Electric meter module automatic assembling method and system based on 3D visual recognition

Abstract

This invention relates to an automated assembly method and system for electricity meter modules based on 3D vision recognition, comprising the following steps: S1: A turnover box containing electricity meters is transported to the production line entrance via an AGV, resulting in a queue of material boxes with identification labels; S2: A depalletizing device breaks down the stacked material boxes into single layers, transports them to a vision grasping station, and corrects the posture of the boxes using a positioning mechanism; S3: Point cloud collection and analysis of the electricity meters within the material boxes is performed using a 3D vision device; S4: Path planning and grasping actions are executed, and a mapping relationship is established between the electricity meter entity and the tooling position for each individual electricity meter positioned on the tooling; S5: The individual electricity meters positioned on the tooling are assembled using an electricity meter module assembly device; S6: Labels are printed in real time according to order information, and automatically affixed by a labeling mechanism to obtain electricity meters containing identification information. This invention effectively improves the production efficiency and stability of automated assembly of electricity meter modules.

Description

Technical Field

[0001] This invention relates to the field of intelligent assembly, and in particular to an automated assembly method and system for electricity meter modules based on 3D vision recognition. Background Technology

[0002] With the rapid development of the State Grid Corporation of China's secondary metering asset warehouse, the power marketing management system has been able to quickly and efficiently manage the electricity meters and other assets of power companies through information technology and networking. However, under the current management model, the secondary metering asset warehouse adopts a separate shipping method for electricity meters and communication modules, which brings significant problems to practical applications. Especially in large-scale meter installation projects such as those in newly built residential areas, material requisition personnel need to manually bind the electricity meters and communication modules. This traditional operation method greatly increases the workload and intensity of the operators, and at the same time, due to the uneven professional skills of the operators, it also significantly increases the error rate of the binding work.

[0003] Traditional manual binding methods have several technical bottlenecks: First, manually identifying and matching electricity meters and communication modules takes a lot of time, with a single binding operation potentially taking several minutes, resulting in extremely low efficiency for large-scale operations; second, manual operation is prone to problems such as identification errors, incorrect label placement, and data entry errors, leading to difficulties in subsequent traceability; third, manual screw removal and module insertion require high skill levels, making it difficult to guarantee consistent quality; and finally, traditional methods cannot achieve real-time data uploading and status monitoring, lacking a complete quality traceability system. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to provide a solution that effectively improves the automated assembly efficiency of electricity meter modules and significantly reduces assembly defect rates and rework rates.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An automated assembly method for electricity meter modules based on 3D vision recognition includes the following steps: S1: The turnover box containing the electricity meter is sent to the production line entrance by AGV. The roller conveyor system, together with the sensor, detects the arrival of the material and reads the turnover box code through the code reader. The turnover box ID, order, and electricity meter batch are initially bound to form a material digital mapping relationship and obtain a material box queue with identity. S2: Based on the material box queue with identification, the depalletizing equipment breaks down the stacked material boxes into single layers, transports them to the vision grasping station, and corrects the posture of the boxes through the positioning mechanism to obtain the meter boxes with standardized posture. S3: Based on 3D vision equipment, point cloud collection and analysis are performed on the meters in the material box to identify the three-dimensional position and posture information of each meter. At the same time, abnormal status is detected. Combined with the barcode reader data, the box and internal materials are reconfirmed to obtain the position and posture data of the meters and the grasping priority strategy. S4: The mechanical gripper performs path planning and gripping actions based on the position data of the meter and the gripping priority strategy. After gripping, it places the meter into the designated fixture of the circular tooling line and establishes a mapping relationship between the meter entity and the tooling position for the single meter located on the tooling. S5: Based on the meter module assembly equipment, the individual meters that have been positioned on the tooling are assembled to obtain the meter with the complete structural assembly. S6: Print labels in real time according to order information, and automatically apply them through a labeling mechanism to obtain a meter containing the labeling information.

[0006] In this embodiment, based on the identified material box queue, the depalletizing equipment breaks down the stacked material boxes into single layers and transports them to the vision grasping station. The positioning mechanism then corrects the posture of the boxes to obtain standardized meter boxes. Specifically: Based on the identified material box queue formed in stage S1, the depalletizing equipment begins automated processing of the stacked meter boxes. Following the FIFO principle, the stacked boxes are sequentially transported to the depalletizing equipment station. Single-layer meter boxes are transported to the 3D vision grasping station via a roller conveyor system. Sensors, including position detection sensors and speed monitoring sensors, monitor the transport process in real time to ensure the boxes run stably along a predetermined trajectory. Once the meter box reaches the designated position, the positioning mechanism is activated for posture correction. After posture correction, the meter box enters the waiting state at the vision grasping station, transmitting the box's identification information, position coordinates, and posture parameter data to the 3D vision system.

[0007] In this embodiment, the positioning mechanism employs a combination of mechanical guidance, multi-point constraint, and servo drive, as detailed below. The system includes a bottom reference positioning platform, adjustable guide baffles on both sides, front and rear positioning pin mechanisms, and a top clamping device. After the meter box enters the positioning station, it is initially centered through a V-shaped guide groove. Subsequently, the guide baffles on both sides tighten synchronously under the drive of a servo motor, so that the box is centered in the X direction. The front and rear positioning pins are inserted into preset reference holes or edges along the Y direction to achieve Y-direction limiting. Finally, the top clamping mechanism applies constraint force in the Z direction to complete three-dimensional full constraint positioning. In terms of attitude detection and error modeling, the attitude deviation of the meter box is obtained through tilt sensors, including the tilt angles around the X and Y axes, denoted as θx and θy respectively, and the in-plane offsets Δx and Δy. When the box is tilted, its height error is expressed as: ; Among them, L x ,L y The characteristic length of the housing in the corresponding direction is defined. A threshold is set to determine if adjustment is needed. When an offset is detected, the servo system drives the guide mechanism to perform compensatory movement, and its target position is represented as: ; Where, x target ,y target x represents the target location coordinates; measured ,y measured To measure the position coordinates; At the same time, the box is restored to a horizontal state by compensating for the Δz error by finely adjusting the height of the clamping mechanism; In the servo control strategy, a combination of closed-loop position control and PID regulation is adopted. Let the target position be r(t), the actual position be y(t), and the error be e(t) = r(t) - y(t). Then the servo control output is: ; Among them, K p ,K i ,K d These are the proportional, integral, and differential coefficients, respectively; a velocity feedforward term is introduced: ; in, This is the differential term of the error; For reference input differential terms; After positioning is completed, the final posture is confirmed to meet the conditions by contact sensors or visual re-inspection. If it does not meet the standards, a secondary correction or abnormal rejection process is triggered.

[0008] In this embodiment, point cloud data of the meters inside the material box is collected and analyzed using a 3D vision device to identify the three-dimensional position and orientation information of each meter. Simultaneously, abnormal states are detected. Combined with barcode reader data, the box and its internal materials are reconfirmed to obtain the meter's pose data and a prioritization strategy for meter acquisition, as detailed below: After the posture standardization is completed, the meter box is transported to the 3D vision recognition station. The meter inside the box is scanned by the 3D vision equipment to obtain high-density point cloud data. The vision system preprocesses the collected raw point cloud in a unified coordinate system, including noise reduction, filtering and coordinate correction. Based on the point cloud reconstruction algorithm, a three-dimensional spatial model of the meter is generated. Based on the 3D spatial model of the electricity meter, feature matching and deep learning algorithms are used to segment and identify the meter. The position and posture information of each meter are extracted one by one, and the completeness and rationality of the identification results are verified. Abnormal states are detected, including meters that are overturned, overlapping or obstructed, missing parts, or whose posture is ungraspable. For abnormal targets, they are automatically marked and their grasping priority is reduced. Combined with the turnover box ID and batch information obtained by the preceding barcode reader, the consistency of the box and the internal meters is verified, realizing secondary confirmation at the material level and ensuring that the data and the physical objects correspond completely. After the identification and verification are completed, the position and posture information of all valid meters is output in a structured manner, and a grasping priority strategy is generated based on the grasping feasibility analysis.

[0009] In this embodiment, feature matching and deep learning algorithms are used to segment and identify the electricity meters, and the location and attitude information of each meter are extracted one by one, as follows: Based on the input point cloud P={p i Based on PointNet++, the set of meter points is obtained through classification: ; Where f(·) is the classification function; p i Let i be the coordinate vector of the i-th point; Then, Euclidean clustering was used to divide the point cloud into multiple independent targets: ; Obtain the point cloud cluster C for each candidate meter k Template matching is introduced for identification, and target confirmation is achieved through feature descriptor matching. In terms of pose estimation, for each meter point cloud cluster C k Perform 6D pose calculation and calculate the centroid as the position coordinate: ; Then, based on principal component analysis, the principal orientation of the point cloud is calculated, and the covariance matrix is ​​constructed: ; Perform eigenvalue decomposition on Σ to obtain the eigenvector v. 1, v2 and v3 are used as the principal axes of the object to construct a rotation matrix R, which is then converted into Euler angles (R2). x , R y , R z ): Let these be the coordinates of the centroid of the point cloud; ; Where α, β, and γ correspond to the rotation angles around the X, Y, and Z axes, respectively, the final pose is obtained: ; Where t=(x,y,z) T .

[0010] In this embodiment, the completeness and reasonableness of the recognition results are checked, and abnormal states are detected, including cases where the meter is overturned, overlapping or obstructing, missing parts, or the posture cannot be grasped, as detailed below: Let the ideal upward normal of the meter be n0 = (0,0,1), and the actual normal be n. Then the flip angle is: θ = arccos(n·n0) When θ>θ th It was determined to be overturned at that time; For two point cloud clusters C i C j Define overlap rate: ; When O ij >O th When it is determined to be occlusion or overlap; Let the number of point clouds for the standard meter be N0, and the current number of points be N. k : ; When the integrity ratio η < the threshold, it is judged as missing part or severe occlusion; Define the scoring function: ; Among them, S stable For attitude stability; S visible S represents the proportion of visible surface area. reach For robot reachability; finally, sort by S to generate grasping priority; w1, w2, and w3 are weight coefficients; The above model outputs the (X, Y, Z, R) values ​​for each meter. x ,R y ,R z The pose data and its status labels are used to form a structured capture sequence.

[0011] In this embodiment, based on the input point cloud P={p i Based on PointNet++, the classification is performed to obtain the meter point set, as follows: The original point cloud acquired by 3D vision is denoted as P={p i |p i =(x i ,y i ,z i ,r i ,g i ,b i )}; Among them, (r i ,g i b) represents color information r i The value for the red channel; g i For green channel values; b i The value for the blue channel; x i ,y i ,z i Let i be the three-dimensional spatial coordinates of the i-th point; Denoising and downsampling are performed to obtain a uniform point set P′. Then, N center points are selected by sampling the farthest points from the point cloud. ; in, The set of selected center points; Let point p be the distance from the center point c. j L2 distance; With each center c k Using the center of the sphere as an example, perform a neighborhood search based on radius r: ; in, Center point c k The set of neighborhood points; Furthermore, the coordinates of neighboring points are relativized to enhance translation invariance. ; in, (x) are relative coordinates; j ,y j ,z j (x) represents the absolute coordinates of a neighboring point; k ,y k ,z k () represents the absolute coordinates of the center point; PointNet++ employs a hierarchical structure of sampling, grouping, and local feature learning for each neighborhood. Features are extracted by sharing an MLP and aggregated using a symmetric function: ; Among them, h j For each point feature vector, the local feature f of each center point is obtained. k ; Finally, the high-level features are interpolated back to the original point through feature propagation: ; The classification head is implemented through a fully connected layer: =Softmax(Wf i +b) Where W is the weight and b is the bias.

[0012] In this embodiment, the mechanical gripper performs path planning and gripping actions based on the meter's pose data and gripping priority strategy. After gripping, it places the meter into a designated fixture on the circular tooling line and establishes a mapping relationship between the meter entity and the tooling position. The single meter is positioned on the tooling, as detailed below: After acquiring the meter pose data (X, Y, Z, Rx, Ry, Rz) and grasping priority strategy output in stage S3, path planning is performed based on the robot's kinematic model to establish the optimal motion trajectory from the current pose to the target pose. The end effector pose is represented by a homogeneous transformation matrix: ; Where R is the attitude rotation matrix and t is the position vector; the path planning uses an improved RRT algorithm to generate a continuous trajectory that satisfies obstacle avoidance constraints, and combines it with a velocity planning function to achieve smooth motion: q(t) = q0 + (q f -q0)·s(t); Where s(t) is the time scale function; After the trajectory is issued, the mechanical gripper moves to the target meter above the pre-grabbing posture according to the planned path, and makes secondary fine adjustments through force feedback; The grippers automatically adjust the gripping width and gripping force based on the meter's geometry, with a gripping force F. grip : ; Where m is the mass of the meter, and μ is the coefficient of friction; After successful grasping, the robot transports the meter to the designated fixture position on the circular tooling line according to a predetermined path and performs a precise placement action. The tooling fixture adopts a standard positioning structure to ensure the consistency of the meter's posture after placement, based on the transformation relationship between the robot's end effector pose and the tooling coordinate system: T fixture =T robot ·T calibration , where T fixture T is the transformation matrix of the tooling fixture coordinate system. robot Let T be the coordinate transformation matrix of the robot's end effector. calibration To calibrate the transformation matrix and achieve precise positioning of the meter on the tooling; Establish a one-to-one mapping relationship between the electricity meter entity ID and the tooling location ID, and write this binding information into the MES or control system database to achieve unique identification and traceability management of a single electricity meter in the production line.

[0013] The automated assembly system for electricity meter modules based on 3D vision recognition includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the automated assembly method for electricity meter modules based on 3D vision recognition as described above.

[0014] A computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the method steps described above.

[0015] The present invention has the following beneficial effects: 1. This invention achieves fully automated operation from material input to assembly completion. Seamless connection between each process is achieved through cycle time matching and data linkage. By introducing path planning and grasping strategies, the robot can adaptively adjust its actions according to the real-time environment, improving equipment utilization and production line flexibility, adapting to the mixed production needs of multiple batches and models of electricity meters, thereby improving overall production efficiency and stability. 2. This invention uses 3D visual point cloud acquisition and analysis technology to accurately acquire the three-dimensional pose information of the meter, and combines it with an anomaly detection mechanism to identify problems such as occlusion, misplacement or missing parts. Compared with traditional 2D vision, it has stronger spatial understanding ability, and with the optimal grasping strategy and precise path planning, it effectively improves the grasping success rate and positioning accuracy. 3. This invention ensures the consistency of assembly of each structural component by using standardized tooling positioning and automated assembly equipment, reduces assembly deviations, and realizes closed-loop quality control from "identification-grabbing-assembly", which greatly improves product consistency and pass rate. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: refer to Figure 1 In this embodiment, an automated assembly method for an electricity meter module based on 3D vision recognition is provided, including the following steps: S1: The turnover box containing the electricity meter is sent to the production line entrance by AGV. The roller conveyor system, together with the sensor, detects the arrival of the material and reads the turnover box code through the code reader. The turnover box ID, order, and electricity meter batch are initially bound to form a material digital mapping relationship and obtain a material box queue with identity. S2: Based on the material box queue with identification, the depalletizing equipment breaks down the stacked material boxes into single layers and transports them to the vision grasping station. The positioning mechanism then corrects the posture of the boxes to obtain standardized meter boxes. S3: Based on 3D vision equipment, point cloud collection and analysis are performed on the meters in the material box to identify the three-dimensional position and posture information of each meter. At the same time, abnormal status is detected. Combined with the barcode reader data, the box and internal materials are reconfirmed to obtain the position and posture data of the meters and the grasping priority strategy. S4: The mechanical gripper performs path planning and gripping actions based on the position data of the meter and the gripping priority strategy. After gripping, it places the meter into the designated fixture of the circular tooling line and establishes a mapping relationship between the meter entity and the tooling position for the single meter located on the tooling. S5: Based on the meter module assembly equipment, the individual meters that have been positioned on the tooling are assembled to obtain the meter with the complete structural assembly. S6: Print labels in real time according to order information, and automatically apply them through a labeling mechanism to obtain a meter containing the labeling information.

[0018] In this embodiment, during the startup phase of the automated assembly line, the secondary warehouse system, based on order information from the Marketing 2.0 system, uses AGVs (Automated Guided Vehicles) to precisely transport the turnover boxes loaded with electricity meters from the warehouse to the production line entrance. The AGV system transports the stacked electricity meter turnover boxes to designated locations according to a preset path, ensuring the continuity and accuracy of material supply. The roller conveyor system, acting as a receiving device, works with sensors to detect material arrival, including photoelectric sensors to detect the presence of the turnover boxes and position sensors to confirm the accuracy of their placement. Once the sensors confirm that the turnover box is accurately in place, the system automatically activates a barcode reader to scan and identify the barcode on the surface of the turnover box, obtaining its unique identification code (ID).

[0019] After obtaining the turnover box code, the system immediately retrieves order information from the database, including key parameters such as order number, meter specifications, production quantity, and barcode rules. A triple data binding is established between "turnover box ID—order information—meter batch," creating a complete digital mapping relationship for materials. This process not only ensures the traceability of the meters within each turnover box but also provides a data foundation for subsequent processes such as 3D vision recognition, robot grasping, and quality inspection. The system also verifies the consistency of the number of meters in the turnover box with the order requirements; if an anomaly is detected, an alarm mechanism is automatically triggered. Furthermore, the system uploads the binding results to the MES system in real time, achieving full digital control of the production process.

[0020] After data binding is completed, each tote box is converted into an identified material box queue, managed according to the First-In-First-Out (FIFO) principle. The queue system records not only basic information for each material box but also detailed data such as meter specifications, assembly requirements, and quality standards. These digitized material box queues provide standardized input for subsequent depalletizing and palletizing equipment, ensuring that each layer of meter boxes has complete identification information and process parameters when entering the visual recognition station. The system also establishes a caching mechanism; when subsequent processes experience temporary shutdowns or anomalies, multiple tote boxes can be temporarily stored on the conveyor belt to ensure production continuity.

[0021] In this embodiment, based on the identified material box queue, the depalletizing equipment breaks down the stacked material boxes into single layers and transports them to the vision grasping station. A positioning mechanism then corrects the posture of the boxes to obtain standardized meter boxes. Specifically: Based on the identified material box queue formed in stage S1, the depalletizing equipment begins automated processing of the stacked meter boxes. Following the FIFO principle, the stacked boxes are sequentially transported to the depalletizing equipment station. The depalletizing machine uses a servo-driven lifting mechanism, combined with a mechanical separation device, to precisely decompose the stacked meter boxes into individual layers, with each layer separation cycle controlled within 5 seconds. During the separation process, the equipment monitors the separation status through multiple sensors to ensure that each layer of boxes is completely separated without damage. After separation, the single-layer meter boxes are automatically transported to the next workstation via a roller conveyor system. The entire separation process requires no manual intervention, ensuring the continuity and stability of material flow. The single-layer meter boxes are then transported to the 3D vision grasping station via the roller conveyor system. Sensors, including position detection sensors and speed monitoring sensors, monitor the transport process in real time, ensuring the boxes run stably along a predetermined trajectory. Once the meter box reaches the designated position, a positioning mechanism is activated for attitude correction. After attitude correction, the meter box enters the waiting state at the vision grasping station, where its identity information, position coordinates, and attitude parameters are transmitted to the 3D vision system. At this point, each meter box possesses a standardized physical state and a complete digital identity, including key data such as the turnover box ID, the number of meters contained, batch information, and order association. A workstation caching mechanism is also established, temporarily storing 2-3 standardized meter boxes at the vision grasping station to ensure production continuity. When subsequent processes require temporary shutdown, the depalletizing equipment can be paused, or excess meter boxes can be temporarily stored on the roller conveyor line.

[0022] In this embodiment, the positioning mechanism employs a combination of mechanical guidance, multi-point constraint, and servo drive, as detailed below. The system includes a bottom reference positioning platform, adjustable guide baffles on both sides, front and rear positioning pin mechanisms, and a top pneumatic / electric clamping device. After the meter box enters the positioning station, it is initially centered through a V-shaped guide groove. Subsequently, the guide baffles on both sides tighten synchronously under the drive of a servo motor, so that the box is centered in the X direction. The front and rear positioning pins are inserted into preset reference holes or edges along the Y direction to achieve Y-direction limiting. Finally, the top clamping mechanism applies constraint force in the Z direction to complete three-dimensional full constraint positioning. In terms of attitude detection and error modeling, the attitude deviation of the meter box is obtained through tilt sensors, including the tilt angles around the X and Y axes, denoted as θx and θy respectively, and the in-plane offsets Δx and Δy. When the box is tilted, its height error is expressed as: ; Among them, L x ,L y The characteristic length of the box in the corresponding direction; by setting a threshold (e.g.) (Or |Δx|>1mm) to determine whether adjustment is needed. When an offset is detected, the servo system drives the guide mechanism to perform compensatory movement, and its target position is represented as: ; Where, x target ,y target x represents the target location coordinates; measured ,y measured To measure the position coordinates; At the same time, the box is restored to a horizontal state by compensating for the Δz error by finely adjusting the height of the clamping mechanism; In the servo control strategy, a combination of closed-loop position control and PID regulation is adopted. Let the target position be r(t), the actual position be y(t), and the error be e(t) = r(t) - y(t). Then the servo control output is: ; Among them, K p ,K i ,K d These are the proportional, integral, and derivative coefficients, respectively; dynamic adjustment is achieved by providing real-time feedback on the positions of the guide baffle and positioning pins through the encoder; to avoid overshoot and oscillation, a speed feedforward term is introduced. ; in, This is the differential term of the error; For reference input differential terms; After positioning is completed, the final posture is confirmed to meet the conditions (e.g., flatness error ≤ 1 mm) through contact sensors or visual re-inspection. If the condition is not met, a secondary correction or anomaly rejection process is triggered. Mechanical limiters ensure rigidity, sensor detection provides error quantification, and servo closed loop enables dynamic correction, forming a stable, calculable, and controllable high-precision positioning system, providing consistent input conditions for 3D visual recognition. In this embodiment, point cloud data of the meters inside the material box is collected and analyzed using a 3D vision device to identify the three-dimensional position and orientation information of each meter. Simultaneously, abnormal states are detected. Combined with barcode reader data, the box and its internal materials are reconfirmed to obtain the meter's pose data and a prioritization strategy for meter acquisition, as detailed below: After the posture standardization is completed, the meter box is transported to the 3D vision recognition station. The 3D vision equipment scans the meters in the box to obtain high-density point cloud data. The vision system preprocesses the collected raw point cloud in a unified coordinate system, including noise reduction, filtering and coordinate correction. Based on the point cloud reconstruction algorithm, a three-dimensional spatial model of the meter is generated to realize the overall spatial distribution model of all meters in the box. Based on the 3D spatial model of the electricity meter, feature matching and deep learning algorithms are used to segment and identify the meter. The position and posture information of each meter are extracted one by one, and the completeness and rationality of the identification results are verified. Abnormal states are detected, including meters that are overturned, overlapping or obstructed, missing parts, or whose posture is ungraspable. For abnormal targets, they are automatically marked and their grasping priority is reduced. Combined with the turnover box ID and batch information obtained by the preceding barcode reader, the consistency of the box and the internal meters is verified, realizing secondary confirmation at the material level and ensuring that the data and the physical objects correspond completely. After the identification and verification are completed, the position and posture information of all valid meters is output in a structured manner, and a grasping priority strategy is generated based on the grasping feasibility analysis.

[0023] In this embodiment, feature matching and deep learning algorithms are used to segment and identify the electricity meters, and the location and attitude information of each meter are extracted one by one, as follows: Based on the input point cloud P={p i Based on PointNet++, the set of meter points is obtained through classification: ; Where f(·) is the classification function; p i Let i be the coordinate vector of the i-th point; Then, Euclidean clustering was used to divide the point cloud into multiple independent targets: ; Obtain the point cloud cluster C for each candidate meter kTemplate matching (such as FPFH features or SHOT features) is introduced for identification, and target confirmation is achieved through feature descriptor matching. In terms of pose estimation, for each meter point cloud cluster C k Perform 6D pose calculation and calculate the centroid as the position coordinate: ; Then, based on principal component analysis, the principal orientation of the point cloud is calculated, and the covariance matrix is ​​constructed: ; Perform eigenvalue decomposition on Σ to obtain the eigenvector v. 1, v2 and v3 are used as the principal axes of the object to construct a rotation matrix R, which is then converted into Euler angles (R2). x , R y , R z ): Let these be the coordinates of the centroid of the point cloud; ; Where α, β, and γ correspond to the rotation angles around the X, Y, and Z axes, respectively, the final pose is obtained: ; Where t=(x,y,z) T .

[0024] In this embodiment, the completeness and reasonableness of the recognition results are checked, and abnormal states are detected, including cases where the meter is overturned, overlapping or obstructing, missing parts, or the posture cannot be grasped, as detailed below: Let the ideal upward normal of the meter be n0 = (0,0,1), and the actual normal be n. Then the flip angle is: θ = arccos(n·n0) When θ>θ th It was determined to be overturned at that time; For two point cloud clusters C i C j Define overlap rate: ; When O ij >O th When it is determined to be occlusion or overlap; Let the number of point clouds for the standard meter be N0, and the current number of points be N. k : ; When the integrity ratio η < the threshold, it is judged as missing part or severe occlusion; Define the scoring function: ; Among them, S stableFor attitude stability; S visible S represents the proportion of visible surface area. reach For robot reachability; finally, sort by S to generate grasping priority; w1, w2, and w3 are weight coefficients; The above model outputs the (X, Y, Z, R) values ​​for each meter. x ,R y ,R z The pose data and its status labels (normal / flipped / occluded / ungrabable) are used to form a structured grabbing sequence.

[0025] In this embodiment, based on the input point cloud P={p i Based on PointNet++, the classification is performed to obtain the meter point set, as follows: The original point cloud acquired by 3D vision is denoted as P={p i |p i =(x i ,y i ,z i ,r i ,g i ,b i )}; Among them, (r i ,g i b) represents color information r i The value for the red channel; g i For green channel values; b i The value for the blue channel; x i ,y i ,z i Let i be the three-dimensional spatial coordinates of the i-th point; Denoising and downsampling are performed to obtain a uniform point set P′. Then, N center points are selected by sampling the farthest points from the point cloud. ; in, The set of selected center points; Let point p be the distance from the center point c. j L2 distance; With each center c k Using the center of the sphere as an example, perform a neighborhood search based on radius r: ; in, Center point c k The set of neighborhood points; Furthermore, the coordinates of neighboring points are relativized to enhance translation invariance. ; in, (x) are relative coordinates; j ,y j ,z j (x) represents the absolute coordinates of a neighboring point; k ,y k ,z k () represents the absolute coordinates of the center point; PointNet++ employs a hierarchical structure of sampling, grouping, and local feature learning for each neighborhood. Features are extracted by sharing an MLP and aggregated using a symmetric function: ; Among them, h j For each point feature vector, the local feature f of each center point is obtained. k Multiple layers are stacked to form feature representations at different scales (MSG or SSG structures): ; Finally, the high-level features are interpolated back to the original point through feature propagation: ; The classification head is implemented through a fully connected layer: =Softmax(Wf i +b) Where W is the weight and b is the bias.

[0026] In this embodiment, the mechanical gripper performs path planning and gripping actions based on the meter's pose data and gripping priority strategy. After gripping, it places the meter into a designated fixture on the circular tooling line and establishes a mapping relationship between the meter entity and the tooling position. The single meter is positioned on the tooling, as detailed below: After acquiring the meter pose data (X, Y, Z, Rx, Ry, Rz) and grasping priority strategy output in stage S3, path planning is performed based on the robot's kinematic model to establish the optimal motion trajectory from the current pose to the target pose. The end effector pose is represented by a homogeneous transformation matrix: ; Where R is the attitude rotation matrix and t is the position vector; the path planning uses an improved RRT algorithm to generate a continuous trajectory that satisfies obstacle avoidance constraints, and combines it with a velocity planning function to achieve smooth motion: q(t) = q0 + (q f -q0)·s(t); Where s(t) is the time scale function; After the trajectory is sent, the mechanical gripper (such as a pneumatic gripper or an adaptive electric gripper) moves to the target meter above the pre-grabbing posture according to the planned path, and performs secondary fine-tuning through force feedback; The grippers automatically adjust the gripping width and gripping force based on the meter's geometry, with a gripping force F. grip : ; Where m is the mass of the meter and μ is the coefficient of friction; force control or impedance control strategies are combined during the grasping process: To reduce contact impact and improve gripping stability; After successful grasping, the robot transports the meter to the designated fixture position on the circular tooling line according to a predetermined path and performs a precise placement action. The tooling fixture adopts a standard positioning structure (such as positioning pins + limiting surfaces) to ensure the consistency of the meter's posture after placement, based on the transformation relationship between the robot's end effector pose and the tooling coordinate system: T fixture =T robot ·T calibration , where T fixture T is the transformation matrix of the tooling fixture coordinate system. robot Let T be the coordinate transformation matrix of the robot's end effector. calibration To calibrate the transformation matrix and achieve precise positioning of the meter on the tooling; Establish a one-to-one mapping relationship between the electricity meter entity ID and the tooling location ID, and write this binding information into the MES or control system database to achieve unique identification and traceability management of a single electricity meter in the production line.

[0027] The automated assembly system for electricity meter modules based on 3D vision recognition includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the automated assembly method for electricity meter modules based on 3D vision recognition as described above.

[0028] A computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the method steps described above.

[0029] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0030] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0031] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0032] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An automated assembly method for electricity meter modules based on 3D vision recognition, characterized in that, Includes the following steps: S1: The turnover box containing the electricity meter is sent to the production line entrance by AGV. The roller conveyor system, together with the sensor, detects the arrival of the material and reads the turnover box code through the code reader. The turnover box ID, order, and electricity meter batch are initially bound to form a material digital mapping relationship and obtain a material box queue with identity. S2: Based on the material box queue with identification, the depalletizing equipment breaks down the stacked material boxes into single layers, transports them to the vision grasping station, and corrects the posture of the boxes through the positioning mechanism to obtain the meter boxes with standardized posture. S3: Based on 3D vision equipment, point cloud collection and analysis are performed on the meters in the material box to identify the three-dimensional position and posture information of each meter. At the same time, abnormal status is detected. Combined with the barcode reader data, the box and internal materials are reconfirmed to obtain the position and posture data of the meters and the grasping priority strategy. S4: The mechanical gripper performs path planning and gripping actions based on the position data of the meter and the gripping priority strategy. After gripping, it places the meter into the designated fixture of the circular tooling line and establishes a mapping relationship between the meter entity and the tooling position for the single meter located on the tooling. S5: Based on the meter module assembly equipment, the individual meters that have been positioned on the tooling are assembled to obtain the meter with the complete structural assembly. S6: Print labels in real time according to order information, and automatically apply them through a labeling mechanism to obtain a meter containing the labeling information.

2. The automated assembly method for an electricity meter module based on 3D vision recognition according to claim 1, characterized in that, According to the material box queue with identification, the depalletizing equipment breaks down the stacked material boxes into single layers and transports them to the vision grasping station. The positioning mechanism corrects the posture of the boxes to obtain standardized meter boxes. Specifically, based on the material box queue with identification formed in stage S1, the depalletizing equipment begins to automatically process the stacked meter boxes. Following the FIFO principle, the stacked boxes are transported to the depalletizing equipment station in sequence. The single-layer meter boxes are transported to the 3D vision grasping station via a roller conveyor system. Sensors are used for real-time monitoring during the transport process, including position detection sensors and speed monitoring sensors, to ensure that the boxes run stably along a predetermined trajectory. When the meter box reaches the designated position, the positioning mechanism is activated to perform posture correction. The meter box that has completed posture correction enters the waiting state of the vision grasping station, and the box's identification information, position coordinates, and posture parameter data are transmitted to the 3D vision system.

3. The automated assembly method for an electricity meter module based on 3D vision recognition according to claim 2, characterized in that, The positioning mechanism employs a combination of mechanical guidance, multi-point constraint, and servo drive, as detailed below. The system includes a bottom reference positioning platform, adjustable guide baffles on both sides, front and rear positioning pin mechanisms, and a top clamping device. After the meter box enters the positioning station, it is initially centered through a V-shaped guide groove. Subsequently, the guide baffles on both sides tighten synchronously under the drive of a servo motor, so that the box is centered in the X direction. The front and rear positioning pins are inserted into preset reference holes or edges along the Y direction to achieve Y-direction limiting. Finally, the top clamping mechanism applies constraint force in the Z direction to complete three-dimensional full constraint positioning. In terms of attitude detection and error modeling, the attitude deviation of the meter box is obtained through tilt sensors, including the tilt angles around the X and Y axes, denoted as θx and θy respectively, and the in-plane offsets Δx and Δy. When the box is tilted, its height error is expressed as: ； Among them, L x ,L y The characteristic length of the housing in the corresponding direction is defined. A threshold is set to determine if adjustment is needed. When an offset is detected, the servo system drives the guide mechanism to perform compensatory movement, and its target position is represented as: ； Where, x target ,y target x represents the target location coordinates; measured ,y measured To measure the position coordinates; At the same time, the box is restored to a horizontal state by compensating for the Δz error by finely adjusting the height of the clamping mechanism; In the servo control strategy, a combination of closed-loop position control and PID regulation is adopted. Let the target position be r(t), the actual position be y(t), and the error be e(t) = r(t) - y(t). Then the servo control output is: ； Among them, K p ,K i ,K d These are the proportional, integral, and differential coefficients, respectively; a velocity feedforward term is introduced: ； in, This is the differential term of the error; For reference input differential terms; After positioning is completed, the final posture is confirmed to meet the conditions by contact sensors or visual re-inspection. If it does not meet the standards, a secondary correction or abnormal rejection process is triggered.

4. The automated assembly method for an electricity meter module based on 3D vision recognition according to claim 1, characterized in that, The 3D vision device is used to collect and analyze point cloud data of the meters inside the material box, identify the three-dimensional position and orientation information of each meter, detect abnormal states, and combine the data from the barcode reader to perform secondary verification of the box and the internal materials, thereby obtaining the position and orientation data of the meters and the grasping priority strategy, as detailed below: After the posture standardization is completed, the meter box is transported to the 3D vision recognition station. The meter inside the box is scanned by the 3D vision equipment to obtain high-density point cloud data. The vision system preprocesses the collected raw point cloud in a unified coordinate system, including noise reduction, filtering and coordinate correction. Based on the point cloud reconstruction algorithm, a three-dimensional spatial model of the meter is generated. Based on the 3D spatial model of the electricity meter, feature matching and deep learning algorithms are used to segment and identify the meter. The position and posture information of each meter are extracted one by one, and the completeness and rationality of the identification results are verified. Abnormal states are detected, including meters that are overturned, overlapping or obstructed, missing parts, or whose posture is ungraspable. For abnormal targets, they are automatically marked and their grasping priority is reduced. Combined with the turnover box ID and batch information obtained by the preceding barcode reader, the consistency of the box and the internal meters is verified, realizing secondary confirmation at the material level and ensuring that the data and the physical objects correspond completely. After the identification and verification are completed, the position and posture information of all valid meters is output in a structured manner, and a grasping priority strategy is generated based on the grasping feasibility analysis.

5. The automated assembly method for an electricity meter module based on 3D vision recognition according to claim 4, characterized in that, The process involves target segmentation and identification of the electricity meters using feature matching and deep learning algorithms, extracting the location and attitude information of each meter individually, as detailed below: Based on the input point cloud P={p i Based on PointNet++, the set of meter points is obtained through classification: ； Where f(·) is the classification function; p i Let i be the coordinate vector of the i-th point; Then, Euclidean clustering was used to divide the point cloud into multiple independent targets: ； Obtain the point cloud cluster C for each candidate meter k Template matching is introduced for identification, and target confirmation is achieved through feature descriptor matching. In terms of pose estimation, for each meter point cloud cluster C k Perform 6D pose calculation and calculate the centroid as the position coordinate: ； Then, based on principal component analysis, the principal orientation of the point cloud is calculated, and the covariance matrix is ​​constructed: ； Perform eigenvalue decomposition on Σ to obtain the eigenvector v. 1, v2 and v3 are used as the principal axes of the object to construct a rotation matrix R, which is then converted into Euler angles (R2). x , R y , R z ): Let these be the coordinates of the centroid of the point cloud; ； Where α, β, and γ correspond to the rotation angles around the X, Y, and Z axes, respectively, the final pose is obtained: ； Where t=(x,y,z) T .

6. The automated assembly method for an electricity meter module based on 3D vision recognition according to claim 4, characterized in that, The process involves verifying the completeness and reasonableness of the identification results and detecting abnormal states, including situations such as the meter being overturned, overlapping or obstructing, missing parts, or being unable to be grasped in a specific posture. Let the ideal upward normal of the meter be n0 = (0,0,1), and the actual normal be n. Then the flip angle is: θ = arccos(n·n0) When θ>θ th It was determined to be overturned at that time; For two point cloud clusters C i C j Define overlap rate: ； When O ij >O th It is determined to be occlusion or overlap at that time; Let the number of point clouds for the standard meter be N0, and the current number of points be N. k : ； When the integrity ratio η < the threshold, it is judged as missing part or severe occlusion; Define the scoring function: ； Among them, S stable For attitude stability; S visible S represents the proportion of visible surface area. reach For robot reachability; finally, sort by S to generate grasping priority; w1, w2, and w3 are weight coefficients; The above model outputs the (X, Y, Z, R) values ​​for each meter. x ,R y ,R z The pose data and its status labels are used to form a structured capture sequence.

7. The automated assembly method for an electricity meter module based on 3D vision recognition according to claim 5, characterized in that, The input point cloud P={p i Based on PointNet++, the classification is performed to obtain the meter point set, as follows: The original point cloud acquired by 3D vision is denoted as P={p i ∣p i =(x i ,y i ,z i ,r i ,g i ,b i )}; Among them, (r i ,g i b) represents color information r i The value for the red channel; g i For green channel values; b i The value for the blue channel; x i ,y i ,z i Let i be the three-dimensional spatial coordinates of the i-th point; Denoising and downsampling are performed to obtain a uniform point set P′. Then, N center points are selected by sampling the farthest points from the point cloud. ； in, The set of selected center points; Let point p be the distance from the center point c. j L2 distance; With each center c k Using the center of the sphere as an example, perform a neighborhood search based on radius r: ； in, Center point c k The neighborhood point set; Furthermore, the coordinates of neighboring points are relativized to enhance translation invariance. ； in, (x) are relative coordinates; j ,y j ,z j (x) represents the absolute coordinates of a neighboring point; k ,y k ,z k () represents the absolute coordinates of the center point; PointNet++ employs a hierarchical structure of sampling, grouping, and local feature learning for each neighborhood. Features are extracted by sharing an MLP and aggregated using a symmetric function: ； Among them, h j For each point feature vector, the local feature f of each center point is obtained. k ; Finally, the high-level features are interpolated back to the original point through feature propagation: ； The classification head is implemented through a fully connected layer: =Softmax(Wf i +b) Where W is the weight and b is the bias.

8. The automated assembly method for an electricity meter module based on 3D vision recognition according to claim 1, characterized in that, The mechanical gripper performs path planning and gripping actions based on the meter's pose data and gripping priority strategy. After gripping, it places the meter into a designated fixture on the circular tooling line and establishes a mapping relationship between the meter entity and the tooling position. The individual meter is positioned on the tooling, as detailed below: After acquiring the meter pose data (X, Y, Z, Rx, Ry, Rz) and grasping priority strategy output in stage S3, path planning is performed based on the robot's kinematic model to establish the optimal motion trajectory from the current pose to the target pose. The end effector pose is represented by a homogeneous transformation matrix: ； Where R is the attitude rotation matrix and t is the position vector; the path planning uses an improved RRT algorithm to generate a continuous trajectory that satisfies obstacle avoidance constraints, and combines it with a velocity planning function to achieve smooth motion: q(t)=q0+(q f -q0)·s(t); Where s(t) is the time scale function; After the trajectory is issued, the mechanical gripper moves to the target meter above the pre-grabbing posture according to the planned path, and makes secondary fine adjustments through force feedback; The grippers automatically adjust the gripping width and gripping force based on the meter's geometry, with a gripping force F. grip : ； Where m is the mass of the meter, and μ is the coefficient of friction; After successful grasping, the robot transports the meter to the designated fixture position on the circular tooling line according to a predetermined path and performs a precise placement action. The tooling fixture adopts a standard positioning structure to ensure the consistency of the meter's posture after placement, based on the transformation relationship between the robot's end effector pose and the tooling coordinate system: T fixture =T robot ·T calibration , where T fixture T is the transformation matrix of the tooling fixture coordinate system. robot Let T be the coordinate transformation matrix of the robot's end effector. calibration To calibrate the transformation matrix and achieve precise positioning of the meter on the tooling; Establish a one-to-one mapping relationship between the electricity meter entity ID and the tooling location ID, and write this binding information into the MES or control system database to achieve unique identification and traceability management of a single electricity meter in the production line.

9. An automated assembly system for electricity meter modules based on 3D vision recognition, characterized in that, It includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the automated assembly method for an electricity meter module based on 3D vision recognition as described in any one of claims 1-8.

10. A computer storage medium, characterized in that, The computer storage medium stores a plurality of instructions adapted for loading by a processor and executing the method steps as claimed in any one of claims 1-8.

Images