Intelligent medium adding method suitable for coal preparation plant and computer equipment

Abstract

The application discloses an intelligent medium adding method and computer equipment suitable for a coal preparation plant, and the method comprises the following steps: cameras are symmetrically arranged on the top center of a medium warehouse and the two side walls in the moving direction of a cross beam of a head sheave, images are synchronously shot and transmitted to an AI controller for preprocessing; when a medium adding operation instruction is received by an industrial intelligent agent, the industrial intelligent agent analyzes the track of the head sheave and the characteristics of a medium pile in the images, associates the track and the characteristics with a position instruction sequence of a common controller, and forms a VLC dynamic association model; based on the model, an optimal suction point and a collision-free shortest path are planned, and the AI controller converts the path into a pulse width instruction to control the movement of the head sheave; after the head sheave grabs the medium and sends the medium into a dense medium barrel, a control system determines the amount of clean water according to the weight of the medium, controls a clean water pipe and an air pipe to inject water and blow air into the barrel, and forms a heavy medium suspension coal preparation. The method can realize medium pile identification, intelligent scheduling of the head sheave and precise medium adding, and solves the problems of low efficiency, poor precision and insufficient reliability of traditional manual medium adding.

Description

Technical Field

[0001] This application relates to the field of coal technology, and in particular to an intelligent medium addition method and computer equipment suitable for coal preparation plants. Background Technology

[0002] Heavy media coal preparation is a method of separating coal using a liquid with a density between that of clean coal and gangue (or middlings) as the medium. It offers advantages such as high separation efficiency, a wide feed particle size range, and ease of operation. The principle of heavy media coal preparation is that clean coal, with a density lower than the medium, floats, while gangue or middlings, with a density higher than the medium, sinks, and are then collected and classified into different products. Therefore, an important aspect of heavy media coal preparation is the preparation of the heavy media suspension. Heavy media suspensions are generally prepared by mixing particulate solid weighting agents with water. Due to the coal's own carrying capacity for the weighting agents, weight loss is unavoidable, necessitating the continuous addition of new media during the production process.

[0003] Currently, coal preparation plants generally employ dedicated personnel to add media, prepare heavy media suspensions, and transport heavy media suspensions. These tasks are primarily performed manually, resulting in significant consumption of manpower and resources. Furthermore, manual operation suffers from poor stability, low media addition efficiency, and an inability to meet production demands in a timely manner, hindering efficient separation. Summary of the Invention

[0004] In view of this, this application provides an intelligent medium addition method and computer equipment suitable for coal preparation plants, which can realize medium pile identification, intelligent crane scheduling and precise medium addition, and solve the problems of low efficiency, poor accuracy and insufficient reliability of traditional manual medium addition.

[0005] According to one aspect of this application, a smart medium-addition method suitable for coal preparation plants is provided, applied to a control system, the control system including a conventional controller, an AI controller, and a medium-addition industrial intelligent agent, the method comprising:

[0006] A top panoramic camera is deployed at the center of the top of the media storage facility, and side wall dynamic tracking cameras are symmetrically deployed on both sides of the media storage facility along the direction of movement of the overhead crane beam. This allows the top panoramic camera and the side wall dynamic tracking cameras to capture images synchronously and transmit the captured images to the AI ​​controller for preprocessing. The camera range of the top panoramic camera covers the central axis area of ​​the overhead crane beam and the entire media stack area in the media storage facility.

[0007] When the additive industrial intelligent agent receives the additive operation command, it uses computer vision technology to analyze the visual features of the crane's movement trajectory and the medium stack shape in the image preprocessed by the AI ​​controller. It then associates the analyzed visual features with the position command sequence issued by the ordinary controller controlling the crane, forming a VLC dynamic association model that includes the mapping relationship between vision, position, and command. The position command sequence is obtained by the additive industrial intelligent agent from the ordinary controller controlling the crane.

[0008] The industrial intelligent agent of Jiajie plans the optimal suction point and the collision-free shortest path of the suction medium based on the VLC dynamic correlation model and feeds it back to the AI ​​controller. The AI ​​controller converts the collision-free shortest path into a pulse width command and controls the ordinary controller of the crane based on the pulse width command, so that the crane moves to the optimal suction point along the collision-free shortest path after being driven by the ordinary controller. The crane has a corresponding suction cup.

[0009] The control system controls the overhead crane to reach the optimal suction point via its ordinary controller. The suction cup then draws the medium from the medium pile and sends it into the thickening tank. The control system determines the required amount of clean water based on the weight of the medium added to the thickening tank. It also controls the flow of the required amount of clean water into the thickening tank via the ordinary controller of the clean water pipe and the blowing of air into the thickening tank via the ordinary controller of the air duct to form a heavy medium suspension for coal preparation. The thickening tank is equipped with a clean water pipe above it and an air duct inside the thickening tank. Each of the overhead crane, clean water pipe, and air duct has its own ordinary controller.

[0010] According to another aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described intelligent medium addition method applicable to coal preparation plants.

[0011] By means of the above technical solution, this application provides an intelligent medium addition method and computer equipment suitable for coal preparation plants, which can realize medium pile identification, intelligent crane scheduling and precise medium addition, and solve the problems of low efficiency, poor accuracy and insufficient reliability of traditional manual medium addition.

[0012] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0013] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0014] Figure 1 A schematic flowchart of an intelligent media addition method for coal preparation plants provided in an embodiment of this application is shown.

[0015] Figure 2 This illustration shows a schematic diagram of a media library architecture provided in an embodiment of this application;

[0016] Figure 3 A schematic flowchart of another intelligent media addition method applicable to coal preparation plants provided by an embodiment of this application is shown;

[0017] Figure 4 This illustration shows a schematic diagram of an AI controller control flow provided in an embodiment of this application;

[0018] Figure 5 This illustration shows a schematic diagram of the association between a general controller and an AI controller provided in an embodiment of this application;

[0019] Figure 6 This illustration shows another schematic diagram of the association between a conventional controller and an AI controller provided in an embodiment of this application.

[0020] Figure 7 This illustration shows a flowchart of another intelligent media addition method for coal preparation plants provided in an embodiment of this application.

[0021] 1-Proximity switch; 2-Overhead crane; 3-Weighing device; 4-Suction cup; 5-Pneumatic agitator valve; 6-Air duct; 7-Clear water flow meter; 8-Clear water valve; 9-Clear water pipe; 10-Transfer pump flow meter; 11-Transfer pump; 12-Grate; 13-Level gauge; 14-Concentrated medium tank; 15-Medium stack; 16-Left side wall dynamic tracking camera; 17-Top panoramic camera; 18-Right side wall dynamic tracking camera. Detailed Implementation

[0022] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0023] This embodiment provides an intelligent medium-addition method suitable for coal preparation plants, applied to a control system. The control system includes a conventional controller, an AI controller, and an industrial intelligent agent for medium addition, such as... Figure 1 As shown, the method includes:

[0024] Step 101: Deploy a top panoramic camera at the center of the top of the media storage facility, and symmetrically deploy side wall dynamic tracking cameras on both sides of the media storage facility along the direction of movement of the overhead crane beam, so that the top panoramic camera and the side wall dynamic tracking cameras can capture images synchronously and transmit the captured images to the AI ​​controller for preprocessing. The camera range of the top panoramic camera covers the central axis area of ​​the overhead crane beam and the entire media stack area in the media storage facility.

[0025] In the above embodiments of this application, such as Figure 2 As shown, three cameras are installed on the top and side walls, respectively, and the data from the cameras is connected to the AI ​​controller of the control system. The top panoramic camera is installed in the center of the top of the media storage tank, covering the entire media stack distribution area and the central axis area of ​​the overhead crane beam. It is mainly used to capture the global outline of the media stack, the media height distribution, and the macroscopic position of the overhead crane. For the side wall dynamic tracking cameras, two are installed on the two side walls of the media storage tank along the direction of the overhead crane beam, focusing on the local details of the media stack (such as areas rich in magnetic materials and the compaction state of the media), forming a complementary view with the top camera, eliminating the blind spots caused by the overlap of the suction cup and the media stack from the top view.

[0026] Optionally, the AI ​​controller includes an edge computing unit. In step 101, the AI ​​controller performs preprocessing, including:

[0027] Step 101-1: The edge computing unit uses a bilateral filtering algorithm to preprocess the captured image.

[0028] In the above embodiments of this application, the edge computing unit first preprocesses the images captured by the three cameras to eliminate interference from the industrial environment. A bilateral filtering algorithm is used to filter high-frequency noise caused by coal dust, while retaining key features such as the suction cup edge and the outline of the dielectric stack.

[0029] Step 102: When the additive industrial intelligent agent receives the additive operation command, the additive industrial intelligent agent uses computer vision technology to analyze the visual features of the crane's movement trajectory and the medium stack shape in the image preprocessed by the AI ​​controller, and associates the analyzed visual features with the position command sequence issued by the ordinary controller controlling the crane to form a VLC dynamic association model containing the mapping relationship between vision, position and command. The position command sequence is obtained by the additive industrial intelligent agent from the ordinary controller controlling the crane.

[0030] Next, when the industrial intelligent agent receives the operation command, it analyzes the visual features of the crane's global coordinate trajectory (crane movement trajectory) and the three-dimensional contour of the dielectric stack from the image preprocessed by the AI ​​controller through multi-frame spatiotemporal alignment, feature point matching, and coordinate transformation. Simultaneously, it collects the position command sequence from the crane's ordinary controller. The visual features are then associated with the corresponding commands to form a dynamic correlation model of "vision-position-command," realizing the modeling of the correspondence between the three. Specifically, computer vision technology can be used to accurately select the two core targets, the "crane" and the "dielectric stack," from a complex industrial background, clarifying their specific position coordinates in the image. Quantifiable key features are extracted, such as the crane's operating speed and hook position, and the dielectric stack's height, volume, and surface flatness.

[0031] Optionally, the industrial intelligent agent includes a lightweight U-Net semantic segmentation model. In step 102, the industrial intelligent agent utilizes computer vision technology to analyze the visual features of the crane's motion trajectory and the dielectric stack morphology in the image preprocessed by the AI ​​controller, including:

[0032] Step 1021: The industrial intelligent agent uses the lightweight U-Net semantic segmentation model to perform pixel-level classification on the image preprocessed by the AI ​​controller, obtaining the suction cup region, media stack region, ground region, and crane region.

[0033] Step 1022: The industrial intelligent agent adds a three-dimensional contour of the dielectric stack in the dielectric stack region using the Otsu threshold contour extraction algorithm, and determines the visual features of the dielectric stack morphology based on the three-dimensional contour.

[0034] Step 1023: In the crane area, the industrial intelligent agent transforms the pixel-level positional changes of the crane into a continuous motion trajectory in physical space through the spatiotemporal correlation of multiple frames of images, and extracts the visual features of the crane's motion trajectory based on the transformed continuous motion trajectory.

[0035] In the above embodiments of this application, the industrial intelligent agent uses computer vision (CV) technology to extract and reconstruct features from images captured by a camera, constructing a static model of the suction cup and the media stack. A lightweight U-Net semantic segmentation model is used to perform pixel-level classification on the images preprocessed by the AI ​​controller, labeling the "suction cup region," "media stack region," "ground region," and "crane region." The Otsu threshold contour extraction algorithm is used to obtain the three-dimensional contour of the media stack, which is used to determine the optimal target location for media acquisition.

[0036] Furthermore, the industrial intelligent agent inputs the preprocessed image into a lightweight improved U-Net model (which reduces computation through depthwise separable convolution and channel pruning). The model extracts features through the encoder and restores spatial resolution through the decoder, outputting a pixel-level classification map that accurately divides the four regions: suction cup, media stack, ground, and crane.

[0037] The industrial intelligent agent first uses the Otsu thresholding algorithm to automatically determine the binarization threshold for the segmented media stack region and extracts the two-dimensional contour of the media stack. Then, it combines multi-view images from the top and side wall cameras and reconstructs the three-dimensional contour of the media stack through multi-view stereo matching (such as SfM structured light). Finally, it extracts morphological visual features (such as stack height, volume, surface undulation, stack shape (conical / flat top)) from the three-dimensional contour.

[0038] The industrial intelligent agent first converts the pixel coordinates of multiple frames of images into physical coordinates through camera extrinsic calibration for the segmented overhead crane area; then it uses optical flow / target tracking algorithms (such as SORT) to associate key feature points of the overhead crane in continuous frames (such as beam endpoints and suction cup hooks) to convert pixel-level position changes into continuous motion trajectories in physical space (such as global XYZ coordinate sequences); finally, it extracts visual features (such as motion speed, direction, acceleration, trajectory shape (straight line / curve)) from the trajectory.

[0039] The entire process achieves end-to-end visual analysis of "image segmentation-region feature-oriented extraction", providing a foundation for path planning and instruction association in subsequent intermediary operations.

[0040] More specifically, the core logic of Jiajie's industrial intelligent agent in parsing the visual features of motion trajectories from the semantically segmented "overhead crane region" is to transform the "pixel-level positional changes" of the overhead crane into "continuous motion trajectories in physical space" through the spatiotemporal correlation of multiple frames of images, and extract the dynamic features of the trajectory (such as speed and direction). The specific process can be divided into the following 5 key steps, and is designed in conjunction with the actual constraints of the industrial scenario (such as the directionality of the overhead crane's movement along the beam):

[0041] 1. Spatiotemporal alignment of multi-frame image sequences:

[0042] The movement of the overhead crane is a continuous time process, and a single frame image cannot show its trajectory. Therefore, the Jiajie industrial intelligent agent first sorts the image data of the continuous frames by timestamp, and performs spatiotemporal calibration based on the extrinsic parameters of the cameras (the calibrated absolute coordinates). For example, the frame data of the top panoramic camera and the side wall tracking camera will be matched with "multi-view images under the same timestamp" to eliminate the positional deviation caused by the difference in viewpoints, and ensure that the position of the overhead crane area in the continuous frames has temporal consistency.

[0043] 2. Dynamic target tracking in the overhead crane area:

[0044] Semantic segmentation only completes the "pixel labeling of the overhead crane region in a single frame," but it needs to lock onto the same overhead crane in consecutive frames (to avoid confusing other equipment or background). Jiajie Industrial Agent uses a lightweight target tracking algorithm (such as a simplified version of SORT or DeepSORT, adapted to the computing power constraints of industrial edge computing) to treat the "overhead crane region" in each frame as the tracking target, through:

[0045] Appearance feature matching (such as the color and shape of the crane, extracting HOG or CNN features from the segmented region).

[0046] Motion model prediction (such as Kalman filtering, based on the crane's historical motion direction / velocity, to predict the position range of the next frame);

[0047] Achieve stable tracking of the crane region in consecutive frames and output the "time series of the bounding box of the crane region" (the position and size of the crane in the image for each frame).

[0048] 3. Extraction and matching of key feature points of the overhead crane:

[0049] To accurately analyze the trajectory, stable physical feature points (rather than the entire bounding box of the entire area) can be extracted from the crane area. These points satisfy the requirements of "fixed parts of the crane structure" (such as the endpoints of the crane beams or the centers of the suction cup hooks), avoiding trajectory errors caused by the crane's own deformation (such as suction cup lifting). Specific operations:

[0050] For the "crane region" of each frame, stable features such as corner points and edge points are extracted using the ORB feature detector (lightweight and suitable for real-time processing);

[0051] By using optical flow methods (such as LK optical flow) or feature point descriptor matching (such as ORB's BRIEF descriptor), the same feature point in consecutive frames can be associated (for example, the "left end of the overhead crane beam" in frame t corresponds to the same position in frame t+1), forming a "time series trajectory of feature points" (such as (x1, y1, t1)-(x2, y2, t2)-...).

[0052] 4. Conversion from pixel coordinates to global physical coordinates:

[0053] The pixel positions in an image lack physical meaning. Multi-view triangulation (combining extrinsic parameters from top and side-wall cameras) is needed to convert the pixel coordinates of feature points into physical coordinates in a global coordinate system (i.e., the actual spatial coordinates of the coal preparation plant, such as the length-width-height of the media storage area corresponding to the XYZ axes). For example:

[0054] The top panoramic camera captures the crane's "top-down position" (XY plane coordinates);

[0055] The side-view tracking camera captures the crane's "side-view height" (Z-axis coordinate);

[0056] Combining the two yields a three-dimensional physical coordinate time series of the crane's feature points (e.g., (X1, Y1, Z1, t1) - (X2, Y2, Z2, t2) - ...), which is the key transformation of the trajectory "from visual to physical".

[0057] 5. Fitting motion trajectories and extracting dynamic features:

[0058] The "three-dimensional coordinate time series" obtained through the above steps is further processed by the industrial intelligent agent:

[0059] Trajectory fitting: Using polynomial fitting or Bézier curves to connect discrete coordinate points into a continuous trajectory curve (such as the linear motion trajectory of an overhead crane along a beam, or the up-and-down motion trajectory when gripping a medium).

[0060] Dynamic feature extraction: Calculate the first derivative (velocity) and second derivative (acceleration) of the trajectory to obtain the motion state features of the crane (such as the moving speed along the crossbeam and the acceleration of the suction cup descending); at the same time, extract the directional features of the trajectory (such as whether the crane is moving to the left or right of the medium pile).

[0061] The final analyzed "visual features of the overhead crane's motion trajectory" are essentially a combination of "continuous path in physical space" and "dynamic motion state," specifically including:

[0062] Trajectory path: The movement route of the overhead crane in the global coordinate system (such as a linear trajectory along the X-axis of the beam, or the Z-axis lifting trajectory during grabbing).

[0063] Motion parameters: the crane's moving speed, acceleration, and direction (e.g., "moving uniformly to the right along the crossbeam at a speed of 0.5 m / s");

[0064] Spatiotemporal correlation: the correspondence between trajectory and time (e.g., "the crane arrives directly above the medium pile at t=10s").

[0065] Since the movement of the overhead crane is limited by the mechanical constraints of the crossbeam (it can only move along the crossbeam direction), the industrial intelligent agent can further improve the accuracy of trajectory analysis by using prior knowledge constraints (such as "the Y-axis coordinate of the overhead crane is fixed, and only the X-axis and Z-axis change") to filter out abnormal points in the image (such as false feature points caused by coal dust obstruction).

[0066] In summary, the core of Jiajie Industrial Intelligent Agent's analysis of the crane's motion trajectory is to transform the "continuous changes in visual images" into the "motion laws of physical space." This step lays the foundation for the subsequent "association between visual features and ordinary controller position commands" (such as the command X=5m corresponding to point (5, Y, Z) in the trajectory), ensuring that the model can understand the mapping relationship between "command-trajectory-position".

[0067] Specifically, when the industrial intelligent agent performs a pre-planned task, that is, when it receives industrial operation instructions, such as... Figure 3 As shown, Figure 3 The intelligent agent in this system, also known as the intermediary industrial intelligent agent, is used by operators to send the command "Start intermediary production and intermediary to intermediary system A" through the interactive interface. The intermediary industrial intelligent agent first determines whether intermediary system A is available. If it is not available, it displays "System in use"; if it is available, it records the status "sys_free=0" and displays "System preparing".

[0068] The industrial intelligent agent plans tasks according to the operator's instructions, prompts "Task planning in progress", and specifies the total amount of medium to be added as x tons and the single suction cup carrying y tons.

[0069] The industrial intelligent agent calculates the estimated number of media scavenging operations, n (rounded up), and indicates "Multiple media scavenging operations are expected, with a total of n operations".

[0070] The industrial intelligent agent uses real-time images to identify the profile of the medium stack, the position of the overhead crane, and the position of the concentrated medium tank. It then performs path planning and sends instructions to the AI ​​controller, displaying the message "Path planning in progress for the No. 1".

[0071] The industrial intelligent agent calls the AI ​​controller to drive the overhead crane from the initial position to the medium suction position. After the suction cup picks up the medium, it moves to the medium concentration position to release the medium and displays "No. Medium suction completed".

[0072] The industrial intelligent agent determines whether all absorbing tasks have been completed (No≤n). If not, it continues to execute task 3. Figure 3 (Note the text in the original). If it has been completed, proceed with the next task.

[0073] The industrial intelligent agent calls the ordinary control system to open the clean water valve so that clean water flows into the concentrate tank, and opens the wind-driven stirring valve. When the level gauge on the tank wall detects that the liquid level is lower than the critical value, the control system closes the clean water valve and displays the message "Concentrate preparation in progress".

[0074] The industrial intelligent agent for adding media calls the ordinary control system to control the delivery pump and the adding valve to deliver the mixed medium to the heavy medium A system. When the level gauge on the wall of the concentrated medium tank detects that the liquid level is lower than the critical value, the control system stops the delivery pump, closes the wind-powered stirring valve and the adding valve, and prompts "Concentrated medium has been added to the heavy medium system".

[0075] The industrial intelligent agent records the status "sys_free=1" and prompts "System interface addition complete", indicating the end of the process.

[0076] Optionally, in step 102, the industrial intelligent agent associates the parsed visual features with the sequence of position commands issued by the ordinary controller controlling the overhead crane, including:

[0077] Step 1024: The industrial intelligent agent, based on the principle of multi-view triangulation, captures the pixel coordinates of the same feature point from the images captured simultaneously by the top panoramic camera and the side wall dynamic tracking camera. Combined with the calibrated extrinsic parameters of the top panoramic camera and the side wall dynamic tracking camera, the coordinates of the feature point in the global coordinate system are calculated until the visual features of the crane movement trajectory and the dielectric stack morphology parsed in the image are unified in the global coordinate system.

[0078] Step 1025: The industrial intelligent agent associates the visual features after unifying coordinates with the position command sequence issued by the ordinary controller controlling the overhead crane.

[0079] In the above embodiments of this application, the Jiajie industrial intelligent agent, based on the principle of multi-view triangulation, calculates the coordinates of the same feature point in the global coordinate system by combining the pixel coordinates of the same feature point captured by the three cameras with the calibrated extrinsic parameters of the top panoramic camera and the extrinsic parameters of the side wall dynamic tracking camera (absolute coordinates of the camera), and then fits the best absorption and collision-free shortest path.

[0080] Specifically, the real-time coordinates of the overhead crane in the global coordinate system (X, Y, Z) are calculated using a multi-view triangulation algorithm and denoted as:

[0081] ,

[0082] For multiple cameras, the least squares method is used to solve for the intersection points of each ray. The error formula is:

[0083] ,

[0084] If the error exceeds the threshold, the feature points are re-extracted and recalculated.

[0085] Step 103: The industrial intelligent agent plans the optimal suction point and the collision-free shortest path for the suction medium based on the VLC dynamic correlation model and feeds it back to the AI ​​controller. The AI ​​controller converts the collision-free shortest path into a pulse width command and controls the ordinary controller of the crane based on the pulse width command, so that the crane moves along the collision-free shortest path to the optimal suction point after being driven by the ordinary controller. The crane has a corresponding suction cup.

[0086] Next, through the VLC dynamic correlation model, the intelligent agent in the substrate addition industry can analyze the substrate pile morphology (such as pile height and density distribution) and the current position of the overhead crane in real time, accurately planning the optimal suction point (such as areas with dense substrate and stable structure), avoiding empty suction or substrate collapse caused by improper selection of suction point, and reducing repetitive operations. The planned collision-free path is directly associated with the physical coordinates of the overhead crane, and can be combined with the RRT algorithm to generate the shortest feasible path, reducing the overhead crane movement time and improving the efficiency of a single substrate addition cycle.

[0087] During the path planning phase, environmental obstacles (such as other equipment and walls) are incorporated into the grid map, resulting in a naturally collision-free path that avoids the collision risks caused by environmental changes in traditional control. The AI ​​controller monitors the crane's position in real time through a vision system. If a path deviation is detected (such as due to mechanical errors or external interference), the pulse width command is immediately adjusted to correct the direction of movement, ensuring that the crane always moves along the planned path and improving operational stability. From optimal pick-up point planning to path generation and command conversion, the entire process is automatically completed by the Jiajie industrial intelligent agent and AI controller, eliminating the need for manual path design or real-time monitoring and reducing labor costs.

[0088] Precise pulse width control makes the crane start-up, acceleration, and deceleration processes smoother, avoiding the impact of sudden stops or overshoots on the motor and transmission mechanism, and extending the service life of the equipment.

[0089] The VLC dynamic correlation model can update the media stack morphology and crane position information in real time. Even if the media stack changes shape due to media addition / removal, the system can still quickly adjust the suction point and path to adapt to complex working conditions. The accumulated "visual feature-position-command" mapping data can feed back into model training, continuously improving the accuracy and efficiency of path planning, and providing a foundation for subsequent expansion to multi-crane collaboration or more complex scenarios. Precise suction point and path control ensures stable media addition each time, avoiding density fluctuations in the concentrated medium tank suspension caused by uneven media distribution, thereby ensuring the stability of heavy media separation in the coal preparation process. Collision-free paths and closed-loop control reduce the crane failure rate, minimize production interruptions caused by collisions or path errors, and improve the continuity of the overall coal preparation process.

[0090] To this end, through the deep integration of visual perception, model association and closed-loop control, the process of adding media has been transformed from "human experience-driven" to "data intelligence-driven". It is superior to traditional control methods in terms of efficiency, safety, automation level and process stability, and provides a highly reliable and low-cost intelligent solution for industrial scenarios such as coal preparation.

[0091] Optionally, in step 103, the industrial intelligent agent plans the optimal absorption point of the absorption medium based on the VLC dynamic correlation model, including:

[0092] Step 1031: The industrial intelligent agent, based on the VLC dynamic correlation model, determines all candidate suction points that the suction cup may move to.

[0093] Step 1032: The industrial intelligent agent uses a multi-objective optimization cost function to select the optimal extraction point from the candidate extraction points that minimizes the optimization cost. The multi-objective optimization cost function is:

[0094]

[0095] P represents the candidate extraction point determined based on the VLC dynamic correlation model, with its real-time coordinates in the global coordinate system (X, Y, Z). The function f(p) is the multi-objective optimization cost function, and argminf(p) is the independent variable P that minimizes the function f(p). This is the Euclidean distance from the current position of the suction cup to the optimal suction point. The magnetic content of the medium at the optimal absorption point. The media height at the optimal absorption point , These are the distance weighting coefficient, the medium quality weighting coefficient, and the medium height weighting coefficient, respectively. .

[0096] In the above embodiments of this application:

[0097] ,

[0098] ,

[0099] In the above formulas, the Euclidean distance formula is:

[0100] ,

[0101] This formula is used to calculate the spatial distance from the current position of the suction cup (crane) to the candidate suction point (its core function is to measure the "movement cost"; the closer the distance, the higher the crane's moving efficiency), P current Let x be the three-dimensional coordinates of the suction cup's current position in the global coordinate system, and let x be the corresponding coordinate components. current y current , z current ),

[0102] Formula for magnetic material content:

[0103] ,

[0104] This formula is used to calculate the magnetic content of the medium at the candidate absorption point P (its core function is to evaluate the quality of the medium; the higher the magnetic content, the better the medium sorting effect). The meanings of the parameters are as follows:

[0105] m(P) represents the content of magnetic material in the medium at the candidate absorption point P (unit: such as mass percentage, %).

[0106] m i The magnetic content of the i-th sampling point pi in the dielectric stack (i.e., the pre-detected sampling point quality data);

[0107] p i (or p) j ) represents the three-dimensional coordinates of the i-th (or j-th) sampling point in the media stack in the global coordinate system;

[0108] d(P, p) i (or d(P, p)) j The candidate sampling point P is the distance from the i-th (or j-th) sampling point p. i (or p) j The three-dimensional Euclidean distance (unit: such as millimeters);

[0109] m is the total number of sampling points in the media pile (i.e., the number of samples involved in the calculation).

[0110] i / j is the index variable for the summation operation (traversing all sampling points from 1 to m).

[0111] Therefore, the Euclidean distance formula measures the "cost" of crane movement through spatial distance. The closer the distance, the shorter the crane movement time and the lower the energy consumption, corresponding to the "distance weight" in multi-objective optimization. The magnetic material content formula calculates the magnetic material content of candidate points through "distance-weighted averaging." Sampling points closer to the candidate points have higher weights (because 1 / d(P, p) i (Increases as distance decreases) to ensure that the calculation results are closer to the actual medium quality of the candidate points, corresponding to the "medium quality weight" (corresponding to β) in multi-objective optimization.

[0112] Specifically, argmin is an abbreviation for "the independent variable that minimizes the function" (full name: argument of the minimum). For example, given a function f(x), argminf(x) is the x value that minimizes f(x). In the formula above, the independent variable P is the real-time coordinate (X, Y, Z) of the candidate suction point in the global coordinate system, which is "all possible positions the suction cup can move to". The function f(P) is a multi-objective optimization "cost function", also known as a "comprehensive evaluation function", and its formula is:

[0113] ,

[0114] Each of these items corresponds to the actual needs of coal preparation plants for adding media:

[0115] d(P, Pcurrent): Euclidean distance from the current position to the target position (the greater the distance, the longer the movement time, the higher the energy consumption, and the greater the cost).

[0116] 1-m(P): The "non-magnetic content" at the target location (m(P) is the magnetic content; the more magnetic material, the better the medium quality, so the smaller 1-m(P), the lower the cost).

[0117] h(P): The height of the medium at the target location (if the material level (i.e., the medium) is too high or too low, it may affect the gripping efficiency and needs to be controlled within a reasonable range. If the height is not appropriate, the cost will increase).

[0118] α, β, γ: Weighting coefficients (totaling 1), used to balance the importance of the three objectives of "distance, medium quality, and medium height" (for example, if medium quality is valued more, the weight of β is increased).

[0119] Therefore, argminf(P) = the position that minimizes the sum of "distance cost + medium quality cost + medium height cost". The intelligent agent for medium addition finds the optimal medium-grabbing position through multi-objective optimization—a position that simultaneously satisfies the comprehensive requirements of "short moving distance, high magnetic content in the medium, and suitable medium height," making the medium addition process most efficient and best suited to the coal preparation process.

[0120] For example, if a suction cup (or overhead crane) is currently at position A, and there are positions B (close but with few magnetic materials), C (with many magnetic materials but far away), and D (with moderate distance and magnetic materials, and suitable material level), then argminf(P) is the position (e.g., D) that minimizes f(P) among B, C, and D, and is considered the optimal suction point. In other words, argminf(P) is the mathematical notation for "selecting the optimal position," and its core principle is to use a weighted comprehensive score to find the most cost-effective gripping point.

[0121] Optionally, the overhead crane only performs orthogonal motion along the X, Y, and Z axes. In step 103, the industrial intelligent agent plans the collision-free shortest path, including:

[0122] Step 1033: The industrial intelligent agent plans the shortest collision-free path based on the improved path optimization algorithm, following the principle of prioritizing the far axis and then the near axis.

[0123] In the above embodiments of this application, the industrial intelligent agent, taking into account the characteristic that the suction cup (overhead crane) can only move orthogonally along the X / Y / Z axes (left-right, forward-backward, up-down movement), first determines the optimal suction point based on the three-dimensional contour of the medium stack, then uses an improved path optimization algorithm to plan the shortest collision-free path, decomposes the motion sequence according to the principle of "far axis first, then near axis", synchronously coordinates the early descent and low-speed stopping of the Z axis, and dynamically adjusts the axis speed in combination with the load and adopts acceleration and deceleration to balance efficiency and safety.

[0124] Optionally, the overhead crane has servo motors, each corresponding to an X, Y, and Z axis. Each axis has a preset pulse distance conversion coefficient, and the overhead crane has preset maximum acceleration and maximum speed. The AI ​​controller converts the collision-free shortest path into pulse width instructions, including:

[0125] Step 1034: The AI ​​controller decomposes the collision-free shortest path into the target distances that need to be moved in the X-axis, Y-axis, and Z-axis directions respectively.

[0126] Step 1035: The AI ​​controller calculates the number of target pulses required for each axis of the servo motor to reach the target distance based on the preset pulse distance conversion coefficient.

[0127] Step 1036: The AI ​​controller determines the speed curve of the crane at different stages based on the collision-free shortest path, the crane's preset maximum acceleration, and the preset maximum speed.

[0128] Step 1037: The AI ​​controller calculates the pulse width modulation duty cycle corresponding to each stage based on the speed curve of the crane at different stages.

[0129] In step 1038, the AI ​​controller encapsulates the calculated target pulse number and the corresponding pulse width modulation duty cycle into a pulse width instruction.

[0130] In the above embodiments of this application, such as Figure 4 As shown, Figure 4 The traditional controller, also known as the ordinary controller, constructs a hierarchical control architecture. The ordinary controller realizes the basic movement (forward, backward, up and down) of the suction cup (crane) through DO pulse commands. The AI ​​controller calculates the pulse width correction value in real time, dynamically adjusts the DO pulse duty cycle according to the position deviation, load and motor temperature, and compensates for the deviation by combining pulse feedback and CV position data closed-loop verification. Compared with the pure ordinary controller, the positioning accuracy is improved, the energy consumption is reduced, and it is compatible with traditional hardware, which greatly reduces the cost of implementation and transformation.

[0131] Specifically, when the suction cup moves to a critical node, the switch is triggered and outputs the absolute reference coordinates, denoted as:

[0132] ,

[0133] The adaptive Kalman filter fusion algorithm achieves weighted fusion of CV data and switch data by dynamically adjusting the observation noise covariance, as shown in the following formula:

[0134] ,

[0135] in, H is the observation vector (CV or switch data), and H is the observation matrix (for extracting position components). To observe the noise, The state vector contains the position and velocity information of the suction cup in the global coordinate system, and fully describes the motion state of the system at time k.

[0136] The observation matrix H can be represented as:

[0137]

[0138] In the prediction phase of Kalman filtering, the state vector It can be represented as:

[0139] ,

[0140] in, It is the optimal estimated state at time k-1. Let be the state transition matrix.

[0141] ,

[0142] in, This represents the sampling time interval.

[0143] A multi-view CV camera (corresponding to the three cameras mentioned above) acquires local images of the target point P. Color-related features are extracted from the images, including the mean and variance of the HSV color space (magnetic materials are typically dark gray, while non-magnetic materials are lighter). Texture features include the energy and entropy value of the gray-level co-occurrence matrix (GLCM) (magnetic materials have rougher surfaces).

[0144] For orthogonal motion constraints (motion only along the axis):

[0145]

[0146] Where g(n) is the actual distance from the starting point to the current grid cell.

[0147] The "segment merging" smoothing technique removes redundant inflection points from the path, resulting in a simplified path composed of straight lines. First, the X-axis is moved to... Then execute the Y-axis movement to Simultaneously initiate the Z-axis descent to achieve coordinated connection between horizontal and vertical movements.

[0148] Calculate the linear velocities in the x and y directions based on the physical displacement and effective time:

[0149] , ,

[0150] When the device leaves the factory, the corresponding relationship between the calibrated pulse and the physical displacement (pulse equivalent q, unit: pulse / mm) has been determined, that is, for every q pulses output by the motor, the actuator moves 1 mm in the x or y direction. This parameter can be regularly calibrated through the CV system (for example, by controlling the motor to output a fixed number of pulses and using vision to measure the actual displacement to calculate the q value by inverse calculation).

[0151] , ,

[0152] The pulse speed (unit: pulse / second) reflects the number of pulses output per unit time, and it is reciprocal to the pulse period (unit: second / pulse).

[0153] , ,

[0154] The pulse width P (i.e., the duration of a single pulse) is jointly determined by the pulse period and the preset duty cycle. The duty cycle (D) is the ratio of the pulse width to the pulse period (usually between 0 and 1), and its value needs to be determined according to the driving characteristics of the actuator (such as a motor). (For example, to avoid overheating of the motor, the duty cycle may be set to 0.5, that is, the pulse width is half of the period). Therefore, the pulse width P:

[0155] , ,

[0156] If the CV system detects a deviation between the actual speed and the theoretical speed of the suction cup (such as v_actual < v_theoretical), the AI controller will first adjust the pulse speed (such as increasing v x pulses), and then automatically update the pulse width (P x decreases as v x pulses increase, or by synchronously adjusting the duty cycle D x to maintain appropriate pulse energy).

[0157] Receive the "target pulse number" and "pulse width modulation duty cycle" from the AI controller; output pulses through the DO interface to drive the servo motor of the corresponding axis to operate, and at the same time, in real-time feedback the actual number of pulses sent and the motor operating status to the AI controller to form an execution closed-loop.

[0158] Dot-matrix machine switches have an inherent limitation: they cannot cover all positions. Because a dot matrix is ​​composed of discrete dots, there are gaps between adjacent dots, and the switch cannot directly reach positions within these gaps. Furthermore, inertia causes positional deviations during movement in dot-matrix machine switches. AI and CV technologies can eliminate these errors through precise image preprocessing. Using inter-frame differencing, the positional changes of the switch are compared across multiple consecutive frames to calculate its acceleration and rate of change. The AI ​​then uses this data to build an inertial model, predicting potential displacement due to inertia. Finally, image registration technology precisely aligns the preprocessed real-time image with a pre-defined dot-matrix standard image. Based on the alignment deviation, the AI ​​adjusts control commands accordingly, thus eliminating positional errors caused by inertia at the image level and providing a basis for precise control.

[0159] To address this, an AI controller is used to integrate multi-view CV full-domain continuous monitoring with high-precision positioning of key nodes and switch references. Through adaptive Kalman filtering, data weights are dynamically allocated. When approaching a node, the CV cumulative error is calibrated using switch data. The position is calculated by calling switch and encoder data, thus adapting to the dynamic monitoring needs of the media stack.

[0160] The AI ​​controller and the execution controller establish a real-time interactive link through an industrial bus. The execution controller performs parameter verification, trajectory interpolation, and drive control on the position, speed, and other commands issued by the AI, transforming them into equipment actions. The CV system provides real-time feedback on the actual position of the suction cup pair, which the AI ​​controller compares with the command position and encoder position. Through basic deviation correction, cumulative error elimination, and dynamic interference suppression, it achieves precise control. At the same time, the AI ​​pre-simulates the trajectory and self-tunes parameters based on the digital twin model, improving the accuracy of suction cup position control and significantly enhancing the accuracy and anti-interference capability of intelligent media addition in heavy media coal preparation plants.

[0161] Step 104: The control system controls the overhead crane to reach the optimal suction point via the ordinary controller of the overhead crane. The suction cup then draws the medium from the medium pile and sends it into the thickening tank. The control system determines the required amount of clean water based on the weight of the medium added to the thickening tank. The ordinary controller of the clean water pipe controls the flow of the required amount of clean water into the thickening tank via the clean water pipe, and the ordinary controller of the air duct controls the blowing of air into the thickening tank via the air duct to form a heavy medium suspension for coal preparation. A clean water pipe is installed above the thickening tank, and an air duct is installed inside the thickening tank. The overhead crane, the clean water pipe, and the air duct each have their own ordinary controller.

[0162] Optionally, in step 104, before the industrial intelligent agent receiving the mediating operation instruction, the method further includes:

[0163] Step 105: On the interactive interface of the additive industrial intelligent agent, the operator inputs the amount of additive and the re-addition system to be added via voice or text input. The additive industrial intelligent agent generates an additive operation command based on the amount of additive and the re-addition system to be added, and controls the AI ​​controller based on the additive operation command, so that the operation result of the AI ​​controller is fed back to the operator through the additive industrial intelligent agent. The additive industrial intelligent agent has an interactive interface.

[0164] Then, as Figure 5 and Figure 6 As shown, Figure 6 In this context, the medium addition command is also known as the medium addition operation command, and the automatic medium addition intelligent agent is also known as the medium addition industrial intelligent agent. The video data is the image captured by the camera, and the model data is the data planned by the VLC dynamic association model. On the interactive interface of the medium addition industrial intelligent agent, the operator inputs the amount of medium to be added and the medium system to be added through voice or text input. The medium addition industrial intelligent agent plans the optimal suction point and the collision-free shortest path for the crane movement. The medium addition industrial intelligent agent uses an AI controller to change the pulse width of the six DOs (forward, backward, left, right, up, down) of the ordinary controller, controlling the crane to move to the optimal suction point of the medium storage to pick up the medium and put it into the concentrated medium tank. The control system obtains the value through a weighing device. The amount of medium added to the concentrate tank is measured. Once the added medium reaches the required amount for the heavy medium system, the AI ​​controller of the control system moves the overhead crane to its initial position and stops working. The control system opens the clear water valve to allow clear water to flow into the concentrate tank through the clear water pipe. The control system opens the air-powered stirring valve to blow air into the concentrate tank to prevent medium sedimentation and promote the mixing of medium and water. When the level gauge on the wall of the concentrate tank detects that the liquid level has reached the set value or the flow meter detects that the required amount of clear water has flowed in, the control system closes the clear water valve. The control system controls the transfer pump to deliver the mixed medium to the heavy medium system. When the level gauge on the wall of the concentrate tank detects that the liquid level is below the critical value, the control system stops the transfer pump and closes the air-powered stirring valve.

[0165] Specifically, the equipment involved in the medium addition process includes, for example, a densitometer, a magnetic content meter, an overhead crane, a suction cup, a weighing device, a concentrated medium tank, a clean water pipe, a clean water valve, a flow meter, an air duct, a pneumatic agitator valve, a level gauge, a transfer pump, a camera, multiple proximity switches, and a control system. The suction cup is equipped with a weighing device to measure the real-time medium addition amount. The suction cup is used to adsorb the medium and, connected to the overhead crane, can move up, down, left, right, forward, and backward. Multiple proximity switches are evenly installed on the overhead crane track and transmit their status to the control system. The control system synthesizes the status of multiple proximity switches at the same time and draws a two-dimensional grid of the medium storage area. Based on the two-dimensional grid, it determines the position information of the suction cup at that moment. The control system controls the suction cup to evenly absorb the medium at a suitable position within the medium storage area and deliver it to the concentrated medium tank.

[0166] In addition, a clean water pipe is installed above the concentrated medium tank, and a clean water valve and flow meter are installed on the clean water pipe;

[0167] The control system determines the required amount of clean water by the weight of the medium added to the thickening tank. The flow meter is used to detect the amount of clean water added. The control system controls the opening and closing of the clean water valve to allow a fixed amount of clean water to flow into the thickening tank through the clean water pipe.

[0168] In addition, an air duct is installed inside the concentrated medium tank, and a wind-driven agitator valve is installed on the air duct;

[0169] The control system opens the air-powered stirring valve to blow air into the concentrated medium tank to prevent media sedimentation and promote the mixing of media and water.

[0170] In addition, a level gauge is installed on the wall of the concentrate tank to monitor the liquid level inside the tank in real time.

[0171] Optionally, in step 104, after the overhead crane is driven by the ordinary controller, it moves along the shortest collision-free path to the optimal suction point, including:

[0172] Step 1041: The crane's ordinary controller receives the pulse width command sent by the AI ​​controller;

[0173] Step 1042: The crane's ordinary controller outputs the target number of pulses according to the pulse width modulation duty cycle through the DO (Digital Output) interface, driving the servo motor of the corresponding axis to run until the crane moves to the optimal suction point along the shortest collision-free path.

[0174] In the above embodiments of this application, such as Figure 7 As shown, the overhead crane is driven by pulses output according to pulse width instructions via the DO interface, achieving precise instruction transmission and fine motor control, ensuring that the crane moves strictly along the planned path to the optimal suction point. This method offers rapid response and high control precision, effectively avoiding path deviations and collision risks, and improving the efficiency and safety of medium addition. Simultaneously, it reduces manual intervention and equipment wear, ensuring stable operation of the coal preparation process and providing an efficient and reliable solution for industrial automation control.

[0175] By applying the technical solution of this embodiment, it is possible to achieve media stack identification, intelligent crane scheduling and precise media addition, solving the problems of low efficiency, poor accuracy and insufficient reliability of traditional manual media addition.

[0176] Based on the above, Figure 1 To achieve the above objectives, this application also provides a computer device, specifically a personal computer, server, network device, etc., as shown in the method. The computer device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figure 1The method shown is a smart media addition method suitable for coal preparation plants.

[0177] Optionally, the computer device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB ports, card reader ports, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Bluetooth interfaces, Wi-Fi interfaces), etc.

[0178] Those skilled in the art will understand that the computer device structure provided in this embodiment does not constitute a limitation on the computer device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0179] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages and stores the hardware and software resources of a computer device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software within the physical device.

[0180] Through the above description of the implementation methods, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or by hardware implementation with cameras symmetrically deployed on the walls at the top center of the media storage and on both sides of the crane beam's movement direction, simultaneously capturing images and transmitting them to the AI ​​controller for preprocessing. When the control system determines the addition of media, the media-adding industrial intelligent agent analyzes the crane trajectory and media pile characteristics in the image, and associates them with the position command sequence of the ordinary controller to form a VLC dynamic association model. Based on this, the optimal absorption point and the collision-free shortest path are planned, and the AI ​​controller converts them into pulse width commands to control the crane movement. After the crane grabs the media and sends it into the thickener tank, the control system determines the amount of clean water according to the weight of the media, and controls the clean water pipe and air pipe to inject water and blow air into the tank, forming a heavy media suspension for coal preparation. It can realize media pile identification, intelligent crane scheduling, and precise media addition, solving the problems of low efficiency, poor accuracy, and insufficient reliability of traditional manual media addition.

[0181] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the system of the embodiment scenario can be distributed throughout the system of the embodiment scenario as described, or they can be modified to reside in one or more systems different from this embodiment scenario. The modules of the above-described embodiment scenario can be combined into one module, or further divided into multiple sub-modules.

[0182] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any modifications that can be made by those skilled in the art should fall within the protection scope of this application.

Claims

1. A smart medium-addition method suitable for coal preparation plants, characterized in that, The method is applied to a control system, which includes a general controller, an AI controller, and an industrial intelligent agent, and includes: A top panoramic camera is deployed at the center of the top of the media storage facility, and side wall dynamic tracking cameras are symmetrically deployed on both sides of the media storage facility along the direction of movement of the overhead crane beam. This allows the top panoramic camera and the side wall dynamic tracking cameras to capture images synchronously and transmit the captured images to the AI ​​controller for preprocessing. The top panoramic camera covers the central axis area of ​​the overhead crane beam and the entire media stack in the media storage facility. Both the top panoramic camera and the side wall dynamic tracking cameras are multi-view CV cameras. The AI ​​controller extracts color and texture features based on the images acquired by the multi-view CV cameras and identifies magnetic features through the color and texture features. When the additive industrial intelligent agent receives the additive operation command, it uses computer vision technology to analyze the visual features of the crane's movement trajectory and the medium stack shape in the image preprocessed by the AI ​​controller. It then associates the analyzed visual features with the position command sequence issued by the ordinary controller controlling the crane, forming a VLC dynamic association model that includes the mapping relationship between vision, position, and command. The position command sequence is obtained by the additive industrial intelligent agent from the ordinary controller controlling the crane. The Jiajie industrial intelligent agent, based on the VLC dynamic correlation model, determines all candidate suction points that the suction cup may move to; The industrial intelligent agent uses a multi-objective optimization cost function to select the optimal extraction point from candidate extraction points that minimizes the optimization cost. The multi-objective optimization cost function is as follows: P represents the real-time coordinates of the candidate extraction point in the global coordinate system, determined based on the VLC dynamic association model, and f(p) is the multi-objective optimization cost function. This represents the Euclidean distance from the current position of the suction cup to the candidate suction point. The magnetic content of the medium at the candidate absorption point. The height of the medium at the candidate absorption point. , These are the distance weighting coefficient, the medium quality weighting coefficient, and the medium height weighting coefficient, respectively. ; The industrial intelligent agent plans the collision-free shortest path based on the VLC dynamic association model and feeds it back to the AI ​​controller. The AI ​​controller converts the collision-free shortest path into a pulse width command and controls the ordinary controller of the overhead crane based on the pulse width command, so that the overhead crane moves along the collision-free shortest path to the optimal suction point after being driven by the ordinary controller. The overhead crane has a corresponding suction cup. The control system controls the overhead crane to reach the optimal suction point via its ordinary controller. The suction cup then draws the medium from the medium pile and sends it into the thickening tank. The control system determines the required amount of clean water based on the weight of the medium added to the thickening tank. It also controls the flow of the required amount of clean water into the thickening tank via the ordinary controller of the clean water pipe and the blowing of air into the thickening tank via the ordinary controller of the air duct to form a heavy medium suspension for coal preparation. The thickening tank is equipped with a clean water pipe above it and an air duct inside the thickening tank. Each of the overhead crane, clean water pipe, and air duct has its own ordinary controller.

2. The method according to claim 1, characterized in that, The Jiajie industrial intelligent agent includes a lightweight U-Net semantic segmentation model. This agent utilizes computer vision technology to analyze the visual features of the crane's motion trajectory and the dielectric stack morphology in images preprocessed by the AI ​​controller, including: The Jiajie Industrial Intelligent Agent uses the lightweight U-Net semantic segmentation model to perform pixel-level classification on the image preprocessed by the AI ​​controller, resulting in suction cup area, media stack area, ground area and crane area. In the dielectric stack region, the industrial intelligent agent obtains the three-dimensional contour of the dielectric stack through the Otsu threshold contour extraction algorithm, and determines the visual features of the dielectric stack morphology based on the three-dimensional contour. In the crane area, the Jiajie industrial intelligent agent transforms the pixel-level positional changes of the crane into a continuous motion trajectory in physical space through the spatiotemporal correlation of multiple frames of images, and extracts the visual features of the crane's motion trajectory based on the transformed continuous motion trajectory.

3. The method according to claim 2, characterized in that, The industrial intelligent agent associates the parsed visual features with the sequence of position commands issued by the ordinary controller controlling the overhead crane, including: The Jiajie Industrial Intelligent Agent is based on the principle of multi-view triangulation. It captures the pixel coordinates of the same feature point from the images captured simultaneously by the top panoramic camera and the side wall dynamic tracking camera. Combined with the calibrated extrinsic parameters of the top panoramic camera and the side wall dynamic tracking camera, it calculates the coordinates of the feature point in the global coordinate system until the visual features of the crane movement trajectory and the medium stack morphology parsed in the image are unified in the global coordinate system. The industrial intelligent agent will associate the visual features after unifying the coordinates with the position command sequence issued by the ordinary controller that controls the overhead crane.

4. The method according to claim 1, characterized in that, The overhead crane only moves orthogonally along the X, Y, and Z axes. The industrial intelligent agent plans the shortest collision-free path, including: The Jiajie industrial intelligent agent plans the shortest collision-free path based on the principle of prioritizing the far axis and then the near axis, using an improved path optimization algorithm.

5. The method according to claim 4, characterized in that, The overhead crane is equipped with servo motors, each corresponding to an X, Y, and Z axis. Each axis has a preset pulse distance conversion coefficient. The overhead crane also has preset maximum acceleration and preset maximum speed. The AI ​​controller converts the collision-free shortest path into pulse width instructions, including: The AI ​​controller decomposes the collision-free shortest path into the target distances that need to be moved in the X, Y, and Z axis directions, respectively. The AI ​​controller calculates the number of target pulses required for each axis of the servo motor to reach the target distance based on the preset pulse distance conversion coefficient. The AI ​​controller determines the speed curve of the crane at different stages based on the collision-free shortest path, the crane's preset maximum acceleration, and the preset maximum speed. The AI ​​controller calculates the pulse width modulation duty cycle corresponding to each stage based on the speed curve of the crane at different stages. The AI ​​controller encapsulates the calculated target number of pulses and the corresponding pulse width modulation duty cycle into a pulse width instruction.

6. The method according to any one of claims 1 to 5, characterized in that, Before the industrial intelligent agent receives the dispensing operation instruction, the method further includes: On the interactive interface of the media processing industrial intelligent agent, the operator inputs the amount of media to be processed and the media reprocessing system to be processed through voice or text input. The media processing industrial intelligent agent generates media processing operation instructions based on the amount of media to be processed and the media reprocessing system to be processed. The AI ​​controller is controlled by the Jiajie operation command, so that the operation result of the AI ​​controller is fed back to the operator through the Jiajie industrial intelligent agent, which has an interactive interface.

7. The method according to claim 6, characterized in that, The AI ​​controller includes an edge computing unit that transmits captured images to the AI ​​controller for preprocessing, including: The captured images are transmitted to the edge computing unit, where a bilateral filtering algorithm is used to preprocess the images.

8. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent media addition method for coal preparation plants as described in any one of claims 1 to 7.

Images