Detection strategy generation method and device, equipment, storage medium and program product
By constructing a two-dimensional to three-dimensional thermal mapping model and a large language model for panel defects to generate detection strategies, the problem of poor adaptability to process differences in panel inspection is solved, and efficient detection strategy generation and improved operation and maintenance efficiency are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-14
- Publication Date
- 2026-04-14
AI Technical Summary
In existing automated optical inspection systems for panel manufacturing, fixed strategy models cannot adapt to the process differences between different batches, resulting in high false alarm rates, insufficient flexibility, difficulty in dynamically responding to production line drift, and cumbersome manual adjustments, which affect inspection efficiency.
By acquiring the panel's risk score and spatial coordinates, a two-dimensional to three-dimensional thermal mapping model is constructed. Combined with a large language model, interpretable classification criteria and structured threshold strategies are generated to dynamically adapt to the testing needs of different production lines and batches, and to adjust the testing strategy in real time.
It enables flexible adaptation to process fluctuations in different production lines and batches, reduces false alarm and false negative rates, and improves the flexibility of detection strategy generation and production line response speed.
Smart Images
Figure CN121860427A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of automated optical inspection, and more specifically, to a method, apparatus, device, storage medium, and program product for generating inspection strategies. Background Technology
[0002] In the field of Automated Optical Inspection (AOI) in panel manufacturing, the accurate identification of particle defects and pixel structure anomalies directly determines product yield and production line operation efficiency. With the improvement of panel resolution, the diversification of pixel structure, and the iteration of production processes, the inspection system needs to adapt to the process fluctuations of different batches and different formulas, while taking into account both inspection accuracy and efficiency.
[0003] In existing technologies, panel defect detection often employs fixed-strategy models, which involve manually pre-setting single template matching rules, uniform threshold parameters, and fixed weight coefficients to identify anomalies in particle and pixel structures. Specifically, the system trains a judgment model based on a small number of offline labeled samples, solidifies core parameters such as judgment thresholds and feature weights, and directly calls the pre-set strategy to complete defect classification during actual detection.
[0004] However, such fixed strategy solutions have significant drawbacks: First, they have poor adaptability and cannot cope with process differences such as pixel structure rotation and geometric deformation in different batches of panels. A single template or threshold can easily misjudge normal process fluctuations as defects, resulting in a high false alarm rate. Second, they lack flexibility. When the production line changes the formula or adjusts the process parameters, the strategy parameters need to be manually recalibrated, which is cumbersome and time-consuming, seriously affecting the production line's response speed. Third, the judgment logic is rigid. Fixed weight coefficients cannot dynamically adjust the contribution ratio of indicators such as deviation degree and rarity according to the production line drift trend. They have weak ability to identify long-tail defects and it is difficult to balance the contradiction between the false alarm rate and the missed detection rate. Summary of the Invention
[0005] The main objective of this application is to provide a detection strategy generation method, apparatus, equipment, storage medium, and program product to solve the technical problems of existing detection strategies being rigid, having poor adaptability, and being unable to dynamically respond to process fluctuations and production line drift. It achieves the technical effects of flexible configuration of strategy parameters, accurate differentiation between normal process differences and true defects, and dynamic adaptation to the detection needs of multi-formula and multi-batch panels.
[0006] To achieve the above objectives, a first aspect of this application proposes a detection strategy generation method, comprising: acquiring a risk score and spatial coordinates of a detection panel, wherein the risk score is a quantitative indicator of the risk level of particle defects or pixel structure anomalies in the panel area, and the spatial coordinates are the positioning information of the detection panel in the production line detection process; constructing a two-dimensional to three-dimensional thermal mapping model based on the risk score and the spatial coordinates, and outputting spatial distribution information of panel defects and production line drift trend information through the thermal mapping model; inputting clustering statistics associated with the spatial distribution information, the production line drift trend information, representative sample index, and preset engineering rule text into a large language model to generate a target detection strategy; wherein the target detection strategy includes an interpretable classification standard, a structured threshold strategy, and an auditable rulebook, wherein the interpretable classification standard refers to the basis for defect level classification, the structured threshold strategy refers to the detection threshold configuration rules adapted to the process fluctuations of different production lines or batches, and the auditable rulebook refers to a standardized document recording the basis for strategy generation and execution logic.
[0007] According to the detection strategy generation method provided in this application, after generating the target detection strategy, the method further includes: distributing the structured threshold strategy to the detection system of the pilot object; collecting the detection data of the pilot object in real time and calculating strategy evaluation indicators, including the defect false alarm rate, the missed detection rate, and the production line adaptability; if the strategy evaluation indicators do not reach the preset evaluation threshold, switching the detection system of the pilot object to the historical detection strategy; if the strategy evaluation indicators reach the preset evaluation threshold, distributing the structured threshold strategy to the detection systems of all production lines.
[0008] According to the detection strategy generation method provided in this application, the risk score is determined by a deviation degree term, a rarity term, and a production line drift penalty term corresponding to the production line metadata; the deviation degree term refers to the minimum value of the Euclidean distances from the high-dimensional embedding vector to the centers of all golden clusters; the golden clusters include particle golden clusters formed by clustering qualified sample embedding vectors without particle defects, and pixel structure golden clusters constructed for each structure family after aggregating structurally equivalent qualified pixel sample embedding vectors into structure families through family clustering; the rarity term refers to the local density of the high-dimensional embedding vector in the feature space of surrounding qualified samples or the outlier probability corresponding to the high-dimensional embedding vector.
[0009] According to the detection strategy generation method provided in this application, the risk score is calculated using the following formula: ; in, Indicates risk score, This indicates normalization processing. Indicates the degree of deviation. Items indicating rarity This indicates the penalty for production line drift. , , This represents the weighting coefficient.
[0010] According to the detection strategy generation method provided in this application, the spatial coordinates include the three-dimensional coordinates of the panel defect on the panel and the production line associated coordinates of the panel.
[0011] This application also provides a detection strategy generation device, comprising the following modules: an acquisition module and a processing module; the acquisition module is used to acquire the risk score and spatial coordinates of the detection panel, wherein the risk score is a quantitative indicator of the risk level of particle defects or pixel structure anomalies in the panel area, and the spatial coordinates are the positioning information of the detection panel in the production line detection process; the processing module is used to construct a two-dimensional to three-dimensional thermal mapping model based on the risk score and the spatial coordinates, and output the spatial distribution information of panel defects and production line drift trend information through the thermal mapping model; the clustering statistical information associated with the spatial distribution information, the production line drift trend information, the representative sample index, and the preset engineering rule text during the panel detection process are input into a large language model to generate a target detection strategy; wherein, the target detection strategy includes an interpretable classification standard, a structured threshold strategy, and an auditable rulebook, wherein the interpretable classification standard refers to the basis for defect level classification, the structured threshold strategy refers to the detection threshold configuration rules adapted to the process fluctuations of different production lines or batches, and the auditable rulebook refers to a standardized document recording the basis for strategy generation and execution logic.
[0012] According to the detection strategy generation device provided in this application, after generating the target detection strategy, the processing module is used to distribute the structured threshold strategy to the detection system of the pilot object; collect the detection data of the pilot object in real time, and statistically analyze the strategy evaluation indicators, including the defect false alarm rate, the missed detection rate, and the production line adaptability; if the strategy evaluation indicators do not reach the preset evaluation threshold, the detection system of the pilot object is switched to the historical detection strategy; if the strategy evaluation indicators reach the preset evaluation threshold, the structured threshold strategy is distributed to the detection systems of all production lines.
[0013] According to the detection strategy generation device provided in this application, the risk score is determined by a deviation degree term, a rarity term, and a production line drift penalty term corresponding to the production line metadata; the deviation degree term refers to the minimum value of the Euclidean distances from the high-dimensional embedding vector to the centers of all golden clusters; the golden clusters include particle golden clusters formed by clustering qualified sample embedding vectors without particle defects, and pixel structure golden clusters constructed for each structure family after aggregating structurally equivalent qualified pixel sample embedding vectors into structure families through family clustering; the rarity term refers to the local density of the high-dimensional embedding vector in the feature space of surrounding qualified samples or the outlier probability corresponding to the high-dimensional embedding vector.
[0014] According to the detection strategy generation device provided in this application, the risk score is calculated using the following formula: ; in, Indicates risk score, This indicates normalization processing. Indicates the degree of deviation. Items indicating rarity This indicates the penalty for production line drift. , , This represents the weighting coefficient.
[0015] According to the detection strategy generation device provided in this application, the spatial coordinates include the three-dimensional coordinates of the panel defect on the panel and the production line associated coordinates of the panel.
[0016] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the detection strategy generation method as described above.
[0017] This application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the detection strategy generation method as described above.
[0018] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the detection strategy generation method as described above.
[0019] The technical solutions provided by the embodiments of this application may include the following beneficial effects: Because a two-dimensional to three-dimensional thermodynamic mapping model is constructed by risk scoring and spatial coordinates, it can output the spatial distribution information of panel defects and the drift trend information of production line. Therefore, it can accurately locate the defect concentration area and capture the fluctuation pattern of production line process, which solves the problem that traditional fixed strategies cannot perceive the dynamic changes of production line. By inputting the aforementioned spatial distribution information, production line drift trend information, clustering statistics, representative sample index, and preset engineering rule text into the large language model, a target detection strategy containing interpretable classification criteria, structured threshold strategies, and auditable rulebooks is generated. Therefore, there is no need to manually preset fixed templates and thresholds, which enables the detection strategy to dynamically adapt to process fluctuations of different production lines and different batches, greatly improving the flexibility and adaptability of strategy generation. Because the generated structured threshold strategy can be specifically adapted to process differences, the interpretable classification standards and auditable rulebooks ensure the transparency and traceability of the judgment logic. Therefore, it can effectively distinguish between normal process fluctuations and real defects, reduce false alarm rate and missed detection rate, and avoid the tedious operation of manually recalibrating parameters, thereby improving the production line response speed and operation and maintenance efficiency. Attached Figure Description
[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 A flowchart illustrating the detection strategy generation method provided in this application; Figure 2 This is a schematic diagram of the detection strategy generation device provided by the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0021] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present application.
[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0023] In this application, the terms "upper," "lower," "left," "right," "front," "rear," "top," "bottom," "inner," "outer," "middle," "vertical," "horizontal," "lateral," and "longitudinal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing this application and its embodiments, and are not intended to limit the indicated device, element, or component to having a specific orientation, or to be constructed and operated in a specific orientation.
[0024] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in some cases to indicate a certain dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.
[0025] Furthermore, the terms "installation," "setup," "equipped with," "connection," "linked," and "socketing" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.
[0026] This application describes some exemplary embodiments for illustrative purposes. It should be understood that this application may be implemented in other ways not specifically shown in the accompanying drawings.
[0027] like Figure 1 As shown, this application provides a detection strategy generation method, which can be applied to a detection strategy generation device. The detection strategy generation method may include steps S101-S103: S101, The detection strategy generation device acquires the risk score and spatial coordinates of the detection panel.
[0028] The aforementioned risk score is a quantitative indicator of the degree of risk of particle defects or abnormal pixel structure in the panel area, and the spatial coordinates are the positioning information of the detection panel in the production line inspection process.
[0029] Specifically, in the strategy iteration scenario of AOI online inspection, to ensure that the generated inspection strategy can both match the characteristics of the defect itself and adapt to the spatial distribution and process rules of the production line, the inspection strategy generation device can acquire two-dimensional basic data: risk score and spatial coordinates. The risk score can be calculated by combining the core feature dimensions of panel defect detection, while the spatial coordinates can determine the specific location of the defect in the panel and production line.
[0030] Optionally, the risk score is determined by a deviation term, a rarity term, and a production line drift penalty term corresponding to the production line metadata; the deviation term refers to the minimum Euclidean distance from the high-dimensional embedding vector to the center of all golden clusters; the golden clusters include particle golden clusters formed by clustering qualified sample embedding vectors without particle defects, and pixel structure golden clusters constructed for each structure family after aggregating structurally equivalent qualified pixel sample embedding vectors into structure families through family clustering; the rarity term refers to the local density of the high-dimensional embedding vector in the feature space of surrounding qualified samples or the outlier probability corresponding to the high-dimensional embedding vector.
[0031] Specifically, by quantitatively assessing defect risks across multiple dimensions, misjudgments or omissions caused by single-dimensional judgments can be avoided. Among these, the gold cluster is a normal feature benchmark constructed based on qualified samples; the particle gold cluster can focus on the normal features of particles without defects; the pixel structure gold cluster can adapt to the rotation and geometric differences of pixel structures through family clustering, ensuring the comprehensiveness and adaptability of the benchmark; the deviation term can measure the deviation of the sample under inspection from the normal benchmark; the rarity term can reflect the rarity of the sample in the normal feature distribution; and the production line drift penalty term can be associated with production line metadata to adapt to feature changes caused by process fluctuations.
[0032] Optionally, the risk score is calculated using the following formula: ; in, Indicates risk score, This indicates normalization processing. Indicates the degree of deviation. Items indicating rarity This indicates the penalty for production line drift. , , This represents the weighting coefficient.
[0033] It should be noted that weighted normalization enables the coordinated quantification of multi-dimensional indicators, ensuring the rationality and comparability of risk scores. Normalization transforms each dimension indicator into a value within the 0-1 range, avoiding the impact of differences in units of measurement on the assessment results; weighting coefficients... , , It can be dynamically adjusted according to production line quality requirements and process characteristics. For example, when the production line has the highest priority for identifying true defects, it can increase... The weight can be increased when process drift is frequent. The percentage.
[0034] Optionally, the spatial coordinates include the three-dimensional coordinates of the panel defect on the panel and the production line-related coordinates of the panel.
[0035] Specifically, the three-dimensional coordinates on the panel can be based on the panel itself, for example, the three-dimensional coordinates are: x=150mm, y=100mm, z=0.1mm, where the z-axis represents the position of the defect in the thickness direction of the panel; the production line-related coordinates are associated with the positioning information of the production line inspection process, such as the inspection machine number, the transfer position of the panel in the production line, and the inspection station. Through dual coordinates, the specific location of the defect on the panel and its corresponding production scenario can be accurately located, providing a basis for analyzing the correlation between defect concentration areas and production line processes.
[0036] S102. The detection strategy generation device constructs a two-dimensional to three-dimensional thermal mapping model based on the risk score and the spatial coordinates, and outputs the spatial distribution information of panel defects and production line drift trend information through the thermal mapping model.
[0037] Specifically, the heat map model can integrate the risk scores corresponding to each spatial coordinate through the aggregation function H(x, y) = Agg(R(e), local statistics). Here, H(x, y) represents the final output three-dimensional heat map value, which corresponds to the comprehensive risk of the planar position (x, y) on the panel. That is, x and y are the two-dimensional coordinates on the panel, and H(x, y) is the risk height of this point, forming the z-axis of the three-dimensional heat map. Agg() is the aggregation calculation function. R(e) is the risk score corresponding to the panel position (x, y). Local statistics refers to the statistical results of all risk scores within a preset area around (x, y), such as the average risk score, the highest risk score, and the percentage of defects in the preset area.
[0038] The spatial distribution information output by the thermal mapping model is presented in the form of a three-dimensional heat map. The color depth or height difference can intuitively reflect the distribution density and severity of defect risks. For example, the central area of the panel shows a dark red high-altitude thermal distribution, indicating that this area is a high-incidence area of particle defects. Production line drift trend information can be obtained by comparing heat map changes according to time dimension, machine dimension, and process formula dimension. For example, if the heat map of 5 consecutive batches shows that the risk distribution gradually shifts along the production line transmission direction (x-axis) and the drift exceeds the preset threshold, it can be determined that there is a systematic process drift.
[0039] For example, if the heat map of a certain batch of panels shows that the average risk score of the edge area is 0.75, which is significantly higher than the historical batches of 0.4, and the corresponding production line associated coordinates are concentrated in "machine 4-transfer section 3", it can be determined that the process parameters of the transfer section may be abnormal, thus providing direction for subsequent strategy adjustments.
[0040] It should be noted that the above steps can transform abstract inspection data into intuitive visualization results, which not only makes it easier for staff to quickly locate defect concentration areas and process drift sources, but also provides a concrete space and trend basis for generating targeted strategies for large language models, thereby improving the traceability of the inspection process and the efficiency of problem investigation.
[0041] S103. The detection strategy generation device inputs the clustering statistics associated with the spatial distribution information, the production line drift trend information, the representative sample index, and the preset engineering rule text into the large language model during the panel detection process to generate a target detection strategy.
[0042] The target detection strategy includes an interpretable classification standard, a structured threshold strategy, and an auditable rulebook. The interpretable classification standard refers to the basis for classifying defect levels. The structured threshold strategy refers to the detection threshold configuration rules that adapt to the process fluctuations of different production lines or batches. The auditable rulebook refers to a standardized document that records the basis for strategy generation and execution logic.
[0043] Specifically, cluster statistics are quantitative data related to spatial distribution information, including the number of golden clusters in each region, the covariance of the sample distribution within the cluster, the proportion and spatial distribution of abnormal clusters, and the tolerance boundary adaptation of different golden clusters. For example, the deviation statistics of particle golden clusters corresponding to defect concentration areas, and the clustering distribution characteristics of pixel structure families in a certain spatial region. Production line drift trend information includes drift direction, drift amplitude, spatial region corresponding to the drift, and related production line metadata, etc. The representative sample index points to a typical sample in the feature space, including high-risk defect samples, interference samples, drift samples, normal samples, etc. Each index can be associated with the corresponding risk score, spatial coordinates, embedded vector features and re-judgment annotation information, which makes it easier for the large language model to associate specific cases for reasoning. The pre-defined engineering rules text is a set of standardized requirements based on the actual needs of the production line, including quality standards, defect judgment specifications, re-judgment requirements, and process constraints.
[0044] After receiving the above input data, the large language model can generate interpretable classification criteria and object detection strategies by deeply understanding the correlation between data distribution patterns and engineering rules.
[0045] Optionally, the generated target detection strategy can be validated using schema verification, range verification, and replay verification. After passing these verifications, it is written to the strategy version repository. The generated strategy can be managed by version and applied through methods such as shadow mode, canary release, comparative replay evaluation, and version rollback. For example, during production line inspection, if process drift is detected, such as equipment parameter deviations or material changes leading to altered defect characteristics, the large language model will automatically output a "solution to adjust the detection strategy." Once this solution is verified to be correct, it will be officially implemented on the production line.
[0046] Optionally, after generating the target detection strategy, the detection strategy generation device can distribute the structured threshold strategy to the detection system of the pilot object; collect the detection data of the pilot object in real time, and statistically analyze the strategy evaluation indicators, including the defect false alarm rate, the missed detection rate, and the production line adaptability; if the strategy evaluation indicators do not reach the preset evaluation threshold, the detection system of the pilot object is switched to the historical detection strategy; if the strategy evaluation indicators reach the preset evaluation threshold, the structured threshold strategy is distributed to the detection systems of all production lines.
[0047] Optionally, the detection strategy generation device can also package key information related to the detection results into a traceable evidence package, which may include representative samples, feature spatial location, risk score, rule trigger chain and version information.
[0048] In this embodiment, by constructing a two-dimensional to three-dimensional thermal mapping model using risk scoring and spatial coordinates, the spatial distribution information of panel defects and production line drift trend information can be output. Therefore, it can accurately locate defect concentration areas and capture the fluctuation patterns of production line processes, solving the problem that traditional fixed strategies cannot perceive dynamic changes in the production line. Since the above spatial distribution information, production line drift trend information, clustering statistics, representative sample indexes, and preset engineering rule text are input into the large language model to generate a target detection strategy containing interpretable classification standards, structured threshold strategies, and auditable rule books, there is no need to manually preset fixed templates and thresholds. This achieves dynamic adaptation of the detection strategy to process fluctuations of different production lines and batches, greatly improving the flexibility and adaptability of strategy generation. Since the generated structured threshold strategy can be specifically adapted to process differences, and the interpretable classification standards and auditable rule books ensure the transparency and traceability of the judgment logic, it can effectively distinguish between normal process fluctuations and true defects, reduce false alarm rates and false negative rates, and avoid the tedious operation of manually recalibrating parameters, thereby improving the production line response speed and operation and maintenance efficiency.
[0049] The foregoing mainly describes the solutions provided by the embodiments of this application from a methodological perspective. To achieve the above functions, it includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, the embodiments of this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0050] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0051] The detection strategy generation method provided in this application can be executed by a detection strategy generation device or a control module within that device for generating detection strategies. This application uses the execution of the detection strategy generation method by a detection strategy generation device as an example to illustrate the detection strategy generation device provided in this application.
[0052] It should be noted that the embodiments of this application can divide the detection strategy generation device into functional modules according to the above method examples. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one processing module. The integrated modules can be implemented in hardware or as software functional modules. Optionally, the module division in the embodiments of this application is illustrative and is only a logical functional division; other division methods may be used in actual implementation.
[0053] like Figure 2 As shown in the figure, this application embodiment provides a detection strategy generation device 200. The detection strategy generation device 200 includes: an acquisition module 201 and a processing module 202.
[0054] The acquisition module 201 is used to acquire the risk score and spatial coordinates of the detection panel. The risk score is a quantitative indicator of the degree of risk of particle defects or pixel structure abnormalities in the panel area. The spatial coordinates are the positioning information of the detection panel in the production line inspection process. The processing module 202 is used to construct a two-dimensional to three-dimensional thermal mapping model based on the risk score and the spatial coordinates, and output the spatial distribution information of panel defects and production line drift trend information through the thermal mapping model; it inputs the clustering statistics information associated with the spatial distribution information, the production line drift trend information, the representative sample index, and the preset engineering rule text into the large language model to generate a target detection strategy; wherein, the target detection strategy includes an interpretable classification standard, a structured threshold strategy, and an auditable rule book, the interpretable classification standard refers to the basis for classifying defect levels, the structured threshold strategy refers to the detection threshold configuration rules adapted to the process fluctuations of different production lines or batches, and the auditable rule book refers to the standardized document that records the basis for strategy generation and execution logic.
[0055] Optionally, after generating the target detection strategy, the processing module 202 is used to distribute the structured threshold strategy to the detection system of the pilot object; collect the detection data of the pilot object in real time, and statistically analyze the strategy evaluation indicators, including the defect false alarm rate, the missed detection rate, and the production line adaptability; if the strategy evaluation indicators do not reach the preset evaluation threshold, the detection system of the pilot object is switched to the historical detection strategy; if the strategy evaluation indicators reach the preset evaluation threshold, the structured threshold strategy is distributed to the detection systems of all production lines.
[0056] Optionally, the risk score is determined by a deviation term, a rarity term, and a production line drift penalty term corresponding to the production line metadata; the deviation term refers to the minimum Euclidean distance from the high-dimensional embedding vector to the center of all golden clusters; the golden clusters include particle golden clusters formed by clustering qualified sample embedding vectors without particle defects, and pixel structure golden clusters constructed for each structure family after aggregating structurally equivalent qualified pixel sample embedding vectors into structure families through family clustering; the rarity term refers to the local density of the high-dimensional embedding vector in the feature space of surrounding qualified samples or the outlier probability corresponding to the high-dimensional embedding vector.
[0057] Optionally, the risk score is calculated using the following formula: ; in, Indicates risk score, This indicates normalization processing. Indicates the degree of deviation. Items indicating rarity This indicates the penalty for production line drift. , , This represents the weighting coefficient.
[0058] Optionally, the spatial coordinates include the three-dimensional coordinates of the panel defect on the panel and the production line-related coordinates of the panel.
[0059] In this embodiment, by constructing a two-dimensional to three-dimensional thermal mapping model using risk scoring and spatial coordinates, the spatial distribution information of panel defects and production line drift trend information can be output. Therefore, it can accurately locate defect concentration areas and capture the fluctuation patterns of production line processes, solving the problem that traditional fixed strategies cannot perceive dynamic changes in the production line. Since the above spatial distribution information, production line drift trend information, clustering statistics, representative sample indexes, and preset engineering rule text are input into the large language model to generate a target detection strategy containing interpretable classification standards, structured threshold strategies, and auditable rule books, there is no need to manually preset fixed templates and thresholds. This achieves dynamic adaptation of the detection strategy to process fluctuations of different production lines and batches, greatly improving the flexibility and adaptability of strategy generation. Since the generated structured threshold strategy can be specifically adapted to process differences, and the interpretable classification standards and auditable rule books ensure the transparency and traceability of the judgment logic, it can effectively distinguish between normal process fluctuations and true defects, reduce false alarm rates and false negative rates, and avoid the tedious operation of manually recalibrating parameters, thereby improving the production line response speed and operation and maintenance efficiency.
[0060] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communications bus 340. The processor 310 can call logic instructions in the memory 330 to execute a detection strategy generation method. This method includes: acquiring a risk score and spatial coordinates of a detection panel, wherein the risk score is a quantitative indicator of the risk level of particle defects or pixel structure anomalies in the panel area, and the spatial coordinates are the positioning information of the detection panel in the production line detection process; constructing a two-dimensional to three-dimensional thermal mapping model based on the risk score and the spatial coordinates, and outputting spatial distribution information of panel defects and production line drift trend information through the thermal mapping model; inputting clustering statistics associated with the spatial distribution information, the production line drift trend information, representative sample indexes, and preset engineering rule texts during panel detection into a large language model to generate a target detection strategy; wherein the target detection strategy includes an interpretable classification standard, a structured threshold strategy, and an auditable rulebook, wherein the interpretable classification standard refers to the basis for defect level classification, the structured threshold strategy refers to detection threshold configuration rules adapted to different production lines or batch process fluctuations, and the auditable rulebook refers to a standardized document recording the basis for strategy generation and execution logic.
[0061] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0062] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the detection strategy generation method provided by the above methods. The method includes: acquiring a risk score and spatial coordinates of a detection panel, wherein the risk score is a quantitative indicator of the degree of risk of particle defects or pixel structure abnormalities in the panel area, and the spatial coordinates are the positioning information of the detection panel in the production line detection process; constructing a two-dimensional to three-dimensional thermal mapping model based on the risk score and the spatial coordinates, and using the thermal... The force mapping model outputs spatial distribution information of panel defects and production line drift trend information. Clustering statistics associated with the spatial distribution information, production line drift trend information, representative sample indexes, and preset engineering rule texts are input into a large language model to generate a target detection strategy. This target detection strategy includes interpretable classification criteria, structured threshold strategies, and auditable rulebooks. The interpretable classification criteria refer to the basis for defect level classification; the structured threshold strategies refer to detection threshold configuration rules adapted to process fluctuations in different production lines or batches; and the auditable rulebooks refer to standardized documents recording the basis for strategy generation and execution logic.
[0063] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a detection strategy generation method provided by the above methods. The method includes: acquiring a risk score and spatial coordinates of a detection panel, wherein the risk score is a quantitative indicator of the degree of risk of particle defects or pixel structure anomalies in the panel area, and the spatial coordinates are the positioning information of the detection panel in the production line detection process; constructing a two-dimensional to three-dimensional thermal mapping model based on the risk score and the spatial coordinates, and outputting spatial distribution information of panel defects and production line drift trend information through the thermal mapping model; inputting clustering statistical information associated with the spatial distribution information, the production line drift trend information, representative sample index, and preset engineering rule text into a large language model to generate a target detection strategy; wherein the target detection strategy includes an interpretable classification standard, a structured threshold strategy, and an auditable rulebook, wherein the interpretable classification standard refers to the basis for defect level classification, the structured threshold strategy refers to the detection threshold configuration rules adapted to the process fluctuations of different production lines or batches, and the auditable rulebook refers to a standardized document recording the basis for strategy generation and execution logic.
[0064] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0065] Obviously, those skilled in the art should understand that the various units or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps into a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.
[0066] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A detection strategy generation method, characterized in that, include: The risk score and spatial coordinates of the detection panel are obtained. The risk score is a quantitative indicator of the degree of risk of particle defects or pixel structure abnormalities in the panel area. The spatial coordinates are the positioning information of the detection panel in the production line inspection process. Based on the risk score and the spatial coordinates, a two-dimensional to three-dimensional thermal mapping model is constructed, and the spatial distribution information of panel defects and production line drift trend information are output through the thermal mapping model. The clustering statistics associated with the spatial distribution information, the production line drift trend information, the representative sample index, and the preset engineering rule text are input into the large language model during the panel detection process to generate a target detection strategy. The target detection strategy includes an interpretable classification standard, a structured threshold strategy, and an auditable rulebook. The interpretable classification standard refers to the basis for classifying defect levels. The structured threshold strategy refers to the detection threshold configuration rules that adapt to the process fluctuations of different production lines or batches. The auditable rulebook refers to a standardized document that records the basis for strategy generation and execution logic.
2. The detection strategy generation method according to claim 1, characterized in that, After generating the target detection strategy, the method further includes: The structured threshold strategy is then distributed to the detection system of the pilot targets; The detection data of the pilot objects are collected in real time, and the strategy evaluation indicators are statistically analyzed. The strategy evaluation indicators include the defect false alarm rate, the missed detection rate, and the production line adaptability. If the strategy evaluation index does not reach the preset evaluation threshold, the detection system of the pilot object will be switched to the historical detection strategy. If the strategy evaluation index reaches the preset evaluation threshold, the structured threshold strategy will be fully distributed to the detection systems of all production lines.
3. The detection strategy generation method according to claim 1, characterized in that, The risk score is determined by a deviation term, a rarity term, and a production line drift penalty term corresponding to the production line metadata. The deviation term refers to the minimum Euclidean distance from the high-dimensional embedding vector to the center of all golden clusters. The golden clusters include particle golden clusters formed by clustering qualified sample embedding vectors without particle defects, and pixel structure golden clusters constructed for each structure family after aggregating structurally equivalent qualified pixel sample embedding vectors into structure families through family clustering. The rarity term refers to the local density of the high-dimensional embedding vector in the feature space of surrounding qualified samples or the outlier probability corresponding to the high-dimensional embedding vector.
4. The detection strategy generation method according to claim 3, characterized in that, The formula for calculating the risk score is as follows: ; in, Indicates risk score, This indicates normalization processing. Indicates the degree of deviation. Items indicating rarity This indicates the penalty for production line drift. , , This represents the weighting coefficient.
5. The detection strategy generation method according to claim 1, characterized in that, The spatial coordinates include the three-dimensional coordinates of the panel defects on the panel, as well as the production line-related coordinates of the panel.
6. A detection strategy generation device, characterized in that, include: Acquisition module and processing module; The acquisition module is used to acquire the risk score and spatial coordinates of the detection panel. The risk score is a quantitative indicator of the degree of risk of particle defects or pixel structure abnormalities in the panel area, and the spatial coordinates are the positioning information of the detection panel in the production line inspection process. The processing module is used to construct a two-dimensional to three-dimensional thermal mapping model based on the risk score and the spatial coordinates, and output the spatial distribution information of panel defects and production line drift trend information through the thermal mapping model; and input the clustering statistics information associated with the spatial distribution information, the production line drift trend information, the representative sample index and the preset engineering rule text into the large language model to generate a target detection strategy. The target detection strategy includes an interpretable classification standard, a structured threshold strategy, and an auditable rulebook. The interpretable classification standard refers to the basis for classifying defect levels. The structured threshold strategy refers to the detection threshold configuration rules that adapt to the process fluctuations of different production lines or batches. The auditable rulebook refers to a standardized document that records the basis for strategy generation and execution logic.
7. The detection strategy generation apparatus according to claim 6, characterized in that, After generating the target detection strategy, the processing module is further configured to distribute the structured threshold strategy to the detection system of the pilot object; collect the detection data of the pilot object in real time, and statistically analyze the strategy evaluation indicators, including the false positive rate, the false negative rate, and the production line adaptability; if the strategy evaluation indicators do not reach the preset evaluation threshold, the detection system of the pilot object is switched to the historical detection strategy; if the strategy evaluation indicators reach the preset evaluation threshold, the structured threshold strategy is distributed to the detection systems of all production lines.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the detection strategy generation method as described in any one of claims 1 to 5.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the detection strategy generation method as described in any one of claims 1 to 5.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the detection strategy generation method as described in any one of claims 1 to 5.