Defect sorting system applied to PCB production

Abstract

The invention relates to the technical field of industrial production, in particular to a defect sorting system applied to PCB production, which comprises a picture acquisition module, a picture processing and analyzing module, a data cloud uploading module, a defect sorting module and an interaction module, and is characterized in that the picture acquisition module acquires PCB pictures and PCB numbers on a conveyor belt; the picture processing and analyzing module is used for improving the network architecture to generate an ONNX model, converting the ONNX model into an OM model, inputting the PCB picture into the OM model, returning the defect type of the current PCB and marking the defect PCB picture; the data cloud uploading module uploads the defective PCB pictures to a platform for collection; the defect sorting module is used for detecting the number of the PCBs, numbering the PCBs and sorting the PCBs with defects out of the conveying belt; the equipment cost and the operation cost of the system are far lower than those of a high-precision X-ray detection system, and compared with a traditional AOI optical detection technology, the hardware cost and the operation and maintenance cost are more controllable, and high equipment updating and calibration cost is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial production, and in particular to a defect sorting system used in PCB production. Background Art

[0002] PCB refers to a circuit board with copper circuit patterns formed on a copper-clad laminate according to a predetermined design. Its main function is to connect various electronic components according to a predetermined circuit and act as an electrical connection.

[0003] However, due to technical and human constraints during the PCB production process, defects such as open circuits, short circuits, gaps, and burrs are unavoidable. These defects not only affect the performance of the PCB but can also pose a serious threat to the safety of electronic products.

[0004] Currently, companies in the market primarily use AOI and X-ray inspection technologies for inspection. However, these two technologies have been slow to achieve breakthroughs due to their high costs and stringent inspection requirements. Furthermore, manual visual inspection is limited in its progress by its high labor costs. Summary of the Invention

[0005] The purpose of the present invention is to provide a defect sorting system for PCB production, aiming to solve the problem of high defect detection cost in existing PCB production.

[0006] To achieve the above-mentioned object, the present invention provides a defect sorting system for PCB production, comprising an image acquisition module, an image processing and analysis module, a data cloud upload module, a defect sorting module, and an interaction module;

[0007] The image acquisition module is used to acquire the PCB image and PCB number on the conveyor belt;

[0008] The image processing and analysis module generates an ONNX model based on the improved network architecture of the YOLOv10 network, converts it into an OM model, inputs the PCB image into the OM model, returns the defect type of the current PCB and marks the defective PCB image;

[0009] The data cloud upload module is used to upload the defective PCB images to the Huawei Cloud IOT platform and collect them;

[0010] The defect sorting module detects the number of PCBs passing through based on the infrared detection mechanism and numbers the PCBs. When the number is the same as that of the defective PCB, the defective PCB is sorted out of the conveyor belt;

[0011] The interactive module is used to obtain the statistical data of all pipelines and the status data of each pipeline, and to display the complete information of the pipeline.

[0012] Wherein, the image acquisition module includes an infrared detection unit and a USB camera module.

[0013] Among them, the image processing and analysis module is based on the Ascend AIPRO development board, and the data cloud upload module is based on the Hi3861v100 chip.

[0014] Wherein, the defect sorting module includes a digital servo and an infrared detection module.

[0015] Among them, the interactive module includes a data preview unit, a pipeline management unit and a detailed pipeline information unit. The data preview unit consists of a List list and a chart icon, the pipeline management unit consists of a List list and a card card, and the detailed pipeline information unit consists of a PCB number and PCB image data.

[0016] The present invention relates to a defect sorting system for PCB production. The image acquisition module acquires PCB images and PCB numbers on a conveyor belt. The image processing and analysis module generates an ONNX model based on a YOLOv10 network with an improved network architecture, converts the model into an OM model, inputs the PCB image into the OM model, returns the defect type of the current PCB, and marks the defective PCB image. The data cloud upload module uploads the defective PCB image to the Huawei Cloud IOT platform and collects the data. The defect sorting module detects the number of PCBs that have passed through the system based on an infrared detection mechanism and numbers the PCBs. When the number is the same as that of a defective PCB, the defective PCB is sorted out of the conveyor belt. The interaction module acquires statistical data of all assembly lines and status data of each assembly line, and displays complete information under the assembly line. The system has a wide range of detection types and can accurately identify and locate various common defects on PCBs, including but not limited to leaks, open circuits, gaps, short circuits, burrs, excess copper, etc. By promptly discovering and eliminating these potential problems, the flow of defective PCBs into the market can be effectively reduced, mitigating the risks posed by defective products. Its accuracy rivals that of traditional AOI optical inspection and X-ray inspection technologies, enabling efficient and accurate identification of minute defects on PCBs, ensuring product quality meets stringent standards. Its inspection accuracy is not only on par with existing technologies, but can even provide superior inspection performance in certain specific application scenarios, significantly satisfying the electronics manufacturing industry's demand for high reliability and high precision. The system's equipment and operating costs are far lower than those of high-precision X-ray inspection systems. Compared to traditional AOI optical inspection technology, hardware costs and operational expenses are more manageable, avoiding high equipment upgrade and calibration costs and resolving the issue of high defect detection costs in existing PCB production. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a structural diagram of a defect sorting system for PCB production provided by the present invention.

[0019] Figure 2 This is the YOLOv10 network structure diagram.

[0020] Figure 3 This is the improved YOLOv10 network structure diagram.

[0021] Figure 4 This is a diagram of the PSA network structure.

[0022] Figure 5 This is the EMA network structure diagram.

[0023] Figure 6 This is the ATC tool conversion process diagram.

[0024] Figure 7 This is a UART communication diagram.

[0025] Figure 8 This is a schematic diagram of the PID algorithm.

[0026] Figure 9 It is a schematic diagram of the Stage model.

[0027] Figure 10 It is a schematic diagram of the FA model.

[0028] Figure 11 The figure is a schematic structural diagram of a defect sorting system for PCB production provided by the present invention.

[0029] In the figure: 1-image acquisition module, 2-image processing and analysis module, 3-data cloud upload module, 4-defect sorting module, 5-interaction module. DETAILED DESCRIPTION

[0030] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0031] See also Figures 1 to 11 , the present invention provides a defect sorting system applied to PCB production, comprising an image acquisition module 1, an image processing and analysis module 2, a data cloud upload module 3, a defect sorting module 4 and an interaction module 5;

[0032] The image acquisition module 1 is used to acquire the PCB image and PCB number on the conveyor belt;

[0033] The image processing and analysis module 2 generates an ONNX model based on the improved network architecture of the YOLOv10 network, converts it into an OM model, inputs the PCB image into the OM model, returns the defect type of the current PCB and marks the defective PCB image;

[0034] The data cloud upload module 3 is used to upload the defective PCB image to the Huawei Cloud IOT platform and collect it;

[0035] The defect sorting module 4 detects the number of PCBs passing through based on the infrared detection mechanism and numbers the PCBs. When the number is the same as that of the defective PCB, the defective PCB is sorted out of the conveyor belt;

[0036] The interactive module 5 is used to obtain the statistical data of all pipelines and the status data of each pipeline, and to display the complete information of the pipeline.

[0037] In an embodiment of the present invention, the image acquisition module 1 acquires PCB images and PCB numbers on a conveyor belt; the image processing and analysis module 2 generates an ONNX model based on the improved network architecture of the YOLOv10 network, converts it into an OM model, inputs the PCB image into the OM model, returns the defect type of the current PCB, and marks the defective PCB image; the data cloud upload module 3 uploads the defective PCB image to the Huawei Cloud IOT platform and collects it; the defect sorting module 4 detects the number of PCBs passing through based on the infrared detection mechanism and numbers the PCBs. When the number is the same as that of a defective PCB, the defective PCB is sorted out of the conveyor belt; the interaction module 5 acquires statistical data of all assembly lines and status data of each assembly line, and displays complete information under the assembly line. The system has a wide range of detection types and can accurately identify and locate various common defects on PCBs, including but not limited to leaks, open circuits, gaps, short circuits, burrs, excess copper, etc. By promptly discovering and eliminating these potential problems, the flow of defective PCBs into the market can be effectively reduced, mitigating the risks posed by defective products. Its accuracy rivals that of traditional AOI optical inspection and X-ray inspection technologies, enabling efficient and accurate identification of minute defects on PCBs, ensuring product quality meets stringent standards. Its inspection accuracy is not only on par with existing technologies, but can even provide superior inspection performance in certain specific application scenarios, significantly satisfying the electronics manufacturing industry's demand for high reliability and high precision. The system's equipment and operating costs are far lower than those of high-precision X-ray inspection systems. Compared to traditional AOI optical inspection technology, hardware costs and operational expenses are more manageable, avoiding high equipment upgrade and calibration costs and resolving the issue of high defect detection costs in existing PCB production.

[0038] Furthermore, the image acquisition module 1 includes an infrared detection unit and a USB camera module.

[0039] In an embodiment of the present invention, the image acquisition module 1 detects the passage of PCBs within a certain range on the conveyor belt through an infrared detection unit, and obtains images of the PCBs within the current range through a USB camera module. The above two pieces of information can be collected in the Ascend AIPRO development board to number each PCB.

[0040] Furthermore, the image processing and analysis module 2 is carried by the Ascend AIPRO development board, and the data cloud upload module 3 is carried by the Hi3861v100 chip.

[0041] In an embodiment of the present invention, the PCB number will be uploaded to the Ascend AIPRO development board to store the information of each PCB and provide corresponding information for subsequent PCB sorting. The Hi3861v100 chip communicates with the Ascend AIPRO development board via UART and uses the MQTT protocol to upload the defective PCB number and defect type to the Huawei Cloud IOT platform for data storage. The Huawei Cloud IOT platform implements MQTT communication with BEARPI by configuring services, properties, data types, access methods, etc., thereby realizing cloud-based data upload. The YOLOv10 network effectively reduces the model size and improves detection accuracy by improving the PSA module to the EMA module. The Ascend Tensor Compiler can convert the ONNX model into an OM model that is more suitable for running on the Ascend AIPRO development board.

[0042] Furthermore, the interaction module 5 includes a data preview unit, a pipeline management unit and a detailed pipeline information unit. The data preview unit consists of a List list and a chart icon, the pipeline management unit consists of a List list and a card card, and the detailed pipeline information unit consists of a PCB number and PCB image data.

[0043] In an embodiment of the present invention, the data preview unit can obtain statistical data for all pipelines and display it through a chart icon. It can also view the number of pipelines, the number of verified PCBs, the number of defective PCBs, and so on. The pipeline management unit can obtain the status data of each pipeline, display the current status of each pipeline in real time, and view more detailed pipeline information by clicking on the pipeline card. The pipeline detailed information unit can display complete information about the pipeline, allowing users to view not only the current status of the pipeline, running time, and other information, but also the PCB inspection list, which contains the number of each PCB, PCB image, and PCB inspection results, and view specific PCB inspection results.

[0044] To better understand the present technical solution, the following examples are provided for further explanation:

[0045] Combine Figures 1 to 8 As shown, this embodiment is a defect sorting system used in PCB production, which is set in the PCB production process in the industrial production field to reduce the number of defective PCBs in the market. Figure 1As shown, the system includes four parts: improved YOLOv10 network, ATC framework conversion, IOT cloud transmission, and PCB defect sorting. These parts cooperate with each other to jointly realize the functions of the present invention.

[0046] Combine Figure 1 、 2 , 3, 4, and 5, this embodiment is an improved YOLOv10 network ( Figure 2 ), the present invention integrates the PSA module ( Figure 4 ) is improved to EMA module ( Figure 5 ), where the EMA module enhances the robustness and stability of feature extraction by introducing the exponential moving average mechanism. Compared with the original PSA module, the EMA module can more effectively capture subtle features when processing PCB defects, improve the accuracy of the model in defect classification and positioning, and generate an improved YOLOv10 network ( Figure 3 ).

[0047] Furthermore, based on the characteristics of PCB defects, this embodiment adjusts the input size of the YOLOv10 network and optimizes the loss function to make it more suitable for a variety of complex defect scenarios.

[0048] Furthermore, the model's recognition accuracy successfully increased from 97.6% to 99.6%, and the detection accuracy was further improved.

[0049] Furthermore, this embodiment constructs local cross-channel interactions in each parallel sub-network without performing channel dimensionality reduction, and fuses the output feature maps of the two parallel sub-networks through a cross-space learning method.

[0050] Combine Figure 1 、 6 As shown, this embodiment is the ATC framework conversion. Based on the improved YOLOv10, the ONNX model is trained. The running speed of this model on the Ascend AIPRO development board is relatively low. The present invention converts the ONNX model into the OM model through the Huawei Ascend tensor compiler. The conversion process is as follows Figure 6 shown.

[0051] Furthermore, an ONNX model refers to a deep learning model file saved in the ONNX format, usually with the ONNX suffix. ONNX is an open neural network exchange format used to promote model interoperability between different deep learning frameworks.

[0052] Furthermore, the OM model is a high-performance model format customized for the Ascend processor, suitable for offline execution of heterogeneous computing units such as NPU.

[0053] Furthermore, the Ascend Tensor Compiler performs the following operations during the model conversion process: operator scheduling optimization, weight data rearrangement, and memory usage optimization.

[0054] Combine Figure 1 、 7 As shown, this embodiment is IOT cloud transmission. After obtaining the detection results and PCB pictures, the present invention transmits the data to the Huawei Cloud IOT platform for data collection based on Huawei's self-developed Hongmeng architecture Hi3861v100 chip for subsequent data processing.

[0055] Furthermore, the detection data of the Hi3861v100 chip comes from the Ascend AIPRO development board, and the data is transmitted to the Hi3861v100 chip through UART communication. The specific communication method of UART communication is as follows: Figure 7 shown.

[0056] Furthermore, the Hi3861v100 chip can connect to the 2.4G network by connecting to the WIFI module, and can connect to the Huawei Cloud IoT platform through the network connection.

[0057] Furthermore, the Hi3861v100 chip connects to the Huawei Cloud IOT platform through the MQTT protocol. By configuring the service type and parameter type of the Huawei Cloud IOT platform, the defective PCB number, corresponding PCB defect type, and corresponding PCB image can be uploaded to the cloud.

[0058] Furthermore, the Huawei Cloud IoT platform provides comprehensive data storage, analysis, and intelligent processing services for this embodiment. Uploaded PCB inspection results and defect images are automatically archived and stored based on information such as equipment number, inspection time, and defect type. Distributed database technology is used to ensure efficient data management and fast retrieval.

[0059] Combine Figure 1 、 8 As shown, this embodiment is for PCB defect sorting. After obtaining the test results, the present invention uses this embodiment to sort whether the PCB contains defects. The sorting device consists of a set of digital servos and crossbars, wherein the digital servos are connected to the Ascend AIPRO development board and are controlled by the PID control algorithm ( Figure 8 ) Control the digital servo to make selections.

[0060] Furthermore, the PID control algorithm is the abbreviation of the proportional, integral, and differential control algorithm. It calculates the deviation between the current position of the servo and the target position to adjust the action of the servo in real time, thereby achieving accurate sorting. The algorithm is as follows Figure 8In this embodiment, the PID control algorithm adjusts the rotation angle of the digital servo in real time, enabling it to quickly respond to inspection results, accurately sorting defective PCBs into designated areas while ensuring that defect-free PCBs pass smoothly through the production line. To meet the requirements of high-speed production, the PID parameters have been repeatedly tuned to achieve rapid system response and stable control.

[0061] Combine Figure 9 、 Figure 10 As shown, this interactive module 5 adopts the Stage model ( Figure 9 ), Stage provides a component-based development mechanism, abstracting components into two categories: UIAbility and ExtensionAbility. Compared with the FA model ( Figure 10 ), the Stage model provides a more flexible development method, lower memory usage and a more standardized system management mechanism.

[0062] Furthermore, the interactive module 5 uses the data collection and transmission capabilities of the Huawei Cloud IoT platform to clearly display the status information of each production line to management personnel in real time. This status information includes but is not limited to PCB inspection progress, defect identification results, sorting operation records, etc.

[0063] Furthermore, the interactive module 5 makes the entire PCB sorting process more transparent and visual, and clearly displays the status of each assembly line to managers in real time, providing managers with clearer work guidance for PCB inspection and assembly line management, which helps to improve the quality control level of the entire production process.

[0064] The above disclosure is merely a preferred embodiment of a defect sorting system for PCB production according to the present invention. It is certainly not intended to limit the scope of the present invention. A person skilled in the art will understand that any equivalent changes made by implementing all or part of the processes of the above embodiment in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A defect sorting system used in PCB production, characterized by ； It includes image acquisition module, image processing and analysis module, data cloud upload module, defect sorting module and interaction module; The image acquisition module is used to obtain the PCB image and PCB number on the conveyor belt; The image processing and analysis module generates an ONNX model based on the improved network architecture of the YOLOv10 network, converts it into an OM model, inputs the PCB image into the OM model, returns the defect type of the current PCB and marks the defective PCB image; The data cloud upload module is used to upload the defective PCB images to the Huawei Cloud IOT platform and collect them; The defect sorting module detects the number of PCBs passing through based on the infrared detection mechanism and numbers the PCBs. When the number is the same as that of the defective PCB, the defective PCB is sorted out of the conveyor belt; The interactive module is used to obtain the statistical data of all pipelines and the status data of each pipeline, and to display the complete information of the pipeline.

2. The defect sorting system for PCB production according to claim 1, characterized in that ； The image acquisition module includes an infrared detection unit and a USB camera module.

3. The defect sorting system for PCB production according to claim 1, It is characterized by: The image processing and analysis module is based on the Ascend AIPRO development board, and the data cloud upload module is based on the Hi3861v100 chip.

4. The defect sorting system for PCB production according to claim 1, characterized in that ； The defect sorting module includes a digital servo and an infrared detection module.

5. The defect sorting system for PCB production according to claim 1, characterized in that ； The interactive module includes a data preview unit, a pipeline management unit and a detailed pipeline information unit. The data preview unit consists of a List and a chart icon, the pipeline management unit consists of a List and a card, and the detailed pipeline information unit consists of a PCB number and PCB image data.

Images