A control system and method for a coal mine haulage rotary lifting beam

Abstract

The present application relates to the field of machine vision learning and Internet of Things sensing technology, in particular to a coal mine transportation rotating lifting beam control system and method; through a data acquisition module, visual, environmental and distributed tension data are synchronously acquired; after fusion preprocessing, a scene perception module based on a machine learning model is used for comprehensive analysis, and a structured result containing swing, wind disturbance and stress distribution is output; a risk assessment module quantifies risks and predicts trends accordingly; a disturbance rejection adjustment module cooperatively generates a disturbance rejection motion instruction and a lifting rope stress balancing adjustment instruction; the execution is cooperatively executed by an execution control module; an optimization module iteratively optimizes a model and a strategy using whole-process data; the present application deeply fuses visual swing analysis, distributed lifting rope tension monitoring and environmental meteorological data, and realizes all-around and quantitative perception of the state, stress and environment of the hoisting system.

Description

Technical Field

[0001] This invention relates to the fields of machine vision learning and Internet of Things sensing technology, specifically to a control system and method for a rotating hoisting beam in coal mine transportation. Background Technology

[0002] In the underground and shaft-head transportation process of coal mines, rotating hoisting beams are used for the hoisting and transfer of heavy materials and equipment components. Current control methods mainly rely on manual experience, facing three major bottlenecks: First, environmental perception is limited to a single dimension; operators rely solely on visual observation and cannot quantify key safety parameters such as wind load, sway, and the internal stress distribution of the hoisting ropes. Second, control is crude, unable to provide rapid and accurate compensation responses to dynamic disturbances (such as sudden gusts of wind) during hoisting. Third, there is a lack of real-time, distributed monitoring of the status of key load-bearing components such as the hoisting ropes, making it difficult to detect potential localized overload hazards in a timely manner. In existing patent literature, there is no intelligent control system that integrates visual swing analysis, distributed tension sensor network and environmental meteorological information, and uses machine learning model to predict risks, generate anti-disturbance control strategies and dynamically adjust the force on the suspension rope. In view of this, in order to overcome the above-mentioned technical problems, the present invention proposes a control system and method for a rotating hoisting beam in coal mine transportation. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention proposes a control system and method for a rotating hoisting beam in coal mine transportation, which solves the problems of insensitivity to environmental disturbances, control lag, and lack of fine monitoring of load-bearing components in existing technologies, and realizes adaptive stable control and active safety protection in the hoisting process under complex environments. The intelligent control system for a rotating hoisting beam in coal mine transportation according to the present invention includes: The data acquisition module is used to simultaneously collect visual images of the work area, environmental wind speed and direction, and distributed tension data of multiple nodes of the hook and rope; The data fusion and preprocessing module is connected to the data acquisition module and is used to perform time synchronization and preprocessing on multi-source heterogeneous data to generate a multimodal feature dataset. The scene perception module is connected to the data fusion and preprocessing module. It has a built-in trained machine learning model for analyzing the multimodal feature dataset and outputting structured perception results that fuse visual swing analysis, environmental disturbance assessment and rope force distribution. The risk assessment module, connected to the scene perception module, is used to perform real-time risk quantification and trend prediction based on the structured perception results. The disturbance rejection adjustment module is connected to the risk assessment module and the scene perception module respectively. It is used to collaboratively generate compensatory motion control commands for actively suppressing the swing of the suspended object and dynamic adjustment commands for optimizing the uniformity of the force on the suspension rope when the risk is controllable. An execution control module, connected to the disturbance rejection adjustment module, is used to coordinate the execution of the motion control command and the dynamic adjustment command; The optimization module connects the data fusion and preprocessing module and the execution control module. It is used to store data of the entire operation process to form a panoramic closed-loop database, and to iteratively optimize the machine learning model and control strategy based on the database. The interaction module connects the risk assessment module and the anti-interference adjustment module, and is used for information visualization and human interaction.

[0004] Preferably, the data acquisition module includes: The multi-view visual acquisition unit consists of multiple explosion-proof cameras installed around the lifting beam and the work area, used to acquire real-time images of the suspended object, hook, rope and environment. The environmental information acquisition unit consists of an explosion-proof anemometer and rain gauge installed in the work area. The distributed tension sensing network unit includes a first tension sensor installed at the hook and multiple second tension sensors evenly distributed at preset intervals along the length of the suspension rope.

[0005] Preferably, the structured perception results output by the scene perception module specifically include: the swing vector of the suspended object obtained based on visual image sequence analysis, the environmental disturbance contribution obtained by fusing environmental data and the swing vector, and the force distribution map and uniformity index of the suspension rope generated based on distributed tension data.

[0006] Preferably, the prediction function of the risk assessment module includes, based on the current swing state of the suspended object, the trend of environmental wind speed and direction, and the change in the force distribution of the suspension rope, using a time-series prediction model to make probabilistic predictions of collision risk, instability risk, and local overload risk of the suspension rope within a set future time period.

[0007] Preferably, the disturbance rejection adjustment module includes: The disturbance rejection decision submodule is used to generate a sequence of compensatory motion control commands for actively suppressing swaying based on the contribution of the environmental disturbance and the sway vector of the suspended object. The adjustment decision submodule is used to generate dynamic adjustment instructions for adjusting the force transmission path of the suspension rope to balance the force on each node, based on the force distribution diagram and uniformity index of the suspension rope.

[0008] Preferably, the execution control module includes: The drive unit is used to execute the compensatory motion control command sequence to control the rotation, lifting, and lateral movement of the lifting beam; The adjustment unit is used to execute the dynamic adjustment command and control the operation of the auxiliary adjustment mechanism connected to the suspension rope or hook to change the geometry or force distribution of the suspension rope.

[0009] Preferably, the optimization module includes: The storage submodule is used to store raw multimodal sensing data, preprocessed feature datasets, structured perception results, risk assessment and prediction reports, and all issued control commands and actuator status feedback in a unified timeline, forming a panoramic closed-loop database. The optimization submodule is used to perform supervised incremental training on the machine learning model in the scene perception module using new data in the panoramic closed-loop database, and to update the control strategy parameters in the anti-disturbance adjustment module through reinforcement learning.

[0010] Preferably, the interactive module provides a graphical interface for real-time visualization of the swing trajectory of the suspended object, environmental wind field information, cloud map of the force distribution of the suspension rope, comprehensive risk level and prediction trend, and system decision logic, and is equipped with an interface for switching between automatic / manual control modes and manual intervention.

[0011] A method for intelligent control of a rotating hoisting beam in coal mine transportation includes the following: S1: Through the data acquisition module, visual images of the work site, environmental wind speed and direction, and tension data of multiple nodes of the hook and rope are collected simultaneously; S2: Through the data fusion and preprocessing module, the data collected by S1 is synchronized, fused and preprocessed in time to form a multimodal feature dataset; S3: Through the scene perception module, the built-in machine learning model is used to analyze the multimodal feature dataset and output a structured perception result that integrates the swing of the suspended object, environmental disturbances and the force distribution of the suspension rope; S4: Based on the structured perception results, the risk assessment module performs real-time comprehensive risk assessment and short-term risk trend prediction. S5: Based on the results of S4, the disturbance rejection adjustment module collaboratively generates disturbance rejection motion control commands for actively suppressing the swing of the suspended object, as well as dynamic adjustment commands for optimizing the uniformity of the force on the suspension rope. S6: By executing the control module, the main motion mechanism and the suspension rope adjustment mechanism are driven to coordinately execute the instructions generated by S5; S7: By optimizing the storage submodule in the module, store the full-link data of this control cycle to the panoramic closed-loop database; S8: Through the optimization submodule in the optimization module, based on the new data in the panoramic closed-loop database, the machine learning model in S3 and the instructions in S5 are iteratively optimized.

[0012] Preferably, generating dynamic adjustment instructions in S5 specifically includes: analyzing the force distribution diagram of the suspension rope in the structured perception results, identifying specific sensor nodes whose tension exceeds the average load setting ratio or absolute safety threshold; calculating the required suspension rope shape fine-tuning amount based on the node's location and overall distribution, and generating an instruction sequence to drive the corresponding adjustment mechanism to transfer part of the load from the high-load node to the low-load node.

[0013] The beneficial effects of this invention are as follows: 1. This invention deeply integrates visual swing analysis, distributed sling tension monitoring, and environmental meteorological data to achieve comprehensive and quantitative perception of the status, stress, and environment of the hoisting system; multi-node tension sensors can provide early warning of local fatigue or overload risks of the sling; and through the fusion prediction model, intervention can be carried out before the suspended object swings significantly or collides, transforming passive response into proactive prevention. 2. This invention can sense environmental changes such as wind disturbance and automatically generate reverse compensation motion and suspension rope force adjustment strategies, which significantly improves the stability of the hoisting process and reduces the difficulty of operation; by monitoring and adjusting the force distribution of the suspension rope, stress concentration is avoided, the service life of the suspension rope is extended, and preventive maintenance is achieved. Attached Figure Description

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] Figure 1 This is a flowchart of the system in this invention; Figure 2 This is a flowchart of the method in this invention; Figure 3 This is a flowchart of the dual-path parallel decision-making process of the intelligent anti-disturbance decision and dynamic adjustment module of the present invention; Figure 4 This is a schematic diagram of the working principle of the closed-loop learning and optimization module of the present invention; Figure 5 This is a flowchart of the dynamic adjustment instruction generation method of the present invention. Detailed Implementation

[0016] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0017] like Figure 1-5 As shown, the embodiments of the present invention are as follows;

[0018] Example 1: The intelligent control system for a rotating hoisting beam in coal mine transportation according to the present invention includes: The data acquisition module is used to simultaneously collect visual images of the work area, environmental wind speed and direction, and distributed tension data of multiple nodes of the hook and rope; The data fusion and preprocessing module is connected to the data acquisition module and is used to perform time synchronization and preprocessing on multi-source heterogeneous data to generate a multimodal feature dataset. The scene perception module is connected to the data fusion and preprocessing module. It has a built-in trained machine learning model for analyzing the multimodal feature dataset and outputting structured perception results that fuse visual swing analysis, environmental disturbance assessment and rope force distribution. The risk assessment module, connected to the scene perception module, is used to perform real-time risk quantification and trend prediction based on the structured perception results. The disturbance rejection adjustment module is connected to the risk assessment module and the scene perception module respectively. It is used to collaboratively generate compensatory motion control commands for actively suppressing the swing of the suspended object and dynamic adjustment commands for optimizing the uniformity of the force on the suspension rope when the risk is controllable. An execution control module, connected to the disturbance rejection adjustment module, is used to coordinate the execution of the motion control command and the dynamic adjustment command; The optimization module connects the data fusion and preprocessing module and the execution control module. It is used to store data of the entire operation process to form a panoramic closed-loop database, and to iteratively optimize the machine learning model and control strategy based on the database. The interaction module connects the risk assessment module and the anti-interference adjustment module, and is used for information visualization and human interaction.

[0019] As attached Figure 1 As shown, this invention constructs a complete information processing chain from physical world perception to intelligent decision output. The system starts with a data acquisition module, which is not simply a physical assembly of cameras, anemometers and tension sensors. Instead, it uses a unified hardware synchronization triggering mechanism to ensure that visual images, environmental meteorological data and distributed tension data are collected simultaneously at the same time reference point. This is the fundamental premise for all subsequent cross-modal fusion analysis to be successful. Specifically, at the start of each system control cycle, each sensing unit in the data acquisition module is synchronously triggered; the multi-view vision acquisition unit captures multiple images reflecting the instantaneous spatial pose and swing state of the hoisted object and hook; the environmental information acquisition unit simultaneously records the wind speed and direction data at that moment as a quantitative description of external disturbances; the distributed tension sensing network unit records the total tension of the hook and the local tension values ​​at each node along the hoisting rope; these three types of data describe the physical state of the hoisting system at the same moment from different dimensions, and after being packaged together and attached with a precise timestamp, they are transmitted to the downstream data fusion and preprocessing module; After receiving the raw data, the data fusion and preprocessing module first performs time synchronization processing. Due to the differences in sampling frequencies of different sensors, the module uses an interpolation algorithm to unify all data onto the same time series. Then, spatial alignment is performed, that is, mapping the tension sensor node number to its actual physical position coordinates on the hoisting rope, and transforming the pixel coordinates in the visual image to the unified world coordinate system of the hoisting operation. After completing the spatiotemporal alignment, the module performs quality improvement processing on various types of data: performing dehazing enhancement and illumination equalization on the visual image to overcome the harsh visual environment in underground or open-pit mines; performing filtering and noise reduction on the mechanical and environmental data to remove abnormal jump values ​​introduced by electrical interference or mechanical vibration. The various types of data processed in the above way are encapsulated into a structured multimodal feature data sample, which contains three components: image feature map, environmental parameter vector, and tension distribution vector, which together constitute the standardized input of the subsequent intelligent perception model. The scene perception module is the cognitive center of the entire system. Internally, it houses a machine learning model pre-trained with a large amount of coal mine hoisting operation data. After receiving the aforementioned multimodal feature data samples, this model performs deep fusion of different modal information in a shared feature extraction layer. The principle of fusion lies in the fact that, through learning from a large amount of data, the model has mastered a deterministic and coordinated change pattern among visual features, environmental wind field data, and hoisting rope tension distribution when the hoisted object undergoes a certain pattern of swaying. Based on these learned cross-modal correlations, the model can extract a unified situational awareness from seemingly unrelated raw data. Finally, through multiple parallel task output heads, the model simultaneously outputs a complete structured perception result, including the hoisted object type and attitude, hoisted object sway vector, environmental disturbance contribution, hoisting rope force distribution heatmap, and uniformity index. This result is a comprehensive, quantitative, and structured mathematical description of the current hoisting operation situation, providing a complete information foundation for subsequent risk assessment and control decisions.

[0020] Example 2: In one embodiment of the present invention, the data acquisition module includes: The multi-view visual acquisition unit consists of multiple explosion-proof cameras installed around the lifting beam and the work area, used to acquire real-time images of the suspended object, hook, rope and environment. The environmental information acquisition unit consists of an explosion-proof anemometer and rain gauge installed in the work area. The distributed tension sensing network unit includes a first tension sensor installed at the hook and multiple second tension sensors evenly distributed at preset intervals along the length of the suspension rope.

[0021] The data acquisition module consists of three specialized sub-units, each responsible for acquiring a type of physical quantity. The three work together to form the complete sensing front end of the system.

[0022] The multi-view visual acquisition unit consists of a network of multiple explosion-proof high-definition network cameras. The specific camera settings follow these principles: a top-down camera is installed below the main beam of the lifting beam to monitor the vertical projection position and attitude of the hook and the hoisted object; a side camera is installed at the rotating support of the lifting beam to monitor the lateral sway of the hoisted object; panoramic cameras are installed at fixed high points around the work area to monitor the relative position of the hoisted object to the surrounding environment and potential obstacle intrusion; all cameras achieve precise synchronization of exposure times through hardware synchronization signal lines or network-based time synchronization protocols to ensure that images captured from different angles are strictly aligned in time; each camera outputs an image stream with a timestamp accurate to milliseconds and a camera number for subsequent multi-view 3D reconstruction and spatial positioning calculations. The environmental information acquisition unit consists of an explosion-proof anemometer and rain gauge, both located in the work area and unobstructed by the equipment itself. The anemometer typically uses the ultrasonic time-of-flight method, simultaneously measuring the two-dimensional wind speed vector in the horizontal plane and outputting instantaneous wind speed scalar values, wind direction angle values, and average wind speed and gust wind speed calculated based on a sliding window. The rain gauge uses a tipping bucket or piezoelectric principle to continuously record the cumulative rainfall per unit time. The output data of this unit includes timestamps and all the aforementioned environmental parameters, providing a quantitative basis for the system to analyze the impact of environmental disturbances on the hoisting process. The distributed tension sensing network unit is configured as follows: a high-range first tension sensor is installed at the connection structure between the hook and the rope. The sensor adopts a spoke or column structure and directly measures the total load force borne by the hook, which serves as the total load reference value of the entire hoisting system. Multiple second tension sensors are installed at preset uniform intervals along the length of the suspension rope, forming a distributed sensing network. These second tension sensors are implemented by using miniaturized fiber Bragg grating sensors or miniature force-measuring pins based on strain gauge principles, embedded or attached to the strand gaps or surface of the suspension rope without damaging its main load-bearing structure. Each second tension sensor is calibrated at the factory, and its reading has a defined linear relationship with the axial tension of the suspension rope at its installation location. All tension sensors form a synchronous acquisition network via a fieldbus (such as a CAN bus or RS-485 bus), simultaneously latching their respective readings in each acquisition cycle to ensure that tension data across the entire rope length is captured at the same time. Each sensor has a unique node number in the bus network, which maps to the sensor's installation position on the suspension rope; therefore, the tension reading at each node naturally corresponds to the local tension value at different locations on the suspension rope. In actual lifting operations, the distributed tension sensing network unit operates on the following principle: When the hook bears a load, the lifting rope is tensioned under the load. The tension value at different locations on the rope varies due to differences in the rope's weight, friction, and degree of bending. Each second tension sensor measures the axial strain at its location in real time and converts it into an electrical or optical signal, which is then uploaded to the bus after analog-to-digital conversion. The system thus obtains a snapshot-like tension distribution map of the lifting rope, showing the tension at each location on the rope at a given moment. Through continuous data acquisition, the system can also obtain the tension change curves of each node over time, providing raw data support for subsequent dynamic stress analysis and stress concentration early warning.

[0023] Example 3: As one embodiment of the present invention, the structured perception results output by the scene perception module specifically include: the swing vector of the suspended object obtained based on visual image sequence analysis, the environmental disturbance contribution obtained by fusing environmental data and the swing vector, and the force distribution map and uniformity index of the suspension rope generated based on distributed tension data.

[0024] As one embodiment of the present invention, the prediction function of the risk assessment module includes, based on the current swing state of the suspended object, the trend of environmental wind speed and direction, and the change of force distribution on the suspension rope, using a time-series prediction model to make probabilistic predictions of collision risk, instability risk, and local overload risk of the suspension rope within a set future time period.

[0025] After receiving the structured perception results output by the scene perception module, the risk assessment module initiates a multi-level risk assessment process. The core design idea of ​​this module is that the safety risks of hoisting operations cannot be judged solely by the threshold of a single indicator, but must be comprehensively assessed from three dimensions: the current state, the evolution trend, and the potential consequences. The first level is rule-based real-time status assessment. The module has a pre-configured safety rule knowledge base, which encapsulates safety procedures and expert experience for coal mine hoisting operations in the form of logical judgment rules. For example, the knowledge base contains typical rules such as: triggering a local overload alarm when the tension value at any node of the hoisting rope exceeds a specific percentage of the rope's rated breaking strength; triggering an instability alarm when the swing amplitude of the hoisted object exceeds a set angle; and triggering an environmental risk alarm when the ambient wind speed exceeds a set threshold. This module matches the values ​​in the structured perception results with the conditions in the rule base in real time, assigns a corresponding risk weight to each triggered rule, and generates a preliminary real-time risk event list. The second level is a comprehensive situation assessment based on semantic association. This level is not satisfied with isolated threshold judgments, but focuses on the coupling effect between multiple risk factors. For example, when the system simultaneously detects three phenomena: a continuous increase in wind speed, an upward trend in the swing amplitude of the suspended load, and a decrease in the uniformity index of the force distribution on the hoisting rope, although each individual indicator may not exceed its respective alarm threshold, the superposition of the three means that the hoisting system is in a rapidly deteriorating dynamic process. This module can identify such multi-factor coupled risk patterns through built-in association analysis logic and integrate them into a higher risk level assessment. The third level is risk trend prediction based on time series forecasting. This module has a built-in lightweight time series forecasting model. This model takes the swing vector sequence, wind speed and direction sequence, and force distribution sequence within the current moment and a past time window as input, and makes probabilistic predictions of key state quantities within a set time period in the future. Specifically, the model will output multiple prediction results: predict the swing amplitude and direction of the suspended object several seconds later, and calculate the predicted distance between it and surrounding obstacles. When the predicted distance is less than the safety threshold, a collision warning is triggered; predict the trend of the tension value change at the maximum tension node of the suspension rope several seconds later, and trigger an overload warning when the predicted value approaches the safety boundary; predict the trend of the change of the attitude angle of the suspended object, and trigger an instability warning when the predicted attitude exceeds the safe operating attitude range. The predictions of these three dimensions together constitute the ability to perceive future risks in advance, enabling the system to take preventive measures before risks actually occur, rather than passively waiting for risks to occur before responding to emergencies.

[0026] Example 4: In one embodiment of the present invention, the anti-interference adjustment module includes: The disturbance rejection decision submodule is used to generate a sequence of compensatory motion control commands for actively suppressing swaying based on the contribution of the environmental disturbance and the sway vector of the suspended object. The adjustment decision submodule is used to generate dynamic adjustment instructions for adjusting the force transmission path of the suspension rope to balance the force on each node, based on the force distribution diagram and uniformity index of the suspension rope.

[0027] In one embodiment of the present invention, the execution control module includes: The drive unit is used to execute the compensatory motion control command sequence to control the rotation, lifting, and lateral movement of the lifting beam; The adjustment unit is used to execute the dynamic adjustment command and control the operation of the auxiliary adjustment mechanism connected to the suspension rope or hook to change the geometry or force distribution of the suspension rope.

[0028] After receiving the risk level and structured perception results output by the risk assessment module, the disturbance rejection adjustment module first makes a decision-making access judgment: the module only enters the active control mode when the comprehensive risk level is lower than the preset operable threshold; if the risk level reaches or exceeds the threshold, the module directly jumps to the safety protection mode and generates an emergency braking or hovering command. In active control mode, the two decision streams within the module work in parallel to solve two core control problems in hoisting operations: one is how to keep the hoisted object stable under environmental disturbances, and the other is how to ensure that the hoisting rope is subjected to uniform force in order to extend its service life. The disturbance rejection decision submodule is responsible for solving the first problem. This submodule uses the sway vector of the suspended object and the contribution of environmental disturbances from the structured perception results as its core inputs. Its decision logic is as follows: First, it determines what factors are primarily causing the sway of the suspended object. If the contribution of environmental disturbances exceeds a set value, it indicates that external wind force is the main cause of the sway, and then it enters the disturbance rejection compensation mode. In this mode, the submodule generates a compensatory motion command in the opposite direction to the disturbance based on the current wind speed and direction and the sway state of the suspended object. For example, under continuous crosswinds, it commands the main motion mechanism of the lifting beam to perform a slight reverse lateral movement to actively counteract the yaw effect of wind load on the suspended object. The essence of this feedforward compensation strategy is to intervene before the disturbance force is fully converted into the displacement of the suspended object, thereby suppressing the sway from the source, rather than passively damping and eliminating it after the sway occurs. If the contribution of environmental disturbances is low, it indicates that the current sway mainly originates from operational factors such as acceleration and deceleration of the lifting beam itself. Then, the submodule switches to a smooth control strategy, optimizing the speed curve of the planned path to reduce additional sway caused by sudden motion changes. The adjustment decision submodule is responsible for solving the second problem, namely optimizing the stress distribution of the suspension rope. This submodule uses the stress distribution heatmap and uniformity index of the suspension rope from the structured sensing results as its core input. Its decision logic is as follows: the module continuously monitors the uniformity index of the suspension rope tension distribution. When this index falls below a preset threshold, it indicates a significant stress concentration phenomenon in the suspension rope, meaning that the tension at some nodes is significantly higher than at other nodes. At this point, the submodule identifies the location of the sensor node with the highest tension and generates a dynamic adjustment command based on this. The specific content of this command is: instructing the auxiliary execution in the adjustment unit... The mechanism makes micro-motions to change the geometry of the lifting ropes or the force transmission path; for example, the command can control a movable guide pulley to make a precise micro-displacement to the side with greater force, redistributing the tension between rope segments by changing the wrap angle or included angle of the lifting ropes; or, for a system using multi-point suspension, the command can independently fine-tune the load-bearing ratio of a certain suspension point, thereby achieving a redistribution of the load among the lifting ropes; the essence of this dynamic adjustment process is to optimize the stress distribution inside the flexible load-bearing body (lifting ropes) by actively intervening in the force transmission path within the lifting system; The compensatory motion control command sequences and dynamic adjustment commands generated by the two decision flows are simultaneously sent to the execution control module. This module contains two independent drive units: the drive unit is responsible for controlling the three degrees of freedom of the lifting beam—rotation, lifting, and lateral movement. It receives disturbance-resistant motion commands and converts them into drive signals for the motors of each degree of freedom. The adjustment unit is responsible for controlling the auxiliary adjustment mechanism connected to the lifting rope or hook. It receives dynamic adjustment commands and converts them into drive signals for the corresponding adjustment motors. The two drive units execute their respective action commands synchronously within the same control cycle, so that macroscopic compensation of the suspended object's motion and microscopic adjustment of the lifting rope's force are completed simultaneously, jointly ensuring the stability of the lifting process under external disturbances and the safety of its internal mechanical state.

[0029] Example 5: As one embodiment of the present invention, the optimization module includes: The storage submodule is used to store raw multimodal sensing data, preprocessed feature datasets, structured perception results, risk assessment and prediction reports, and all issued control commands and actuator status feedback in a unified timeline, forming a panoramic closed-loop database. The optimization submodule is used to perform supervised incremental training on the machine learning model in the scene perception module using new data in the panoramic closed-loop database, and to update the control strategy parameters in the anti-disturbance adjustment module through reinforcement learning.

[0030] The optimization module consists of two tightly coupled sub-modules. The storage sub-module is the data foundation layer of this module, which is responsible for the complete recording and organization of the system's operating data. At the physical level, this submodule consists of a large-capacity industrial-grade solid-state storage array, which has high write speed and large storage capacity, and can meet the high-frequency data full storage needs for several days or even several weeks. At the logical level, this submodule continuously monitors and subscribes to all key data streams transmitted on the system's collaborative communication bus, and its data collection range covers every link in the entire chain of system operation. Specifically, for each system control cycle, this submodule fully records the following seven categories of data: The first category is the raw sensor data output by the data acquisition module, including but not limited to the raw image frames captured by each camera, the instantaneous wind speed and direction data output by the anemometer, the rainfall intensity data output by the rain gauge, and the synchronous readings of the first tension sensor and all second tension sensors in the distributed tension sensor network at the same time; the second category is the feature dataset output by the data fusion and preprocessing module after spatiotemporal alignment and preprocessing, including the standardized image feature map, environmental parameter vector, and tension distribution vector; the third category is the structured perception results output by the scene perception module, including the type and attitude information of the suspended object. The data includes: information, load swing vector, contribution of environmental disturbance, heat map of rope force distribution, and tension uniformity index; the fourth category is the comprehensive risk level, risk source list, and risk trend prediction results for each dimension output by the risk assessment module; the fifth category is all control commands output by the disturbance rejection adjustment module, including the complete content and issuance time of compensatory motion control command sequences and dynamic adjustment commands; the sixth category is the actual execution status fed back by the execution control module, including the actual position, speed, current feedback value of each drive motor, and actual displacement of the rope adjustment mechanism; the seventh category is the operator operation behavior recorded by the interaction module, including mode switching operations, decision confirmation or rejection operations, and emergency stop button trigger records; The seven types of data mentioned above are organized according to a unified timeline, with each complete hoisting operation as the basic storage unit, forming a complete time-series data block. Each data block not only records the system state evolution and control behavior sequence throughout the entire operation, but also includes the final execution result of the operation, such as whether the hoisting was successfully completed, whether a risk event occurred, the total operation time, and the maximum stress value of the hoisting rope, which can serve as evaluation labels. The unique value of this data organization method lies in its close correlation of four key pieces of information—what the system perceived, what the system decided, what the system executed, and the final result—within the same time series, making each piece of historical data a naturally labeled machine learning training sample. The data collection and storage behavior of this submodule is proactive and uninterrupted, and can be completed automatically without human intervention, ensuring the objectivity, integrity, and traceability of the data, providing a solid data foundation for accident retrospective analysis, system performance evaluation, and iterative training of machine learning models.

[0031] Example 6: As one embodiment of the present invention, the interactive module provides a graphical interface for real-time visualization of the swing trajectory of the suspended object, environmental wind field information, cloud map of the force distribution of the suspension rope, comprehensive risk level and prediction trend, and system decision logic, and is equipped with an interface for switching between automatic / manual control modes and manual intervention.

[0032] The interaction module serves as a bridge connecting the intelligent system and the human operator. Its design goal is to enable the operator to understand the system's judgments, trust the system's decisions, and take over control when necessary. This module achieves the above goals through a graphical cockpit interface, which is laid out according to the information hierarchy of situational awareness → decision understanding → action authorization. At the situational awareness level, the interface provides a multi-window video and data fusion display area. The main window of this area overlays camera footage labeled with visual recognition algorithms. Specific labels include: the type and bounding box of the hoisted object, the real-time position tracking point of the hook, the predicted trajectory of the hoisted object's swing, and warning signs for automatically identified obstacles and personnel. In the sidebar of the video window, the interface displays the rope tension distribution in a real-time updated color cloud map, using a gradient from green to red to map the readings of each tension node, allowing operators to easily identify stress concentration areas. Furthermore, the interface visualizes current wind speed and direction data in the form of a dynamic wind field streamline diagram, transforming abstract environmental parameters into intuitive graphical representations. At the decision understanding level, the interface includes a decision log and explanation panel. This panel displays the decisions the system is executing and their rationale in real-time using natural language text. For example: "Northeast wind level 3 detected, wind disturbance contribution 78%, execute windward pre-compensation lateral movement or rope node 4 tension exceeds average by 32%, execute pulley displacement adjustment." This decision interpretability mechanism is key to building human-machine trust. Operators do not need to understand the details of the underlying control algorithm, but can understand why the system does what it does by reading the logic expressed in natural language. When the system issues a risk warning, the panel will also highlight the specific risk source that triggered the warning with a bright color and flashing icon, along with a brief explanation of the predicted trend, such as: "It is expected that the swing amplitude will reach 18 degrees in 5 seconds, posing a risk of collision with the eastern support." At the action authorization level, the interface prominently displays a control mode switch and an emergency stop button. Control modes include three switchable states: fully automatic, decision suggestion, and manual. In fully automatic mode, the system operates completely autonomously, with the operator only responsible for monitoring. In decision suggestion mode, the system generates a decision plan and displays an execution suggestion button on the interface; the system executes the corresponding command only after the operator confirms it. In manual mode, the system degenerates into a pure information enhancement tool; all control commands are manually issued by the operator via a joystick or button panel, but situational awareness functions such as visual annotations, risk alarms, and force cloud maps remain operational, providing auxiliary information for the operator's manual operations. The emergency stop button employs a dual redundancy design of physical hardware and software, independent of the system's main control flow, ensuring that the operator can directly cut off power output via a physical button in any abnormal situation, achieving a forced safe shutdown. Through the information organization and functional design of the above three levels, the interaction module ensures the transparency and trustworthiness of the system during fully automated operation, while also providing flexible options for the allocation of responsibilities for operators of different levels and preferences, thus leveraging the system's intelligent advantages while retaining the ultimate safety guarantee of human supervision.

[0033] Example 7: A method for intelligent control of a rotating hoisting beam in coal mine transportation includes the following: S1: Through the data acquisition module, visual images of the work site, environmental wind speed and direction, and tension data of multiple nodes of the hook and rope are collected simultaneously; S2: Through the data fusion and preprocessing module, the data collected by S1 is synchronized, fused and preprocessed in time to form a multimodal feature dataset; S3: Through the scene perception module, the built-in machine learning model is used to analyze the multimodal feature dataset and output a structured perception result that integrates the swing of the suspended object, environmental disturbances and the force distribution of the suspension rope; S4: Based on the structured perception results, the risk assessment module performs real-time comprehensive risk assessment and short-term risk trend prediction. S5: Based on the results of S4, the disturbance rejection adjustment module collaboratively generates disturbance rejection motion control commands for actively suppressing the swing of the suspended object, as well as dynamic adjustment commands for optimizing the uniformity of the force on the suspension rope. S6: By executing the control module, the main motion mechanism and the suspension rope adjustment mechanism are driven to coordinately execute the instructions generated by S5; S7: By optimizing the storage submodule in the module, store the full-link data of this control cycle to the panoramic closed-loop database; S8: Through the optimization submodule in the optimization module, based on the new data in the panoramic closed-loop database, the machine learning model in S3 and the instructions in S5 are iteratively optimized.

[0034] The online optimization submodule for machine learning models is the data analysis and evolution component of the optimization module. This submodule automatically starts during system idle periods (such as work breaks, nighttime, or non-working windows) without requiring manual triggering by the operator. After starting, the submodule first reads the data records added since the last optimization from the storage submodule and performs quality screening on these new data, removing invalid records caused by sensor failures or communication anomalies, and retaining only complete, reliable, and valid samples. Subsequently, the submodule starts two independent optimization tasks in parallel, iteratively improving the perception system and the decision-making system respectively. The first optimization task targets the machine learning model in the scene perception module. The underlying principle of this optimization is supervised incremental learning. Specifically, the optimization submodule extracts the original multimodal features (i.e., fused preprocessed feature data) from the newly added data in the panoramic closed-loop database as model input, and simultaneously extracts the objective results of this task as supervision labels. The objective result labels are constructed as follows: if the task is completed successfully without any risk events, the result is marked as a successful sample, and the corresponding perception result is considered correct; if the task triggers a risk alarm or an emergency stop event occurs, the result is marked as a risk sample, and the corresponding perception result is considered correct. Known results are considered edge cases that require focused learning; if the operator makes a manual correction during the operation, the operator's correction is considered an indirect negation of the system's perception or decision at that time, and the sample is marked as a sample to be corrected; the optimization submodule merges these newly added labeled samples with the original training data, fine-tunes the existing model, and updates the model's network weight parameters; through this continuous learning mechanism, the model can gradually adapt to new types of hoisting objects, new ambient lighting conditions, new climate conditions, and edge scenarios that were not fully covered in the previous training data, and its recognition accuracy and generalization ability steadily improve with the accumulation of running time; The second optimization task targets the control strategy parameters in the disturbance rejection adjustment module. The technical principle of this optimization is strategy search and parameter optimization based on historical data, and its implementation borrows from the offline policy evaluation concept in reinforcement learning. Specifically, the optimization submodule treats each complete hoisting operation as a decision round, calculating a comprehensive performance score as a reward signal for that round. This comprehensive performance score is a weighted composite multi-objective function, including the following four evaluation indicators: operation completion time (reflecting operation efficiency, shorter time, higher score), peak risk level during the operation (reflecting safety, lower risk, higher score), and the uniformity index of the hoisting rope force throughout the entire operation. The average value during the operation (reflecting equipment wear and tear; better uniformity results in a higher score) and total energy consumption (reflecting economic efficiency; lower energy consumption results in a higher score) are analyzed. The optimization submodule extracts a large amount of historical operation data from the panoramic closed-loop database and analyzes the correlation pattern between the control strategy parameter combination adopted by the system in each round and the comprehensive performance score of that round. Based on this correlation analysis, the submodule uses a parameter search algorithm to find the strategy parameter configuration that maximizes the comprehensive performance score. The better parameter configuration found is automatically applied to subsequent hoisting operations, so that the system's control strategy continuously approaches Pareto optimality in four dimensions: safety, efficiency, equipment life, and economy. After the two optimization tasks are completed, the optimization submodule generates an optimization report, recording the number of samples optimized, the changes in model performance indicators, and the adjustment values ​​of strategy parameters. The report is then pushed to the interactive module for operators to view. Through the closed-loop mechanism of panoramic data recording and dual-path model optimization, this invention realizes a complete self-evolutionary cycle from data acquisition → model training → on-site setup → new data acquisition → model retraining. The system continuously accumulates experience and optimizes decisions during continuous operation, and its intelligence level continues to improve over time, becoming more intelligent with use.

[0035] Example 8: As one embodiment of the present invention, generating dynamic adjustment instructions in S5 specifically includes: analyzing the force distribution diagram of the suspension rope in the structured perception result, identifying specific sensor nodes whose tension exceeds the average load setting ratio or absolute safety threshold; calculating the required suspension rope shape fine adjustment amount based on the node position and overall distribution, and generating an instruction sequence to drive the corresponding adjustment mechanism to transfer part of the load of the high-load node to the low-load node.

[0036] The generation of dynamic adjustment commands is a data-driven, progressively refined decision-making process. Its core lies in transforming the abstract diagnosis of uneven force distribution into specific, executable mechanism adjustment parameters. Step 1: Receiving and preprocessing force distribution data; The adjustment decision submodule extracts the force distribution map dataset of the suspension rope from the structured sensing results. This dataset contains the tension values ​​corresponding to the node numbers along the suspension rope. The submodule first verifies these raw tension data, removing obvious outliers caused by sensor failures or communication errors. Then, it calculates the following three key statistics: the average tension of the entire rope, the percentage deviation of the tension at each node from the average value, and the overall tension uniformity index (defined as the ratio of the standard deviation of the tension value at each node to the average value). When the uniformity index is lower than a preset threshold, the dynamic adjustment program is triggered. Step 2: Identification and location of the node with the highest stress; After confirming the existence of uneven stress, the submodule traverses the tension data of all nodes and identifies the node number with the highest tension value and its corresponding tension value; Since the number of each sensor node is precisely mapped to its actual physical position on the suspension rope, the identified node number with the highest tension can be directly matched to the specific position on the suspension rope—for example, the position of sensor number 4 located near the end of the suspension rope close to the hook; At the same time, the submodule also identifies the node position with the lowest tension value as the target reference area for subsequent load transfer; Step 3: Generation of adjustment strategy and calculation of fine-tuning amount; Based on the location of the maximum stress node and the degree of tension deviation, the submodule selects an applicable adjustment scheme from the preset adjustment strategy library; When the maximum stress point is located in the area of ​​the lifting rope close to the hook, it indicates that the stress concentration is caused by the force line convergence effect at the hook connection point. At this time, the strategy is to adjust the position of the balance pulley or adjustable guide wheel above the hook to change the wrap angle and force line direction of the lifting rope at the hook connection point, thereby transferring part of the load to other sections of the rope; The specific implementation of this strategy is: control the balance pulley to move precisely along the preset guide rail in the opposite direction to the maximum stress node (i.e., the side with smaller tension). The displacement amount is proportional to the current deviation percentage. The more severe the deviation, the greater the pulley displacement amount; When the maximum stress point is located in the middle section of the lifting rope, it indicates that the stress concentration may be caused by the local bending effect at the contact point between the lifting rope and the pulley group. At this time, the strategy is to fine-tune the height or lateral position of the corresponding guide pulley on the lifting beam to change the bending radius and contact pressure of the lifting rope at that point; Step 4: Encapsulation and transmission of dynamic adjustment commands; The calculated adjustment strategy and fine-tuning amount are encapsulated according to the command format required by the execution control module to generate a complete dynamic adjustment command; This command contains three elements: target actuator identifier (specifying which adjustment motor or actuator is moving), direction of movement and displacement, and execution speed or acceleration parameters; This command is sent to the adjustment unit in the execution control module through the system cooperative communication bus, and the unit converts it into a specific drive signal for execution; After the adjustment command is executed, the system will re-collect the tension distribution data in the next control cycle to evaluate the adjustment effect; If the uniformity of force is improved but not to the ideal state, the system can continue to issue fine-tuning commands in subsequent cycles to form a closed-loop iterative adjustment; If the uniformity index deteriorates after adjustment, the system has a rollback mechanism that can cancel the current adjustment command and restore the state before adjustment to prevent risks caused by erroneous adjustment.

[0037] To facilitate understanding, each module of this invention will be described in detail and illustrated with examples; Principles of each module: The data acquisition module, which is the physical sensing layer, is responsible for the synchronous acquisition of raw data and directly determines the accuracy and real-time performance of the upper-layer cognition. It includes the following units: Multi-view visual acquisition unit: Composed of multiple high-resolution explosion-proof network cameras networked together; installation points have been precisely simulated and calculated to ensure coverage of the following key visual domains: Full-range view of suspended object: At least two cameras observe the suspended object from different angles for 3D attitude calculation; Hook tracking view: A dedicated camera closely monitors the hook to track its relative position to the lifting point and the details of its swing. Environmental monitoring perspective: Monitor fixed and moving obstacles along the lifting beam's running path; Body status perspective: Monitor the arrangement of the wire rope on the drum and the key weld areas of the structure; All cameras achieve frame-level synchronization through hardware synchronization signals or precision network clock protocols, and the output image streams are accompanied by timestamps accurate to the millisecond level. Environmental information acquisition unit: Typically installed at a high, open point in the work area; includes wind speed and direction sensors, usually employing ultrasonic principles, capable of instantaneously measuring three-dimensional wind speed vectors, not just scalar values; and rain sensors recording cumulative rainfall intensity; this unit outputs structured data packets at a fixed frequency (e.g., 10Hz), containing data such as instantaneous wind speed, average wind speed, wind direction angle, and rainfall intensity; quantifying the dynamic input disturbance of the external natural environment on the large-scale suspended object of the hoisting system; The distributed tension sensing network unit includes: The first tension sensor (total load sensor) is a high-precision spoke-type or shear beam type sensor, installed on the upper support beam of the hook assembly; the second tension sensor group is integrated into the wire rope in an array; a flexible encapsulated fiber optic grating sensor string is embedded into the core gap of the wire rope or fixed parallel to the outside of the strands using a specific process. The sensor spacing is non-uniform or uniform depending on the length of the rope and the key areas of concern (such as near the end, bending section); each sensor node has a built-in microprocessor, which can perform local analog-to-digital conversion and temperature compensation, and realize the synchronous acquisition and uploading of data from all nodes through industrial wireless networking; a rope tension distribution spectrum is generated in real time, directly reflecting the load transmission state and stress concentration on the flexible rope; Data fusion and preprocessing module: For each input image frame, each packet of environmental data, and each tension sample, the engine dynamically compensates and aligns them based on the deviation between their built-in timestamps and the global clock. Data with different sampling rates (e.g., 30Hz for images, 100Hz for tension) are unified to a series of discrete, synchronous fusion time points through interpolation or sampling. Simultaneously, using pre-calibrated sensor spatial position parameters, the tension sensor readings are bound to their position coordinates on the 3D suspension rope model, transforming the target position identified in the image from the pixel coordinate system to the hoisting operation coordinate system associated with the world coordinate system. For visual images, the process is as follows... Execution: Defogging and Low-Light Enhancement: Applying physically based image restoration technology to improve image contrast and detail visibility in dimly lit and dusty underground environments; Dynamic Region of Interest (ROI) Extraction: Based on the recognition results of the previous moment, adaptively adjusting the digital pan-tilt or image cropping area of ​​each camera to focus computing resources on key targets and improve processing efficiency; For environmental and mechanical data, the following are performed sequentially: Outlier Removal: Using statistical methods (such as the 3σ principle based on moving windows) to remove obviously erroneous transient jump data; Smoothing Filtering: Applying low-pass filters to data such as wind speed and tension to retain true trend changes and filter out high-frequency measurement noise; After preprocessing, all data corresponding to each fusion time point are packaged into a standardized multimodal feature tensor. This tensor contains: the deep semantic features of the image feature map extracted by a lightweight convolutional network, rather than the original pixels, the normalized environmental parameter vector, and the tension distribution vector composed of all tension sensor readings. The multimodal feature tensor is a unified language that downstream machine learning models can directly process. The scene perception module is essentially a multimodal fusion machine learning model trained on massive amounts of operational data. It achieves the mapping from low-level data features to high-level semantic understanding; the model employs a multi-task learning framework. The input is the aforementioned multimodal feature tensor; the model internally has a shared feature extraction layer and multiple dedicated output heads for specific tasks; the model outputs the following structured perception results in parallel: Visual analysis results include fine classification of hoisted objects such as coal mining machine drums and hydraulic supports, accurate 2D bounding boxes, and 3D attitude estimation based on key point detection, including pitch, yaw, roll angle, and pixel-level position of the hook. Motion state variables: By comparing the visual analysis results of consecutive frames, the model calculates the swing vector of the suspended object in three-dimensional space, including instantaneous linear velocity, angular velocity, swing trajectory and dominant frequency; Disturbance attribution analysis: Receives the current sway vector and real-time environmental wind speed and direction, and outputs a quantitative evaluation value based on its internally learned dynamic prior knowledge. This value is used to determine the extent to which the current sway is caused by wind disturbance and the extent to which it is caused by the acceleration and deceleration of the lifting beam, etc. Mechanical State Map: Directly processes the raw tension distribution vector from the distributed tension sensing network unit, and outputs a more advanced mechanical state description through a dedicated sub-network, including: tension distribution heatmap, visualization of stress magnitude at each point on the rope, location of the maximum stress point, tension uniformity index, calculation of the variance or Gini coefficient of the entire distribution curve, and abnormal fluctuation pattern recognition, such as identifying tension fluctuations at a specific frequency, which may be related to some kind of mechanical resonance; The key to achieving the complex output mentioned above lies in the deep cross-modal correlations learned during its training process. For example, it learned that when the suspended object in the image is tilted at a specific angle and the wind direction is from one side, the sensor readings on the corresponding side of the tension distribution heatmap will show characteristic gradient changes. This cognitive ability enables the system to have a unified and quantitative understanding of the complex phenomenon of uneven force caused by wind blowing on the suspended object. The risk assessment module has an embedded layered, configurable rule base; First layer: Real-time threshold alarm; based on simple logical judgment, such as: if any tension node reading is greater than 85% of the breaking tension, any overload risk event is triggered; The second layer: situational correlation assessment; based on more complex combinations of conditions, such as: if the swing amplitude is >15° and the predicted collision time is <5 seconds, and the wind disturbance contribution is >0.7, then a high wind load collision risk event is triggered. The third layer: time series trend prediction; using a lightweight time series prediction model, such as a network based on LSTM units, the model takes into account the perception results of the swing vector, wind speed, force distribution, etc. within the current and a short historical time window, and makes probabilistic predictions for key states in the next 3-10 seconds, such as swing angle, distance to the nearest obstacle, and maximum tension value; for example, it outputs that after 3 seconds, the predicted distance between the suspended object and the right side of the tunnel wall has a 70% probability of being less than the safety threshold of 1 meter. Risk Integration and Decision Triggering: Based on the combined assessment results of the three layers mentioned above, a quantitative comprehensive risk level is generated, for example, a score of 0-100 or four levels: low, medium, high, and critical, along with a detailed list of risk sources and prediction details. This risk level directly serves as the master switch for subsequent decision-making modules: under low risk, the system executes an efficient operation strategy; under medium to high risk, the system executes a conservative or protective strategy; under critical risk, an emergency shutdown is immediately triggered. The disturbance rejection module translates safety constraints and efficiency objectives into specific control commands; internally, two decision streams run in parallel: Disturbance-resistant motion decision flow: This decision flow is responsible for generating instructions to control the rotation, movement, and lifting of the lifting beam's main mechanism; it is activated when the system is in operation and the risk is controllable; it receives the target position of the suspended object and the current status provided by the sensing module, including position, sway, and wind disturbance contribution; it contains a motion planner and an disturbance-resistant controller. The motion planner plans a collision-free trajectory in space based on the current position, the target position, and a real-time updated map of environmental obstacles. The core of the disturbance rejection controller is to generate feedforward compensation commands. For example, when a stable crosswind (with a high contribution of wind disturbance) is detected, the controller will superimpose a small, continuous reverse lateral movement speed command on the planned trajectory. The magnitude of this command is proportional to the empirical model of the current wind speed and the wind-received area of ​​the suspended object. The purpose is to actively cancel the influence of wind disturbance before it causes displacement, thereby suppressing the swaying from the source, rather than waiting for the swaying to occur and then damping it to eliminate it. Dynamic adjustment decision flow: This decision flow is a design feature of this invention specifically for the safety characteristics of the suspension rope; it is independent of the main motion decision and specifically handles the force distribution diagram of the suspension rope; its decision logic is: Monitoring and analysis: Continuously monitor the tension uniformity index and the location of the maximum stress point; Strategy generation: If the uniformity index is found to be consistently below the threshold (e.g., 0.8), or the tension at a certain node remains high for an extended period, the decision flow is initiated; it generates dynamic adjustment commands, such as: instructing the adaptive balance pulley above the hook to move a distance ΔL towards the side with higher tension, or instructing a certain electric lifting point in the multi-point spreader to fine-tune the lifting amount; Control objective: The purpose of this instruction is not to change the spatial position of the suspended object, but to change the internal force transmission path by fine-tuning the geometry or force distribution mechanism of the suspension rope system, so that the load can be more evenly distributed to multiple parts of the rope, thereby reducing local stress peaks, delaying fatigue, and preventing the risk of breakage due to local damage. The execution control module is responsible for translating high-level instructions into low-level drive signals; it includes the following units. Drive unit: Receives a sequence of instructions from the disturbance-resistant motion decision flow, such as rotating to angle α while moving to coordinate X, with the lifting speed maintained at V; contains servo drives or frequency converters for each degree of freedom; internally, it typically employs classic PID position / speed loop control, but the received setpoints are already ideal values ​​after disturbance-resistant compensation; this unit ensures that large mechanisms execute complex motions accurately and smoothly. The adjustment unit is an independent and precise auxiliary actuator that receives instructions from the dynamic adjustment decision flow. For example, it may control a precision trolley driven by a servo motor, which can move laterally on a fixed track to change the position of the hoisting rope guide point; or it may control an intelligent hook with multiple hydraulic cylinders working together to independently fine-tune the length of each hoisting point. This unit has a small amplitude of action and a fast response. Its actions are superimposed on the main motion unit in time and coupled with the spatial effect to jointly achieve the dual goals of sway suppression and load equalization. The optimization module consists of two closely cooperating parts: Storage submodule: In the form of a ring buffer or database, it continuously and synchronously records the entire link information from raw sensor data to the final execution state; each record in the stored panoramic closed-loop database contains: timestamp, raw / preprocessed data snapshot, structured sensing results, risk assessment report, all issued control commands, and actual feedback values ​​of each actuator (such as actual position, current); these data are organized according to work tasks, and key events (such as successful wind-resistant operation, abnormal emergency stop) can be marked manually or automatically; The optimization submodule is an offline or low-priority background task that periodically (e.g., daily or nightly) scans the panoramic closed-loop database to find new data samples. For the model in the scene perception module, the optimization sub-module uses supervised learning to perform incremental updates. It takes the original features in the new data as input and the results of manual post-calibration or the more accurate state derived through multi-sensor information fusion as new teacher signals to fine-tune the model, so that it has a better ability to recognize and understand new types of hoisting objects and new environmental conditions. For the strategy parameters in the disturbance rejection adjustment module, the optimization submodule can use the reinforcement learning approach for evaluation and optimization. It treats each hoisting operation as a round, and uses operation efficiency (time), safety indicators (no risk events), energy consumption, etc. as reward functions. By analyzing a large amount of historical round data, it automatically adjusts the control parameters in the decision flow (such as disturbance rejection compensation coefficient and uniformity adjustment threshold) to optimize the system strategy towards better overall performance. The interactive modules include: The interface provides operators with a fused panoramic view through a high-resolution display; the interface may include: video footage overlaid with dynamic recognition boxes and swing trajectory prediction lines; real-time 3D scene reproduction, displaying virtual models of the suspended object, lifting beam, and obstacles; dynamically updated color tension cloud map of the lifting rope; dynamic flow field visualization of wind speed and direction; and prominently marked comprehensive risk levels and prediction warnings. The system not only displays what to do, but also explains why; for example, when the system starts to perform wind disturbance lateral shift, the interface will pop up a prompt: "Continuous easterly wind level 3 detected. In order to suppress the predicted oscillation, an eastward pre-compensation lateral shift is being performed." It offers multiple modes, including fully automatic, semi-automatic, and manual. In fully automatic mode, the system operates completely autonomously. In semi-automatic mode, the system provides decision suggestions, which are then implemented after confirmation by the operator. In manual mode, the system only provides enhanced perception information (such as visual guidance and risk alarms), and control is completely handed over to the operator. This ensures that the final control of the system remains in the hands of the staff in complex and unknown situations. Take, for example, a typical high-risk scenario of installing large equipment next to an open-pit mine; The operator selects the fully automatic mode on the terminal of the interactive module and sets the task: to lift the large crusher rotor (weighing 80 tons) located at point A on the edge of the mine pit to the installation foundation at point B at the bottom of the pit; the system passes the self-test and enters the ready state; Phase 1: Initial Lifting Stage Data Acquisition: The data acquisition module is activated; the visual camera locks onto the suspended load; the environmental sensor displays a wind speed of 1.5 m / s (Level 1 wind) with a stable wind direction; the distributed tension network shows that the tension at each node is rising steadily, with a total load of 80 tons and even distribution. Fusion cognition: The data fusion and preprocessing module aligns the data; the scene perception module identifies the target as the crusher rotor, with a horizontal attitude, a near-zero sway vector, low wind disturbance contribution, and a tension uniformity index of 0.95; the output status is normal. Risk assessment: The risk assessment module evaluates the risk as low.

[0038] Decision-making and execution: The disturbance rejection adjustment module's disturbance rejection motion decision flow plans the path for vertical lifting to a safe height, and the dynamic adjustment decision flow remains silent due to uniform force; the execution control module smoothly executes the lifting action; Monitoring: The operator sees the load rising steadily on the monitoring screen, with all parameters showing green; Phase Two: Unexpected Situations Occur When the load is raised to a height of 50 meters, the airflow at the pit opening becomes turbulent. Data perception: Environmental sensors showed that the wind speed suddenly increased to 8 m / s (level 5 wind), and the wind direction was turbulent and changeable; visual analysis showed that the suspended object began to swing periodically in the east-west direction, and the amplitude rapidly increased to 10 degrees; tension network data showed that the tension fluctuation of the sensor node group on the west side intensified, and the peak value was significantly higher than that of the nodes on the east side. The reasons for the fusion analysis and summary are as follows: The scene perception module quickly updated the structured results: the swing vector (amplitude 12 degrees, frequency 0.8 Hz, east-west direction), the wind disturbance contribution jumped to 0.9, the rope stress distribution map showed stress concentration on the west side, and the uniformity index dropped to 0.6; Risk Prediction and Alarm: Based on the current severe swaying and wind speed trend, the risk assessment module uses a time series model to predict that the swaying amplitude may reach 20 degrees in 5 seconds, causing the distance between the suspended load and the maintenance platform on the east side to be less than 2 meters (the safe distance is 5 meters); the module immediately raises the comprehensive risk level to high and issues a predictive collision risk alarm. Intelligent dual-path decision generation: The anti-disturbance motion decision flow was triggered by high risk and immediately intervened; it was calculated that the current main wind disturbance came from the west, causing the suspended object to swing eastward; the decision flow generated a compensatory motion command: control the lifting beam trolley to move eastward (in the windward direction) at a speed of 0.1 m / s in a small and continuous lateral movement, and adjust the descent path to move it away from the east platform; The dynamic adjustment decision flow is simultaneously triggered by uneven force distribution; after analyzing the force diagram, a dynamic adjustment command is generated: control the adaptive balance pulley above the hook to move 3 cm to the west (the side with greater tension) to improve the tension balance on both sides of the rope; Effect feedback: The drive unit and adjustment unit of the execution control module act simultaneously; the operator sees on the screen that the swaying trend of the suspended load in the wind is effectively curbed, and the amplitude changes from the predicted increase to a decrease; at the same time, the color of the tension distribution cloud map gradually changes from dark red (high stress) on the west side to orange-yellow (medium stress), and the uniformity index recovers. Process Recording: The storage submodule of the optimization module completely records the data of the entire process of coordinated stability control under this sudden wind disturbance, and marks it as a successful high-risk handling case; Phase Three: Post-event Learning and Intelligent Optimization After completing the day's tasks, the optimization submodule is launched; it retrieves the data from this case study. The model of the scene perception module was fine-tuned using a series of data on strong winds, large swaying, uneven force, and successful regulation, making it more sensitive to swaying and force correlation patterns under similar wind conditions in the future. The control effect of the combined action of the anti-disturbance motion decision flow and the dynamic adjustment decision flow was analyzed, and the compensation algorithm parameters and adjustment logic were optimized to make the response faster and more accurate when encountering similar situations in the future.

[0039] 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 illustrative of the principles of 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 control system for a rotating hoisting beam in coal mine transportation, characterized in that, include: The data acquisition module is used to simultaneously collect visual images of the work area, environmental wind speed and direction, and distributed tension data of multiple nodes of the hook and rope; The data fusion and preprocessing module is connected to the data acquisition module and is used to perform time synchronization and preprocessing on multi-source heterogeneous data to generate a multimodal feature dataset. The scene perception module is connected to the data fusion and preprocessing module. It has a built-in trained machine learning model for analyzing the multimodal feature dataset and outputting structured perception results that fuse visual swing analysis, environmental disturbance assessment and rope force distribution. The risk assessment module, connected to the scene perception module, is used to perform real-time risk quantification and trend prediction based on the structured perception results. The disturbance rejection adjustment module is connected to the risk assessment module and the scene perception module respectively. It is used to collaboratively generate compensatory motion control commands for actively suppressing the swing of the suspended object and dynamic adjustment commands for optimizing the uniformity of the force on the suspension rope when the risk is controllable. An execution control module, connected to the disturbance rejection adjustment module, is used to coordinate the execution of the motion control command and the dynamic adjustment command; The optimization module connects the data fusion and preprocessing module and the execution control module. It is used to store data of the entire operation process to form a panoramic closed-loop database, and to iteratively optimize the machine learning model and control strategy based on the database. The interaction module connects the risk assessment module and the anti-interference adjustment module, and is used for information visualization and human interaction.

2. The intelligent control system for a rotating hoisting beam in coal mine transportation according to claim 1, characterized in that, The data acquisition module includes: The multi-view visual acquisition unit consists of multiple explosion-proof cameras installed around the lifting beam and the work area, used to acquire real-time images of the suspended object, hook, rope and environment. The environmental information acquisition unit consists of an explosion-proof anemometer and rain gauge installed in the work area. The distributed tension sensing network unit includes a first tension sensor installed at the hook and multiple second tension sensors evenly distributed at preset intervals along the length of the suspension rope.

3. The intelligent control system for a rotating hoisting beam in coal mine transportation according to claim 1, characterized in that, The structured perception results output by the scene perception module specifically include: the swing vector of the suspended object obtained based on visual image sequence analysis, the environmental disturbance contribution obtained by fusing environmental data and the swing vector, and the force distribution map and uniformity index of the suspension rope generated based on distributed tension data.

4. The intelligent control system for a rotating hoisting beam in coal mine transportation according to claim 3, characterized in that, The prediction function of the risk assessment module includes using a time-series prediction model to make probabilistic predictions of collision risk, instability risk, and local overload risk of the suspension rope within a set future time period, based on the current swing state of the suspended object, the trend of environmental wind speed and direction, and the changes in the force distribution of the suspension rope.

5. The intelligent control system for a rotating hoisting beam in coal mine transportation according to claim 1, characterized in that, The disturbance rejection adjustment module includes: The disturbance rejection decision submodule is used to generate a sequence of compensatory motion control commands for actively suppressing swaying based on the contribution of the environmental disturbance and the sway vector of the suspended object. The adjustment decision submodule is used to generate dynamic adjustment instructions for adjusting the force transmission path of the suspension rope to balance the force on each node, based on the force distribution diagram and uniformity index of the suspension rope.

6. The intelligent control system for a rotating hoisting beam in coal mine transportation according to claim 5, characterized in that, The execution control module includes: The drive unit is used to execute the compensatory motion control command sequence to control the rotation, lifting, and lateral movement of the lifting beam; The adjustment unit is used to execute the dynamic adjustment command and control the operation of the auxiliary adjustment mechanism connected to the suspension rope or hook to change the geometry or force distribution of the suspension rope.

7. The intelligent control system for a rotating hoisting beam in coal mine transportation according to claim 1, characterized in that, The optimization module includes: The storage submodule is used to store raw multimodal sensing data, preprocessed feature datasets, structured perception results, risk assessment and prediction reports, and all issued control commands and actuator status feedback in a unified timeline, forming a panoramic closed-loop database. The optimization submodule is used to perform supervised incremental training on the machine learning model in the scene perception module using new data in the panoramic closed-loop database, and to update the control strategy parameters in the anti-disturbance adjustment module through reinforcement learning.

8. The intelligent control system for a rotating hoisting beam in coal mine transportation according to claim 1, characterized in that, The interactive module provides a graphical interface for real-time visualization of the swing trajectory of the suspended object, environmental wind field information, cloud map of the force distribution of the suspension rope, comprehensive risk level and prediction trend, as well as system decision logic. It also has an interface for switching between automatic / manual control modes and manual intervention.

9. A method for intelligent control of a rotating hoisting beam in coal mine transportation, characterized in that, An intelligent control system for a coal mine transport rotating lifting beam, as described in any one of claims 1-8, comprises the following: S1: Through the data acquisition module, visual images of the work site, environmental wind speed and direction, and tension data of multiple nodes of the hook and rope are collected simultaneously; S2: Through the data fusion and preprocessing module, the data collected by S1 is synchronized, fused and preprocessed in time to form a multimodal feature dataset; S3: Through the scene perception module, the built-in machine learning model is used to analyze the multimodal feature dataset and output a structured perception result that integrates the swing of the suspended object, environmental disturbances and the force distribution of the suspension rope; S4: Based on the structured perception results, the risk assessment module performs real-time comprehensive risk assessment and short-term risk trend prediction. S5: Based on the results of S4, the disturbance rejection adjustment module collaboratively generates disturbance rejection motion control commands for actively suppressing the swing of the suspended object, as well as dynamic adjustment commands for optimizing the uniformity of the force on the suspension rope. S6: By executing the control module, the main motion mechanism and the suspension rope adjustment mechanism are driven to coordinately execute the instructions generated by S5; S7: By optimizing the storage submodule in the module, store the full-link data of this control cycle to the panoramic closed-loop database; S8: Through the optimization submodule in the optimization module, based on the new data in the panoramic closed-loop database, the machine learning model in S3 and the instructions in S5 are iteratively optimized.

10. The intelligent control method for a rotating hoisting beam in coal mine transportation according to claim 9, characterized in that, The specific steps for generating dynamic adjustment commands in S5 include: analyzing the force distribution diagram of the suspension rope in the structured perception results, identifying specific sensor nodes whose tension exceeds the average load setting ratio or absolute safety threshold; calculating the required suspension rope shape fine-tuning amount based on the node's location and overall distribution, and generating a command sequence to drive the corresponding adjustment mechanism to transfer part of the load from the high-load node to the low-load node.

Images