Intelligent sorting device and method for waste concrete blocks

Abstract

The application provides a kind of abandoned concrete block intelligent sorting equipment and method, relating to solid waste resource processing field.The abandoned concrete block intelligent sorting equipment includes conveying mechanism, the conveying belt of conveying mechanism is supported by face increasing wheel, and the upper part of conveying belt is convex, face increasing wheel is supported by the cooperation of support and spring, fixed frame is arranged on both sides of conveying mechanism, rotating rod is rotatably installed on two fixed frames, another end of rotating rod is fixedly provided with flattening plate, and two flattening plates are arranged in the shape of "eight".The abandoned concrete block intelligent sorting equipment and method can accurately identify concrete and various impurities, target grabbing and sorting, solve the problems of low efficiency, poor environment, large precision fluctuation of manual sorting, incomplete impurity removal of traditional physical sorting, insufficient purity, and high dependence on manual operation.

Description

Technical Field

[0001] This invention relates to the field of solid waste resource utilization, specifically to an intelligent sorting device and method for waste concrete blocks. Background Technology

[0002] During the demolition and reconstruction of existing roads and bridges, there is a common problem of waste concrete being indiscriminately dumped as garbage. This not only occupies a large amount of land resources and pollutes the environment, but also wastes resources. Relevant data shows that construction waste and the raw materials needed to produce building materials such as cement and concrete have similar components and minerals, making it a resource that has been misplaced. Therefore, the use of waste concrete to prepare recycled aggregates and recycled concrete is strongly advocated by many experts and scholars.

[0003] The efficient sorting and purification of waste concrete is the core bottleneck to improving the comprehensive utilization rate. However, among the existing sorting technologies, manual sorting is inefficient and causes serious dust and noise pollution in the workshop. Physical sorting methods such as magnetic separation and air separation can only remove some impurities in a targeted manner, and it is difficult to completely remove wood, plastic and non-ferrous metals, resulting in insufficient sorting purity, which seriously restricts the quality of recycled aggregates from waste concrete. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent sorting device and method for waste concrete blocks, which solves the problems of low efficiency in manual sorting and insufficient purity in physical sorting methods such as magnetic separation and air separation.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: an intelligent sorting device for waste concrete blocks, including a conveying mechanism, wherein the conveyor belt of the conveying mechanism is supported by a booster wheel, and the upper part of the conveyor belt is raised, and the booster wheel is supported by a bracket and a spring. The conveying mechanism has fixed frames on both sides, and a rotating rod is rotatably mounted on each of the two fixed frames. A flat plate is fixedly mounted on the other end of the rotating rod. The two flat plates are arranged in a figure-eight shape. A rack is mounted on the increasing wheel, and a gear meshes on the rack. The rotating rod is located at the axis of the gear cavity, and a coil spring is fixedly mounted between the two. The rack moves with the increasing wheel and drives the gear to rotate, which is used to adjust the initial torque of the coil spring. A gantry frame is installed on the outside of the conveying mechanism, and a vision recognition module is installed on the gantry frame. An industrial robot is installed at the tail end of the conveying mechanism, and the output end of the industrial robot is used to install grippers.

[0006] Preferably, the belt-enlarging wheel includes two wheel bodies and two connecting plates, and both wheel bodies are fixed to the two connecting plates. The lower wheel body is used to support the bottom of the conveyor belt and make it concave.

[0007] Preferably, both connecting plates are provided with sliding grooves, and fixed blocks are slidably connected in the sliding grooves. Fixed members are fixedly connected to the two fixed blocks on their opposite sides, and the positions of the fixed members are always fixed.

[0008] Preferably, the two connecting plates are fixed to the two racks respectively, and the connecting plates are parallel to the racks.

[0009] Preferably, the paving plate is parallel to the inclined surface at the top of the conveyor belt, and the paving plate is located at the inclined surface at the rear end of the conveyor belt.

[0010] Preferably, the conveying mechanism further includes two rotating wheels and several support frames. The two rotating wheels are connected by a conveyor belt and driven to rotate by an external motor. Both ends of the two rotating wheels are rotatably connected to the support frames adjacent to them.

[0011] Preferably, the visual recognition module includes a color industrial camera, a near-infrared spectral camera, and an LED lighting source. The color industrial camera and the near-infrared spectral camera are both installed below the gantry beam. The LED lighting source consists of two symmetrically arranged high color rendering index light sources. The color industrial camera and the near-infrared spectral camera are triggered to capture images simultaneously, respectively collecting the appearance characteristics and spectral characteristics of the material.

[0012] Preferably, the spring is fixed between the bottom of the fixing member and the connecting member, and the extension and contraction direction of the spring is consistent with the movement direction of the connecting member.

[0013] Preferably, there is a gap between the two flat surfaces.

[0014] A sorting method for a smart sorting device for waste concrete blocks includes the following steps: S1: Feed the construction waste mixture, which has been initially crushed to 50-300mm, into the conveyor belt of the conveying mechanism, start the conveying mechanism, and drive the conveyor belt to run at a speed of 0.5-1.2m / s by rotating the wheel; S2: During the operation of the conveyor belt, the bulging wheel supports the conveyor belt in a raised shape with the cooperation of the bracket and spring. When the material moves to the spreading plate area, the spreading plate arranged in a figure-eight shape spreads the material under the action of the coil spring torque. At the same time, the bulging wheel moves relative to the fixed block through the slide groove according to the thickness of the material, driving the rack drive gear to rotate and dynamically adjusting the initial torque of the coil spring to ensure that the material forms a single layer of material flow with a thickness of 50-100mm. S3: When the material passes under the gantry, the color industrial camera and near-infrared spectral camera of the vision recognition module are triggered to shoot simultaneously, and the LED lighting source provides uniform illumination, respectively collecting the appearance features such as color, texture, and shape of the material and the spectral features of the material. S4: The visual recognition module transmits the acquired image data to the industrial-grade GPU computing unit. The industrial-grade GPU computing unit has a built-in deep learning model based on the improved YOLOv7 architecture. Through the deep learning model based on the improved YOLOv7 architecture, multimodal feature fusion and intelligent recognition are performed to output the material category. S5: The central control system receives the identification results, combines them with the conveyor belt encoder data to predict the time when the material will arrive at the industrial robot's working area, generates sorting instructions based on the impurity category, and controls the industrial robot to switch grippers or adjust gripping parameters. S6: When the impurity reaches the predetermined working position, the industrial robot performs a grasping action according to the planned motion trajectory, removes the impurity from the conveyor belt and throws it into the corresponding classified impurity collection bin. If the single grasping fails, a second grasping is triggered. S7: Target concrete blocks that are not identified as impurities are continuously transported to the end of the conveyor belt and fall into the target concrete collection bin for subsequent crushing or processing. S8: During system operation, the system receives material sampling results through the online learning module, incrementally trains and updates the parameters of the deep learning model, and at the same time, the central control system monitors the operating status of each module in real time to ensure stable sorting efficiency and purity.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This intelligent sorting equipment and method for waste concrete blocks utilizes a figure-eight shaped spreader plate. This not only lays the material flat, increasing its exposure rate, but also prevents it from being thrown out from the sides during transport, drawing it towards the center. This facilitates detection by the visual recognition module. Through the cooperation of the visual recognition module and an industrial robot, it can accurately identify concrete and various impurities, targeting and sorting them to achieve high-purity resource recovery from waste concrete. The entire sorting process is automated, balancing environmental protection and economic benefits. It solves the problems of low efficiency, harsh environment, and large accuracy fluctuations in manual sorting, as well as the incomplete removal of impurities and insufficient purity in traditional physical sorting, and high dependence on manual labor. When the material weight increases, the rack and pinion drive the gears to rotate, thereby adjusting the initial torque of the coil spring. When the material is heavier, the coil spring torque increases, enhancing the flattening force of the spreader plate; when the material is lighter, the torque decreases, avoiding excessive compression and damage to the material.

[0016] This intelligent sorting equipment and method for waste concrete blocks utilizes a conveyor belt supported by rollers, which creates a raised upper section. This allows the material to pass through both uphill and downhill sections during transport. During the uphill section, the material's gravity helps to loosen the stacked material, reducing adhesion and lowering resistance for the "eight"-shaped slab, making slab leveling easier and more uniform. During the downhill section (the inclined surface at the rear of the conveyor belt), the material moves along the inclined surface under gravity, allowing the torque applied by the slab to act more directly on the material surface. This avoids incomplete leveling caused by material rolling during planar transport, ensuring a single-layer material flow. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a schematic diagram of the structure of the flat plate of the present invention; Figure 3 This is a cross-sectional view of the gear, coil spring, and rotating rod of the present invention; Figure 4 This is a cross-sectional view of the front view of the conveyor belt of the present invention; Figure 5 This is a cross-sectional view of the side view of the connecting plate of the present invention; Figure 6 This is a schematic diagram of the structure of the surface-enhancing wheel and rack of the present invention; Figure 7 This is a schematic diagram of the meshing of the rack and gear of the present invention; Figure 8 This is a schematic diagram of the connecting plate and rack of the present invention.

[0018] Among them, 1. Conveying mechanism; 101. Conveyor belt; 102. Rotating wheel; 2. Surface-enlarging wheel; 201. Wheel body; 202. Connecting plate; 3. Rotating rod; 4. Spreading plate; 5. Rack; 6. Gear; 7. Coil spring; 8. Gantry frame; 9. Vision recognition module; 10. Industrial robot; 11. Slide; 12. Fixing block; 13. Fixing component; 14. Spring; 15. Baffle. Detailed Implementation

[0019] like Figures 1-8As shown, an intelligent sorting device for waste concrete blocks includes a conveying mechanism 1. The conveyor belt 101 of the conveying mechanism 1 is supported by a convex wheel 2, making the conveyor belt 101 convex. The convex wheel 2 is supported by a bracket and a spring 14. The spring 14 is fixed between the bottom of the fixing member 13 and the connecting member, and the extension and retraction direction of the spring 14 is consistent with the movement direction of the connecting member. The convex wheel 2 includes two wheel bodies 201 and two connecting plates 202, and both wheel bodies 201 are fixed to the two connecting plates 202. The lower wheel 201 supports the bottom of the conveyor belt 101 and makes it concave. The convex wheel 2 uses a combination of double wheels 201 and connecting plates 202. The lower wheel 201 supports the conveyor belt 101 to form a concave shape. Since the upper wheel 201 supports the conveyor belt 101 to form a convex shape, and the lower wheel 201 presses against the conveyor belt 101 to form a concave shape, the supporting forces of the two on the conveyor belt 101 are opposite in direction but move synchronously, avoiding the conveyor belt (101) from slackening due to unilateral force. Both connecting plates 202 are provided with grooves 11. A fixed block 12 is slidably connected within the chute 11. Two fixed blocks 12 are each fixedly connected to a fixing element 13 on their opposite sides, ensuring the fixing element 13 remains in a fixed position. The fixing element 13 can be fixed to the bearing base. Two connecting plates 202 are respectively fixed to two racks 5, and the connecting plates 202 are parallel to the racks 5. The fixed and parallel connection of the connecting plates 202 and racks 5 ensures smooth transmission of the racks 5 when the spreading wheel 2 moves, thereby achieving linear adjustment of the torque of the coil spring 7, making the spreading pressure of the spreading plate 4 on the material uniform and controllable. To avoid localized material stacking or excessive compression due to transmission misalignment, the conveying mechanism 1 also includes two rotating wheels 102 and several support frames. The two rotating wheels 102 are connected by a transmission belt 101. The rotating wheels 102 are driven to rotate by an external motor. Both ends of the two rotating wheels 102 are rotatably connected to the support frames adjacent to them. A baffle 15 is fixedly installed between a set of support frames at the feed end of the conveyor belt 101. The baffle 15 is located at the end of the conveyor belt 101 to prevent material from sliding out from one end of the conveyor belt 101.

[0020] The conveying mechanism 1 has fixed frames on both sides, and rotating rods 3 are rotatably mounted on both fixed frames. A spreading plate 4 is fixedly mounted on the other end of the rotating rod 3. The two spreading plates 4 are arranged in a figure-eight shape with a gap between them. A rack 5 is provided on the expanding wheel 2, and a gear 6 meshes on the rack 5. The rotating rod 3 is located at the axis of the cavity of the gear 6, and a coil spring 7 is fixedly installed between the two. The rack 5 moves with the expanding wheel 2 and drives the gear 6 to rotate, which is used to adjust the initial torque of the coil spring 7. The spreading plate 4 is parallel to the upper inclined surface of the conveyor belt 101 and is located at the rear inclined surface of the conveyor belt 101. The spreading plate 4 is parallel to the upper inclined surface of the conveyor belt 101 and is located at the rear inclined surface, which fits the material conveying path and realizes the precise spreading of materials in the inclined area, avoiding the accumulation of materials due to gravity sliding down the inclined surface.

[0021] A gantry frame 8 is provided on the outer side of the conveying mechanism 1, and a vision recognition module 9 is provided on the gantry frame 8. An industrial robot 10 is provided at the tail end of the conveying mechanism 1. The output end of the industrial robot 10 is used to install grippers. The upper part of the conveyor belt 101 is raised by the cooperation of the surface wheel 2 and the spring 14. Combined with the torque adjustment structure of the "eight"-shaped flat plate 4 and the rack 5, gear 6, and coil spring 7, the material is self-adaptively and evenly spread, avoiding stacking and obstruction. The vision recognition module 9 and the industrial robot 10 are linked to achieve full automation of the sorting process without human intervention. This solves the problems of low efficiency and large accuracy fluctuations in traditional sorting, and greatly improves the stability and accuracy of waste concrete block sorting. The vision recognition module 9 includes color Industrial cameras, near-infrared spectral cameras, and LED lighting sources are used. The color industrial camera and the near-infrared spectral camera are both installed below the gantry frame 8 crossbeams, 3-5m away from the upper surface of the conveyor belt 101. The color industrial camera has a resolution of no less than 5 million pixels and a frame rate of no less than 30fps. The near-infrared spectral camera has a spectral range of 900-1700nm and a spectral resolution of no more than 10nm. The LED lighting source consists of two sets of symmetrically arranged high color rendering index light sources with a color rendering index Ra≥90 and illuminance uniformity≥85%. They illuminate the conveyor belt 101 at an angle of 30-60°. The color industrial camera and the near-infrared spectral camera are triggered synchronously to capture the appearance characteristics and spectral characteristics of the material, respectively.

[0022] A method for intelligent sorting equipment for waste concrete blocks further includes the following steps: S1: The construction waste mixture, initially crushed to 50-300mm, is fed into the conveyor belt 101 of the conveying mechanism 1. The conveying mechanism 1 is started, and the conveyor belt 101 is driven to run at a speed of 0.5-1.2m / s by the rotating wheel 102. S2: During the operation of the conveyor belt 101, the convex roller 2 supports the conveyor belt 101 in a raised shape with the cooperation of the bracket and the spring 14. When the material moves to the area of ​​the spreading plate 4, the spreading plate 4 arranged in a figure-eight shape spreads the material under the torque of the coil spring 7. At the same time, the convex roller 2 moves relative to the fixed block 12 through the slide groove 11 according to the thickness of the material, driving the rack 5 to drive the gear 6 to rotate, dynamically adjusting the initial torque of the coil spring 7, and ensuring that the material forms a single layer of material flow with a thickness of 50-100mm. S3: When the material passes under the gantry 8, the color industrial camera and near-infrared spectral camera of the vision recognition module 9 are triggered to shoot simultaneously, and the LED lighting source provides uniform illumination, respectively collecting the appearance features such as color, texture, and shape of the material and the spectral features of the material. S4: The visual recognition module 9 transmits the acquired image data to the industrial-grade GPU computing unit. The industrial-grade GPU computing unit has a built-in deep learning model based on the improved YOLOv7 architecture. Through the deep learning model based on the improved YOLOv7 architecture, multimodal feature fusion and intelligent recognition are performed to output the material category. S5: The central control system receives the identification results, combines the encoder data of the conveyor belt 101 to predict the time when the material arrives at the working area of ​​the industrial robot 10, generates sorting instructions according to the impurity category, and controls the industrial robot 10 to switch grippers or adjust gripping parameters. S6: When the impurity reaches the predetermined working position, the industrial robot 10 performs a grasping action according to the planned motion trajectory, removes the impurity from the conveyor belt 101 and throws it into the corresponding classified impurity collection bin. If the single grasping fails, a second grasping is triggered. S7: Target concrete blocks that are not identified as impurities are continuously transported to the end by conveyor belt 101 and fall into the target concrete collection bin for subsequent crushing or processing. S8: During system operation, the system receives material sampling results through the online learning module, incrementally trains and updates the parameters of the deep learning model, and at the same time, the central control system monitors the operating status of each module in real time to ensure stable sorting efficiency and purity.

[0023] The standardized steps throughout the entire process cover feeding, adaptive leveling, multimodal acquisition, AI recognition, precise sorting, material collection, and model optimization, achieving automated and intelligent sorting. Through dynamic leveling, precise recognition, fault-tolerant capture, and online model updates, it solves the problems of low efficiency, low purity, and high reliance on manual labor in traditional sorting, while reducing labor costs and equipment wear and tear, balancing economic benefits and environmental protection requirements.

[0024] In use, an external motor drives the rotating wheel 102 of the conveying mechanism 1 to rotate, causing the conveyor belt 101 to run smoothly. Construction waste mixture initially crushed to 50-300mm is fed into the conveyor belt 101. The support frame provides stable support for the rotating wheel 102 and the conveyor belt 101, ensuring no deviation or shaking during the conveying process. The expanding wheel 2 supports the conveyor belt 101 through the cooperation of the double wheel body 201 and the connecting plate 202. The upper wheel body 201 makes the upper part of the conveyor belt 101 convex, and the lower wheel body 201 makes the bottom of the conveyor belt 101 concave, forming a convex-concave conveying trajectory. The connecting plate 202 of the expanding wheel 2 is slidably connected to the fixing block 12 through the sliding groove 11. The fixing member 13 is fixed to the bearing base surface to provide a stable reference. One end of the spring 14 is connected to the bottom of the fixing member 13, and the other end is connected to the connecting plate 202. The extension and contraction direction of the spring 14 is perpendicular to the connecting plate 202. The moving direction of plate 202 is consistent, providing elastic support for the surface-enhancing wheel 2. When the material moves to the inclined surface at the rear end of the conveyor belt 101, the figure-eight shaped spreading plate 4 can prevent the material from being thrown out from both sides during the conveying process, and bring the material closer to the middle, which is beneficial for the subsequent detection by the visual recognition module 9. Moreover, the figure-eight shaped spreading plate 4 (parallel to the inclined surface of the conveyor belt 101) applies a spreading force to the material under the torsion of the coil spring 7, avoiding material stacking and occlusion, increasing the surface area exposure range of each material block, allowing the color camera to fully capture color, texture, and shape features, and the near-infrared spectral camera to accurately collect material spectral information, avoiding feature loss and AI misjudgment caused by occlusion, and reducing mutual interference between materials, avoiding the overlap of impurities with similar appearances with concrete blocks, which would cause the AI ​​model to be unable to distinguish individual categories, reducing the probability of missed judgment and misjudgment.

[0025] Next, as the weight of the material increases, the surface roller 2 slides relative to the fixed block 12 along the chute 11 under the pressure of the material, driving the rack 5, which is fixed and parallel to it, to move synchronously. At this time, the spring 14 begins to compress, and the rack 5 meshes with the drive gear 6 to rotate, thereby adjusting the initial torque of the coil spring 7. When the material is heavier, the torque of the coil spring 7 increases, and the flattening force of the flattening plate 4 is enhanced; when the material is lighter, the torque decreases to avoid excessive compression and damage to the material. A gap is reserved between the two flattening plates 4 to prevent small materials from getting stuck and to avoid excessive compression of the material. Finally, the material forms a single-layer material flow on the conveyor belt 101, increasing the surface area exposure range of each material block and clearing obstacles for subsequent visual recognition.

[0026] It should be noted that, since the conveyor belt 101 is supported by the surface roller 2 and the upper part of the conveyor belt 101 is convex, when the conveyor belt 101 is conveying materials, the materials will go through the uphill and downhill stages. During the uphill stage, the gravity component of the materials is used to make the stacked materials naturally loosen, reduce the tight adhesion between materials, reduce the resistance of the "eight"-shaped slab 4 for the slab spreading action, and make the slab spreading more labor-saving and uniform. During the downhill stage (the inclined surface of the rear end of the conveyor belt 101) is arranged parallel to the slab 4. The materials move along the inclined surface of the conveyor belt 101 under the action of gravity. The torque applied by the slab 4 can act more directly on the surface of the materials, avoiding the incomplete slab spreading caused by the easy rolling of materials during flat conveying, and ensuring the formation of a single-layer material flow.

[0027] Next, 3-5m below the gantry frame's 8 crossbeams and from the upper surface of conveyor belt 101, a color industrial camera and a near-infrared spectral camera are deployed simultaneously. Two sets of high color rendering index LED light sources are symmetrically arranged on both sides to effectively eliminate ambient light interference, material surface reflections, and shadows, ensuring the uniformity and stability of the acquired image brightness. The color industrial camera captures the material's color (RGB three-channel information), texture (sand grain distribution, cracks), and shape (outline, aspect ratio) appearance features in real time; the near-infrared spectral camera captures the molecular structure spectral features of the material. Different materials have different absorption or reflectivity of near-infrared light (e.g., concrete has strong absorption peaks at 1450nm and 1940nm, while plastic has a characteristic peak at 1720nm), achieving essential differentiation at the material level. The image acquisition unit is linked with the encoder of conveyor belt 101, binding each frame of acquired image to the real-time position of conveyor belt 101, providing a precise coordinate reference for subsequent material positioning and grasping timing planning. Color images and near-infrared spectral images are simultaneously transmitted to an industrial-grade GPU computing unit, which incorporates a deep learning model based on an improved YOLOv7 architecture. The model input layer uses a multimodal feature fusion layer to weightedly fuse appearance features and material spectral features (combining an attention mechanism to strengthen key difference features), breaking through the limitations of traditional vision that relies solely on appearance for recognition. The model analyzes the fused features using trained parameters, accurately classifying materials into "target concrete blocks" or "wood, plastic, non-ferrous metals, bricks" impurities. It also outputs the precise bounding box of each material through a bounding box regression algorithm. Combined with encoder data from conveyor belt 101, it calculates the real-time coordinates of the material in the width direction (X-axis) and length direction (Y-axis) of conveyor belt 101, and predicts the time it takes for the material to arrive at the working area of ​​industrial robot 10. The central control system receives the "material category, location coordinates, and arrival time" data output by the AI ​​model and generates sorting instructions according to preset logic: if it is determined to be an impurity with a size ≥ 5mm, it specifies the type of end effector (vacuum suction cup or mechanical gripper) according to the type of impurity (such as lightweight wood or plastic, hard metal or brick) and matches the corresponding gripping parameters; if it is the target concrete block, no sorting instructions are generated and its conveying trajectory is retained.

[0028] After receiving sorting instructions from the central control system, the industrial robot 10 generates the optimal motion trajectory through a motion planning algorithm (such as the RRT* algorithm) and completes posture adjustment before the impurity reaches the predetermined working position. When the impurity arrives at the gripping area, the robot switches its end effector according to the instructions to perform gripping or suction actions with a repeatability accuracy of ≤±0.1mm, quickly removing the impurity from the conveyor belt 101 and throwing it into the corresponding classified impurity collection bin (such as wood / paper bin, plastic bin, non-ferrous metal bin, brick bin). The end effector is equipped with a vacuum pressure sensor (suction type) or a force sensor (clamping type) to detect the gripping status in real time: if a single gripping fails (e.g., the pressure does not reach the threshold), the control system immediately triggers a second gripping instruction; if two consecutive gripping failures occur, the backup pneumatic pusher at the end of the conveyor belt 101 initiates a second interception to ensure the impurity rejection rate. The target concrete blocks that are not identified as impurities are continuously conveyed to the end by conveyor belt 101 and fall into the target concrete collection bin. The bin is protected from arching by a vibration unloading device at the bottom. The blocks are then conveyed to the next crushing or processing step to produce high-purity recycled aggregate.

[0029] It should be noted that the original YOLOv7 C3 module was replaced with a lightweight C2f module, and a CBAM attention mechanism was embedded after each C2f module. This enhanced the texture or material differences between concrete and impurities through channel attention branches (weights calculated by MLP after compressing feature maps) and spatial attention branches (convolution after concatenation of global average pooling and global max pooling). The sampled labeled data was divided into an incremental dataset containing newly added material samples (such as new plastic or wood impurities) and a retention dataset containing historical training samples of concrete and common impurities, with random sampling used as the retention set. The training strategy was to freeze the first 8 layers of the improved YOLOv7 backbone network (only fine-tuning the last 4 layers, the Neck, and the detection head) to avoid destroying the learned basic features. LWF knowledge distillation was employed: the currently deployed model was used as the "teacher model," and the newly trained model was used as the "student model." The KL divergence loss between the student model output and the teacher model output is calculated to constrain the model to retain historical knowledge. EWC weight constraints are used to avoid parameter updates causing a decrease in historical task accuracy for important parameters (calculated through an information matrix derived from the mean squared gradient of the dataset). After incremental training and model validation, the corresponding layer parameters of the original model are incrementally replaced with the new model parameters (only fine-tuning layers are updated, no full replacement is needed). The replacement process uses a "hot update" method to avoid interrupting system operation. Simultaneously, the central control system monitors the operating status of each module in real time: key indicators such as conveyor belt speed deviation, visual recognition module frame rate, industrial robot grasping success rate, and sorting purity are monitored. When these indicators exceed thresholds, alarms are triggered and parameters are automatically adjusted (e.g., reducing conveyor belt speed or recalibrating visual coordinates) to ensure stable sorting efficiency and purity.

[0030] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent sorting device for waste concrete blocks, comprising a conveying mechanism (1), characterized in that: The conveyor belt (101) of the conveying mechanism (1) is supported by the face-enlarging wheel (2), and the upper part of the conveyor belt (101) is raised. The face-enlarging wheel (2) is supported by the cooperation of the bracket and the spring (14). The conveying mechanism (1) has fixed frames on both sides, and rotating rods (3) are rotatably installed on both fixed frames. A flat plate (4) is fixedly installed at the other end of the rotating rod (3). The two flat plates (4) are arranged in a figure-eight shape. A rack (5) is provided on the surface-enlarging wheel (2). A gear (6) meshes on the rack (5). The rotating rod (3) is located at the axis of the cavity of the gear (6), and a coil spring (7) is fixedly installed between the two. The rack (5) moves with the surface-enlarging wheel (2) and drives the gear (6) to rotate, which is used to adjust the initial torque of the coil spring (7). A gantry (8) is provided on the outside of the conveying mechanism (1), a vision recognition module (9) is provided on the gantry (8), and an industrial robot (10) is provided at the tail end of the conveying mechanism (1). The output end of the industrial robot (10) is used to install grippers. The surface-enlarging wheel (2) includes two wheel bodies (201) and two connecting plates (202), and both wheel bodies (201) are fixed to the two connecting plates (202). The lower wheel body (201) is used to support the bottom of the conveyor belt (101) and make it concave. Both of the connecting plates (202) are provided with a sliding groove (11), and a fixing block (12) is slidably connected in the sliding groove (11). A fixing member (13) is fixedly connected to one side of the two fixing blocks (12) that are far apart from each other. The position of the fixing member (13) is always fixed.

2. The intelligent sorting equipment for waste concrete blocks according to claim 1, characterized in that: The two connecting plates (202) are fixed to the two racks (5) respectively, and the connecting plates (202) are parallel to the racks (5).

3. The intelligent sorting equipment for waste concrete blocks according to claim 1, characterized in that: The paving plate (4) is parallel to the inclined surface of the upper part of the conveyor belt (101), and the paving plate (4) is located at the inclined surface at the rear end of the conveyor belt (101).

4. The intelligent sorting equipment for waste concrete blocks according to claim 1, characterized in that: The conveying mechanism (1) also includes two rotating wheels (102) and several support frames. The two rotating wheels (102) are connected by a conveyor belt (101). The rotating wheels (102) are driven to rotate by an external motor. Both ends of the two rotating wheels (102) are rotatably connected to the support frames that are close to them.

5. The intelligent sorting equipment for waste concrete blocks according to claim 1, characterized in that: The visual recognition module (9) includes a color industrial camera, a near-infrared spectral camera and an LED lighting source. The color industrial camera and the near-infrared spectral camera are both installed below the crossbeam of the gantry (8). The LED lighting source consists of two sets of symmetrically arranged high color rendering index light sources. The color industrial camera and the near-infrared spectral camera are triggered to capture images simultaneously, respectively collecting the appearance characteristics of the material and the spectral characteristics of the material.

6. The intelligent sorting equipment for waste concrete blocks according to claim 1, characterized in that: The spring (14) is fixed between the bottom of the fixing member (13) and the connecting plate (202), and the extension and retraction direction of the spring (14) is consistent with the movement direction of the connecting plate (202).

7. The intelligent sorting equipment for waste concrete blocks according to claim 1, characterized in that: A gap is left between the two spreading plates (4).

8. A sorting method for a smart sorting device for waste concrete blocks as described in any one of claims 1-7, comprising the following steps: S1: The construction waste mixture initially crushed to 50-300mm is fed into the conveyor belt (101) of the conveying mechanism (1), the conveying mechanism (1) is started, and the conveyor belt (101) is driven to run at a speed of 0.5-1.2m / s by the rotating wheel (102); S2: During the operation of the conveyor belt (101), the convex roller (2) supports the conveyor belt (101) in a raised shape under the cooperation of the bracket and the spring (14). When the material moves to the area of ​​the slab (4), the slab (4) arranged in the shape of the figure eight flattens the material under the torque of the coil spring (7). At the same time, the convex roller (2) moves relative to the fixed block (12) through the slide groove (11) according to the thickness of the material, driving the rack (5) to drive the gear (6) to rotate, dynamically adjusting the initial torque of the coil spring (7) to ensure that the material forms a single layer of material flow with a thickness of 50-100mm. S3: When the material passes under the gantry (8), the color industrial camera and near-infrared spectral camera of the vision recognition module (9) are triggered to shoot simultaneously, and the LED lighting source provides uniform illumination, respectively collecting the material color, texture, shape and material spectral characteristics; S4: The visual recognition module (9) transmits the collected image data to the industrial-grade GPU computing unit. The industrial-grade GPU computing unit has a built-in deep learning model based on the improved YOLOv7 architecture. Through the deep learning model based on the improved YOLOv7 architecture, multimodal feature fusion and intelligent recognition are performed to output the material category. S5: The central control system receives the identification results, combines the encoder data of the conveyor belt (101) to predict the time when the material arrives at the working area of ​​the industrial robot (10), generates sorting instructions according to the impurity category, and controls the industrial robot (10) to switch grippers or adjust gripping parameters. S6: When the impurity reaches the predetermined working position, the industrial robot (10) performs a grasping action according to the planned motion trajectory, removes the impurity from the conveyor belt (101) and throws it into the corresponding classified impurity collection bin. If the single grasping fails, a second grasping is triggered. S7: Target concrete blocks that are not identified as impurities are continuously transported to the end by the conveyor belt (101), fall into the target concrete collection bin, and enter the subsequent crushing or processing process. S8: During system operation, the system receives material sampling results through the online learning module, incrementally trains and updates the parameters of the deep learning model, and at the same time, the central control system monitors the operating status of each module in real time to ensure stable sorting efficiency and purity.

Images