PCB intelligent sorting process and system based on unmanned factory

Abstract

The application discloses a kind of based on unmanned factory's PCB intelligent sorting process and system, including the surface image and internal structure parameter of real-time acquisition PCB board piece in image acquisition module and scanning module on sorting equipment, generate standardization feature dataset;Lightweight classification model combines process grade determination rule and is classified to PCB board piece and quality grade division, and based on MES system order demand dynamic distribution sorting priority, generate task queue;Path planning algorithm calculates the optimal grabbing path of mechanical arm and AGV conveying route according to priority queue, forms sorting instruction set and is issued to execution unit through industrial communication network;Mechanical arm completes PCB grabbing and placing operation according to instruction, while visual check module detects sorting result in real time, and outputs sorting state report and quality data;The process is realized through automated data acquisition, intelligent classification decision and closed-loop optimization mechanism, the efficiency, accuracy and unmanned of PCB sorting.

Description

Technical Field

[0001] This invention relates to the field of PCB sorting, and more particularly to a PCB intelligent sorting process and system based on unmanned factories. Background Technology

[0002] Printed circuit boards (PCBs), as core components of electronic devices, are undergoing rapid development in manufacturing processes towards high density and miniaturization. With the surge in demand for PCBs from consumer electronics, automotive electronics, and other sectors, traditional production models face severe challenges. Particularly in the sorting stage, PCBs need to be precisely classified according to model, process parameters (such as plating thickness and pad layout), and quality grade to meet subsequent packaging, warehousing, and order delivery requirements. Currently, under the trend of smart manufacturing, unmanned factories (lights-out factories) require fully automated operations, but PCB sorting still relies on manual visual inspection or semi-automated equipment, resulting in low efficiency, high error rates, and insufficient capacity to handle mixed-specification production lines, severely restricting enterprises' large-scale flexible production capabilities.

[0003] Current PCB sorting technologies primarily rely on mechanical positioning combined with basic visual recognition. For example, images of the PCB surface are captured using a fixed industrial camera, and then simple classification is performed based on threshold segmentation algorithms. This type of technology can only identify a single model or limited features (such as size and color) and requires a pre-set fixed gripping path, leading to the following core drawbacks: insufficient sorting accuracy and compatibility. For instance, in scenarios involving mixed types of PCBs (such as rigid boards, flexible boards, and high-frequency boards), traditional systems lack the ability to detect internal structural parameters (such as copper layer thickness and aperture distribution), making it impossible to distinguish boards with similar process complexity. This results in a sorting error rate exceeding 5%, necessitating manual re-inspection. Furthermore, fixed path planning algorithms struggle to adapt to dynamic order demands, causing sorting efficiency to stagnate below 5,000 pieces per day, failing to meet the requirements of intelligent manufacturing for high throughput (≥20,000 pieces / day) and zero-defect sorting.

[0004] Therefore, it is necessary to improve the existing PCB sorting technology to solve the technical problems of high sorting error rate and insufficient ability to handle mixed circuits of multiple board specifications. Summary of the Invention

[0005] The purpose of this invention is to provide a PCB intelligent sorting process and system based on unmanned factories, thereby solving the above-mentioned technical problems.

[0006] To achieve this objective, the present invention adopts the following technical solution: A PCB intelligent sorting process based on an unmanned factory includes the following steps: By deploying image acquisition and scanning modules on sorting equipment, surface image data and internal structural parameters of PCB boards are acquired in real time, and the model identification, pad distribution characteristics and process parameters of PCB boards are extracted to generate a standardized feature dataset. The feature dataset is input into the lightweight classification model, and the PCB board is classified by model and quality level according to the preset process level judgment rules. At the same time, the sorting priority is dynamically allocated based on the order requirements of the MES system to generate a sorting task priority queue. Based on the sorting task priority queue, the optimal grasping path of the robotic arm and the AGV conveying route are calculated by the path planning algorithm, a sorting instruction set is generated, and it is sent to the execution unit through the industrial communication network. Based on the sorting instruction set, the robotic arm is controlled to complete the gripping and placement of PCB boards. Simultaneously, the sorting results are inspected in real time by the vision verification module. If an abnormal board is detected, a rework signal is triggered, and a sorting completion status report and quality verification data are output.

[0007] Optionally, the image acquisition module and scanning module deployed on the sorting equipment acquire surface image data and internal structural parameters of the PCB board in real time, and extract the model identifier, pad distribution characteristics, and process parameters of the PCB board to generate a standardized feature dataset, specifically including: The industrial camera and scanning module deployed on the sorting equipment are used to acquire multi-angle images and scan the internal structure of the PCB board on the conveyor belt, simultaneously obtaining surface image data and internal structure parameters, and generating the original data packet based on the timestamp alignment mechanism; the internal structure parameters include the internal copper layer thickness distribution data and aperture characteristic parameters. The preprocessing of the original data packet specifically includes: performing adaptive denoising and edge enhancement operations on the surface image data to eliminate image distortion caused by uneven lighting and mechanical vibration; using a dynamic filtering algorithm to remove environmental interference signals from the internal scan data, and matching the reference value through a preset PCB process parameter library to generate denoised standard data units. Feature extraction of the standard data unit specifically includes: identifying model identification characters, positioning pad coordinates and spacing distribution from the surface image data, and extracting the average copper layer thickness, aperture distribution density and microcrack defect parameters from the internal scan data to form a multidimensional feature vector.

[0008] Optionally, feature extraction is performed on the standard data units, followed by: The multidimensional feature vector is input into the dynamic template generation module, and the heterogeneous feature parameters are converted into structured data of a unified dimension according to the preset standardization rules, and a unique identifier and timestamp label are added. The data validation engine performs integrity checks and logical verification on the standardized structured data, removes outliers, generates a standardized feature dataset containing complete metadata, and stores it in the database of the edge computing node.

[0009] Optionally, the feature dataset is input into a lightweight classification model, and the PCB boards are classified by model and quality grade according to the preset process grade judgment rules. At the same time, the sorting priority is dynamically allocated based on the order requirements of the MES system to generate a sorting task priority queue. The specific steps include: The feature dataset is input into a lightweight classification model. The model identification characters of the PCB board are semantically parsed through a pre-trained PCB model feature matching network to generate an initial model classification label set and associate it with the corresponding process parameter threshold range. Based on the preset process grade judgment rule library, each PCB board in the initial model classification label set is classified into quality grades, including: calculating the quality score and mapping it to the preset process grade range according to the uniformity of pad distribution, copper layer thickness deviation value and microcrack defect density, and generating an enhanced classification result with quality grade label; The system acquires order demand data from the MES system in real time, extracts order urgency coefficients, warehouse location status, and customer-customized parameters, constructs a multi-objective optimization function, dynamically calculates the sorting priority weight of each PCB board, and generates a priority weight matrix. The enhanced classification results are associated and matched with the priority weight matrix, and a sorting task priority queue is generated by a dynamic weighted sorting algorithm. The queue is then compiled into a task instruction package and transmitted to the path planning module.

[0010] Optionally, based on the sorting task priority queue, the optimal grasping path of the robotic arm and the AGV conveying route are calculated using a path planning algorithm to generate a sorting instruction set, which is then sent to the execution unit via an industrial communication network. This specifically includes the following steps: Based on the sorting task priority queue, the target storage location, process level constraints and time window requirements of the PCB board are extracted by the task parsing module to generate a sorting task parameter matrix containing spatiotemporal constraints. A hybrid path planning algorithm is adopted, which combines the kinematic model of the robotic arm with the dynamic obstacle avoidance rules of the AGV. With the optimization objective of minimizing the total sorting time and energy consumption, the robotic arm's grasping trajectory sequence and the AGV's conveying path map are calculated to generate an initial path planning scheme. The initial path planning scheme is input into the instruction compilation engine, which encodes the robotic arm joint angle, AGV speed curve and collaborative timing based on the preset industrial protocol standard, generates an executable sorting instruction set, and embeds the equipment status self-check verification code. The sorting instruction set is transmitted to the execution unit through the industrial communication network. The unit receives instruction confirmation signals from the robotic arm and AGV in real time and dynamically adjusts the instruction timing offset according to the network delay to complete the issuance of sorting instructions.

[0011] Optionally, based on the sorting instruction set, the robotic arm is controlled to complete the gripping and placement of PCB boards. Simultaneously, the sorting results are inspected in real time by a vision verification module. If an abnormal board is detected, a rework signal is triggered, and a sorting completion status report and quality verification data are output. Specifically, the following steps are included: The robot arm gripping coordinate sequence and AGV target placement coordinates in the sorting instruction set are analyzed. The robot arm end effector is driven to move to the designated position through the motion control module. The PCB board is gripped non-destructively based on the composite gripping mechanism of electromagnetic chuck and flexible gripper, and a gripping action log is generated. The system collects joint torque data of the robotic arm and pressure feedback signals from the end effector in real time, and dynamically adjusts the gripping force in combination with preset mechanical tolerance thresholds. If gripping offset or abnormal pressure fluctuations are detected, the gripping retry mechanism is triggered and the gripping action log is updated.

[0012] Optionally, if abnormal fluctuations in the grabbing offset or pressure are detected, the grabbing retry mechanism is triggered and the grabbing action log is updated, which further includes: The AGV collaborative scheduling module transports the captured PCB board to the target storage location, and simultaneously activates the vision verification module deployed on the AGV vehicle to scan the surface integrity and placement posture of the PCB board in real time, generating the original verification image data stream. Anomaly analysis is performed on the original verification image data stream based on a dynamic adaptive detection algorithm, including: identifying scratches, warpage and pad oxidation defects through a defect detection model optimized by transfer learning, comparing the process parameter thresholds in the standardized feature dataset, and outputting anomaly judgment results and defect type labels. If the anomaly determination result is marked as a defective board, a rework instruction is sent to the MES system through the rework signal generation module, triggering the AGV to automatically return to the sorting start station.

[0013] Optionally, the synchronization involves real-time quality inspection of the sorting results via a visual verification module. If an abnormal board is detected, a rework signal is triggered, and a sorting completion status report and quality verification data are output. This process also includes: The sorting completion status report and quality verification data are sent back to the MES system. The data analysis module statistically analyzes sorting efficiency and error types, dynamically updates the parameter library of the lightweight classification model and the optimization weights of the path planning algorithm, and forms a closed-loop iterative mechanism.

[0014] This invention also provides a PCB intelligent sorting system based on an unmanned factory, which uses the PCB intelligent sorting process based on an unmanned factory as described above to achieve sorting. The PCB intelligent sorting system specifically includes: The sorting equipment is equipped with an industrial camera and a scanning module to acquire surface image data and internal structural parameters of PCB boards in real time. The robotic arm, equipped with an end effector consisting of an electromagnetic chuck and a flexible gripper, is used to support non-destructive gripping of PCBs of various thicknesses; wherein, the robotic arm integrates a joint torque sensor and an end pressure sensor; The AGV conveying assembly, equipped with a vision verification module and positioning device, enables directional conveying of sheet metal. Industrial communication networks are used to transmit sorting instruction sets; The control center integrates a lightweight classification model, a dynamic path planning algorithm, and a multi-objective optimization engine to drive the collaborative operation of the sorting system. The closed-loop management module enables the feedback of sorting data and iterative updates of process parameters through the MES system interface.

[0015] Compared with existing technologies, this invention has the following advantages: First, the image acquisition module and scanning module on the sorting equipment acquire surface images and internal structural parameters of PCB boards in real time, extract model identifiers, pad distribution features, and process parameters, and generate a standardized feature dataset. Then, a lightweight classification model, combined with process grade determination rules, classifies and grades PCB boards, and dynamically allocates sorting priorities based on MES system order requirements, generating a task queue. A path planning algorithm calculates the optimal gripping path for the robotic arm and the AGV conveying route based on the priority queue, forming a sorting instruction set and sending it to the execution unit via the industrial communication network. The robotic arm completes PCB gripping and placement operations according to the instructions, while a vision verification module monitors the sorting results in real time. Abnormal boards trigger rework signals, and a sorting status report and quality data are output. This process, through automated data acquisition, intelligent classification decision-making, and a closed-loop optimization mechanism, achieves a deep integration of high efficiency, accuracy, and unmanned adaptation in PCB sorting. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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] The structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0018] Figure 1 This is one of the flowcharts illustrating the PCB intelligent sorting process based on an unmanned factory in this embodiment. Figure 2 This is the second flowchart illustrating the PCB intelligent sorting process based on an unmanned factory in this embodiment one. Figure 3 This is a layout diagram of the PCB intelligent sorting system in Embodiment 2. Detailed Implementation

[0019] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0020] In the description of this invention, it should be understood that the terms "upper," "lower," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. It should be noted that when a component is considered to be "connected" to another component, it can be directly connected to the other component or there may be a component positioned centrally in the connection.

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0022] Example 1: Combination Figure 1 and Figure 2 As shown, this embodiment of the invention provides a PCB intelligent sorting process based on an unmanned factory, including the following steps: S1, through the image acquisition module and scanning module deployed on the sorting equipment, acquires surface image data and internal structural parameters of PCB board in real time, and extracts the model identification, pad distribution characteristics and process parameters of PCB board to generate a standardized feature dataset; By deploying a multispectral industrial camera array and an infrared scanning module, multimodal data acquisition of PCB board surface images and internal structural parameters is achieved. The image acquisition module captures surface features (such as pad distribution and character markings), while the infrared scanning module penetrates the surface to obtain internal parameters such as copper layer thickness and aperture distribution. Combined with a timestamp alignment mechanism, spatiotemporal consistency of the data is ensured.

[0023] S2, input the feature dataset into the lightweight classification model, combine the preset process grade judgment rules, classify the PCB board by model and quality grade, and dynamically allocate sorting priority based on the order requirements of the MES system to generate a sorting task priority queue. A lightweight classification model (such as the improved YOLOv7-PCB network) is used to classify the feature dataset by model and quality level. The quality level is determined by combining the preset process rule base (such as pad uniformity threshold and copper layer thickness tolerance). At the same time, parameters such as order urgency coefficient and warehouse location status are obtained in real time through the MES system, and a multi-objective optimization function is constructed to dynamically allocate sorting priority.

[0024] S3, based on the priority queue of sorting tasks, calculates the optimal grasping path of the robotic arm and the AGV conveying route through the path planning algorithm, generates a sorting instruction set, and sends it to the execution unit through the industrial communication network; A hybrid path planning algorithm (such as a fusion of RRT and Dijkstra's algorithm) is employed to calculate the robotic arm's grasping trajectory and the AGV's conveying path, aiming to minimize the total sorting time and energy consumption, and generating a spatiotemporally coordinated sorting instruction set. An industrial communication network (TSN protocol) ensures low-latency instruction transmission, and CRC checksums guarantee data integrity.

[0025] S4, based on the sorting instruction set, controls the robotic arm to complete the picking and placing of PCB boards. Simultaneously, the visual verification module performs real-time quality inspection on the sorting results. If an abnormal board is detected, a rework signal is triggered, and a sorting completion status report and quality verification data are output.

[0026] The robotic arm performs grasping operations based on sorting instructions, and the end effector uses a composite mechanism of electromagnetic chuck and flexible gripper to adapt to PCB boards of different thicknesses. Simultaneously, the multispectral vision module on the AGV carrier is activated, and the defect detection model optimized through transfer learning identifies anomalies such as scratches and warping in real time, triggering rework signals and associating them with unique identifiers to trace the source of defects.

[0027] S5 sends the sorting completion status report and quality verification data back to the MES system. The data analysis module then statistically analyzes sorting efficiency and error types, dynamically updates the parameter library of the lightweight classification model and the optimization weights of the path planning algorithm, forming a closed-loop iterative mechanism.

[0028] The sorting completion status report and quality verification data are sent back to the MES system. The data analysis module statistically analyzes sorting efficiency and error type distribution, and dynamically updates the classification model parameter library (such as defect feature weights) and path planning algorithm optimization rules (such as obstacle avoidance priority adjustment). This is achieved through a data-driven parameter iteration mechanism (optimizing the model based on historical sorting data) and edge-cloud collaborative computing (achieving low-latency feedback). This forms a closed-loop process, continuously improving the system's adaptability and solving the problem of traditional sorting processes being unable to be dynamically optimized.

[0029] The working principle of this invention is as follows: First, the image acquisition module and scanning module on the sorting equipment acquire surface images and internal structural parameters of PCB boards in real time, extract model identifiers, pad distribution characteristics, and process parameters, and generate a standardized feature dataset. Then, a lightweight classification model, combined with process grade judgment rules, classifies and grades the PCB boards, and dynamically allocates sorting priorities based on order requirements from the MES system, generating a task queue. The path planning algorithm calculates the optimal gripping path of the robotic arm and the AGV conveying route based on the priority queue, forming a sorting instruction set and sending it to the execution unit through the industrial communication network. The robotic arm completes the PCB gripping and placement operations according to the instructions, while the vision verification module detects the sorting results in real time. Abnormal boards trigger a rework signal, and the system outputs a sorting status report and quality data. This process achieves a deep integration of high efficiency, accuracy, and unmanned adaptation in PCB sorting through automated data acquisition, intelligent classification decision-making, and a closed-loop optimization mechanism.

[0030] In this embodiment, step S1 specifically includes: S11 uses industrial cameras and scanning modules deployed on the sorting equipment to perform multi-angle image acquisition and internal structure scanning of PCB boards on the conveyor belt, simultaneously acquiring surface image data and internal structure parameters, and generating raw data packets based on a timestamp alignment mechanism; internal structure parameters include internal copper layer thickness distribution data and aperture characteristic parameters. Industrial cameras and scanning modules on the sorting equipment are used to acquire multi-angle images (including surface texture, character markings, etc.) and scan the internal structure (such as copper layer thickness and aperture distribution) of PCB boards on the conveyor belt. The industrial cameras use multispectral imaging technology to capture visible light and near-infrared data, while the infrared scanning module acquires internal parameters based on the principle of penetration detection. A timestamp alignment mechanism (ensuring spatiotemporal synchronization between surface images and internal scan data) generates a standardized data package containing the original surface images and internal parameters, providing a complete data source for subsequent processing and solving the problem of insufficient feature dimensions caused by traditional single data acquisition.

[0031] S12, preprocessing the raw data packet, specifically including: performing adaptive denoising and edge enhancement operations on the surface image data to eliminate image distortion caused by uneven lighting and mechanical vibration; using a dynamic filtering algorithm to remove environmental interference signals from the internal scan data, and matching the reference value through a preset PCB process parameter library to generate denoised standard data units; The original data packets are processed in different modes: the surface image is de-blurred by an adaptive denoising algorithm (such as wavelet transform) to eliminate the blurring caused by uneven lighting or mechanical vibration, and the pad outline is enhanced by edge enhancement technology; the internal scan data is de-blurred by a dynamic filtering algorithm (such as Kalman filtering) to remove environmental electromagnetic interference signals, and the reference value is matched based on a preset PCB process parameter library (such as copper layer thickness tolerance ±0.05mm) to correct the data offset.

[0032] S13, based on the improved YOLOv7-PCB model, performs feature extraction on standard data units, specifically including: identifying model identification characters, positioning pad coordinates and spacing distribution from surface image data, and extracting the average copper layer thickness, aperture distribution density and microcrack defect parameters from internal scan data to form a multi-dimensional feature vector.

[0033] Based on the improved YOLOv7-PCB model (adding a PCB process feature layer to the original YOLOv7), feature extraction is performed on standard data cells. The surface image is used to identify model identification characters through the OCR module, and to locate the pad coordinates and spacing distribution. The internal scan data is used to extract the mean copper layer thickness and aperture distribution density through regression analysis, and to detect microcrack defects.

[0034] S14. Input the multidimensional feature vector into the dynamic template generation module. According to the preset standardization rules (including data format, unit unification and normalization processing), convert the heterogeneous feature parameters into structured data of uniform dimension and add unique identifiers and timestamp labels. The multidimensional feature vectors are input into the dynamic template generation module, which standardizes the heterogeneous data according to preset rules (such as normalization). For example, the copper layer thickness unit is unified to micrometers, the pad spacing is normalized to a percentage value, and unique identifiers (such as UUIDs) and timestamp labels are added.

[0035] S15 uses a data verification engine to perform integrity checks and logical verifications on the standardized structured data, removes outliers, generates a standardized feature dataset containing complete metadata, and stores it in the database of the edge computing node.

[0036] The data validation engine performs integrity checks (such as field missing checks) and logical verifications (such as whether the copper layer thickness exceeds the process library threshold) on the standardized structured data, removes outliers, and triggers a re-sampling process. Data that passes the validation is tagged with metadata (such as sorting batch number) and stored in the distributed database of the edge computing node.

[0037] In this embodiment, step S2 specifically includes the following steps: S21. Input the feature dataset into the lightweight classification model, and use the pre-trained PCB model feature matching network to perform semantic parsing on the model identification characters of the PCB board, generate an initial model classification label set, and associate it with the corresponding process parameter threshold range. A pre-trained PCB model feature matching network performs semantic parsing (e.g., OCR recognition technology) on model identifier characters in the feature dataset, generating an initial model classification label set. The classification results are then dynamically associated with preset process parameter threshold ranges (e.g., copper layer thickness tolerance, aperture allowable deviation). This involves a pre-trained feature matching network (based on Transformer architecture to optimize character recognition) and a dynamic process parameter binding mechanism (associating model numbers with process library parameters).

[0038] S22, based on the preset process grade judgment rule library, classify each PCB board in the initial model classification label set into quality grades, including: calculating quality scores and mapping them to preset process grade ranges based on pad distribution uniformity, copper layer thickness deviation value and microcrack defect density, and generating enhanced classification results with quality grade labels; Based on a process grade determination rule library, the uniformity of pad distribution (calculated using the coefficient of variation), copper layer thickness deviation (compared to the baseline value in the process library), and microcrack defect density (number of defects per unit area) of each PCB component are scored from multiple dimensions and mapped to a preset grade range (e.g., A / B / C). This involves a multi-dimensional scoring algorithm (weighted fusion of different parameters) and grade mapping rules (preset threshold ranges). Enhanced classification results with quality grade identifiers are generated, providing quality dimension input for priority allocation and overcoming the limitations of traditional single-threshold determination.

[0039] S23: Real-time acquisition of order demand data from the MES system, extraction of order urgency coefficient, warehouse location status and customer-customized parameters, construction of multi-objective optimization function, dynamic calculation of sorting priority weight for each PCB board, and generation of priority weight matrix. The system acquires real-time order urgency coefficients (e.g., remaining delivery time), warehouse location status (e.g., target location capacity), and customer-customized parameters (e.g., special process requirements) from the MES system. It then constructs a multi-objective optimization function (objectives include minimizing delivery delay and maximizing warehouse location utilization) and dynamically calculates sorting priority weights. This includes a multi-objective optimization algorithm (e.g., NSGA-II genetic algorithm) and a dynamic weight adjustment mechanism (responding to order changes in real time). A priority weight matrix is ​​generated to achieve flexible scheduling of sorting tasks, resolving the rigidity of resource allocation caused by traditional static rules.

[0040] S24. The enhanced classification results are associated and matched with the priority weight matrix. A sorting task priority queue is generated through a dynamic weighted sorting algorithm and then collected into a task instruction package and transmitted to the path planning module.

[0041] The enhanced classification results are spatiotemporally correlated and matched with the priority weight matrix (e.g., aligned by sorting station number and time window). A sorting task priority queue is generated using a dynamic weighted sorting algorithm (based on Pareto front optimization), and then encapsulated into a task instruction package (including target coordinates and process level constraints). This package includes a dynamic weighted sorting algorithm (balancing quality and timeliness) and an instruction encapsulation protocol (e.g., JSON structured data). The output task instruction package directly drives the path planning module, ensuring accurate issuance of sorting instructions.

[0042] In this embodiment, step S3 specifically includes the following steps: S31, based on the sorting task priority queue, extracts the target storage location, process level constraints and time window requirements of PCB boards through the task parsing module, and generates a sorting task parameter matrix containing spatiotemporal constraints. This step uses the task parsing module to extract the target storage location (e.g., warehouse coordinates), process level constraints (e.g., anti-static requirements), and time window requirements (e.g., urgent orders need to be sorted within a specified time period) of the PCB board from the sorting task priority queue, constructing a sorting task parameter matrix that includes spatiotemporal constraints. This multi-dimensional task parameter matrix provides precise input for path planning, improving its adaptability and efficiency.

[0043] S32 employs a hybrid path planning algorithm, combining the kinematic model of the robotic arm with the dynamic obstacle avoidance rules of the AGV, with the optimization objective of minimizing the total sorting time and energy consumption. It calculates the grasping trajectory sequence of the robotic arm and the conveying path map of the AGV to generate an initial path planning scheme. A hybrid path planning algorithm is used to calculate the robotic arm grasping trajectory sequence and AGV conveying path map. Combining the robotic arm kinematic model (joint angular velocity limit) with AGV dynamic obstacle avoidance rules (such as real-time detection by LiDAR), the optimization goal is to minimize the total sorting time and energy consumption. This achieves spatiotemporal coordination of equipment movements, avoids the risk of equipment collision caused by fixed paths, and reduces energy consumption costs.

[0044] S33 inputs the initial path planning scheme into the instruction compilation engine, encodes the robot arm joint angle, AGV speed curve and collaborative timing based on the preset industrial protocol standard, generates an executable sorting instruction set, and embeds the equipment status self-check verification code. The initial path planning scheme is input into the instruction compilation engine, which encodes the robot arm joint angles (unit: radians), AGV speed curves (such as Bezier curve control points), and collaborative timing into an executable instruction set based on industrial protocol standards (such as OPC UA), and embeds the device status self-check verification code (such as CRC check code).

[0045] S34 transmits the sorting instruction set to the execution unit through the industrial communication network, receives instruction confirmation signals from the robotic arm and AGV in real time, and dynamically adjusts the instruction timing offset according to the network delay to complete the issuance of sorting instructions.

[0046] The sorting instruction set is broadcast to the execution unit via an industrial communication network, and confirmation signals from the robotic arm and AGV are received in real time. If network latency (e.g., >5ms) is detected, the instruction timing offset is dynamically adjusted to ensure precise synchronization of the actions of multiple devices.

[0047] In this embodiment, step S4 specifically includes the following steps: S41, parse the robotic arm grasping coordinate sequence and AGV target placement coordinate in the sorting instruction set, drive the robotic arm end effector to move to the specified position through the motion control module, and complete the non-destructive grasping of PCB board based on the composite grasping mechanism of electromagnetic chuck and flexible gripper, and generate grasping action log. By analyzing the robotic arm's grasping coordinate sequence and the AGV's target placement coordinates from the sorting instruction set, the motion control module positions the robotic arm's end effector to the target location. A composite grasping mechanism combining an electromagnetic chuck and a flexible gripper completes the non-destructive grasping, generating a grasping action log containing timestamps, grasping coordinates, and force feedback data. This achieves precise positioning and flexible grasping, solving the problem of plate damage caused by traditional rigid grippers and providing a reliable grasping foundation for subsequent processes.

[0048] S42 collects real-time joint torque data of the robotic arm and pressure feedback signals of the end effector, and dynamically adjusts the gripping force in combination with the preset mechanical tolerance threshold. If gripping offset or abnormal pressure fluctuation is detected, the gripping retry mechanism is triggered and the gripping action log is updated.

[0049] The system collects real-time joint torque data of the robotic arm (reflecting the load status) and pressure signals of the end effector (monitoring adsorption stability), and dynamically adjusts the gripping force based on preset mechanical tolerance thresholds (such as pressure fluctuation range ±5N). If gripping offset or pressure abnormality is detected, a gripping retry mechanism is triggered and the log is updated.

[0050] S43, through the AGV collaborative scheduling module, transports the grabbed PCB board to the target storage location, and simultaneously activates the vision verification module deployed on the AGV vehicle to scan the surface integrity and placement posture of the PCB board in real time, generating the original verification image data stream; The AGV collaborative scheduling module transports PCB boards according to the coordinates of the target storage location, and simultaneously activates the multispectral vision verification module (visible light + near-infrared band) mounted on the AGV to perform real-time scanning of the PCB surface integrity (scratches, stains) and placement posture, generating a high frame rate verification image data stream. This enables parallel conveying and quality inspection, solving the rework delay problem caused by post-sorting quality inspection.

[0051] S44, based on the dynamic adaptive detection algorithm, performs anomaly analysis on the original verification image data stream, including: identifying scratches, warpage and pad oxidation defects through a defect detection model optimized by transfer learning, comparing with the process parameter thresholds in the standardized feature dataset, and outputting anomaly judgment results and defect type labels; A defect detection model optimized based on transfer learning (with a pre-trained dataset containing multiple PCB defect samples) analyzes the verification image data stream to identify scratches, warpage, and pad oxidation defects, and compares them with process parameter thresholds in the standardized feature dataset (e.g., pad oxidation area ≤ 0.1 mm). 2 It outputs anomaly detection results and defect type labels, providing a quantitative basis for rework decisions and improving the defect detection rate.

[0052] S45. If the abnormality judgment result is marked as a defective board, a rework instruction is sent to the MES system through the rework signal generation module, triggering the AGV to automatically return to the sorting start station.

[0053] If the anomaly assessment indicates a defective board, the rework signal generation module sends a rework instruction (including a unique identifier, defect type, and capture log) to the MES system. This triggers the AGV to automatically plan a return path to the sorting start station, with the rework path dynamically planned to avoid the current sorting path. This achieves closed-loop processing of defective boards, avoids efficiency losses caused by manual intervention, and ensures the complete unmanned sorting process.

[0054] Example 2: Combination Figure 3 As shown, the present invention also provides a PCB intelligent sorting system based on an unmanned factory, which adopts the PCB intelligent sorting process based on an unmanned factory as described in Embodiment 1 to achieve sorting. The PCB intelligent sorting system specifically includes: The sorting equipment 10 is equipped with an industrial camera 11 and a scanning module 12, which are used to acquire surface image data and internal structural parameters of PCB boards in real time.

[0055] The robotic arm 20 is equipped with an end effector consisting of an electromagnetic chuck and a flexible gripper 21, which is used to support non-destructive gripping of PCBs of various thicknesses; the robotic arm integrates a joint torque sensor and an end pressure sensor.

[0056] The AGV conveying assembly 30 is equipped with a vision verification module and a positioning device 31 to realize the directional conveying of plates; the industrial communication network 40 is used to transmit the sorting instruction set.

[0057] The control center 50 integrates a lightweight classification model, a dynamic path planning algorithm, and a multi-objective optimization engine to drive the collaborative operation of the sorting system.

[0058] The closed-loop management module enables the transmission of sorting data and iterative updates of process parameters through the MES system interface.

[0059] The above-described 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.

Claims

1. A PCB intelligent sorting process based on an unmanned factory, characterized in that, Includes the following steps: By deploying image acquisition and scanning modules on sorting equipment, surface image data and internal structural parameters of PCB boards are acquired in real time, and the model identification, pad distribution characteristics and process parameters of PCB boards are extracted to generate a standardized feature dataset. The feature dataset is input into a lightweight classification model, and PCB boards are classified by model and quality level according to preset process grade determination rules. Simultaneously, sorting priorities are dynamically allocated based on order requirements from the MES system, generating a sorting task priority queue, specifically including: The feature dataset is input into a lightweight classification model. The model identification characters of the PCB board are semantically parsed through a pre-trained PCB model feature matching network to generate an initial model classification label set and associate it with the corresponding process parameter threshold range. Based on the preset process grade judgment rule library, each PCB board in the initial model classification label set is classified into quality grades, including: calculating the quality score and mapping it to the preset process grade range according to the uniformity of pad distribution, copper layer thickness deviation value and microcrack defect density, and generating an enhanced classification result with quality grade label; The system acquires order demand data from the MES system in real time, extracts order urgency coefficients, warehouse location status, and customer-customized parameters, constructs a multi-objective optimization function, dynamically calculates the sorting priority weight of each PCB board, and generates a priority weight matrix. The enhanced classification results are associated and matched with the priority weight matrix, and a sorting task priority queue is generated by a dynamic weighted sorting algorithm. The queue is then compiled into a task instruction package and transmitted to the path planning module. Based on the sorting task priority queue, the optimal grasping path of the robotic arm and the AGV conveying route are calculated using a path planning algorithm, a sorting instruction set is generated, and then sent to the execution unit via the industrial communication network, specifically including: Based on the sorting task priority queue, the target storage location, process level constraints and time window requirements of the PCB board are extracted by the task parsing module to generate a sorting task parameter matrix containing spatiotemporal constraints. A hybrid path planning algorithm is adopted, which combines the kinematic model of the robotic arm with the dynamic obstacle avoidance rules of the AGV. With the optimization objective of minimizing the total sorting time and energy consumption, the robotic arm's grasping trajectory sequence and the AGV's conveying path map are calculated to generate an initial path planning scheme. The initial path planning scheme is input into the instruction compilation engine, which encodes the robotic arm joint angle, AGV speed curve and collaborative timing based on the preset industrial protocol standard, generates an executable sorting instruction set, and embeds the equipment status self-check verification code. The sorting instruction set is transmitted to the execution unit through the industrial communication network, and the instruction confirmation signals from the robotic arm and AGV are received in real time. The instruction timing offset is dynamically adjusted according to the network delay to complete the issuance of sorting instructions. Based on the sorting instruction set, the robotic arm is controlled to complete the gripping and placement of PCB boards. Simultaneously, a vision verification module performs real-time quality checks on the sorting results. If an abnormal board is detected, a rework signal is triggered, and a sorting completion status report and quality verification data are output. Specifically, the process includes the following steps: The robot arm gripping coordinate sequence and AGV target placement coordinates in the sorting instruction set are analyzed. The robot arm end effector is driven to move to the designated position through the motion control module. The PCB board is gripped non-destructively based on the composite gripping mechanism of electromagnetic chuck and flexible gripper, and a gripping action log is generated. The system collects joint torque data of the robotic arm and pressure feedback signals from the end effector in real time, and dynamically adjusts the gripping force based on a preset mechanical tolerance threshold. If a gripping offset or abnormal pressure fluctuation is detected, a gripping retry mechanism is triggered and the gripping action log is updated. The gripping action log includes timestamps, gripping coordinates, and force feedback data. The AGV collaborative scheduling module transports the captured PCB board to the target storage location, and simultaneously activates the vision verification module deployed on the AGV vehicle to scan the surface integrity and placement posture of the PCB board in real time, generating the original verification image data stream. Anomaly analysis is performed on the original verification image data stream based on a dynamic adaptive detection algorithm, including: identifying scratches, warpage and pad oxidation defects through a defect detection model optimized by transfer learning, comparing the process parameter thresholds in the standardized feature dataset, and outputting anomaly judgment results and defect type labels. If the anomaly determination result is a defective board, the rework signal generation module sends a rework instruction to the MES system. The rework instruction includes a unique identifier, defect type, and the grabbing action log, triggering the AGV to automatically plan a return path to the sorting start station. The rework path is dynamically planned to avoid the current sorting path.

2. The PCB intelligent sorting process based on an unmanned factory according to claim 1, characterized in that, The image acquisition and scanning modules deployed on the sorting equipment acquire surface image data and internal structural parameters of the PCB board in real time, and extract the model identification, pad distribution characteristics, and process parameters of the PCB board to generate a standardized feature dataset, specifically including: The industrial camera and scanning module deployed on the sorting equipment are used to acquire multi-angle images and scan the internal structure of the PCB board on the conveyor belt, simultaneously obtaining surface image data and internal structure parameters, and generating the original data packet based on the timestamp alignment mechanism; the internal structure parameters include the internal copper layer thickness distribution data and aperture characteristic parameters. The preprocessing of the original data packet specifically includes: performing adaptive denoising and edge enhancement operations on the surface image data to eliminate image distortion caused by uneven lighting and mechanical vibration; using a dynamic filtering algorithm to remove environmental interference signals from the internal structural parameters, and matching the reference value through a preset PCB process parameter library to generate denoised standard data units. Feature extraction of the standard data unit specifically includes: identifying model identification characters, positioning pad coordinates and spacing distribution from the surface image data, and extracting the average copper layer thickness, aperture distribution density and microcrack defect parameters from the internal scan data to form a multidimensional feature vector.

3. The PCB intelligent sorting process based on an unmanned factory according to claim 2, characterized in that, Feature extraction is performed on the standard data unit, followed by: The multidimensional feature vector is input into the dynamic template generation module, and the heterogeneous feature parameters are converted into structured data of a unified dimension according to the preset standardization rules, and a unique identifier and timestamp label are added. The data validation engine performs integrity checks and logical verification on the standardized structured data, removes outliers, generates a standardized feature dataset containing complete metadata, and stores it in the database of the edge computing node.

4. The PCB intelligent sorting process based on an unmanned factory according to claim 1, characterized in that, The synchronous process uses a visual verification module to perform real-time quality checks on the sorting results. If an abnormal board is detected, a rework signal is triggered, and a sorting completion status report and quality verification data are output. This process also includes: The sorting completion status report and quality verification data are sent back to the MES system. The data analysis module statistically analyzes sorting efficiency and error types, dynamically updates the parameter library of the lightweight classification model and the optimization weights of the path planning algorithm, and forms a closed-loop iterative mechanism.

5. A PCB intelligent sorting system based on an unmanned factory, characterized in that, Sorting is achieved using the PCB intelligent sorting process based on an unmanned factory as described in any one of claims 1 to 4, wherein the PCB intelligent sorting system specifically includes: The sorting equipment is equipped with an industrial camera and a scanning module to acquire surface image data and internal structural parameters of PCB boards in real time. The robotic arm, equipped with an end effector consisting of an electromagnetic chuck and a flexible gripper, is used to support non-destructive gripping of PCBs of various thicknesses; wherein, the robotic arm integrates a joint torque sensor and an end pressure sensor; The AGV conveying assembly, equipped with a vision verification module and positioning device, enables directional conveying of sheet metal. Industrial communication networks are used to transmit sorting instruction sets; The control center integrates a lightweight classification model, a dynamic path planning algorithm, and a multi-objective optimization engine to drive the collaborative operation of the sorting system. The closed-loop management module enables the transmission of sorting data and iterative updates of process parameters through the MES system interface.

Images