AOI detection collaborative optimization method and system applied to AI re-judgment and intelligent screening mechanism

By establishing a persistent communication connection between the AI ​​server and the business rule management module in SMT production, AOI detection data is transmitted and filtered in real time. Combined with multi-level confidence thresholds and data caching mechanisms, the problems of high complexity of AOI detection results and unreasonable timing of AI re-judgment are solved, realizing efficient collaboration between AI re-judgment and manual review, and improving the automation and intelligence level of the production process.

CN121680170APending Publication Date: 2026-03-17苏州旗开得电子科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511790772.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-01
Publication Date
2026-03-17

AI Technical Summary

Technical Problem

In the SMT production process, AOI inspection results are highly complex, and it is difficult to maintain the accuracy and consistency of manual review. The timing of AI review is unreasonable, which can lead to data transmission delays or production interruptions. The lack of an effective closed-loop feedback mechanism can result in misjudgments or omissions.

Method used

By establishing a persistent communication connection between the AI ​​server and the business rule management module, setting up an automatic reconnection mechanism, transmitting and filtering detection data in real time, filtering results by combining multi-level confidence thresholds, and constructing a data caching and delay control mechanism, we can achieve automated, intelligent, and closed-loop optimization of data, forming a collaborative mechanism between AI review and manual review.

Benefits of technology

It effectively reduced the amount of manual review, improved the recognition accuracy of AI models, ensured stable data flow in high-paced production environments, reduced misjudgments, missed judgments and delayed responses, and achieved automated and intelligent optimization of the production process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121680170A_ABST
    Figure CN121680170A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of electronic manufacturing, and discloses an AOI detection collaborative optimization method and system applied to an AI re-judgment and intelligent screening mechanism, and the method comprises the following steps: S1, automatic optical detection data collection; s2, establishing AI server communication; s3, data extraction and transmission; s4, a data caching and delay control mechanism; s5, checking bad data; s6, carrying out secondary re-judgment by the re-judgment station equipment; s7, data file generation and tracing query; and S8, performing closed-loop feedback and model optimization. According to the AOI detection collaborative optimization method based on the AI re-judgment and intelligent screening mechanism, efficient transmission and scheduling of detection data, intelligent screening of AI results and closed-loop feedback of manual re-check are achieved, the problems that in the prior art, the re-judgment process depends on manual work, the judgment opportunity is uncertain, and system response is lagged are effectively solved, and the AOI detection collaborative optimization method based on the AI re-judgment and intelligent screening mechanism is achieved. And the intelligent level and the production efficiency of production line quality management are obviously improved.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic manufacturing, and in particular to an AOI detection collaborative optimization method and system applied to AI re-inspection and intelligent screening mechanism. BACKGROUND

[0002] With the development of electronic products towards high density and miniaturization, SMT has become the core production process in the electronic manufacturing industry. In the SMT production process, due to factors such as equipment precision and process parameter deviation, component offset and virtual soldering are prone to occur. In order to ensure product quality, AOI equipment is usually configured on the production line to detect the mounted PCB. The detected abnormal boards need to be transmitted to the re-inspection station for manual re-inspection by the operator.

[0003] However, in actual production, the SMT production line has high rhythm and high throughput characteristics, and the components on the PCB are small in size and numerous in quantity. Long-time re-inspection by the operator is prone to visual fatigue and distraction, resulting in misjudgment or missed judgment. In addition, the combination of different production lines and equipment models has different differences, resulting in high complexity of AOI detection results, and it is difficult to maintain the accuracy and consistency of manual re-inspection.

[0004] The industry has gradually tried to introduce AI re-inspection technology to intelligently analyze and judge the NG pattern in the AOI detection data through an AI model, to realize automatic re-inspection and trend quality analysis; but the existing AI re-inspection technology has the following problems: 1. The AI judgment timing is unreasonable, which may cause data transmission delay or production rhythm interruption; 2. The AI re-inspection result is directly fed back to the operator without effective secondary screening, which may cause misjudgment information diffusion or operation delay; 3. There is no effective closed-loop feedback mechanism, and the AI model cannot be continuously optimized according to the actual re-inspection result. SUMMARY

[0005] The present application aims to provide an AOI detection collaborative optimization method applied to AI re-inspection and intelligent screening mechanism to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions: An AOI detection collaborative optimization method applied to AI re-inspection and intelligent screening mechanism, comprising the following steps: S1, automatic optical detection data acquisition: automatically optically detecting the PCB by an automatic optical detection unit, and collecting and storing the detection data; S2, establishing AI server communication: establishing a persistent communication connection between the business rule management module and the AI server, and setting an automatic reconnection mechanism; S3. Data Extraction and Transmission: The business rule management module extracts the detection data and performs standardized processing, filters out the set of defective component data that the AI ​​model can process, and transmits it to the AI ​​server in real time; the AI ​​server performs intelligent analysis on the received data, and filters and dynamically filters the judgment results based on multi-level confidence thresholds; S4. Data caching and delay control mechanism: Set up a data caching queue and asynchronous scheduling mechanism; S5. Verification of Inappropriate Data: The business rules management module parses the results returned by the AI ​​server and verifies them against the detection results in the local database; data that does not meet the conditions is updated, rewritten, and recorded in a timely manner. S6. Secondary re-judgment by the re-judgment station equipment: The defective data with AI judgment marks is displayed on the terminal of the re-judgment station equipment, and the operator performs secondary confirmation; S7. Data File Generation and Traceability Query: After the re-judgment station equipment performs a second re-judgment on the PCB board, it stores the corresponding data and results in the local database and generates a one-to-one file for subsequent production process data statistics, management, and traceability; thereby enabling the testing process to achieve automation, intelligence, and closed-loop optimization. S8. Closed-loop feedback and model optimization: Collect the difference data between the conclusions of manual review and the initial judgment results of AI to form an error sample set, and periodically send it back to the AI ​​training server for model optimization.

[0007] As a further aspect of the present invention: in step S1, the collected and stored detection data includes: PCB board barcode, panel number, component coordinates, material code, detection conclusion, and defective image path data.

[0008] As a further aspect of the present invention: in step S2, the business rule management module establishes a persistent communication connection with the AI ​​server in the following manner: When the system detects that the automatic optical inspection equipment has entered the working state, the business rules management module automatically initiates a connection request to the AI ​​review server; If the connection is interrupted, the system will automatically execute a reconnection mechanism to ensure stable network interaction throughout the entire production cycle.

[0009] As a further aspect of the present invention: in step S4, the data caching and latency control mechanism includes: When the AI ​​server response latency exceeds the threshold, the unprocessed data will be temporarily stored in the local cache. Once the server recovers, data will be retransmitted from the cache queue according to the first-in, first-out (FIFO) principle.

[0010] On the other hand, this application also provides an AOI detection collaborative optimization system applied to AI review and intelligent screening mechanisms, including: An automated optical inspection unit is used to perform PCB solder joint inspection and generate inspection data. The data processing unit is used to standardize the detection data and communicate with the AI ​​server; The AI ​​judgment unit, based on image feature extraction and matching algorithms, is used to intelligently analyze defective images and output re-judgment results. The review station equipment terminal is used to display the AI ​​review results and other adverse results, and to perform secondary manual confirmation. The data management and traceability unit is used to store judgment records, generate traceability documents, and feed them back to the AI ​​model for retraining.

[0011] As a further embodiment of the present invention: the data processing unit includes: The data standardization module is used to perform structured processing on the detection data; the communication management module is used to manage the network connection with the AI ​​server. The cache management module is used to implement data caching and latency control.

[0012] As a further aspect of the present invention: the AI ​​determination unit includes: The feature extraction module is used to extract feature information from defective images; The confidence score calculation module is used to calculate the confidence score of the judgment result. The results filtering module is used to filter and classify results based on multi-level thresholds.

[0013] As a further embodiment of the present invention: the data management and traceability unit includes: The error sample collection module is used to collect the difference data between manual review and AI judgment; The model training module is used to perform model fine-tuning and retraining; The traceability file generation module is used to generate one-to-one data traceability files.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention dynamically determines the timing of AI re-judgment by comprehensively analyzing production cycle time, inspection task queues, and AI server load status. It can adjust data flow in real time according to production status to ensure AI judgment is performed at the optimal time. Based on AI feature matching, it filters and selects output results, automatically identifying high-confidence qualified boards and low-confidence suspicious boards. High-confidence results are directly approved, while low-confidence results are pushed to manual review. This strategy effectively reduces the number of manual re-judgments. A dynamic feedback relationship is established between AI judgment results and manual review conclusions, feeding back manual review results to the AI ​​model for parameter retraining and judgment logic optimization. Through a periodic self-learning mechanism, the AI ​​model's recognition accuracy continuously improves over long-term operation. Through the organic integration of the AI ​​re-judgment module with AOI inspection equipment and re-judgment workstations, this invention maintains stable data flow and task allocation in high-cycle production environments, reducing fluctuations caused by differences in manual operations. An intelligent collaborative mechanism is formed between AI judgment, manual review, and result feedback, effectively reducing misjudgments, missed judgments, and delayed responses. Attached Figure Description

[0015] Fig. 1 This is a schematic diagram of the framework structure of the AOI detection collaborative optimization method applied to AI review and intelligent screening mechanisms; Fig. 2 This is a schematic diagram illustrating the workflow of the AOI detection collaborative optimization method applied to AI review and intelligent screening mechanisms. Fig. 3 This is a schematic diagram of the framework structure of existing technology. Detailed Implementation

[0016] Please see Figs. 1-3 In this embodiment of the invention, the AOI detection collaborative optimization method applied to AI review and intelligent screening mechanisms includes the following steps: S1. Automated Optical Inspection (AOI) Data Acquisition: The Automated Optical Inspection (AOI) unit performs automatic optical inspection on the PCB board, collecting and storing the inspection data. For example, all AOI equipment on the production lines in the workshop automatically performs optical inspection on each PCB board during production, detecting and identifying common anomalies such as solder joint defects, component misalignment, and cold solder joints, and collecting and storing the corresponding PCB board data in real time. The collected and stored inspection data includes: PCB board barcode, panel number, component coordinates, material code, inspection conclusion (e.g., GOOD or NG), and defect image path data. The collected inspection data is processed by the equipment's internal program and stored in a local database, providing a data source for subsequent secondary review by personnel at the re-evaluation station. The processing methods within the equipment's internal program include: image preprocessing, feature extraction and measurement, and data association and structuring. S2. Establish AI Server Communication: Establish a persistent communication connection between the Business Rules Management (BRM) module and the AI ​​server, and configure an automatic reconnection mechanism; the method for establishing a persistent communication connection between the Business Rules Management module and the AI ​​server is as follows: When the system detects that the Automated Optical Inspection (AOI) equipment has entered the working state, the Business Rules Management (BRM) module automatically initiates a connection request to the AI ​​review server; If the connection is interrupted, the system will automatically execute a reconnection mechanism to ensure stable network interaction throughout the entire production cycle. S3. Data Extraction and Transmission: The Business Rules Management (BRM) module extracts the detection data and performs standardized processing, filters out the set of defective component data that the AI ​​model can process, and transmits it to the AI ​​server in real time. The defective component dataset includes fields such as barcode number, timestamp, defective component name, defective image path, and detection result. The AI ​​server intelligently analyzes the received data, filtering and dynamically selecting the judgment results based on multi-level confidence thresholds. The multi-level confidence thresholds are set as follows: Confidence level ≥ 0.95: Automatic clearance; 0.7 ≤ confidence level < 0.95: Enter manual review. This screening mechanism enables dynamic allocation and optimization of review resources. If the confidence level is <0.7, it is marked as a suspicious item and handled specially; S4. Data Caching and Delay Control Mechanism: A data caching queue and asynchronous scheduling mechanism are set up to prevent data backlog under high load conditions for the AI ​​server. The data caching and delay control mechanism includes: When the AI ​​server response latency exceeds a threshold, unprocessed data will be temporarily stored in the local cache; the threshold can be set based on production cycle time or server performance reliability. The method for setting a threshold based on production cycle time is as follows: First, obtain the key parameters of the production line: (1) Production cycle time: The interval between production of one PCB board on the production line; assumed to be 3.0 seconds / board; (2) AOI inspection time: The time required for the AOI equipment to inspect one board; assumed to be 2.0 seconds; (3) Buffer size: The number of boards that the physical buffer between the AOI and the review station can hold; assumed to be 5 boards; The maximum allowable delay time is calculated as follows: Response delay threshold < buffer capacity × production cycle time - AOI detection time; that is, threshold < 5 blocks × 3.0 seconds / block - 2.0 seconds = 15 seconds - 2.0 seconds = 13.0 seconds; To prevent the system from constantly operating at the limit, a safety factor needs to be set (e.g., 0.6 to 0.8); when the safety factor is 0.7, the final threshold setting is: 13.0 seconds × 0.7 = 9.0 seconds; In this scenario, the response latency threshold can be set to 9.0 seconds; this means that any AI decision that does not return within 9 seconds will have its corresponding PCB board cached in advance to avoid clogging the production line. The method for setting thresholds based on server performance reliability is as follows: The threshold is dynamically adjusted based on the load of the AI ​​server at different times; for example, the threshold is set to 5.0 seconds on normal working days; and a stricter threshold is set to pursue high real-time performance due to the expectation of good server performance. During peak periods or model retraining, the threshold is set to 10.0 seconds; as server response is expected to slow down, the threshold is relaxed to avoid frequent triggering of the caching mechanism due to short-term performance fluctuations. In nighttime or low-load mode, the threshold is set to 7.0 seconds; thus, while ensuring the production line rhythm, a moderate response expectation is provided. Once the server recovers (i.e., the response latency is below the threshold), data is retransmitted from the cache queue according to the first-in-first-out principle, thereby ensuring that the production cycle is not disrupted. S5. Verification of Non-Performing (NG) Data: The Business Rules Management (BRM) module parses the results returned by the AI ​​server and verifies them against the detection results in the local database; data that does not meet the conditions is updated, rewritten, and recorded in a timely manner. S6. Secondary re-judgment at the re-judgment station: Defective data marked with AI judgment is displayed on the terminal of the re-judgment station equipment, and the operator performs secondary confirmation; this can reduce the amount of invalid re-judgment. S7. Data File Generation and Traceability Query: After the re-judgment station equipment performs a second re-judgment on the PCB board, it stores the corresponding data and results in the local database and generates a one-to-one file for subsequent production process data statistics, management, and traceability; thereby enabling the testing process to achieve automation, intelligence, and closed-loop optimization. S8. Closed-loop feedback and model optimization: Collect the difference data between the conclusions of manual review and the initial judgment results of AI to form an error sample set, and regularly send it back to the AI ​​training server for model optimization; continuously improve the recognition accuracy of AI model through a periodic self-learning mechanism.

[0017] An AOI detection collaborative optimization system applied to AI review and intelligent screening mechanisms includes: An automated optical inspection (AOI) unit is used to perform PCB solder joint inspection and generate inspection data; A data processing unit is used to standardize the detection data and communicate with the AI ​​server; the data processing unit includes: The data standardization module is used to perform structured processing on the test data; that is, to convert the heterogeneous and non-standard raw test data generated by AOI equipment of different models and brands into a unified, structured data format that can be recognized by AI models. The communication management module is used to manage the network connection with the AI ​​server; that is, to establish, maintain and manage a stable, efficient and reliable network connection with the AI ​​server, and to be responsible for all data transmission transactions. The cache management module is used to implement data caching and latency control; that is, when the AI ​​server is slow to respond or unavailable, the detection data is temporarily stored and the transmission is delayed to ensure the continuity of the production cycle; and the data is retransmitted in an orderly manner after the server recovers. Suppose that for SMT Line 1, the product is a motherboard PCB, model number MB-20231027-001; there is a defect: component U5 has a solder bridge. During the testing process, the AI ​​server experienced slow response due to instantaneous network fluctuations and excessive load; (1) The data standardization module periodically retrieves new records from the database, maps and normalizes the fields, adds context information, and outputs a clean, unified, and information-rich standardized data package; This ensures that the data received by the subsequent AI model and the manual review are in the same format, eliminating the need for additional parsing, improving processing efficiency, and laying the foundation for orderly flow. (2) After the standardized data packet is sent to the communication management module, the communication management module sends the data packet to the AI ​​server through the established WebSocket connection and starts a 9-second timer; at this time, the AI ​​server fails to respond in time due to excessive load; If the timer times out, the cache management module is immediately triggered; the cache management module securely saves the entire data packet to the local SQLite database, marking the status as "pending"; at the same time, the system immediately releases resources and continues processing the data of the next PCB board; This ensures that the production cycle does not stop at all, and the production line continues to run at a speed of 3 seconds per board. Subsequent PCB board inspection data is not affected by the failure of the AI ​​server at the front end, thus ensuring the stability of the entire system. (3) When the network fluctuations stop after 2 minutes and the AI ​​server load decreases, the heartbeat detection of the communication management module resumes, confirming that the connection is stable; the cache management module detects the connection recovery and retrieves the cached data packets in first-in-first-out order, including the request MB-20231027-001; the communication management module resends these cached data packets to the AI ​​server; the AI ​​server processes them normally this time and returns the judgment result; Therefore, despite the delay, no data was lost, the processing order was maintained, and all system states were eventually synchronized and consistent. When the operator sees this board at the review station, the AI's intelligent conclusion is already clearly visible, and he only needs to confirm it easily.

[0018] The AI ​​judgment unit, based on image feature extraction and matching algorithms, is used to intelligently analyze defective images and output re-judgment results; the AI ​​judgment unit includes: The feature extraction module is used to extract feature information from defective images; that is, to transform the original, pixel-level defective images into a set of high-dimensional, digital feature vectors that can accurately characterize the nature of their defects. Suppose that component U5 has obvious solder bridging, showing a clear and continuous solder connection between two adjacent solder joints in the image; then the output feature vector will have extremely high activation values ​​in dimensions representing "irregular protrusions" and "large bright area connections"; this vector is very similar to the "bridging" features learned by the AI ​​model during training. Suppose that component U5 has a slight spike, which is shown in the image as a small, spike-like solder wire at the top of the solder joint; then the output feature vector will have higher activation values ​​in dimensions representing "small linear protrusions" and "local highlights", while the activation values ​​in dimensions related to "bridging" or "insufficient solder" will be lower. The confidence calculation module is used to calculate the confidence score of the judgment result; that is, it receives the feature vector output by the feature extraction module, passes it through a classifier (usually a Softmax layer or a support vector machine SVM), calculates the probability that the feature vector belongs to each predefined defect category, and uses a quantified score (confidence) to characterize the credibility of the judgment result. For example, the feature vector [0.12, -0.45, 0.88, ... , 0.67] from the feature extraction module has the following probability distribution: P(solder bridging) = 0.02; P(component offset) = 0.01; P(cold solder joint) = 0.07; P(pull-up) = 0.90; take the highest probability value among all categories as the confidence score for this judgment; for example, if the judgment result is "pull-up", the confidence score is 0.90. Assume that component U5 has a typical "pulling tip" characteristic (high confidence); the feature vector is highly matched with the "pulling tip" feature; the probability distribution is [0.02, 0.01, 0.07, 0.90]; then the confidence level is 0.90. Assume that component U5 has a defect that is difficult to distinguish (low confidence); display a fuzzy case on the image that falls between "cold solder joint" and "insufficient solder"; the feature vector contains certain features that simultaneously contain both types of defects and does not perfectly match any standard prototype; The probability distribution is [0.05, 0.40, 0.45, 0.10]; the highest probability is 0.45, which is judged as "cold solder joint", but the confidence level is 0.45; this indicates that the model is "uncertain" and the image features are blurred. The result filtering module is used to filter and classify results based on multi-level thresholds; that is, based on the score output by the confidence calculation module, a preset multi-level threshold strategy is executed to automatically divert and filter the AI ​​judgment results, thereby realizing the intelligent allocation of review resources. (1) In the case of confidence level ≥ 0.95 (automatic release); assuming input: judgment result = GOOD (i.e. AOI false alarm), confidence level = 0.97; Filtering action: If 0.97 ≥ 0.95, then the "automatic release" rule will be triggered; Result: The PCB board was marked as "AI review passed" in the system and can be directly moved to the next process without manual intervention; (2) In the case of 0.70 ≤ confidence level < 0.95 (pushing to manual review); assuming input: judgment result = solder bridging, confidence level = 0.85; Filtering action: 0.70 ≤ 0.85 < 0.95; then the "push for manual review" rule is triggered. Result: The defect entry was sent to the operator interface of the review station with the label "AI suggestion: bridging", and the operator quickly confirmed it. In cases with a confidence level < 0.70 (marked as suspicious); assuming input: decision result = component offset, confidence level = 0.55; Filtering action: 0.55 < 0.70; then the "mark as suspicious" rule is triggered; Result: The entry was sent to a special "difficult case" queue, which may be reviewed by senior engineers, or its image may be automatically collected as a difficult sample for subsequent model retraining.

[0019] The review station terminal is used to display the AI ​​review results and other adverse results, and to perform secondary manual confirmation.

[0020] The data management and traceability unit is used to store judgment records, generate traceability files, and feed them back to the AI ​​model for retraining; the data management and traceability unit includes: The error sample collection module is used to collect the difference data between human review and AI judgment; that is, to systematically and automatically capture and build a golden dataset for AI model optimization; and to identify those that AI judged incorrectly or uncertainly, but which human review gave the correct answer. Assuming the AI's judgment is: Judgment = NG, Defect Type = Cold Solder, Confidence Level = 0.88; Manual judgment: Final judgment = GOOD, Note: Shadow misjudgment; Comparison results: Inconsistent; AI identified it as a "cold solder joint," while human confirmation indicated it was actually a "good product." Then proceed as follows: (1) Difference identification and classification: Identifying the specific types of differences, there are two common types: False positive: AI false alarm; both AOI and AI identify it as defective, but it is confirmed as good by humans; False negative: AI misses a test; the AOI identifies a defect, but the AI ​​mistakenly allows it to pass, which is then detected and corrected by human workers. If AI identifies a good product as having "poor soldering", this is a false positive error. (2) Package all relevant data of the discrepancy cases into a structured “error sample” and output a continuously growing, high-quality error sample library, in which each sample clearly points out the cognitive blind spot or error of the current AI model; The model training module is used to perform model fine-tuning and retraining; that is, by using the "error notebook" provided by the error sample collection module, the AI ​​model is periodically retrained in a targeted manner to correct its errors, optimize its performance, and achieve continuous self-evolution of the model. Suppose the system has been running for a week and has collected 500 error samples; the training module mixes these 500 new samples with the original 100,000 basic training samples to form a new training set containing 100,500 samples. During training, the model repeatedly sees the shadow images that it misclassified as "cold solder joints," but this time these images are labeled "GOOD." Through the backpropagation algorithm, the model gradually adjusts its millions of internal parameters and learns that "this texture feature is more likely to be a shadow than a cold solder joint." The new model reduced the false positive rate on the test set from 1.5% to 0.8%, while maintaining the original high recall rate. After passing the evaluation, the system automatically updated the new model file to the AI ​​server on the production line. The traceability file generation module is used to generate one-to-one data traceability files; that is, to generate a complete, independent and tamper-proof "inspection medical record" for each PCB board that has undergone the re-inspection process, so as to realize quality traceability throughout the entire production process. Suppose that for a PCB board with barcode MB-20231027-001, the following data is aggregated: Basic information (board code, production batch, timestamp), AOI raw data (all defects detected by AOI, original image path), AI judgment data (AI judgment result and confidence score for each defect), manual review data (operator's final confirmation result, review time, operator ID), closed-loop data (whether the board is included in the error sample library); and generate a traceability file from the aggregated data and store it in an easily searchable manner; If a customer reports a problem with poor soldering in a certain batch of products, the factory can directly access the traceability files for all PCBs in that batch, quickly locate the boards with low confidence in the AI ​​assessment but which were manually approved, analyze the basis for the reassessment at that time, and determine responsibility. For example, if engineers discover that a certain type of capacitor has a high misjudgment rate, they can analyze the traceability files to identify the AI ​​misjudgment patterns for that component, and then optimize the AOI detection parameters or AI training samples accordingly.

[0021] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An AOI detection collaborative optimization method applied to AI re-trial and intelligent screening mechanism, characterized in that, The method comprises the following steps: S1, automatic optical inspection data acquisition: automatic optical inspection is performed on the PCB by an automatic optical inspection unit, and detection data is collected and stored; S2, establishing AI server communication: establishing a persistent communication connection between the business rule management module and the AI server, and setting up an automatic reconnection mechanism; S3, data extraction and transmission: the business rule management module extracts the detection data and performs standardization processing, selects a set of defective component data that can be processed by the AI model, and transmits it to the AI server in real time; the AI server intelligently analyzes the received data, and filters and dynamically filters the determination results based on a confidence level multi-level threshold; S4, data caching and delay control mechanism: setting up a data caching queue and an asynchronous scheduling mechanism; S5, verification of defective data: the business rule management module analyzes the AI server feedback results, and checks and verifies them with the detection results in the local database; the data that do not meet the conditions are updated and rewritten in time and recorded; S6, secondary review of the review station equipment: the defective data with AI determination marks are displayed on the review station equipment terminal, and are confirmed by the operator; S7, data file generation and traceability query: after the review station equipment performs secondary review on the PCB, the corresponding data and results are stored in the local database, and a one-to-one file is generated for subsequent production process data statistics and management traceability; thus, the detection process is realized in an automatic, intelligent and closed-loop optimized manner; S8, closed-loop feedback and model optimization: collecting the difference data between the artificial review conclusion and the AI initial determination result, forming an error sample set, and periodically feeding back to the AI training server for model optimization.

2. The AOI detection collaborative optimization method applied to an AI re-trial and intelligent screening mechanism according to claim 1, characterized in that, In the S1 step, the collected and stored detection data includes: PCB barcode, board number, component coordinates, material code, detection conclusion, and defective image path data. 3.The AOI detection collaborative optimization method applied to an AI re-trial and intelligent screening mechanism according to claim 1, wherein, In the S2 step, the business rule management module and the AI server establish a persistent communication connection in the following manner: When the system detects that the automatic optical inspection equipment enters a working state, the business rule management module automatically initiates a connection request to the AI review server; If the connection is interrupted, the system automatically executes a reconnection mechanism to ensure stable network interaction during the entire production cycle. 4.The AOI detection collaborative optimization method applied to an AI re-trial and intelligent screening mechanism according to claim 1, wherein, In the S4 step, the data caching and delay control mechanism includes: When the AI server response delay exceeds the threshold, the unprocessed data is temporarily stored in the local cache area; When the server recovers; the data is retransmitted from the cache queue according to the first-in-first-out principle.

5. An AOI detection collaborative optimization system applied to AI rejudication and intelligent screening mechanism, characterized in that, It comprises: An automatic optical inspection unit for performing PCB solder joint detection and generating detection data; A data processing unit for standardizing the detection data and communicating with the AI server; An AI determination unit based on image feature extraction and matching algorithm, for intelligently analyzing the defective images and outputting review results; A review station equipment terminal for displaying AI review results and other defective results and performing artificial secondary confirmation; A data management and traceability unit for storing determination records, generating traceability files, and feeding back to the AI model for retraining. 6.The AOI detection collaborative optimization system applied to AI re-trial and intelligent screening mechanism according to claim 5, wherein, The data processing unit comprises: A data standardization module for structuring the detection data; a communication management module for managing network connection with the AI server; A cache management module for implementing data caching and delay control. 7.The AOI detection collaborative optimization system applied to AI re-trial and intelligent screening mechanism of claim 5, wherein, The AI determination unit includes: A feature extraction module for extracting feature information from defective images; A confidence score calculation module for calculating the confidence score of the determination result; A result screening module for filtering and classifying the results according to multiple thresholds. 8.The AOI detection collaborative optimization system applied to AI re-trial and intelligent screening mechanism of claim 5, wherein, The data management and traceability unit includes: An error sample collection module for collecting difference data between manual review and AI determination; A model training module for performing model fine-tuning and retraining; A traceability file generation module for generating one-to-one data traceability files.