Board passing processing system and method of circuit board assembly

Abstract

The invention relates to the technical field of PCBs, and discloses a board passing processing system and method of a circuit board assembly. According to the system, a multi-surface array camera carries out real-time image acquisition on a circuit board assembly on a production line conveyor belt; the AI reasoning server is used for analyzing the images acquired by the multi-surface array camera so as to identify whether the circuit board assembly has any one abnormal condition of too close assembly distance, assembly overlapping, assembly folding or assembly deflection; the communication module sends an exception handling instruction generated when the AI reasoning detects that an exception state exists to a conveyor belt control end and a mobile mechanical arm control end; the conveyor belt control end is used for controlling the conveyor belt to stop running when receiving the exception handling instruction sent by the communication module, and recovering running of the conveyor belt when receiving a completion instruction; and when the conveyor belt stops running, the mobile mechanical arm identifies, positions, grabs and homes the abnormal circuit board assembly according to a coordinate range given by AI reasoning, and outputs a completion instruction after processing is completed.

Description

Technical Field

[0001] This invention relates to the field of PCB processing technology, and in particular to a circuit board assembly over-board processing system and method. Background Technology

[0002] In the electronics assembly industry, circuit board assemblies require numerous processing steps, including solder paste application, visual inspection, and final inspection. During the final inspection or internal inspection stages in the printed circuit board industry, surface quality checks are performed on the top and bottom sides of the circuit board. Circuit board assemblies are typically transported via guide rails. In actual production, issues such as wear on the guide rail components and inadequate lubrication often lead to collisions or overlaps between adjacent circuit board assemblies, preventing them from proceeding to the next process.

[0003] While the PCB industry has seen significant improvements in its level of automation over the years, manual intervention is still required in non-standard automated board processing processes. This necessitates coordination among production line leaders, engineers, and personnel responsible for upstream and downstream processes, resulting in significant variability and instability. Therefore, in the context of promoting intelligent manufacturing, fully automated processes, and "dark factories," there is an urgent need to design a fully automated, unmanned board processing solution to address the cumbersome manual intervention required in existing PCB board processing procedures. Summary of the Invention

[0004] Therefore, it is necessary to provide a circuit board assembly board-passing system and method to address the above-mentioned technical problems, providing a fully automated board-passing solution that can solve the problems caused by the need for manual intervention in board-passing.

[0005] A circuit board assembly through-board processing system, comprising: Multi-plane array camera is used to acquire real-time images of circuit board assemblies on the production line conveyor belt; The AI ​​inference server is used to analyze the images acquired by the multi-plane array camera to identify any abnormalities in the circuit board components, such as components being too close together, components overlapping, components being folded, or components being skewed. The communication module is used to send the exception handling instructions generated by the AI ​​inference server when it detects an abnormal state to the conveyor belt control terminal and the mobile robotic arm control terminal. The conveyor belt control terminal is used to control the conveyor belt to stop running when it receives an abnormal handling command from the communication module, and to resume the operation of the conveyor belt when it receives a completion command. A mobile robotic arm is used to identify, locate, grasp, and return abnormal circuit board components to their original positions based on the coordinate range given by the AI ​​inference server during periods when the conveyor belt is not running, and outputs a completion command after processing.

[0006] Furthermore, the AI ​​inference server is configured to recognize at least one of the following anomalies: The distance between adjacent circuit board assemblies is too close, less than a preset threshold. The edge regions of the circuit board assemblies overlap with each other; The circuit board assembly has bends or deformed folds; The circuit board assembly is placed on the conveyor belt at an angle that deviates from the preset reference angle range.

[0007] Furthermore, the mobile robotic arm includes: The control host is used to receive exception handling instructions issued by the AI ​​inference server; The motion actuator has multiple joint degrees of freedom to perform movement, rotation, and grasping actions; End grippers are used to grip and place abnormal circuit board assemblies; The binocular vision positioning module is used to collect the spatial position information of the circuit board assembly and transmit it to the control host to realize the positioning operation of the circuit board assembly.

[0008] Furthermore, the mobile robotic arm further includes: A torque sensor is used to detect the force generated by the end gripper during gripping or handling; Anti-collision mechanism, used to sense contact with external obstacles during movement and limit the movement of robotic arm; The control feedback unit is used to transmit the data collected by the torque sensor and the anti-collision mechanism to the control host.

[0009] Furthermore, the AI ​​inference server further includes: The model management unit is used to store multiple deep learning-based AI visual recognition models; The update unit is used to receive externally input model update data and replace, expand, or upgrade the AI ​​visual recognition model. The feature extraction unit is used to preprocess image data from various production lines and generate corresponding image feature vectors; The model call scheduling unit is used to match the image feature vector generated by the feature extraction unit with the preset production line model index table, determine the AI ​​visual recognition model corresponding to the current production line, and call the model to execute the image recognition task of the circuit board assembly of the corresponding production line. The model switching unit is used to switch the recognition model used by the AI ​​inference server to the corresponding target model when image data input from different production lines is detected.

[0010] Furthermore, the communication module is a WiFi6-NCP low-power communication module, comprising: A wireless data access unit is used to receive anomaly handling instructions output by the AI ​​inference server; A wireless data transmission unit is used to send the abnormality handling command to the conveyor belt control module and the mobile robotic arm.

[0011] Furthermore, the conveyor belt control terminal includes: A stop command receiving unit is used to stop the conveyor belt when it receives a stop command sent by the communication module; A recovery command receiving unit is used to restart the conveyor belt when a recovery command is received from the communication module; The control interface unit is used to connect to the conveyor belt driver via electrical signals to switch the operating status of the conveyor belt.

[0012] Furthermore, the system further includes a background log management module, which includes: A data receiving unit is used to receive the anomaly determination results from the AI ​​inference server; A data storage unit is used to store the anomaly determination results; The report generation unit is used to generate reports of abnormal events related to the production line based on the stored abnormality determination results.

[0013] Furthermore, the multi-path array camera is an industrial camera and includes: Imaging sensors are used to acquire images of the surface of circuit board assemblies; The data interface module is used to transmit the image data acquired by the imaging sensor to the AI ​​inference server via a network interface; Mounting brackets are used to fix the area array cameras at different locations on the production line.

[0014] A method for handling circuit board assemblies includes: a multi-plane array camera acquiring real-time images of the circuit board assemblies on a production line conveyor belt and transmitting the acquired image data to an AI inference server; the AI ​​inference server analyzing the image data to determine if the circuit board assemblies exhibit any of the following abnormalities: components are too close together, components overlap, components are folded, or components are skewed; when an abnormality is detected, the AI ​​inference server generates an abnormality handling instruction and sends it to the conveyor belt control unit and a mobile robotic arm via a communication module; upon receiving the abnormality handling instruction, the conveyor belt control unit stops the conveyor belt; the mobile robotic arm, based on the abnormality handling instruction and a given coordinate range, identifies, locates, grasps, and repositions the abnormal circuit board assembly; after completing the handling of the abnormal circuit board assembly, the mobile robotic arm sends a completion instruction to the conveyor belt control unit; upon receiving the completion instruction, the conveyor belt control unit resumes conveyor belt operation.

[0015] In any of the solutions provided above, the time required for anomaly handling is shortened through fully automated system collaboration, and 24-hour uninterrupted operation is supported. This fully adapts to the intelligent manufacturing requirements of fully automated processes, eliminating the variability and instability of manual intervention. It also reduces the risk of panel damage and scrap. Manual intervention can easily lead to losses such as localized deformation and breakage of panels due to improper operation. In this embodiment, a mobile robotic arm is used, and the grasping and repositioning process is contactless, reducing the secondary damage rate of panels and directly reducing production losses from the scrapping of a large number of panels in a short period. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a circuit board assembly board-passing system illustrated in an exemplary embodiment of this application; Figure 2 This is a schematic diagram illustrating the process of transporting circuit boards on a conveyor belt in a circuit board assembly overpass system, which is an exemplary embodiment of this application. Figure 3 This is a schematic diagram illustrating the positional relationship between a robotic arm and a camera in a circuit board assembly over-board processing system, as shown in another exemplary embodiment of this application. Figure 4 A schematic diagram of a robotic arm shown as another exemplary embodiment of this application; Figure 5This is a schematic diagram illustrating a circuit board assembly over-board processing method as another exemplary embodiment of this application. Detailed Implementation

[0018] To make the technical problems solved by the present invention, the technical solutions and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0019] It should be understood that the embodiments described below represent essential information to enable those skilled in the art to implement the embodiments and to illustrate the best mode of implementation. Upon reading the following description in conjunction with the accompanying drawings, those skilled in the art will understand the concepts of this disclosure and recognize the applications of these concepts not specifically mentioned herein. It should be understood that these concepts and applications fall within the scope of this disclosure and the appended claims.

[0020] It should also be understood that the terms “upper,” “lower,” “left,” “right,” “front,” “back,” “bottom,” “middle,” “top,” etc., may be used herein to describe various elements, and the orientation or positional relationship indicated is based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, these elements should not be limited by these terms.

[0021] To be further understood, the terms “comprising” or “including” as used herein specify the presence of the said feature, integer, step, operation, component and / or device, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, devices and / or groups thereof.

[0022] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It will be further understood that the terms used herein should be interpreted as having the same meaning as they mean in the context of this specification and related art, and will not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0023] In one embodiment, please refer to the following: Figures 1-4 As shown, a circuit board assembly through-board processing system is provided, comprising: Multi-plane array camera is used to acquire real-time images of circuit board assemblies on the production line conveyor belt; The AI ​​inference server is used to analyze the images acquired by the multi-plane array camera to identify any abnormalities in the circuit board components, such as components being too close together, components overlapping, components being folded, or components being skewed. The communication module is used to send the exception handling instructions generated by the AI ​​inference server when it detects an abnormal state to the conveyor belt control terminal and the mobile robotic arm control terminal. The conveyor belt control terminal is used to control the conveyor belt to stop running when it receives an abnormal handling command from the communication module, and to resume the operation of the conveyor belt when it receives a completion command. A mobile robotic arm is used to identify, locate, grasp, and return abnormal circuit board components to their original positions based on the coordinate range given by the AI ​​inference server during periods when the conveyor belt is not running, and outputs a completion command after processing.

[0024] Multi-plane array cameras, such as Figure 3 As shown, industrial-grade area scan cameras are deployed at key locations along each conveyor belt to acquire high-definition images of the Printed Circuit Board Assembly (PCBA) in real time. These cameras support data transmission via a network interface. The PCBA can also be referred to as a circuit board component, hereinafter simply referred to as a component or assembly. The AI ​​inference server is the backend processing center integrating a deep learning visual model, responsible for receiving images acquired by the cameras and analyzing anomalies. For example, the core model can be based on the Yolov10 object detection algorithm. The communication module handles the instruction transmission between the AI ​​inference server and the conveyor belt control unit and the mobile robotic arm. The conveyor belt control unit is the control unit electrically connected to the conveyor belt driver, which receives instructions from the communication module to start, stop, and resume the conveyor belt. Figure 2 As shown, it includes a loading area and a unloading area. A conveyor belt is located between the loading area and the unloading area. The conveyor belt is used to transport the plates from the loading area to the unloading area. The conveyor belt can transport the plates through torque guide rails.

[0025] A mobile robotic arm is a robotic device with autonomous movement and grasping capabilities. It can intervene and handle abnormal parts based on the coordinate range given by the AI ​​inference server. For example... Figure 3 As shown, the mobile robotic arm can move around the conveyor belt.

[0026] In this embodiment, abnormal situations specifically refer to board transfer problems caused by at least one of the following four types of effects during the transfer of circuit board assemblies: components being too close together, components overlapping, components folding, and components being skewed.

[0027] For example, suppose a scenario includes three parallel fully automated conveyor belts responsible for transporting circuit board assemblies to subsequent processes such as solder paste inspection and visual inspection. In actual production, due to issues such as wear on the conveyor belt rails and inadequate lubrication, circuit board assemblies often become too close together, overlapping, folded, or skewed, leading to jamming. Traditional solutions require manual coordination with production line supervisors and engineers, resulting in response delays and a high risk of board breakage.

[0028] In this embodiment, a 12-megapixel industrial area scan camera is deployed at the inlet, middle, and outlet ends of three conveyor belts, for a total of nine area scan cameras. These cameras are fixed above and to the sides of the conveyor belts using mounting brackets to ensure coverage of the entire component transport process. Each camera acquires images of the TOP and BOTTOM surfaces of the circuit board assembly in real time at a set frequency, and the high-definition image data is synchronously transmitted to the AI ​​inference server.

[0029] After receiving nine streams of image data, the AI ​​inference server first performs preprocessing such as noise reduction, cropping, and scale normalization to extract key features such as board edges, spacing, and pose. Then, it calls a PCB anomaly detection model optimized based on Yolov10 to perform real-time inference on the preprocessed images. For example, this includes the following detections: If the distance is too close: the distance between adjacent plates is less than 5cm (preset threshold), such as the distance between two plates in the middle section of the second conveyor belt being only 3cm; Determine if the overlapping area of ​​the edges of the plates is greater than 10%, such as when the plates at the feed end of the first conveyor belt partially overlap due to offset. Determining folding: The sheet metal has obvious bending deformation (bending angle > 15°), such as the sheet metal at the discharge end of the 3rd conveyor belt folding due to collision; Determine if the plate placement angle deviates from the preset reference (0° horizontal placement) by more than ±10°, such as the plate on the second conveyor belt being skewed by 18°.

[0030] When an anomaly of excessively close panel spacing is detected in the middle section of the second conveyor belt, the AI ​​inference server generates an anomaly handling instruction containing the anomaly production line number (2), the anomaly type (excessively close spacing), and the anomaly coordinate range. The communication module performs bidirectional instruction transmission: it receives the anomaly handling instruction output by the AI ​​inference server through the wireless data access unit; and simultaneously sends the anomaly handling instruction to both the control terminal of the second conveyor belt and the control terminal of the mobile robotic arm through the wireless data transmission unit, ensuring synchronous and delay-free instruction delivery.

[0031] Upon receiving the anomaly handling command, the control unit of conveyor belt No. 2 immediately sends an electrical signal to the conveyor belt driver via the control interface unit, stopping the conveyor belt to prevent further compression of the abnormal board, which could lead to increased overlap or folding. After receiving the anomaly handling command, the mobile robotic arm autonomously moves to the middle section of conveyor belt No. 2 based on the coordinate range provided by the AI ​​inference server, confirming the position and spacing of the abnormal board using its built-in vision module. The robotic arm's end effector, made of flexible material, precisely grasps the PCB that is too close to the previous board and moves it to the standard position with the previous board, completing the repositioning operation. After intervention, the mobile robotic arm sends an anomaly handling completion command to the conveyor belt control unit via the communication module. Upon receiving the completion command, the conveyor belt control unit immediately sends a start signal to the driver via the control interface unit, and the conveyor belt resumes board passing. The entire anomaly handling process is short and does not cause production interruption.

[0032] As can be seen, in this embodiment, existing technologies for achieving unmanned production rely on manual coordination between production line leaders and engineers to handle stuck boards, resulting in response delays and significant operational randomness. This embodiment, through fully automated system collaboration, shortens the time required for anomaly handling and supports 24 / 7 uninterrupted operation, fully adapting to the intelligent manufacturing requirements of fully automated processes and eliminating the variability and instability of manual intervention. It also reduces the risk of board damage and scrap. Manual intervention can easily lead to losses such as localized deformation and breakage of boards due to improper operation. This embodiment utilizes a mobile robotic arm, eliminating human contact during the gripping and repositioning process, reducing the rate of secondary damage to boards, and directly reducing production losses from the scrapping of a large number of boards in a short period.

[0033] In one embodiment, the AI ​​inference server is configured to recognize at least one of the following anomalies: The distance between adjacent circuit board assemblies is too close, less than a preset threshold. The edge regions of the circuit board assemblies overlap with each other; The circuit board assembly has bends or deformed folds; The circuit board assembly is placed on the conveyor belt at an angle that deviates from the preset reference angle range.

[0034] In this embodiment, "too close" refers to a scenario where the distance between adjacent circuit board assemblies on the conveyor belt is less than a preset threshold, which is a pre-existing risk scenario that may lead to stacking due to subsequent board pushing. "Overlap" refers to a situation where the edge areas of circuit board assemblies overlap each other, with the covered area accounting for ≥15% of the surface area of ​​a single board, which may be a direct anomaly type that may prevent the board from entering the next process. "Folding" refers to a core risk where the circuit board assembly is bent or deformed due to collision or compression, which may lead to damage to the board circuit and scrap the board. "Skewing" refers to a situation where the placement angle of the circuit board assembly on the conveyor belt deviates from the preset reference angle range (e.g., the reference is 0° horizontal placement, with an allowable deviation of ±12°), which may cause the board to rub against the conveyor belt guardrail and cause jamming, which is a posture anomaly in the board passing process.

[0035] AI inference servers can accurately determine the above four types of anomalies through image feature extraction and comparison, providing type labels and coordinate data for subsequent intervention.

[0036] As examples, the AI ​​inference server can use edge detection algorithms to locate the long edges of adjacent boards and calculate the edge spacing. For instance, if the edge spacing between two circuit board assemblies in the middle of the conveyor belt is 3.2cm, which is less than the preset threshold of 5cm, it is determined to be too close; a warning label that is likely to cause overlap is marked, and the abnormal coordinates are output simultaneously.

[0037] As examples, the AI ​​inference server can extract the outline of board components using contour segmentation algorithms and calculate the pixel area ratio of overlapping areas. For instance, if the edge overlap area of ​​two circuit board assemblies on a conveyor belt reaches 22% (the area of ​​a single board is 150cm², and the overlap area is 33cm²), it is determined to be an overlap; an emergency label indicating high risk of board jamming is marked, and the coordinates of the overlap center are output.

[0038] As examples, the AI ​​inference server can acquire the three-dimensional shape of a board component using a 3D contour reconstruction algorithm (combined with angle offset acquisition from an area array camera) and calculate the angle between the bending vertex and the reference plane. For instance, if a circuit board assembly on a conveyor belt bends by 28° due to a collision, exceeding the 20° threshold, it is determined to be folded; a high-risk label is marked on the board component indicating a high risk of scrapping, and the coordinates of the bending vertex are output.

[0039] As examples, the AI ​​inference server can use the Hough transform algorithm to detect the angle between the long side of a board and the conveyor belt's transport direction (X-axis). For instance, if a circuit board assembly on the conveyor belt is placed at an angle of -25° (skewed to the left), exceeding the ±12° range, it is determined to be skewed; a warning label is marked, and the center coordinates of the board are output.

[0040] The AI ​​inference server integrates the identified anomaly type, risk level, and coordinate range into anomaly handling instructions, which are then synchronously sent to the corresponding conveyor belt control unit and the mobile robotic arm via the communication module. For example, for an anomaly indicating that the conveyor belt is too close, the instruction may include: anomaly type: too close, coordinates: X850-1100 / Y280-350, intervention priority: medium. Upon receiving the instruction, the conveyor belt control unit immediately stops operating, and the mobile robotic arm adjusts the end effector gripper posture according to the anomaly type.

[0041] In this embodiment, the four types of anomaly identification are highly accurate through the Yolov10 optimization model and quantization threshold determination, clearly distinguishing the anomaly types and providing a precise basis for intervention by the mobile robotic arm, avoiding secondary jamming caused by misjudgment and incorrect repair. The AI ​​inference server has a fast single-frame image processing time, and the response speed for anomaly type determination is significantly improved compared to manual methods. Furthermore, it synchronously outputs anomaly coordinates (such as the bending vertex of a folded plate), eliminating the need for secondary positioning by the robotic arm and further shortening the intervention preparation time, thus meeting the core requirement of rapidly handling jamming issues.

[0042] In addition, in one embodiment, by quantifying the risk of anomalies, the production line preventive maintenance system can mark anomalies with warning / emergency / high-risk labels (such as warning for being too close and high-risk for folding). The system can accurately locate the root causes of problems such as slide rail wear and oil drying by statistically analyzing the frequency of different anomaly types in the background, and optimize the maintenance process in advance to reduce the problem of jamming caused by inadequate equipment maintenance.

[0043] In one embodiment, the mobile robotic arm includes: The control host is used to receive exception handling instructions issued by the AI ​​inference server; The motion actuator has multiple joint degrees of freedom to perform movement, rotation, and grasping actions, such as... Figure 4 As shown, it may include a motion mechanism for the air pump pipe, which has the ability to rotate about the RX / RY / RZ directions.

[0044] End grippers are used to grip and place abnormal circuit board assemblies; The binocular vision positioning module is used to collect the spatial position information of the circuit board assembly and transmit it to the control host to realize the positioning operation of the circuit board assembly.

[0045] The mobile robotic arm in this embodiment, combined with binocular vision positioning and a multi-degree-of-freedom actuator, can achieve millimeter-level positioning and flexible grasping of abnormal plates, solving the problems of inaccurate positioning and stiff operation that may exist when traditional robotic arms handle plate-crossing issues.

[0046] The control host is the core control unit of the mobile robotic arm, used to receive anomaly handling instructions (including production line number, anomaly coordinates, and board model) from the AI ​​inference server. The motion actuator can be a robotic arm body with 6 degrees of freedom, capable of movement along the X / Y / Z axes and 360° rotation, adapting to the intervention needs of boards at different positions and angles. The end effector uses a flexible silicone gripper assembly with adjustable gripping force to avoid damaging the surface circuitry of the board during gripping. The binocular vision positioning module consists of two industrial cameras spaced apart, which acquire the spatial coordinates of the board through triangulation.

[0047] For example, in this robotic arm, the control host receives instructions from the AI ​​inference server via the communication module, including the following information: production line 2, anomaly type: overlap, coordinate range X: 1000-1200mm / Y: 300-350mm, board model: PCB-2024A. The mobile robotic arm travels to the middle section of production line 2, and the binocular vision positioning module acquires images of the overlapping boards, calculating the center coordinates (X: 1100mm / Y: 325mm / Z: 50mm) and tilt angle (10°) of the upper board. This data is transmitted to the control host in real time. Based on the positioning data, the control host schedules the motion actuator to move along the X-axis to 1100mm, the robotic arm body rotates 10° to match the board angle, and the end effector opens to the appropriate board width (15cm). The end effector grips the upper board with a 10N gripping force, and the motion actuator moves it to a temporary placement area (coordinates X: 800mm / Y: 325mm). After returning to its original position, the gripper is released.

[0048] In this embodiment, millimeter-level positioning and flexible operation are achieved, ensuring effective positioning and preventing slippage of the panels during intervention. Binocular vision positioning accurately captures the center coordinates of overlapping panels (e.g., X: 1100mm / Y: 325mm), the 6-DOF mechanism adapts to the panel tilt angle, and the flexible clamp grips with appropriate force, reducing the potential for panel breakage due to human intervention. Furthermore, this embodiment, through positioning, path planning, and gripping collaborative logic, can handle anomalies from different production lines and with different postures (e.g., overlapping panels in production line 2, skewed panels in production line 4), adapting to complex scenarios involving multi-line monitoring without requiring separate panel intervention programs for different anomalies. The control host can automatically parse the panel model and anomaly coordinates from the AI ​​instructions, eliminating the need for manual programming to adjust the robotic arm parameters, reducing operational complexity, and ensuring high repeatability of intervention actions, guaranteeing consistency in batch processing.

[0049] In one embodiment, the mobile robotic arm further includes: A torque sensor is used to detect the force generated by the end gripper during gripping or handling; Anti-collision mechanism, used to sense contact with external obstacles during movement and limit the movement of robotic arm; The control feedback unit is used to transmit the data collected by the torque sensor and the anti-collision mechanism to the control host.

[0050] A torque sensor, installed at the joint between the end effector and the robotic arm, is a force-sensing element that detects torque changes during gripping and handling in real time and converts them into gripping force data. The anti-collision mechanism consists of an array of pressure sensors on the surface of the robotic arm; when the contact force is greater than or equal to a set value, a braking signal is triggered. The control feedback unit transmits data from the torque sensor and the anti-collision mechanism to the control host in real time, supporting dynamic adjustments.

[0051] For example, when a mobile robotic arm performs a task of repositioning a tilted plate, the end effector contacts the ultra-thin plate, and the torque sensor detects an initial gripping force of A. The control feedback unit transmits this data to the control host. The control host determines that A exceeds the safety threshold of the ultra-thin plate and immediately issues a command to reduce the gripping force to B. The torque sensor provides real-time feedback of the adjusted data to ensure stable and damage-free gripping. When the robotic arm moves with the plate, the anti-collision mechanism contacts the conveyor belt guardrail, immediately triggering a braking signal. The control feedback unit transmits the collision position data to the control host. The control host schedules the motion actuator to fine-tune the path (offset along the Y-axis), avoiding the guardrail and resuming movement to complete the plate repositioning. This protects ultra-thin / fragile plates. Ultra-thin plates can withstand gripping forces ≤ B, while the fixed force (A) of traditional robotic arms easily leads to bending deformation. In this embodiment, the torque sensor detects the gripping force in real time, and the control host can adjust the force value from A to B or less than B, reducing the plate damage rate and solving the problem of plate scrapping due to manual / mechanical operations. In addition, in this embodiment, the anti-collision mechanism triggers braking when the contact external force is greater than or equal to the set value. The control feedback unit synchronously transmits the collision position, and the host quickly fine-tunes the path, reducing the equipment collision failure rate and eliminating production interruptions caused by manual maintenance.

[0052] It is also worth noting that the real-time data feedback from the torque sensor and the anti-collision mechanism allows the robotic arm to dynamically adjust its movements based on the condition of the plate (such as weight and surface hardness) and environmental obstacles, thereby improving the success rate of intervention and ensuring the need for stable plate intervention.

[0053] Furthermore, such as Figure 4 As shown, the mobile robotic arm also has a moving device such as moving wheels.

[0054] In one embodiment, the AI ​​inference server further includes: The model management unit is used to store multiple deep learning-based AI visual recognition models; The update unit is used to receive externally input model update data and replace, expand, or upgrade the AI ​​visual recognition model. The feature extraction unit is used to preprocess image data from various production lines and generate corresponding image feature vectors; The model call scheduling unit is used to match the image feature vector generated by the feature extraction unit with the preset production line model index table, determine the AI ​​visual recognition model corresponding to the current production line, and call the model to execute the image recognition task of the circuit board assembly of the corresponding production line. The model switching unit is used to switch the recognition model used by the AI ​​inference server to the corresponding target model when image data input from different production lines is detected.

[0055] In this embodiment, the model management unit stores four model-specific models optimized based on Yolov10 (corresponding to four types of board parts). The update unit receives external model update packages via an Ethernet interface to replace models or upgrade versions (such as importing models adapted to new board parts). The feature extraction unit performs edge detection and texture analysis on the images to generate board size and layout feature vectors. The model call scheduling unit stores a production line-model-model index table and determines the target model through feature vector matching. The model switching unit completes model switching when different types of board part images are input.

[0056] In this embodiment, during implementation, a multi-plane array camera acquires PCB images of production line 1, and the feature extraction unit generates feature vectors. The model call scheduling unit matches the feature vectors with an index table to determine the corresponding motherboard model and calls the dedicated recognition model (model A). When production line 1 switches to conveying power supply boards, the feature extraction unit generates new vectors, and the model switching unit switches the current model from A to the power supply board-specific model B. An update package of an external new control board model C is received, the update unit automatically replaces the old model, and the model management unit records the version iteration log.

[0057] In this embodiment, dedicated models are deployed according to each board type. Precise feature vector calls are used to accurately identify anomalies in the four types of boards, resolving the issue of poor adaptability in mixed-model production lines. Furthermore, automatic model switching and updates are supported, meeting the goal of full automation with reduced manual intervention. When a new type of control board is added, there is no need to replace the AI ​​inference server hardware; simply importing the dedicated model through the update unit allows for rapid adaptation to the new scenario, reducing equipment iteration costs for enterprises.

[0058] In one embodiment, the communication module is a WiFi6-NCP low-power communication module, comprising: A wireless data access unit is used to receive anomaly handling instructions output by the AI ​​inference server; A wireless data transmission unit is used to send the abnormality handling command to the conveyor belt control module and the mobile robotic arm.

[0059] WiFi6-NCP Low Power Communication Module: A wireless communication component using the IEEE 802.11ax protocol, with a transmission rate of up to 1.2Gbps and standby power consumption ≤5mA (defined as "WiFi6-NCP Low Power Communication Module").

[0060] Wireless data access unit: Equipped with 2.4G / 5G dual-band receiving capability, used to receive anomaly handling instructions output by the AI ​​inference server.

[0061] Wireless data transmission unit: Supports simultaneous communication of multiple devices (up to 16 terminals), and distributes instructions to the conveyor belt control terminal and the mobile robotic arm.

[0062] Traditional WiFi modules have high transmission latency, causing the conveyor belt to stop and the robotic arm to start asynchronously, which can easily lead to continued sliding and collisions of components. The WiFi6-NCP low-power communication module in this embodiment has low transmission latency, allowing AI commands to be sent synchronously to both the conveyor belt control unit and the robotic arm. The short response time difference between the two prevents secondary anomalies caused by asynchronous commands.

[0063] In one embodiment, the conveyor belt control terminal includes: A stop command receiving unit is used to stop the conveyor belt when it receives a stop command sent by the communication module; A recovery command receiving unit is used to restart the conveyor belt when a recovery command is received from the communication module; The control interface unit is used to connect to the conveyor belt driver via electrical signals to switch the operating status of the conveyor belt.

[0064] The conveyor belt control terminal of this embodiment can achieve accurate command reception and state switching, solve the problem of chaotic linkage control, and adapt to the needs of multi-production line collaborative operation.

[0065] The stop command receiving unit can receive stop commands from the communication module via an RS485 interface and has a command verification function to avoid false triggering. The recovery command receiving unit is independent of the stop unit's receiving module and only responds to the completion command sent by the moving robotic arm to prevent premature recovery. Control interface unit: directly electrically connected to the conveyor belt driver.

[0066] For example, if a board overlap anomaly occurs on production line 2, the stop command receiving unit receives the command from the communication module. After verification, it sends a stop control signal to the driver through the control interface unit, and the conveyor belt stops within a certain time. The control terminal automatically identifies the associated production lines (1 and 3) of production line 2, and the stop command receiving unit synchronously sends stop commands to them to ensure that the linkage is paused. After the mobile robotic arm completes its intervention, it sends a completion command. The recovery command receiving unit independently receives and verifies the command, and after verification, sends a recovery signal through the control interface unit. The recovery command receiving units of the associated production lines synchronously receive the signals, and the four conveyor belts resume linkage operation within a certain time.

[0067] In this embodiment, the stop / recovery unit is designed independently, and by verifying the integrity of the instructions, the rate of false stop / start is reduced, and the number of meaningless pauses of related production lines is effectively reduced, directly improving capacity utilization. In this embodiment, the control terminal can automatically identify the correlation (such as production line 2 abnormally suspending production lines 1 and 3 simultaneously), with short start / stop response time and improved linkage and coordination.

[0068] In one embodiment, the system further includes a background log management module, the background log management module comprising: A data receiving unit is used to receive the anomaly determination results from the AI ​​inference server; A data storage unit is used to store the anomaly determination results; The report generation unit is used to generate reports of abnormal events related to the production line based on the stored abnormality determination results.

[0069] The background log management module in this application embodiment can realize automatic storage of abnormal data and report generation, solve the problems of difficult fault tracing and lack of optimization basis, and adapt to the needs of data-driven optimization in intelligent manufacturing.

[0070] The data receiving unit receives abnormal data (including production line, type, time, and processing result) from the AI ​​inference server via an Ethernet interface, with the receiving frequency synchronized with the anomaly detection. The data storage unit uses a database to store data and supports indexing by production line, date, and anomaly type. The report generation unit supports automatic generation of daily / weekly / monthly anomaly reports, including dimensions such as anomaly frequency, type percentage, and processing time.

[0071] For example, this embodiment takes a PCB production workshop in the electronic assembly industry as the application object. The workshop has 8 production lines and about 200 board jamming abnormalities occur every month. Traditional manual recording of faults (paper ledgers) has problems such as incomplete information, difficulty in tracing, and time-consuming report generation. It is impossible to accurately analyze the reasons for the high incidence of abnormalities (such as specific production lines or specific time periods), resulting in a lack of data support for production line optimization.

[0072] In this example, a "distance too close" anomaly occurred on production line 5. The data receiving unit received data from the AI ​​inference server: Production line 5, type: distance too close, time: 2024-09-20 14:30:22, processing time: 25 seconds. The data storage unit wrote the data into the database, creating an index based on "Production line 5-20240920-distance too close" for easy retrieval. At 23:59 daily, the report generation unit automatically compiled the data for the day and generated the "August 20th Production Line Anomaly Report," which showed 3 instances of "distance too close" anomalies on production line 5, accounting for 20% of the total anomalies for the day.

[0073] As can be seen, in this embodiment, indexing by production line-date-type (e.g., production line 5-20240920-too close distance) shortens traceability time and allows for the reconstruction of the entire anomaly handling process (e.g., anomaly at 14:30, processed in 25 seconds), which is beneficial for troubleshooting anomalies and allows engineers to optimize accordingly. It provides data support for production management and quality auditing, and meets the compliance requirements of digital traceability in intelligent manufacturing.

[0074] In one embodiment, the multi-path array camera is an industrial camera and includes: Imaging sensors are used to acquire images of the surface of circuit board assemblies; The data interface module is used to transmit the image data acquired by the imaging sensor to the AI ​​inference server via a network interface; Mounting brackets are used to fix the area array cameras at different locations on the production line.

[0075] The industrial camera features a dustproof and oil-resistant design, making it suitable for harsh workshop environments. The imaging sensor can utilize a CMOS sensor to capture minute details of the workpiece. The data interface module supports a Gigabit Ethernet interface and transmits images via the RTSP protocol. The mounting bracket is an adjustable metal bracket for securing the camera to the side or above the production line.

[0076] Camera Deployment and Debugging: One industrial camera was deployed at the inlet, middle, and outlet ends of each of the four conveyor belts. The cameras were adjusted to a height of 1.5m and an angle of 45° using mounting brackets to ensure coverage of the top and bottom surfaces of the boards. The imaging sensor acquired images of the boards at 16 megapixels and 30 frames per second, capturing details of skewness and overlap in the three boards on the production line. The data interface module transmitted the images to the AI ​​inference server via Gigabit Ethernet using the RTSP protocol, resulting in reduced transmission latency and no image distortion.

[0077] In this embodiment, flexible deployment covers the entire scene, reducing the number of cameras deployed. The adjustable bracket supports height / angle adjustment, and the monitoring range of a single camera is increased by 2 times compared to a fixed camera. Nine cameras can cover the entire length of three conveyor belts. Compared to traditional dense deployment solutions, the number of cameras is reduced, thus lowering hardware costs.

[0078] In one embodiment, such as Figure 5 As shown, a method for processing circuit board assemblies includes the following steps: S11, a multi-plane array camera performs real-time image acquisition of the circuit board assembly on the production line conveyor belt and transmits the acquired image data to an AI inference server; S12, the AI ​​inference server analyzes the image data to determine whether the circuit board assembly has any of the following abnormal conditions: components are too close together, components overlap, components are folded, or components are skewed; S13, when an abnormal condition is detected, the AI ​​inference server generates an abnormality handling instruction and sends it to the conveyor belt control terminal and the mobile robotic arm through a communication module; S14, the conveyor belt control terminal stops the conveyor belt after receiving the abnormality handling instruction; S15, the mobile robotic arm identifies, locates, grasps, and returns the abnormal circuit board assembly according to the abnormality handling instruction and a given coordinate range; S16, after the mobile robotic arm completes the processing of the abnormal circuit board assembly, it sends a completion instruction to the conveyor belt control terminal; S17, the conveyor belt control module resumes the operation of the conveyor belt after receiving the completion instruction.

[0079] It is understood that, based on the corresponding beneficial effects of the above embodiments, the method of this embodiment is obtained from the processing system described in the above embodiments and should also have the corresponding technical effects. To avoid repetition, it will not be described again here.

[0080] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A circuit board assembly board-passing processing system, characterized in that, include: Multi-plane array camera is used to acquire real-time images of circuit board assemblies on the production line conveyor belt; The AI ​​inference server is used to analyze the images acquired by the multi-plane array camera to identify any abnormalities in the circuit board components, such as components being too close together, components overlapping, components being folded, or components being skewed. The communication module is used to send the exception handling instructions generated by the AI ​​inference server when it detects an abnormal state to the conveyor belt control terminal and the mobile robotic arm control terminal. The conveyor belt control terminal is used to control the conveyor belt to stop running when it receives an abnormal handling command from the communication module, and to resume the operation of the conveyor belt when it receives a completion command. A mobile robotic arm is used to identify, locate, grasp, and return abnormal circuit board components to their original positions based on the coordinate range given by the AI ​​inference server during periods when the conveyor belt is not running, and outputs a completion command after processing.

2. The system according to claim 1, characterized in that, The AI ​​inference server is configured to recognize at least one of the following anomalies: The distance between adjacent circuit board assemblies is too close, less than a preset threshold. The edge regions of the circuit board assemblies overlap with each other; The circuit board assembly has bends or deformed folds; The circuit board assembly is placed on the conveyor belt at an angle that deviates from the preset reference angle range.

3. The system according to claim 1, characterized in that, The mobile robotic arm includes: The control host is used to receive exception handling instructions issued by the AI ​​inference server; The motion actuator has multiple joint degrees of freedom to perform movement, rotation, and grasping actions; End grippers are used to grip and place abnormal circuit board assemblies; The binocular vision positioning module is used to collect the spatial position information of the circuit board assembly and transmit it to the control host to realize the positioning operation of the circuit board assembly.

4. The system according to claim 3, characterized in that, The mobile robotic arm further includes: A torque sensor is used to detect the force generated by the end gripper during gripping or handling; Anti-collision mechanism, used to sense contact with external obstacles during movement and limit the movement of robotic arm; The control feedback unit is used to transmit the data collected by the torque sensor and the anti-collision mechanism to the control host.

5. The system according to claim 1, characterized in that, The AI ​​inference server further includes: The model management unit is used to store multiple deep learning-based AI visual recognition models; The update unit is used to receive externally input model update data and replace, expand, or upgrade the AI ​​visual recognition model. The feature extraction unit is used to preprocess image data from various production lines and generate corresponding image feature vectors; The model call scheduling unit is used to match the image feature vector generated by the feature extraction unit with the preset production line model index table, determine the AI ​​visual recognition model corresponding to the current production line, and call the model to execute the image recognition task of the circuit board assembly of the corresponding production line. The model switching unit is used to switch the recognition model used by the AI ​​inference server to the corresponding target model when image data input from different production lines is detected.

6. The system according to claim 1, characterized in that, The communication module is a WiFi6-NCP low-power communication module, comprising: A wireless data access unit is used to receive anomaly handling instructions output by the AI ​​inference server; A wireless data transmission unit is used to send the abnormality handling command to the conveyor belt control module and the mobile robotic arm.

7. The system according to claim 1, characterized in that, The conveyor belt control terminal includes: A stop command receiving unit is used to stop the conveyor belt when it receives a stop command sent by the communication module; A recovery command receiving unit is used to restart the conveyor belt when a recovery command is received from the communication module; The control interface unit is used to connect to the conveyor belt driver via electrical signals to switch the operating status of the conveyor belt.

8. The system according to claim 1, characterized in that, The system further includes a background log management module, which includes: A data receiving unit is used to receive the anomaly determination results from the AI ​​inference server; A data storage unit is used to store the anomaly determination results; The report generation unit is used to generate reports of abnormal events related to the production line based on the stored abnormality determination results.

9. The system according to claim 1, characterized in that, The multi-path array camera is an industrial camera and includes: Imaging sensors are used to acquire images of the surface of circuit board assemblies; The data interface module is used to transmit the image data acquired by the imaging sensor to the AI ​​inference server via a network interface; Mounting brackets are used to fix the area array cameras at different locations on the production line.

10. A method for processing circuit board assemblies, characterized in that, include: A multi-plane array camera captures real-time images of circuit board assemblies on the production line conveyor belt and transmits the captured image data to an AI inference server. The AI ​​inference server analyzes the image data to determine if any of the following anomalies exist: components are too close together, components overlap, components are folded, or components are skewed. When an anomaly is detected, the AI ​​inference server generates an anomaly handling command and sends it to the conveyor belt control unit and the mobile robotic arm via a communication module. Upon receiving the anomaly handling command, the conveyor belt control unit stops the conveyor belt. Based on the anomaly handling command and a given coordinate range, the mobile robotic arm identifies, locates, grasps, and repositions the abnormal circuit board assembly. After processing the abnormal circuit board assembly, the mobile robotic arm sends a completion command to the conveyor belt control unit. Upon receiving the completion command, the conveyor belt control unit resumes operation.

Images