Digital twin modeling and dynamic control method and system in electric arc furnace installation process

Abstract

The invention discloses a digital twinning modeling and dynamic control method and system in an electric arc furnace installation process, and belongs to the technical field of crossing of metallurgical engineering and digital twinning technology, and the digital twinning modeling and dynamic control method comprises the following steps: S1, initialization and model deployment; s2, electric arc furnace component point cloud data collection and preprocessing; s3, component identification and position extraction; s4, virtual and real data transmission and coordinate conversion; and S5, virtual model dynamic mapping, collision early warning and installation adjustment are carried out. Through the high-precision point cloud recognition technology, category distinguishing and three-dimensional position accurate extraction of key components (such as electrodes and furnace shells) of the electric arc furnace are achieved, a low-delay data transmission link is constructed, and real-time synchronization of physical equipment position information and a virtual model is ensured; and through collision detection and early warning logic in the virtual scene, installation personnel are assisted in accurate positioning, and misoperation is avoided.

Description

Digital twin modeling and dynamic control methods and systems for electric arc furnace installation Technical Field

[0001] This invention relates to the interdisciplinary field of metallurgical engineering and digital twin technology, specifically to a digital twin modeling and dynamic control method and system for the installation process of an electric arc furnace. Background Technology

[0002] As a core piece of equipment in the metallurgical industry, the installation of electric arc furnaces is characterized by high precision, high risk, and multi-process coordination. Traditional installation methods rely on manual measurement and experience-based judgment, which presents the following problems:

[0003] Low installation position accuracy: It relies on manual measurement with tools such as laser rangefinders, which is easily affected by environmental vibration and high temperature interference, with errors reaching ±5mm or more, affecting the stability of subsequent smelting.

[0004] Insufficient process visualization: The installation process lacks real-time 3D modeling capabilities, making it difficult for managers to remotely monitor the spatial relationships of key components (such as electrode columns and furnace shells);

[0005] Low collaboration efficiency: The installation team, the supervisor, and the equipment manufacturer need to collaborate on-site, but information transmission is delayed, leading to delays in the connection of work processes;

[0006] Slow response to abnormalities: When components approach the installation reference position, there is a lack of real-time early warning mechanism, which can easily lead to equipment collisions or positioning deviations due to operational errors.

[0007] Therefore, there is an urgent need for a digital twin method based on high-precision point cloud recognition to achieve real-time modeling, precise positioning, dynamic early warning and collaborative control of the electric arc furnace installation process. Summary of the Invention

[0008] The technical problem this invention aims to solve is: how to address the issues of low accuracy in digital twin modeling, high virtual-physical synchronization delay, and lack of dynamic interactive early warning mechanisms during electric arc furnace installation. This invention provides a digital twin modeling and dynamic control method for electric arc furnace installation. This method uses high-precision point cloud recognition technology to classify and accurately extract the three-dimensional positions of key components of the electric arc furnace (such as electrodes and furnace shell), and constructs a low-latency data transmission link to ensure real-time synchronization between the physical equipment position information and the virtual model. Furthermore, it assists installers in accurate positioning and avoids operational errors through collision detection and early warning logic in the virtual scene.

[0009] The present invention solves the above-mentioned technical problems through the following technical solution, and the present invention includes the following steps:

[0010] S1: Initialization and Model Deployment

[0011] Start the Jetson embedded computing platform to complete the loading of the operating system kernel, hardware driver adaptation and pre-allocation of computing resources; read the weight file of the pre-trained 3D point cloud segmentation model, which is generated by training the 3D point cloud segmentation model through sample point clouds of electric arc furnace components;

[0012] S2: Point cloud data acquisition and preprocessing of electric arc furnace components

[0013] The ZED binocular depth camera is activated, and the built-in binocular vision matching algorithm is used to generate 3D point cloud data of the electric arc furnace components. The data is then stored in ply format to a specified path. Noise filtering and downsampling are performed on the 3D point cloud data to obtain preprocessed 3D point cloud data.

[0014] S3: Component Identification and Location Extraction

[0015] The preprocessed 3D point cloud data is input into the 3D point cloud segmentation model inference instance, and the identified electric arc furnace component categories and 3D position information are output.

[0016] S4: Virtual and Real Data Transmission and Coordinate Transformation

[0017] The 3D position information is encapsulated into data frames using the UDP protocol and transmitted to the Unity digital twin platform. After parsing the data frames, the Unity digital twin platform converts the 3D position information in the camera coordinate system into coordinates in the Unity digital twin platform's world coordinate system.

[0018] S5: Virtual Model Dynamic Mapping, Collision Warning, and Installation Adjustment

[0019] The Unity digital twin platform uses an interpolation algorithm to drive the virtual model of the electric arc furnace to move gradually according to the transformed coordinates, achieving smooth synchronization with the physical components. The BoxCollider component for installing the reference position in the virtual model of the electric arc furnace is activated, and collision detection is enabled. When the virtual component enters the BoxCollider range, a red warning line flashing effect is triggered. The furnace position is adjusted on-site according to the warning until all components of the electric arc furnace are installed, completing the dynamic mapping and interactive closed loop of objects in the digital twin scene.

[0020] Furthermore, in step S1, the electric arc furnace components include electrode columns, electrode arms, worktable, furnace shell, furnace cover, and furnace bottom.

[0021] Furthermore, in step S1, the 3D point cloud segmentation model is constructed based on the PointNet++ network.

[0022] Furthermore, in step S2, the ZED binocular depth camera establishes a data link with the Jetson embedded computing platform via a USB interface; noise filtering uses a statistical filtering algorithm to remove outliers, and downsampling uses voxel grid filtering to reduce the point cloud density to a set value.

[0023] Furthermore, in step S3, in the three-dimensional point cloud segmentation model, through model forward propagation calculation, the electric arc furnace component category label, identification confidence, and three-dimensional center coordinates (Xc, Yc, Zc) in the camera coordinate system are output.

[0024] Furthermore, in step S4, the coordinate transformation process is as follows:

[0025] The coordinates (Xu, Yu, Zu) in the world coordinate system of the Unity digital twin platform are obtained by transforming the camera coordinate system (Xc, Yc, Zc) with the preset hand-eye calibration matrix M.

[0026] Furthermore, in step S5, the specific process of using an interpolation algorithm to drive the virtual model of the electric arc furnace to move step by step according to the transformed coordinates is as follows:

[0027] S51: Let the current position of the electric arc furnace virtual model be... The target location is The total duration of the exercise is T;

[0028] S52: Calculate the model position at time t using a linear interpolation algorithm. :

[0029] ;

[0030] ;

[0031] ;

[0032] Where 0 ≤ t ≤ T;

[0033] S53: Update the position of the electric arc furnace virtual model every frame. This enables a smooth transition from the current position to the target position.

[0034] S54: When a virtual component in the electric arc furnace virtual model enters the BoxCollider range, a red warning line flashes, and the physical component position is adjusted on-site based on the warning.

[0035] Furthermore, in step S5, the formula for calculating the deviation d between the virtual model of the electric arc furnace and the hazardous area is as follows:

[0036]

[0037] The danger zone is the area covered by the BoxCollider. The real-time position coordinates of the electric arc furnace virtual model. When d≥100mm, the virtual model of the electric arc furnace is determined to have left the danger zone, the red warning line stops flashing and switches to yellow.

[0038] This invention also provides a digital twin modeling and dynamic control system for the electric arc furnace installation process, applied to the above-mentioned method, including:

[0039] The initialization and model deployment module is used to start the Jetson embedded computing platform, complete the loading of the operating system kernel, hardware driver adaptation and pre-allocation of computing resources; read the pre-trained 3D point cloud segmentation model weight file, the 3D point cloud segmentation model is generated by training the point cloud of the electric arc furnace component sample;

[0040] The data acquisition and preprocessing module is used to start the ZED binocular depth camera, generate 3D point cloud data of the electric arc furnace components through the camera's built-in binocular vision matching algorithm, and store it in ply format to a specified path; perform noise filtering and downsampling processing on the 3D point cloud data to obtain preprocessed 3D point cloud data;

[0041] The component identification and location extraction module is used to input the preprocessed 3D point cloud data into the 3D point cloud segmentation model inference instance and output the identified electric arc furnace component category and 3D location information.

[0042] The data transmission and coordinate transformation module is used to encapsulate 3D position information into data frames according to the UDP protocol and transmit them to the Unity digital twin platform. After parsing the data frames, the Unity digital twin platform converts the 3D position information in the camera coordinate system into coordinates in the Unity digital twin platform's world coordinate system.

[0043] The dynamic mapping and collision warning module is used in the Unity digital twin platform to drive the electric arc furnace virtual model to move step by step according to the transformed coordinates using an interpolation algorithm, achieving smooth synchronization with physical components; it activates the BoxCollider component for installing the reference position in the electric arc furnace virtual model and enables collision detection; when a virtual component enters the BoxCollider range, a red warning line flashing effect is triggered, and the furnace position is adjusted on-site according to the warning until all components of the electric arc furnace are installed, completing the dynamic mapping and interaction loop of objects in the digital twin scene.

[0044] The present invention has the following advantages over the prior art:

[0045] 1. Fast synchronization between virtual and real worlds

[0046] By optimizing lightweight inference and using high-speed UPD transmission, the end-to-end latency from physical component location acquisition to virtual model update is ≤250ms, meeting the real-time requirements of the installation process.

[0047] 2. Dynamic early warning capability

[0048] By combining BoxCollider collision detection with a warning line flashing mechanism, proactive warnings can be issued when components approach the mounting reference position, reducing the operational error rate by more than 60%.

[0049] 3. Improved collaboration efficiency

[0050] The digital twin platform supports remote access from multiple terminals, enabling real-time collaboration between the installation team and the supervisor, and reducing the time for process connection by 30%. Attached Figure Description

[0051] Figure 1 is a flowchart illustrating the digital twin modeling and dynamic control method for the electric arc furnace installation process in an embodiment of the present invention.

[0052] Figure 2 is a digital twin timing diagram of the entire process of electric arc furnace component installation in an embodiment of the present invention;

[0053] Figure 3 is an example diagram of collision detection and warning line effects in a Unity virtual scene according to an embodiment of the present invention. Detailed Implementation

[0054] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0055] Example 1

[0056] This embodiment provides a technical solution: a digital twin modeling and dynamic control method for the installation process of an electric arc furnace, including the following steps:

[0057] Step 1: System Initialization and Model Deployment

[0058] 1.1 Start the Jetson embedded computing platform, complete the loading of the operating system kernel, hardware driver adaptation and pre-allocation of computing resources to ensure that the real-time requirements of point cloud data processing and model inference are met;

[0059] 1.2 Call the model loading module to read the pre-trained PointNet++ 3D point cloud segmentation model weight file. This model is generated by training on point clouds of electric arc furnace components (including 6 key components: electrode column, electrode arm, worktable, furnace shell, furnace cover, and furnace bottom), with a category recognition accuracy of ≥90%.

[0060] Step 2: Point cloud data acquisition and preprocessing of electric arc furnace components

[0061] 2.1. Start the ZED binocular depth camera and establish a high-speed data link with the Jetson embedded computing platform via the USB 3.0 interface;

[0062] 2.2 Control the binocular depth camera to scan the electric arc furnace components at the installation site in real time, generate 3D point cloud data through binocular vision matching algorithm, and store it in ply format to the SSD cache path ( / data / pointcloud / ) of the Jetson embedded computing platform.

[0063] In this embodiment, the specific process of generating point cloud data using the binocular vision matching algorithm (built into the camera) is as follows:

[0064] Phase 1:

[0065] Hardware Sync: The left and right CMOS sensors of the ZED camera must achieve shutter synchronization at the microsecond level.

[0066] ISP processing: The raw RAW data is processed by the ISP (Image Signal Processor) for demosaicing, white balance, and automatic exposure.

[0067] At this point, we have two ordinary RGB images, called... (Left image) and (Right image). However, at this point, they are filled with the physical distortion of the lens (barrel or pincushion distortion) and cannot be directly used for calculation.

[0068] Phase Two:

[0069] Distortion removal: using the factory-calibrated internal parameters (Intrinsics, ... Using distortion coefficients, we flatten the image from a “curved surface”.

[0070] Epipolar Alignment: Using extrinsics (the rotation and translation relationship between the left and right cameras), we reproject the left and right images onto two completely parallel virtual planes.

[0071] Result: After correction, any pixel in the left image... The corresponding matching points in the right-hand image must be located in the same row. This greatly reduces the algorithm's complexity.

[0072] Phase Three:

[0073] Feature extraction and matching:

[0074] The corrected left and right images are fed into a convolutional neural network (CNN). The network does not simply compare pixels, but rather "understands" the context of the image patches. Even on a textureless white wall, the network can infer the parallax in the middle based on the surrounding environment (table edges, shadows).

[0075] Disparity Estimation:

[0076] The network outputs a disparity map. (Disparity) Defined as the difference in horizontal coordinates between corresponding points in the left and right images:

[0077] ;

[0078] Sub-pixel accuracy: Physical pixels are integers, but real-world offsets are continuous. ZED pixel algorithms calculate floating-point disparity values ​​(e.g., d = 24.56 pixels), which is key to achieving high-precision (millimeter-level) measurements.

[0079] Phase Four:

[0080] With parallax We can then calculate the depth using geometric principles. The formula is very classic:

[0081]

[0082] in, The vertical depth of the object from the camera. Corrected focal length (in pixels). Baseline length (i.e., the distance between the optical centers of the left and right lenses, ZED 2 is 120mm). Parallax (unit: pixels).

[0083] Phase 5: 3D Spatial Reprojection

[0084] Now we have a depth map, but this is not a point cloud. A point cloud needs to know the location of each point in real-world space. coordinate.

[0085] We used a pinhole camera model to reverse engineer:

[0086] Coordinate system definition: The ZED SDK uses a right-handed coordinate system by default, with the Y-axis pointing downwards, the X-axis pointing to the right, and the Z-axis pointing forwards (ImageCoordinate System).

[0087] Calculation formula:

[0088] ;

[0089] ;

[0090] ;

[0091] in, : The coordinates of a pixel on the image The optical center (principal point) of the image.

[0092] Phase 6: Data Packing (XYZRGBA Packing)

[0093] The Mat object (MEASURE::XYZRGBA) generated by the ZED SDK uses an efficient memory layout:

[0094] Each point occupies 128 bits (16 bytes):

[0095] float32 X (4 bytes)

[0096] float32 Y (4 bytes)

[0097] float32 Z (4 bytes)

[0098] uint32 Color (4 bytes)

[0099] Here, bitwise operations are used to compress the color information of the four channels: B, G, R, and A. Thus, a 1920x1080 point cloud image is essentially a huge... The graphics card can directly read and render a colorful 3D world from the array.

[0100] 2.3 Preprocessing of point cloud data:

[0101] Noise filtering: Outliers are removed using a statistical filtering algorithm (neighborhood number = 50, standard deviation threshold = 1.0).

[0102] Downsampling: The point cloud density is reduced to 200 points / m² by voxel grid filtering, balancing accuracy and computational efficiency;

[0103] Step 3: Part recognition and location extraction based on PointNet++

[0104] The preprocessed point cloud data is input into the PointNet++ 3D point cloud segmentation model inference instance, and the segmented electric arc furnace component categories and 3D position information are output.

[0105] Step 4: Virtual and Real Data Transmission and Coordinate Transformation

[0106] 4.1 The data transmission module encapsulates the effective location information (the 3D location information segmented from the point cloud data) into data frames according to the UPD protocol and transmits them to the Unity digital twin platform via IP address;

[0107] 4.2 After the Unity digital twin platform parses the data frame, it drives the components of the electric furnace to move.

[0108] Step 5: Dynamic mapping and collision warning of virtual model

[0109] 5.1 The Unity word twin platform drives the virtual model of the electric arc furnace (this model is the SolidWorks construction model designed by the manufacturer. The STL format model file exported from SolidWorks is imported into the software using SketchUp modeling software. Then, redundant resources are deleted, processed, groups are created, corresponding materials are applied, and the model is exported as an FBX format model and imported into the Unity platform to ensure that the model is mapped 1:1) to move in real time according to the converted coordinates to achieve synchronization with the position of physical components;

[0110] 5.2 Activate the BoxCollider component for the mounting reference position in the electric arc furnace virtual model and enable collision detection;

[0111] 5.3 When the virtual component enters the BoxCollider range, the red warning line flashing effect is triggered, prompting the staff to make adjustments and issuing new instructions. Under the instructions, the furnace body slowly adjusts and gradually moves away from the danger zone. The warning line then stops flashing, and the scene returns to stability, completing the digital twin mapping of the component.

[0112] Example 2

[0113] As shown in Figures 1-3, the specific implementation steps of the method of the present invention are described in detail below, taking into account the actual scenario of the electric arc furnace electrode column being connected and installed with the worktable:

[0114] I. System Preparation and Initialization

[0115] 1. Hardware Deployment: Deploy a Jetson embedded computing platform at the electric arc furnace installation site, connect and fix a ZED binocular camera to ensure that the camera's field of view covers the electrode column hoisting area and the workbench installation reference position; deploy a Unity server in the monitoring center, preload a 1:1 virtual model of the electric arc furnace (including electrode columns, workbench and installation reference markings).

[0116] 2. Software Startup: Start the Jetson platform, complete the loading of the operating system and hardware drivers, call the model loading module to read the pre-trained PointNet++ 3D point cloud segmentation model weight file, and initialize the point cloud recognition inference environment; start the ZED camera and establish a data transmission link with the Jetson platform; start the Unity server, load the virtual scene and activate the communication receiving module, and wait to receive position data.

[0117] II. Point Cloud Acquisition and Component Recognition

[0118] 1. Point cloud acquisition: Control the ZED camera to scan the electrode column in real time during the hoisting process, generate three-dimensional point cloud data through binocular vision matching, store it in the preset format to the specified path of the Jetson platform, and record the scanning time information synchronously.

[0119] 2. Point cloud preprocessing: Noise filtering is performed on the collected point cloud data to remove noise points in the environment that are unrelated to the electrode posts; the amount of point cloud data is reduced by downsampling to improve processing efficiency while ensuring recognition accuracy.

[0120] 3. Component identification and location extraction: The preprocessed point cloud data is input into the PointNet++ 3D point cloud segmentation model inference instance. The model identifies the "electrode post" category through feature extraction and analysis, and calculates its 3D center position in the camera coordinate system.

[0121] III. Virtual and Real Data Transmission and Model Synchronization

[0122] 1. Data transmission: The Jetson platform encapsulates the valid location information into data frames according to the UDP protocol and transmits them to the Unity server over the network.

[0123] 2. Coordinate Transformation and Model-Driven Operation: The Unity server parses the data frame, extracts the position information of the electrode pillars, and converts the position in the camera coordinate system to the coordinates in the virtual scene world coordinate system through a preset coordinate transformation relationship; it then drives the virtual electrode pillar model to move in real time according to the transformed coordinates, achieving synchronization with the position of the physical electrode pillars.

[0124] IV. Collision Detection and Early Warning Control

[0125] 1. Collision Range Monitoring: Activate the BoxCollider component of the workbench mounting reference position in the Unity virtual scene to monitor in real time whether the virtual electrode post enters the installation allowable range defined by the component.

[0126] 2. Warning Trigger: When the virtual electrode column enters the BoxCollider range, the system automatically triggers a red warning line flashing effect to alert on-site installers that there may be a risk of collision.

[0127] 3. Adjustment and Warning Cancellation: Installers slowly adjust the physical electrode post positions according to the warning prompts, and the virtual model updates synchronously; when the electrode post reaches the preset installation position (meets the accuracy requirements), an "Installation Complete" signal is sent on-site. After receiving the signal, the Unity server turns off the flashing warning line, and the virtual electrode post locks in its final position, completing the digital twin mapping of the component.

[0128] In summary, the digital twin modeling and dynamic control method for the electric arc furnace installation process described in the above embodiments constructs a real-time mapping mechanism between the physical installation process and the virtual model through system initialization and model deployment, point cloud acquisition and segmentation, virtual and real data transmission synchronization, and collision detection and early warning. This effectively solves the problems of insufficient accuracy, inefficient collaboration, and delayed early warning in electric arc furnace installation, and provides intelligent technical support for the high-precision installation of large-scale metallurgical equipment.

[0129] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A digital twin modeling and dynamic control method for the installation process of an electric arc furnace, characterized in that, Includes the following steps: S1: Initialization and Model Deployment. Start the Jetson embedded computing platform and complete the loading of the operating system kernel, hardware driver adaptation, and pre-allocation of computing resources. S1: Read the pre-trained 3D point cloud segmentation model weight file. The 3D point cloud segmentation model is generated through training on the point cloud samples of the electric arc furnace component. S2: Electric arc furnace component point cloud data acquisition and preprocessing. Start the ZED binocular depth camera and generate 3D point cloud data of the electric arc furnace component through the camera's built-in binocular vision matching algorithm. Store the data in ply format to the specified path. Perform noise filtering and downsampling on the 3D point cloud data to obtain the preprocessed 3D point cloud data. S3: Component identification and location extraction inputs the preprocessed 3D point cloud data into the 3D point cloud segmentation model inference instance, and outputs the identified electric arc furnace component category and 3D location information; S4: Virtual-to-real data transmission and coordinate transformation: The 3D position information is encapsulated into data frames according to the UDP protocol and transmitted to the Unity digital twin platform. After parsing the data frames, the Unity digital twin platform converts the 3D position information in the camera coordinate system into coordinates in the Unity digital twin platform's world coordinate system. S5: Virtual model dynamic mapping, collision warning, and installation adjustment: The Unity digital twin platform uses an interpolation algorithm to drive the electric arc furnace virtual model to move gradually according to the converted coordinates, achieving smooth synchronization with physical components. The BoxCollider component for the installation reference position in the electric arc furnace virtual model is activated, and collision detection is enabled. When a virtual component enters the BoxCollider range, a red warning line flashing effect is triggered. The furnace position is adjusted on-site according to the warning until all components of the electric arc furnace are installed, completing the dynamic mapping and interactive closed loop of objects in the digital twin scene.

2. The digital twin modeling and dynamic control method for the electric arc furnace installation process according to claim 1, characterized in that, In step S1, the electric arc furnace components include electrode columns, electrode arms, worktable, furnace shell, furnace cover, and furnace bottom.

3. The digital twin modeling and dynamic control method for the electric arc furnace installation process according to claim 1, characterized in that, In step S1, the 3D point cloud segmentation model is constructed based on the PointNet++ network.

4. The digital twin modeling and dynamic control method for the electric arc furnace installation process according to claim 1, characterized in that, In step S2, the ZED binocular depth camera establishes a data link with the Jetson embedded computing platform via a USB interface; noise filtering uses a statistical filtering algorithm to remove outliers, and downsampling uses voxel grid filtering to reduce the point cloud density to a set value.

5. The digital twin modeling and dynamic control method for the electric arc furnace installation process according to claim 1, characterized in that, In step S3, the electric arc furnace component category label, identification confidence score, and three-dimensional center coordinates (Xc, Yc, Zc) in the camera coordinate system are output through forward propagation calculation in the three-dimensional point cloud segmentation model.

6. The digital twin modeling and dynamic control method for the electric arc furnace installation process according to claim 5, characterized in that, In step S4, the coordinate transformation process is as follows: the coordinates (Xu, Yu, Zu) in the world coordinate system of the Unity digital twin platform are transformed by the camera coordinate system (Xc, Yc, Zc) and the preset hand-eye calibration matrix M.

7. The digital twin modeling and dynamic control method for the electric arc furnace installation process according to claim 1, characterized in that, In step S5, the specific process of using an interpolation algorithm to drive the electric arc furnace virtual model to move step by step according to the transformed coordinates is as follows: S51: Let the current position of the electric arc furnace virtual model be... The target location is The total duration of the motion is T; S52: The model position at time t is calculated using a linear interpolation algorithm. ： ； ； Where 0≤t≤T; S53: Update the position of the electric arc furnace virtual model to [value] in each frame. This enables a smooth transition from the current position to the target position; S54: When a virtual component in the electric arc furnace virtual model enters the BoxCollider range, a red warning line flashes, and the physical component position is adjusted on-site based on the warning.

8. The digital twin modeling and dynamic control method for the electric arc furnace installation process according to claim 7, characterized in that, In step S5, the formula for calculating the deviation d between the virtual model of the electric arc furnace and the hazardous area is as follows: The danger zone is the area covered by the BoxCollider. For the real-time position coordinates of the virtual model of the electric arc furnace, When d≥100mm, the virtual model of the electric arc furnace is determined to have left the danger zone, the red warning line stops flashing and switches to yellow.

9. A digital twin modeling and dynamic control system for the installation process of an electric arc furnace, characterized in that, The method described in any one of claims 1 to 8, comprising: an initialization and model deployment module, used to start the Jetson embedded computing platform, complete the loading of the operating system kernel, hardware driver adaptation, and pre-allocation of computing resources; read the weight file of the pre-trained 3D point cloud segmentation model, wherein the 3D point cloud segmentation model is generated through training point clouds of electric arc furnace component samples; a data acquisition and preprocessing module, used to start the ZED binocular depth camera, generate 3D point cloud data of the electric arc furnace component through the camera's built-in binocular vision matching algorithm, and store it in ply format to a specified path; perform noise filtering and downsampling processing on the 3D point cloud data to obtain preprocessed 3D point cloud data; a component identification and location extraction module, used to input the preprocessed 3D point cloud data into the 3D point cloud segmentation model inference instance, and output the identified electric arc furnace component category and 3D location information; data The transmission and coordinate transformation module encapsulates 3D position information into data frames using the UDP protocol and transmits them to the Unity digital twin platform. After parsing the data frames, the Unity digital twin platform converts the 3D position information in the camera coordinate system into coordinates in the Unity digital twin platform's world coordinate system. The dynamic mapping and collision warning module is used by the Unity digital twin platform to drive the electric arc furnace virtual model to move gradually according to the converted coordinates using an interpolation algorithm, achieving smooth synchronization with physical components. It activates the BoxCollider component for installing the reference position in the electric arc furnace virtual model and enables collision detection. When a virtual component enters the BoxCollider's range, a red warning line flashing effect is triggered, and the furnace position is adjusted on-site according to the warning until all components of the electric arc furnace are installed, completing the dynamic mapping and interactive closed loop of objects in the digital twin scene.

Images