An industrial internet-based cross-product quality inspection method and device

Abstract

The embodiment of the application discloses a cross-product quality inspection method and device based on an industrial internet, relates to the technical field of household appliance production quality analysis, and comprises the following steps: a first quality inspection knowledge base for guiding source product quality inspection is constructed; the first quality inspection knowledge base comprises first node chain information corresponding to quality inspection items of the source product and a first causal edge strategy; a target product to be constructed in a library is queried, and second node chain information corresponding to quality inspection items of the target product is acquired; if the first node chain information and the second node chain information are the same, key parameters of a causal relationship logic function in the first causal edge strategy are adjusted based on quantitative differences of the source product and the target product in physical characteristics, production environments and process control plans, and a second causal edge strategy is obtained; a second quality inspection knowledge base is constructed according to the second node chain information and the second causal edge strategy; and quality inspection of quality inspection items of the target product is performed based on the second quality inspection knowledge base, so that the construction efficiency of the quality inspection knowledge base is improved.

Description

Technical Field

[0001] This application relates to the field of home appliance production quality analysis technology, and in particular to a cross-product quality inspection method and device based on the Industrial Internet. Background Technology

[0002] In the product manufacturing process, products typically undergo processing on multiple production lines, with each line treated as an independent quality inspection item. Only after manufacturing is complete and all production lines have passed quality inspection is the product allowed to leave the factory. During the quality inspection of each production line, quality inspectors can repair any faulty points to ensure the product meets quality standards. Traditional quality inspection methods do not trace the root cause of faults, preventing quality inspectors from addressing the problem at its source. This leads to recurring and frequent faults, increasing quality inspection costs and reducing production efficiency.

[0003] To improve quality inspection and production efficiency, the quality inspection system can build a quality inspection knowledge base based on product failure mode and effects analysis (FMEA) databases, electronic control plans, historical failure cases, and other resources. This knowledge base guides the node topology and causal relationships of the quality inspection processes on each production line. In this way, the quality inspection system can trace the root causes of failures based on the knowledge base, thereby identifying problems at their source and resolving them more effectively.

[0004] For different types of products (such as air conditioners and televisions), there are differences in quality inspection items, node chain topology, causal relationships between nodes, and testing standards. The quality inspection system needs to build an independent quality inspection knowledge base for each product based on its own production line and quality inspection requirements. However, building a quality inspection knowledge base requires a large amount of computing power from the Industrial Internet and is time-consuming, which will reduce the efficiency of product quality inspection and production. Summary of the Invention

[0005] Some embodiments of this application provide a cross-product quality inspection method and apparatus based on the Industrial Internet. By performing topological matching on the node chains of quality inspection items of the source product and the target product, when the migration conditions are met, the relevant causal relationship logic functions in the quality inspection knowledge base of the source product are fine-tuned based on the quantitative differences in quality inspection standards / environments between the source product and the target product, thereby obtaining a quality inspection knowledge base adapted to the target product. This enables cross-product migration and reuse of the quality inspection knowledge base, quickly builds the quality inspection knowledge base of the target product, improves the efficiency of database construction, quality inspection, and production, and reduces the consumption of industrial internet computing power in building the quality inspection knowledge base.

[0006] Firstly, some embodiments of this application provide a cross-product quality inspection method based on the Industrial Internet, including:

[0007] Construct a first quality inspection knowledge base for guiding the quality inspection of source products; wherein, the first quality inspection knowledge base includes first node chain information and first causal edge strategy corresponding to the quality inspection items of the source products, the first node chain information includes the topology and node type of the first node chain, and the first causal edge strategy includes the causal relationship logic function between each node in the first node chain;

[0008] Query the target product in the database to be built, and obtain the second node chain information corresponding to the quality inspection items of the target product; wherein, the target product and the source product have different equipment types, quality inspection items and quality inspection standards, and the second node chain information includes the topology and node type of the second node chain;

[0009] If the first node chain information and the second node chain information are the same, then based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment and process control plan, the key parameters of the causal relationship logic function in the first causal edge strategy are adjusted to obtain the adjusted second causal edge strategy.

[0010] Based on the second node chain information and the second causal edge strategy, a second quality inspection knowledge base is constructed;

[0011] Based on the second quality inspection knowledge base, quality inspection is carried out on the quality inspection items of the target product.

[0012] The beneficial effects of the above-mentioned first aspect of the embodiments are as follows: First, a first quality inspection knowledge base is constructed for the source product, and prior knowledge in the first quality inspection knowledge base is transferred to the target product. However, due to differences in equipment type, quality inspection items, and quality inspection standards (including environmental standards, process standards, etc.) between the source product and the target product, there is a pain point in the difficulty of reusing knowledge between heterogeneous products / production lines. Therefore, this application analyzes cross-product production lines with the same node chain topology. By stripping away the superficial differences in products, prior knowledge of the same topology production line is transferred to the target production line, maintaining the causal edge logic. The objective process laws of the function remain unchanged. Only the values ​​of certain key parameters in the function need to be fine-tuned to achieve the reuse of quality inspection knowledge across products / production lines. By quantifying the differences between the source product production line and the target product production line, these differences are integrated into the changes of the function's logical parameters, improving the accuracy of database construction. This enables the rapid construction of a quality inspection knowledge base for the target product, improving database construction efficiency, quality inspection efficiency, and production efficiency. It overcomes the challenges of knowledge transfer across products, production lines, processes, and fields. Furthermore, the overall control process is based on the automated execution of the Industrial Internet, reducing the computing power load and overhead of the Industrial Internet.

[0013] In some embodiments of the first aspect, the node includes: a first input node representing a usage intensity index, a second input node representing an environmental disturbance index, and an intermediate node representing a product degradation state; the causal relationship logic function includes: a first function representing the causal relationship between the first input node and the intermediate node in the first node chain, and a second function representing the causal relationship between the second input node and the intermediate node in the first node chain; if the first node chain information and the second node chain information are the same, then based on the quantitative differences between the source product and the target product in physical characteristics, production environment, and process control plan, the key parameters of the causal relationship logic function in the first causal edge strategy are adjusted, specifically including: based on the quantitative differences between the source product and the target product in physical characteristics, production environment, and process control plan, adjusting the first degradation sensitivity parameter and the first critical usage amount included in the first function; based on the quantitative differences between the source product and the target product in the production environment, adjusting the first environmental sensitivity parameter included in the second function.

[0014] The beneficial effects of this embodiment are as follows: The causal logic and operational relationships of the first function remain unchanged. Only the quantitative differences between the source product and the target product in terms of physical characteristics, production environment, and process control plan need to be adjusted to adjust relevant influencing factors (such as degradation sensitivity parameters and critical usage amounts), thereby adapting the first function to the quality inspection requirements of the target product production line. Similarly, the causal logic and operational relationships of the second function remain unchanged. Only the environmental sensitivity parameters need to be adjusted based on the quantitative differences between the source product and the product in terms of production environment, thereby adapting the second function to the quality inspection requirements of the target product production line. In this way, while keeping the objective process laws of the causal edges unchanged, the key parameter values ​​of the causal edge strategy on the target product production line are automatically adapted and rewritten based on multi-dimensional quantitative differences, improving the adaptability of transferred knowledge to the target product production line, thereby overcoming the knowledge transfer barriers across products / production lines, and balancing the effectiveness of database construction with the adaptability of the quality inspection knowledge base.

[0015] In some embodiments of the first aspect, adjusting the first degradation sensitivity parameter and the first critical usage amount included in the first function based on the quantitative differences in physical characteristics, production environment, and process control plan between the source product and the target product specifically includes: calculating a first coefficient to characterize the quantitative differences in the production environment based on a first environmental parameter benchmark value mapped to the quality inspection items of the source product and a second environmental parameter benchmark value mapped to the quality inspection items of the target product; calculating a second coefficient to characterize the quantitative differences in physical characteristics based on a first equipment sensitivity score mapped to the source product and a second equipment sensitivity score mapped to the target product; calculating a third coefficient to characterize the quantitative differences in the process control plan based on a first process control threshold used in the quality inspection items of the source product and a second process control threshold used in the quality inspection items of the target product; calculating a second degradation sensitivity parameter based on the first coefficient, the second coefficient, the third coefficient, and the first degradation sensitivity parameter; obtaining a second critical usage amount by performing function fitting through a logistic regression algorithm based on the second degradation sensitivity parameter; adjusting the first degradation sensitivity parameter in the first function to the second degradation sensitivity parameter; and adjusting the first critical usage amount to the second critical usage amount.

[0016] The beneficial effects of this embodiment are as follows: It quantifies the differences in production environment across products by using environmental parameter benchmarks for the source and target products; it quantifies the differences in physical characteristics across products by using equipment sensitivity scores for the source and target products; and it quantifies the differences in processes across products by using process control thresholds for the source and target products. This quantifies the objective differences between the original and target products, and then uses these quantified differences to drive the rewriting of the degradation sensitivity parameters and critical usage in the first function. The causal logic of equipment usage intensity → state degradation remains unchanged before and after migration, improving the adaptability of the migrated knowledge to the target product production line, and enhancing migration feasibility and database accuracy. Through multi-dimensional physical parameter correction, the experiential knowledge of the source production line is accurately transferred to the target production line, avoiding engineering distortions caused by "numerical transfer" of prior knowledge.

[0017] In some embodiments of the first aspect, adjusting the first environmental sensitivity parameter included in the second function based on the quantitative difference between the source product and the target product in the production environment specifically includes: calculating a second environmental sensitivity parameter based on the first coefficient and the first environmental sensitivity parameter; and adjusting the first environmental sensitivity parameter in the second function to the second environmental sensitivity parameter.

[0018] The beneficial effects of this embodiment are as follows: By quantifying the differences in production environments across products using benchmark values ​​of environmental parameters for the source and target products, the changes in environmental sensitivity parameters in the second function are driven by these quantified differences. The causal logic of environmental disturbance → state degradation remains unchanged before and after migration, improving the adaptability of the transferred knowledge to the target product production line, and enhancing migration feasibility and database accuracy. Through multi-dimensional physical parameter correction, the experiential knowledge of the source production line is accurately transferred to the target production line, avoiding engineering distortions caused by "numerical transfer" of prior knowledge.

[0019] In some embodiments of the first aspect, the node further includes a first output node characterizing the missed detection rate index and a second output node characterizing the final inspection defect rate index; the causal relationship logic function further includes a third function characterizing the causal relationship between the intermediate node and the first output node in the first node chain, and a fourth function characterizing the causal relationship between the first output node and the second output node in the first node chain, the third function including a first detectivity benchmark value, and the fourth function including a first severity value; after adjusting the key parameters of the causal relationship logic function in the first causal edge strategy based on the quantitative differences in physical characteristics, production environment, and process control plan between the source product and the target product if the first node chain information and the second node chain information are the same, the method further includes: obtaining a second detectivity benchmark value and a second severity value mapped to the quality inspection items of the target product from the failure mode and effects analysis database of the target product; adjusting the first detectivity benchmark value in the third function to the second detectivity benchmark value; and adjusting the first severity value in the fourth function to the second severity value.

[0020] The second causal edge strategy is generated based on the updated first function, second function, third function, and fourth function.

[0021] The beneficial effects of this embodiment are as follows: by obtaining the detectivity D value and severity S value corresponding to the target product production line from the FMEA database, keeping the basic logical relationship and operation relationship of the third and fourth functions unchanged, only the S value and D value in the function need to be changed, thereby improving the adaptability of the transferred knowledge to the target product production line, improving the feasibility of the transfer and the accuracy of the database construction, and avoiding the engineering distortion caused by "numerical transfer" of prior knowledge.

[0022] In some embodiments of the first aspect, the node further includes a constraint node for constraining the process control of product quality inspection items; before adjusting the key parameters of the causal relationship logic function in the first causal edge strategy based on the quantitative differences in physical characteristics, production environment, and process control plan between the source product and the target product if the first node chain information and the second node chain information are the same, the method further includes: obtaining the second process control threshold used for the quality inspection items of the target product from the electronic control plan file of the target product; and setting the adjustment range of the control parameters of the constraint node in the second node chain based on the second process control threshold.

[0023] The beneficial effects of this embodiment are as follows: In addition to modifying the key parameters in the logic function of the causal edge, for the constraint node, the control threshold of the constraint node is also changed based on the electronic control plan of the target product, thereby ensuring that the target product production line can accurately constrain the input node according to the control threshold, improving the feasibility of migration and the accuracy of database construction, and avoiding engineering distortion caused by "numerical transfer" of prior knowledge.

[0024] In some embodiments of the first aspect, after constructing a second quality inspection knowledge base based on the second node chain information and the second causal edge strategy, the method further includes: obtaining a perturbation variable affecting the final inspection defect rate of the quality inspection items of the target product; sampling the historical distribution data of the perturbation variable to generate a preset number of first test samples, wherein the first test samples include the sampled value of the perturbation variable and a first target value of the parameter to be deduced; using the second causal edge strategy in the second quality inspection knowledge base to deduce the preset number of first test samples to obtain a first final inspection defect rate mapped to each first test sample; calculating the mean increase of the first final inspection defect rate mapped to the preset number of first test samples; if the mean increase is greater than a first threshold, then selecting a second target value between the migration value of the parameter to be deduced in the second quality inspection knowledge base and the first target value, and changing the value of the parameter to be deduced to the second target value.

[0025] The beneficial effects of this embodiment are as follows: By performing computer simulation and deduction on the transferred causal graph knowledge, and by quantifying uncertainty disturbances, the deduction is driven by the causal graph, enabling the overall quality inspection knowledge base system to have sustainable learning and evolution capabilities. Through continuous improvement and optimization of knowledge parameters, the accuracy and engineering adaptability of the quality inspection knowledge base are enhanced. If the mean increase exceeds a first threshold, the quality deduction results are automatically converted into a quantitative, risk-controllable, and executable fixed threshold optimization strategy, which is easy to implement and monitor.

[0026] In some embodiments of the first aspect, after constructing a second quality inspection knowledge base based on the second node chain information and the second causal edge strategy, the method further includes: obtaining a perturbation variable affecting the final inspection defect rate of the quality inspection items of the target product; sampling the historical distribution data of the perturbation variable to generate a preset number of first test samples, wherein the first test samples include the sampled value of the perturbation variable and a first target value of the parameter to be deduced; using the second causal edge strategy in the second quality inspection knowledge base to deduce the preset number of first test samples to obtain a first final inspection defect rate mapped to each first test sample; calculating the mean and slope of the first final inspection defect rate mapped to the preset number of first test samples; if the mean is greater than a second threshold and the slope of the change is greater than a third threshold, then obtaining a sensitive critical value of the perturbation variable; calculating a third target value based on the migration value of the parameter to be deduced in the second quality inspection knowledge base, the current value of the perturbation variable, and the sensitive critical value, and changing the value of the parameter to be deduced to the third target value.

[0027] The beneficial effects of this embodiment are as follows: By performing computer simulation and deduction on the transferred causal graph knowledge, and by quantifying uncertainty disturbances, the deduction is driven by the causal graph, enabling the overall quality inspection knowledge base system to have sustainable learning and evolution capabilities. Through continuous improvement and optimization of knowledge parameters, the accuracy and engineering adaptability of the quality inspection knowledge base are enhanced. If the mean is greater than the second threshold and the slope of change is greater than the third threshold, it indicates the existence of a high-risk sensitive area. To avoid this risk, the dynamic value of the parameter to be deduced can be calculated using the transferred value of the parameter to be deduced, the current value of the disturbance variable, and the sensitive critical value. This allows for dynamic and adaptive allocation of values ​​to the parameter to be deduced, following the disturbance factors. The strategy is more refined and has stronger anti-interference capabilities.

[0028] In some embodiments of the first aspect, after performing quality inspection on the quality inspection items of the target product based on the second quality inspection knowledge base, the method further includes: obtaining a target final inspection defect rate obtained after performing quality inspection on the quality inspection items of the target product; calculating the deviation ratio between the target final inspection defect rate and the mean of the second final inspection defect rate; the second final inspection defect rate is obtained by extrapolating a preset number of second test samples through the second causal edge strategy in the second quality inspection knowledge base, the second test samples including the sampled values ​​of the disturbance variable and the changed values ​​of the parameter to be extrapolated; analyzing the target root cause that causes the deviation ratio to be greater than a fourth threshold; locating the target function that matches the target root cause in the second causal edge strategy; and refitting the target function using a logistic regression algorithm to update the key parameters in the target function.

[0029] The beneficial effects of this embodiment are as follows: the quality inspection system or cross-product quality inspection device can compare the target final inspection defect rate with the average final inspection defect rate derived by the inference engine. If the target final inspection defect rate deviates significantly from the average final inspection defect rate, automatic correction is triggered. By analyzing the root causes of the large deviation in the final inspection defect rate, the target function to be refitted can be located through these root causes. In this way, not only is the quantification of the final inspection defect rate deviation considered, but also the root causes of the deviation are located. By analyzing the source of the deviation, the root cause of the inference distortion can be accurately located, achieving efficient and accurate correction of the causal graph and timely deviation compensation. This "deviation-driven correction" mechanism gradually evolves from a simple reactive strategy into an intelligent self-learning process with judgment capabilities, root cause analysis, and self-evolution.

[0030] Secondly, some embodiments of this application also provide a cross-product quality inspection device based on the Industrial Internet, including:

[0031] The communicator is configured to communicate with the quality inspection execution unit via the Industrial Internet;

[0032] The quality inspection execution unit is configured to perform quality inspection on the target product;

[0033] The controller is configured as follows:

[0034] Construct a first quality inspection knowledge base for guiding the quality inspection of source products; wherein, the first quality inspection knowledge base includes first node chain information and first causal edge strategy corresponding to the quality inspection items of the source products, the first node chain information includes the topology and node type of the first node chain, and the first causal edge strategy includes the causal relationship logic function between each node in the first node chain;

[0035] Query the target product in the database to be built, and obtain the second node chain information corresponding to the quality inspection items of the target product; wherein, the target product and the source product have different equipment types, quality inspection items and quality inspection standards, and the second node chain information includes the topology and node type of the second node chain;

[0036] If the first node chain information and the second node chain information are the same, then based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment and process control plan, the key parameters of the causal relationship logic function in the first causal edge strategy are adjusted to obtain the adjusted second causal edge strategy.

[0037] Based on the second node chain information and the second causal edge strategy, a second quality inspection knowledge base is constructed;

[0038] The second quality inspection knowledge base is synchronized to the quality inspection execution unit via the communicator, so that the quality inspection execution unit can perform quality inspection on the quality inspection items of the target product based on the second quality inspection knowledge base.

[0039] The beneficial effects of the embodiments in the second aspect above are as follows: First, a first quality inspection knowledge base is constructed for the source product. Prior knowledge from this knowledge base is then transferred to the target product. However, due to differences in equipment type, quality inspection items, and quality inspection standards (including environmental standards, process standards, etc.) between the source and target products, there is a pain point in the difficulty of reusing knowledge between heterogeneous products / production lines. Therefore, this application analyzes cross-product production lines with the same node chain topology. By stripping away the superficial differences in products, prior knowledge from the same topology production line is transferred to the target production line, maintaining the causal edge logic. The objective process laws of the function remain unchanged. Only the values ​​of certain key parameters in the function need to be fine-tuned to achieve the reuse of quality inspection knowledge across products / production lines. By quantifying the differences between the source product production line and the target product production line, these differences are integrated into the changes of the function's logical parameters, improving the accuracy of database construction. This enables the rapid construction of a quality inspection knowledge base for the target product, improving database construction efficiency, quality inspection efficiency, and production efficiency. It overcomes the challenges of knowledge transfer across products, production lines, processes, and fields. Furthermore, the overall control process is based on the automated execution of the Industrial Internet, reducing the computing power load and overhead of the Industrial Internet. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in some embodiments of this application or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A schematic diagram illustrating the construction of a quality inspection knowledge base for source products based on a data source, provided for some embodiments of this application;

[0042] Figure 2 A schematic diagram of the node chain topology of the air conditioning-air tightness testing project provided in some embodiments of this application;

[0043] Figure 3 A flowchart illustrating a cross-product quality inspection method based on the Industrial Internet provided for some embodiments of this application;

[0044] Figure 4 A schematic diagram of the node chain topology of a television-automatic optical inspection camera visual inspection project provided in some embodiments of this application;

[0045] Figure 5A schematic diagram illustrating the dynamic adaptation principle of cross-product / production line knowledge transfer provided for some embodiments of this application;

[0046] Figure 6 A flowchart illustrating the deduction and optimization method of the quality inspection knowledge base provided in some embodiments of this application;

[0047] Figure 7 A schematic diagram illustrating the principle of causal graph correction triggered by inference-actual deviation quantization provided for some embodiments of this application;

[0048] Figure 8 This is a structural block diagram of a cross-product quality inspection device based on the Industrial Internet, provided for some embodiments of this application. Detailed Implementation

[0049] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described below do not represent all embodiments consistent with this application, but are merely examples of systems and methods consistent with some aspects of this application as detailed in the claims.

[0050] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0051] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.

[0052] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.

[0053] In the product manufacturing process, products typically undergo processing on multiple production lines, with each line treated as an independent quality inspection item. Only after manufacturing is complete and all production lines have passed quality inspection is the product allowed to leave the factory. During the quality inspection of each production line, quality inspectors can repair any faulty points to ensure the product meets quality standards. Traditional quality inspection methods do not trace the root cause of faults, preventing quality inspectors from addressing the problem at its source. This leads to recurring and frequent faults, increasing quality inspection costs and reducing production efficiency.

[0054] In some embodiments, to improve quality inspection and production efficiency, the quality inspection system can build a quality inspection knowledge base based on the product's FMEA database, electronic control plans, historical failure cases, and other content. This knowledge base guides the node topology and causal relationships between nodes in the quality inspection process of each production line. In this way, the quality inspection system can trace the root causes of failures based on the knowledge base, thereby identifying problems at their source and resolving them more effectively.

[0055] Figure 1 This is a schematic diagram illustrating the construction of a quality inspection knowledge base for source products based on a data source, as provided in some embodiments of this application.

[0056] In some embodiments, such as Figure 1 As shown, when building a quality inspection knowledge base for source products, it is necessary to acquire reserve knowledge from various data sources and build the source product quality inspection knowledge base based on this reserve knowledge. These data sources include, but are not limited to: FMEA databases, electronic control plans, root cause knowledge bases, and equipment parameter logs.

[0057] In some embodiments, the FMEA database is used to analyze the failure modes, potential causes, effects, and hazards of products / processes, forming a structured processing flow that includes modules such as functional analysis and failure mode assessment, and integrates a fault database with customized specifications. The FMEA database assists in product quality inspection, identifying potential failure modes in products / processes, quantifying the risks of these potential failure modes, listing the causes / mechanisms of failures, and identifying preventative or improvement measures. The quality inspection system can extract "potential causes" as root cause nodes and "failure modes" as defect nodes from the FMEA database to build a root cause knowledge base.

[0058] In some embodiments, the FMEA database records S / O / D values, where S represents Severity, characterizing the degree of impact of the failure mode on product quality, customer experience, or safety; O represents Occurrence, characterizing the likelihood of the failure mode occurring during manufacturing or use; and D represents Detection, characterizing the ease with which current control measures can detect the failure mode.

[0059] In some embodiments, within the manufacturing field, an Electronic Control Plan (ECP) is typically used to define the control characteristics, control methods, and response plans of a product / process to prevent defects and ensure product quality meets requirements. Control characteristics refer to the key parameters that need to be monitored during product manufacturing. Control methods refer to the specific methods used during production to monitor and ensure that control characteristics are under control. Response plans refer to the corrective and containment measures that quality control personnel should take when anomalies are detected during product quality inspection, to prevent non-conforming products from flowing into the next process or leaving the factory.

[0060] In some embodiments, the root cause knowledge base (also known as the Ka root cause knowledge base) includes descriptions of a-nodes, descriptions of K-nodes, defect phenomena, handling records, and historical case data. Here, a-nodes are the source nodes causing product defects, and K-nodes are the failure checkpoints. After tracing back to a-node, starting from a-node and ending at the node where the defect was discovered, all intermediate nodes are traversed forward along the process topology network. From these intermediate nodes, nodes belonging to the quality control type are selected. Quality control types include verification, testing, and audit / review. The actual execution records of the selected nodes are compared with the expected interception action library. If there are cases of non-execution, invalid execution, or missing standards, the selected node is marked as a K-node.

[0061] In some embodiments, the device parameter log includes usage intensity-related parameters of the product device, calibration records, historical data from environmental sensors, etc. The root cause knowledge base and device parameter log can be used to fit the coefficients of the causal relationship logic function in the causal edge strategy, ensuring that the causal edge strategy possesses physical interpretability and engineering credibility.

[0062] In some embodiments, such as Figure 1 As shown, the quality inspection system can construct a quality inspection knowledge base based on the reserve knowledge obtained from various data sources. The quality inspection knowledge base can include M causal graph files. Assuming that i represents the production line number of the source product, 1≤i≤M, and M represents the total number of production lines of the source product, then the causal graph file i maps to the quality inspection item corresponding to the i-th production line of the source product to be inspected. The causal graph file i includes node chain information i, causal edge strategy i, and metadata i.

[0063] In some embodiments, node chain information i includes the topology and node information of node chain i. Node chain i contains N nodes, where j represents the node index number, 1 ≤ j ≤ N. Each node j has independent node information. The relationships (e.g., causal relationships, constraint relationships, etc.) and flow directions among the N nodes constitute the topology of node chain i. Figure 1In the example, the node chain information i includes nodes S1, S2, S3, S4, S5, and S6, i.e., N=6.

[0064] In some embodiments, node information includes node types. Node types may include: a1 input nodes, a2 intermediate nodes, a3 output nodes, and a4 constraint nodes, where a1 + a2 + a3 + a4 = N.

[0065] In some embodiments, the quality inspection system can extract "control characteristics" as input nodes from the electronic control plan and usage intensity indicators as input nodes from the equipment parameter log. For example, taking an air conditioner as the source product and an airtightness test as the quality inspection item, this quality inspection item includes three input nodes. Input node 1 is named "calibration cycle (h)", input node 2 is named "cumulative number of tests (times)", and input node 3 is named "workshop temperature and humidity (°C / %)". The calibration cycle and cumulative number of tests are indicators of air conditioner usage intensity, while workshop temperature and humidity are control characteristic indicators.

[0066] In some embodiments, the quality inspection system can extract equipment status indicators based on the "potential causes" in the FMEA database and the 'a' node in the Ka root cause knowledge base, thereby generating intermediate nodes. The standardized naming structure of intermediate nodes is, for example, [Equipment Type]_[Status Indicator Description], ensuring semantic consistency of nodes when migrating across products / production lines. Taking an air conditioner as the source product and airtightness testing as the quality inspection item, the intermediate node is named "Helium Detector_Sensor Drift Probability," representing the impact of the input node on the sensor drift probability of the helium detector.

[0067] In some embodiments, the quality inspection system can extract indicators to measure production line quality based on the "failure modes" in the FMEA database and the "defect phenomena" in the Ka root cause knowledge base, thereby generating output nodes. Taking air conditioners as the source product and airtightness testing as the quality inspection item, the name of output node 1 is "missed inspection rate (%)", and the name of output node 2 is "commercial inspection_refrigerant leakage rate". Output node 1 represents the impact of the intermediate node's status indicator on the missed inspection rate, and output node 2 represents the impact of the missed inspection rate of output node 1 on the commercial inspection_refrigerant leakage rate (which belongs to the final inspection defect rate).

[0068] In some embodiments, the quality inspection system can extract threshold parameters from the "reaction plan" in the electronic control plan, thereby generating constraint nodes. These constraint nodes are used to constrain and control relevant processes in the quality inspection items of the source product. The names of the constraint nodes can adopt a standardized structure, such as parameter name ≤ threshold parameter. Taking an air conditioner as the source product and an airtightness test as an example, the name of the constraint node could be "calibration cycle ≤ 500h". This constraint node indicates that the parameter of the "calibration cycle" node is controlled within a range not exceeding 500h.

[0069] In some embodiments, the quality inspection system can obtain the causal relationships (or causal edges) between nodes based on the failure chains corresponding to "potential causes-failure modes" in the FMEA database. For example, helium detector sensor drift (potential cause) leads to missed detection of airtightness (failure mode).

[0070] In some embodiments, the quality inspection system can obtain the causal relationship between nodes based on the causal labeling of node a and defect phenomena in the Ka root cause database. For example, helium detector sensor drift (node ​​a) leads to missed detection in airtightness testing (defect phenomenon), and the missed detection ultimately leads to refrigerant leakage during commercial inspection (defect phenomenon).

[0071] In some embodiments, the quality inspection system can obtain the constraint relationships and control plans between nodes based on the trigger conditions of the "control method-response plan" in the electronic control plan. For example, equipment locking is triggered when the calibration cycle exceeds 500 hours. By converting documents from different data sources into a deductive and evolutionary causal graph model, quality knowledge is digitally stored, facilitating retrieval and execution of quality inspections.

[0072] Figure 2 This is a schematic diagram of the node chain topology of the air conditioning-air tightness testing project provided in some embodiments of this application.

[0073] In some embodiments, the quality inspection system can construct the topology of node chain i based on the relationships (e.g., causal relationships, constraint relationships, etc.) and flow directions among N nodes. Taking an air conditioner as the source product and an airtightness inspection item as the quality inspection item as an example, see [link to relevant documentation]. Figure 2 The node chain topology of this quality inspection project includes input node 1 (calibration cycle), input node 2 (cumulative number of tests), input node 3 (workshop temperature and humidity), intermediate node (helium detector_sensor drift probability), output node 1 (missed detection rate), output node 2 (commercial inspection_refrigerant leakage rate) and constraint node (calibration cycle ≤ 500h).

[0074] In some embodiments, such as Figure 2As shown, the calibration cycle is one of the metrics for usage intensity. Higher usage intensity leads to greater product degradation, resulting in a higher probability of sensor drift in the helium detector. Therefore, input node 1 and the intermediate node form a causal edge. The cumulative number of tests is another metric for usage intensity. More cumulative tests result in a higher probability of sensor drift in the helium detector. Therefore, input node 2 and the intermediate node also form a causal edge. Workshop temperature and humidity are environmental factors on the production line. For example, high temperature and high humidity environments can exacerbate product degradation, leading to an increased probability of sensor drift in the helium detector. Therefore, input node 3 and the intermediate node also form a causal relationship.

[0075] In some embodiments, such as Figure 2 As shown, the sensor drift probability of the helium detector is an indicator of the deterioration state of the source product. The higher the sensor drift probability of the helium detector, the more severe the deterioration of the source product, which leads to a higher false negative rate and a higher final inspection defect rate. Therefore, the intermediate node and output node 1 form a causal edge. False negative products flow into the commodity inspection department, leading to the detection of refrigerant leaks. Therefore, the false negative rate affects the commodity inspection refrigerant leak rate, so output node 1 and output node 2 also form a causal edge.

[0076] In some embodiments, such as Figure 2 As shown, the constraint node is used to constrain the calibration cycle and control the calibration cycle to not exceed a threshold (e.g., 500h).

[0077] In some embodiments, the quality inspection system can bind causal relationship logic functions from the quality causal function library to each causal edge based on the causal edge type. The causal edge types include, but are not limited to: ① equipment usage intensity → condition degradation, ② environmental disturbance → condition degradation, ③ condition degradation → missed detection rate, and ④ missed detection rate → final inspection defect rate. The functions called for each type of causal edge are shown in Table 1.

[0078] Table 1

[0079]

[0080] In Table 1, the degradation sensitivity parameter *k* characterizes the degree of influence of equipment usage intensity on equipment degradation; a larger *k* value indicates faster equipment degradation. *X0* is the critical usage level; when the equipment usage intensity exceeds *X0*, the risk of degradation increases dramatically. *α* represents the state change caused by a unit environmental disturbance, characterizing the degree of influence of environmental disturbance on equipment degradation; a larger *α* value indicates faster equipment degradation. The quality causal function library is specifically designed for the manufacturing quality field. Its function forms are based on engineering physics principles (e.g., using the Logistic regression function to describe the nonlinear characteristics of equipment degradation), the parameters have clear physical meanings, and the coefficient fitting strictly depends on historical case data for *Ka*. The use of random initialization models without a source or black-box machine learning models is prohibited.

[0081] In some embodiments, the coefficient fitting of the "equipment usage intensity → condition deterioration" function is taken as an example:

[0082] 1. Fitting data screening: Select several relevant cases from the Ka root cause knowledge base that contain "cumulative number of tests before calibration" and "verification results after calibration".

[0083] 2. Label definition: If the verification fails after calibration, set the label to 1 (e.g., drift has occurred); if the verification passes after calibration, set the label to 0 (e.g., no drift has occurred).

[0084] 3. Logistic Regression: Using the cumulative number of detections X as the independent variable and the label value as the dependent variable, a function is fitted to obtain P_drift=1 / (1+ e^(-0.02·(X-420))), thus outputting k= 0.02 and X0=420.

[0085] 4. Confidence Level Labeling: Record the number of cases and the fitted R² = 0.87, labeling "Confidence Level" as "High". R² (Coefficient of Determination) is a core statistical indicator for measuring the goodness of fit of a regression model to observed data, representing the proportion of the variance in the dependent variable that can be explained by the independent variable. Its value ranges from [0, 1], with a value closer to 1 indicating a stronger explanatory power and better fit of the model.

[0086] 5. Source Binding: Write the case ID of the case from which the fitted coefficients originate in the attributes of the causal edge, for example, Ka-2025-112 to Ka-2026-034. The coefficient fitting process is forcibly associated with the ID of the specific Ka case, and the fitting result includes statistical indicators (such as R², number of cases, etc.) to ensure that the causal edge strategy is auditable and verifiable, avoiding the subjective defects of setting parameters out of thin air.

[0087] In some embodiments, the causal mapping file i can be stored as a structured document. For example... Figure 1 As shown, the causal graph file i includes: metadata i, node chain information i, and causal edge strategy i. The content of the causal graph file i can be described in the form of lists, tables, etc., to ensure readability, auditability, and system parsability. The naming convention for the causal graph file is: CG_[product line abbreviation]_[detection stage]_V[version number].txt, for example, CG_AC2026_LeakTest_V1.txt.

[0088] In some embodiments, metadata i is used to describe the basic attributes and source information of causal graph file i, and serves as the fixed content of the header of causal graph file i.

[0089] In some embodiments, metadata i includes, but is not limited to: ① Cause-effect graph ID (e.g., CG_AC2026_LeakTest_V1), ② Source production line name / ID (e.g., air conditioning assembly line / AC2026), ③ Applicable testing process (helium detector_air tightness test), ④ Construction time, ⑤ Construction basis document, ⑥ Construction subject (e.g., quality knowledge engine V2.1), ⑦ Current status (e.g., draft, under review, effective, archived, etc.), ⑧ Quality system compliance statement.

[0090] In some embodiments, each causal graph file contains a node chain, which includes N nodes arranged according to logical relationships (e.g., causal relationships, constraint relationships, etc.) to form a node chain topology. Node chain information i includes the node information of the N nodes in causal graph file i.

[0091] In some embodiments, node information includes, but is not limited to: node sequence number, node identifier / ID, node name, node type, data type (continuous or discrete), unit, data source and source location, associated Ka case, confidence level, and last update time. Only intermediate nodes and output nodes need to have associated Ka cases filled in; the associated Ka cases for input nodes and constraint nodes can be marked as "none".

[0092] In some embodiments, if the number of Ka cases associated with node j is greater than or equal to a first quantity (e.g., 20), and R 2 If the confidence level is greater than 0.8, then node j is marked as having high confidence. If the number of Ka cases associated with node j is within the range of [second number, first number], for example, the number of cases is between 10 and 19, and R... 2 If the number of cases associated with node j is less than the second number, and R... 2 If the confidence level is less than 0.6, then node j is marked as having low confidence.

[0093] In some embodiments, the causal edge strategy i is used to describe the causal relationships between nodes and the function form, etc. The causal edge strategy i includes, but is not limited to: ① causal edge sequence number, ② starting and target nodes of the causal edge (e.g., calibration cycle → helium detector sensor drift probability), ③ description of the causal relationship (e.g., extended calibration cycle leading to increased helium detector sensor drift probability), ④ causal edge type (e.g., equipment usage intensity → condition deterioration), ⑤ function form, ⑥ key parameter values ​​of the function, ⑦ fitting source of the function coefficients (e.g., Ka cases Ka-2025-112 to Ka-2026-034, historical environmental data and drift records, etc.), ⑧ fitting statistical indicators (e.g., number of cases, R², etc.). 2 (etc.), ⑨ confidence level, ⑩ trigger threshold of constraint node.

[0094] In some embodiments, for causal edges of the type "equipment usage intensity → condition deterioration", the key parameter values ​​of the function in causal edge strategy i should be labeled k and X0. For causal edges of the type "environmental disturbance → condition deterioration", the key parameter values ​​of the function should be labeled α and Env_base. For causal edges of the types "condition deterioration → missed detection rate" and "missed detection rate → final inspection defect rate", the key parameter values ​​of the function should be labeled with the S / O / D values ​​of the corresponding entries in the FMEA database.

[0095] For different types of products (such as air conditioners and televisions), there are differences in their quality inspection items, node chain topology, causal relationships between nodes, and quality inspection standards. The quality inspection system needs to build an independent quality inspection knowledge base for each product based on its own production line characteristics and quality inspection requirements. Referring to the above example, the quality inspection system needs to pre-build an FMEA database, a Ka root cause database, and electronic control plans, etc. After the multi-source data sources are ready, the quality inspection system also needs to coordinate the reserve knowledge from various different data sources to build node chains and causal edges, bind the functions called by each causal edge, and perform function coefficient fitting using algorithms such as logistic regression, configuring the key parameters of each function, etc. Therefore, building a quality inspection knowledge base is a very complex and high-load process, typically consuming a large amount of industrial internet computing power, increasing enterprise computing power costs, and taking a long time, reducing product quality inspection efficiency. Low quality inspection efficiency, in turn, leads to a decrease in enterprise production efficiency.

[0096] In response, this application provides a cross-product quality inspection method and apparatus based on the Industrial Internet. The technical concept is as follows: by performing topological matching on the node chain of quality inspection items of the source product and the target product, when it is determined that the migration conditions are met, the parameters of the relevant causal graph in the quality inspection knowledge base of the source product are fine-tuned based on the differences in quality inspection standards / environments between the source product and the target product, so as to obtain a quality inspection knowledge base adapted to the target product. This enables cross-product migration and reuse of the quality inspection knowledge base, quickly builds the quality inspection knowledge base of the target product, shortens the quality ramp-up time of new production lines, reduces the computing power consumption of the Industrial Internet, and improves the efficiency of database construction, quality inspection and production, thus providing a virtuous cycle for enterprise production and manufacturing.

[0097] In this embodiment of the application, the product for which a quality inspection knowledge base has been built is called the "source product", and the quality inspection knowledge base of the source product is named the "first quality inspection knowledge base"; the product for which a database to be built is to reuse the common prior knowledge of the quality inspection knowledge base of the source product is called the "target product", and the quality inspection knowledge base of the target product is named the "second quality inspection knowledge base".

[0098] The source product and the target product differ in equipment type, production line / quality inspection items, and quality inspection standards / requirements. The source product can be "white goods" (or "white appliances"), while the target product can be "black goods" (or "black appliances"). "White goods" typically refer to appliances used to replace household chores, such as washing machines, refrigerators, and air conditioners. These appliances were initially mostly white, hence the name. Their main function is to improve the living environment and quality of life, typically involving cooling, cleaning, and cooking. "Black goods," on the other hand, initially referred to entertainment and audiovisual products such as televisions, radios, and stereos. Because their casings were often black or dark-colored, they were called "black goods." With technological advancements, the scope of "black goods" has expanded to include electronic devices such as computers, game consoles, and DVD players, whose primary function is to provide entertainment and information.

[0099] The following examples demonstrate how by stripping away the superficial differences between products and migrating common logical genes, a logical-level migration from white goods experience to black goods strategy is achieved, thus overcoming the pain point of the difficulty in reusing knowledge between heterogeneous products / production lines.

[0100] Figure 3 A flowchart of a cross-product quality inspection method based on the Industrial Internet provided for some embodiments of this application.

[0101] The method can be implemented by a quality inspection system or a cross-product quality inspection device, and includes the following steps:

[0102] Step S31: Construct a first quality inspection knowledge base to guide the quality inspection of the source product. The first quality inspection knowledge base includes the first node chain information and the first causal edge strategy corresponding to the quality inspection items of the source product.

[0103] The first quality inspection knowledge base includes causal graph files corresponding to the M production lines of the source product. Each causal graph file i includes at least the first node chain information i and the first causal edge strategy i on the source product side, and may also include first metadata i. The first node chain information i includes information such as the topology and node type of the first node chain. The specific implementation of step S31 has been described in detail in the previous related embodiments and will not be repeated here. Given that the first quality inspection knowledge base of the source product has been constructed, the following steps are used to achieve cross-product migration and reuse of quality inspection logic knowledge.

[0104] Step S32: Query the target product to be built in the database and obtain the second node chain information corresponding to the current quality inspection item of the target product.

[0105] In some embodiments, the quality inspection system / cross-product quality inspection device can obtain information such as the Q production lines involved in the target product, as well as the equipment type, technical parameters, status information, and operation and maintenance records of each production line, through the equipment ledger of the Manufacturing Execution System (MES). The MES equipment ledger is used to digitally record and manage the entire lifecycle information of production equipment. It serves as the fundamental database connecting the physical entities of equipment with the information system. The MES equipment ledger integrates the static attributes, dynamic status, and operation and maintenance records of equipment in a structured manner, providing real-time and accurate data support for production scheduling, quality control, and maintenance decisions.

[0106] In some embodiments, the quality inspection system / cross-product quality inspection device can also obtain environmental baseline data for each production line of the target product from the environmental monitoring system. This environmental baseline data includes historical averages and fluctuation ranges of environmental parameters. These environmental parameters include, but are not limited to, dust concentration, temperature, and humidity in the workshop.

[0107] In some embodiments, the quality inspection system / cross-product quality inspection device can also obtain the electronic control plan for each production line of the target product from the quality management system (QMS). The electronic control plan includes current control thresholds, response plans (for adapting to constraint nodes), etc.

[0108] In some embodiments, the quality inspection system / cross-product quality inspection device can perform feature extraction on data obtained from the MES equipment ledger, environmental monitoring system, and QMS system, extracting indicators related to usage intensity, environmental disturbance, condition deterioration, and defects to obtain node information of the k-th production line of the target product to be inspected. Based on the relationships (e.g., causal relationships, constraint relationships, etc.) and flow directions between these nodes, a topology of the second node chain is constructed to obtain the second node chain information k. Here, k represents the production line number of the target product, and 1≤k≤Q.

[0109] Figure 4 This is a schematic diagram of the node chain topology of a television-automatic optical inspection camera visual inspection project provided in some embodiments of this application.

[0110] In some embodiments, taking a television as the target product and an automated optical inspection (AOI) camera vision inspection item as an example, see [link to relevant documentation]. Figure 4The node chain topology of this quality inspection project includes input node 1 (cleaning cycle), input node 2 (cumulative number of photos taken by the lens), input node 3 (dust concentration), intermediate node (AOI camera_lens contamination probability), output node 1 (missed detection rate), output node 2 (screen dust ingress rate), and constraint node (cleaning cycle ≤ 500 times).

[0111] In some embodiments, such as Figure 4 As shown, the cleaning cycle is one of the metrics for usage intensity. Higher usage intensity leads to greater product degradation, resulting in a higher probability of lens contamination for the AOI camera. Therefore, input node 1 and the intermediate node form a causal edge. The cumulative number of times the lens takes photos is another metric for usage intensity. The more times the lens takes photos, the higher the probability of lens contamination for the AOI camera. Therefore, input node 2 and the intermediate node also form a causal edge. Dust is a production line environmental factor. For example, a high-concentration dust environment can easily exacerbate product degradation, leading to an increased probability of lens contamination for the AOI camera. Therefore, input node 3 and the intermediate node also form a causal relationship.

[0112] In some embodiments, such as Figure 4 As shown, the lens contamination probability of the AOI camera is one of the indicators for measuring the degradation state of the target product. The higher the lens contamination probability of the AOI camera, the more severe the degradation of the target product, which will lead to a higher missed detection rate and screen dust ingress rate. Therefore, the intermediate node and output node 1 form a causal edge. Missed products lead to screen dust ingress detection, so the missed detection rate will affect the screen dust ingress rate. Therefore, output node 1 and output node 2 also form a causal edge.

[0113] In some embodiments, such as Figure 4 As shown, the constraint nodes in the second node chain are used to constrain the cleaning cycle, controlling the cleaning cycle to not exceed a threshold (e.g., 500 times).

[0114] It should be noted that when quality inspection systems / cross-product quality inspection devices extract features from the production line of a target product, they should focus on the "causal role" rather than the "equipment name" and identify the quality inspection links that have the characteristics of "usage intensity input → condition deterioration intermediate → missed detection rate output". That is, strip away the superficial differences of products and transfer the common logical genes, thereby laying the foundation for the reuse of quality inspection knowledge across products / production lines.

[0115] Step S33: Match the information of the first node chain and the information of the second node chain to determine whether the migration conditions are met.

[0116] In some embodiments, the quality inspection system / cross-product quality inspection device can encode the first node chain information and the second node chain information into topological features to generate a topological fingerprint describing the node type sequence and the causal edge direction.

[0117] In some embodiments, the topological fingerprint obtained by encoding the information of the first node chain is called the "first topological fingerprint," and the topological fingerprint obtained by encoding the information of the second node chain is called the "second topological fingerprint." The quality inspection system / cross-product quality inspection device can determine whether the topological structure between node chains and the node types of each node are consistent by matching the first and second topological fingerprints, thereby determining whether the node chains across products / production lines have logical isomorphism.

[0118] In some embodiments, see Figure 2 Taking the air conditioning air tightness test as an example, the first node chain contains three causal logic chains: ① [calibration cycle] → [helium detector_sensor drift probability] → [missing detection rate] → [commercial inspection_refrigerant leakage rate], ② [cumulative number of tests] → [helium detector_sensor drift probability] → [missing detection rate] → [commercial inspection_refrigerant leakage rate], ③ [workshop temperature and humidity] → [helium detector_sensor drift probability] → [missing detection rate] → [commercial inspection_refrigerant leakage rate]. These three causal logic chains are converted into the first topological fingerprint as [input] → [intermediate] → [output] → [output].

[0119] In some embodiments, see Figure 4 Taking the TV AOI camera visual inspection project as an example, the second node chain contains three causal logic chains: ① [Cleaning cycle] → [AOI camera_lens contamination probability] → [Missed detection rate] → [Screen dust ingress rate], ② [Cumulative number of lens photos] → [AOI camera_lens contamination probability] → [Missed detection rate] → [Screen dust ingress rate], ③ [Dust concentration] → [AOI camera_lens contamination probability] → [Missed detection rate] → [Screen dust ingress rate]. These three causal logic chains are converted into the second topological fingerprint as [Input] → [Intermediate] → [Output] → [Output]. Therefore, it can be seen that for both the air conditioning airtightness inspection project and the TV AOI camera visual inspection project, the first and second topological fingerprints are completely matched (100% matching degree).

[0120] In some embodiments, assuming the second topological fingerprint corresponding to the TV aging test item is [input] → [output], this second topological fingerprint lacks intermediate nodes. Compared with the first topological fingerprint [input] → [intermediate] → [output] → [output] corresponding to the air conditioner airtightness test item, the matching degree of the two topological fingerprints is only 50%, resulting in a topological fingerprint mismatch.

[0121] In some embodiments, assuming the second topological fingerprint corresponding to the TV packaging line inspection item is [Input] → [End], this second topological fingerprint lacks intermediate nodes and output nodes. Compared with the first topological fingerprint [Input] → [Intermediate] → [Output] → [Output] corresponding to the air conditioning air tightness inspection item, the matching degree of the two topological fingerprints is 0%, that is, the topological fingerprints do not match at all.

[0122] In some embodiments, when the first and second topological fingerprints match perfectly (i.e., the matching degree is 100%), the quality inspection system / cross-product quality inspection device can determine that the migration conditions are met. This discrimination method can quickly screen quality inspection items of the target product that meet the migration conditions, and the discrimination efficiency is high. It can quickly filter out items that obviously do not meet the migration conditions. However, this method is a coarse screening with low discrimination accuracy. That is, two cross-product / production line quality inspection items with perfectly matching topological fingerprints may not be able to migrate.

[0123] In some embodiments, to improve the accuracy of migration condition discrimination, other dimension-based discrimination rules can be added. By fusing multi-dimensional discrimination rules, fine screening can be achieved, improving the accuracy and stability of migration, as shown in Table 2:

[0124] Table 2

[0125]

[0126] Table 2 shows that the transfer criteria include four matching dimensions: ① topological fingerprint, ② consistency of causal edge direction, ③ semantics of output nodes, and ④ existence of constraint nodes. When the topological fingerprint is perfectly matched, it is also necessary to consider whether the causal edge direction is "predecessor → successor". For example, in the causal edge "cleaning cycle → AOI camera_lens contamination probability", "cleaning cycle" is the predecessor, and "AOI camera_lens contamination probability" is the successor. If both quality inspection items contain constraint nodes, such as "calibration cycle" and "cleaning cycle", and the constraint nodes have clear control thresholds, such as calibration cycle ≤ 500h and cleaning cycle ≤ 500 times, then the constraint node existence principle is satisfied. If the final nodes are both final inspection defect rate indicators, it indicates that the final nodes are affected by the causal chain of [input] → [intermediate] → [output]. The relevant function in this logical strategy is also transferable prior knowledge.

[0127] As can be seen from the above-mentioned migration condition discrimination mechanism, the migration matching process of this application only relies on the node chain topology, causal logic and node role type, completely avoiding superficial features such as device name, node name and parameter value, and realizing logical-level knowledge transfer across products (e.g. white goods → black goods) and across production lines / processes (e.g. airtightness inspection → AOI camera vision inspection).

[0128] If the quality inspection system / cross-product quality inspection device determines that the migration conditions are not met, then proceed to step S34; if the quality inspection system / cross-product quality inspection device determines that the migration conditions are met, then proceed to step S35.

[0129] Step S34: Continue iterating through the next quality inspection item of the target product.

[0130] If the quality inspection item k corresponding to the kth production line of the target product does not meet the migration conditions, it means that the quality inspection item k cannot reuse the prior knowledge in the quality inspection knowledge base of the source product. Therefore, skip the currently traversed quality inspection item k, continue to traverse the quality inspection item k+1 corresponding to the k+1th production line, obtain the second node chain information corresponding to the quality inspection item k+1, and re-determine whether the migration conditions are met.

[0131] Step S35: Based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment and process control plan, adjust the key parameters of the causal relationship logic function in the first causal edge strategy to obtain the adjusted second causal edge strategy.

[0132] When the quality inspection item i of the source product and the quality inspection item k of the target product meet the transfer conditions, it indicates that transferring the logical-level prior knowledge of the quality inspection item i of the source product in the first quality inspection knowledge base to the quality inspection item k of the target product is feasible. However, due to the differences in equipment type, process, and production environment between the source product and the target product, the transfer and reuse of prior knowledge across products / production lines becomes a challenge in this field. If the transferred prior knowledge has low compatibility with the quality inspection item k of the target product, it will reduce the accuracy of the second quality inspection knowledge base and cause false detection problems in the target product production line.

[0133] This application quantifies the differences between the source and target products in terms of physical characteristics, production environment, and processes. Based on these quantified differences, it adjusts the key parameters of the relevant functions in the first causal edge strategy, thereby improving the adaptability of the transferred logical knowledge to the target product production line and avoiding problems such as low database accuracy and false detection. Based on equipment physical characteristics (e.g., equipment sensitivity scores), production line environmental data (e.g., environmental parameter baselines), and production line process requirements (e.g., threshold constraints), it adapts the migration coefficients to avoid black-box operations. Through multi-dimensional coefficient correction, it achieves logical gene transfer. For example, the degradation patterns of equipment associated with usage intensity and environmental disturbances remain unchanged, but differences in environment and processes are compensated for through migration coefficients.

[0134] In some embodiments, the first node chain includes: ① a first input node representing the intensity index used, ② a second input node representing the environmental disturbance index, and ③ an intermediate node representing the product degradation state. See also Figure 2 The first input node includes input node 1 (calibration cycle) and input node 2 (cumulative number of tests), the second input node includes input node 3 (workshop temperature and humidity), and the intermediate node is the helium detector_sensor drift probability.

[0135] In some embodiments, the causal relationship logic function in the first causal edge strategy may include: ① a first function representing the causal logic relationship between the first input node and the intermediate node in the first node chain; ② a second function representing the causal logic relationship between the second input node and the intermediate node in the first node chain; ③ a third function representing the causal logic relationship between the intermediate node and the output node in the first node chain; and ④ a fourth function representing the causal logic relationship between the output nodes. Referring to Table 1, the first function is, for example, P=1 / (1+e^(-k×(X-X0))), the second function is, for example, ΔP=α×(Env-Env_base), the third function is, for example, the false negative rate=P_state×D_base, and the fourth function is, for example, the final defect rate=false negative rate×(S / 10). Each function represents an objective process law of a causal edge. This objective process law can be a common law of most products and things, with the same function law, but the values ​​of key parameters (e.g., k and X0) have individual differences.

[0136] Figure 5 This is a schematic diagram illustrating the dynamic adaptation principle of cross-product / production line knowledge transfer provided in some embodiments of this application.

[0137] In some embodiments, such as Figure 5 As shown, when the first node chain and the second node chain are successfully matched, i.e. the migration conditions are met, cross-product / production line knowledge transfer can be performed. The quality inspection system or cross-product quality inspection device can adjust the key parameters in the logic function to adapt to the quality inspection requirements of cross-product / production line.

[0138] In some embodiments, such as Figure 5 As shown, the quality inspection system or cross-product quality inspection device can adjust the first degradation sensitivity parameter (k_source) in the first function to the second degradation sensitivity parameter (k_target) and adjust the first critical usage amount (X0_source) to the second critical usage amount (X0_target) based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment and process control plan.

[0139] In some embodiments, such as Figure 5As shown, the quality inspection system or cross-product quality inspection device can adjust the first environmental sensitivity parameter (α_source) in the second function to the second environmental sensitivity parameter (α_target) based on the quantitative differences between the source product and the target product in the production environment. In this way, while keeping the objective process law of the causal edge unchanged, the key parameter values ​​of the causal edge strategy on the k-th production line of the target product are automatically adapted and rewritten based on multi-dimensional quantitative differences, improving the adaptability of transferred knowledge to the target product production line, thereby overcoming the knowledge transfer barrier across products / production lines and balancing the database construction effect and the adaptability of the quality inspection knowledge base.

[0140] In some embodiments, such as Figure 5 As shown, the quality inspection system or cross-product quality inspection device can query the second detectivity benchmark value D_base2 adapted to the target product based on the FMEA database of the target product, and adjust the first detectivity benchmark value D_base1 of the source product in the third function to D_base2.

[0141] In some embodiments, such as Figure 5 As shown, the quality inspection system or cross-product quality inspection device can query the severity value S2 of the target product based on the FMEA database of the target product, and adjust the severity value S1 of the source product in the fourth function to S2. Through the above parameter correction method, the second causal edge strategy adapted to the target product production line after migration is obtained.

[0142] In some embodiments, the quality inspection system or cross-product quality inspection device can obtain a first environmental parameter benchmark value mapped to the quality inspection items of the source product and a second environmental parameter benchmark value mapped to the quality inspection items of the target product from the environmental monitoring system. The first and second environmental parameter benchmark values ​​are standard values ​​for the same environmental indicator on different production lines; for example, the benchmark value for dust concentration required in an air conditioning tightness testing environment is 0.3 mg / m³. 3 The baseline dust concentration required in a television AOI camera visual inspection environment is 0.5 mg / m³. 3 .

[0143] In some embodiments, the quality inspection system or cross-product quality inspection device calculates a first coefficient, Env_coeff (or environmental severity coefficient), based on a first environmental parameter baseline value and a second environmental parameter baseline value, to characterize the quantitative difference in the production environment between the source product and the target product. Env_coeff = Second environmental parameter baseline value / First environmental parameter baseline value. Taking the migration of an air conditioning airtightness inspection item to a television AOI camera visual inspection item as an example, assuming the first environmental parameter baseline value (Env_base1) corresponding to the dust concentration in the air conditioning production line is 0.3 mg / m³,... 3The baseline value (Env_base2) for the second environmental parameter corresponding to the dust concentration in the TV production line is 0.5 mg / m³. 3 Therefore, Env_coeff = 0.5 / 0.3 = 1.67.

[0144] In some embodiments, the quality inspection system or cross-product quality inspection device may calculate a second coefficient, Dev_coeff (or device sensitivity coefficient), to characterize the quantitative difference in physical properties based on the first device sensitivity score of the source product mapping and the second device sensitivity score of the target product mapping. Dev_coeff = second device sensitivity score / first device sensitivity score.

[0145] In some embodiments, the equipment sensitivity score is preset by a quality engineer based on the equipment's physical characteristics, equipment technical manual, and historical failure data. For example, as shown in Table 3, if the first equipment sensitivity score is 3.0 and the second equipment sensitivity score is 4.5, then Dev_coeff = 4.5 / 3.0 = 1.5.

[0146] Table 3

[0147]

[0148] In some embodiments, the quality inspection system or cross-product quality inspection device can calculate a third coefficient, Proc_coeff (or process window coefficient), to characterize the quantitative difference in the process control plan based on a first process control threshold used for the quality inspection items of the source product and a second process control threshold used for the quality inspection items of the target product. Proc_coeff = second process control threshold / first process control threshold. For example, if the first process control threshold corresponding to the calibration cycle in the airtightness testing item is 500 hours, and the second process control threshold corresponding to the cleaning cycle in the television AOI camera visual inspection item is 500 times, then Proc_coeff = 500 / 500 = 1.

[0149] In some embodiments, the quality inspection system or cross-product quality inspection device can calculate a second degradation sensitivity parameter (k_target) based on a first coefficient (Env_coeff), a second coefficient (Dev_coeff), a third coefficient (Proc_coeff), and a first degradation sensitivity parameter (k_source), where k_target = k_source × Env_coeff × Dev_coeff × Proc_coeff. For example, in the first causal edge strategy, k_source = 0.02, and according to the aforementioned migration calculation example, k_target = 0.02 × 1.67 × 1.5 × 1 = 0.050.

[0150] k_target (also known as the comprehensive migration coefficient) is a core parameter for measuring the causal relationship migration adaptability between the target production line and the source production line. It represents the adjustment factor of the causal function model of the source production line to the target production line under specific environmental, equipment, and process conditions. Essentially, it accurately transfers the experiential knowledge of the source production line to the target production line through multi-dimensional physical parameter correction, avoiding engineering distortions caused by "numerical transfer" of prior knowledge. Before migration, k_source represents the degree of influence of usage intensity on the sensor drift probability of the air conditioner helium detector; after migration, k_target represents the degree of influence of usage intensity on the lens contamination probability of the TV AOI camera. That is, the k value before and after migration both represent the sensitivity of the production line equipment to the degradation of usage intensity; the physical meaning of the k value and the causal logic of equipment usage intensity → state degradation remain unchanged. By stripping away the superficial differences between products and transferring common logical genes, a logical-level migration from white goods experience to black goods strategy is achieved, solving the pain point of difficult knowledge reuse between heterogeneous products / production lines.

[0151] In some embodiments, for the first function P=1 / (1+e^(-k×(X-X0))), after the value of k is adjusted from k_source to k_target, the value of X0 also needs to be adjusted synchronously to adapt to the risk assessment of equipment degradation caused by the usage intensity under the target product / production line environment. The quality inspection system or cross-product quality inspection device can refit the first function based on k_target and the measured data of the target production line, such as the equipment usage intensity index and historical cases related to "whether the AOI camera lens is contaminated", using the Logistic regression algorithm to obtain the second critical usage amount X0_target, and rewrite the X0 value in the first function from the first critical usage amount X0_source to X0_target, for example, changing X0 from 420 to 380.

[0152] This application's embodiments reconstruct the causal logic of "equipment usage intensity → condition deterioration" based on quantitative differences between products, causal inference, and fitting of actual data, rather than simply "numerical transfer" or "hard numerical replacement." This improves the adaptability and stability of cross-product / production line knowledge transfer, enhances the accuracy of database construction, and enables the rapid construction of a quality inspection knowledge base for the target product. It also improves database construction efficiency, quality inspection efficiency, and production efficiency, overcoming the challenges of cross-product, cross-production line, cross-process, and cross-domain knowledge transfer. Furthermore, the overall control process is based on the automated execution of the Industrial Internet, reducing the computing power load and overhead of the Industrial Internet.

[0153] In some embodiments, the quality inspection system or cross-product quality inspection device can calculate the second environmental sensitivity parameter α_target based on the first coefficient Env_coeff and the first environmental sensitivity parameter α_source, where α_target = α_source × Env_coeff, and adjust the α value in the second function ΔP = α × (Env - Env_base) from α_source to α_target. For example, α_source = 0.005, α_target = 1.67 × 0.005 = 0.008. Before migration, α_source characterizes the influence of workshop temperature and humidity on the sensor drift probability of the air conditioning helium detector, while after migration, α_target characterizes the influence of dust concentration on the lens contamination probability of the TV AOI camera. That is, the α value before and after migration both characterize the degradation sensitivity of production line equipment to environmental disturbances, meaning that the physical meaning of the α value and the causal logic of environmental disturbance → state degradation remain unchanged. By stripping away the superficial differences between products and migrating common logical genes, a logical-level migration from white goods experience to black goods strategy is achieved, solving the pain point of difficult knowledge reuse between heterogeneous products / production lines.

[0154] In some embodiments, the quality inspection system or cross-product quality inspection device can obtain the second detectivity benchmark value (D_base2) mapped to the quality inspection items of the target product based on the FMEA database of the target product. This allows the D_base value in the third function (missed detection rate = P_state × D_base) to be adjusted from the first detectivity benchmark value (D_base1) mapped to the quality inspection items of the source product to D_base2. In this way, the causal logic of "state degradation → missed detection rate" remains unchanged before and after the migration. By stripping away the superficial differences between products and migrating common logical genes, a logical-level migration from white goods experience to black goods strategy is achieved, overcoming the pain point of difficulty in reusing knowledge between heterogeneous products / production lines.

[0155] In some embodiments, the quality inspection system or cross-product quality inspection device can also query the second severity value S2 mapped to the quality inspection items of the target product based on the FMEA database of the target product, and adjust the first severity value S1 corresponding to the quality inspection items of the source product in the fourth function to S2. In this way, the causal logic of "missed inspection rate → final inspection defect rate" remains unchanged before and after the migration. By stripping away the superficial differences between products and migrating common logical genes, a logical-level migration from white goods experience to black goods strategy is achieved, solving the pain point of difficult knowledge reuse between heterogeneous products / production lines.

[0156] In some embodiments, the quality inspection system or cross-product quality inspection device can also obtain a second process control threshold used for the quality inspection items of the target product from the electronic control plan file of the target product, and set the adjustment range of the control parameters of the constraint nodes in the second node chain based on the second process control threshold. For example, the constraint node in the second node chain is used to constrain the upper limit of the "cleaning cycle". If the control threshold of the cleaning cycle is obtained as 500 times from the reaction plan of the electronic control plan file of the television, then the constraint node sets the cleaning cycle ≤ 500 times. In this way, the capability and processing logic of the constraint node remain unchanged, and only the control threshold of the constraint node is updated.

[0157] Taking the knowledge transfer from the air conditioning air tightness inspection project to the TV AOI camera visual inspection project as an example, the updated representation of key content in the causal graph file before and after the transfer is shown in Table 4. In the process of reconstructing the causal graph file after the transfer, the transfer source and confidence level can be marked. The new causal graph retains the topological structure of the source causal graph and only corrects the key function parameters with physical significance to ensure the engineering credibility of the transfer result.

[0158] Table 4

[0159]

[0160] In some embodiments, the quality inspection system or cross-product quality inspection device can automatically verify the rationality of the reconstructed second causal edge strategy after migration based on triple verification rules. The verification rules include: ① the k_target after migration should be within a reasonable range (e.g., between 0.01 and 0.1); ② the control threshold of the constraint nodes should be greater than 0 and conform to process standards / common sense; ③ the number of nodes and causal edges should remain consistent before and after migration. For example, in the aforementioned example, a k_target value of 0.050 is reasonable, a cleaning cycle of ≤500 times is reasonable, and both before and after migration have 7 nodes and 6 causal edges, with the same and complete topology, thus the verification passes. The quality inspection system or cross-product quality inspection device can utilize the verified second causal edge strategy to construct a second quality inspection knowledge base, thereby balancing the feasibility, accuracy, and efficiency of migration-based knowledge base construction, and improving quality inspection efficiency and production efficiency.

[0161] In some embodiments, manual review by quality engineers is also possible. For example, a quality inspection system or cross-product quality inspection device can indicate in a visual interface that the node chain topology of the first quality inspection item of the source product and the second quality inspection item of the target product are completely matched, such as "The node chain topology of the air conditioning air tightness test item and the TV AOI camera visual inspection item are 100% matched, migration is recommended." In this way, the quality engineer can choose to approve or reject to manually trigger whether to execute the migration process.

[0162] In some embodiments, the quality inspection system or cross-product quality inspection device can provide risk warnings through a visual interface, such as "Equipment sensitivity score depends on preset value, and it is recommended to strengthen monitoring in the first week", thereby providing risk warnings.

[0163] In some embodiments, the quality inspection system or cross-product quality inspection device can display an activation prompt message on a visual interface to indicate that the generated second causal graph file is about to take effect, such as "The CG_TV2026_AOI_V1 file will be automatically generated after the review is passed, and the file will be automatically pushed to the inference engine after it takes effect." Thus, after the quality engineer confirms the activation, the quality inspection system or cross-product quality inspection device can perform quality inspection based on the activated second causal graph file, and simultaneously push the second causal graph file to the inference engine. The inference engine then infers and optimizes the key parameters involved in the second causal graph file, thereby enabling the second causal graph file to iterate and evolve, for example, generating the CG_TV2026_AOI_V2 file.

[0164] By using automated review by quality inspection systems or cross-product quality inspection devices, along with manual review assisted by quality engineers, the basis for migration can be strengthened, avoiding low-confidence and high-risk migration data, thereby preventing blind migration and improving the accuracy of database construction. This balances the efficiency and accuracy of building quality inspection knowledge bases for new products / production lines, thereby improving quality inspection efficiency and production efficiency. It provides positive guidance for enterprise manufacturing, enabling the migration and reuse of prior knowledge across products, production lines, processes, and domains in the industrial internet-related fields. By stripping away the superficial differences between products and migrating common logical genes, it solves the pain point of the difficulty in reusing knowledge between heterogeneous products / production lines.

[0165] Step S36: Construct a second quality inspection knowledge base based on the second node chain information and the second causal edge strategy.

[0166] In some embodiments, see Figure 5 The quality inspection system or cross-product quality inspection device can construct a second causal graph file based on the second node chain information, the second causal edge strategy reconstructed during migration, and the second metadata. After traversing Q production lines of the target product, the quality inspection system or cross-product quality inspection device aggregates the second causal edge strategies corresponding to the target production lines that meet the migration conditions and constructs a second quality inspection knowledge base. This enables the rapid construction of a new product quality inspection knowledge system, balancing the efficiency and accuracy of database construction, and solving the pain point of difficulty in reusing knowledge between heterogeneous products / production lines.

[0167] Step S37: Based on the second quality inspection knowledge base, perform quality inspection on the quality inspection items of the target product.

[0168] In some embodiments, the migrated second quality inspection knowledge base is a prototype that can be used in early quality inspection testing. Subsequently, it can be further evolved based on the data source and actual test data of the target product, combined with manual correction and computer virtual world deduction and optimization, so as to gradually improve the accuracy and product adaptability of the second quality inspection knowledge base.

[0169] In some embodiments, after the quality inspection system or cross-product quality inspection device pushes the reconstructed second causal graph file to the inference engine, the inference engine can pre-infer "parameter change → quality result" in the virtual world based on the counterfactual inference mechanism, generate a quantifiable and executable optimization strategy, and thus guide the evolution of the strategy by virtual inference.

[0170] In some embodiments, the quality inspection system or cross-product quality inspection device can automatically control the simulation engine to perform simulations when a new target production line or process change is detected, such as simulations of recommended values ​​for the cleaning cycle of the AOI camera in a new TV production line.

[0171] In some embodiments, the inference engine can automatically infer key parameters (such as cleaning cycles) involved in the second causal graph when it receives the second causal graph file, and output parameter optimization strategies.

[0172] Figure 6 A flowchart illustrating the deduction and optimization method of the quality inspection knowledge base provided in some embodiments of this application.

[0173] The execution entity of this method is a quality inspection system or a cross-product quality inspection device, specifically the inference engine within the quality inspection system or cross-product quality inspection device. This method includes:

[0174] Step S61: Sample the historical distribution data of the disturbance variables that affect the final inspection defect rate of the target product and generate a preset number of test samples.

[0175] In some embodiments, before executing the simulation program, the simulation engine can pre-set simulation configuration information, which includes, but is not limited to, the current execution parameter value of the parameter to be simulated, the first target value, the perturbation variable, the perturbation range, the number of simulations, and the output indicators. Taking the cleaning cycle simulation of a TV AOI camera visual inspection project as an example, where the parameter to be simulated is the cleaning cycle, its simulation configuration information is shown in Table 5:

[0176] Table 5

[0177]

[0178] The simulation configuration information defines a comparison framework between the baseline strategy and the target strategy, and specifies the environmental variables to be perturbed and their statistical distribution, ensuring that the simulation results have engineering comparability and a basis for uncertainty quantification. This application's embodiment implements a Monte Carlo simulation mechanism, quantifying uncertainty perturbations and driving the simulation with a causal graph, enabling the overall quality inspection knowledge base system to have sustainable learning and evolution capabilities.

[0179] In some embodiments, the inference engine generates a preset number of test samples (also known as perturbation samples) by sampling the historical distribution data of perturbation variables that affect the final inspection defect rate of the target product. Specifically, based on the historical normal distribution of the perturbation variables, a preset number of random perturbation variable values ​​are generated, and each test sample contains the first target value of the parameter to be inferred and the perturbation variable value.

[0180] In some embodiments, taking the cleaning cycle simulation of a TV AOI camera visual inspection project as an example, the perturbation variable is dust concentration. Assuming that the historical normal distribution of dust concentration is N(0.5, 0.15²), based on the number of Monte Carlo simulation samples set in the simulation configuration information, 10,000 sets of random dust concentration values ​​are generated. Each set of test samples includes cleaning cycles (700 times) and dust concentration (random value).

[0181] Step S62: Using the second causal graph file in the second quality inspection knowledge base, a preset number of test samples are deduced to obtain the first final inspection defect rate mapped to each test sample.

[0182] In some embodiments, the inference engine performs layer-by-layer inference and calculation for each group of test samples based on the node chain and causal edge strategy in the second causal graph file to obtain the first final inspection defect rate mapped to each group of test samples.

[0183] In some embodiments, taking the cleaning cycle simulation of a TV AOI camera visual inspection project as an example, the simulation engine performs logical simulation and calculation layer by layer along CG_TV2026_AOI_V1: Cleaning cycle (700 times) + dust concentration (random value) → AOI camera_lens contamination probability = f(cleaning cycle, dust concentration) / / call causal function library → missed detection rate = AOI camera_lens contamination probability × D_base2 → screen dust ingress rate = missed detection rate × S_weight. Wherein, S_weight is the normalized value of S2, i.e., S_weight = S2 / 10. In this simulation example, the first final inspection defect rate is the "screen dust ingress rate". The simulation calculation results are shown in Table 6:

[0184] Table 6

[0185]

[0186] When the simulation engine executes the simulation program, it strictly calculates forward along the topological structure of the causal graph. Each step calls functions from the pre-defined causal function library to ensure that the simulation logic is interpretable and auditable, avoiding black-box operations. The sampling of perturbation variables is based on historical distribution data to quantify the impact of environmental uncertainties on the quality simulation results, ensuring the reliability and stability of the simulation results.

[0187] Step S63: Obtain the quality projection result based on the first final inspection defect rate mapped by a preset number of test samples.

[0188] In some embodiments, the first final inspection defect rate mapped to a preset number of test samples is statistically aggregated to obtain quality inference results. The quality inference results include, but are not limited to, the mean, mean increase, confidence interval, and risk-sensitive area distribution of the preset number of first final inspection defect rates.

[0189] In some embodiments, taking the cleaning cycle simulation of the aforementioned TV AOI camera visual inspection project as an example, the average screen gray-incidence rate under the baseline strategy (500 cleaning cycles) is 0.6%, and the average gray-incidence rate of 10,000 screens under the target strategy (700 cleaning cycles) is 1.82%. Therefore, the average increase = (1.82% - 0.60%) / 0.60% ≈ 203%, thus quantifying the cost of parameter changes so that the engine system can decide on the output of the parameter optimization strategy. When the average increase exceeds a first threshold (e.g., 15%), it indicates that the simulation shows a significant increase in the final inspection defect rate after parameter tuning, posing an incremental risk.

[0190] In some embodiments, the confidence level (1-α) is: with the same sample size, a higher confidence level results in a wider confidence interval. To ensure that the defect rate has a higher probability of falling within the interval, the confidence interval needs to be widened. Although this sacrifices accuracy, it reduces the risk of Type I errors, ensuring that the true risk value is still included within the interval and avoiding safety incidents caused by sample fluctuations. A wider confidence interval makes the system more "conservative," effectively preventing aggressive risky operations when data is insufficient, thereby protecting production line quality. Under the baseline strategy (500 cleaning cycles), the 95% confidence interval is [0.52%, 0.68%], and under the target strategy (700 cleaning cycles), the 95% confidence interval is [1.51%, 2.15%].

[0191] In some embodiments, the inference engine can identify "risk-sensitive areas" and analyze the derivative of the final inspection defect rate with respect to the disturbance variable through a sliding window. This allows it to locate the critical interval (i.e., the risk-increase interval) where a small change in the disturbance variable leads to a sharp increase in the final inspection defect rate. This provides a precise basis for optimizing the output parameters of the inference engine, solving the decision blind spot of "when to strengthen monitoring", thereby predicting and avoiding risks.

[0192] In some embodiments, the simulation engine can identify a high-risk sensitive area based on the mean and slope of the final inspection defect rate mapped from a preset number of test samples. When the mean is greater than a second threshold and the slope is greater than a third threshold, the engine obtains the lower limit value of the perturbation variable in the high-risk sensitive area as the sensitive threshold. When the actual perturbation variable value exceeds the sensitive threshold, it indicates a clear high-risk scenario. Taking the cleaning cycle simulation of the aforementioned TV AOI camera visual inspection project as an example, Table 7 illustrates a risk-sensitive area identification table:

[0193] Table 7

[0194]

[0195] Table 7 shows that when the dust concentration is in the range of [0.1, 0.4] mg / m³, based on the deduction from the migrated second causal map file, it is identified as a low-risk sensitive area, requiring only conventional monitoring measures. When the dust concentration is in the range of [0.4, 0.65] mg / m³, it is identified as a medium-risk sensitive area, requiring enhanced monitoring, such as increasing the frequency of dust monitoring. When the dust concentration is in the range of [0.65, 1.2] mg / m³, the average screen dust ingress rate is as high as 2.8%, with a change slope as high as 3.5% / 0.1mg, indicating a high-risk sensitive area. [0.65, 1.2] mg / m³ is the high-risk sensitive area for dust concentration, and 0.65 mg / m³ is the sensitive threshold for dust concentration. If the actual dust concentration in the workshop exceeds 0.65 mg / m³, it is determined to be a high-risk scenario, and monitoring measures can be further upgraded, such as triggering "daily standard version verification".

[0196] Step S64: Determine the target optimization strategy that matches the quality simulation results, and execute the target optimization strategy to change the values ​​of the parameters to be simulated.

[0197] In some embodiments, if the average increase of a preset number of first final inspection defect rates is greater than a first threshold, the simulation engine executes a first optimization strategy corresponding to the incremental risk, thereby automatically converting the quality simulation results into a quantifiable and executable static optimization strategy, achieving a leap from "analysis" to "decision-making". The first optimization strategy (also known as the fixed threshold optimization strategy) involves the simulation engine selecting a second target value between the migration value of the parameter to be simulated (e.g., 500 cleaning cycles) and the first target value (700 cleaning cycles), and changing the value of the parameter to be simulated to this second target value.

[0198] In some embodiments, when the simulation engine executes the first optimization strategy, it can obtain the risk inflection point value between the migration value of the parameter to be simulated and the first target value by finding the "risk inflection point" through Monte Carlo simulation and risk analysis. This risk inflection point value is then used as the second target value, and the value of the parameter to be simulated is changed to this second target value. By analyzing risk trends and inflection points, the mean increase is reduced or compressed to near the first threshold to reduce incremental risk and, while ensuring quality, maximize efficiency and extend the maintenance cycle. The fixed threshold optimization strategy is equivalent to establishing an automatic balancing mechanism between efficiency and quality risk, ensuring that any process changes are within a controllable risk range and avoiding blind and hasty decisions. The fixed threshold optimization strategy recommends a compromise, risk-controllable, and fixed new parameter value; the strategy is simple, easy to implement, and easy to monitor.

[0199] In some embodiments, when a high-risk sensitive area exists, the simulation engine can execute a second optimization strategy (also known as a dynamic threshold optimization strategy) corresponding to the high-risk sensitive area risk. This automatically transforms the quality simulation results into a quantifiable and executable dynamic optimization strategy, achieving a leap from "analysis" to "decision-making." When executing the second optimization strategy, the simulation engine calculates a third target value based on the migration values ​​of the parameters to be simulated, the current values ​​of the perturbation variables, and the sensitive critical values ​​in the second quality inspection knowledge base, and changes the values ​​of the parameters to be simulated to this third target value. The dynamic threshold optimization strategy can dynamically and adaptively assign values ​​to the parameters to be simulated following perturbation factors (such as dust concentration). Its strategy is more refined and has strong anti-interference capabilities.

[0200] In some embodiments, the third target value = migration value + R × (1 - perturbation variable value / sensitivity threshold). Here, the migration value is the value of the parameter to be deduced under the baseline strategy (e.g., 500 cleaning cycles), serving as the "anchor" of the formula and representing the baseline with the lowest risk. R is the maximum allowable adjustment calculated through counterfactual deduction and optimization objectives, reflecting the maximum extent to which the parameter to be deduced can be safely extended. When the perturbation variable value reaches the sensitivity threshold, the parameter to be deduced should revert to the baseline value (e.g., 500 cleaning cycles) without increasing. When the perturbation variable value is below the sensitivity threshold, the parameter to be deduced can be appropriately increased, and the lower the perturbation variable value, the greater the increase in the parameter to be deduced. This formula uses the perturbation variable value as an input variable and, through a simple linear controller, outputs an optimized parameter to be deduced, balancing risk control and quality knowledge optimization.

[0201] Taking the cleaning cycle simulation of the aforementioned TV AOI camera visual inspection project as an example, Table 8 illustrates a strategy generation rule base for a simulation engine. The simulation engine can match target optimization strategies based on the quantitative characteristics of the quality simulation results (such as incremental risk, risk-sensitive area, and confidence interval), and automatically generate three types of strategies: fixed threshold, dynamic threshold, and enhanced monitoring.

[0202] Table 8

[0203]

[0204] In some embodiments, the inference engine can generate a strategy package corresponding to the target optimization strategy. This strategy package includes: ① a unique strategy identifier / ID (e.g., STRAT_TV2026_AOI_20260805_001), ② inference basis (e.g., based on CG_TV2026_AOI_V1, 10,000 Monte Carlo simulations), ③ a recommended strategy (e.g., a dynamic threshold optimization strategy), ④ a strategy formula (e.g., cleaning cycle = 500 + 150 × (1 - dust concentration / 0.65)), and ⑤ applicable conditions (e.g., dust concentration ∈ [0.1, ...). The strategy package includes the following parameters: 0.65 mg / m³, ⑥ Expected results (e.g., average screen dust ingress rate reduced to 0.95%, monthly defect reduction of 65 units), ⑦ Validation recommendations (monitor screen dust ingress rate daily during the first week and compare and extrapolate the confidence interval), and ⑧ Corresponding measures for high-risk scenarios (e.g., trigger daily standard version validation when dust > 0.65 mg / m³). The strategy package clearly indicates the strategy formula, applicable conditions, and expected results to ensure the strategy is executable and verifiable, preventing distortion during strategy implementation.

[0205] In some embodiments, the simulation engine can generate a simulation report and push it to the terminal device used by quality engineers, enabling them to make decision support based on the report, including risk comparison and strategy recommendations. The simulation engine can also send simulation cases to a second quality inspection knowledge base, where they are accumulated into "simulation-actual effect" cases. These cases are used for subsequent correction of causal graph files and to provide simulation reference cases for new production lines, achieving a closed-loop mechanism for knowledge accumulation and group reuse.

[0206] In some embodiments, when constructing a causal graph for a new production line, the quality inspection system or cross-product quality inspection device can automatically retrieve deduction cases from the quality inspection knowledge base that have a similar topological structure to the node chain of the new production line, and recommend the deduction cases to the new production line, thereby facilitating the optimization and iteration of the causal graph of the new production line and accelerating the deduction and evolution of the quality knowledge system of the new production line.

[0207] In some embodiments, after the second quality inspection knowledge base is deduced and iteratively optimized, the iterative second quality inspection knowledge base becomes effective and is used to assist in quality inspection and management.

[0208] In some embodiments, during production and quality inspection on the production line, the quality inspection system or cross-product quality inspection device can calculate the target final inspection defect rate based on the total output for the current period and the number of defective products actually found in the final inspection stage. For example, if a TV production line produces 10,000 TVs on a given day and finds dust in the screens of 12 TVs during the final inspection stage, then the target final inspection defect rate is 0.12%.

[0209] In some embodiments, the quality inspection system or cross-product quality inspection device can compare the target final inspection defect rate with the average final inspection defect rate derived by the inference engine. If the target final inspection defect rate deviates significantly from the average final inspection defect rate, or the target final inspection defect rate does not fall within the 95% confidence interval derived by the inference engine, or the ratio of the first change slope to the second change slope is not within the threshold range, then the second causal edge strategy is corrected based on the inference-actual deviation quantification mechanism. This allows the quantified deviation between virtual world inference and real-world monitoring to drive the evolution of the quality knowledge base. Here, the first change slope is the slope of the target final inspection defect rate as a function of the disturbance variable, and the second change slope is the slope of the derived final inspection defect rate.

[0210] Figure 7 The schematic diagram illustrates the principle of causal graph correction based on inference-actual deviation quantization, which is provided for some embodiments of this application.

[0211] In some embodiments, such as Figure 7 As shown, the quality inspection system or cross-product quality inspection device obtains the target final inspection defect rate after performing quality inspection on the quality inspection items of the target product.

[0212] In some embodiments, such as Figure 7 As shown, the quality inspection system or cross-product quality inspection device calculates the deviation ratio between the target final inspection defect rate and the mean of the second final inspection defect rate. The second final inspection defect rate is obtained by extrapolating from a preset number of second test samples using a second causal edge strategy in a second quality inspection knowledge base. The second test samples include sampled values ​​of disturbance variables and the changed values ​​of the extrapolated parameters. The deviation ratio is calculated as: (Target final inspection defect rate - Mean of second final inspection defect rate) / Mean of second final inspection defect rate. For example, if the actual screen grayscale rate is 1.12% and the extrapolated mean screen grayscale rate is 0.95%, then the deviation ratio is 17.9%.

[0213] In some embodiments, such as Figure 7 As shown, the quality inspection system or cross-product quality inspection device calculates the first slope of the target final inspection defect rate as a function of the disturbance variable, and obtains the confidence interval and second slope of the final inspection defect rate derived from prior simulation cases. The second slope is the slope of the derived final inspection defect rate.

[0214] In some embodiments, such as Figure 7 As shown, the causal graph correction process is determined based on the following three conditions:

[0215] Deviation discrimination condition ①: If the deviation ratio is greater than the fourth threshold, it indicates that the results of virtual world simulation and real world monitoring are significantly different.

[0216] Deviation judgment condition ②: The target final inspection defect rate is not within the confidence interval of the derived final inspection defect rate, that is, the confidence interval does not cover the target final inspection defect rate, indicating that the actual final inspection result exceeds the expected fluctuation range of the deduced result, and a forced correction needs to be triggered.

[0217] Deviation criterion ③: If the ratio of the first change slope to the second change slope is not within the threshold range, it indicates that the direction of change predicted by the inference engine and its sensitivity to disturbances are incorrect, that is, there is a problem with the causal logic, and correction needs to be triggered and coefficient fitting needs to be performed again.

[0218] The causal graph correction process is triggered when at least one of the deviation criteria ①, ②, and ③ is met. If none of these criteria are met, the causal graph correction process is not triggered.

[0219] In some embodiments, such as Figure 7 As shown, when the graph correction process is triggered, the quality inspection system or cross-product quality inspection device can analyze the target root cause of the deviation and locate the objective function matching the target root cause in the second causal edge strategy. Assuming the deviation ratio is 17.9% and the fourth threshold is 15%, if the deviation ratio is greater than the fourth threshold, it is necessary to analyze the target root cause causing the deviation.

[0220] In some embodiments, taking a television AOI visual inspection project as an example, see [link to relevant documentation]. Figure 7 The node chain topology of this quality inspection project includes node S1 (cleaning cycle), node S2 (cumulative number of lens photos), node S3 (dust concentration), node S4 (AOI camera_lens contamination probability), node S5 (missed detection rate), node S6 (screen dust ingress rate), and node S... y (Constraint node, cleaning cycle ≤ 500 times). If the dust concentration monitoring value is consistent with the extrapolation assumption, then the error of environmental disturbance is excluded; if the actual AOI camera_lens contamination probability is greater than the extrapolated AOI camera_lens contamination probability, the root cause is determined to be the abnormality of the previously extrapolated cleaning cycle index, which leads to the deviation of the AOI camera_lens contamination probability. Therefore, the causal edge corresponding to the root cause is located as "cleaning cycle → AOI camera_lens contamination probability" (i.e., ...). Figure 7The objective function corresponding to the causal edge marked by the medium-thick line segment S1→S4 is P=1 / (1+e^(-0.050×(X-380))).

[0221] In some embodiments, such as Figure 7 As shown, a quality inspection system or cross-product quality inspection device can refit the objective function using a logistic regression algorithm to update the key parameters in the objective function. For example, new monitoring data (e.g., 168 sets of data from the first week's 7 days × 24 hours) can be integrated with 30 historical Ka cases. The objective function can then be refitted using a logistic regression algorithm, resulting in a refitted objective function P = 1 / (1 + exp(-0.058 × (X - 360))). In this case, k = 0.058 (originally 0.050), and X0 = 360 (originally 380). The fitting index R is recorded. 2 =0.91 (0.87 before refit), number of cases 198 (30 before refit).

[0222] Figure 7 The core idea behind the example's judgment rules is to not only consider the quantification of deviations but also to pinpoint the root causes of these deviations and analyze their sources. Through multi-dimensional judgment, it avoids overreacting to single data fluctuations while accurately locating the root causes of inference distortions, thus achieving efficient and accurate correction of the causal graph.

[0223] This approach avoids a "one-size-fits-all" approach, as not all deviations require correction. The system can distinguish between random fluctuations and systemic problems; it can preliminarily diagnose the root cause of deviations, such as uncovered confidence intervals or inconsistent trends, and guide subsequent corrections by locating the objective function. It saves computational resources and reduces computing power overhead. When any one of the deviation discrimination conditions ①, ②, and ③ is met, it triggers coefficient refitting of the objective function, thereby timely compensation for deviations and updating the causal graph. It ensures system stability, preventing frequent changes to validated causal graph knowledge due to occasional data anomalies, thus guaranteeing reliability in production environments. This approach transforms the system's "deviation-driven correction" mechanism from a simple reactive strategy into an intelligent self-learning process with judgment, root cause analysis capabilities, and self-evolution.

[0224] The "deviation-driven correction" mechanism employs a triple-judgment system of deviation threshold, confidence interval coverage, and consistency of change trends to avoid frequent corrections caused by noise interference. Coefficient refitting forces the fusion of new monitoring data with historical Ka cases, ensuring the causal graph continuously approximates physical reality while retaining a complete version change log to meet IATF 16949 change control requirements. After deviation correction, a new version of the causal graph file is generated, such as CG_TV2026_AOI_V2.

[0225] In some embodiments, the cross-product quality inspection device of the Industrial Internet can communicate and interact with the control terminals of each production line through the Industrial Internet. Each production line's control terminal, acting as a quality inspection execution unit, can monitor and manage its own product production line. Assuming a quality inspection knowledge base for air conditioners has already been built, the cross-product quality inspection device of the Industrial Internet can transfer matching causal logic knowledge from the air conditioner knowledge base to the television production line, quickly building a television quality inspection knowledge base and achieving a logical-level migration from white goods experience to black goods strategies. As the television quality inspection knowledge base is built and continuously evolves, matching causal logic knowledge can also be transferred to the refrigerator production line, achieving a logical-level migration from black goods experience to white goods strategies, thus overcoming the pain point of difficulty in reusing knowledge between heterogeneous products / production lines / processes.

[0226] By performing topological matching of the node chains of quality inspection items for the source and target products, and upon determining that the migration conditions are met, the parameters of the relevant causal relationship logic functions in the source product's quality inspection knowledge base are fine-tuned based on the quantitative differences in quality inspection standards / environments between the source and target products. This results in a quality inspection knowledge base adapted to the target product, enabling cross-product migration and reuse of the quality inspection knowledge base. This allows for the rapid construction of the target product's quality inspection knowledge base, improving efficiency in database construction, quality inspection, and production, while reducing the consumption of industrial internet computing power for building the quality inspection knowledge base. Furthermore, by extrapolating and optimizing the migrated and reconstructed quality inspection knowledge base, it is possible for the knowledge base to continuously evolve, iterate, and improve, thereby enhancing its accuracy.

[0227] Figure 8 This is a structural block diagram of a cross-product quality inspection device based on the Industrial Internet, provided for some embodiments of this application.

[0228] In some embodiments, such as Figure 8 As shown, the cross-product quality inspection device based on the Industrial Internet includes a communicator, a controller, and a quality inspection execution unit. The communicator is configured to communicate with the quality inspection execution unit via the Industrial Internet. The quality inspection execution unit is configured to perform quality inspection on the target product. The controller is configured as follows:

[0229] Build a primary quality inspection knowledge base to guide the quality testing of source products;

[0230] Query the target product in the database to be built, and obtain the second node chain information corresponding to the quality inspection items of the target product;

[0231] If the information of the first node chain and the information of the second node chain match, then based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment and process control plan, the key parameters of the causal relationship logic function in the first causal edge strategy are adjusted to obtain the adjusted second causal edge strategy.

[0232] Construct a second quality inspection knowledge base based on the second node chain information and the second causal edge strategy;

[0233] The second quality inspection knowledge base is synchronized to the quality inspection execution unit via a communicator, so that the quality inspection execution unit can perform quality inspection on the quality inspection items of the target product based on the second quality inspection knowledge base.

[0234] In some embodiments, such as Figure 8 As shown, the cross-product quality inspection device based on the Industrial Internet also includes a deduction engine. This engine is used to deduce key parameters in the quality inspection knowledge base to optimize them. The device also includes a memory configured to store quality inspection knowledge bases for multiple different products, serving as a cloud-based knowledge reserve. This provides data support for the subsequent deduction, optimization, and iteration of the deduction engine, and supports operations such as traceability and editing of the quality inspection knowledge base.

[0235] This application also provides a computer storage medium that can store a program. When the computer storage medium is configured in a cross-product quality inspection device based on the Industrial Internet, the program, when executed, can include the program steps involved in the cross-product quality inspection methods based on the Industrial Internet in the above embodiments. The computer storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0236] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0237] For ease of explanation, the above description has been provided in conjunction with specific embodiments. However, the foregoing exemplary discussion is not intended to be exhaustive or to limit the embodiments to the specific forms disclosed above. Various modifications and variations can be made based on the foregoing teachings. The selection and description of the above embodiments are for the purpose of better explaining the contents of this disclosure, thereby enabling those skilled in the art to better utilize the described embodiments.

Claims

1. A cross-product quality inspection method based on the Industrial Internet, characterized in that, include: A first quality inspection knowledge base is constructed to guide the quality inspection of source products. This knowledge base includes first node chain information and a first causal edge strategy corresponding to the quality inspection items of the source products. The first node chain information includes the topology and node types of the first node chain, and the first causal edge strategy includes causal relationship logic functions between nodes in the first node chain. Each node includes: a first input node representing a usage intensity index, a second input node representing an environmental disturbance index, and an intermediate node representing the product degradation state. The causal relationship logic functions include: a first function representing the causal logical relationship between the first input node and the intermediate node in the first node chain, and a second function representing the causal logical relationship between the second input node and the intermediate node in the first node chain. Query the target product in the database to be built, and obtain the second node chain information corresponding to the quality inspection items of the target product; wherein, the target product and the source product have different equipment types, quality inspection items and quality inspection standards, and the second node chain information includes the topology and node type of the second node chain; If the first node chain information and the second node chain information are the same, then based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment and process control plan, the key parameters of the causal relationship logic function in the first causal edge strategy are adjusted to obtain the adjusted second causal edge strategy. Based on the second node chain information and the second causal edge strategy, a second quality inspection knowledge base is constructed; Based on the second quality inspection knowledge base, quality inspection is carried out on the quality inspection items of the target product; Wherein, if the first node chain information and the second node chain information are the same, then based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment, and process control plan, the key parameters of the causal relationship logic function in the first causal edge strategy are adjusted, specifically including: Based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment and process control plan, the first degradation sensitivity parameter and the first critical usage amount included in the first function are adjusted. Based on the quantitative differences between the source product and the target product in the production environment, the first environmental sensitivity parameter included in the second function is adjusted; The adjustment of the first degradation sensitivity parameter and the first critical usage amount included in the first function based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment, and process control plan specifically includes: Based on the first environmental parameter benchmark value mapped to the quality inspection items of the source product and the second environmental parameter benchmark value mapped to the quality inspection items of the target product, a first coefficient is calculated to characterize the quantitative differences in the production environment. Based on the first device sensitivity score of the source product mapping and the second device sensitivity score of the target product mapping, a second coefficient is calculated to characterize the quantitative differences in physical characteristics; A third coefficient is calculated to characterize the quantitative difference in the process control plan based on the first process control threshold used for the quality inspection items of the source product and the second process control threshold used for the quality inspection items of the target product. The second degradation sensitivity parameter is calculated based on the first coefficient, the second coefficient, the third coefficient, and the first degradation sensitivity parameter. Based on the second degradation sensitivity parameter, the second critical usage amount is obtained by fitting a function through a logistic regression algorithm. The first degradation sensitivity parameter in the first function is adjusted to the second degradation sensitivity parameter, and the first critical usage amount is adjusted to the second critical usage amount; The adjustment of the first environmental sensitivity parameter included in the second function based on the quantitative differences in the production environment between the source product and the target product specifically includes: Based on the first coefficient and the first environmental sensitivity parameter, calculate the second environmental sensitivity parameter; Adjust the first environmental sensitivity parameter in the second function to the second environmental sensitivity parameter.

2. The method of claim 1, wherein, The node also includes a first output node representing the missed detection rate index and a second output node representing the final inspection defect rate index; the causal relationship logic function also includes a third function representing the causal logic relationship between the intermediate node and the first output node in the first node chain and a fourth function representing the causal logic relationship between the first output node and the second output node in the first node chain, the third function including a first detectivity benchmark value and the fourth function including a first severity value. After adjusting the key parameters of the causal relationship logic function in the first causal edge strategy based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment, and process control plan, if the first node chain information and the second node chain information are the same, the method further includes: From the failure mode and effects analysis database of the target product, obtain the second detectivity benchmark value and the second severity value mapped to the quality inspection items of the target product; Adjust the first detectivity reference value in the third function to the second detectivity reference value; Adjust the first severity value in the fourth function to the second severity value; The second causal edge strategy is generated based on the updated first function, second function, third function, and fourth function.

3. The method of claim 1, wherein, The node further includes constraint nodes for constraining process control of product quality inspection items; before adjusting the key parameters of the causal relationship logic function in the first causal edge strategy based on the quantitative differences between the source product and the target product in physical characteristics, production environment, and process control plan if the first node chain information and the second node chain information are the same, the method further includes: Obtain the second process control threshold used for the quality inspection items of the target product from the electronic control plan file of the target product; Based on the second process control threshold, the adjustment range of the control parameters of the constraint nodes in the second node chain is set.

4. The method of claim 1, wherein, After constructing the second quality inspection knowledge base based on the second node chain information and the second causal edge strategy, the method further includes: Obtain the perturbation variables that affect the final inspection defect rate of the quality inspection items of the target product; The historical distribution data of the disturbance variable is sampled to generate a preset number of first test samples. The first test samples include the sampled values ​​of the disturbance variable and the first target value of the parameter to be deduced. Using the second causal edge strategy in the second quality inspection knowledge base, a preset number of first test samples are deduced to obtain the first final inspection defect rate mapped to each first test sample. Calculate the mean increase of the first final inspection defect rate mapped to a preset number of first test samples; If the mean increase is greater than the first threshold, then a second target value is selected from the migration value of the parameter to be deduced in the second quality inspection knowledge base and the first target value, and the value of the parameter to be deduced is changed to the second target value.

5. The method of claim 1, wherein, After constructing the second quality inspection knowledge base based on the second node chain information and the second causal edge strategy, the method further includes: Obtain the perturbation variables that affect the final inspection defect rate of the quality inspection items of the target product; The historical distribution data of the disturbance variable is sampled to generate a preset number of first test samples. The first test samples include the sampled values ​​of the disturbance variable and the first target value of the parameter to be deduced. Using the second causal edge strategy in the second quality inspection knowledge base, a preset number of first test samples are deduced to obtain the first final inspection defect rate mapped to each first test sample; the mean and slope of the first final inspection defect rate mapped to the preset number of first test samples are calculated. If the mean is greater than the second threshold and the slope of change is greater than the third threshold, then the sensitive critical value of the disturbance variable is obtained. Based on the migration value of the parameter to be deduced in the second quality inspection knowledge base, the current value of the disturbance variable, and the sensitive critical value, a third target value is calculated, and the value of the parameter to be deduced is changed to the third target value.

6. The method according to claim 4 or 5, characterized in that, After performing quality inspections on the target product's quality inspection items based on the second quality inspection knowledge base, the method further includes: Obtain the target final inspection defect rate after quality inspection of the quality inspection items of the target product; Calculate the deviation ratio between the target final inspection defect rate and the mean of the second final inspection defect rate; the second final inspection defect rate is obtained by extrapolating a preset number of second test samples through the second causal edge strategy in the second quality inspection knowledge base, and the second test samples include the sampled values ​​of the disturbance variable and the changed values ​​of the parameter to be extrapolated; Analyze the root causes that lead to the deviation ratio exceeding the fourth threshold; In the second causal edge strategy, the objective function that matches the target root cause is located; The objective function is refitted using a logistic regression algorithm to update the key parameters in the objective function.

7. An industrial internet-based cross-product quality inspection device, comprising: include: The communicator is configured to communicate with the quality inspection execution unit via the Industrial Internet; The quality inspection execution unit is configured to perform quality inspection on the target product; The controller is configured as follows: A first quality inspection knowledge base is constructed to guide the quality inspection of source products. This knowledge base includes first node chain information and a first causal edge strategy corresponding to the quality inspection items of the source products. The first node chain information includes the topology and node types of the first node chain, and the first causal edge strategy includes causal relationship logic functions between nodes in the first node chain. Each node includes: a first input node representing a usage intensity index, a second input node representing an environmental disturbance index, and an intermediate node representing the product degradation state. The causal relationship logic functions include: a first function representing the causal logical relationship between the first input node and the intermediate node in the first node chain, and a second function representing the causal logical relationship between the second input node and the intermediate node in the first node chain. Query the target product in the database to be built, and obtain the second node chain information corresponding to the quality inspection items of the target product; wherein, the target product and the source product have different equipment types, quality inspection items and quality inspection standards, and the second node chain information includes the topology and node type of the second node chain; If the first node chain information and the second node chain information are the same, then based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment and process control plan, the key parameters of the causal relationship logic function in the first causal edge strategy are adjusted to obtain the adjusted second causal edge strategy. Based on the second node chain information and the second causal edge strategy, a second quality inspection knowledge base is constructed; The second quality inspection knowledge base is synchronized to the quality inspection execution unit through the communicator, so that the quality inspection execution unit can perform quality inspection on the quality inspection items of the target product based on the second quality inspection knowledge base; Wherein, if the first node chain information and the second node chain information are the same, the controller adjusts the key parameters of the causal relationship logic function in the first causal edge strategy based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment, and process control plan. Specifically, it is configured as follows: Based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment and process control plan, the first degradation sensitivity parameter and the first critical usage amount included in the first function are adjusted; Based on the quantitative differences between the source product and the target product in the production environment, the first environmental sensitivity parameter included in the second function is adjusted; The controller performs adjustments to the first degradation sensitivity parameter and the first critical usage amount included in the first function based on the quantitative differences between the source product and the target product in terms of physical characteristics, production environment, and process control plan. Specifically, it is configured as follows: Based on the first environmental parameter benchmark value mapped to the quality inspection items of the source product and the second environmental parameter benchmark value mapped to the quality inspection items of the target product, a first coefficient is calculated to characterize the quantitative differences in the production environment. Based on the first device sensitivity score of the source product mapping and the second device sensitivity score of the target product mapping, a second coefficient is calculated to characterize the quantitative differences in physical characteristics; A third coefficient is calculated to characterize the quantitative difference in the process control plan based on the first process control threshold used for the quality inspection items of the source product and the second process control threshold used for the quality inspection items of the target product. The second degradation sensitivity parameter is calculated based on the first coefficient, the second coefficient, the third coefficient, and the first degradation sensitivity parameter. Based on the second degradation sensitivity parameter, the second critical usage amount is obtained by fitting a function through a logistic regression algorithm. The first degradation sensitivity parameter in the first function is adjusted to the second degradation sensitivity parameter, and the first critical usage amount is adjusted to the second critical usage amount; The controller adjusts the first environmental sensitivity parameter included in the second function based on the quantitative differences between the source product and the target product in the production environment, specifically configured as follows: Based on the first coefficient and the first environmental sensitivity parameter, calculate the second environmental sensitivity parameter; Adjust the first environmental sensitivity parameter in the second function to the second environmental sensitivity parameter.

Images