A method and system for sorting of residual carbon blocks based on intelligent identification and control

Abstract

The application discloses a kind of residual pole carbon block sorting method and system based on intelligent identification and control, including the electrolyte block and residual pole carbon block obtained by breaking residual pole and falling on conveying belt;The electrolyte block and residual pole carbon block of conveying belt are photographed to obtain picture;Picture is identified using NanoDet network model to obtain identification information;Based on identification information, residual pole carbon block is clamped by mechanical arm.The application realizes the operation of accurate detection, positioning and rapid sorting of the material flow composed of electrolyte block and residual pole carbon block on the high-speed running conveying belt, which can not only accurately distinguish residual pole carbon block, but also identify its size specification and position coordinates simultaneously.

Description

Technical Field

[0001] This invention relates to the field of residual carbon block sorting technology, and in particular to a method and system for sorting residual carbon blocks based on intelligent identification and control. Background Technology

[0002] In the existing electrolytic aluminum production process, the anode carbon block is extremely critical, as it is the core component of the electrolytic cell, responsible for conductivity and participation in the reaction. As electrolysis progresses, the anode carbon block is subjected to high temperatures, strong currents, and electrolyte erosion, leading to internal structural deterioration and a decline in physicochemical properties. For example, high temperatures alter the carbon block's crystal structure, resulting in decreased strength, and electrolyte erosion causes surface oxidation, weakening conductivity and stability, ultimately reducing it to anode waste and rendering it ineffective as an anode. The recycling and processing of anode waste is of great significance to the electrolytic aluminum industry. In terms of cost, anode waste contains a large amount of unconsumed carbon resources and other valuable components; recycling can significantly reduce raw material procurement costs. Medium-sized electrolytic aluminum enterprises can save millions of yuan annually through efficient recycling. Furthermore, if the anode waste is not sorted, it will enter the crushing system along with the electrolyte, being crushed and mixed together before entering the electrolytic cell. This indirectly increases the amount of carbon slag during electrolytic cell production. The anode waste also increases the contact resistance between the anode carbon block and the electrolytic cell, further reducing electrolysis efficiency, increasing energy consumption in electrolytic aluminum production, and ultimately raising the production cost of electrolytic aluminum.

[0003] Traditional sorting of residual anode carbon blocks relies on manual labor, which presents numerous problems. The working environment is harsh, with workers enduring high temperatures, dust, and strong magnetic fields, while also frequently handling heavy residual anodes—averaging several tons per day—resulting in high labor intensity, fatigue, and injury. In terms of efficiency, relying on visual observation and limited experience leads to slow sorting speeds and a high risk of errors. Missorting can result in substandard products being mixed into reuse batches, causing unstable performance of new anode carbon blocks and affecting the quality and efficiency of electrolytic aluminum.

[0004] Patent CN117680382A discloses an automatic sorting system based on visual recognition and multi-stage pneumatic pushers. A camera is installed diagonally above the material inflow direction of the conveyor belt to capture real-time images of the electrolyte and carbon block mixture on the belt. The captured image data is sent to an algorithm server, which deploys the YOLOv5 target detection algorithm to identify residual carbon blocks on the conveyor belt and calculate their physical location. The device's actuator consists of multiple cylinders connected to pushers. Each pneumatic pusher has a baffle connected to its lower end. When a residual carbon block is transported to the pusher position, the cylinders actuate, driving the pusher and baffle to move perpendicular to the belt, pushing the residual carbon block off the belt. The residual carbon block falls into the sorting box, completing the sorting process.

[0005] The sorting systems based on vision recognition and pneumatic push rods mentioned in existing patents have the following drawbacks:

[0006] Push rods and baffles can easily push out electrolyte along with the carbon residue: When using pneumatic push rods to sort residual carbon blocks, if the residual carbon blocks and electrolyte blocks are close together, when the residual carbon blocks reach the position of the pneumatic push rod, the pneumatic push rod starts to move, pushing the residual carbon blocks out of the conveyor belt while also pushing out the electrolyte. This results in the sorted residual carbon blocks containing a lot of electrolyte, increasing the workload of secondary sorting for workers.

[0007] The actuator lacks identification feedback: Because the speed of the conveyor belt is not uniform and varies with the amount of material on it, and the system in this patent operates based on a preset time, if the belt speed changes, errors will occur. This can cause the pusher to activate before the residual carbon blocks reach the pusher position. Since there is no feedback system at the pusher, it will push out electrolyte blocks that should not be sorted, while the residual carbon blocks that should be sorted will be missed.

[0008] The lack of a speed detection system and real-time control system makes it prone to errors: The speed of the conveyor belt varies with the amount of material on it. When the material is heavy, the belt speed slows down; when the material is light, the belt speed is normal. However, the pneumatic pusher-based residual electrode sorting system lacks a speed detection sensor. It relies on a fixed, preset delay time to determine whether the residual electrode block has reached the pusher position and when the pusher should activate. When the belt speed changes, the system still relies on the original delay time, leading to errors and preventing the correct sorting of residual electrode blocks.

[0009] The actuator lacks a protection system, making it prone to jamming and damaging equipment: the shape and size of the electrolyte blocks and residual carbon blocks on the conveyor belt are random and unpredictable. Thin, sheet-like electrolyte blocks or residual carbon blocks, or large quantities of small pieces, are frequently found. Furthermore, there is a gap between the baffle at the end of the pneumatic push rod and the conveyor belt. When the pneumatic push rod actuates, there is a certain probability that the sheet-like or small pieces of material will get stuck in the gap, causing the pneumatic push rod to be pulled by the belt. This can damage the pneumatic push rod or cylinder, or even the conveyor belt, resulting in a serious production accident. Summary of the Invention

[0010] To address the shortcomings mentioned above, this invention provides a method and system for sorting residual carbon blocks based on intelligent identification and control.

[0011] To achieve the above objectives, the present invention provides a sorting method for residual carbon blocks based on intelligent identification and control, which includes crushing the residual electrode and dropping the resulting electrolyte blocks and residual carbon blocks onto a conveyor belt.

[0012] Images were obtained by taking photographs of the electrolyte block and the residual carbon block on the conveyor belt;

[0013] The image was then used to perform an algorithmic recognition process using a NanoDet network model to obtain recognition information.

[0014] Based on the identification information, the residual carbon block is picked up by a robotic arm.

[0015] Preferably, the image is used to perform algorithmic recognition using a NanoDet network model to obtain recognition information, including:

[0016] The image is preprocessed by adjusting it to a fixed size and normalizing it.

[0017] After processing, the image features are extracted using a lightweight Backbone neural network;

[0018] The features at different scales are fused using a BiFPN neural network;

[0019] After fusion, target classification and bounding box regression are performed to predict the target's category and location;

[0020] For each target, perform center point detection and output the bounding box, category, and confidence score of the target.

[0021] Preferably, the identification information includes the size information, classification information, coordinate information, timestamp information, and speed information of the residual carbon block.

[0022] Preferably, the coordinate and size information in the identification information is converted into real-world coordinate information and target size, which includes:

[0023] Obtain the original image resolution (w*h) and the size of the corresponding real-world object region in the original image. ;

[0024] Simultaneously calculate the size of the image input to the NanoDet network model. The scaling factor relative to the resolution of the original image and ;

[0025] Obtain the size of the target detection box from the recognition information. Calculate the actual size of the target;

[0026] The actual width of the target is:

[0027] ;

[0028] The actual height of the target is:

[0029] ;

[0030] In the formula: .

[0031] Preferably, based on the identification information, the residual carbon block is gripped by a robotic arm, including:

[0032] The real-time sensor acquires the velocity of the residual carbon block and uses a camera to capture images, while simultaneously recording the acquisition time;

[0033] Calculate the displacement of the residual carbon block during the image recognition time;

[0034] Based on the displacement, the relative distance between the target and the robotic arm gripper is calculated;

[0035] Calculate the time required for the robotic arm gripper to grasp the target;

[0036] Calculate the adjustment time of the robotic arm gripper;

[0037] The difference between the time required to achieve the target and the adjustment time is the waiting time;

[0038] If the waiting time is positive, the robotic arm gripper will promptly remove the residual carbon block.

[0039] If the waiting time is negative, then the data cannot be captured and must wait for the next capture.

[0040] Preferably, the top of the robotic arm gripper is provided with a force feedback device.

[0041] Preferably, during the gripping process of the robotic arm, supplemental lighting is provided to the residual carbon block.

[0042] Preferably, the gripper of the robotic arm includes a gripper, a linear cylinder, and a valve for controlling the air source. The linear cylinder is connected to the upper part of the mechanical movable link. The two air outlets of the valve are connected to the two air holes of the linear cylinder. The control signal line of the valve is connected to the main control board. When the main control board controls the valve to open or close, high-pressure gas enters the linear cylinder through the connecting air pipe. The linear cylinder push rod moves, driving the mechanical movable link to move, thereby realizing the gripping and releasing action of the gripper.

[0043] This application also provides a residual carbon block sorting system based on intelligent identification and control, including:

[0044] The crushing module is used to crush the residual electrode and drop the resulting electrolyte block and residual electrode carbon block onto the conveyor belt.

[0045] The imaging module is used to take pictures of the electrolyte block and the residual carbon block on the conveyor belt to obtain images;

[0046] The recognition module is used to perform algorithmic recognition on the image using a NanoDet network model to obtain recognition information;

[0047] The gripping module is used to grip the residual carbon block using a robotic arm based on the identification information.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] This invention enables precise detection, positioning, and rapid sorting of a material flow consisting of a mixture of electrolyte blocks and residual carbon blocks on a high-speed conveyor belt. It not only accurately identifies the residual carbon blocks but also simultaneously recognizes their size, dimensions, and location coordinates. Simultaneously, based on information from the detection model, it precisely grasps the residual carbon blocks requiring sorting, ensuring they fall accurately into a specially designed sorting bin. This invention adapts to changes in belt speed, accurately grasping residual carbon blocks even with varying belt speeds, preventing the accidental grasping of electrolyte blocks. In the event of a scrape, it activates protection logic to safeguard the sorting equipment and the conveyor belt. Current technologies have significant shortcomings in the anode residual carbon block cleaning process. When lumpy residual carbon blocks accidentally fall into the conveyor belt, subsequent processes cannot effectively identify and sort them, resulting in these lumpy residual carbon blocks being directly transported to the crushing system along with electrolyte crusts. Attached Figure Description

[0050] Figure 1 The flowchart shows a method for sorting residual carbon blocks based on intelligent identification and control.

[0051] Figure 2 This is a diagram of the internal structure of a device for sorting residual carbon blocks based on intelligent identification and control.

[0052] Figure 3 This is a structural diagram of an apparatus for sorting residual carbon blocks based on intelligent identification and control. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Reference Figure 1 The present invention provides a sorting method for residual carbon blocks based on intelligent identification and control, including crushing the residual electrode and dropping the resulting electrolyte block and residual carbon block onto a conveyor belt.

[0055] Images were taken of the electrolyte block and residual carbon block on the conveyor belt.

[0056] The image is used to perform algorithmic recognition to obtain recognition information;

[0057] Based on the identification information, the residual carbon block is picked up by a robotic arm.

[0058] This application also provides a residual carbon block sorting system based on intelligent identification and control, including:

[0059] The crushing module is used to crush the residual electrode and drop the resulting electrolyte block and residual electrode carbon block onto the conveyor belt.

[0060] The imaging module is used to take pictures of the electrolyte block and residual carbon block on the conveyor belt to obtain images;

[0061] The recognition module is used to perform algorithmic recognition on images using the NanoDet network model to obtain recognition information.

[0062] The gripping module is used to grip the residual carbon block using a robotic arm based on identification information.

[0063] In this embodiment, the residual carbon block sorting system based on intelligent identification and control includes a material identification system, a sorting identification system, a speed measurement system, a sorting execution system, a control server, a background management system, a monitoring system, sorting boxes, and a conveyor belt device.

[0064] Reference Figure 2The material identification system includes a housing frame 12, a wide-angle industrial camera 11, and four 1-meter strip light sources 10. The wide-angle industrial camera 11 is installed 1.1m above the conveyor belt, perpendicular to the conveyor belt plane, with the camera lens facing the conveyor belt. The four strip light sources 10 are also installed 1m above the conveyor belt, with their long sides perpendicular to the belt's running direction and their emitting surfaces facing and parallel to the conveyor belt surface. The four strip light sources are spaced 330mm apart. The wide-angle industrial camera 11 is installed between the second and third strip light sources. Due to the high dust levels on site, to prevent dust from entering the equipment and causing malfunctions, dustproof protective shells are installed on the industrial camera 11 and the strip light sources 10. The dustproof protective shell for the wide-angle industrial camera is cylindrical, with a circular high-transmittance germanium glass top surface. The wide-angle industrial camera is installed inside the dustproof protective shell. The cleaning device, mainly consisting of air nozzles and a wiper system, is installed on the outside of the circular high-transmittance germanium glass. Regularly clean the protective glass to ensure the camera lens is clean and the images captured are clear. The strip light dustproof protective case is rectangular, with the strip light installed inside. The top surface of the strip light dustproof protective case is made of a highly transparent material, and the emitting side of the strip light is installed facing the top surface of the protective case.

[0065] Reference Figure 3The sorting execution system consists of two sets of sorting execution devices for residual carbon blocks. These devices are installed 3 meters in front of the material identification system (with the conveyor belt's direction of travel as the forward direction). Each set includes a parallel robotic arm device 6, a force feedback device 5, a gripper device 9, and a sorting device support frame 13. The sorting device support frame 13 is a square frame with four support legs mounted above the conveyor belt, measuring 2m x 2m. The four support legs of the sorting device support frame 13 are fixed to the ground with chemical bolts. The top surface of the sorting device support frame 13 is parallel to the plane of the conveyor belt 4, and the top surface of the sorting device support frame 13 is 1.8 meters above the plane of the conveyor belt. The sorting device support frame 13 is constructed of 100*100mm steel pipes connected by bolts and can bear a weight of 300kg. The parallel robotic arm device 6 is mounted inside the top of the sorting device support frame 13 using trapezoidal high-strength connectors. The bottom of the parallel robotic arm device 6 is a circular mounting plate. The force feedback device 5 is installed below the circular mounting plate of the parallel robotic arm device 6 via a connector. The gripper device 9 is a gripping device designed with a mechanical linkage. It consists of a gripper with a mechanical linkage, a high-power linear cylinder, and a valve for controlling the air source. The high-power linear cylinder is connected to the upper part of the mechanical linkage. The two air outlets of the valve are connected to the two air holes of the cylinder. The control signal line of the valve is connected to the main control board. When the main control board controls the valve to open or close, high-pressure gas enters the cylinder through the connecting air pipe. The cylinder push rod moves, driving the mechanical linkage to move, realizing the gripping and releasing action of the gripper device 9. The top surface of the gripper device 9 is connected to the bottom surface of the force feedback device 5 by bolts. The parallel robotic arm device 6, the force feedback device 5, and the gripper device 9 are connected in stages to form a sorting execution device. The sorting execution system has two sets of sorting execution devices, installed side by side in the direction of the conveyor belt, with a spacing of 0.9m. The force feedback device 5 can detect the force on the gripper device in real time. Once the force feedback device 5 detects an abnormal force on the gripper device, it will determine the abnormal situation through an algorithm. If it is determined that the gripper device 9 is impacted or scrapes against the conveyor belt 4, it will immediately control the parallel robotic arm device 6 to lift up and simultaneously control the gripper device 9 to open up, protecting the sorting execution system equipment from damage and the conveyor belt 4 from damage.

[0066] The sorting and identification system is used by the sorting execution device to confirm whether the sorted target is the correct one during the sorting process. It consists of two sets, each containing a strip light source 7 and a wide-angle industrial camera 8. The strip light source 7 is mounted on the upper side of the gripper device 9, with its long side perpendicular to the belt running direction and its emitting surface angled downwards towards the gripper device. The wide-angle industrial camera 8 is mounted on the upper side of the gripper device 9, on the same side as the strip light source 7, at a straight-line distance of 0.8m from the bottom of the gripper device 9. The lens of the wide-angle industrial camera 8 angles downwards towards the gripper device 9. Due to the high dust levels in the area, to prevent dust from entering the equipment and causing malfunctions, a dustproof protective shell is installed on the industrial camera and the strip light source. The dustproof protective shell for the wide-angle industrial camera is cylindrical, with a circular high-transmittance germanium glass top surface. The wide-angle industrial camera is installed inside the dustproof protective shell. The cleaning device, mainly consisting of air nozzles and a wiper system, is installed on the outside of the circular high-transmittance germanium glass. Regularly clean the protective glass to ensure the camera lens is clean and the images captured are clear. The strip light dustproof protective case is rectangular, with the strip light installed inside. The top surface of the strip light dustproof protective case is made of a highly transparent material, and the emitting side of the strip light is installed facing the top surface of the protective case.

[0067] The speed detection device 3 provides accurate, real-time speed data for the entire system, eliminating sorting errors. It includes a high-precision speed sensor and an adaptive bracket. The high-precision speed sensor is screwed onto the adaptive bracket. The adaptive bracket is connected to the sorting device support frame via a fixed crossbar. Its installation position is below the conveyor belt, flush against the inner surface of the conveyor belt, and able to adapt to belt vibrations without loosening.

[0068] The monitoring system uses high-definition night vision cameras as cameras. The cameras are installed around the conveyor belt and their field of view covers the material identification system and sorting execution device, enabling remote monitoring of the overall operation of the entire equipment.

[0069] The sorting box is installed on one side of the conveyor belt and consists of two chutes and a collection box. The chutes are U-shaped and roofless, with their long sides perpendicular to the belt's running direction and angled downwards at a 60-degree angle to the conveyor belt plane. The upper edge of the chutes is level with the edge of the conveyor belt, while the lower edge is 50cm lower than the conveyor belt plane. The lower end is welded integrally with the collection box. The chutes are 55cm wide. The sorting system picks up the residual carbon blocks from the conveyor belt and slides them from the chutes into the collection box.

[0070] The control server is installed near the sorting execution system and carries the backend management system, recognition algorithms, and control software. The control server hardware includes an electrical control cabinet, algorithm and management servers, a main control board, a robotic arm controller, a light source controller, a display screen, a communication module, a power supply module, and control buttons. The control buttons, including system start, reset, stop, and emergency stop buttons, are connected to the main control board via signal cables. The main control board communicates with the algorithm and management server via RS-485 communication. The robotic arm controller communicates with the algorithm and management server via RJ45 Ethernet communication. The display screen connects to the algorithm and management server via an HDMI interface. The light source controller communicates with the main control board via control cables. The power supply module connects to other modules via power cords, providing power to them. The algorithm and management server, equipped with a GPU and a deep learning artificial intelligence network model, enables fast and accurate image target recognition.

[0071] The backend management system runs on the control server. The system uses a visual interface to present test results and real-time monitoring footage, allowing operators to obtain key information and control the equipment's operating status. The visual interface is divided into four sections: system settings, testing functions, system management, and data analysis.

[0072] The system settings section covers equipment control, equipment status display, and system parameter settings; operators can remotely control industrial cameras, sorting actuators, gripping devices, and the system's start / stop and reset indicators; the operating status is presented using dynamic icons and real-time data to show the equipment system network and software status, helping operators to identify problems; the parameter settings can be adjusted according to production tasks and material characteristics.

[0073] The detection function connects to front-end industrial cameras and monitoring cameras to achieve low-latency, high-definition display of material dynamics in key areas. It also displays the category, location information, and other results output by the material identification system in real time, providing a basis for sorting decisions and providing operators with an intuitive, real-time, dynamic display of identification and sorting results.

[0074] The data analysis module mines system data and includes functions for historical image / data query, abnormal data recording, and statistical analysis of test results. Operators can use this to review production scenes and conduct data recaps; abnormal data records are archived to assist in troubleshooting; and statistical analysis can calculate material sorting accuracy, equipment failure rate, etc.

[0075] The system management section focuses on user and permission management, and sets up a user authentication system and permission allocation mechanism. Operators in different positions can access functional modules and data according to their responsibilities. Ordinary operators have basic operation and screen viewing permissions, while system administrators have full advanced permissions, ensuring hierarchical and orderly management.

[0076] Under the innovative system architecture constructed in this invention, precise detection, positioning, and rapid sorting of a material flow consisting of a mixture of electrolyte blocks and residual carbon blocks on a high-speed conveyor belt can be achieved. This system can not only accurately identify the residual carbon blocks but also simultaneously recognize their size, dimensions, and location coordinates. Simultaneously, the accompanying block-shaped residual carbon sorting execution system, based on information fed back from the detection model, drives the sorting execution device to precisely grasp the residual carbon blocks to be sorted, ensuring they fall accurately into a specially designed sorting bin. Furthermore, this system can adapt to changes in belt speed, accurately grasping residual carbon blocks even when the belt speed changes, without mistakenly grasping electrolyte blocks. In the event of a scrape, protection logic can be activated to protect the sorting equipment and the conveyor belt. This system employs a special gripper with a mechanically movable linkage design, which solves the problem of sorting out electrolyte blocks together when they are adjacent to residual carbon blocks. Current technologies have significant shortcomings in the anode residual carbon block cleaning process. When lumpy residual carbon blocks accidentally fall into the conveyor belt, subsequent processes cannot effectively identify and sort them, resulting in these blocks being directly transported to the crushing system along with electrolyte crust blocks. Once in the crushing system, the crushed material mixes into the electrolytic cell. This situation causes several negative effects: firstly, the crushed residual electrodes cannot be sold as complete products, directly leading to a sharp decline in residual electrode sales; secondly, the mixing of residual electrode fragments into the electrolytic cell drastically increases the amount of carbon slag during production, and the residual electrode fragments also increase the contact resistance between the anode carbon blocks and the electrolytic cell, significantly reducing electrolysis efficiency and thus increasing energy consumption. The system designed in this invention specifically overcomes the above problems, ensuring the high efficiency and stability of the production process from the root.

[0077] Reference Figure 2-3 The present invention discloses a method for sorting residual carbon blocks based on intelligent identification and control, comprising:

[0078] Step 1: The conveyor belt starts, and the upstream process performs preliminary crushing of the residual electrode. The crushed electrolyte blocks and residual electrode carbon blocks are mixed together and fall onto the conveyor belt. The conveyor belt transports the mixture of electrolyte blocks and residual electrode carbon blocks to the next stage process, namely the residual electrode carbon block sorting system.

[0079] Step 2: The operator turns the power knob on the control cabinet to power on the equipment, then presses the start button. The start indicator light illuminates, and the equipment enters the automatic start-up process, sequentially activating the material recognition supplementary light source 10, the sorting recognition supplementary light source 7, the material recognition industrial camera 11, the sorting recognition industrial camera 8, the speed detection system 3, the sorting execution system 13, the algorithm, and the system management software.

[0080] Step 3: The industrial camera 11 in the material recognition system takes continuous pictures. Simultaneously, the industrial camera drives the control of the material recognition supplementary lighting source 10, which illuminates the conveyor belt 4 and the transported materials while the industrial camera 11 is imaging. The timing of the light source 10's emission and the industrial camera 11's imaging is controlled to ensure that the industrial camera 11 can acquire clear object images with a relatively short exposure time. This also solves the motion blur problem caused by the high-speed operation of the conveyor belt 4. After the wide-angle industrial camera 11 captures high-definition images, it transmits the high-definition images to the algorithm server via an RJ45 network cable. The algorithm server performs image preprocessing on the acquired high-definition images; specifically, the acquired high-definition images are first scaled down to 640*640 pixels. This ensures that the image size is reduced without losing important information, saving time for subsequent preprocessing and deep learning model calculations. Gaussian filtering is then applied to the scaled images to reduce noise.

[0081] After preprocessing, an improved NanoDet network model is used for algorithmic recognition. However, the NanoDet network's accuracy is insufficient in scenarios with dense small targets, such as scrap sorting. Therefore, this patent proposes an improved loss function to enhance the model's accuracy in scrap sorting scenarios. Specifically, the Quality Focal Loss function is used to replace the original loss function, allowing the network to focus not only on targets of easy and difficult classification but also on the quality of the targets. This is particularly important for small target detection, as it allows for greater focus on small, difficult-to-classify objects. When an image is input into the NanoDet model, it first undergoes image preprocessing, adjusting the image to a fixed size and normalizing it. The lightweight Backbone (MobileNetV3) extracts image features. After feature extraction, the model uses BiFPN (Bidirectional Feature Pyramid Network) to fuse features at different scales, enhancing the model's ability to recognize multi-scale targets, especially small targets. Subsequently, the model uses a detection head to perform target classification and bounding box regression, predicting the target's category and location. Because NanoDet employs an anchor-free method, the prediction process directly regresses the target's location and category without requiring predefined anchor boxes. In the output stage, the model performs center point detection for each target, calculates the target confidence score, and uses non-maximum suppression (NMS) to remove redundant boxes. Finally, the model outputs the NMS-processed bounding boxes, category, and confidence score, representing the recognition result.

[0082] After obtaining the recognition results, the algorithm retrieves the preset image size parameters from the configuration file and converts the coordinate and size information from the acquired recognition information into real-world coordinates and target size. It also integrates the classification information from the recognition results; specifically, it obtains the original camera image resolution w*h from the configuration file and the size of the corresponding real-world object region in the original camera image. Simultaneously calculate the size of the input image to the model. The scaling factor to the resolution of the original camera image (Width scaling factor) and (Height proportionality coefficient); where,

[0083] ;

[0084] Obtain the size of the target detection box from the recognition results information. Then the actual size and width of the target can be calculated.

[0085] ;

[0086] Actual height dimensions of the target:

[0087] ;

[0088] Based on the same calculation principle, the target's coordinates in the real world can be calculated from the target's coordinates in the target detection results.

[0089] The identification result information is sent to the sorting execution system 13, and the result information is simultaneously plotted on a high-definition image and displayed on the interactive interface. The entire process, from industrial camera imaging to high-definition image transmission, algorithm recognition and coordinate system transformation, and result plotting and display, takes an extremely short time, achieving real-time performance comparable to the belt speed.

[0090] Step 4: The sorting execution algorithm obtains the identification result information, including the size, classification, coordinates, timestamp, and speed of the residual carbon block. This information is input into the sorting execution algorithm, which uses real-time sensors to acquire the object's speed *v* and an industrial camera to capture images, while simultaneously recording the acquisition time. By processing images, the time taken to process objects in the image is calculated. Displacement within:

[0091] ;

[0092] Calculate the relative distance between the target object and the gripper using the displacement formula:

[0093] ;

[0094] Where s is the initial distance, and Δs is the displacement of the object. The time required for the target object to reach the grab point is:

[0095] ;

[0096] To ensure synchronization between the grasping action and the target object, the robotic arm needs to be adjusted into position within time t3, where the robotic arm's movement time can be determined by its speed. The distance traveled is obtained as follows:

[0097] ;

[0098] Calculate synchronization wait time:

[0099] ;

[0100] After calculating the synchronization waiting time, it is determined whether the synchronization waiting time is positive. If it is negative, it means that the movement time of the residual carbon block is less than the action time of the sorting mechanism. The sorting mechanism will not be able to sort the carbon block. The algorithm controls the sorting mechanism to skip the current sorting and wait for the next sorting. If the synchronization waiting time is positive, the execution thread waits for △t2.

[0101] Specifically, when the residual carbon block is transported on the conveyor belt at 1.2 m / s through the material identification system, the material identification algorithm calculates the identification result information and the calculated object identification coordinate position information. According to the above algorithm, the synchronous waiting time Δt2 is calculated, and it is determined whether the Δt2 time is greater than 0. If Δt2 is greater than 0, it means that the movement of the robotic arm and the gripper device can sort out the residual carbon block in time. At this time, the program controls the robotic arm to wait for Δt2, and then controls the robotic arm to move directly from the initial position to the sorting position. At the same time, during the movement of the robotic arm, the program controls the gripper device to open. When the robotic arm moves to the sorting position, the target residual carbon block is just transported to the sorting position by the conveyor belt. The robotic arm and the gripper device are in time and space synchronization with the residual carbon block. At this time, the gripper device is controlled to clamp. After the gripper device clamps, the robotic arm is controlled to move up to the gripper detection position. Waiting for the sorting and identification system to sort and identify; if △t2 is less than 0, it means that the action time of the robotic arm and the action time of the gripper device are greater than the time it takes for the residual carbon block to be transported to the sorting position, and it will be impossible to sort in time. At this time, the sorting device is controlled not to move, and waits for the next sorting.

[0102] Step 5: During the movement of the sorting execution device according to the algorithm calculated in Step 4, based on the speed and target coordinates, and during the gripper device 9's gripping and releasing process, if the sorting execution device unexpectedly collides with an obstacle, or scrapes or collides with a large irregular electrolyte on the belt or the conveyor belt itself, the force feedback device 5 located on the sorting execution device will promptly detect the force on the sorting execution device and determine whether the force is abnormal. If the force feedback device 5 detects a scraping or collision, the control algorithm will immediately terminate the actions of the parallel robotic arm 6 and the gripper device 9, and immediately control the parallel robotic arm 6 to retract and control the gripper device 9 to open, preventing further scraping and collisions and protecting the safety of the sorting equipment and the conveyor belt.

[0103] Step Six: During the sorting process of the sorting execution device gripping the residual carbon block according to the sorting execution flow calculated in Step Four, when the gripper device 9 grasps the target object and the algorithm controls the parallel robotic arm 6 to move to the detection position, the algorithm controls the sorting identification supplementary light source 7 to work, providing supplementary light to the grasped target object. Simultaneously, it controls the sorting identification industrial camera 8 to acquire images. The sorting identification camera 8 is a high-resolution wide-angle industrial camera capable of acquiring clear images. The acquired images are sent to the algorithm server via an RJ45 Ethernet cable. The algorithm server deploys a lightweight deep learning object detection network, which can quickly and accurately identify the acquired high-definition image information and output the identification results. After obtaining the identification results, the algorithm controls the sorting execution device to perform subsequent operations. If the identification result information shows that the grasped target object is a residual carbon block, the sorting execution device is controlled to move to the release position according to the given motion path, speed, and target position. Once in position, the gripping device 9 is released, releasing the residual carbon block. The residual carbon block slides down the chute 2 into the residual carbon block collection box 1. The sorting execution device then moves to the starting position and waits for the next sorting. If the identification result information shows that the target object being gripped is not a residual carbon block or no object has been gripped, the algorithm controls the gripper device to immediately release, releasing the object back onto the conveyor belt. Then, the parallel robotic arm 6 is controlled to move to the starting position and wait for the next sorting. Example:

[0104] The material identification system is installed 1m directly above the conveyor belt, and the sorting execution device is installed 3m away from the material identification system along the direction of belt travel. The chute 2 is installed to the side of the sorting execution device, with the chute opening aligned with the gripper device. The collection box 1 is installed below the chute.

[0105] The conveyor belt operates normally, transporting the residual carbon blocks and electrolyte mixture on conveyor belt 4. Simultaneously, the speed detection system 3 detects the belt speed, i.e., the material's running speed.

[0106] The residual carbon block and electrolyte mixture are transported by conveyor belt to the material identification system. The material identification system collects images, performs algorithm processing, and outputs the identification results, including the target location and target size of the residual carbon block.

[0107] The material recognition algorithm sends the recognition result to the sorting execution algorithm. The sorting execution algorithm calculates the delay waiting time of the parallel robotic arm 6 based on the current position, speed and size of the material, and then executes the delay waiting.

[0108] After the parallel robotic arm 6 has finished waiting, it performs the sorting action. First, the parallel robotic arm 6 moves from the initial position to the sorting position, and at the same time, the gripper device 9 opens. When both the parallel robotic arm 6 and the gripper device are in position, the residual carbon block has just moved to the sorting position. At this time, the algorithm controls the gripper device 9 to clamp.

[0109] After the gripper device 9 clamps the object, the parallel robotic arm 6 rises 200mm to the detection position. At this time, the sorting and identification system (sorting and identification camera 8 and strip lighting source 7) starts working and acquires images. Simultaneously, the sorting algorithm calculates whether the object being gripped is the target object. If the gripping is empty or incorrect, the gripper device 9 is opened, allowing the object to fall back onto the conveyor belt. Then, the parallel robotic arm 6 moves back to its initial position to await the next sorting action. If the sorting algorithm determines that the object being gripped is a piece of residual carbon, the parallel robotic arm 6 moves above the chute 2, and the gripper device 9 is opened, allowing the residual carbon to fall into the chute and slide down into the collection box. The algorithm then controls the parallel robotic arm 6 to move back to its initial position to await the next sorting action.

[0110] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for sorting residual carbon blocks based on intelligent identification and control, characterized in that, include: The residual electrode is broken, and the resulting electrolyte block and residual electrode carbon block fall onto the conveyor belt. Images were obtained by taking photographs of the electrolyte block and the residual carbon block on the conveyor belt; The image is used to perform algorithmic recognition to obtain recognition information; the recognition information includes the size information, classification information, coordinate information, timestamp information, and speed information of the residual carbon block; Based on the identification information, the residual carbon block is gripped by a robotic arm, including: The real-time sensor acquires the velocity of the residual carbon block and uses a camera to capture images, while simultaneously recording the acquisition time; Calculate the displacement of the residual carbon block during the image recognition time; Based on the displacement, the relative distance between the target and the robotic arm gripper is calculated; Calculate the time required for the robotic arm gripper to grasp the target; Calculate the adjustment time of the robotic arm gripper; The difference between the time required to achieve the target and the adjustment time is the waiting time; If the waiting time is positive, the robotic arm gripper will promptly remove the residual carbon block. If the waiting time is negative, then it cannot be captured and must wait for the next capture. The process of using the NanoDet network model to perform algorithmic recognition on the image to obtain recognition information includes: The image is preprocessed by adjusting it to a fixed size and normalizing it. After processing, the image features are extracted using a lightweight Backbone neural network; The features at different scales are fused using a BiFPN neural network; After fusion, target classification and bounding box regression are performed to predict the target's category and location; For each target, perform center point detection and output the bounding box, category, and confidence score of the target; Based on the coordinate and size information in the identification information, it is converted into real-world coordinate information and target size, which includes: Obtain the original image resolution (w*h) and the size of the corresponding real-world object region in the original image. ; Simultaneously calculate the size of the image input to the NanoDet network model. The scaling factor relative to the resolution of the original image ; Obtain the size of the target detection box from the recognition information. Calculate the actual size of the target; Wherein, the actual width of the target is; ； The actual height of the target is; ； In the formula: ; It also includes the following steps: The identification information is input into the sorting execution algorithm. The object's velocity v is obtained through real-time sensors, and images are acquired using an industrial camera, while the acquisition time is recorded. By processing images, the time taken to process objects in the image is calculated. Displacement within; ； Calculate the relative distance between the target and the gripper based on the displacement formula; ； Where s is the initial distance, Δs is the displacement of the object, and the time required for the target to reach the grab point is; ； To ensure synchronization between the grasping action and the target object, the robotic arm needs to be adjusted into position within time t3, where the robotic arm's movement time can be determined by its speed. The distance traveled is obtained; ； Calculate the synchronization wait time; ； After calculating the synchronization waiting time, it is determined whether the synchronization waiting time is positive. If it is negative, it means that the movement time of the residual carbon block is less than the action time of the sorting mechanism. The sorting mechanism will not be able to sort the carbon block. The algorithm controls the sorting mechanism to skip the current sorting and wait for the next sorting. If the synchronization waiting time is positive, the execution thread waits for △t2.

2. The method for sorting residual carbon blocks based on intelligent identification and control according to claim 1, characterized in that, The robotic arm gripper is equipped with a force feedback device at its top.

3. The method for sorting residual carbon blocks based on intelligent identification and control according to claim 2, characterized in that, During the gripping process of the robotic arm, supplemental lighting is provided to the residual carbon block.

4. The method for sorting residual carbon blocks based on intelligent identification and control according to claim 3, characterized in that, The gripper of the robotic arm includes a gripper, a linear cylinder, and a valve for controlling the air source. The linear cylinder is connected to the upper part of the mechanical moving link. The two air outlets of the valve are connected to the two air holes of the linear cylinder. The control signal line of the valve is connected to the main control board. When the main control board controls the valve to open or close, high-pressure gas enters the linear cylinder through the connecting air pipe. The linear cylinder push rod moves, driving the mechanical moving link to move, thereby realizing the gripper's gripping and releasing action.

5. A sorting system for residual carbon blocks based on intelligent identification and control, characterized in that, include: The crushing module is used to crush the residual electrode and drop the resulting electrolyte block and residual electrode carbon block onto the conveyor belt. The imaging module is used to take pictures of the electrolyte block and the residual carbon block on the conveyor belt to obtain images; The recognition module is used to perform algorithmic recognition on the image using a NanoDet network model to obtain recognition information; The gripping module is used to grip the residual carbon block using a robotic arm based on the identification information; The process of using the NanoDet network model to perform algorithmic recognition on the image to obtain recognition information includes: The image is preprocessed by adjusting it to a fixed size and normalizing it. After processing, the image features are extracted using a lightweight Backbone neural network; The features at different scales are fused using a BiFPN neural network; After fusion, target classification and bounding box regression are performed to predict the target's category and location; For each target, perform center point detection and output the bounding box, category, and confidence score of the target; Based on the coordinate and size information in the identification information, it is converted into real-world coordinate information and target size, which includes: Obtain the original image resolution (w*h) and the size of the corresponding real-world object region in the original image. ; Simultaneously calculate the size of the image input to the NanoDet network model. The scaling factor relative to the resolution of the original image ; Obtain the size of the target detection box from the recognition information. Calculate the actual size of the target; Wherein, the actual width of the target is; ； The actual height of the target is; ； In the formula: .

Images