A defect root cause localization method, system, device, and storage medium

Abstract

The application provides a defect root cause positioning method, system, device and storage medium, and relates to the field of industrial automation.The method flow is as follows: the quality detection data and processing history data of products are cleaned and pretreated to form process flow data; a suspicious degree measurement model is constructed based on the process flow data; node importance analysis is carried out based on the suspicious degree measurement model, and suspicious nodes are initially determined based on the node importance analysis result; the value difference of the process parameters of the suspicious nodes between normal products and abnormal products is measured, and the suspicious nodes are determined again based on the value difference; and the suspicious node set causing the product quality defect and the related process parameters are output.The application describes the relationship between different nodes in the process flow through a complex network, fully considers the time sequence and correlation between the nodes, and determines the suspicious nodes in combination with the importance analysis mode, so that the defect root cause can be more accurately identified and positioned.

Description

Technical Field

[0001] This invention relates to the field of industrial automation, and more specifically, to a method, system, device, and storage medium for locating the root cause of defects. Background Technology

[0002] In industrial manufacturing, identifying the root causes of defects is crucial for improving production efficiency and product quality. Existing methods for identifying the root causes of defects mainly include traditional statistical analysis methods and methods based on human experience.

[0003] Existing methods for identifying the root causes of defects have some limitations and shortcomings, specifically: (1) Traditional statistical analysis methods require a large amount of data processing and complex calculations in identifying the root causes of defects, resulting in long analysis and location time and low efficiency; in addition, traditional statistical analysis methods often require specific data distribution and assumptions, which limits their applicability and flexibility; (2) Human experience judgment methods are easily affected by individual subjective opinions and experience, resulting in a lack of objectivity and consistency in the results; in addition, human experience judgment methods are often limited by the individual's cognitive and experience level, resulting in numerous factors and complex relationships in the identification of defects, which limits their ability to analyze the root causes of defects. Summary of the Invention

[0004] This invention provides a method, system, device, and storage medium for locating the root cause of defects, which solves the problem that existing technologies struggle to accurately locate the root cause of defects in industrial products due to numerous process parameters, high correlation between process parameters, and complex product processes.

[0005] In a first aspect, embodiments of the present invention provide a method for locating the root cause of a defect, the method comprising the following steps:

[0006] Clean and preprocess the product's quality inspection data and processing history data to form process flow data;

[0007] A suspicion level measurement model was constructed based on process flow data;

[0008] Node importance analysis is performed based on a suspiciousness measurement model, and suspicious nodes are initially identified based on the results of the node importance analysis.

[0009] The process parameters of suspicious nodes are measured to differ between normal and abnormal products, and the suspicious nodes are re-identified based on the differences in values.

[0010] Output a set of suspected nodes that could lead to product quality defects, along with the relevant process parameters.

[0011] In the above embodiments, complex networks are used to characterize the relationships between different nodes in the process flow, fully considering the temporal sequence and correlation between nodes, and combining importance analysis to determine suspicious nodes, which can more accurately identify and locate the root cause of defects.

[0012] As some optional embodiments of this application, the process for preprocessing product quality inspection data and processing history data is as follows:

[0013] Correlation analysis is performed on numerical process parameters to eliminate those with correlation coefficients exceeding the numerical threshold.

[0014] Non-numerical process parameters are encoded to convert the parameter type of the process parameters.

[0015] In the above embodiments, preprocessing the data can avoid serious collinearity problems and the problem that the type selection parameters cannot be directly used for subsequent analysis, thereby improving the accuracy and efficiency of data analysis.

[0016] As some optional embodiments of this application, the process flow data includes the product's process steps, flow relationships, and process parameters.

[0017] As some optional implementations of this application, the process for constructing a suspicion level measurement model based on process flow data is as follows:

[0018] The product's technological processes are treated as nodes in a complex network, and the product's circulation relationships are treated as edges in the complex network.

[0019] A suspicion level measurement model is constructed based on the association relationships of nodes, where the association relationships of nodes are determined based on the temporal order of nodes or the correlation coefficient.

[0020] In the above embodiments, complex networks are used to characterize the relationships between different nodes in the process flow, and the temporal relationships between nodes in the process flow are fully considered to determine the connection relationships between nodes. Mapping the process flow to a complex network can more accurately capture the propagation path of defects and the degree of influence of different process links on the overall production quality, making the process flow more intuitive and easier to analyze.

[0021] As some optional implementations of this application, the process of performing node importance analysis based on a suspiciousness measurement model and determining suspicious nodes based on the node importance analysis results is as follows:

[0022] Calculate the importance index of all nodes in the suspicion metric model;

[0023] Based on the importance index of all nodes, several nodes are selected as suspicious nodes.

[0024] In the above embodiments, the suspiciousness measurement model is analyzed from the perspective of measuring the importance of nodes and suspicious process links are explored. This can quantitatively assess the importance of nodes in the entire processing process and thus identify suspicious nodes that cause defects. It can better consider the correlation and temporal sequence between nodes, thereby more accurately identifying the root causes of potential process defects.

[0025] As some optional embodiments of this application, the formula for calculating the importance index is as follows:

[0026]

[0027] Among them, S i K represents the importance index of node i. out PV represents the in-degree and out-degree of a node. i The entropy value represents the distribution of the number of products on different branches of the node, where N represents the total number of products and N' represents the number of products entering the node.

[0028] As some optional embodiments of this application, the formula for calculating the entropy value is as follows:

[0029]

[0030] Among them, PV i Let j represent the entropy value of the product quantity distribution on different branches of node i, j represent the branches of node i, and w represent the total number of products on all branches of node i. j This represents the number of products flowing from node i to branch j.

[0031] As some optional embodiments of this application, the process for measuring the difference in process parameters of suspicious nodes between normal and abnormal products, and determining suspicious nodes based on the difference in values, is as follows:

[0032] The process parameters of the suspicious nodes are obtained for normal and abnormal products, and two probability distribution functions are constructed based on the values.

[0033] Calculate the divergence of the two probability distribution functions. If the divergence exceeds the divergence threshold, continue to trace the predecessor node of the suspicious node; otherwise, it is determined to be a suspicious node again.

[0034] In the above embodiments, while outputting the set of suspicious nodes, data such as the connection relationships, path information, and process parameters between these suspicious nodes are also provided to better understand the interaction and influence between the nodes in the process flow.

[0035] As some optional embodiments of this application, the divergence is KL divergence or JS divergence.

[0036] In a second aspect, the present invention provides a defect root cause localization system, the system comprising:

[0037] The data processing unit is used to clean and preprocess the product's quality inspection data and processing history data to form process flow data.

[0038] A model building unit, which builds a suspicion degree measurement model based on process flow data;

[0039] The importance analysis unit performs node importance analysis based on a suspiciousness measurement model, and initially identifies suspicious nodes based on the node importance analysis results.

[0040] A suspicious node location unit is used to measure the difference in the values ​​of process parameters of suspicious nodes between normal and abnormal products, and to determine the suspicious node again based on the difference in values.

[0041] The data output unit is used to output a set of suspected nodes that lead to product quality defects and related process parameters.

[0042] In a third aspect, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the defect root cause localization method.

[0043] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the defect root cause localization method.

[0044] The beneficial effects of this invention are as follows:

[0045] 1. This invention uses complex networks to characterize the relationships between different nodes in the process flow, fully considering the temporal relationships between nodes in the process flow to determine the connection relationships between nodes. By mapping the process flow into a complex network, the propagation path of defects and the degree of influence of different process links on the overall production quality can be captured more accurately, making the process flow more intuitive and easier to analyze.

[0046] 2. This invention analyzes the suspiciousness measurement model from the perspective of measuring the importance of nodes, and explores suspicious process flow links. This model can quantitatively assess the importance of nodes in the entire processing flow and thus identify suspicious nodes that lead to defects. It can better consider the correlation and temporal sequence between nodes, thereby more accurately identifying potential root causes of process defects. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of the defect root cause localization method described in an embodiment of the present invention;

[0049] Figure 2 This is a flowchart illustrating the determination of suspicious nodes using KS divergence as described in an embodiment of the present invention;

[0050] Figure 3 This is a network diagram of the glass panel manufacturing process described in an embodiment of the present invention;

[0051] Figure 4 This is a flowchart of the 115600 node production process according to an embodiment of the present invention;

[0052] Figure 5 This is a distribution diagram of parameter values ​​115600 as described in an embodiment of the present invention;

[0053] Figure 6 This is a distribution diagram of parameter values ​​115400 as described in an embodiment of the present invention;

[0054] Figure 7 This is a distribution diagram of parameter values ​​115402 as described in an embodiment of the present invention;

[0055] Figure 8 This is a structural block diagram of the defect root cause localization system described in an embodiment of the present invention. Detailed Implementation

[0056] To better understand the above technical solutions, the technical solutions of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solutions of the present invention, rather than limitations thereof. In the absence of conflict, the technical features in the embodiments and embodiments can be combined with each other.

[0057] It should also be understood that, in order to simplify the description of the invention and thus aid in the understanding of at least one embodiment, multiple features may sometimes be grouped into a single embodiment, drawing, or description thereof in the foregoing description of the embodiments of the invention. However, this method of disclosure does not imply that the subject matter of the invention requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of the single embodiment disclosed above.

[0058] In industrial manufacturing processes, numerous process parameters, high correlation between these parameters, and complex product processes often present technical challenges, making it difficult to accurately pinpoint the root cause of defects in industrial products. Therefore, this invention provides a defect root cause localization method, aiming to effectively solve the above technical problems and help industrial manufacturing enterprises quickly and accurately locate and resolve defective product issues.

[0059] Example 1

[0060] This invention provides a method for locating the root cause of defects. Please refer to [link / reference]. Figure 1 , Figure 1 The flowchart of the method is as follows:

[0061] (1) Clean and preprocess the product quality inspection data and processing history data to form complete process flow data.

[0062] In this embodiment of the invention, the processing history data includes, but is not limited to, the following aspects: the flow relationship and process parameters of each process step in the production process. The process parameters mainly include process status parameters monitored by the production control equipment, such as temperature, humidity, pressure, and current during the production process; intermediate state parameters of the product monitored during the sampling inspection process, such as the thickness and surface cleanliness of glass during panel production, and the pin quality during electronic component processing; and material and time data during the production process, such as the proportion of different materials and the processing time of a certain step.

[0063] In this embodiment of the invention, in order to ensure the feasibility and accuracy of large-scale quality inspection data and processing history data, it is necessary to perform data cleaning on the collected product quality inspection data and processing history data. Specifically, the data cleaning process is as follows:

[0064] (1.1) Based on the product's quality inspection data, the product is labeled with different quality levels to obtain the labeling data and the corresponding processing history data;

[0065] (1.2) Analyze the number of times the same product processing record is repeated in the processing history data, and delete the processing record data that is exactly the same;

[0066] (1.3) Sort all product processing record data by product number and time, and integrate processing record data of the same product at different times;

[0067] (1.4) Remove process parameters with many missing values, process parameters with fixed values, and processing record data with serious missing values.

[0068] In this embodiment of the invention, when performing data cleaning, additional conditions in specific fields (such as process flow codes) can be used to assist in screening abnormal data in order to achieve data cleaning. This embodiment of the invention does not limit this.

[0069] In this embodiment of the invention, in order to ensure the feasibility and accuracy of large-scale process parameter analysis, the cleaned process parameters need to undergo data preprocessing. Specifically, the data preprocessing process is as follows:

[0070] (1.5) In order to avoid serious collinearity problems and improve the accuracy and efficiency of data analysis, Pearson correlation coefficient is used to calculate the correlation between all numerical process parameters and process parameters with correlation coefficients exceeding the numerical threshold are removed.

[0071] (1.6) In order to solve the problem that type selection parameters cannot be directly used for subsequent analysis and improve the feasibility and accuracy of data analysis, non-numerical process parameters are encoded to transform the parameter type of process parameters. That is, the One-Hot encoding method is used to convert a single type selection parameter into multiple independent type selection variables.

[0072] In this embodiment of the invention, when performing data preprocessing, other preprocessing steps can be added or reduced according to the data quality and characteristics of the process parameters in the industrial manufacturing process. For example, for information on parallel processing, information such as the number of parallel equipment can be processed as process parameters. This embodiment of the invention does not limit this.

[0073] (2) Construct a suspiciousness measurement model based on complete process flow data, wherein the preprocessed process flow data includes, but is not limited to, process steps, flow relationships and process parameters.

[0074] In this embodiment of the invention, modeling the complete process flow data involves using a complex network to sequentially connect the processing steps of each product during production according to the processing flow path of all products. In this complex network, the process links of the product are used as nodes of the complex network, and the flow relationship of the product is used as the edge of the complex network. The final suspiciousness measurement model is constructed based on the association relationship of the nodes.

[0075] Specifically, the relationships between different nodes are determined based on their temporal sequence, that is, the production order of the product at different nodes. It should be noted that other factors can also be used, such as sampling the Pearson correlation coefficient, to characterize the relationships between nodes, thereby determining the connections between them and ensuring the accuracy and reliability of the suspicion metric model.

[0076] (3) Node importance analysis is performed based on the suspiciousness measurement model, and suspicious nodes are determined based on the results of the node importance analysis; that is, by calculating the node importance index, nodes that are highly related to defects are screened out, and these nodes are selected as the core nodes (suspicious nodes) of the network. These core nodes have a greater impact on the quality defects of the product, so they should be given greater importance and more attention when conducting root cause analysis of product quality defects.

[0077] In this embodiment of the invention, the node importance analysis process is as follows:

[0078] (3.1) Calculate the importance index of all nodes; wherein the formula for calculating the importance index is as follows:

[0079]

[0080] Among them, S i K represents the importance index of node i. out PV represents the in-degree and out-degree of a node. i This represents the entropy value of the product quantity distribution on different branches of the node, where N represents the total number of products and N′ represents the number of products entering the node. In other words, the importance of a node (i.e., the technological step in the production process) is assessed by calculating the node's in-degree, out-degree, and entropy value.

[0081] Specifically, the number of products N′ entering a node can reflect the node's location information, that is, the node's hierarchical position in the network model. The more products entering a node, the more important the node's position in the network model, indicating that this process is an important processing step, and a large number of products must go through this process.

[0082] Specifically, the in-degree and out-degree K of a node out It can reflect the connectivity of nodes, that is, the number of connections between node i and other nodes. The greater the in-degree of a node, the more connections the node has in the network model, which means that the process link is more important.

[0083] Specifically, the PV of a node i The entropy value reflects the distribution of the number of products on different branches of a node. According to the definition and calculation method of entropy, the larger the entropy value, the more evenly the number of products flowing to different subsequent processing branches at that node is. In other words, a large number of products flow to more branches at that node, and the flow of products to subsequent processing branches at that process stage varies greatly. This means that the production process control at that process stage is not good, and that process stage is more likely to be an important link that leads to product defects.

[0084] Specifically, the formula for calculating the entropy value is as follows:

[0085]

[0086] Where j represents the branch of node i, w represents the total number of products on all branches of node i, and w j This represents the number of products flowing from node i to branch j.

[0087] (3.2) Select several nodes as suspicious nodes based on the importance index of all nodes. Generally, since the importance index S i The importance index S is a comprehensive calculation value that combines multiple factors such as node in-degree and out-degree and the quantity distribution of products on different process paths. It reflects the importance of the node in the entire process network. i Larger nodes are often associated with critical steps or abnormal situations in the production process. Therefore, according to the importance index S... i The size of the data can determine the importance of each node in the production process, thereby identifying the suspected processing steps that may lead to product defects.

[0088] In this embodiment of the invention, the importance index S is selected. i The first k nodes are considered as suspicious nodes and are further analyzed and diagnosed in subsequent steps to promptly identify and resolve potential product defect root causes. The value of k can be set according to business needs, and this invention does not impose any restrictions on it.

[0089] Specifically, the construction of a complex network for the entire process flow can be achieved through data visualization, that is, by using relevant tools (including but not limited to Tableau, Power BI, etc.) to automate the process. By using a complex network to present the entire process flow, and in conjunction with the suspicion degree measurement model constructed in this embodiment of the invention, it can further help users understand and analyze key nodes and abnormal situations in the production process flow.

[0090] (4) Measure the difference in the values ​​of process parameters of suspicious nodes between normal and abnormal products, and determine the suspicious nodes again based on the difference in values;

[0091] Specifically, the process for re-identifying suspicious nodes based on value differences is as follows:

[0092] (4.1) Obtain the values ​​of the process parameters of the suspicious nodes in normal products and abnormal products respectively, and construct two probability distribution functions based on the values;

[0093] (4.2) Calculate the divergence of the two probability distribution functions. If the divergence exceeds the divergence threshold, continue to trace the predecessor node of the suspicious node; otherwise, it is determined to be a suspicious node again.

[0094] In the implementation of this invention, KS divergence is used as a statistical index to quantify the difference in the values ​​of process parameters of suspicious nodes between normal and abnormal products.

[0095] In the implementation of this invention, for each suspicious node, two probability distribution functions are constructed by comparing the values ​​of their process parameters between normal products and abnormal products, respectively reflecting the values ​​of process parameters of normal products and abnormal products at each suspicious node.

[0096] Specifically, the formula for calculating KS divergence is as follows:

[0097] D = max|F1x - F2x|

[0098] Where D represents the KS divergence, F1(x) and F2(x) represent the cumulative distribution functions of a certain process parameter on normal and abnormal products, respectively, and x represents the value of the process parameter. By comparing the cumulative distribution functions of the two sets of data, the KS divergence can capture the maximum deviation between them and provide an effective way to evaluate the difference in process parameters between normal and abnormal products at suspicious nodes.

[0099] In particular, the KS divergence calculation method is not only applicable to the difference analysis of a single process step, but can also be extended to the joint difference analysis of multiple process steps. By constructing a joint probability distribution function of the process parameter values ​​of multiple process steps, the difference in the values ​​of process parameters of multiple process steps between normal and abnormal products can be measured to assess the comprehensive impact of process parameters of multiple process steps on product quality defects.

[0100] Furthermore, if, considering the characteristics of the business, it is assumed that the values ​​of a certain process parameter across all products should follow a normal distribution (i.e., the "theoretical distribution" of the process parameter), the KS divergence can be used to calculate the difference between the distribution of the process parameter across all products and the "theoretical distribution." This embodiment of the invention is not limiting; this part is not the focus of the invention and will not be elaborated further.

[0101] It should be emphasized that the present invention uses KS divergence to measure the difference in the distribution of process parameters of suspicious nodes between normal and abnormal products. Other mathematical indicators can also be selected to measure the difference in the distribution of process parameters between normal and abnormal products, including but not limited to: KL divergence and JS divergence. The embodiments of the present invention do not limit this. Preferably, the present invention chooses to use KS divergence.

[0102] If the KS divergence is greater than a preset divergence threshold d, then the process parameters of the suspicious node are considered to differ significantly between normal and abnormal products. Please refer to [link / reference]. Figure 2 , Figure 2 The flowchart for identifying suspicious nodes using KS divergence is as follows:

[0103] ① If the difference in the distribution of process parameters between normal and abnormal products at the suspicious node M1 exceeds the divergence threshold d, it indicates that the suspicious node may have a significant impact on product quality defects and requires further analysis and processing.

[0104] ② If the difference in the distribution of process parameters of the suspected node M1 between normal and abnormal products is less than the divergence threshold d, then the preceding node M of the suspected node will be traced back. n This is to determine whether there are other preceding process steps that may affect product quality defects. This allows for analysis of the problem from the perspective of the entire process flow, identifying potential sources of problems, and taking corresponding corrective measures. In the implementation of this invention, the process continues until the preceding node exhibits a KS divergence exceeding a preset divergence threshold.

[0105] Furthermore, the KS divergence of each parameter in the process steps that cause the problem is sorted, thereby identifying a set of suspicious nodes with significant distribution differences in these process steps based on the sorting of KS divergence. These suspicious nodes and their corresponding process parameters may have a significant impact on product quality defects.

[0106] (5) Output the set of suspected nodes that cause product quality defects and related process parameters. That is, based on the above steps (3) and (4), it is possible to determine which suspected nodes have a significant impact on product quality defects and find the process flow links that may cause problems.

[0107] That is, when outputting the set of suspicious nodes, the connection relationship, path information and process parameters between these suspicious nodes are also given, so as to better understand the interaction and influence between the nodes in the process flow.

[0108] To more clearly demonstrate the purpose, technical solution, and advantages of this invention, the following section uses the problem of locating the root cause of defective products in glass panel manufacturing as an example, and elaborates on the invention in detail with reference to the accompanying drawings and specific embodiments.

[0109] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0110] In the glass panel manufacturing process, fluctuations in the processing parameters of the equipment can lead to the production of defective glass. The processing parameters of glass panels have two prominent characteristics: (1) high degree of automation, the process flow is set in advance by engineers, and the processing flow is highly automated; (2) large amount of data, because each process equipment generates a large number of processing parameters, so a whole process flow will generate a massive amount of data.

[0111] Traditional statistical analysis and manual experience-based judgment methods are clearly insufficient for effectively locating the root causes of problems in this scenario. To address these issues, this invention uses the aforementioned defect root cause localization method for defect root cause localization.

[0112] The following is an analysis of product data related to a certain defect that occurred during the production of a specific type of glass panel. The detailed process is as follows:

[0113] S1. Clean and preprocess the product's quality inspection data and processing history data to form complete process flow data.

[0114] Specifically, the data cleaning process is as follows:

[0115] S11. Labeling Data: Based on the quality inspection data, the products are divided into two categories: "good" and "bad", and the labeling data is obtained.

[0116] S12. Preliminary data cleaning: Remove duplicate data from the processing records of the same product and retain the non-duplicate processing records;

[0117] S13. Construct a wide parameter table: Sort the product processing records by product number and time, and integrate the parameters in the processing records of different stages and nodes of the same product to form a wide table of product parameters indexed by product number; this step can distinguish process parameters with the same parameter name, such as by adding node number or process flow stage name;

[0118] S14. Deep Data Cleaning:

[0119] ① Eliminate process parameters with severe deficiencies: Calculate the product coverage rate of each process parameter and eliminate process parameters with a coverage rate lower than the preset value.

[0120] ② Eliminate process parameters with fixed values: Eliminate process parameters with the same value for all products to reduce redundant information.

[0121] ③ Remove product processing records with severe deficiencies: Calculate the coverage rate of process parameters in each product processing record and remove product processing records with a coverage rate lower than the preset value.

[0122] Specifically, the data preprocessing process is as follows:

[0123] S15. Delete highly correlated process parameters: For all numerical process parameters, calculate the Pearson correlation coefficient between each numerical process parameter and all other numerical process parameters. If the correlation coefficient between any two numerical process parameters exceeds a preset value, one of the parameters will be selectively deleted. This reduces collinearity among parameters and improves the speed of subsequent model analysis and calculation.

[0124] S16. Feature Encoding: For situations with multiple parallel nodes, devices, chambers, etc., multiple independent type selection variables are used to replace a single type selection parameter, i.e., one-hot encoding is used to generate type variables.

[0125] S2. Construct a suspiciousness measurement model based on complete process flow data, and select the top ten nodes in importance as suspicious nodes based on importance analysis;

[0126] S21. Constructing complex networks based on the sequence of process production nodes;

[0127] For the labeled data and process flow data obtained in step S1, the processing of each glass panel is sorted based on the production time. Each node in the production process (including processing nodes and inspection nodes) is represented as a node in the network. The edges of the network are constructed according to the flow relationship between nodes (process links) in the process flow, forming the entire production process path of the glass panel. Since different glass panels will experience different processing steps during the processing, some processing paths will overlap and some will not overlap. In the end, the processing flow of all panels will form a network.

[0128] Please see Figure 3 , Figure 3 This is a network diagram illustrating the entire manufacturing process for a certain type of glass panel. In the diagram, type A nodes represent process nodes, type B nodes represent inspection nodes, and type C nodes represent nodes with missing site type information due to incomplete data records. The process flow represents the main manufacturing process. Figure 3 It can be seen that from the "start node" (the first process step that all products must go through) to the "end node" (the last process step that almost all products except scrapped products will reach), not only is the processing flow of a single panel long, but the processing paths of different panels are also different (there are multiple branches in the network). This network diagram fully, vividly and intuitively shows the complexity of glass panel products in the production and processing process.

[0129] S22. The importance index of each node can be calculated based on its in-degree and entropy values.

[0130] For example, as shown in Table 1, the importance index of each node in the glass panel process is calculated (only the data of the top five nodes is shown here).

[0131] Based on the calculation results, node 115600 is initially suspected to be the node most likely to cause production defects, followed by node 11B600, which is the second most likely node to cause production defects, and so on, to obtain the top ten suspected nodes.

[0132]

[0133]

[0134] Table 1

[0135] S3. Measure the difference in process parameters of suspicious nodes between normal and abnormal products, and determine suspicious nodes based on the difference in values;

[0136] S31. Analyze each identified suspicious node: For example, for the data in this embodiment, based on the identified suspicious nodes, first select node 115600, which ranks first in suspicion, for analysis. Please refer to [link to relevant documentation]. Figure 4 , Figure 4 The diagram shows a local processing step of this node in the complex network diagram of the entire process flow. It can be seen that the following problems exist for node 115600: there are a large number of detection nodes in the subsequent steps of this node, indicating that this node is very important; moreover, the branches at this node are very complicated and require special attention.

[0137] S32. The difference in the values ​​of the processing parameters at node 115600 on good and defective products is measured by calculating the KS divergence of the processing parameters on good and defective products.

[0138] Please see Figure 5 , Figure 5 The graph shows the value distribution of parameter A at node 115600, representing the value distribution of parameter A for good and defective products. It is clear that the value distribution for parameter A at node 115600 is not significantly different between good and defective products.

[0139] Furthermore, its KS divergence value is calculated to be 0.1, which is less than the preset threshold (in this embodiment, the preset threshold is set to 0.2). The KS divergence of all parameters of this node is traversed in turn, and it is found that they are all less than the preset threshold. Then, its previous node 115400 is traced back.

[0140] S33. Calculate the KS divergence of each parameter at node 115400. Please refer to [link / reference]. Figure 6 , Figure 6 The parameter value distribution diagram for node 115400 represents the value distribution of parameter B in good and defective products. After calculation, its KS divergence is 0.19, which is less than the preset threshold (in this embodiment, the preset threshold is set to 0.2). The KS divergence of all parameters of this node is traversed in turn, and it is found that they are all less than the preset threshold. Then, the previous node 115402 is traced back.

[0141] S34. Calculate the KS divergence of each parameter at node 115402. Please refer to [link / reference]. Figure 7 , Figure 7 The graph shows the value distribution of parameter 115402, representing the value distribution of parameter C1 at node 115402 for good and defective products.

[0142] Calculations show that the KS divergence is greater than the threshold (the preset threshold is set to 0.2 in this embodiment), indicating that the parameter values ​​of node 115402 have a very large difference and are most likely to be abnormal.

[0143] The reason why node 115600 is considered important is likely due to the abnormal parameters caused by the anomaly of node 115402. Thus, from the initial suspicious node 115600, the suspicious node 115402 from its preceding steps has been located. Therefore, node 115600 is replaced with 115402 in Table 1.

[0144] S4. Output the set of suspected nodes that lead to product quality defects and the relevant process parameters.

[0145] Based on the analysis results of the above nodes, the set of nodes in the process flow that are most important to product quality defects is directly output. According to the ranking of the KS divergence of different processing parameters on the nodes, the relevant key influencing parameters on the nodes are determined. When outputting the set of nodes, the connection relationship and path information between these key nodes, as well as the set of key influencing parameters, are also given to better understand the interaction and influence between the nodes in the process flow.

[0146] For example, as shown in Table 2, the final results of the root cause analysis of the process defects of this batch of panels are presented. The table shows the suspected sites that caused the product defects and the suspected processing parameters at the sites, as well as the local processing path of the suspected sites, so as to further assist relevant personnel in further analysis in conjunction with specific business.

[0147]

[0148] Table 2

[0149] In summary, this invention characterizes the relationships between different nodes in a process flow using complex networks, fully considering the temporal sequence and correlation between nodes. Furthermore, it combines importance analysis to evaluate node importance based on in-degree and entropy values. Therefore, it can more accurately identify and locate the root cause of defects, solving the problem that existing technologies struggle to accurately locate the root cause of defects in industrial products due to numerous process parameters, high correlation between some parameters, and complex product processes.

[0150] Example 2

[0151] This invention provides a defect root cause localization system; please refer to [link / reference]. Figure 8 , Figure 8The diagram shows the structure of the system, which corresponds one-to-one with the method described in Embodiment 1. The system includes:

[0152] The data processing unit is used to clean and preprocess the product's quality inspection data and processing history data to form process flow data.

[0153] A model building unit, which builds a suspicion degree measurement model based on process flow data;

[0154] The importance analysis unit performs node importance analysis based on a suspiciousness measurement model, and preliminarily identifies suspicious nodes based on the node importance analysis results.

[0155] A suspicious node location unit is used to measure the difference in the values ​​of process parameters of suspicious nodes between normal and abnormal products, and to determine the suspicious node again based on the difference in values.

[0156] The data output unit is used to output a set of suspected nodes that lead to product quality defects and related process parameters.

[0157] Example 3

[0158] The present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the defect root cause localization method described in Embodiment 1.

[0159] The computer device provided in this embodiment can implement the method described in Embodiment 1. To avoid repetition, it will not be described again here.

[0160] Example 4

[0161] This invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the defect root cause localization method described in Embodiment 1.

[0162] The computer-readable storage medium provided in this embodiment can implement the method described in Embodiment 1. To avoid repetition, it will not be described again here.

[0163] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0164] The memory can be used to store the computer program and / or modules. The processor implements various functions of the defect root cause localization system of the invention by running or executing the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart memory card, secure digital card, flash memory card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0165] If a defect root cause localization system is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program that can be stored in a computer-readable storage medium. When executed by a processor, this computer program can implement the steps of the various method embodiments described above. The computer program includes computer program code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory, random access memory, dot carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0166] The basic concepts of this invention have been described. Obviously, for those skilled in the art, the above detailed disclosure is merely illustrative and does not constitute a limitation of this specification. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this specification. Such modifications, improvements, and corrections are suggested in this specification and therefore remain within the spirit and scope of the exemplary embodiments described herein.

Claims

1. A method for locating the root cause of a defect, characterized in that, The method includes the following steps: Clean and preprocess the product's quality inspection data and processing history data to form process flow data; A suspicion level measurement model was constructed based on process flow data; The process for constructing a suspicion level measurement model based on process flow data is as follows: The product's technological processes are treated as nodes in a complex network, and the product's circulation relationships are treated as edges in the complex network. A suspiciousness measurement model is constructed based on the association relationship of nodes, wherein the association relationship of nodes is determined based on the temporal order of nodes or the correlation coefficient; Node importance analysis is performed based on a suspiciousness measurement model, and suspicious nodes are initially identified based on the results of the node importance analysis. The process of performing node importance analysis based on the suspiciousness measurement model, and initially identifying suspicious nodes based on the results of the node importance analysis, is as follows: Calculate the importance index of all nodes in the suspicion metric model; Based on the importance index of all nodes, select several nodes as suspicious nodes; The formula for calculating the importance index is as follows: ； in, This represents the importance index of node i. Indicates the in-degree and out-degree of a node. This represents the entropy value of the product quantity distribution on different branches of the node. Indicates the total number of products. This indicates the number of products entering this node; The formula for calculating the entropy value is as follows: ; in, Let represent the entropy value of the product quantity distribution on different branches of node i, j represent the branches of node i, and w represent the total number of products on all branches of node i. This represents the number of products flowing from node i to branch j; The process parameters of suspicious nodes are measured to differ between normal and abnormal products, and the suspicious nodes are re-identified based on the differences in values. Output a set of suspected nodes that could lead to product quality defects, along with the relevant process parameters.

2. The defect root cause localization method according to claim 1, characterized in that, The preprocessing procedure for product quality inspection data and processing history data is as follows: Correlation analysis is performed on numerical process parameters to eliminate those with correlation coefficients exceeding the numerical threshold. Non-numerical process parameters are encoded to convert the parameter type of the process parameters.

3. The defect root cause localization method according to claim 1, characterized in that, The process flow data includes the product's process steps, flow relationships, and process parameters.

4. The defect root cause localization method according to claim 1, characterized in that, The process for measuring the differences in process parameters for suspicious nodes between normal and abnormal products, and then re-identifying suspicious nodes based on these differences, is as follows: The process parameters of the suspicious nodes are obtained for normal and abnormal products, and two probability distribution functions are constructed based on the values. Calculate the divergence of the two probability distribution functions. If the divergence exceeds the divergence threshold, it is identified as a suspicious node. Otherwise, continue to trace the preceding nodes of the suspicious node to determine whether there are other preceding process steps that affect product quality defects.

5. The defect root cause localization method according to claim 4, characterized in that, The divergence is either KL divergence or JS divergence.

6. A defect root cause localization system, characterized in that, The system includes: The data processing unit is used to clean and preprocess the product's quality inspection data and processing history data to form process flow data. A model building unit, which builds a suspicion degree measurement model based on process flow data; The process for constructing a suspicion level measurement model based on process flow data is as follows: The product's technological processes are treated as nodes in a complex network, and the product's circulation relationships are treated as edges in the complex network. A suspiciousness measurement model is constructed based on the association relationship of nodes, wherein the association relationship of nodes is determined based on the temporal order of nodes or the correlation coefficient; The importance analysis unit performs node importance analysis based on a suspiciousness measurement model, and initially identifies suspicious nodes based on the node importance analysis results. The process of performing node importance analysis based on the suspiciousness measurement model, and initially identifying suspicious nodes based on the results of the node importance analysis, is as follows: Calculate the importance index of all nodes in the suspicion metric model; Based on the importance index of all nodes, select several nodes as suspicious nodes; The formula for calculating the importance index is as follows: ; in, This represents the importance index of node i. Indicates the in-degree and out-degree of a node. This represents the entropy value of the product quantity distribution on different branches of the node. Indicates the total number of products. This indicates the number of products entering this node; The formula for calculating the entropy value is as follows: ; in, Let represent the entropy value of the product quantity distribution on different branches of node i, j represent the branches of node i, and w represent the total number of products on all branches of node i. This represents the number of products flowing from node i to branch j; A suspicious node location unit is used to measure the difference in the values ​​of process parameters of suspicious nodes between normal and abnormal products, and to determine the suspicious node again based on the difference in values. The data output unit is used to output a set of suspected nodes that lead to product quality defects and related process parameters.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes a computer program, it implements a defect root cause localization method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements a defect root cause localization method according to any one of claims 1-5.

Images