

[0046] The point cloud topology analysis module detects the surface undulations of the coal pile based on the consistency of normal vectors. A normal vector is a vector perpendicular to the point cloud surface. The direction of the point cloud normal vector changes drastically at protrusions in the coal pile, while the direction is relatively consistent at depressions. The module calculates the angle between adjacent point cloud normal vectors and quantifies the angle using the vector dot product formula. ,in and Two adjacent points and The unit normal vector, " represents the vector dot product operation. The angle between the two normal vectors, with a preset angle threshold. By calibrating 100 sets of coal pile convexity / concave samples, protrusions are defined as those with an angle greater than a preset threshold, and depressions as those less than the threshold. This process generates a surface convexity / concave distribution map of the coal pile, restoring its true morphology. A spatial interpolation algorithm is then used to reconstruct the geometry of occluded areas. Since the unloading environment may contain car pillars and equipment supports, resulting in missing point cloud data for some coal pile areas, the algorithm selects unoccluded point cloud data within a range of five times the surrounding occluded area and uses an inverse distance weighted interpolation formula to complete the missing data. ,in The coordinates of the point to be interpolated are: The number of nearby unobstructed points is set to 10-15 to balance accuracy and efficiency. For the first Coordinates of neighboring points, For weights ( for arrive Euclidean distance, The squared term ensures that points closer to each other have higher weights. By using the coordinates and normal vector trends of neighboring points, the point cloud of the occluded area is gradually interpolated and completed to ensure the integrity of the coal pile shape without breaks. The thickness inversion model dynamically calculates the thickness of the residual coal seam based on the mapping relationship between the point cloud reflection intensity and the coal density. The greater the coal density, the stronger the reflection intensity of the laser. The model establishes a correspondence table between reflection intensity and coal density through laboratory calibration, and then, combined with the correlation law between the density and thickness of the coal seam, it inversely derives the thickness value of the residual coal seam on the bottom of the car. At the same time, the model eliminates the point cloud distortion caused by dust generated during operation through a dust interference compensation filter. Dust generated during coal unloading will weaken the laser reflection intensity. If there is no dust, the reflection intensity will be lower. Dust concentration Intensity measured at time The filter uses an exponential decay correction formula. ,in To correct the reflected intensity, Through comparative experiments with five different dust concentrations, the deviation between the corrected intensity and the intensity without dust was less than or equal to 5%, ensuring the accuracy of the thickness calculation results. Through the above steps, the aligned multi-source data, segmented semantic regions, improved detailed information on concave and convex shapes, occlusion completion, and residual thickness were integrated into a three-dimensional digital model, namely a real-time digital twin model. This model can accurately map the spatial location and morphological characteristics of the car body, coal pile, and residues in the physical coal unloading environment, providing a clear environmental topological representation for the autonomous decision-making planning unit 2 to plan the coal unloading sequence and cleaning path.

[0047] The specific implementation method of coal pile feature mapping in Environment Fusion Perception Unit 1: After the basic data fusion and semantic segmentation of the real-time digital twin model are completed in Environment Fusion Perception Unit 1, in order to further accurately restore the key dynamic features of the coal unloading operation environment, it is also necessary to carry out special mapping for the surface morphology, spatial distribution, and thickness changes of the residual coal seam during the cleaning stage. These three types of features directly determine the coal unloading strategy of Autonomous Decision Planning Unit 2, so they need to be meticulously characterized through targeted methods. The specific implementation method is as follows:

[0048] For coal pile surface morphology mapping, the core approach is to fit the point cloud curvature field using the moving least squares method to generate a heat map of coal pile undulation characterized by Gaussian curvature. The moving least squares method is an algorithm capable of smoothly fitting discrete point cloud data without requiring a fixed grid. It can adaptively generate continuous surface models based on the point cloud distribution, making it particularly suitable for processing point cloud data with irregular shapes, such as coal piles, during unloading operations. In practice, local point clouds are first selected from the semantically segmented coal pile point cloud data according to a preset window. The algorithm then fits the local surface within each window, and the curvature value of each point on the surface is calculated, forming a point cloud curvature field covering the entire coal pile—that is, continuous data reflecting the degree of curvature of the coal pile surface. The field is then used to extract the Gaussian curvature from the curvature field. Gaussian curvature is a key indicator for measuring the concavity and convexity of a surface. Its value can directly distinguish between convex, concave, and flat areas on the coal pile surface. A positive value indicates a convex surface, a negative value indicates a concave surface, and a value close to zero indicates a flat surface. Finally, the Gaussian curvature value is correlated with the color gradient to generate a heat map of the coal pile undulation. This heat map can intuitively present the three-dimensional morphology of the coal pile surface, avoiding morphological misjudgments caused by point cloud discretization. After completing the fine mapping of the coal pile surface morphology, in order to clarify the overall distribution pattern of the coal pile within the carriage, it is necessary to further carry out the spatial distribution mapping of the coal pile. Specifically, a topological mesh of the coal pile volume distribution is constructed based on the Deloni triangulation, and the volume distribution of the coal pile is calculated. The centroid offset of point clouds within a grid cell quantifies the coal pile aggregation degree. Deloni triangulation is an algorithm that transforms discrete point cloud data into a uniform triangular mesh. Its advantage lies in its ability to adaptively adjust the mesh size according to point cloud density; the mesh is finer in dense point cloud areas and coarser in sparse areas, efficiently covering the entire car space. In operation, the coal pile point cloud data is first input into the algorithm to generate a triangular topological mesh covering the coal pile area within the car. Then, the centroid coordinates of all point clouds within each grid cell are calculated, i.e., the average of the X, Y, and Z coordinates of all point clouds within that cell. These centroid coordinates are then compared with the geometric center coordinates of the grid cell to obtain the point cloud centroid offset. The larger the offset, the higher the centroid offset. The more uneven the distribution of coal piles within a grid cell and the higher the density, the spatial distribution of coal piles is transformed into quantifiable density data. This provides a basis for subsequent global-level unloading grid cell division and traversal priority planning. When the unloading operation enters the cleaning stage, i.e., when the overall volume of the coal pile is significantly reduced and the remaining coal seam in the car needs to be cleaned, the thickness change of the residual coal seam directly affects the cleaning path planning. Therefore, it is necessary to perform residual coal seam thickness change mapping for this stage. Specifically, in combination with the thickness inversion model, a high-frequency scanning mode is started in the cleaning stage. The elevation difference between the bottom reference surface and the bottom surface of the coal pile is dynamically extracted using a multi-layer point cloud difference algorithm, and the thickness contour cloud map is updated to the real-time digital twin model in real time.First, the coal pile thickness is thin and unevenly distributed during the cleaning stage. Therefore, a high-frequency scanning mode needs to be activated, increasing the scanning frequency of the 3D laser scanner from 1 scan / second during the conventional coal unloading stage to 5 scans / second. This ensures that subtle thickness changes in the residual coal seam can be captured. Simultaneously, the thickness inversion model constructed earlier is invoked, using the mapping relationship between point cloud reflection intensity and coal density to calculate the thickness, providing basic parameters for thickness calculation. Then, a multi-layer point cloud difference algorithm is employed. First, a reference surface for the bottom of the cargo compartment is constructed using point cloud data from the cargo compartment floor, i.e., a 3D planar model of the cargo compartment floor. Disturbing points such as stains and protrusions on the floor need to be removed to ensure the reference surface is flat. Next, the point cloud of the bottom surface of the coal pile, i.e., the surface point cloud of the residual coal seam in contact with air, is extracted from the high-frequency scanned point cloud data. The elevation difference between the point cloud of the bottom surface of the coal pile and the reference surface of the cargo compartment floor in the vertical direction (Z-axis direction) is calculated. This difference represents the thickness of the residual coal seam at the corresponding location. For example, an elevation difference of 10cm at a certain point represents a residual coal seam thickness of 10cm at that location. Finally, the residual thickness data of all locations are generated into a thickness contour map by drawing a contour line every 2 cm according to the contour line rules. The map is updated every 5 seconds and synchronized to the real-time digital twin model, so that the autonomous decision-making and planning unit 2 can grasp the thickness distribution of the residual coal seam in real time and avoid omissions or equipment damage during cleaning.

[0049] The specific implementation of the hierarchical optimization strategy of the autonomous decision-making and planning unit 2 is as follows: After the environmental fusion perception unit 1 generates a real-time digital twin model containing a thermal map of the coal pile spatial distribution and a Gaussian curvature thermal map, the autonomous decision-making and planning unit 2 needs to generate a coal unloading plan from two dimensions: global operation sequence and local operation details. The global layer ensures optimal overall operation efficiency and avoids wasting time by the grab bucket traveling back and forth. The local layer focuses on the safe cleaning of complex coal pile shapes to prevent collapse or missed unloading. The two work together to achieve accurate and efficient coal unloading operations. The specific implementation is as follows:

[0050] The execution of the hierarchical optimization strategy begins at the global layer. The core of the global layer is to divide the coal unloading grid units and plan the priority of grab bucket traversal. Its operation is entirely based on the coal pile spatial distribution heat map generated earlier through Deloni triangulation and centroid offset. The darker the color in the map, the higher the coal pile aggregation degree. First, an adaptive region growing algorithm is used to decompose the coal car into several unloading grid units. The adaptive region growing algorithm is an algorithm that automatically divides regions based on data characteristics. It does not require a preset fixed grid size. During operation, points in high-density areas are selected as seed points in the thermal map of the coal pile spatial distribution, such as the points in the top 20% of the density values. Then, with the seed points as the center, the surrounding point clouds are gradually merged to form initial grid units according to the rule that the density difference between adjacent areas is less than a preset threshold, such as 5%. If the area of ​​the initial grid unit is too large, such as exceeding 1.5 times the coverage area of ​​a single grab bucket operation, it is automatically split into smaller units. If the point cloud of a certain area is sparse and has low density, adjacent small units are merged to finally generate unloading grid units with a size that fits the grab bucket operation range, such as 2m×2m or 3m×3m, to ensure that each unit can be cleaned in 1-2 grab bucket operations.

[0051] Subsequently, a grab bucket traversal priority sequence is generated using the coal pile volume density within the grid cell as the weight. The grab bucket traversal priority sequence refers to the ordered instruction queue for grab bucket accessing the grid cell generated by the path optimization algorithm based on the comprehensive weight value dynamically calculated based on the effective volume of the coal pile, spatial position relationship, and stability parameters of the coal unloading grid cell of the car body. This sequence is updated in real time based on changes in the coal pile morphology during operation to ensure optimal global coal unloading efficiency. The coal pile volume density refers to the ratio of the actual volume of the coal pile in each grid cell, calculated from point cloud data, to the total volume of the grid cell, such as 2m × 2m × the average height of the coal pile. The higher the ratio, the more abundant the coal in the cell. The system sorts all grid cells from high to low volume density and fine-tunes the order by combining the Manhattan distance of the grab bucket's current position, i.e., the horizontal straight-line distance between the grab bucket's current coordinates and the center coordinates of the grid cell. If two grid cells have similar volume densities with a difference of less than 3%, the closer cell is selected first to avoid long-distance movement of the grab bucket. If a high volume density cell is too far away, exceeding 2 / 3 of the length of the car body, 1-2 medium-density, close-range cells are inserted to balance efficiency and path rationality, ultimately forming a grab bucket traversal priority sequence, such as cell A → cell C → cell B → cell D, guiding the grab bucket to operate sequentially. After determining the traversal order at the global layer, the local layer needs to plan the cleaning trajectory for the complex coal pile morphology within each grid cell, especially for high-curvature areas. Specifically, based on the Gaussian curvature heatmap generated earlier using the moving least squares method, high-curvature bulges are marked in red and high-curvature depressions in dark blue. A spiral progressive cleaning trajectory is then initiated for high-curvature areas. First, the curvature extrema of the high-curvature area are located. These extrema are the points where the Gaussian curvature value is the maximum, the bulge is the sharpest, or the depression is the minimum, the depression is the deepest. The coal pile structure around these points is unstable and prone to collapse when the grab bucket cuts in. Therefore, these points are used as the initial entry points to ensure orderly cleaning starting from the most dangerous area. Next, a constant-pitch spiral path is generated along the gradient descent direction of the coal pile. The gradient descent direction refers to the direction from the extreme point of curvature towards the surrounding decreasing Gaussian curvature value, i.e., from convex to gentler areas, and from concave to gentler areas. The system uses the extreme point as the spiral center and generates a spiral line with a fixed pitch, such as 0.5m, adapted to the width of the grab bucket's cutting edge. This ensures that when the grab bucket operates along the spiral line, the curvature change of the coal pile area is gradual each time it cuts in, avoiding sudden contact with high-curvature structures. At the same time, the spiral step spacing is adaptively adjusted through the curvature radius. The spiral step spacing refers to the distance between two adjacent spiral turns. When the grab bucket moves to an area with a small curvature radius, such as the top of a convex area with a curvature radius of 1m, the step spacing automatically decreases, such as from 0.5m to 0.3m, to avoid the grab bucket's cutting edge contacting too many high-curvature coal piles at the same time, resulting in excessive resistance. When it moves to an area with a large curvature radius, such as a gentle slope with a curvature radius of 5m, the step spacing automatically increases, such as from 0.5m to 0.8m, improving operating efficiency. In addition, the trajectory smoothing module uses Bézier curves to fit discrete spiral point sequences to generate a continuous and differentiable grab bucket trajectory, avoiding sudden stops or turns during grab bucket movement and reducing the impact on the coal pile.The global and local layers collaborate through a dynamic priority arbitrator: the arbitrator compares the global traversal sequence and the execution status of the local trajectory in real time. If the local layer detects a sudden change in the coal pile density of a certain grid cell during operation, such as the actual coal pile volume density being 30% higher than the initial calculated value, it indicates that there was an omission in the initial point cloud scan. In this case, the global sequence replanning is immediately triggered, the current cell operation is paused, the volume density and priority of all grid cells are recalculated, and the traversal sequence is adjusted. For example, the current high-density cell is prioritized to be assigned to the next operation position to avoid the collapse of the coal pile in that cell after the subsequent operation of other cells. If the local trajectory execution is smooth with no density changes and no risk of collapse, the operation continues according to the original global sequence to ensure a balance between global efficiency and local safety.

[0052] The specific implementation of the heuristic spatial segmentation algorithm in the autonomous decision-making and planning unit 2: When the autonomous decision-making and planning unit 2 executes the global layer operation of the hierarchical optimization strategy, the heuristic spatial segmentation algorithm undertakes the core tasks of accurate decomposition of the car body grid and optimization of the grab bucket traversal path. This algorithm breaks through the limitations of traditional fixed-size grid segmentation, and achieves coordinated optimization of grid division and path planning by deeply combining the physical structure of the car body and the actual distribution characteristics of the coal pile, laying the foundation for subsequent efficient grab bucket operation. The specific implementation is as follows:

[0053] The heuristic spatial segmentation algorithm first introduces a mesh decomposition strategy based on the spatial structure constraints of the car body to ensure that the mesh division does not deviate from the actual operation boundary. The system first calls the point cloud data of the car body sidewalls generated by the environment fusion perception unit 1. This data has been semantically segmented by the point cloud density clustering algorithm and can accurately reflect the three-dimensional coordinates of the front, rear, left, and right sidewalls of the car body, such as the X-axis range of the left sidewall and the Y-axis range of the front sidewall. Based on these coordinates, a three-dimensional bounding box of the car body is constructed using a 3D modeling tool: taking the four corner points of the car body floor as the reference and combining the sidewall height data, such as a car body height of 3 meters, the length, width, and height range of the bounding box are determined, such as 15 meters long, 3 meters wide, and 3 meters high. This bounding box strictly fits the physical structure of the car body, limiting the coal unloading operation space to inside the car body and avoiding subsequent meshes exceeding the car body range, which could lead to the grab bucket grabbing empty or colliding with the car body. After determining the bounding box, the optimal mesh size is calculated based on the entropy value of the coal pile volume distribution. The entropy value of the coal pile volume distribution is an indicator that quantifies the unevenness of the coal pile distribution within the wagon. A higher entropy value indicates a more scattered coal pile distribution, such as dense coal piles in some areas and sparse coal piles in others. A lower entropy value indicates a more uniform distribution. During the calculation, the three-dimensional bounding box of the wagon is initially divided into several temporary small meshes, such as 1m×1m×1m. The coal pile volume within each temporary mesh is calculated using point cloud data, and then substituted into the entropy formula to obtain the overall coal pile volume distribution entropy value. If the entropy value is higher than the preset threshold, such as 0.8, it indicates that the coal pile distribution is chaotic, and a fine-grained grid, such as 1.5m × 1.5m, should be used to ensure accurate coverage of scattered coal pile areas. If the entropy value is lower than the threshold, such as 0.5, it indicates that the coal pile distribution is uniform, and a coarse-grained grid, such as 2.5m × 2.5m, can be used to reduce the number of grids and improve operational efficiency. In this way, the grid division granularity is adapted to the coal pile distribution characteristics, avoiding the problems of insufficient density and waste in sparse areas in traditional fixed grids. After the grid division is completed, the dynamic weight allocation module assigns a comprehensive weight coefficient to each grid cell. This coefficient is the core basis for determining the priority of grab bucket traversal and is jointly determined by the effective volume of the coal pile within the grid, the Manhattan distance of the grab bucket's current position, and the coal pile stability factor. Among them, the effective volume of the coal pile refers to the actual volume of coal pile that can be excavated by the grab bucket within the grid unit, excluding the internal voids of the coal pile and the ungrabable parts that are stuck to the wall of the truck bed. It is calculated by removing invalid areas from point cloud data. The larger the effective volume, the more workable the grid can be, and the higher the weight coefficient. For example, the Manhattan distance is the straight-line distance between the current three-dimensional coordinates of the grab bucket and the geometric center coordinates of the grid unit on the horizontal plane, ignoring the height difference. Since the height of the grab bucket is adjustable, the closer the distance, the shorter the time it takes for the grab bucket to move. The weight coefficient is the second highest, such as 0.3. The coal pile stability factor is determined based on the Gaussian curvature heat map of the coal pile within the grid unit. High curvature convex or concave areas have low stability and small factor values. Flat areas have high stability and large factor values. The higher the stability, the lower the risk of collapse during grab bucket operation. The weight coefficient is the lowest, such as 0.3.The three factors are weighted and summed according to the above proportions to obtain the comprehensive weight coefficient of each grid cell. For example, if the effective volume proportion of a grid is 0.4, the comprehensive weight coefficient is 0.32; the distance proportion is 0.3, the comprehensive weight coefficient is 0.24; and the stability factor proportion is 0.3, the comprehensive weight coefficient is 0.27, and the comprehensive weight coefficient is 0.83. The higher the coefficient, the higher the priority of the grid cell. Finally, an improved ant colony algorithm is used to generate the shortest path coverage sequence of the grab bucket using the comprehensive weight coefficient as heuristic information. The improved ant colony algorithm is an algorithm that optimizes the pheromone update mechanism based on the traditional ant colony algorithm. The traditional algorithm is prone to slow path convergence. After improvement, the pheromone concentration of high-weight grid cells is increased. For example, the pheromone concentration of a grid with a comprehensive weight coefficient of 0.83 is twice that of a grid with a coefficient of 0.5. This guides the search individuals in the ant algorithm to prioritize high-weight grids and adjust the pheromone evaporation coefficient at the same time, avoiding the algorithm from getting trapped in local optima. For example, if a certain path is short but covers low-weight grids, the algorithm can adjust it in time through pheromone evaporation. When the algorithm runs, the ant starts from the current position of the grab bucket and selects the grid cell with high pheromone concentration (i.e., high comprehensive weight coefficient) and no coverage at each step until all grid cells are covered. The final generated path sequence satisfies the requirement of prioritizing high-weight grid cells and ensures that the total distance of the grab bucket is minimized, thus achieving dual optimization of work efficiency and priority.

[0054] The specific implementation method of the spiral progressive cleaning trajectory in the autonomous decision-making planning unit 2: After the autonomous decision-making planning unit 2 completes the global layer coal unloading grid cell division and traversal priority planning through the heuristic spatial segmentation algorithm, for each grid cell, especially in high curvature areas such as the complex shape of coal piles with sharp protrusions and deep depressions, a spiral progressive cleaning trajectory is needed to ensure the safety and thoroughness of the operation. Traditional straight cleaning trajectories are prone to causing coal pile collapse due to cutting too deep at once, or causing missed unloading due to incomplete path coverage. The spiral progressive trajectory can dynamically adjust the path parameters around the curvature characteristics of the coal pile, taking into account both safety and efficiency. The specific implementation method is as follows:

[0055] The generation of this cleaning trajectory is based on the curvature gradient field of the coal pile. This field is derived from the Gaussian curvature heatmap generated by the environmental fusion sensing unit 1 mentioned earlier. It not only includes the Gaussian curvature value of each point to distinguish the degree of concavity and convexity, but also marks the direction and rate of curvature change. For example, the direction in which the curvature value gradually decreases from the apex of a convex area to a flat area is the gradient descent direction, providing a precise morphological reference for trajectory generation. Based on this, the system designs a curvature-driven trajectory generator. The core logic of this generator is that curvature characteristics determine path density: in high-curvature areas, such as sharp convexities with Gaussian curvature values ​​> 0.8 and deep depressions < -0.8, the coal pile structure is loose and unstable. Using a wide-spacing path can easily lead to the collapse of unclogged coal. Therefore, the trajectory generator automatically adopts a small-pitch, high-density spiral. The pitch refers to the distance between two adjacent spiral turns. The small pitch is set to 0.3-0.5m to adapt to the grab bucket cutting edge width. For example, if the cutting edge width is 0.6m, the small pitch ensures that there is slight overlap between adjacent turns. To avoid missing any coal, high density is reflected in the greater number of spiral coils per unit area, such as 3-4 coils per square meter. This allows the grab bucket to excavate only a small amount of coal each time, reducing the impact on the overall structure of the coal pile through gradual cleaning and fundamentally avoiding the risk of collapse. In low-curvature areas, such as gentle slopes with Gaussian curvature values ​​between -0.3 and 0.3, the coal pile structure is stable and has a regular shape. The trajectory generator switches to a large-pitch spiral with a pitch of 0.8-1.2m, reducing the number of spiral coils per unit area to 1-2. This expands the coverage area of ​​a single operation, accelerating the cleaning process and balancing local operation efficiency with the global traversal rhythm. To further improve the adaptability of the grab bucket and the coal pile, the entry angle optimization module dynamically adjusts the initial entry angle of the grab bucket based on the curvature direction vector. The curvature direction vector is the vector along the direction of the most drastic curvature change at a point on the coal pile surface. For example, at a high-curvature protrusion, the direction vector points to the outer, deeper depression away from the center of the protrusion, while at the center of the depression, the direction vector points to the center of the depression. This vector directly reflects the tilt direction of the coal pile's cutting plane. The module calculates the required angle for adjusting the grab's cutting edge by reading the curvature direction vector at each entry point on the trajectory in real time. For example, in a high-curvature protruding area, if the curvature direction vector shows that the coal pile cutting plane is tilted at 30°, the initial entry angle of the grab is adjusted to 30° with the horizontal direction, making the grab's cutting edge completely perpendicular to the cutting plane. If the entry area is a low-curvature, gentle surface, and the curvature direction vector is close to horizontal, the entry angle is adjusted to 0° for horizontal entry. This dynamic adjustment ensures that the grab's cutting edge always contacts the coal pile at the optimal angle, avoiding excessive force on one side of the cutting edge that could cause the coal pile to collapse, maximizing the amount of material excavated in a single operation, and reducing ineffective work.Considering that abrupt changes in the trajectory during grab bucket operation, such as excessively large angle differences between discrete points, could easily cause grab bucket vibration or wire rope swaying, the trajectory smoothing module fits a discrete spiral point sequence using a Bézier curve. This discrete spiral point sequence consists of key points of the spiral path initially generated by the curvature-driven trajectory generator. For example, a point is marked at 0.2m intervals, forming a series of discrete coordinates. The Bézier curve, through the traction effect of the control vertices, connects these discrete points into a continuous and arbitrarily differentiable smooth curve. During fitting, the module sets the control vertices of the Bézier curve based on the curvature change trend of adjacent discrete points: if the curvature change of adjacent points is gentle, such as in a low-curvature region, the control vertices are closer to the discrete points, and the curve is closer to a straight line; if the curvature change of adjacent points is drastic, such as at the transition from a high-curvature region to a low-curvature region, the control vertices are farther from the discrete points, and the curve transition is smoother. The resulting continuous guideable trajectory perfectly matches the motion characteristics of the grab bucket, ensuring that the speed and direction of the grab bucket remain stable as it runs along the trajectory. This reduces wear and tear on the mechanical structure and allows the grab bucket to precisely conform to the shape of the coal pile to complete the cleaning operation, laying a stable foundation for the real-time correction of the subsequent dynamic posture adjustment unit 3.

[0056] The specific implementation of the mechanical feedback-driven strategy of the dynamic posture adjustment unit 3: After the autonomous decision-making and planning unit 2 generates the spiral progressive cleaning trajectory, although the grab bucket has a clear operating path, the actual working conditions of the coal pile during coal unloading operations often have dynamic changes that are not covered by the preset trajectory. For example, there may be hard coal cores hidden inside the coal pile, or the thickness of the local coal seam may exceed the inversion value of the environmental fusion perception unit 1. Relying solely on a fixed trajectory can easily lead to overload of the grab bucket or incomplete excavation. Therefore, the dynamic posture adjustment unit 3 needs to use a mechanical feedback-driven strategy to capture the excavation resistance signal in real time and correct the grab bucket's action parameters to ensure operational safety and efficiency. The specific implementation is as follows:

[0057] The core data source for this strategy is a six-dimensional force sensor embedded in the grab bucket's hinge. This six-dimensional force sensor is a high-precision sensor that can simultaneously acquire the tensile and compressive forces in the X, Y, and Z axes of three-dimensional space, as well as the rotational torque of three-dimensional forces around the X, Y, and Z axes. It is installed at the hinge where the grab bucket connects to the wire rope and can directly capture the digging resistance torque that the grab bucket experiences when it cuts into the coal pile, i.e., the resisting torque around the grab bucket's rotation axis. The greater the resistance torque, the harder the coal or the deeper the digging. The acquisition frequency is set to 100Hz to ensure real-time response to changes in resistance. The system integrates the excavation resistance torque data collected by sensors with the coal pile thickness distribution map generated by the environmental fusion sensing unit 1, which indicates the coal pile thickness in the current working area. For example, if the thickness in a certain area is 1.2m, a correlation analysis is performed to construct a coal hardness grading model. The model uses resistance torque-thickness as the core dimension to divide the coal quality into three levels. When the resistance torque is less than the preset low threshold, such as 40% of the rated resistance torque of the grab bucket and the corresponding coal pile thickness is less than 1m, it is determined to be a soft coal seam. When the resistance torque is between the low threshold and 70% of the rated resistance torque of the high threshold, the thickness is 1-2m, the resistance torque is determined to be a medium coal seam, and the thickness is greater than 2m, the thickness is determined to be a hard coal seam that may contain a hard coal core. This model provides a hardness basis for subsequent grab bucket action adjustments. The angle adjustment module is a key component for dealing with hard coal cores. Its working logic revolves around the resistance torque threshold and the curvature direction of the entry point: The system pre-sets the resistance torque threshold based on the rated load capacity of the grab bucket, such as 80% of the rated resistance torque. Exceeding this threshold may cause grab bucket deformation or motor overload. When the six-dimensional force sensor detects that the digging resistance torque exceeds this threshold, it immediately determines that there is a hard coal core in the current area. The module synchronously calls the curvature direction of the current entry point output by the autonomous decision planning unit 2. For example, the curvature direction of high curvature protrusions points to the outside of the coal pile, and the curvature direction of low curvature depressions points to the inside. Based on this, the grab bucket's forward tilt angle is increased. The grab bucket's forward tilt angle is the angle between the grab bucket's cutting edge and the horizontal direction. Increasing the forward tilt angle allows the cutting edge to cut into the coal pile at a sharper angle. For example, at high curvature protrusions, the curvature direction points to the outside. The module increases the forward tilt angle from the conventional 15° to 25°, allowing the cutting edge to conform to the curvature direction and penetrate deep into the gap of the hard coal core. By concentrating the force, the hard coal is broken, and the grab bucket is prevented from stopping due to excessive overall force.The depth controller focuses on preventing the grab bucket from overloading. Its control is based on the linear relationship between the coal pile thickness inversion result and the resistance torque. The thickness inversion result generated by the environmental fusion sensing unit 1 mentioned above has clearly defined the theoretical thickness of the coal pile in the current working area. In actual excavation, the resistance torque will increase linearly with the excavation depth. For example, when the thickness is 1m, the resistance torque increases by an average of 5% for every 0.1m increase in excavation depth. The depth controller dynamically limits the maximum excavation depth of the grab bucket by calculating this linear relationship in real time. If the inversion thickness of the coal pile in a certain area is 1.5m, when the grab bucket excavation depth reaches 1.2m, the resistance torque is close to 75% of the rated threshold. The controller will automatically limit the maximum excavation depth to 1.3m to avoid the resistance torque exceeding the threshold due to continued deep excavation. If the resistance torque decreases due to the softening of the coal, the depth limit will be gradually relaxed to ensure that the grab bucket always operates within the safe load range and prevent motor overload or wire rope wear. In addition, the swing strategy generator is used to address the resistance fluctuation problem caused by the uneven looseness of the coal seam. When the six-dimensional force sensor detects periodic fluctuations in the excavation resistance torque, such as a high-resistance-low-resistance cycle every 2 seconds, it indicates that there is an alternating distribution of loose interlayers and dense regions within the coal seam. This immediately triggers the resonant excavation mode: This mode controls the grab bucket to swing left and right around the hinge axis, using the impact force generated by the swing to loosen the coal seam. For example, increasing the swing amplitude from the conventional 5° to 10° causes the grab bucket to simultaneously generate a combined cutting and swinging motion during the excavation process, breaking up the dense structure of the coal seam. Meanwhile, to avoid excessive swaying that could damage the stability of the coal pile, the system uses the curvature data of the spiral progressive clearing trajectory for boundary constraints. In high curvature areas, such as bulges and depressions, where the absolute value of the curvature is >0.8, the swaying amplitude is limited to 5-8° to prevent the swaying from causing the coal pile to collapse. In low curvature areas where the absolute value of the curvature is <0.3, the swaying amplitude can be widened to 10-12° to maximize the loosening effect. Finally, through a closed-loop logic of real-time mechanical feedback, dynamic parameter correction, and structural stability constraints, the grab bucket can still operate efficiently and safely under complex coal pile conditions.

[0058] The specific implementation of the closed-loop vibration suppression control unit 4 with active damping controller and optimal energy consumption: After the dynamic posture adjustment unit 3 corrects the grab's entry angle, digging depth, and swing amplitude in real time through a mechanical feedback drive strategy, the grab can adapt to the coal pile working conditions to complete the digging action. However, during operation, vibration will still occur due to mechanical inertia, such as the tension and rebound of the wire rope, sudden changes in the reaction force of the coal pile, such as the instantaneous impact force when breaking hard coal cores. If the vibration continues, it will lead to grab position deviation, accelerated wear of the wire rope, and even safety hazards. Therefore, the closed-loop vibration suppression control unit 4 needs to suppress vibration through active damping controller. At the same time, in order to avoid excessive increase in energy consumption during vibration suppression, energy consumption also needs to be optimized to ensure the dual goals of stable operation and energy-saving operation. The specific implementation is as follows:

[0059] The core of the active damping controller of the closed-loop vibration suppression control unit 4 adopts a fusion framework of Lyapunov stability theory and online spectrum analysis. Lyapunov stability theory is a theory that judges the stability of a system by constructing an energy function. Its core logic is that if the vibration energy of the system can be continuously decayed, stable vibration suppression can be achieved. Online spectrum analysis is used to analyze the frequency characteristics of the vibration signal in real time and accurately locate the vibration source that needs to be suppressed. When the controller is working, it first collects the acceleration and angular velocity signals of the grab bucket in real time through the IMU sensor inertial measurement unit installed on the top of the grab bucket, which indirectly reflects the vibration state and obtains the vibration signal. The signal is then input into the wavelet packet decomposition module. Wavelet packet decomposition is a signal processing technology that can finely divide frequency ranges. Compared with the traditional Fourier transform, it can capture the time domain and frequency domain characteristics of vibration at the same time. Through decomposition, two types of key frequencies can be accurately extracted from the complex vibration signal: one is the longitudinal vibration frequency of the wire rope, such as 2-5Hz, which is caused by the tension and rebound of the wire rope and manifests as the grab bucket bobbing up and down; the other is the lateral vibration frequency of the swing angle, such as 1-3Hz, which is caused by the left and right swing of the grab bucket and manifests as the grab bucket deflecting horizontally. These two types of frequencies directly determine the intensity and degree of damage of the vibration. Next, the adaptive observer constructs a state-space equation based on the dual-frequency coupled vibration model. The dual-frequency coupled vibration model is a mathematical model that considers the mutual influence between the longitudinal vibration of the wire rope and the lateral vibration of the grab bucket. For example, longitudinal vibration will aggravate lateral swaying, and vice versa. The adaptive observer will dynamically adjust the model parameters, such as the wire rope stiffness coefficient and the grab bucket rotational inertia, according to the two types of dominant frequencies extracted in real time, and construct a state-space equation that can accurately describe the current vibration state. The equation includes state quantities such as vibration displacement, velocity, and acceleration, ensuring accurate characterization of the vibration system. Subsequently, the Lyapunov energy function constructor constructs an energy function based on the principle that the kinetic energy of vibration is determined by the mass and vibration velocity of the grab bucket, and the potential energy is determined by the deformation of the wire rope and the swing angle of the grab bucket. The larger the value of the energy function, the more intense the vibration. Through mathematical derivation, the constructor generates a compensating torque that makes the derivative of the energy function negative. A negative derivative means that the energy function is continuously decreasing, that is, the vibration energy is constantly decaying. The magnitude and direction of the compensating torque will be dynamically adjusted according to the dominant vibration frequency. For example, when the dominant longitudinal vibration frequency is 3Hz, the upward compensating torque is output to offset the downward rebound of the wire rope. When the dominant transverse vibration frequency is 2Hz, the reverse compensating torque is output to suppress the swing. The compensation torque needs to be processed by a bandwidth limiting filter first. The filter is used to remove high-frequency noise in the compensation torque, such as instantaneous pulse torque caused by signal interference, to avoid frequent start-stop of the frequency converter driver caused by high-frequency signals. Then, the filtered compensation torque is superimposed on the torque command of the frequency converter driver. For example, if the original torque command is 100 N·m, a compensation torque of -15 N·m is superimposed to suppress vibration. The motor torque of the wire rope hoisting mechanism is controlled by the frequency converter driver, thereby adjusting the tension and speed of the wire rope, and finally achieving smooth tracking of the grab bucket's running trajectory, so that the vibration energy converges exponentially, such as the vibration amplitude decreasing from 5 cm to less than 1 cm within 1 second.Building upon the stable operation achieved through the active damping controller, to prevent a surge in energy consumption due to excessive output compensation torque during vibration suppression, the closed-loop vibration suppression control unit 4 further optimizes energy consumption. This is achieved by using a frequency converter as the core, reducing ineffective energy consumption through refined control of motor actions. The frequency converter first employs an acceleration curve smoothing optimization algorithm to decompose the vibration suppression compensation signal generated by the active damping controller into coordinated control commands for the wire rope hoisting motor and the slewing motor. This algorithm transforms abrupt changes in the original compensation signal, such as sudden large torque adjustments, into a continuous and smooth curve, preventing additional energy consumption due to sudden torque changes, such as the inrush current during motor startup. Simultaneously, it ensures coordinated action between the two motors; for example, when the hoisting motor adjusts the wire rope tension, the slewing motor synchronously fine-tunes its speed, avoiding mutual interference that could increase energy consumption. For the wire rope winch motor, the system dynamically adjusts the acceleration torque slope based on the potential energy change rate of the grab bucket. The potential energy change rate refers to the rate of change of the potential energy of the grab bucket when it moves vertically. When the grab bucket rises, the potential energy increases, requiring the motor to output more torque. When the grab bucket falls, the potential energy decreases, allowing the motor to reduce torque or even recover energy. When the grab bucket rises, such as moving from the bottom of the car body to the unloading point, the potential energy change rate is positive, and the frequency converter driver gradually increases the acceleration torque slope, such as from 5 N·m / s to 8 N·m / s, to ensure that the grab bucket accelerates smoothly without wasting torque. When the grab bucket falls, such as returning from the unloading point to the car body, the potential energy change rate is negative, and the driver immediately activates the potential energy recovery mode. At this time, the gravitational potential energy of the grab bucket will drive the winch motor to rotate in the opposite direction, and the motor switches to generator mode, feeding the generated electrical energy back to the grid. For example, the electrical energy recovered during the falling process can meet 30% of the energy consumption during the rising phase, fundamentally reducing the energy consumption during the falling phase. To control the horizontal rotation of the grab bucket using a rotary motor to adjust the working position, the system predicts centripetal force changes based on the curvature data of the spiral progressive cleaning trajectory generated by the autonomous decision-making and planning unit 2. The greater the curvature of the spiral trajectory, such as the sharper turns in high-curvature convex areas, the greater the centrifugal force generated during grab bucket rotation. Without compensation, this can lead to load fluctuations and increased energy consumption in the rotary motor. Therefore, the driver pre-calculates the required compensation torque based on the curvature data. For example, when the curvature value is 0.8, it outputs a compensation torque of 15 N·m in advance to counteract the centrifugal force, avoiding frequent output adjustments by the motor due to sudden load changes and reducing ineffective energy consumption. Simultaneously, this compensation torque works in conjunction with the vibration suppression requirements of the active damping controller to ensure that centrifugal vibration is suppressed without additional burden on the motor.Finally, the power equalizer coordinates the operating points of the two motors. The power equalizer collects the actual power consumption of the hoist motor and the slewing motor in real time. For example, if the current power of the hoist motor is 5kW and the power of the slewing motor is 3kW, and the power of one motor exceeds 80% of the rated power, such as the hoist motor reaching 4.8kW, then the unnecessary compensation torque of the other motor is appropriately reduced. For example, the compensation torque of the slewing motor is reduced from 15N·m to 12N·m, so that the total power is controlled within a safe and energy-saving range. If the power of both motors is low, such as the total power <5kW, then the vibration suppression compensation effect can be appropriately enhanced to achieve a dynamic balance of strengthening stability when energy consumption is redundant and prioritizing energy saving when the load is high. Ultimately, while ensuring the smooth tracking of the grab bucket's running trajectory, optimal energy consumption is achieved.

[0060] In this invention, the environmental fusion perception unit 1 integrates coal pile point cloud data from a 3D laser scanner with texture information from an RGB-D camera. After hierarchical semantic segmentation, a real-time digital twin model is constructed, mapping the surface morphology, spatial distribution, and residual coal layer thickness during the cleaning stage of the coal pile. The autonomous decision-making and planning unit 2 adopts a hierarchical optimization strategy. The global layer uses a heuristic spatial segmentation algorithm to divide the coal unloading grid and plan the grab bucket. The local layer generates a spiral progressive cleaning trajectory with an adaptive entry point based on the curvature gradient of the coal pile. The dynamic posture adjustment unit 3 collects excavation resistance torque data and builds a coal hardness grading model based on the coal pile thickness. It dynamically corrects the grab bucket entry angle, excavation depth, and swing amplitude. The closed-loop vibration suppression control unit 4 uses an active damping controller based on Lyapunov stability theory to generate a vibration suppression signal, achieving optimal energy consumption of the grab bucket.

[0061] The second objective of this invention is to provide a method for implementing an intelligent coal unloading machine control system including any of the above-mentioned features, comprising the following steps:

[0062] S1. Real-time point cloud data of coal pile in the car is collected by a 3D laser scanner, and texture information from an RGB-D camera is fused with a hierarchical semantic segmentation strategy to construct a real-time digital twin model, which dynamically maps the surface morphology, spatial distribution, and residual coal layer thickness changes during the cleaning stage of the coal pile.

[0063] S2. Based on the real-time digital twin model, a hierarchical optimization strategy is adopted to generate a global coal unloading sequence and a local cleaning path. The global layer decomposes the car into coal unloading grid units through a heuristic spatial segmentation algorithm and plans the traversal priority of the grab bucket. The local layer generates a spiral progressive cleaning trajectory with an adaptive entry point based on the coal pile curvature gradient analysis.

[0064] S3. During the coal unloading operation of the grab bucket, the coal pile shape change data is analyzed in real time, and the grab bucket cutting angle, digging depth and swing amplitude are dynamically corrected through the mechanical feedback drive strategy. The action parameters are adaptively adjusted according to the coal hardness grading model.

[0065] S4. Combining the grab bucket's position and vibration spectrum, an active damping controller based on Lyapunov stability theory is used to generate a vibration suppression compensation signal. The acceleration curve of the wire rope winch mechanism is controlled by a frequency converter to achieve smooth tracking of the grab bucket's running trajectory and optimal energy consumption.

[0066] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent coal unloading machine unloading control system, characterized in that, include: The environmental fusion perception unit (1) acquires the three-dimensional point cloud data of the coal pile in the car in real time through a three-dimensional laser scanner, and uses a three-dimensional point cloud reconstruction and semantic segmentation fusion algorithm to construct a real-time digital twin model of the coal unloading operation environment of the car, which maps the surface morphology, spatial distribution and thickness change of the residual coal layer during the cleaning stage of the coal pile. The specific method for constructing a real-time digital twin model using the environmental fusion sensing unit (1) is as follows: Through the spatiotemporal fusion mechanism of multi-source heterogeneous data, the point cloud data collected by the three-dimensional laser scanner and the texture information obtained by the RGB-D camera are spatiotemporally aligned. Using the hierarchical semantic segmentation strategy, the boundary of the carriage structure is first segmented by the point cloud density clustering algorithm, and then the semantic region of the coal pile and the residue is identified by the convolutional neural network. Among them, the point cloud topology analysis module detects the concave and convex features of the coal pile surface based on the normal vector consistency, the spatial interpolation algorithm reconstructs the geometric shape of the occluded area, and the thickness inversion model dynamically calculates the thickness of the residual coal layer based on the mapping relationship between the point cloud reflection intensity and the coal density. The model eliminates the point cloud distortion caused by the dust of the operation through the dust interference compensation filter, and finally generates a real-time digital twin model to provide environmental topology representation for the autonomous decision planning unit (2). The autonomous decision-making planning unit (2) generates a global unloading coal sequence and a local cleaning path based on the real-time digital twin model through a hierarchical optimization strategy. The global layer uses a heuristic spatial segmentation algorithm to decompose the car into unloading grid units and plan the grab bucket traversal priority. The local layer generates a spiral progressive cleaning trajectory with an adaptive entry point based on the coal pile curvature gradient analysis. The dynamic posture adjustment unit (3) analyzes the coal pile shape change data in real time during the coal unloading operation of the grab bucket, and dynamically corrects the grab bucket's cutting angle, digging depth and swing amplitude through the mechanical feedback drive strategy. The closed-loop vibration control unit (4) combines the grab bucket posture vibration spectrum collected by the sensor and uses an active damping controller based on Lyapunov stability theory to generate a vibration compensation signal. The acceleration curve of the wire rope winch mechanism is controlled by the frequency converter to achieve smooth tracking of the grab bucket's running trajectory and optimal energy consumption.

2. The intelligent coal unloading machine unloading control system according to claim 1, characterized in that, The environmental fusion sensing unit (1) maps the surface morphology, spatial distribution, and thickness changes of the residual coal seam during the cleaning stage of the coal pile in the following way: For the surface morphology mapping of the coal pile, the moving least squares method is used to fit the curvature field of the point cloud to generate a thermal map of the coal pile undulation characterized by Gaussian curvature. For the spatial distribution mapping, a topological grid of the coal pile volume distribution is constructed based on the Deloni triangulation. The coal pile aggregation degree is quantified by calculating the centroid offset of the point cloud in each grid cell. For the thickness change mapping of the residual coal seam during the cleaning stage, a high-frequency scanning mode is started during the cleaning stage in combination with the thickness inversion model. The elevation difference between the bottom reference surface and the bottom surface of the coal pile is dynamically extracted using a multi-layer point cloud difference algorithm, and the thickness contour cloud map is updated to the real-time digital twin model in real time.

3. The intelligent coal unloading machine unloading control system according to claim 1, characterized in that: The specific execution process of the hierarchical optimization strategy of the autonomous decision-making planning unit (2) is as follows: At the global level, based on the thermal map of the spatial distribution of coal piles, an adaptive region growing algorithm is used to decompose the car into several unloading grid units, and a grab bucket traversal priority sequence is generated with the coal pile volume density in the grid as the weight. In the local layer, based on the Gaussian curvature heat map, a spiral progressive cleaning trajectory planning is initiated for high curvature areas. This includes generating a spiral path with equal pitch along the coal pile gradient descent direction, using the curvature extreme point as the initial entry point, and adaptively adjusting the spiral step spacing through the curvature radius. The global layer and the local layer collaborate through a dynamic priority arbitrator, triggering global sequence replanning when the coal pile density in a local area changes abruptly.

4. The intelligent coal unloading machine unloading control system according to claim 3, characterized in that: The heuristic space partitioning algorithm in the autonomous decision-making planning unit (2) specifically includes: A grid decomposition strategy with spatial structural constraints of the car body is introduced. A three-dimensional bounding box of the car body is established based on the point cloud data of the car body sidewalls. Then, the optimal grid division granularity is calculated based on the entropy value of the coal pile volume distribution. The dynamic weight allocation module assigns a comprehensive weight coefficient to each grid cell. This coefficient is jointly determined by the effective volume of the coal pile in the grid, the Manhattan distance between the grab bucket and the current position, and the coal pile stability factor. Finally, an improved ant colony algorithm is used to generate the shortest path coverage sequence of the grab bucket using the comprehensive weight coefficient as heuristic information.

5. The intelligent coal unloading machine unloading control system according to claim 3, characterized in that: The spiral-progressive cleanup trajectory of the autonomous decision-making planning unit (2) specifically includes: Based on the curvature gradient field of the coal pile, a curvature-driven trajectory generator is designed. In the high curvature region, a small-pitch, high-density spiral is used to avoid the risk of collapse, while in the low curvature region, a large-pitch spiral is used to accelerate coverage. The entry angle optimization module dynamically adjusts the initial entry angle of the grab bucket according to the curvature direction vector, so that the grab bucket blade is always perpendicular to the coal pile cutting plane. The trajectory smoothing module uses Bezier curve fitting of discrete spiral point series to generate a continuous and differentiable grab bucket running trajectory.

6. The intelligent coal unloading machine unloading control system according to claim 1, characterized in that: The mechanical feedback driving strategy of the dynamic posture adjustment unit (3) specifically includes: A six-dimensional force sensor embedded in the grab bucket hinge collects excavation resistance torque data in real time. Combined with the coal pile thickness distribution map, a coal hardness grading model is constructed. When the resistance torque exceeds the threshold, the angle adjustment module increases the grab bucket tilt angle according to the curvature direction of the current entry point to break the hard coal core. The depth controller dynamically limits the excavation depth to avoid overloading based on the linear relationship between the coal pile thickness inversion result and the resistance torque. When the swing strategy generator detects periodic fluctuations in the resistance torque, it triggers a resonant excavation mode to increase the swing amplitude to loosen the coal seam, and constrains the swing amplitude boundary through the spiral trajectory curvature data.

7. The intelligent coal unloading machine unloading control system according to claim 1, characterized in that: The active damping controller of the closed-loop vibration suppression control unit (4) specifically includes: Employing a framework that integrates Lyapunov stability theory and online spectrum analysis, the vibration signal acquired by the IMU sensor is first decomposed using wavelet packets to extract the dominant longitudinal vibration frequency and the dominant transverse vibration frequency of the wire rope. An adaptive observer constructs the state-space equation based on a dual-frequency coupled vibration model. The Lyapunov energy function constructor generates a compensating torque that makes the derivative of the energy function negative definite, based on the sum of the vibration kinetic energy and potential energy. This compensating torque is then superimposed onto the torque command of the frequency converter after passing through a bandwidth limiting filter, achieving exponential convergence of vibration energy.

8. The intelligent coal unloading machine unloading control system according to claim 7, characterized in that: The optimal energy consumption of the closed-loop vibration damping control unit (4) is achieved as follows: The variable frequency drive uses an acceleration curve smoothing optimization algorithm to decompose the vibration suppression compensation signal into coordinated control commands for the wire rope hoisting motor and the slewing motor. For the hoisting mechanism, the acceleration torque slope is dynamically adjusted according to the change rate of the grab bucket potential energy, and the potential energy recovery mode is activated in the descent phase. For the slewing mechanism, the centripetal force change is predicted based on the spiral trajectory curvature data to compensate for centrifugal vibration in advance. Finally, the power equalizer coordinates the operating points of the two motors.

9. A method for implementing an intelligent coal unloading machine control system according to any one of claims 1-8, characterized in that: Includes the following steps: S1. Real-time point cloud data of coal pile in the car is collected by a 3D laser scanner, and texture information from an RGB-D camera is fused with a hierarchical semantic segmentation strategy to construct a real-time digital twin model, which dynamically maps the surface morphology, spatial distribution, and residual coal layer thickness changes during the cleaning stage of the coal pile. S2. Based on the real-time digital twin model, a hierarchical optimization strategy is adopted to generate a global coal unloading sequence and a local cleaning path. The global layer decomposes the car into coal unloading grid units through a heuristic spatial segmentation algorithm and plans the traversal priority of the grab bucket. The local layer generates a spiral progressive cleaning trajectory with an adaptive entry point based on the coal pile curvature gradient analysis. S3. During the coal unloading operation of the grab bucket, the coal pile shape change data is analyzed in real time, and the grab bucket cutting angle, digging depth and swing amplitude are dynamically corrected through the mechanical feedback drive strategy. The action parameters are adaptively adjusted according to the coal hardness grading model. S4. Combining the grab bucket's position and vibration spectrum, an active damping controller based on Lyapunov stability theory is used to generate a vibration suppression compensation signal. The acceleration curve of the wire rope winch mechanism is controlled by a frequency converter to achieve smooth tracking of the grab bucket's running trajectory and optimal energy consumption.

Images