Product quality inspection method

Abstract

The invention discloses a product quality inspection method and device. The method comprises the following steps: collecting quality inspection result information of a first product set at each quality inspection site in the production process and production process information corresponding to each piece of quality inspection result information; for each quality inspection site, determining at least one related production process parameter type related to the quality inspection result information according to the quality inspection result information and the first product parameter value set; for each quality inspection site, according to the quality inspection result information and the first product parameter value set, determining an anomaly detection strategy set and the accuracy rate and recall rate of the anomaly detection strategy set for the first product set; and for each quality inspection site, based on the anomaly detection strategy set and the accuracy rate and recall rate of the anomaly detection strategy set for the first product set, determining a sampling inspection strategy of the quality inspection site for the second product set. According to the product quality inspection method, the production efficiency and the quality inspection effect can be improved by dynamically adjusting and updating the sampling inspection strategy of the product in the production process.

Description

Product quality inspection methods Technical Field

[0001] The present application relates to the field of product quality inspection, and specifically to a product quality inspection method and apparatus, computing equipment, computer-readable storage medium, and computer program product. Background Art

[0002] In the production of products with complex processes (e.g., dozens or even hundreds of processes) (such as assembly line products like display panels), in addition to the final product quality inspection station, at least some important intermediate processes (e.g., production stations) in the production process (e.g., within the assembly line) also need to be set up for product inspection to promptly detect defective products and rework or repair them. Because the quality inspection process will extend the production cycle, these quality inspection stations generally adopt the form of random inspections.

[0003] There are generally two commonly used sampling inspection strategies: one is to set a sampling inspection cycle, and after each sampling inspection cycle, one (or a batch of) products will enter the quality inspection site for inspection; the other is random sampling, that is, a fixed ratio of a number of products in each batch are randomly selected for inspection. When problems are found during the sampling inspection process, sometimes further processing is carried out by manual or automated strategies. For example, when a defective product is found during the sampling inspection of a batch of products, the entire batch of products will be fully inspected. However, since the sampling inspection cycle or sampling inspection ratio is generally a fixed value, if the sampling inspection coverage rate is too high, it will affect the product production process and reduce production capacity and production efficiency; if the sampling inspection coverage rate is too low, the inspection effect will be poor. For example, there is a high probability that defective products produced in the early process will not be discovered in time, and it will be difficult to trace back if defects are found in the later processes.

[0004] Summary of the Invention

[0005] In view of this, the present application provides a product quality inspection method and apparatus, a computing device, a computer-readable storage medium, and a computer program product, which are expected to alleviate or overcome some or all of the above-mentioned defects and other possible defects.

[0006] According to a first aspect of the present application, a product quality inspection method is provided, comprising: collecting quality inspection result information of a first product set at each quality inspection site during the product production process and production process information corresponding to the quality inspection result information of each quality inspection site, wherein the production process information corresponding to the quality inspection result information of each quality inspection site includes at least one type of production process parameter involved in the production process of the first product set before entering the quality inspection site and a first product parameter value set of each production process parameter type; for each quality inspection site, according to the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each production process parameter type A value set is provided, and at least one relevant production process parameter type related to the quality inspection result information is determined from at least one production process parameter type corresponding to the quality inspection result information; for each quality inspection site, based on the quality inspection result information of the quality inspection site and the first product parameter value set of each of the at least one relevant production process parameter type, an anomaly detection strategy set corresponding to the at least one relevant production process parameter type and its precision and recall rate for the first product set are determined; for each quality inspection site, based on the anomaly detection strategy set and its precision and recall rate for the first product set, a sampling inspection strategy of the quality inspection site for the second product set is determined.

[0007] In the product quality inspection method according to some embodiments of the present application, the quality inspection result information includes at least one of the following items: identification information of whether the product is good or bad, product identifier of good products, product identifier of bad products, bad location, bad type, and bad parameters.

[0008] In a product quality inspection method according to some embodiments of the present application, for each quality inspection site, based on the quality inspection result information of the quality inspection site and the first product parameter value set corresponding to each production process parameter type, at least one related production process parameter type related to the quality inspection result information is determined from at least one production process parameter type corresponding to the quality inspection result information, including: for each quality inspection site, constructing a quality inspection result variable of the first product set at the quality inspection site based on the quality inspection result information of the quality inspection site; for each quality inspection site, constructing a parameter variable of each production process parameter type for the first product set based on the first product parameter value set of each production process parameter type corresponding to the quality inspection result information of the quality inspection site; for each quality inspection site, calculating a correlation measure between the quality inspection result variable corresponding to the quality inspection result information of the quality inspection site and the parameter variable of each production process parameter type; for each quality inspection site, determining at least one related production process parameter type related to the quality inspection result information from the at least one production process parameter type based on the correlation measure between the quality inspection result variable corresponding to the quality inspection result information of the quality inspection site and the parameter variable of each production process parameter type.

[0009] In a product quality inspection method according to some embodiments of the present application, for each quality inspection site, a correlation measure between the quality inspection result variable corresponding to the quality inspection result information of the quality inspection site and the parameter variable of each production process parameter type is calculated, including: for the quality inspection result information of each quality inspection site, the following steps are performed: using statistical hypothesis testing, calculating the first correlation between the quality inspection result variable corresponding to the quality inspection result information and the parameter variable of each production process parameter type; calculating the correlation coefficient between the quality inspection result variable corresponding to the quality inspection result information and the parameter variable of each production process parameter type to determine the second correlation; training a product quality inspection model based on training samples constructed according to the first product parameter value set of each production process parameter type and sample labels constructed according to the quality inspection result information to determine the third correlation between the quality inspection result variable corresponding to the quality inspection result information and the parameter variable of each production process parameter type; determining the correlation measure between the quality inspection result variable and the parameter variable of each production process parameter type based on at least one of the first correlation, second correlation and third correlation between the quality inspection result variable corresponding to the quality inspection result information and the parameter variable of each production process parameter type.

[0010] In the product quality inspection method according to some embodiments of the present application, the correlation measure between the quality inspection result variable and the parameter variable of each production process parameter type is determined based on at least one of the first correlation, second correlation and third correlation between the quality inspection result variable corresponding to the quality inspection result information and the parameter variable of each production process parameter type, including: calculating the weighted sum of the first correlation, the second correlation and the third correlation based on a preset weight set; and determining the correlation measure between the quality inspection result variable and the parameter variable of each production process parameter type based on the weighted sum.

[0011] In the product quality inspection method according to some embodiments of the present application, for each quality inspection site, based on the quality inspection result information of the quality inspection site and the first product parameter value set of each of the at least one related production process parameter type, the anomaly detection strategy set corresponding to the at least one related production process parameter type and its precision and recall rate for the first product set are determined, including: for each quality inspection site, based on the quality inspection result information of the quality inspection site and the first product parameter value set of each corresponding related production process parameter type, the anomaly detection strategy of the related process parameter type is determined; for each quality inspection site, based on the anomaly detection strategy of each of the at least one related production process parameter type, the anomaly detection strategy set corresponding to the at least one production process parameter type is determined; for each quality inspection site, based on the quality inspection result information of the quality inspection site, the anomaly detection strategy set corresponding to the at least one related production process parameter type and the first product parameter value set of each related production process parameter type, the precision and recall rate of the anomaly detection strategy set for the first product set is determined.

[0012] In the product quality inspection method according to some embodiments of the present application, for each quality inspection site, based on the quality inspection result information of the quality inspection site and the first product parameter value set of each relevant production process parameter type, an anomaly detection strategy for the relevant process parameter type is determined, including: for each quality inspection site, performing at least one of the following steps: constructing a normal distribution based on the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each relevant production process parameter type to determine the first anomaly value range of the relevant production process parameter type; constructing a decision tree classification model based on the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each relevant production process parameter type to determine the second anomaly value range of the relevant production process parameter type; constructing an anomaly detection model based on the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each relevant production process parameter type, which is used to determine whether the parameter value of the relevant production process parameter type is abnormal.

[0013] In the product quality inspection method according to some embodiments of the present application, an anomaly detection model is constructed based on the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each relevant production process parameter type, including: constructing an initial anomaly detection model based on a clustering algorithm and / or an isolation forest algorithm based on the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each relevant production process parameter type; optimizing the parameters of the initial anomaly detection model based on a Bayesian optimization algorithm to obtain an anomaly detection model for determining whether the parameter value of the relevant production process parameter type is abnormal.

[0014] In the product quality inspection method according to some embodiments of the present application, for each quality inspection site, based on the quality inspection result information of the quality inspection site, the anomaly detection strategy set corresponding to the at least one relevant production process parameter type and the first product parameter value set of each relevant production process parameter type, the precision and recall rate of the anomaly detection strategy set for the first product set are determined, including: for the quality inspection result information of each quality inspection site, performing the following steps: determining a first anomaly parameter value set from the first product parameter value set of the relevant production process parameter type according to the anomaly detection strategy set of the at least one relevant production process parameter type and the first product parameter value set of each relevant production process parameter type; determining a first anomaly parameter value set according to each relevant According to the first abnormal parameter value set of the production process parameter type, first information of the defective products determined in the first product set is determined, and the first information includes a first identifier set and a first quantity of the defective products determined; according to the quality inspection result information, second information of the actual defective products in the first product set is determined, and the second information includes a second identifier set and a second quantity of the actual defective products; according to the first identifier set and the second identifier set, a third quantity of the correctly determined defective products in the first product set is determined; the quotient of the third quantity and the first quantity is calculated to obtain the precision of the abnormal detection strategy set for the first product set; and the quotient of the third quantity and the second quantity is calculated to obtain the recall rate of the abnormal detection strategy set for the first product set.

[0015] In the product quality inspection method according to some embodiments of the present application, for each quality inspection site, based on the anomaly detection strategy set and its precision and recall rate for the first product set, the sampling inspection strategy of the quality inspection site for the second product set is determined, including: for each quality inspection site, performing the following steps: obtaining a second product parameter value set of each of the at least one related production process parameter type involved in the production process of the second product set before entering the quality inspection site; determining a second abnormal parameter value set from the second product parameter value set of each related production process parameter type according to the anomaly detection strategy set; determining the abnormalities determined in the second product set according to the second abnormal parameter value set of each related production process parameter type. The third information of the product and the fourth information of the determined normal product, wherein the third information includes the third identifier set and the fourth quantity of the determined abnormal product, and the fourth information includes the fourth identifier set and the fifth quantity of the determined normal product; based on the product of the fourth quantity and the precision of the abnormal detection strategy set for the first product set, the first random inspection quantity of the quality inspection site for the determined abnormal product is calculated; based on the product of the fifth quantity and the recall rate of the abnormal detection strategy set for the first product set and the fifth quantity, the second random inspection quantity of the quality inspection site for the determined normal product is calculated; according to the first random inspection quantity and the second random inspection quantity and the third identifier set and the fourth identifier set, the random inspection strategy of the quality inspection site for the second product set is determined.

[0016] In the product quality inspection method according to some embodiments of the present application, the production process parameter type includes at least one of the following: product history information, production site type, production equipment parameters, production environment parameters, key time node information, and product quality monitoring parameters.

[0017] In the product quality inspection method according to some embodiments of the present application, collecting the quality inspection result information of each quality inspection site for the first product set during the product production process and the production process information corresponding to the quality inspection result information of each quality inspection site includes: collecting the production process data of each production site and the quality inspection result data of each quality inspection site during the product production process in real time and synchronizing them to the message queue to form a process database and a quality inspection database respectively; regularly fusing the process data and the data in the quality inspection database according to the product identifier to obtain a fused database; and collecting the quality inspection result information of each quality inspection site for the first product set and the production process information corresponding to each quality inspection result information from the fused database.

[0018] According to a second aspect of the present application, a product quality inspection device is provided, comprising: a data acquisition module, which is configured to collect quality inspection result information of a first product set at each quality inspection site during the product production process and production process information corresponding to the quality inspection result information of each quality inspection site, wherein the production process information corresponding to the quality inspection result information of each quality inspection site includes at least one type of production process parameter involved in the production process of the first product set before entering the quality inspection site and a first product parameter value set of each production process parameter type; a correlation analysis module, which is configured to, for each quality inspection site, collect quality inspection result information of the first product set at the quality inspection site and the first product parameter value set of each corresponding production process parameter type according to the quality inspection result information of the quality inspection site , determine at least one relevant production process parameter type related to the quality inspection result information from at least one production process parameter type corresponding to the quality inspection result information; an abnormality detection strategy establishment module is configured to, for each quality inspection site, determine the abnormality detection strategy set corresponding to the at least one relevant production process parameter type and its precision and recall rate for the first product set based on the quality inspection result information of the quality inspection site and the first product parameter value set of each of the at least one relevant production process parameter type; a sampling strategy determination module is configured to, for each quality inspection site, determine the sampling strategy of the quality inspection site for the second product set based on the abnormality detection strategy set and its precision and recall rate for the first product set.

[0019] According to a third aspect of the present application, a computing device is provided, comprising: a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor is prompted to perform a method according to some embodiments of the present application.

[0020] According to a fourth aspect of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed, the methods according to some embodiments of the present application are implemented.

[0021] According to a fifth aspect of the present application, a computer program product is provided, comprising a computer program, which implements the steps of the method according to some embodiments of the present application when executed by a processor.

[0022] In the product quality inspection method according to some embodiments of the present application, based on the bad correlation analysis (i.e., the correlation analysis between the quality inspection results and the production process parameter types) of the (inspected) product set (i.e., the first product set) corresponding to the current acquisition cycle in each quality inspection site, one or more related production process parameter types with high correlation with the product quality inspection result information are dynamically obtained, and the precision and recall rate of the related production process parameter types for the first product set are obtained based on the anomaly detection strategy set. Finally, based on the anomaly detection strategy set and its precision and recall rate, the sampling strategy of the latest (to be inspected) product set (i.e., the second product set) at the corresponding quality inspection site is determined, thereby realizing the dynamic adjustment and update of the quality inspection or sampling strategy during the product production process (such as assembly line products); because the anomaly detection strategy set is determined for several related production parameter types with high bad correlation, the sampling strategy obtained thereby is more targeted and more accurate; and because the precision and recall rate of the anomaly detection strategy set for historically inspected products (i.e., the first product set) are simultaneously referred to when determining the extraction strategy, the impact of missed judgments and misjudgments on the sampling accuracy is reduced. In summary, the product quality inspection method according to the present application is a method for determining a dynamic sampling strategy for products in the production process based on defect correlation analysis. It significantly improves the quality inspection effect (accuracy) and efficiency while ensuring the production efficiency (capacity) of the (assembly line) product production process, and avoids the occurrence of missed detection and false detection of product defects to the greatest extent.

[0023] These and other advantages of the application will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Embodiments of the present application will now be described in more detail and with reference to the accompanying drawings, in which:

[0025] FIG1 shows an exemplary application scenario based on a product quality inspection method according to some embodiments of the present application;

[0026] FIG2 schematically shows a principle block diagram of a product quality inspection method according to some embodiments of the present application;

[0027] FIG3 shows a flow chart of a product quality inspection method according to some embodiments of the present application;

[0028] FIG4 shows an example process of a data collection step in a product quality inspection method according to some embodiments of the present application;

[0029] 5A-5B illustrate an example process of a correlation analysis step in a product quality inspection method according to some embodiments of the present application;

[0030] 6A-6C illustrate an example process of establishing an abnormality detection strategy in a product quality inspection method according to some embodiments of the present application;

[0031] FIG7 shows an example process of determining a sampling strategy in a product quality inspection method according to some embodiments of the present application;

[0032] FIG8 shows an exemplary structural block diagram of a product quality inspection device according to some embodiments of the present application;

[0033] FIG9 schematically shows an example block diagram of a computing device according to some embodiments of the present application. DETAILED DESCRIPTION

[0034] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. Like reference numerals in the drawings represent like or similar parts, and thus repetitive description thereof will be omitted.

[0035] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner.In the following description, many specific details are provided so as to provide a full understanding of the embodiments of the present application. However, it will be appreciated by those skilled in the art that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps etc. can be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0036] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0037] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, while others may be combined or partially combined. Therefore, the actual execution order may vary depending on the actual situation.

[0038] It should be understood that although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. Thus, the first component discussed below could be referred to as the second component without departing from the teachings of the present invention. As used herein, the term "and / or" and similar terms include all combinations of any, multiple, and all of the associated listed items.

[0039] Those skilled in the art will understand that the drawings are merely schematic diagrams of example embodiments, and the modules or processes in the drawings are not necessarily necessary for implementing the present application, and therefore cannot be used to limit the scope of protection of the present application.

[0040] Figure 1 shows an exemplary application scenario 100 of a product quality inspection method according to some embodiments of the present application. As shown in Figure 1 , the application scenario 100 may include a production line 110 , a network 120 , and a server 130 , and may optionally include an external database 140 .

[0041] As shown in Figure 1, the production line 110 can be an assembly line. The assembly line (also known as the assembly line) is a production method in industry, which means that each production unit only focuses on processing a certain segment of work to improve work efficiency and output. In the assembly line production process of display panels, since the complete production process is very complicated, it often involves dozens or even hundreds of processes. Therefore, in addition to the quality inspection station for the final product, after some important processes in the assembly line are completed (that is, important production stations), quality inspection stations will also be set up to inspect the products at that stage, so as to promptly detect defects and rework or repair them. In this article, the production line 110 may include an assembly line for display panels. Optionally, the production line 110 may also be a production line in other forms besides an assembly line.

[0042] Therefore, in the product quality inspection method according to some embodiments of the present application, as shown in Figure 1, the production line 110 may not only include a production site set 111 consisting of multiple production sites 111a, 111b, 111c, 111d, etc., but also include multiple quality inspection sites 112a, 112b, and 112c. As shown in Figure 1, in order to improve production efficiency and avoid the impact of excessive quality inspection processes on production capacity, a quality inspection site is not set after each production site. Instead, quality inspection sites, such as 112a, 112b, and 112c, are only set at production sites involved in important processes, such as 111a, 111c, 111d, etc.

[0043] The product quality inspection method according to some embodiments of the present application can be deployed on the server 130 and implemented through the server 130. The server 130 can be configured to: first, collect the quality inspection result information of each quality inspection site for the first product set in the product production process and the production process information corresponding to the quality inspection result information of each quality inspection site, wherein the production process information corresponding to the quality inspection result information of each quality inspection site includes at least one type of production process parameter involved in the production process of the first product set before entering the quality inspection site and the first product parameter value set of each production process parameter type; second, for each quality inspection site, according to the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each production process parameter type, collect the quality inspection result information of the first product set from the quality inspection site; Determine at least one relevant production process parameter type related to the quality inspection result information from at least one production process parameter type corresponding to the quality inspection result information; again, for each quality inspection site, determine the anomaly detection strategy set corresponding to the at least one relevant production process parameter type and its precision and recall rate for the first product set based on the quality inspection result information of the quality inspection site and the first product parameter value set of each of the at least one relevant production process parameter type; finally, for each quality inspection site, determine the sampling inspection strategy of the quality inspection site for the second product set based on the anomaly detection strategy set and its precision and recall rate for the first product set.

[0044] Exemplarily, the server 130 can store and run instructions that can execute the various methods described herein. The server 130 can be a single server or a server cluster, or can be a cloud server or cloud server cluster that can provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. It should be understood that the server mentioned herein can typically be a server computer with a large amount of memory and processor resources, but other embodiments are also possible. In addition, the server 130 is shown only as an example. In fact, other devices or combinations of devices with computing power and storage capacity can also be used instead or in addition to provide corresponding services.

[0045] As shown in FIG1 , the application scenario 100 may further include an external database 140 and a network 120. The server 130 may be connected to the production line 110 via the network 120 to exchange data therewith, and may be connected to the external database 140 via the network 120 to, for example, obtain regularly collected product-related data from the database 140, and, for example, store anomaly detection strategies and their corresponding precision and recall rates in the database 140. For example, the database 140 may be an independent data storage device or device group, or may be a backend data storage device or device group associated with other online services (such as online services that provide intelligent customer service, voice assistants, and other functions).

[0046] Examples of the network 120 include, for example, a local area network (LAN), a wide area network (WAN), a personal area network (PAN), and / or a combination of communication networks such as the Internet. Each of the server 130, the database 140, and the production line 110 may include at least one communication interface (not shown) capable of communicating via the network 120. Such a communication interface may be one or more of the following: any type of network interface (e.g., a network interface card (NIC)), a wired or wireless (such as an IEEE 802.11 wireless LAN (WLAN)) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and the like.

[0047] It should be understood that although the server 130 and the database 140 are shown and described as separate structures in this document, they may also be different components of the same device. Among them, for example, the server 130 may provide background computing functions, and the database 140 may provide data exchange, storage, and acquisition functions. The server 130 and the database 140 may also be integrated in the production line 110, so that data exchange and processing as well as direct application of various computing functions may be performed directly without going through the network 120. Optionally, the product quality inspection method according to some embodiments of the present application is not limited to being implemented on the server side shown in FIG1 , but may also be implemented on the production line side, or may also be implemented jointly on the production line 110 side and the server 130 side.

[0048] Figure 2 schematically shows a principle block diagram of a product quality inspection method according to some embodiments of the present application. As shown in Figure 2, compared with Figure 1, in addition to the server 130 and the external database 140, the principle block diagram further includes a message queue 150 (such as a distributed messaging system Kafka) for real-time transmission and synchronization of data. As shown in Figure 2, the message queue 150 can be a part of the external database 140, used for data exchange or transmission (especially real-time exchange and synchronization) between the external data 140 (and / or the server 130) and the production line 110; optionally, the message queue 150 can also be a separate message system separated from the external database. As shown in Figure 2, the product quality inspection method according to the present application can be implemented mainly based on the bad correlation analysis module 130a, the abnormality detection strategy establishment module 130b and the sampling strategy determination module 130c.

[0049] More specifically, as shown in Figure 2, in the product quality inspection method according to some embodiments of the present application, in the data collection step, the production process data of each production site in the production line 110 and the quality inspection result data of each quality inspection site can be first recorded and synchronized in real time through the message queue 150 (such as Kafka), and then the real-time recorded data are stored in the external database 140, wherein the production process data is stored in the process database 140a therein, and the quality inspection result data is stored in the quality inspection database 140b; secondly, the production process data in the process database 140a in the external database 140 and the quality inspection result data in the quality inspection database 140b are regularly fused according to the product identifier (such as ID) to form a fusion database indexed by the product identifier (ID), which can include the quality inspection result data and production process data of the current collection cycle (such as the current batch of products) for use by the bad correlation analysis module 130a.

[0050] As shown in Figure 2, in the correlation analysis step, the poor correlation analysis module 130a can first obtain the quality inspection result information (or poor result information) of the currently collected first product set at each quality inspection site and the first product parameter value set of each of the various production parameter types involved in the growth site that the first product set passed through during the production process before entering the quality inspection site from the fusion database; secondly, the poor correlation analysis module 130a determines at least one relevant production parameter type related to each quality inspection site or quality inspection result information through the correlation analysis between the quality inspection result information and the various production parameter types.

[0051] As shown in FIG2 , in the abnormality detection strategy establishment step, the abnormality detection strategy establishment module 130b is used to combine the first product parameter value sets of various production parameter types and the corresponding quality inspection result information to first determine the abnormality detection strategy sets corresponding to the various relevant production parameter types. For example, the abnormality threshold is determined by constructing a normal distribution, a decision tree classification model, or an unsupervised abnormality detection model is constructed to realize the abnormality detection of the parameter values ​​of the relevant production parameter types. Subsequently, the abnormality detection strategy set can be used to determine whether each parameter value in the first product parameter value set of the relevant production parameter type at each production site is abnormal (for example, the first abnormal parameter value set is determined from the first product parameter value set), so that the products in the first product set corresponding to the abnormal parameter values ​​(in the first abnormal parameter value set) are determined as defective products. Then, based on the defective product information determined by the abnormality detection strategy set and the actual defective product information in the current first product set, the precision and recall of the abnormality detection strategy set for the first product set are calculated for use in the subsequent sampling strategy determination process.

[0052] In the sampling strategy determination step, as shown in Figure 2, the sampling strategy determination module 130c first uses the message queue 150 to collect and synchronize the second product parameter value sets of various relevant production parameter types of the latest product set to be tested (i.e., the second product set) in real time at each production site; secondly, the abnormal detection strategy set determined by the abnormal detection strategy establishment module 130b is used to obtain the abnormal information in the second product parameter value sets of various relevant production parameter types (for example, the second abnormal parameter value set is determined from the second product parameter value set); then, based on the abnormal information (i.e., the second abnormal parameter value set), the abnormal (e.g., determined defective) product information (e.g., including abnormal product ID and / or quantity) and normal product information are obtained; finally, the accuracy and recall rate of the abnormal detection strategy for the previous historical product set (i.e., the first product set) are combined with the abnormal product information and normal product information of the second product set to obtain the sampling strategy of the latest product set to be tested (second product combination) at the corresponding quality inspection site, such as the number of abnormal products sampled and / or the number of normal products sampled.

[0053] Figure 3 is a flow chart of a product quality inspection method according to some embodiments of the present application. The product quality inspection method shown in Figure 3 can be implemented in the application scenario shown in Figure 1 , and its execution entity can be the server 130 shown in Figure 1 or can optionally include an external database 140 .

[0054] As shown in Figure 3, according to some embodiments of the present application, the product quality inspection method may include the following steps: S310, data collection step; S320, correlation analysis step; S330, anomaly detection strategy establishment step; S340, sampling strategy determination step.

[0055] The execution process of the above steps S310-S340 is described in detail below with reference to FIG. 2 .

[0056] In step S310 (data collection step), the quality inspection result information of the first product set at each quality inspection site during the product production process and the production process information corresponding to the quality inspection result information of each quality inspection site are collected, wherein the production process information corresponding to the quality inspection result information of each quality inspection site includes at least one production process parameter type involved in the production process of the first product set before entering the quality inspection site and the first product parameter value set of each production process parameter type.

[0057] The first product set in step S310 refers to a set of products that have passed at least one quality inspection station within a certain period of time (that is, a set of products that have been inspected by at least one corresponding quality inspection station). It can be a set of products from a batch, or a set of part of a batch of products, or a set of products from more than one batch. For example, the second product set in the subsequent step S340 is a set of products to be tested that are produced within a certain period of time after the first product set (that is, a set of products that have not yet been inspected by at least one corresponding quality inspection station). It can be the latest set of products produced within a certain period of time after the first product set.

[0058] In some embodiments, as shown in the principle block diagram of Figure 2, according to the concept of the present application, in order to dynamically adjust or update the sampling inspection strategy of products in the production process based on the bad correlation analysis, it is first necessary to regularly obtain the production process data and quality inspection result data of the first product set in the production line (that is, the product set that has been inspected in the corresponding quality inspection station), so as to analyze the bad correlation from past historical data (that is, the correlation between the quality inspection results of the detected first product set (such as bad products or bad inspection results) and the production process data or parameters of the corresponding products).

[0059] In some embodiments, production process information may include but is not limited to: product history information, production site type, production equipment parameters, environmental parameters, key time node information, product quality monitoring parameters, product shape parameters, etc.; quality inspection result information may include but is not limited to: identification information of good or bad products, product identifiers of good products, product identifiers of bad products, defect types, locations where defects occur, etc.

[0060] It should be noted that the so-called quality inspection result information and corresponding production process information of each quality inspection site here are for the same product set (i.e., the first product set), wherein the quality inspection result information of each quality inspection site includes the quality inspection result information of the first product set detected at the quality inspection site, and the production process information corresponding to each quality inspection result includes at least one type of production process parameter involved in the production process (for example, at each production site) of each product in the first product set before entering the quality inspection site, and the first product parameter value set of each production process parameter type. For example, as shown in Figure 2, the production process information corresponding to the quality inspection result information at the quality inspection site 112a includes at least one type of production process parameter involved in the first product set at the production site 111a (the production site before the quality inspection site 112a) and the first product parameter value set corresponding to each of them.

[0061] The various production process parameter types involved in the production process or each production site refer to the various parameter types related to the product production dynamics in the production site, such as current environmental parameters, key time node information, product quality monitoring parameters, product shape parameters, etc.; the first product parameter value set of each production process parameter type refers to the product parameter value set of this production process parameter type for each product in the first product set.

[0062] Table 1 - Examples of first product parameter value sets for various production process parameter types

[0063] Table 1 shows an example of a first product parameter value set. As shown in Table 1, the quality inspection result information of the current quality inspection site (e.g., 112a) corresponds to k production process parameter types (i.e., the first product set involves or includes k production process parameter types in the production process (at each production site) before the quality inspection site), i.e., A1, A2, ..., A k ; The first product set includes m products, whose ID or serial number is 1, 2, ..., m, so each production process parameter type A i The first product parameter value set (i=1,2,…,k) is {a i1 ,a i2 ,…,a im}, where a ij (j=1,2,…,m) represents parameter type A iParameter value for product with serial number j. In some embodiments, it should be noted that during the production process of the same batch of products, the number of products in the first product set corresponding to the quality inspection result information of each quality inspection site may be different, because when the batch of products passes through the quality inspection sites located between the production sites, some defective products may be detected and eliminated. For example, as shown in Figure 1, assuming that the number of products in the first product set corresponding to the first quality inspection site 112a is m (i.e., m products pass through the first quality inspection site 112a), of which 2 defective products are detected, then when the current batch of products enters the second quality inspection site 112b, the number of products in the first product set corresponding to the batch of products becomes m-2.

[0064] Herein, the quality inspection result information may simply include simple inspection results regarding which products were detected as "bad" or "good," such as the identifiers (e.g., IDs) of products detected as bad, identifiers of good products, and the number of good and bad products. The quality inspection result information may also further include more specific bad result information or bad parameters, such as the type of bad product or the location of occurrence. In some embodiments, the quality inspection result information detected at the quality inspection site may only include the quality inspection result information (i.e., bad result information) corresponding to each product in the first product set detected as bad. For example, if, during the current collection cycle, three bad product identifiers, e.g., ID1, ID2, and ID3, were detected at quality inspection site 112a, the quality inspection result information at quality inspection site 112a would include the product identifiers ID1-ID3 of these three bad products, or alternatively, could further include at least one of the three bad product types, locations of occurrence, or other bad-related parameters. Alternatively, the quality inspection result information may also include information about each product in the corresponding product set detected as good (e.g., normal or good) at the quality inspection site, such as the identifiers of good products or the number of good products.

[0065] FIG4 shows an example process of the data collection step in the product quality inspection method according to some embodiments of the present application shown in FIG3. As shown in FIG4, in some embodiments, the data collection step S310 may include:

[0066] S311, real-time collection of production process data of each production site and quality inspection result data of each quality inspection site during the product production process and synchronization of them to the message queue to form a process database and a quality inspection database respectively;

[0067] S312, regularly merging the process data with the data in the quality inspection database according to the product identifier to obtain a fused database; and

[0068] S313 , collecting quality inspection result information of each quality inspection site for the first product set and production process information corresponding to each quality inspection result information from the fusion database.

[0069] As shown in FIG2 , in the quality inspection process implemented during the product production process, each production site 111a-111d records its production process data and synchronizes it to a message queue (e.g., Kafka) 150 in real time; each quality inspection site 112a-112c records its quality inspection result data and synchronizes it to the message queue 150 in real time, so that the data in the message queue can be cumulatively stored in an external database 140 to form a process database 140a and a quality inspection database 140b, wherein the production process data is stored in the process database 140a, and the quality inspection result data is stored in the quality inspection database 140b. Subsequently, the data in the process database 140a and the quality inspection database 140b can be periodically (i.e., periodically) fused according to the product identifier to form fused data indexed by the product identifier for use in subsequent updates to the sampling inspection strategy. As shown in Figure 2, when the sampling or quality inspection strategy needs to be updated or adjusted, the quality inspection result information of a certain product set (i.e., a product set, such as a batch of products corresponding to the current collection cycle) at each quality inspection site and the production process information (at each production site) corresponding to these quality inspection result information can be collected regularly (i.e., periodically) from the fusion database for further processing by the subsequent bad correlation analysis module 130a.

[0070] In step S320 (correlation analysis step), for each quality inspection site, based on the quality inspection result information of the quality inspection site and the first product parameter value set of each corresponding production process parameter type, at least one related production process parameter type related to the quality inspection result information is determined from at least one production process parameter type corresponding to the quality inspection result information.

[0071] According to the concept of the present application, the sampling strategy of the quality inspection site during the production process can be implemented based on the bad correlation analysis, where the bad correlation analysis refers to finding one or more production process parameter types that are most relevant (or have the strongest correlation) to the quality inspection result information (or bad result information) of the products of the first product set corresponding to the current acquisition cycle at the corresponding quality inspection site, that is, finding (related to the quality inspection result information) at least one related production process parameter type among the various production process parameter types involved in the production process (each production site) before the quality inspection site corresponding to the quality inspection result information. Bad correlation analysis can also be called correlation analysis of quality inspection results. Subsequently, corresponding anomaly detection strategies and their precision and recall rates for the first product set are established for determining or adjusting the sampling strategies of subsequent products (i.e., the second product set) at the corresponding quality inspection sites in the production process.

[0072] Regarding the calculation of the correlation between quality inspection result information and various production process parameter types, the quality inspection result information (based on different quality inspection results of "poor" or "excellent") can be first constructed into quality inspection result variables, and each production process parameter type can be constructed into a parameter variable based on the first parameter value set for the first product set; secondly, the correlation measurement calculation method between the quality inspection result variable and the parameter variable is used to realize the measurement of the degree of correlation between the quality inspection result information and the production process parameter type.

[0073] In some embodiments, statistical hypothesis tests in mathematical statistics (e.g., non-parametric tests in significance tests, such as chi-square tests) can be used to sequentially test or predict the correlation measures between parameter variables corresponding to various production process parameter types and quality inspection result variables. For example, the chi-square value of the test result in the chi-square test can be used as the final correlation measure. Optionally, the correlation coefficient calculation method between variables can be used to obtain the correlation measures between parameter variables corresponding to various production process parameter types and quality inspection result variables.

[0074] In some embodiments, machine learning models (such as neural network) models, especially various classification models (for example, random forest classifier models, XGBoost (eXtreme Gradient Boosting) models, LightGBM (Light Gradient Boosting Machine) models, etc.) or regression models can also be used to predict the degree of influence of various production process parameter types or corresponding parameter variables on quality inspection result information or quality inspection result variables or classifications, because the importance of various production process parameter types or their corresponding parameter variables to the predicted quality inspection result information can be output during the model training process.

[0075] In step S330 (anomaly detection strategy establishment step), for each quality inspection site, based on the quality inspection result information of the quality inspection site and the first product parameter value set of each of the at least one related production process parameter type, the anomaly detection strategy set corresponding to the at least one related production process parameter type and its precision and recall rate for the first product set are determined.

[0076] Based on the concept of this application, after determining the relevant production process parameter types corresponding to each quality inspection site, a corresponding anomaly detection strategy set can be established for each relevant production process parameter type, wherein each anomaly detection strategy in the anomaly detection strategy set can correspond one-to-one to each relevant production process parameter type (that is, each anomaly detection strategy can be used to determine or judge whether the parameter value of the corresponding production process parameter type is abnormal, thereby determining whether the corresponding product is abnormal or defective). Subsequently, the precision and recall of the anomaly detection strategy set corresponding to each relevant production process parameter type for the first product set can be calculated for use in subsequent sampling inspection strategy updates.

[0077] As shown in FIG2 , the anomaly detection strategy establishment step S330 can be implemented by combining the first product parameter value sets of various relevant production parameter types and the corresponding quality inspection result information through the anomaly detection strategy establishment module 130b. Anomaly detection strategies for various relevant production parameter types can first be determined based on various anomaly detection algorithms. For example, an anomaly threshold can be determined by constructing a normal distribution, a decision tree classification model, or the like to determine whether an anomaly is present, or an anomaly detection model can be constructed (unsupervised) to implement parameter value anomaly detection for relevant production parameter types. Subsequently, an anomaly detection strategy set can be determined based on the anomaly detection strategy for each relevant production process parameter type. Subsequently, the anomaly detection strategy set is used to determine whether each parameter value in the first product parameter value of the relevant production parameter type in the production process (at each production site) before the corresponding quality inspection site is abnormal, thereby determining the products in the first product set corresponding to the abnormal parameter value as defective products. Furthermore, the precision and recall of the corresponding anomaly detection strategy set for at least one production type for the first product set are calculated based on the determined defective product information and the actual defective product information in the first product set.

[0078] Specifically, the anomaly detection algorithm may include a first anomaly threshold determination method based on a normal distribution, an unsupervised anomaly detection model trained based on clustering and / or isolation forests, and a second anomaly range determination method based on a decision tree classification model. Accordingly, the resulting anomaly detection strategy may include, but is not limited to: a first anomaly range derived from a normal (Gaussian) distribution, a trained (unsupervised) anomaly detection model based on clustering and / or isolation forests, and a second anomaly range derived from a decision tree classification model.

[0079] Regarding the precision and recall of the anomaly detection strategy set corresponding to each quality inspection site for the first product set, the overall precision and recall can be calculated using the following precision calculation formula (1) and recall calculation formula (2) as well as the above-mentioned actual defective product information and the determined defective product information: A=CN / DNx100% (1) R=CN / TNx100% (2)

[0080] In the above formulas (1) and (2), A represents the precision, R represents the recall, CN represents the number of products correctly judged as defective, DN represents the number of products judged as defective, and TN represents the actual number of defective products.

[0081] For the specific calculation method of the precision and recall of the anomaly detection strategy for the first product set, please refer to Figures 6A and 6C and their corresponding descriptions.

[0082] In step S340 (sampling strategy determination step), for each quality inspection site, based on the anomaly detection strategy set and its precision and recall rate for the first product set, the sampling strategy of the quality inspection site for the second product set is determined.

[0083] According to the concept of the present application, for each quality inspection site, after the abnormality detection strategy establishment step is completed, it is necessary to determine the quality inspection or sampling strategy of the second product set to be inspected at the quality inspection site based on the abnormality detection strategy set corresponding to the quality inspection site and its precision and recall rate for the first product set. Specifically, as shown in Figure 2, when the quality inspection site (for example, 112a) of the production line 110 processes a product of a new product set (i.e., the second product set), it first uses the message queue 150 to collect and synchronize in real time the second product parameter value set of various relevant production parameter types of each production site in the production process of the second product set before the quality inspection site 112a; secondly, it can obtain the latest abnormality detection strategy set (for example, the abnormality detection strategy therein includes an abnormal value range or an abnormality detection model) and its precision and recall rate for the corresponding first product set from the message queue; thirdly, it uses the latest abnormality detection strategy set to determine the abnormal information of the second product set, i.e., various relevant production parameter types. Each of the abnormal information in the second product parameter value set for the second product set, such as the second abnormal parameter value set; then, based on the second abnormal parameter value set, the abnormal (determined to be defective) product information (for example, it may include the ID and / or quantity of the determined defective or abnormal products) and normal product (or determined to be excellent or good) information of the second product set about the quality inspection site 112a are obtained; finally, the precision and recall rate of the abnormal product strategy for the first product set are combined with the above-mentioned abnormal product information and normal product information to obtain the sampling strategy of the new product set (i.e., the second product set) at the quality inspection site 112a, such as the number of abnormal products sampled and / or the number of normal products sampled.

[0084] In the product quality inspection method according to some embodiments of the present application, based on the bad correlation analysis (i.e., the correlation analysis between the quality inspection results and the production process parameter types) of the (inspected) product set (i.e., the first product set) corresponding to the current acquisition cycle in each quality inspection site, one or more related production process parameter types with high correlation with the product quality inspection result information are dynamically obtained, and the precision and recall rate of the related production process parameter types for the first product set are obtained based on the anomaly detection strategy set. Finally, based on the anomaly detection strategy set and its precision and recall rate, the sampling strategy of the latest (to be inspected) product set (i.e., the second product set) at the corresponding quality inspection site is determined, thereby realizing the dynamic adjustment and update of the quality inspection or sampling strategy during the product production process (such as assembly line products); because the anomaly detection strategy set is determined for several related production parameter types with high bad correlation, the sampling strategy obtained thereby is more targeted and more accurate; and because the precision and recall rate of the anomaly detection strategy set for historically inspected products (i.e., the first product set) are simultaneously referred to when determining the extraction strategy, the impact of missed judgments and misjudgments on the sampling accuracy is reduced. In summary, the product quality inspection method according to the present application is a method for determining a dynamic sampling strategy for products in the production process based on defect correlation analysis. It significantly improves the quality inspection effect (accuracy) and efficiency while ensuring the production efficiency (capacity) of the (assembly line) product production process, and avoids missed detection and false detection of product defects to the greatest extent.

[0085] 5A and 5B illustrate an example process of the correlation analysis step in the product quality inspection method shown in FIG. 3 according to some embodiments of the present application.

[0086] FIG5A shows an example process of the correlation analysis step S320 of FIG3. As shown in FIG5A, in some embodiments, the correlation analysis step S320 may include:

[0087] S321: For each quality inspection site, construct a quality inspection result variable for the first product set at the quality inspection site based on the quality inspection result information of the quality inspection site;

[0088] S322: For each quality inspection site, construct parameter variables for each production process parameter type for the first product set based on the first product parameter value set for each production process parameter type corresponding to the quality inspection result information of the quality inspection site;

[0089] S323, for each quality inspection site, calculating a correlation measure between the quality inspection result variable corresponding to the quality inspection result information of the quality inspection site and the parameter variable of each production process parameter type;

[0090] S324, for each quality inspection site, based on the correlation measurement between the quality inspection result variable corresponding to the quality inspection result information of the quality inspection site and the parameter variable of each production process parameter type, determine at least one related production process parameter type related to the quality inspection result information from at least one production process parameter type.

[0091] Specifically, as described in S321, the quality inspection result information of each quality inspection site can be constructed as a quality inspection result variable, whose value can be represented as 1 or 0 according to the inspection results (i.e., quality inspection result information) corresponding to different products in the first product set, where 1 indicates that the product is detected as defective and 0 indicates that the product is not detected as good or normal. For example, if m products in the first product set pass through production site 111a, the value of the quality inspection result variable can be represented as an m-dimensional vector, such as (1, 0, 0, 1, 0, ..., 0), where m represents the number of products in the first product set that pass through the quality inspection site.

[0092] Similarly, as described in S322, each type of production process parameter corresponding to the quality inspection result information can also be constructed as a parameter variable through its first product parameter value set, and its value can be each value of the first product parameter value set. For example, when the first product set passes through the production site 111a, each type of production process parameter A shown in Table 1 i The first product parameter value set (i=1,2,…,k) is {a i1 ,a i2 ,…,a im}, then A i The value of the corresponding parameter variable can also be expressed as an m-dimensional vector (a i1 ,a i2 ,…,a im ).

[0093] Therefore, as described in S323, after the above-mentioned variable construction process, the quality inspection result information and various production parameter types can be constructed into quality inspection result variables and parameter variables that can be represented by vectors of the same dimension according to the first product set, so that the correlation measurement (correlation) calculation method between variables can be used to calculate the correlation measurement between the parameter variables corresponding to various production process parameter types and the quality inspection result variables.

[0094] Finally, as described in S324, the correlation metrics corresponding to the various parameter variables can be sorted in descending order, and the production process parameter types corresponding to one or more parameter variables with the highest correlation metrics can be selected as at least one relevant production process parameter type related to the quality inspection result information.

[0095] Figure 5B shows an example flow of the correlation metric calculation step S323 shown in Figure 5 A. As shown in Figure 5B, the correlation metric calculation step S323 shown in Figure 5A may include: executing steps S323a-S323d for each quality inspection result information of the quality inspection node.

[0096] In step S323a, a statistical hypothesis test is used to calculate a first correlation between the quality inspection result variable corresponding to the quality inspection result information and the parameter variable of each production process parameter type.

[0097] Hypothesis testing, also known as statistical hypothesis testing, is a statistical inference method used to determine whether differences between samples or between a sample and the population are due to sampling error or underlying differences. Significance testing is the most commonly used hypothesis testing method and the most basic form of statistical inference. Its basic principle is to first make a hypothesis about the characteristics of the population and then, through statistical reasoning based on sampling studies, infer whether the hypothesis should be rejected or accepted. Common hypothesis testing methods include the Z test, t-test, chi-square test, and F-test.

[0098] Here, the chi-square test can be used to calculate the correlation measure between the quality inspection result variable and the parameter variable, because the chi-square test can be used to analyze the correlation between two categorical variables. The so-called categorical variable refers to a variable with a discrete value. The value of the parameter variable (i.e., the production process parameter type) in this application can be regarded as a categorical variable, and its value for the first product set is the first product parameter value set, such as the parameter variable A shown in Table 1. i The value of (i=1,2,…,k) is the first product parameter value set {a i1 ,a i2 ,…,a im}; the quality inspection result variable can also be regarded as a categorical variable, whose values ​​can include "1 (bad)" and "0 (excellent)". In this way, the chi-square test method can be used to calculate the parameter variable A corresponding to each production process parameter type in turn. i (i=1,2,…,k) Whether there is correlation between the quality inspection result variables corresponding to the quality inspection result information and the degree of correlation or correlation measurement (for example, the chi-square value can represent the correlation measurement, and the larger the chi-square value, the greater the possibility that the hypothesis that the two are related is true).

[0099] In step S323b, the correlation coefficient between the quality inspection result variable corresponding to the quality inspection result information and the parameter variable of each production process parameter type is calculated to determine a second correlation.

[0100] The correlation coefficient is a statistical measure that measures the degree of linear correlation between variables. There are various definitions of the correlation coefficient, with the Pearson, Spearman, and Kendall correlation coefficients being the most commonly used. Depending on the characteristics of the correlation phenomenon, the statistical measure's name varies. For example, a statistical measure that reflects a linear correlation between two variables is called a correlation coefficient (the square of the correlation coefficient is called the coefficient of determination); a statistical measure that reflects a curvilinear correlation between two variables is called a nonlinear correlation coefficient or a nonlinear coefficient of determination; and a statistical measure that reflects a multivariate linear correlation is called a multiple correlation coefficient or a multiple coefficient of determination. Taking the Pearson correlation coefficient as an example, its value ranges from -1 to 1, where -1 indicates a negative correlation between the two variables; 0 indicates no correlation; and 1 indicates a positive correlation.

[0101] Therefore, in some embodiments, the degree of correlation or correlation measure between the quality inspection result variable and each parameter variable (corresponding to various production process parameter types) can be calculated using the correlation coefficient between the variables. In this article, at least one of the Pearson correlation coefficient, the Spearman correlation coefficient, and the Kendall correlation coefficient can be used to calculate the degree of correlation or correlation measure between the quality inspection result variable and each parameter variable.

[0102] In step S323c, the product quality inspection model is trained based on the training samples constructed according to the first product parameter value set corresponding to the quality inspection result information and the sample labels constructed according to the quality inspection result information to determine the third correlation between the quality inspection result variable corresponding to the quality inspection result information and the parameter variable of each production process parameter type.

[0103] In some embodiments, the product quality inspection model can be a neural network model of machine learning, whose input can be the value of various production process parameter types of the product, and the output can be the quality inspection result of the product (for example, bad or excellent). Therefore, for each quality inspection site, the first product parameter value set of all production process parameter types of each production site in the production process before the quality inspection site can be used to construct a product training sample, and the product sample label is constructed using the test results or quality inspection result information of the quality inspection site to train the model. Depending on the specific content of the quality inspection result information, the product quality inspection model can be a classification model or a regression model. Because the quality inspection results or quality inspection result information of the product can not only be binary or multi-classification targets, such as bad and good or excellent, or multiple bad types (for example, including first-level bad, second-level bad, third-level bad, etc. in order of increasing degree of badness); but also other indicators, such as specific values ​​such as bad parameters and bad locations. For the former (classification target or discrete value), the product quality inspection model can adopt a classification model, and for the latter (parameter or continuous value), the product quality inspection model can adopt a regression model.

[0104] It's important to note that during model training, the importance (e.g., normalized numerical value) of each input feature (i.e., various production process parameter types or corresponding parameter variables) in the training samples to the predicted results can be output. This allows the third correlation between the corresponding parameter variable and the quality inspection result variable to be determined based on the importance or importance metric of each production process parameter type (or corresponding parameter variable) to the predicted results (e.g., importance is directly used as a correlation metric). Specifically, taking a decision tree classification model as an example, at each decision node, the optimal feature is selected for splitting to further distinguish the samples reaching that decision node. With each split, we move closer to the final decision (i.e., the leaf node). Therefore, selected features are more important than unselected features that play no role in the decision process. Feature importance can be quantified during training by recording the features selected at each decision node, the number of splits, the cumulative information gain, and the sample ratio for each split (i.e., the percentage of samples in the left and right subtrees relative to the total samples). In this way, the importance of the model input features (i.e., various production process parameter types) to the prediction results (i.e., quality inspection result variables) can be obtained based on the quantitative importance of the output of the last round of iteration at the end of training, thereby determining the third correlation between the quality inspection result variables corresponding to the quality inspection result information and the parameter variables of each production process parameter type.

[0105] In step S323d, the correlation measure between the quality inspection result variable and the parameter variable of each production process parameter type is determined based on at least one of the first correlation, the second correlation and the third correlation between the quality inspection result variable corresponding to the quality inspection result information and the parameter variable of each production process parameter type.

[0106] Generally, after calculating the first, second, and third correlations between the quality inspection result variable and the parameter variable, a final correlation measure between the two can be determined based on any one of the correlations. For example, any one of the first, second, and third correlations can be directly used as the correlation measure between the two. Alternatively, a correlation measure between the two can be calculated based on any two or three of the first, second, and third correlations, such as taking the arithmetic mean or geometric mean.

[0107] In some embodiments, step S323d may include: calculating the weighted sum of the first, second, and third correlations based on a preset weight set; and determining the correlation measure between the quality inspection result variable and the parameter variable of each production process parameter type based on the weighted sum. In practical applications, as described above, the weighted sum of the first, second, and third correlations of each production process parameter type with the quality inspection result information can be calculated based on the three methods in S323a-S323c above to obtain the correlation measure between the production process parameter type and the quality inspection result information (parameter variable and quality inspection result variable). The weight of each correlation in the weighted sum can be predetermined based on conditions such as the specific application scenario. In this way, the advantages of the above three methods can be combined into one, making the obtained results more accurate.

[0108] 6A-6C illustrate an example process of an abnormality detection strategy establishment step in a product quality inspection method according to some embodiments of the present application.

[0109] The steps for establishing anomaly detection strategies can be divided into three parts: first, establishing corresponding anomaly detection strategies for various relevant production process parameter types (i.e., determining strategies for determining whether corresponding parameter values ​​are abnormal (such as anomaly value ranges or anomaly detection models), thereby determining whether corresponding products are abnormal or defective); second, forming an anomaly detection strategy set from anomaly detection strategies for various relevant production process parameter types; and third, calculating the precision and recall of the anomaly detection strategy set for the first product set.

[0110] FIG6A shows an example process of the abnormality detection strategy establishment step S330 shown in FIG3. As shown in FIG6A, the abnormality detection strategy establishment step S330 may include:

[0111] S331, for each quality inspection site, determine an anomaly detection strategy for each relevant process parameter type based on the quality inspection result information of the quality inspection site and the corresponding first product parameter value set for each relevant production process parameter type;

[0112] S332, for each quality inspection site, determining a set of anomaly detection strategies corresponding to at least one type of production process parameter based on anomaly detection strategies for each of at least one related production process parameter type; and

[0113] S333, for each quality inspection site, based on the quality inspection result information of the quality inspection site, the anomaly detection strategy set, and the first product parameter value set of each relevant production process parameter type, determine the precision and recall of the anomaly detection strategy set for the first product set.

[0114] For the quality inspection result information of each quality inspection site, based on the first product parameter value set of each relevant production process parameter type for the first product set and the data set of the corresponding quality inspection result information, different anomaly detection algorithms can be used to determine the anomaly detection strategy. Depending on the anomaly detection algorithm adopted, the corresponding anomaly detection strategy can be expressed in different forms, for example, it can be expressed as an abnormal value range of the parameter value of each relevant production process parameter type, or it can be expressed as an anomaly detection model for determining whether the parameter value to be detected of each relevant production process parameter type is abnormal. In some embodiments, the anomaly detection algorithm includes but is not limited to: an anomaly detection algorithm based on statistical distribution, an anomaly detection algorithm based on distance, an anomaly detection algorithm based on clustering, an anomaly detection algorithm based on tree, and an anomaly detection algorithm based on density. As described in S331, at least one of the anomaly detection algorithms can be used to first determine the anomaly detection strategy for each relevant production process parameter type based on the quality inspection result information of the quality inspection site and the first product parameter value set of each corresponding relevant production process parameter type. Subsequently, as described in S332 , various anomaly detection strategies corresponding to each relevant production process parameter type are grouped together to form an anomaly detection strategy set, which corresponds to various relevant production process parameter types one by one.

[0115] The overall precision and recall in S333 can be calculated in the following way: first, based on the actual inspection result data (i.e., quality inspection result information) of the first product set at each quality inspection site, the defective product information (e.g., including defective product ID and quantity) actually detected in the first product set at each quality inspection site is determined; secondly, the abnormality detection strategy set can be used to determine whether there is an abnormality in the first product parameter value set of at least one production process parameter type in the production process of the first product set before the corresponding quality inspection site (at each production site), and then determine the information of abnormal products (or products determined to be defective) in the first product set; finally, the precision calculation formula (1) and the recall calculation formula (2) as well as the above-mentioned actual defective product information and the determined defective product information are used to calculate the precision and recall.

[0116] Fig. 6B shows an example process of step S331 shown in Fig. 6A. As shown in Fig. 6B, step S331 may include: executing at least one of steps S331a-S331c for each quality inspection site.

[0117] In step S331a, a normal distribution is constructed based on the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each relevant production process parameter type to determine the first abnormal value range of the relevant production process parameter type.

[0118] Statistical distribution-based methods rely on the assumption that the dataset follows a certain probability distribution (such as a normal distribution or a Gaussian distribution). Anomaly detection is achieved by determining whether a data point conforms to this distribution. For example, an anomaly detection algorithm based on a normal distribution can be based on the following idea: if the current data point deviates from the population mean by a certain number of standard deviations (3 or 6), it can be considered an outlier (the number of standard deviations can be adjusted according to the actual situation).

[0119] Therefore, in some embodiments, for each relevant production process parameter type, it is assumed that the relevant production process parameter is a random variable and obeys a normal distribution (μ, σ 2 ), where μ is the mean and σ is the standard deviation. The mean μ and standard deviation σ can be calculated based on the first product parameter value set corresponding to the relevant production parameter; then, based on the actual range of the parameter (i.e., the value in the first parameter value set), the abnormal threshold is defined as μ+kσ (upper limit) and μ-kσ (lower limit). Among them, different k values ​​can be traversed to find the k value that maximizes the difference in product defect rates of the data on both sides of the abnormal threshold (i.e., greater than the upper limit or less than the lower limit), thereby obtaining the abnormal value range. For example, if when k=6, the difference in product defect rates corresponding to the data on both sides is the largest, then the abnormal threshold can be determined as μ+6σ (upper limit) and μ-6σ (lower limit). In other words, the abnormal value range of the current relevant production process parameter type is (μ+6σ, +∞)∪(-∞, μ-6σ), that is, when the corresponding parameter value is greater than μ+6σ or less than μ-6σ, the parameter value is determined to be an abnormal value.

[0120] In step S331b, a decision tree classification model is constructed based on the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each relevant production process parameter type to determine the second abnormal value range of the relevant production process parameter type.

[0121] The anomaly detection algorithm based on the decision tree classification model uses the classification of normal data and abnormal data to achieve anomaly detection. Therefore, for each relevant production process parameter type, a decision tree classification model can be constructed based on its corresponding first product parameter value set, and the split value of the decision tree at the parameter type node can be output during the construction process as the anomaly threshold.

[0122] In step S331c, an anomaly detection model is constructed based on the quality inspection result information of the quality inspection site and the corresponding first product parameter value set of each relevant production process parameter type, which is used to determine whether the parameter value of the relevant production process parameter type is abnormal.

[0123] In some embodiments, for each relevant production process parameter type, an unsupervised model, such as a clustering algorithm model or an isolation forest model, can be constructed (or trained) using the first product parameter value set for the first product set to obtain an anomaly detection model for that production process parameter type. Optionally, the model accuracy can be evaluated using detection labels.

[0124] In the Isolation Forest algorithm, data is repeatedly partitioned based on a specific attribute. Outliers are often partitioned early, effectively isolating them. However, normal points, due to their large population, require more partitioning. Multiple isolation trees are constructed using the following method: An attribute and a value are randomly selected at the current node, dividing all data at the current node into two leaf nodes. If the leaf node depth is small or there are still many data points in the leaf node, the partitioning continues. Outliers are characterized by an average low tree depth across all isolation trees. Clustering algorithms exploit the vertical structure of data and classify it into different categories. They automatically group similar data points and separate different groups. Clustering algorithms or models can group data points into clusters, where a cluster consists of a core point and all points within a certain distance from the core point. If a data point does not belong to any cluster, it is labeled an outlier.

[0125] In some embodiments, the anomaly detection model constructed in step S331c can further utilize a Bayesian optimization algorithm to find optimal model parameters to maximize model accuracy. Therefore, step S331c can include: constructing an initial anomaly detection model based on the quality inspection results of the quality inspection site and the corresponding first product parameter value set for each relevant production process parameter type using a clustering algorithm and / or an isolation forest algorithm; and optimizing the parameters of the initial anomaly detection model using a Bayesian optimization algorithm to obtain an anomaly detection model for determining whether the parameter value of the relevant production process parameter type is abnormal.

[0126] Figure 6C illustrates an example process flow for step S333 shown in Figure 6A. As shown in Figure 6C, step S333 (determining, for each quality inspection site, the precision and recall of the anomaly detection strategy set for the first product set based on the quality inspection result information at that quality inspection site, the anomaly detection strategy set, and the first product parameter value set for each relevant production process parameter type) may include: executing the following steps S333a-S333f for the quality inspection result information at each quality inspection site.

[0127] In step S333a, based on the anomaly detection strategy set of the at least one relevant production process parameter type and the first product parameter value set of each relevant production process parameter type, a first abnormal parameter value set is determined from the first product parameter value set of the relevant production process parameter type.

[0128] First, for each relevant production process parameter type corresponding to the quality inspection result information of each quality inspection site, the corresponding anomaly detection strategy in the anomaly detection strategy set (such as an anomaly value range or an anomaly detection model) is used to detect all parameter values ​​in the first product parameter value set, and abnormal parameter values ​​are detected therefrom to form a first abnormal parameter value set, so as to find abnormal products corresponding to each abnormal parameter value in the first abnormal parameter value set from the first product set.

[0129] In step S333b, first information of defective products in the first product set is determined based on the first abnormal parameter value set of each relevant production process parameter type, where the first information includes a first identifier set and a first quantity of the defective products.

[0130] After determining the first abnormal parameter value set, abnormal products corresponding to each abnormal parameter value in the first abnormal parameter value set can be identified from the first product set, that is, products determined to be defective or determined to be defective products, and the first information of the determined defective products, that is, the first identifier set of each determined defective product or abnormal product and its total number (that is, the first number), can be recorded for subsequent calculation of precision and recall rate.

[0131] In step S333c, second information of actual defective products in the first product set is determined based on the quality inspection result information, where the second information includes a second identifier set and a second quantity of the actual defective products.

[0132] In order to calculate the precision and recall rate, it is also necessary to determine the actual defective products from the first product set based on the quality inspection result information actually detected in the quality inspection site, record their identifier set (i.e., the second identifier set) and the total number of all actual defective products (i.e., the second number), thereby forming the second information of the actual defective products.

[0133] In step S333d, a third number of correctly determined defective products in the first product set is determined based on the first identifier set and the second identifier set.

[0134] Based on the above formulas (1) and (2), it is also necessary to obtain the number of correctly determined defective products among all the determined defective products (i.e., the third number). Because the defective products determined based on the anomaly detection strategy cannot guarantee a 100% determination accuracy, the determined defective products may include correctly determined defective products and incorrectly determined defective products. For the quality inspection result information of each quality inspection site, the so-called correctly determined defective products refer to those determined defective products that match the actual defective products corresponding to the corresponding quality inspection result information among all the defective products determined using the corresponding anomaly detection strategy set. The third number can be obtained based on the comparison or matching process of each identifier in the first identifier set of the determined defective product and each identifier in the second identifier set of the actual defective product, that is, the third number can be determined as the number of first identifiers in the first identifier set that match each second identifier in the second identifier set.

[0135] In step S333e, the quotient of the third quantity and the first quantity is calculated to obtain the accuracy of the anomaly detection strategy set for the first product set.

[0136] For each quality inspection site and its corresponding quality inspection result information and anomaly detection strategy set, according to the precision calculation formula (1), by calculating the quotient between the third number of correctly identified defective products and the first number of identified defective products, the precision of the anomaly detection strategy set corresponding to the quality inspection site for the first product set can be obtained. As shown in formula (1), the precision A refers to the correct judgment rate of the anomaly detection strategy set for the first product set, that is, the percentage of correctly judged defective products among all identified defective products. Correspondingly, the judgment error rate (i.e., the false positive rate) can be defined as the percentage of false positives among all identified defective products, that is, 1-A.

[0137] In step S333f, the quotient of the third quantity and the second quantity is calculated to obtain the recall rate of the anomaly detection strategy set for the first product set.

[0138] For each quality inspection site and its corresponding quality inspection result information and anomaly detection strategy set, according to the recall rate calculation formula (2), by calculating the quotient between the third number of correctly identified defective products and the second number of actual defective products, the recall rate of the anomaly detection strategy set corresponding to the quality inspection site for the first product set can be obtained. As shown in formula (2), the recall rate R refers to the correct determination coverage rate of the actual defective products of the anomaly detection strategy set for the first product set, that is, the percentage of all correctly identified defective products to all actual defective products. Correspondingly, the omission rate of the determination of actual defective products (i.e., the missed determination rate) can be defined as 1-R.

[0139] Figure 7 shows an example process of the sampling strategy determination step in the product quality inspection method according to some embodiments of the present application. As shown in Figure 7, the sampling strategy determination step S340 may include executing the following steps S341-S346 for each quality inspection site.

[0140] In step S341, a second product parameter value set of each of at least one relevant production process parameter type involved in the production process of the second product set before entering the quality inspection site is obtained.

[0141] In order to update the sampling strategy of a designated quality inspection site, before the new product set to be inspected (i.e., the second product set) enters the quality inspection site, it is first necessary to collect a second product parameter value set of various (at least one) relevant production process parameter types for the second product set in the production process before entering the quality inspection site, so as to use the corresponding anomaly detection strategy set to find or judge the abnormal parameter value (i.e., the second abnormal parameter value set), thereby obtaining the abnormal product information in the second product set for judging the abnormal products in the second product set.

[0142] Table 2 - Examples of Second Product Parameter Value Sets for Various Production Process Parameter Types

[0143] Table 2 shows an example of a second product parameter value set of various production process parameter types. As shown in Table 2, the quality inspection result information of the current quality inspection site (for example, the quality inspection site 112a) corresponds to k production process parameter types (that is, the second product set includes k production process parameter types in each production site in the production process before entering the quality inspection site), namely A1, A2, ..., A k , which is the same as the production process parameter type for the first product set in Table 1. As shown in Table 2, the second product set includes n products in total, whose ID or serial number is 1, 2, ..., n, so each production process parameter type A i (i=1,2,…,k) The second product parameter value set for the second product set is {b i1 ,bi2 ,…,b in}, where b ij (j=1,2,…,n) represents parameter type A i Parameter values ​​for the product with serial number j in the second product set. Obviously, the first product set and the second product set are both produced on the same production line 110 (and each production site) shown in Figure 1, so the production process parameter types in Table 2 are the same as those in Table 1, but the number of products m (first product set) and n (second product set) may be different (of course, they can also be the same). It should be pointed out that what is required in step S341 is a second product parameter value set of at least one related production process parameter type. Therefore, after obtaining the information in Table 2, it is necessary to select (at least one) second product parameter value set of the relevant production process parameter type from the second product parameter value sets of various (at least one) production process parameter types shown in Table 2 based on the at least one related production process parameter type determined in step S320 (correlation analysis step).

[0144] In step S342, according to the anomaly detection strategy set, a second anomaly parameter value set is determined from the second product parameter value set of each relevant production process parameter type.

[0145] After obtaining the second product parameter value set corresponding to the second product set, the previously obtained anomaly detection strategy set for various related production process parameter types can be used to determine abnormal parameter values ​​in the second parameter value set, thereby obtaining a second abnormal parameter value set to determine information about abnormal and normal products in the second product set. The specific method for determining the second abnormal parameter value set is similar to the determination of the first abnormal parameter value set in step S333a shown in Figure 6C and is not further described here.

[0146] In step S343, based on the second abnormal parameter value set of each relevant production process parameter type, the third information of the abnormal products determined in the second product set and the fourth information of the normal products determined are determined, wherein the third information includes the third identifier set and the fourth quantity of the abnormal products determined, and the fourth information includes the fourth identifier set and the fifth quantity of the normal products determined.

[0147] In order to determine the number of samples to be inspected for products of different properties in the sampling strategy, it is necessary to determine the abnormal products and normal products in the second product set based on the second abnormal parameter value set and record the corresponding identifier and quantity information, that is, the third information of the abnormal products determined, which may include the third identifier set and fourth quantity of the abnormal products, and the fourth information of the normal products determined, which may include the fourth identifier set and fifth quantity of the normal products. In this way, different sampling strategies (such as proportions or quantities) are determined for products of different properties (normal or abnormal). The specific execution method of step S343 is similar to S333b shown in Figure 6C.

[0148] In step S344, the first sampling quantity of the determined abnormal products at the quality inspection site is calculated based on the product of the fourth quantity and the accuracy of the abnormality detection strategy set for the first product set.

[0149] In some embodiments, the number of samples of different types of products (i.e., abnormal products and normal products) of the second product set at the corresponding quality inspection site can be calculated according to the following formulas (3) and (4): MN = EN * A (3) MM = NN * (1-R) ​​(4)

[0150] In the above formulas (3) and (4), MN represents the number of samples of abnormal products, MM represents the number of samples of normal products, A represents the precision, R represents the recall rate, EN represents the number of abnormal products, and NN represents the number of normal products.

[0151] Based on the above formula (3), the first sampling quantity of the abnormal products determined by the corresponding quality inspection site can be equal to the product of the fourth quantity and the accuracy rate.

[0152] In step S345 , the second sampling quantity for the normal products determined by the quality inspection site is calculated based on the product of the fifth quantity and the recall rate of the abnormality detection strategy set for the first product set and the fifth quantity.

[0153] Based on the above formula (4), the second sampling quantity of the normal products determined by the corresponding quality inspection site can be equal to 1 minus the difference in recall rate multiplied by the fifth quantity.

[0154] In step S346 , the sampling inspection strategy of the quality inspection site for the second product set is determined according to the first sampling inspection quantity, the second sampling inspection quantity, the third identifier set, and the fourth identifier set.

[0155] In some embodiments, the quality inspection site can identify abnormal and normal products from the first product set based on the third and fourth identifier sets, and randomly sample a corresponding number of products from the abnormal and normal products, respectively, according to the first and second sampling quantities, for testing at the quality inspection site. Thus, the dynamic sampling strategy for the second product set implemented by the quality inspection site can be represented as the aforementioned quality inspection or sampling process.

[0156] FIG8 is an exemplary block diagram of a product quality inspection device 800 according to some embodiments of the present application. As shown in FIG8 , the product quality inspection device 800 may include a data acquisition module 810 , a correlation analysis module 820 , an anomaly detection strategy establishment module 830 , and a sampling strategy determination module 840 .

[0157] The data acquisition module 810 can be configured as a data acquisition module, which is configured to collect the quality inspection result information of the first product set at each quality inspection site during the product production process and the production process information corresponding to the quality inspection result information of each quality inspection site, wherein the production process information corresponding to the quality inspection result information of each quality inspection site includes at least one production process parameter type involved in the production process of the first product set before entering the quality inspection site and the first product parameter value set of each production process parameter type.

[0158] The correlation analysis module 820 can be configured to determine, for each quality inspection site, at least one related production process parameter type related to the quality inspection result information from at least one production process parameter type corresponding to the quality inspection result information based on the quality inspection result information of the quality inspection site and the first product parameter value set of each corresponding production process parameter type.

[0159] The anomaly detection strategy establishment module 830 can be configured to determine, for each quality inspection site, a set of anomaly detection strategies corresponding to at least one relevant production process parameter type and its precision and recall rate for the first product set based on the quality inspection result information of the quality inspection site and the first product parameter value set of each of the at least one relevant production process parameter type.

[0160] The sampling strategy determination module 840 can be configured to determine, for each quality inspection site, a sampling strategy for the second product set based on the anomaly detection strategy set and its precision and recall for the first product set.

[0161] It should be noted that the various modules described above can be implemented in software or hardware or a combination of both. Multiple different modules can be implemented in the same software or hardware structure, or one module can be implemented by multiple different software or hardware structures.

[0162] In the product quality inspection device according to some embodiments of the present application, since the anomaly detection strategy set is determined for several types of production process parameters with high bad correlation, the sampling inspection strategy obtained thereby is more targeted and more accurate; and because the precision and recall rate of the historical anomaly detection strategy set for the first product set are referenced when determining the extraction strategy, the impact of missed detection and misjudgment on the sampling inspection accuracy is reduced. In short, the product quality inspection method according to the present application is a method for determining a dynamic sampling inspection strategy in the product production process based on bad correlation analysis, which significantly improves the quality inspection effect (accuracy) and efficiency while ensuring the production efficiency (capacity) in the product production process, and avoids missed detection and misdetection of defective products to the greatest extent.

[0163] FIG9 schematically illustrates an example block diagram of a computing device 900 according to some embodiments of the present application. The computing device 900 may represent a device for implementing the various devices or modules described herein and / or performing the various methods described herein. The computing device 900 may be, for example, a server, a desktop computer, a laptop computer, a tablet, a smart phone, a smart watch, a wearable device, or any other suitable computing device or computing system, which may include various levels of devices ranging from full-resource devices with a large amount of storage and processing resources to low-resource devices with limited storage and / or processing resources. In some embodiments, the flow product quality inspection device 800 described above with respect to FIG8 may be implemented in one or more computing devices 900, respectively.

[0164] As shown in Figure 9, example computing device 900 includes a processing system 901, one or more computer-readable media 902, and one or more I / O interfaces 903 that are communicatively coupled to each other. Although not shown, computing device 900 can also include a system bus or other data and command transmission system that couples various components to each other. The system bus can include any one or combination of different bus structures, and the bus structure can be such as a memory bus or memory controller, a peripheral bus, a universal serial bus, and / or utilize any processor or local bus in a variety of bus architectures. Alternatively, it can also include such as control and data lines.

[0165] Processing system 901 represents functionality that performs one or more operations using hardware. Thus, processing system 901 is illustrated as including hardware elements 904 that may be configured as processors, functional blocks, and the like. This may include implementation in hardware as application-specific integrated circuits or other logic devices formed using one or more semiconductors. Hardware elements 904 are not limited by the materials from which they are formed or the processing mechanisms employed therein. For example, a processor may be comprised of (a plurality of) semiconductors and / or transistors (e.g., electronic integrated circuits (ICs)). In such a context, processor-executable instructions may be electronically executable instructions.

[0166] The computer-readable medium 902 is illustrated as including a memory / storage device 905. The memory / storage device 905 represents a memory / storage device associated with one or more computer-readable media. The memory / storage device 905 may include volatile media (such as random access memory (RAM)) and / or non-volatile media (such as read-only memory (ROM), flash memory, optical disk, magnetic disk, etc.). The memory / storage device 905 may include fixed media (e.g., RAM, ROM, fixed hard drive, etc.) and removable media (e.g., flash memory, removable hard drive, optical disk, etc.). Exemplarily, the memory / storage device 905 can be used to store the production process data and quality inspection result data mentioned in the above embodiments, etc. The computer-readable medium 902 can be configured in various other ways as further described below.

[0167] One or more I / O (input / output) interfaces 903 represent functionality that allows a user to enter commands and information into the computing device 900 and also allows information to be displayed to the user and / or sent to other components or devices using various input / output devices. Examples of input devices include a keyboard, a cursor control device (e.g., a mouse), a microphone (e.g., for voice input), a scanner, touch functionality (e.g., a capacitive or other sensor configured to detect physical touch), a camera (e.g., that can detect motion that does not involve touch as gestures using visible or invisible wavelengths (such as infrared frequencies)), a network card, a receiver, and the like. Examples of output devices include a display device, a speaker, a printer, a tactile response device, a network card, a transmitter, and the like.

[0168] Computing device 900 also includes a product quality inspection strategy 906. This strategy can be stored as computer program instructions in memory / storage device 905, or can be implemented as hardware or firmware. Together with processing system 901 and other components, this strategy can implement all of the functionality of the various modules of product quality inspection device 800 described with respect to FIG. 8 .

[0169] Various techniques may be described herein in the general context of software, hardware, elements, or program modules. Generally, these modules include routines, programs, objects, elements, components, data structures, etc. that perform specific tasks or implement specific abstract data types. As used herein, the terms "module," "function," etc. generally refer to software, firmware, hardware, or a combination thereof. A feature of the techniques described herein is that they are platform-independent, meaning that these techniques can be implemented on a variety of computing platforms with a variety of processors.

[0170] An implementation of the described modules and techniques may be stored on or transmitted across some form of computer-readable media. Computer-readable media may include various media accessible by the computing device 900. By way of example and not limitation, computer-readable media may include "computer-readable storage media" and "computer-readable signal media."

[0171] As opposed to a simple signal transmission, carrier wave, or signal itself, "computer-readable storage medium" refers to a medium and / or device, and / or tangible storage device, capable of persistently storing information. Thus, a computer-readable storage medium refers to a non-signal-bearing medium. Computer-readable storage media includes hardware such as volatile and non-volatile, removable and non-removable media and / or storage devices implemented with methods or technologies suitable for storing information (such as computer-readable instructions, data structures, program modules, logic elements / circuits, or other data). Examples of computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage devices, hard disks, cassettes, magnetic tape, magnetic disk storage devices or other magnetic storage devices, or other storage devices, tangible media, or articles of manufacture suitable for storing desired information and accessible by a computer.

[0172] "Computer-readable signal media" refers to signal-bearing media configured to transmit instructions to the hardware of computing device 900, such as via a network. Signal media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, data signal, or other transport mechanism. Signal media also includes any information transmission media. By way of example, and not limitation, signal media include wired media such as a wired network or direct connection, and wireless media such as acoustic, RF, infrared, and other wireless media.

[0173] As previously mentioned, hardware elements 904 and computer-readable medium 902 represent instructions, modules, programmable device logic and / or fixed device logic implemented in hardware form, which can be used to implement at least some aspects of the technology described herein in some embodiments. Hardware elements can include other implementations in integrated circuits or systems on a chip, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), complex programmable logic devices (CPLDs) and silicon or components of other hardware devices. In this context, hardware elements can be used as processing equipment for executing program tasks defined by the instructions, modules and / or logic embodied by the hardware elements, and hardware devices for storing instructions for execution, such as the computer-readable storage media previously described.

[0174] The aforementioned combinations may also be used to implement the various techniques and modules described herein. Thus, software, hardware or program modules and other program modules may be implemented as one or more instructions and / or logic embodied on some form of computer-readable storage medium and / or by one or more hardware elements 904. The computing device 900 may be configured to implement specific instructions and / or functions corresponding to the software and / or hardware modules. Thus, for example, by using a computer-readable storage medium and / or hardware elements 904 of a processing system, a module may be implemented as a module executable by the computing device 900 as software, at least in part, in hardware. Instructions and / or functions may be executed / operable by, for example, one or more computing devices 900 and / or processing systems 901 to implement the techniques, modules, and examples described herein.

[0175] The techniques described herein may be supported by these various configurations of computing device 900 and are not limited to the specific examples of the techniques described herein.

[0176] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts may be implemented as computer programs. For example, embodiments of the present application provide a computer program product comprising a computer program carried on a computer-readable medium, the computer program including program code for executing at least one step of the method embodiments of the present application.

[0177] In some embodiments of the present application, one or more computer-readable storage media are provided on which computer-readable instructions are stored, and when executed, the computer-readable instructions implement the product quality inspection method according to some embodiments of the present application. The various steps of the product quality inspection method according to some embodiments of the present application can be converted into computer-readable instructions through programming and stored in a computer-readable storage medium. When such a computer-readable storage medium is read or accessed by a computing device or a computer, the computer-readable instructions therein are executed by a processor on the computing device or the computer to implement the method according to some embodiments of the present application.

[0178] In the description of this specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0179] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a sequence other than as shown or discussed (including in a substantially simultaneous manner or in reverse order depending on the functions involved), which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0180] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0181] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, any one of the following technologies known in the art or a combination thereof can be used to implement the present invention: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combinational logic gate circuit, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0182] Those skilled in the art will appreciate that all or part of the steps of the method of the above embodiment may be accomplished through hardware associated with program instructions, and the program may be stored in a computer-readable storage medium, which, when executed, includes executing one or a combination of the steps of the method embodiment.

[0183] In addition, the functional units in the various embodiments of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into a module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

Claims

1. A product quality inspection method, characterized in that, Including: Collecting the quality inspection result information of each quality inspection station for the first product set and the production process information corresponding to the quality inspection result information of each quality inspection station, wherein the production process information corresponding to the quality inspection result information of each quality inspection station includes at least one type of production process parameter involved in the production process before the first product set enters this quality inspection station and the first product parameter value set of each type of production process parameter; For each quality inspection station, according to the quality inspection result information of this quality inspection station and the first product parameter value set of each type of production process parameter corresponding thereto, determining at least one type of relevant production process parameter related to this quality inspection result information from the at least one type of production process parameter corresponding to this quality inspection result information; For each quality inspection station, according to the quality inspection result information of this quality inspection station and the first product parameter value set of each of the at least one type of relevant production process parameter, determining the abnormal detection strategy set corresponding to the at least one type of relevant production process parameter and its precision rate and recall rate for the first product set; For each quality inspection station, based on the abnormal detection strategy set and its precision rate and recall rate for the first product set, determining the sampling inspection strategy of this quality inspection station for the second product set.

2. The method according to claim 1, wherein The quality inspection result information includes at least one of the following: identification information of whether the product is good or bad, product identifier of good products, product identifier of bad products, bad position, bad type, bad parameter.

3. The method according to claim 1, wherein The step of, for each quality inspection station, determining at least one type of relevant production process parameter related to this quality inspection result information from the at least one type of production process parameter corresponding to this quality inspection result information according to the quality inspection result information of this quality inspection station and the first product parameter value set of each type of production process parameter corresponding thereto includes: For each quality inspection station, constructing a quality inspection result variable of the first product set at this quality inspection station according to the quality inspection result information of this quality inspection station; For each quality inspection station, constructing a parameter variable of each type of production process parameter for the first product set according to the first product parameter value set of each type of production process parameter corresponding to the quality inspection result information of this quality inspection station; For each quality inspection station, calculating the correlation measure between the quality inspection result variable corresponding to the quality inspection result information of this quality inspection station and the parameter variable of each type of production process parameter; For each quality inspection station, determining at least one type of relevant production process parameter related to this quality inspection result information from the at least one type of production process parameter according to the correlation measure between the quality inspection result variable corresponding to the quality inspection result information of this quality inspection station and the parameter variable of each type of production process parameter.

4. The method according to claim 3, characterized in that, The step of, for each quality inspection station, calculating the correlation measure between the quality inspection result variable corresponding to the quality inspection result information of this quality inspection station and the parameter variable of each type of production process parameter includes: for the quality inspection result information of each quality inspection station, performing the following steps: Using statistical hypothesis testing, calculate the first correlation degree between the quality inspection result variable corresponding to the quality inspection result information and the parameter variables of each type of production process parameter; Calculate the correlation coefficient between the quality inspection result variable corresponding to the quality inspection result information and the parameter variables of each type of production process parameter to determine the second correlation degree; Based on the training samples constructed according to the first product parameter value sets of each type of production process parameter and the sample labels constructed according to the quality inspection result information, train the product quality inspection model to determine the third correlation degree between the quality inspection result variable corresponding to the quality inspection result information and the parameter variables of each type of production process parameter; According to at least one of the first correlation degree, the second correlation degree, and the third correlation degree between the quality inspection result variable corresponding to the quality inspection result information and the parameter variables of each type of production process parameter, determine the correlation measure between the quality inspection result variable and the parameter variables of each type of production process parameter.

5. The method according to claim 4, characterized in that The step of determining the correlation measure between the quality inspection result variable and the parameter variables of each type of production process parameter according to at least one of the first correlation degree, the second correlation degree, and the third correlation degree between the quality inspection result variable corresponding to the quality inspection result information and the parameter variables of each type of production process parameter includes: Calculate the weighted sum of the first correlation degree, the second correlation degree, and the third correlation degree based on a preset weight set; Determine the correlation measure between the quality inspection result variable and the parameter variables of each type of production process parameter according to the weighted sum.

6. The method according to claim 1, characterized in that, For each quality inspection site, according to the quality inspection result information of the quality inspection site and the first product parameter value sets of each of the at least one type of relevant production process parameter, determine the set of anomaly detection strategies corresponding to the at least one type of relevant production process parameter and its precision rate and recall rate for the first product set, including: For each quality inspection site, according to the quality inspection result information of the quality inspection site and the first product parameter value sets of each corresponding type of relevant production process parameter, determine the anomaly detection strategy for the relevant process parameter type; For each quality inspection site, based on the anomaly detection strategies of each of the at least one type of relevant production process parameter, determine the set of anomaly detection strategies corresponding to the at least one type of production process parameter; For each quality inspection site, based on the quality inspection result information of the quality inspection site, the set of anomaly detection strategies corresponding to the at least one type of relevant production process parameter, and the first product parameter value sets of each type of relevant production process parameter, determine the precision rate and recall rate of the set of anomaly detection strategies for the first product set.

7. The method according to claim 6, characterized in that, The step of determining the anomaly detection strategy for the relevant process parameter type for each quality inspection site according to the quality inspection result information of the quality inspection site and the first product parameter value sets of each type of relevant production process parameter includes: For each quality inspection site, perform at least one of the following steps: Construct a normal distribution according to the quality inspection result information of the quality inspection site and the first product parameter value sets of each corresponding type of relevant production process parameter to determine the first outlier range for the relevant production process parameter type; Construct a decision tree classification model based on the quality inspection result information of the quality inspection site and the first set of product parameter values for each corresponding type of relevant production process parameter to determine the second outlier range of the relevant production process parameter type; Construct an anomaly detection model based on the quality inspection result information of the quality inspection site and the first set of product parameter values for each corresponding type of relevant production process parameter, which is used to determine whether the parameter values of the relevant production process parameter type are abnormal.

8. The method according to claim 7, wherein The constructing of the anomaly detection model based on the quality inspection result information of the quality inspection site and the first set of product parameter values for each corresponding type of relevant production process parameter includes: Based on the quality inspection result information of the quality inspection site and the first set of product parameter values for each corresponding type of relevant production process parameter, construct an initial anomaly detection model based on the clustering algorithm and / or the isolation forest algorithm; Optimize the parameters of the initial anomaly detection model based on the Bayesian optimization algorithm to obtain an anomaly detection model for determining whether the parameter values of the relevant production process parameter type are abnormal.

9. The method according to claim 6, wherein For each quality inspection site, based on the quality inspection result information of the quality inspection site, the set of anomaly detection strategies corresponding to the at least one type of relevant production process parameter, and the first set of product parameter values for each type of relevant production process parameter, determine the precision and recall rate of the set of anomaly detection strategies for the first product set, including: For the quality inspection result information of each quality inspection site, perform the following steps: Based on the set of anomaly detection strategies for the at least one type of relevant production process parameter and the first set of product parameter values for each type of relevant production process parameter, determine the first set of abnormal parameter values from the first set of product parameter values of the relevant production process parameter type; Based on the first set of abnormal parameter values for each type of relevant production process parameter, determine the first information of the defective products determined in the first product set, where the first information includes the first identifier set and the first quantity of the defective products determined; Based on the quality inspection result information, determine the second information of the actual defective products in the first product set, where the second information includes the second identifier set and the second quantity of the actual defective products; Based on the first identifier set and the second identifier set, determine the third quantity of the defective products correctly determined in the first product set; Calculate the quotient of the third quantity and the first quantity to obtain the precision rate of the set of anomaly detection strategies for the first product set; Calculate the quotient of the third quantity and the second quantity to obtain the recall rate of the set of anomaly detection strategies for the first product set.

10. The method according to claim 1, wherein For each quality inspection site, based on the set of anomaly detection strategies and its precision rate and recall rate for the first product set, determine the sampling inspection strategy of the quality inspection site for the second product set, including: For each quality inspection site, perform the following steps: Obtain the second set of product parameter values for each of the at least one type of relevant production process parameter involved in the production process before the second product set enters the quality inspection site; Based on the set of anomaly detection strategies, determine the second set of abnormal parameter values from the second set of product parameter values for each type of relevant production process parameter; Based on the second set of abnormal parameter values for each type of relevant production process parameter, determine the third information of the abnormal products determined in the second product set and the fourth information of the normal products determined, where the third information includes the third identifier set and the fourth quantity of the abnormal products determined, and the fourth information includes the fourth identifier set and the fifth quantity of the normal products determined; Calculate the first sampling quantity for the abnormal products determined at this quality inspection site based on the product of the fourth quantity and the precision rate of the abnormal detection strategy set for the first product set; Based on the product of the fifth quantity and the recall rate of the abnormal detection strategy set for the first product set and the fifth quantity, calculate the second sampling quantity for the normal products determined at this quality inspection site; Determine the sampling strategy for the second product set at this quality inspection site according to the first sampling quantity, the second sampling quantity, the third identifier set, and the fourth identifier set. The type of production process parameter includes at least one of the following: product resume information, production site type, production equipment parameters, production environment parameters, key time node information, product quality monitoring parameters.

11. The method according to claim 1, wherein The collection of the quality inspection result information of each quality inspection site for the first product set and the production process information corresponding to the quality inspection result information of each quality inspection site during the product production process includes:

12. The method according to claim 1, wherein Real-time collect the production process data of each production site and the quality inspection result data of each quality inspection site during the product production process and synchronize them to the message queue to respectively form a process database and a quality inspection database; Regularly fuse the data in the process database and the quality inspection database according to the product identifier to obtain a fused database; and Collect the quality inspection result information of each quality inspection site for the first product set and the production process information corresponding to each quality inspection result information from the fused database.

13. A product quality inspection device, comprising: A data collection module configured to collect the quality inspection result information of each quality inspection site for the first product set and the production process information corresponding to the quality inspection result information of each quality inspection site during the product production process, where the production process information corresponding to the quality inspection result information of each quality inspection site includes at least one type of production process parameter involved in the production process before the first product set enters this quality inspection site and the first product parameter value set of each type of production process parameter; A correlation analysis module configured to, for each quality inspection site, determine at least one type of relevant production process parameter type related to the quality inspection result information from at least one type of production process parameter type corresponding to the quality inspection result information according to the quality inspection result information of this quality inspection site and the first product parameter value set of each type of production process parameter; An abnormal detection strategy establishment module configured to, for each quality inspection site, determine the abnormal detection strategy set corresponding to at least one type of relevant production process parameter type, its precision rate and recall rate for the first product set according to the quality inspection result information of this quality inspection site and the first product parameter value set of each type of production process parameter in at least one type of relevant production process parameter type; ​ ​ A sampling inspection strategy determination module, configured to determine, for each quality inspection site, a sampling inspection strategy of the quality inspection site for a second product set based on the abnormal detection strategy set and its precision rate and recall rate for a first product set.

14. A computing device, comprising: a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, it causes the processor to execute the method according to any one of claims 1-12.

15. A computer-readable storage medium, storing computer-readable instructions thereon, and when the computer-readable instructions are executed, the method according to any one of claims 1-12 is implemented.

16. A computer program product, comprising a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1-12 are implemented.

Images