An automatic defect detection system

Abstract

This invention provides an automatic defect detection system, relating to the field of automatic defect detection technology. The system includes a switch, a client, a DFS server, and a network server. The network server includes an application server, an AI image judgment server, and a database server. The DFS server uses Detail field information matching to scan and locate image files. The network server is configured with a shortest task time image judgment mode, a maximum task volume image judgment mode, and a comprehensive image judgment mode. Because the DFS server uses Detail field information matching to scan and locate image files, it can accurately match the files needed for image judgment, reducing the resource overhead of the DFS server. Furthermore, because the network server adopts a multi-scenario resource adaptation strategy, limited resources can play a greater role in different business scenarios. That is, by comprehensively improving the DFS server and the network server, the system's automatic image judgment speed is improved.

Description

Technical Field

[0001] This invention relates to the field of automatic defect detection technology, and more specifically, to an automatic defect detection system. Background Technology

[0002] PCB substrate processing plants generate numerous defects during production. However, the entire PCB manufacturing process is complex and has a long cycle, often requiring considerable time from substrate to completion. Therefore, defects generated at each stage of the process need constant monitoring to prevent them from flowing into the next process and reducing yield. Defect detection requires identifying and classifying defects. Traditional defect detection relies on manual image interpretation, which is costly and suffers from low accuracy due to human subjectivity and fatigue.

[0003] Currently, many PCB substrate processing plants have introduced automated defect detection systems to replace manual inspection. Specifically, the existing ADC system's defect detection process is as follows: 1. Read the ***IDX.CSV file (directory file) and parse it to obtain different production information parameters (including material, batch, and sequence); 2. Combine the material, batch, and sequence to obtain the ***LOG.TXT directory; 3. Enter the ***LOG.TXT directory and parse to locate the image position; 4. Obtain the ***.BMP image based on the image position for AI automatic image judgment. The ***IDX.CSV file is a notification and index file, stored in a fixed directory and generated by date. Each time a PCB substrate is photographed, a line of data is added to the ***IDX.CSV file, including time, machine number, material, batch, sequence, product information, and site information. By concatenating the material, batch, and sequence, a PCB substrate file directory prefix is ​​obtained, and based on fuzzy matching of the prefix, the directory of the PCB substrate can be obtained, leading to the corresponding image. Considering the existing automated defect detection system's process flow and hardware / software performance, the main technical challenges are as follows:

[0004] (1) Due to the use of fuzzy matching prefixes for image search and the large number of automatically judged images, the DFS server has a large load and the automatic image judgment speed of the automatic defect detection system is slow. Furthermore, due to the diverse business scenarios of the automatic defect detection system, the image judgment mode of the automatic defect detection system cannot meet the needs of the business scenarios, and the automatic image judgment speed of the automatic defect detection system cannot be significantly improved for different business scenarios.

[0005] (2) Since the automatic defect detection system cannot be shut down except for maintenance, the automatic defect detection system cannot meet the production needs.

[0006] (3) Due to issues such as upload delays, file format errors, and network anomalies, automatic defect detection systems are prone to freezing. Summary of the Invention

[0007] This invention provides a hierarchical label text classification method, system, device, and storage medium, which effectively solves the problem of poor hierarchical label classification performance in traditional hierarchical label classification tasks due to the lack of full utilization of the correlation between hierarchical labels and the potential limitation of coarse-grained labels on fine-grained labels.

[0008] This invention provides an automatic defect detection system. The system adopts a B / S architecture to automatically detect defects on a PCB carrier board. The system includes a switch, a client connected to the switch via a network, a DFS server, and a network server.

[0009] The network server includes an application server for web front-end services, manual image interpretation services, and automatic image interpretation services; an AI image interpretation server for AI image interpretation inference services; and a database server for data storage services.

[0010] The DFS server uses Detail field information matching to scan and locate image files.

[0011] The network server adopts a multi-scenario resource adaptation strategy and is equipped with a shortest task time map judgment mode, a maximum task volume map judgment mode, and a comprehensive map judgment mode.

[0012] In the above embodiments, since the DFS server uses Detail field information matching to scan and locate image files, it can accurately match the files needed for image judgment, greatly reducing the resource consumption of the DFS server and alleviating its load. At the same time, since the network server adopts a multi-scenario resource adaptation strategy, that is, the network server is set with shortest task time image judgment mode, maximum task volume image judgment mode, and comprehensive image judgment mode, it can effectively utilize the network server's resources, so that limited resources can play a greater role in different application scenarios. That is, comprehensive improvements are made to the DFS server and the network server to improve the automatic image judgment speed of the ADC system.

[0013] As some optional implementations of this application, the shortest task time image judgment mode uses a single thread to parse image files and judge image defects.

[0014] In the above embodiments, the shortest task time image judgment mode is that one thread of the application server corresponds to multiple AI image judgment services of the AI ​​image judgment server. This mode is suitable for scenarios with a small number of tasks but a huge number of images per task, by centralizing all AI image judgment instances to process the same task, thereby reducing the processing time of a single task.

[0015] As some optional implementations of this application, the maximum task volume image judgment mode adopts a multi-threaded parallel approach to parse image files and judge image defects.

[0016] In the above embodiments, the maximum task volume image judgment mode is to correspond one-to-one with the same number of AI image judgment services of the AI ​​image judgment server in multiple threads of the application server. This mode is suitable for scenarios with large task volume but small number of images per task. The AI ​​image judgment instances process different tasks respectively, thereby increasing the number of parallel image judgment tasks.

[0017] As some optional implementations of this application, the comprehensive image judgment mode adopts a combined approach of single-threaded and multi-threaded parallelism to parse image files and judge image defects.

[0018] In the above embodiments, the comprehensive image judgment mode combines the shortest task time image judgment mode and the maximum task volume image judgment mode. This mode is suitable for scenarios with many and complex tasks, and it compromises the image judgment task volume and single task image judgment time.

[0019] As some optional implementations of this application, the application server, AI image judgment server, and database server all adopt a dual-machine mode.

[0020] In the above embodiments, the dual-machine mode is adopted by setting up two application servers, two AI image judgment servers and two database servers respectively, and each server adopts a dual-active high availability strategy, so that the failure of any one server will not affect the normal operation of the automatic defect detection system.

[0021] As some optional implementations of this application, the DFS server, application server, AI image judgment server, and database server are all equipped with fault tolerance mechanisms.

[0022] In the above embodiments, since abnormal situations such as image file upload delays, file format errors, and slow networks are relatively common, adding a fault tolerance mechanism at the software layer can effectively prevent the automatic defect detection system from freezing.

[0023] As some optional implementations of this application, the AI ​​image judgment server adopts a dual-graphics card configuration. Since the automatic image judgment speed depends on the file parsing time of the application server and the AI ​​image judgment time of the AI ​​image judgment server, and the time for the program to parse files and summarize results is usually much shorter than the AI ​​image judgment time, it is necessary to focus on the resource consumption of AI image judgment. A V100 series graphics card is used, and a single graphics card can run 16 AI image judgment instances at the same time, while dual graphics cards can run 32 AI image judgment instances at the same time, which meets the requirement of the system being in the shortest task time image judgment mode.

[0024] As some optional implementations of this application, the application server, AI image judgment server and database server are all configured with dual CPUs, with each CPU containing 16 cores.

[0025] In the above embodiments, in order to run 32 AI image judgment instances simultaneously, from a software perspective, 32 threads are needed to correspond to the AI ​​image judgment instances to perform tasks such as file parsing, database saving, and result pushing. Therefore, the application server, AI image judgment server, and database server all adopt a dual-CPU configuration, with each CPU containing 16 cores. That is, the number of threads that can run in parallel in the best performance mode of a single server is 64.

[0026] As some optional embodiments of this application, the overall system adopts a B / S architecture to automatically detect defects on the PCB carrier board, and the process is as follows:

[0027] (1) The DFS server uses the Detail field information matching method to scan and locate image files in order to obtain the tasks to be processed, and generates a list of tasks to be processed based on several tasks to be processed.

[0028] (2) The network server obtains the list of tasks to be processed, parses the located image files, and performs AI image judgment reasoning service based on the selection of the shortest task time image judgment mode, the maximum task volume image judgment mode and the comprehensive image judgment mode, so as to realize the image defect judgment.

[0029] As some optional embodiments of this application, the path for scanning and locating the task to be processed is an absolute path formed by concatenating fields.

[0030] As some optional implementations of this application, the process of scanning and locating image files based on the DFS server using Detail field information matching to obtain tasks to be processed, and generating a list of tasks to be processed based on several tasks to be processed, is as follows:

[0031] (1.1) Asynchronous timer A calls the DFS scanning program;

[0032] (1.2) The DFS scanning program connects to the DFS server, reads the directory files within a preset time range, and parses them to obtain several lines of production information data;

[0033] (1.3) Add a column of ISON format distinguishing data after each row of production information data, that is, realize the image file scanning and positioning based on the splicing of production information data and ISON format distinguishing data, so as to obtain the task to be processed;

[0034] (1.4) Generate a list of tasks to be processed based on several tasks to be processed.

[0035] As some optional implementations of this application, the network server obtains a list of tasks to be processed, parses the located image files, and performs AI image judgment inference service based on the selection of the shortest task time image judgment mode, the maximum task volume image judgment mode, and the comprehensive image judgment mode, so as to realize the following process of image defect judgment:

[0036] (2.1) Asynchronous timer B calls the automatic image determination program;

[0037] (2.2) The automatic image judgment program traverses the list of tasks to be processed and assigns the tasks to threads in the thread pool;

[0038] (2.3) The thread pool parses the task to be processed and downloads the image based on an absolute path obtained by concatenating the data according to the production information data and the ISON format;

[0039] (2.4) Calculate the number of images and select the image judgment mode according to the number of images, that is, select the shortest task time image judgment mode, the maximum task volume image judgment mode or the comprehensive image judgment mode.

[0040] (2.5) Defect judgment of images based on the selected image judgment mode.

[0041] The beneficial effects of this invention are as follows:

[0042] 1. The DFS server of this invention uses Detail field information matching to scan and locate image files, which can accurately match the files needed for image judgment, greatly reducing the resource consumption and load of the DFS server. At the same time, because the network server adopts a multi-scenario resource adaptation strategy, that is, the network server is set with a shortest task time image judgment mode, a maximum task volume image judgment mode, and a comprehensive image judgment mode, it can effectively utilize the network server resources, so that limited resources can play a greater role in different scenarios. That is, by comprehensively improving the DFS server and the network server, the automatic image judgment speed of the automatic defect detection system can be improved to meet the needs of different business scenarios.

[0043] 2. The application server, AI image judgment server and database server of the present invention all adopt a dual-machine mode, that is, two application servers, two AI image judgment servers and two database servers are set up respectively, and each server adopts a dual-active high availability strategy, so that the failure of any one server will not affect the normal operation of the automatic defect detection system.

[0044] 3. The DFS server, application server, AI image judgment server and database server of the present invention are all equipped with fault tolerance mechanisms, that is, fault tolerance mechanisms are added at the software layer, which can effectively prevent the automatic defect detection system from freezing or other situations. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is an overall structural block diagram of the automatic defect detection system according to an embodiment of the present invention;

[0047] Figure 2 This is a structural block diagram of the shortest task time decision-making mode according to an embodiment of the present invention;

[0048] Figure 3 This is a structural block diagram of the maximum task load determination mode according to an embodiment of the present invention;

[0049] Figure 4 This is a structural block diagram of the integrated image judgment mode according to an embodiment of the present invention;

[0050] Figure 5 This is a flowchart of the DFS scanning procedure according to an embodiment of the present invention;

[0051] Figure 6 This is a flowchart of the automatic image judgment procedure according to an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0053] The components of the embodiments of the invention described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0054] In the following, the terms “comprising,” “having,” and their cognates, which may be used in various embodiments of the invention, are intended only to indicate a particular feature, number, step, operation, component, or combination thereof, and should not be construed as excluding, firstly, the presence of one or more other features, numbers, steps, operations, components, or combinations thereof, or adding the possibility of one or more combinations thereof.

[0055] This invention provides an automatic defect detection system. Please refer to [link / reference]. Figure 1 The system adopts a B / S architecture to automatically detect defects on PCB carrier boards. The system includes a switch, a client connected to the switch via a network, a DFS server, and a network server.

[0056] The network server includes an application server for web front-end services, manual image interpretation services, and automatic image interpretation services; an AI image interpretation server for AI image interpretation inference services; and a database server for database services.

[0057] The DFS server uses Detail field information matching to scan and locate image files.

[0058] The network server is configured with a shortest task time map judgment mode, a maximum task volume map judgment mode, and a comprehensive map judgment mode.

[0059] In this embodiment of the invention, since the system adopts a B / S architecture, that is, a working mode in which the browser requests data and the server responds, this working mode has advantages such as simple operation, easy maintenance, easy expansion, and strong distribution.

[0060] In this embodiment of the invention, since the DFS server uses Detail field information matching to scan and locate image files, the load on the DFS server is reduced by modifying business rules. The core optimization logic is to modify ambiguous files and file locations into data with explicit names and locations. An ISON format data column is added after the path information of the original directory file (***IDX.CSV file) to clarify the data contained in the task. In this way, when parsing the data, an absolute path set can be formed by concatenating fields. Through this absolute path set, all files required for image judgment can be directly located, which can greatly reduce resource consumption and improve performance.

[0061] At the same time, because the network server adopts a multi-scenario resource adaptation strategy, that is, the network server is set with the shortest task time graph judgment mode, the maximum task volume graph judgment mode, and the comprehensive graph judgment mode, it can effectively utilize the network server resources, so that the limited resources can play a greater role in different application scenarios.

[0062] This involves comprehensive improvements to the DFS server and network server to enhance the automatic image interpretation speed of the automatic defect detection system.

[0063] In the embodiments of the present invention, please refer to Figure 2 The shortest task time image judgment mode adopts a single-threaded approach to parse image files and judge image defects. The shortest task time image judgment mode means that one thread of the application server corresponds to multiple AI image judgment services of the AI ​​image judgment server. This mode is suitable for scenarios with a small number of tasks but a huge number of images per task. It centralizes all AI image judgment instances to process the same task, reducing the processing time of a single task.

[0064] In the embodiments of the present invention, please refer to Figure 3 The maximum task volume image judgment mode adopts a multi-threaded parallel approach to parse image files and judge image defects. The maximum task volume image judgment mode means that multiple threads on the application server correspond one-to-one with the same number of AI image judgment services on the AI ​​image judgment server. This mode is suitable for scenarios with a large workload but a small number of images per task. The AI ​​image judgment instances handle different tasks separately, increasing the number of parallel image judgment tasks.

[0065] In the embodiments of the present invention, please refer to Figure 4 The comprehensive image judgment mode uses a combination of single-threaded and multi-threaded parallel methods to parse image files and determine image defects. This comprehensive image judgment mode combines the shortest task time image judgment mode and the maximum task volume image judgment mode. This mode is suitable for scenarios with many and complex tasks, and it strikes a balance between the number of image judgment tasks and the single task image judgment time.

[0066] In this embodiment of the invention, the application server, AI image judgment server, and database server all adopt a dual-machine mode. The dual-machine mode means that two application servers, two AI image judgment servers, and two database servers are set up respectively, and each server adopts a dual-active high-availability strategy, so that the failure of any one server will not affect the normal operation of the automatic defect detection system.

[0067] In this embodiment of the invention, the DFS server, application server, AI image judgment server, and database server are all equipped with fault tolerance mechanisms. Since image file upload delays, file format errors, slow network speeds, and other abnormal situations are relatively common, adding fault tolerance mechanisms at the software layer can effectively prevent the automatic defect detection system from freezing. For example: ① adding a retry strategy when the network is disconnected; ② adding file format verification and error prompts; ③ setting a waiting time for file upload delays before retrying; ④ terminating any abnormal program promptly and assigning it to an abnormal task, waiting for a period of time before retrying execution.

[0068] In this embodiment of the invention, the AI ​​image judgment server adopts a dual-graphics card configuration. Since the automatic image judgment speed depends on the file parsing time of the application server and the AI ​​image judgment time of the AI ​​image judgment server, and the time for the program to parse files and summarize results is usually much shorter than the AI ​​image judgment time, it is necessary to focus on the resource consumption of AI image judgment. V100 series graphics cards are used, and a single graphics card can run 16 AI image judgment instances at the same time, while dual graphics cards can run 32 AI image judgment instances at the same time, which meets the requirement of the automatic defect detection system to be in the shortest task time image judgment mode.

[0069] In this embodiment of the invention, the application server, AI image judgment server, and database server all adopt a dual-CPU configuration, with each CPU containing 16 cores. In order to run 32 AI image judgment instances simultaneously, from a software perspective, 32 threads are needed to correspond to the AI ​​image judgment instances to perform tasks such as file parsing, database saving, and result pushing. Therefore, the application server, AI image judgment server, and database server all adopt a dual-CPU configuration, with each CPU containing 16 cores. That is, the number of threads that can run in parallel under the best performance mode of a single server is 64.

[0070] Specifically, the system adopts a B / S architecture to automatically detect defects on the PCB carrier board. The process is as follows:

[0071] (1) The DFS server uses the Detail field information matching method to scan and locate image files in order to obtain the tasks to be processed, and generates a list of tasks to be processed based on several tasks to be processed.

[0072] The path for scanning and locating the task to be processed is an absolute path formed by concatenating fields.

[0073] (2) The network server obtains the list of tasks to be processed, parses the located image files, and performs AI image judgment reasoning service based on the selection of the shortest task time image judgment mode, the maximum task volume image judgment mode and the comprehensive image judgment mode to realize image defect judgment.

[0074] In the embodiments of the present invention, please refer to Figure 5 The process of using the DFS server to scan and locate image files based on Detail field information matching to obtain tasks to be processed, and generating a list of tasks to be processed based on several tasks to be processed, is as follows:

[0075] (1.1) Asynchronous timer A calls the DFS scanning program periodically according to the specified period, and locks the current scan and writes the lock data, including the execution time of the current scan, the expiration time, the service name of the executing program, etc. This information can be used as the basis for judging timeout and program abnormality, so as to monitor the system status.

[0076] (1.2) The DFS scanning program connects to the DFS server, reads the directory files within a preset time range, parses and obtains several lines of production information data, and judges whether each line of production information data is newly added: In order to reduce the pressure on the DFS database and improve efficiency, the DFS scanning program first queries the cache. If it is newly added, it proceeds to the next step; otherwise, the DFS scanning program completes the execution and releases the lock information.

[0077] (1.3) Add a column of ISON format distinguishing data after each row of production information data, that is, realize the image file scanning and positioning based on the splicing of production information data and ISON format distinguishing data, so as to obtain the task to be processed;

[0078] (1.4) Generate a list of tasks to be processed based on several tasks to be processed.

[0079] In the embodiments of the present invention, please refer to Figure 6 The network server obtains the list of tasks to be processed, parses the located image files, and performs AI image judgment inference service based on the selection of the shortest task time image judgment mode, the maximum task volume image judgment mode, and the comprehensive image judgment mode. The process of realizing image defect judgment is as follows:

[0080] (2.1) Asynchronous timer B calls the automatic image judgment program and locks the current scan, writing lock data, including the execution time of the current scan, expiration time, service name of the executing program, etc. This information can be used as the basis for judging timeout and program abnormality, so as to monitor the system status.

[0081] (2.2) The automatic image judgment program traverses the list of tasks to be processed and assigns the tasks to threads in the thread pool;

[0082] (2.3) The thread pool parses the task to be processed and downloads the image based on an absolute path obtained by concatenating the data according to the production information data and the ISON format;

[0083] (2.4) Calculate the number of images and select the image judgment mode according to the number of images, that is, select the shortest task time image judgment mode, the maximum task volume image judgment mode or the comprehensive image judgment mode.

[0084] (2.5) Based on the selected image judgment mode, image defects are judged. At the same time, according to the confidence level of image defect judgment, images with confidence levels below the threshold are converted to manual image judgment, realizing an effective combination of automatic and manual image judgment, and improving the flexibility and accuracy of the system.

[0085] In summary, the automatic defect detection system of this invention, through comprehensive improvements to the defect detection process, software layer, hardware layer, and business model, effectively reduces system load and increases data processing speed, enabling the automatic defect detection system to meet the needs of more business scenarios. Therefore, this invention effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0086] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. An automatic defect detection system, characterized in that: The system adopts a B / S architecture to automatically detect defects on PCB carrier boards. The system includes a switch, a client connected to the switch via a network, a DFS server, and a network server. The system adopts a B / S architecture to automatically detect defects on PCB carrier boards. The process is as follows: (1) The DFS server uses the Detail field information matching method to scan and locate image files in order to obtain the tasks to be processed, and generates a list of tasks to be processed based on several tasks to be processed. (2) The network server obtains the list of tasks to be processed, parses the located image files, and performs AI image judgment reasoning service based on the selection of the shortest task time image judgment mode, the maximum task volume image judgment mode and the comprehensive image judgment mode, so as to realize the image defect judgment. The path for scanning and locating the task to be processed is an absolute path formed by concatenating fields. The network server includes an application server for web front-end services, manual image interpretation services, and automatic image interpretation services; an AI image interpretation server for AI image interpretation inference services; and a database server for data storage services. The DFS server uses Detail field information matching to scan and locate image files. Furthermore, the process of using the DFS server to scan and locate image files based on Detail field information matching to obtain tasks to be processed, and generating a list of tasks to be processed based on several tasks to be processed, is as follows: (1.1) Asynchronous timer A calls the DFS scanning program; (1.2) The DFS scanning program connects to the DFS server, reads the directory files within a preset time range, and parses them to obtain several lines of production information data; (1.3) Add a column of ISON format distinguishing data after each row of production information data, that is, realize the image file scanning and positioning based on the splicing of production information data and ISON format distinguishing data, so as to obtain the task to be processed; (1.4) Generate a list of tasks to be processed based on several tasks to be processed; The network server adopts a multi-scenario resource adaptation strategy and is equipped with a shortest task time graph judgment mode, a maximum task volume graph judgment mode, and a comprehensive graph judgment mode. The process involves the network server acquiring a list of tasks to be processed, parsing the located image files, and performing AI image judgment inference services based on the selection of the shortest task time image judgment mode, the maximum task volume image judgment mode, and a comprehensive image judgment mode. The process for determining image defects is as follows: (2.1) Asynchronous timer B calls the automatic image determination program; (2.2) The automatic image judgment program traverses the list of tasks to be processed and assigns the tasks to threads in the thread pool; (2.3) The thread pool parses the task to be processed and downloads the image based on an absolute path obtained by concatenating the data according to the production information data and the ISON format; (2.4) Calculate the number of images and select the image judgment mode according to the number of images, that is, select the shortest task time image judgment mode, the maximum task volume image judgment mode or the comprehensive image judgment mode. (2.5) Defect judgment of images based on the selected image judgment mode.

2. The automatic defect detection system according to claim 1, characterized in that: The shortest task time image judgment mode uses a single thread to parse image files and judge image defects.

3. The automatic defect detection system according to claim 1, characterized in that: The maximum workload image judgment mode uses a multi-threaded parallel approach to parse image files and judge image defects.

4. The automatic defect detection system according to claim 1, characterized in that: The comprehensive image judgment mode uses a combination of single-threaded and multi-threaded parallel methods to parse image files and judge image defects.

5. The automatic defect detection system according to claim 1, characterized in that: The application server, AI image processing server, and database server all adopt a dual-machine mode.

6. The automatic defect detection system according to claim 1, characterized in that: The DFS server, application server, AI image analysis server, and database server are all equipped with fault tolerance mechanisms.

7. The automatic defect detection system according to claim 1, characterized in that: The AI ​​image processing server uses a dual-graphics card configuration.

8. An automatic defect detection system according to claim 1, characterized in that: The application server, AI image analysis server, and database server all adopt a dual-CPU configuration, with each CPU containing 16 cores.

Images