Agent task scheduling model and scheduling method of automatic optical defect detection equipment

Abstract

The present disclosure provides an agent task scheduling model and method of an automatic optical defect detection device, and relates to the field of defect detection, which comprises: acquiring a detected image of a detected device; acquiring a defect detection task requirement; identifying a plurality of detection tasks based on the task requirement; scheduling a task execution order of the plurality of detection tasks, the task execution order comprising a plurality of task sets executed in series, each task set comprising a plurality of detection tasks executed in parallel; configuring an expert model recipe for the detected image and each task set from a plurality of expert models, the expert model recipe comprising at least part of the expert models selected from the plurality of expert models; and performing defect detection on the detected image based on the expert model recipe, so as to sequentially execute the plurality of task sets in series according to the task execution order, and execute the plurality of detection tasks in a task set in parallel when the task set is executed. The detection rate of the automatic optical defect detection device when executing multiple detection tasks is improved.

Description

Technical Field

[0001] This disclosure relates to the field of defect detection technology, specifically to an intelligent agent task orchestration model and orchestration method for an automated optical defect detection device. Background Technology

[0002] In the semiconductor manufacturing process, automated optical defect inspection equipment is typically used to automatically detect manufacturing defects in semiconductors and other devices. However, the defect detection models of these automated optical defect inspection equipment have low efficiency in performing defect detection tasks on the input images, making it difficult to meet the time requirements for defect detection. Summary of the Invention

[0003] The main objective of this disclosure is to provide an intelligent agent task orchestration model and method for an automated optical defect detection device, so as to improve the detection rate of the automated optical defect detection device when performing multiple detection tasks.

[0004] To achieve the above objectives, the first aspect of this disclosure provides a defect detection management model for an automated optical defect detection device, the agent task orchestration model comprising: The data access module is used to acquire the image of the device under test acquired by the image acquisition device; The task acquisition module is used to acquire the task requirements for defect detection of the image being detected. The task identification module is used to identify multiple detection tasks based on the task requirements. The task orchestration module is used to orchestrate the task execution order of the multiple detection tasks, wherein the task execution order includes multiple task sets executed serially, and each task set includes multiple detection tasks executed in parallel. A router module is configured to configure expert model recipes for the image being inspected and each of the task sets from a plurality of expert models included in the automated optical defect detection device, the expert model recipes including at least a subset of expert models selected from the plurality of expert models; An actuator module is used to perform defect detection on the image to be detected based on the expert model recipe, so as to execute the multiple task sets sequentially according to the task execution order, and execute multiple detection tasks in the task set in parallel while executing one of the task sets.

[0005] In some embodiments of this disclosure, the agent task orchestration model further includes: The task orchestration evaluation module is used to obtain the execution results of the executor module for each detection task, analyze the execution results to obtain the parallel execution interference between multiple detection tasks in each task set, and send the parallel execution interference to the task orchestration module. The task orchestration module dynamically adjusts the detection tasks included in each task set in the task execution order based on the parallel execution interference.

[0006] In some embodiments of this disclosure, the task orchestration evaluation module analyzes the parallel execution interference between multiple detection tasks in each task set based on the execution results, including: Obtain reference detection result data for each detection task when performing each detection task individually on the image being detected; Based on the execution results, obtain the actual detection result data corresponding to each detection task when each detection task is executed in parallel with other detection tasks in its current task set; Based on the reference detection result data and the actual detection result data for each detection task, the parallel execution interference between the multiple detection tasks in each task set is obtained.

[0007] In some embodiments of this disclosure, the task orchestration evaluation module includes a scoring module; The scoring module scores the reference detection result data corresponding to each detection task to obtain a reference score value for each detection task. The scoring module also scores the actual detection result data corresponding to each detection task to obtain the actual score value corresponding to each detection task; When the reference score value corresponding to the detection task is greater than the actual score value corresponding to the detection task, the task orchestration and evaluation module determines that, in the current task execution order, the detection task has parallel execution interference with other detection tasks in the task set to which it currently belongs.

[0008] In some embodiments of this disclosure, the task orchestration module dynamically adjusts the detection tasks included in each task set in the task execution order based on the parallel execution interference, including: When the task orchestration evaluation module determines that the detection task has parallel execution interference with other detection tasks in its current task set, the detection task is adjusted from the current task set to another task set.

[0009] In some embodiments of this disclosure, when the reference score value corresponding to the detection task is less than or equal to the actual score value corresponding to the detection task, the task orchestration evaluation module determines that in the current task execution order, the detection task does not have parallel execution interference with other detection tasks in the task set to which it currently belongs.

[0010] In some embodiments of this disclosure, the task orchestration module dynamically adjusts the detection tasks included in each task set in the task execution order based on the parallel execution interference, including: When the task orchestration evaluation module determines that the detection task does not interfere with the parallel execution of other detection tasks in its current task set, the detection task remains unchanged in the current task set.

[0011] In some embodiments of this disclosure, the task orchestration module adopts a canary release strategy, which dynamically adjusts the task execution order obtained after the detection tasks included in each task set in the task execution order, and applies it to the defect detection process of the detected image acquired by the data access module thereafter.

[0012] In some embodiments of this disclosure, the agent task orchestration model further includes: The caching and idempotency module is used to acquire and store intermediate result data generated by the actuator module for each detection task during the defect detection of the image being detected by the actuator module, and send it to the task orchestration and evaluation module. The task orchestration and evaluation module obtains the execution result of each detection task in the current task execution order based on the intermediate result data.

[0013] The second aspect of this disclosure provides an automatic optical defect detection device, which includes an agent task orchestration model of any of the automatic optical defect detection devices provided in the first aspect of this disclosure.

[0014] A third aspect of this disclosure provides an agent task orchestration method for an automated optical defect detection device, the agent task orchestration method comprising: Acquire the image of the device under test captured by the image acquisition device; Obtain the task requirements for defect detection of the image being detected; Multiple detection tasks were identified based on the aforementioned task requirements; The task execution order of the multiple detection tasks is arranged, wherein the task execution order includes multiple task sets executed serially, and each task set includes multiple detection tasks executed in parallel; The automated optical defect detection device configures an expert model recipe for the image being inspected and for each of the task sets from a plurality of expert models, the expert model recipe comprising at least a subset of the expert models selected from the plurality of expert models; Based on the expert model formula, defect detection is performed on the detected image, and the multiple task sets are executed sequentially according to the task execution order, and multiple detection tasks in the task set are executed in parallel while executing one of the task sets; The execution results of each detection task are obtained, and the parallel execution interference between multiple detection tasks in each task set is analyzed based on the execution results. Based on the parallel execution interference, the detection tasks included in each task set in the task execution order are dynamically adjusted.

[0015] The intelligent agent task orchestration model and method of the automatic optical defect detection device provided in this disclosure embodiment identify multiple detection tasks based on the task requirements by a task identification module. The task orchestration module then orchestrates the execution order of the multiple detection tasks, which includes multiple task sets executed serially, each task set including multiple detection tasks executed in parallel. A router module then configures an expert model recipe for each task set from multiple expert models. Subsequently, an actuator module performs defect detection on the image to be detected based on the expert model recipe. This is done by sequentially executing the multiple task sets according to the task execution order and executing multiple detection tasks in parallel within each task set. This combination of serial and parallel execution improves the efficiency of defect detection task execution on the input image to be detected, thereby meeting the defect detection time requirements. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of this disclosure, the accompanying drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A schematic block diagram of a defect detection management model for an automated optical defect detection device provided in an embodiment of this disclosure; Figure 2 A flowchart illustrating a defect detection management model for an automated optical defect detection device provided in an embodiment of this disclosure; Figure 3A schematic block diagram of a defect detection management model for an automated optical defect detection device provided in another embodiment of this disclosure; Figure 4 This is a flowchart illustrating the process of determining the parallel execution interference between multiple detection tasks in each task set, as provided in an embodiment of this disclosure. Figure 5 A flowchart illustrating the dynamic adjustment of task execution order provided in an embodiment of this disclosure; Figure 6 A schematic block diagram of a defect detection management model for an automated optical defect detection device provided in another embodiment of this disclosure; Figure 7 A schematic flowchart illustrating a defect detection management method for an automated optical defect detection device provided in an embodiment of this disclosure; Figure 8 A flowchart illustrating a defect detection management method for an automated optical defect detection device provided in another embodiment of this disclosure; Figure 9 A flowchart illustrating a defect detection management method for an automated optical defect detection device provided in another embodiment of this disclosure; Figure 10 A flowchart illustrating a defect detection management method for an automated optical defect detection device provided in another embodiment of this disclosure; Figure 11 This diagram illustrates a comparison of the effects of a model on defect detection management and related technologies of an automated optical defect detection device provided in an embodiment of this disclosure. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present disclosure, the technical solutions of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present disclosure, and not all embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present disclosure.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure 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 disclosure 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.

[0020] In this disclosure, the terms “upper,” “lower,” “left,” “right,” “front,” “rear,” “top,” “bottom,” “inner,” “outer,” and “middle,” etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings. These terms are primarily for the purpose of better describing this disclosure and its embodiments, and are not intended to limit the indicated devices, elements, or components to having a specific orientation, or to be constructed and operated in a specific orientation.

[0021] 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 dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this disclosure according to the specific circumstances.

[0022] Furthermore, the terms "set up," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral structure; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection via an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of these terms in this disclosure according to the specific circumstances.

[0023] It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] Example 1 To address at least some of the aforementioned problems, embodiments of this disclosure provide a defect detection management model for an automated optical defect detection device, with reference to... Figure 1 and Figure 2The intelligent agent task orchestration model includes: a data access module, a task acquisition module, a task recognition module, a task orchestration module, a router module, and an actuator module. Specifically, the data access module acquires images of the device under test acquired by the image acquisition device; the task acquisition module acquires task requirements for defect detection on the images; the task recognition module identifies multiple detection tasks based on the task requirements; the task orchestration module orchestrates the execution order of multiple detection tasks, where the execution order includes multiple task sets executed sequentially, each task set including multiple detection tasks executed in parallel; the router module configures expert model recipes for the images under test and each task set from multiple expert models included in the automated optical defect detection equipment, where each expert model recipe includes at least a portion of the expert models selected from the multiple expert models; and the actuator module performs defect detection on the images under test based on the expert model recipes, sequentially executing multiple task sets according to the task execution order, and executing multiple detection tasks in parallel within a task set while executing the task set.

[0025] In the above scheme, the task identification module identifies multiple detection tasks based on task requirements, and the task orchestration module arranges the execution order of the multiple detection tasks. This task execution order includes multiple task sets executed serially, and each task set includes multiple detection tasks executed in parallel. Then, the router module configures expert model recipes for each task set from multiple expert models. After that, the executor module performs defect detection on the image to be detected based on the expert model recipes. This is to execute multiple task sets serially according to the task execution order, and execute multiple detection tasks in the task set in parallel while executing a task set. Thus, by combining serial execution and parallel execution, the efficiency of defect detection tasks on the input image to be detected is improved to meet the defect detection time requirements.

[0026] The following is in conjunction with the appendix Figures 1 to 11 The agent task orchestration model of the automatic optical defect detection device disclosed herein is described in detail.

[0027] The device being inspected can be, but is not limited to, semiconductor devices such as display panels and wafers. The aforementioned inspection tasks can be, but are not limited to, any defect inspection task that can be visually inspected, such as, conductive line defects, contact hole defects, void defects, flatness defects, groove defects, etc.

[0028] refer to Figure 8 and Figure 9The data access module can retrieve the detected images and metadata and perform integrity verification. Then, it identifies the task scenario and target, generating a structured task description, which can specify the task execution order. Next, the router module selects expert model recipes from a multi-expert model set to generate an execution graph (parallel / serial).

[0029] For example, refer to Figure 2 and Figure 3 The intelligent agent task orchestration model may further include a task orchestration evaluation module. This module obtains the execution results of each detection task from the executor module, analyzes the results to determine the parallel execution interference between multiple detection tasks in each task set, and sends this interference information to the task orchestration module. The task orchestration module can also dynamically adjust the detection tasks included in each task set based on the parallel execution interference. Because different detection tasks have different execution conditions (e.g., different light source parameters), parallel execution of multiple detection tasks may lead to mutual interference due to the limitations of their execution conditions, thus reducing the accuracy of the detection task execution results. The embodiments described above avoid the reduced accuracy of the detection task execution results caused by interference inherent in the execution methods of different detection tasks when they are divided into the same task set for parallel execution.

[0030] There are several ways to obtain information about the parallel execution interference between multiple detection tasks in each task set based on the analysis of execution results in the task orchestration evaluation module. Some of these methods are illustrated below.

[0031] For example, refer to Figure 4 The task orchestration and evaluation module analyzes the execution results to obtain the parallel execution interference between multiple detection tasks in each task set. This can include: obtaining reference detection result data for each detection task when each detection task is executed individually on the image being detected; obtaining actual detection result data for each detection task when it is executed in parallel with other detection tasks in its current task set based on the execution results; and obtaining the parallel execution interference between multiple detection tasks in each task set based on the reference detection result data and the actual detection result data for each detection task.

[0032] Specifically, reference detection result data for each detection task can be obtained in advance by performing the detection on a portion of the detected images separately for each detection task. Since the reference detection result data is the detection result obtained by performing the detection on a single detection task separately, it is not affected by other detection tasks and can be used as a basis for evaluating whether there is interference when the detection task is executed in parallel with other detection tasks in its current task set.

[0033] In the process of obtaining the actual detection result data of each detection task when it is executed in parallel with other detection tasks in its current task set, the task orchestration and evaluation module obtains the actual detection result data of each detection task when it is executed in parallel with other detection tasks in its current task set. Since the execution result is based on the current task execution order, it can obtain the actual detection result data of each detection task when it is executed in parallel with other detection tasks in its current task set. The actual detection result data can reflect whether each detection task is interfered with by other detection tasks in the same task set that are executed in parallel. Thus, it can accurately obtain the interference between multiple detection tasks in each task set during parallel execution, and provide an accurate basis for subsequent dynamic adjustment of the execution task order.

[0034] For example, refer to Figure 5 and Figure 6 The task orchestration and evaluation module may include a scoring module. The scoring module scores the reference detection result data for each detection task to obtain a reference score value for each task. The scoring module also scores the actual detection result data for each detection task to obtain an actual score value for each task. When the reference score value for a detection task is greater than the actual score value, the task orchestration and evaluation module determines that, in the current task execution order, the detection task interferes with the parallel execution of other detection tasks in its current task set. This allows for accurate quantitative identification of whether each detection task interferes with the parallel execution of other detection tasks in its current task set.

[0035] The scoring module specifically scores the reference detection result data for each detection task to obtain a reference score value. This is achieved by comparing the reference detection result data for that task with the actual defect data of the inspected image. For example, the closer the reference detection result data is to the actual defect data, the more accurately the reference detection result data detects the defects of the inspected device corresponding to the inspected image for that detection task, thus allowing for a higher reference score value. Conversely, the greater the difference between the reference detection result data and the actual defect data, the less accurately the reference detection result data detects the defects of the inspected device corresponding to the inspected image for that detection task, thus allowing for a lower reference score value. This results in a positive correlation between the reference score value and the accuracy of the detection task.

[0036] Similarly, when the scoring module scores the actual detection result data for each detection task to obtain the actual score value, it can also compare the actual detection result data for that task with the actual defect data of the inspected image to obtain the actual score value. For example, the closer the actual detection result data is to the actual defect data, the more accurately the actual detection result data detects the defect of the inspected device corresponding to the inspected image for that detection task, thus allowing for a higher actual score value. Conversely, the greater the difference between the actual detection result data and the actual defect data, the less accurately the actual detection result data detects the defect of the inspected device corresponding to the inspected image for that detection task, thus allowing for a lower actual score value. This results in a positive correlation between the actual score value and the accuracy of the detection task.

[0037] Therefore, when the reference score value corresponding to the detection task is greater than the actual score value corresponding to the detection task, it indicates that in the current task execution order, the detection task does indeed have parallel execution interference with other detection tasks in its current task set. Thus, it can accurately identify whether each detection task has parallel execution interference with other detection tasks in its current task set through quantitative methods.

[0038] For example, refer to Figure 5 and Figure 6The task orchestration module dynamically adjusts the detection tasks included in each task set in the task execution order based on parallel execution interference. This can include: when the task orchestration evaluation module determines that a detection task is interfering with the parallel execution of other detection tasks in its current task set, the detection task is moved from the current task set to another task set. This avoids the interference caused by the parallel execution of the detection task with other detection tasks in its current task set, thereby improving the defect detection accuracy for that detection task.

[0039] For example, refer to Figure 5 and Figure 6 The task orchestration and evaluation module determines that, in the current task execution order, there is no parallel execution interference between the detection task and other detection tasks in its current task set when the reference score value corresponding to the detection task is less than or equal to the actual score value. In other words, if the reference score value corresponding to the detection task is less than or equal to the actual score value, it indicates that in the current task execution order, there is indeed no parallel execution interference between the detection task and other detection tasks in its current task set. This allows for accurate quantitative identification of whether each detection task experiences parallel execution interference with other detection tasks in its current task set.

[0040] For example, refer to Figure 5 and Figure 6 The task orchestration module dynamically adjusts the detection tasks included in each task set in the task execution order based on the interference of parallel execution. This can include: keeping the detection task unchanged in the current task set when the task orchestration evaluation module determines that there is no parallel execution interference between the detection task and other detection tasks in its current task set. Since the aforementioned method can accurately identify whether each detection task has parallel execution interference with other detection tasks in its current task set through quantitative means, when the judgment result is that there is no parallel execution interference between the detection task and other detection tasks in its current task set, it means that the accuracy of the detection task will not be affected by interference when the task set containing the detection task is executed in parallel. Therefore, the detection task can be kept unchanged in the current task set, improving the efficiency of parallel execution and the rationality of task set partitioning.

[0041] For example, the task orchestration module can adopt a canary release strategy, applying the task execution order obtained after dynamically adjusting the detection tasks included in each task set in the task execution order to the defect detection process of the images subsequently acquired by the data access module. That is, by using a canary release strategy, the dynamically adjusted task sets are gradually applied to the defect detection process of the images subsequently acquired by the data access module. This allows for gradual monitoring of the overall detection effect of the adjusted task sets relative to the original task sets, providing more time for monitoring and evaluation of the effect of the dynamically adjusted task sets. This avoids the problem of sudden and uncontrollable deterioration of detection effect caused by sudden and large-scale adjustments to the task sets.

[0042] For example, refer to Figure 6 The intelligent agent task orchestration model may further include a caching and idempotency module. This module acquires and stores intermediate result data generated by the executor module for each detection task during defect detection of the image being detected, and sends this data to the task orchestration evaluation module. The task orchestration evaluation module then uses this intermediate result data to determine the execution result of each detection task in the current task execution order. In other words, during defect detection, the intermediate result data generated by the executor module for each detection task is not deleted but stored, serving as the execution result for the task orchestration evaluation module to determine the current task execution order, thus facilitating optimization of the task execution order. For example, refer to... Figure 10 The caching and idempotency module can perform detection, segmentation, classification, and caching of intermediate result data to support idempotency and playback.

[0043] It should be noted that the above-mentioned functional modules can be configured through hardware such as memory and processor, as well as software code running on them.

[0044] In summary, for reference Figure 11 Compared with related technologies, the defect detection management model of the automatic optical defect detection equipment provided in this disclosure has the following advantages: it upgrades the detection capability from a fixed link to an intelligent agent system with programmable task experts and reasoning, adapting to the needs of multiple scenarios and continuous changes; it improves traceability and engineering controllability through the recording, evidence package and version management of the router module; the task execution order is reusable and replayable, reducing on-site maintenance costs and accelerating iteration.

[0045] It should be noted that, in addition to the structure described above, the intelligent agent task orchestration model of the automatic optical defect detection device in this embodiment may also include the related structures described in part of the embodiment, all of which are within the protection scope of the intelligent agent task orchestration model of the automatic optical defect detection device provided in this embodiment.

[0046] Example 2 This disclosure provides an automatic optical defect detection device, which includes any of the agent task orchestration models of the automatic optical defect detection device provided in Embodiment 1 of this disclosure.

[0047] Example 3 This disclosure also provides an agent task orchestration method for an automated optical defect detection device, referencing... Figure 7 The task orchestration method for this intelligent agent mainly includes the following steps: In step 310: the image of the device under test is acquired by the image acquisition device; In step 320: Obtain the task requirements for defect detection of the image to be detected; In step 330: Multiple detection tasks are identified based on task requirements; In step 340: the execution order of multiple detection tasks is arranged, wherein the task execution order includes multiple task sets executed serially, and each task set includes multiple detection tasks executed in parallel; In step 350: an expert model recipe for the image to be inspected and for each task set is configured from a plurality of expert models contained in the automated optical defect detection device, the expert model recipe including at least a portion of the expert models selected from the plurality of expert models; In step 360: Defect detection is performed on the image to be detected based on the expert model recipe, so that multiple task sets are executed sequentially in the order of task execution, and multiple detection tasks in the task set are executed in parallel while executing a task set; In step 370: the execution result of each detection task is obtained, and the parallel execution interference between multiple detection tasks in each task set is analyzed based on the execution result; In step 380: Based on the interference of parallel execution, the detection tasks included in each task set in the task execution order are dynamically adjusted.

[0048] It should be noted that the above-described agent task orchestration method for the automated optical defect detection device has the same technical effects as the agent task orchestration model of the automated optical defect detection device shown in Embodiment 1, and will not be repeated here. It should also be noted that the agent task orchestration method for the automated optical defect detection device shown in this disclosure embodiment may include other steps besides the steps described above, and these are all within the protection scope of this disclosure embodiment.

[0049] Although embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. An agent task orchestration model for an automated optical defect detection device, characterized in that, include: The data access module is used to acquire the image of the device under test acquired by the image acquisition device; The task acquisition module is used to acquire the task requirements for defect detection of the image being detected. The task identification module is used to identify multiple detection tasks based on the task requirements. The task orchestration module is used to orchestrate the task execution order of the multiple detection tasks, wherein the task execution order includes multiple task sets executed serially, and each task set includes multiple detection tasks executed in parallel. A router module is configured to configure expert model recipes for the image being inspected and each of the task sets from a plurality of expert models included in the automated optical defect detection device, the expert model recipes including at least a subset of expert models selected from the plurality of expert models; An actuator module is used to perform defect detection on the image to be detected based on the expert model recipe, so as to execute the multiple task sets sequentially according to the task execution order, and execute multiple detection tasks in the task set in parallel while executing one of the task sets.

2. The agent task orchestration model as described in claim 1, characterized in that, Also includes: The task orchestration evaluation module is used to obtain the execution results of the executor module for each detection task, analyze the execution results to obtain the parallel execution interference between multiple detection tasks in each task set, and send the parallel execution interference to the task orchestration module. The task orchestration module dynamically adjusts the detection tasks included in each task set in the task execution order based on the parallel execution interference.

3. The agent task orchestration model as described in claim 2, characterized in that, The task orchestration and evaluation module analyzes the execution results to obtain the parallel execution interference between multiple detection tasks in each task set, including: Obtain reference detection result data for each detection task when performing each detection task individually on the image being detected; Based on the execution results, obtain the actual detection result data corresponding to each detection task when each detection task is executed in parallel with other detection tasks in its current task set; Based on the reference detection result data and the actual detection result data for each detection task, the parallel execution interference between the multiple detection tasks in each task set is obtained.

4. The agent task orchestration model as described in claim 3, characterized in that, The task orchestration and evaluation module includes a scoring module; The scoring module scores the reference detection result data corresponding to each detection task to obtain a reference score value for each detection task. The scoring module also scores the actual detection result data corresponding to each detection task to obtain the actual score value corresponding to each detection task. When the reference score value corresponding to the detection task is greater than the actual score value corresponding to the detection task, the task orchestration and evaluation module determines that, in the current task execution order, the detection task has parallel execution interference with other detection tasks in the task set to which it currently belongs.

5. The agent task orchestration model as described in claim 3, characterized in that, The task orchestration module dynamically adjusts the detection tasks included in each task set in the task execution order based on the parallel execution interference, including: When the task orchestration evaluation module determines that the detection task has parallel execution interference with other detection tasks in its current task set, the detection task is adjusted from the current task set to another task set.

6. The agent task orchestration model as described in claim 4, characterized in that, When the reference score value corresponding to the detection task is less than or equal to the actual score value corresponding to the detection task, the task orchestration and evaluation module determines that in the current task execution order, the detection task does not have parallel execution interference with other detection tasks in the task set to which it currently belongs.

7. The agent task orchestration model as described in claim 6, characterized in that, The task orchestration module dynamically adjusts the detection tasks included in each task set in the task execution order based on the parallel execution interference, including: When the task orchestration evaluation module determines that the detection task does not interfere with the parallel execution of other detection tasks in its current task set, the detection task remains unchanged in the current task set.

8. The agent task orchestration model as described in claim 5 or 7, characterized in that, The task orchestration module adopts a canary release strategy, which dynamically adjusts the task execution order obtained after the detection tasks included in each task set in the task execution order. This order is then applied to the defect detection process of the detected image acquired by the data access module afterward.

9. The agent task orchestration model as described in claim 2, characterized in that, Also includes: The caching and idempotency module is used to acquire and store intermediate result data generated by the actuator module for each detection task during the defect detection of the image being detected by the actuator module, and send it to the task orchestration and evaluation module. The task orchestration and evaluation module obtains the execution result of each detection task in the current task execution order based on the intermediate result data.

10. A method for scheduling intelligent agent tasks in an automated optical defect detection device, characterized in that, include: Acquire the image of the device under test captured by the image acquisition device; Obtain the task requirements for defect detection of the image being detected; Multiple detection tasks were identified based on the aforementioned task requirements; The task execution order of the multiple detection tasks is arranged, wherein the task execution order includes multiple task sets executed serially, and each task set includes multiple detection tasks executed in parallel; The automated optical defect detection device configures an expert model recipe for the image being inspected and for each of the task sets from a plurality of expert models, the expert model recipe comprising at least a subset of the expert models selected from the plurality of expert models; Based on the expert model formula, defect detection is performed on the detected image, and the multiple task sets are executed sequentially according to the task execution order, and multiple detection tasks in the task set are executed in parallel while executing one of the task sets; The execution results of each detection task are obtained, and the parallel execution interference between multiple detection tasks in each task set is analyzed based on the execution results. Based on the parallel execution interference, the detection tasks included in each task set in the task execution order are dynamically adjusted.

Images