LCD display screen production quality collaborative management and control method based on detection data chain

Abstract

The invention discloses an LCD display screen production quality collaborative management and control method based on a detection data chain, particularly relates to the technical field of industrial production quality management, and is used for solving the problem of quality root cause misjudgment caused by complex association of multi-source data in the existing liquid crystal display screen manufacturing process. The method comprises the following steps of: constructing a detection data chain covering an array, a box forming process and a module process, firstly performing multi-source data correlation analysis to obtain a variable pair set, then analyzing an intermediary influence path between variable pairs to eliminate false correlation, and then determining a real causal direction by applying a causal discovery algorithm based on conditional independence. Carrying out unsupervised clustering analysis on the basis of defect characteristic indexes, verifying the consistency of causal directions in different data subsets, and finally outputting a stable quality root cause analysis result to realize accurate positioning and effective management and control of key process parameters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial production quality management, and particularly relates to a LCD display screen production quality collaborative management method based on detection data chain. BACKGROUND

[0002] In the field of LCD display screen manufacturing, constructing a detection data chain covering array, cell and module processes is a basic means for modern factories to carry out quality management. The prior art usually deploys various types of automated detection equipment, collects detection data corresponding to each station and product unit, and stores it in a database. Based on the detection data, quality management personnel generally use statistical process control or correlation analysis methods to locate the root cause of the final product defects in order to develop targeted process improvement strategies.

[0003] However, due to the complex interrelation between environmental parameters, equipment states, material properties and process variables in the manufacturing process, the existing analysis methods based on isolated or simple linear correlation face significant difficulties in identifying real causal relationships, which leads to the misjudgment of variables that are statistically related but have no actual causal relationship as key influencing factors in root cause analysis, thereby misleading the quality improvement direction, causing resource waste and yield improvement bottlenecks. SUMMARY

[0004] The present application provides a LCD display screen production quality collaborative management method based on detection data chain to solve the technical problems in the prior art.

[0005] The technical solution of the present application to solve the above technical problems is as follows: The LCD display screen production quality collaborative management method based on detection data chain comprises: S1, acquiring a detection data chain covering array, cell and module processes in the manufacturing process of the LCD display screen; S2, performing correlation analysis on the multi-source data in the detection data chain to obtain a variable pair set related to the statistics of the final product defects; S3, analyzing whether there is an indirect, mediated influence path through at least one third variable between each variable pair in the variable pair set, and identifying and excluding false correlation variable pairs caused by mediation effects based on the existence of the mediated influence path; S4, applying a conditional independence-based causal discovery algorithm to determine the real causal direction for the remaining variable pair set after identifying and excluding the false correlation variable pairs; S5, performing unsupervised clustering analysis on the detection data chain based on multiple characteristic indexes of the final product defects to divide data subsets, and determining whether the real causal direction remains consistent in the data subsets corresponding to different defect characteristic clusters; S6, output the real cause-effect direction determined to be consistent as the final quality root cause analysis result.

[0006] Further, a detection data chain covering array, cell and module processes in the manufacturing process of the liquid crystal display is acquired, including: Acquiring environmental parameter data from a manufacturing execution system, acquiring equipment state data from an equipment monitoring system, and acquiring process variable data from a quality detection system; Associating the environmental parameter data, the equipment state data and the process variable data through product unit identification to form a detection data chain.

[0007] Further, correlation analysis is performed on the multi-source data in the detection data chain to obtain a variable pair set related to the final product defect statistics, including: Calculating the correlation coefficients between the environmental parameter data, the equipment state data and the process variable data and the final product defect statistics respectively; Wherein, the implementation of the final product defect statistics includes: calculating the total good rate of each production batch as the batch good rate; calculating the occurrence frequency of each type of defect on the unit area as the defect density index; and combining the batch good rate and the defect density index to form the final product defect statistics; Screening variable combinations with absolute correlation coefficients greater than a preset correlation threshold; Combining the screened variable combinations with corresponding final product defect statistics to form a variable pair set.

[0008] Further, whether there is an indirect, mediated influence path through at least one third variable between each variable pair in the variable pair set is analyzed, and false correlation variable pairs caused by mediation effects are identified and excluded based on the existence of the mediated influence path, including: Predefining potential mediation relationships between variables based on a process knowledge base; For variable pairs in the variable pair set that do not satisfy the pre-defined potential mediation relationship, searching for candidate third variables in the detection data chain that have statistical correlations with both variables in the variable pair; Determine whether the candidate third variable constitutes a complete mediation or partial mediation path by checking the change in correlation coefficients between the variable pairs before and after the candidate third variable; Identifying variable pairs that form complete mediation paths as false correlation variable pairs caused by mediation effects; Removing all false correlation variable pairs caused by mediation effects from the variable pair set to form a remaining variable pair set.

[0009] Further, the potential mediating relationships between variables are predefined based on the process knowledge base, including: extracting known physical and chemical mechanisms of action between variables from the process knowledge base; identifying pairs of variables having direct process correlations based on the mechanisms of action; and recording the pairs of variables having direct process correlations as potential mediating relationships.

[0010] Further, the determination of whether the candidate third variable constitutes a complete mediating or partial mediating path by checking the change in the correlation coefficient between the pairs of variables before and after the control of the candidate third variable includes: calculating the correlation coefficient between the pairs of variables before the control of the candidate third variable; calculating the partial correlation coefficient between the pairs of variables after the control of the candidate third variable; comparing the numerical difference between the correlation coefficient and the partial correlation coefficient; determining a complete mediating path when the partial correlation coefficient tends to zero; and determining a partial mediating path when the partial correlation coefficient significantly decreases but is still significantly greater than zero.

[0011] Further, the application of a causal discovery algorithm based on conditional independence to the remaining set of pairs of variables after the identification and exclusion of false correlation pairs to determine real causal directions includes: constructing an initial causal structure diagram of the causal discovery algorithm based on the time sequence constraints of the manufacturing process; performing a conditional independence test on the remaining set of pairs of variables, and orienting the edges of the initial causal structure diagram based on the test results; verifying the consistency of the orientation results in combination with the physical interaction relationships between variables recorded in the process knowledge base; outputting the verified causal directions as real causal directions.

[0012] Further, unsupervised clustering analysis is performed on the detection data chain based on multiple characteristic indicators of the final product defects to divide data subsets, and it is determined whether the real causal directions remain consistent in the data subsets corresponding to different defect characteristic clusters, including: extracting multiple characteristic indicators of the final product defects to form a defect characteristic vector; performing unsupervised clustering analysis on the defect characteristic vector to form multiple defect characteristic clusters; dividing data belonging to the same defect characteristic cluster in the detection data chain into the same data subset; verifying whether the real causal directions determined from the remaining set of pairs of variables remain the same causal directions in each data subset; recording the real causal directions that remain the same causal directions in all data subsets as the real causal directions that are determined to remain consistent.

[0013] Further, verifying whether the true causal directions determined from the set of remaining variable pairs remain the same causal directions on each data subset respectively, comprising: reapplying the conditional independence-based causal discovery algorithm on each data subset; comparing whether the causal directions discovered on different data subsets are consistent; recording the causal directions pointing to the consistency in all data subsets.

[0014] Further, outputting the true causal directions determined to remain consistent as the final quality root cause analysis result, comprising: associating the true causal directions determined to remain consistent with the position information of the involved process variables in the manufacturing process; generating a quality root cause analysis report identifying the key control process and process parameters according to the association result; outputting the quality root cause analysis report to a quality management terminal for guiding process improvement.

[0015] The beneficial effects of the present application are: 1. By constructing a detection data chain covering multiple processes and implementing a multi-stage data analysis process, the accuracy and reliability of quality root cause identification in the manufacturing process of liquid crystal display screens are effectively improved. First, by integrating the detection data of array, cell and module processes to form a complete data chain, the limitations of data isolation in traditional analysis are overcome. Then, by correlation analysis to screen potential associated variable pairs and introducing an intermediary effect test mechanism to eliminate false associations, subsequent analysis is focused on real causal chains. Finally, combining conditional independence-based causal discovery and unsupervised clustering verification, the causal directions that are stable under different defect modes are systematically identified, not only solving the misjudgment problem caused by complex correlation of multiple data sources, but also significantly reducing the risk of misleading process improvement due to noise variables, thereby providing more targeted decision support for quality management personnel.

[0016] 2. By optimizing the quality management process in a data-driven manner, the intermediary effect exclusion step effectively identifies and filters pseudo-correlation relationships indirectly affected by the third variable, avoiding the common mistake of misjudging statistical correlation as causal correlation in traditional methods. The conditional independence-based causal discovery algorithm combined with process knowledge verification ensures the physical reasonableness of the causal directions. Unsupervised clustering analysis verifies the universality of causal relationships through defect feature subdivision, preventing local data bias from affecting the overall conclusion. The final output of the quality root cause analysis result directly associates key processes and process parameters, enabling process improvement measures to accurately focus on core factors that truly affect product quality, improving resource utilization efficiency and yield optimization effect. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 Flowchart of the present application based on detection data chain LCD display screen production quality collaborative control method; Figure 2 This is a flowchart for determining whether the true causal direction of the present invention remains consistent. Detailed Implementation

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

[0019] Example: Figure 1 The present invention provides a collaborative quality control method for LCD display production based on a detection data link, comprising: S1. Obtain the detection data chain covering the array, cell assembly and module processes in the manufacturing process of the LCD display; S2. Perform correlation analysis on the multi-source data in the detection data chain to obtain a set of variable pairs related to the final product defect statistics; S3. Analyze the set of variable pairs to see if there is an indirect mediating path between each variable pair through at least one third variable, and identify and exclude spurious related variable pairs caused by the mediating effect based on the existence of the mediating path. S4. For the set of remaining variable pairs after identifying and eliminating spurious association pairs, apply a causal discovery algorithm based on conditional independence to determine the true causal direction. S5. Based on multiple characteristic indicators of final product defects, perform unsupervised clustering analysis on the detection data chain to divide the data into subsets and determine whether the true causal direction is consistent in the data subsets corresponding to different defect feature clusters. S6. The output is determined to be the consistent true causal direction as the final quality root cause analysis result.

[0020] S1. Obtain the detection data chain of the array, cell assembly and module processes in the manufacturing process of the covered liquid crystal display screen, specifically implemented as follows: In practice, acquiring the detection data chain covering the array, cell assembly, and module processes in LCD manufacturing begins with obtaining environmental parameter data from the Manufacturing Execution System (MES). The MES stores historical and real-time environmental parameter data corresponding to each product unit's production batch. These environmental parameters include, for example, the internal temperature data of the exposure machine in the array process, the humidity data of the cleanroom in the cell assembly process, and the bonding pressure data in the module process. Environmental parameter data is acquired through the application programming interface (API) provided by the MES, extracting data in batches according to the product unit identifier and timestamp range. The product unit identifier is a unique string code that identifies an individual LCD panel throughout the manufacturing process, and the timestamp range is determined based on the actual processing time of each product unit in each process.

[0021] Equipment status data is acquired from the equipment monitoring system through its built-in data acquisition module. This data includes, for example, vacuum level data for the coating machine in the array process, nozzle movement speed data for the alignment film coating machine in the cassette process, and pressure head flatness data for the bonding machine in the module process. The equipment status data is collected in real-time by sensors installed on the corresponding equipment and transmitted to the real-time database of the equipment monitoring system via industrial Ethernet. Each data point is accompanied by a timestamp accurate to milliseconds and a corresponding equipment number. The equipment number and product unit identifier are mapped through a work order tracking table. When establishing this mapping, it is essential to ensure that the transmission time interval between adjacent processes for the same product unit meets the process standard requirements.

[0022] Obtaining process variable data from the quality inspection system involves extracting measurement values ​​recorded by online inspection stations at each stage of the process. Process variable data includes, for example, film thickness uniformity measurements in the array process, cell thickness gap measurements in the cell assembly process, and anisotropic conductive adhesive coating width measurements in the module process. The quality inspection system uses automated testing equipment to perform measurements immediately after each product unit completes a specific process. The measurement results, along with the product unit identifier, are stored in the quality inspection system's structured query language database. Data acquisition is achieved by executing standard query statements to extract data from the measurement result table in product unit identifier order, automatically filtering out obvious erroneous data caused by sensor malfunctions during extraction.

[0023] Linking environmental parameter data, equipment status data, and process variable data through product unit identifiers is a key step in establishing a unified data view. The linking operation is performed on the data processing server using a multi-table join query. Specifically, the product unit identifier serves as the unique join key, connecting the environmental parameter data table from the manufacturing execution system, the equipment status data table from the equipment monitoring system, and the process variable data table from the quality inspection system. Time alignment is also considered during the linking process to ensure the continuity and comparability of data for the same product unit across different processes. For example, the time difference between the end time of the array process and the start time of the cell assembly process must be within the allowable range of the process standard; this time difference range is set to 10 to 30 minutes depending on the type of LCD screen.

[0024] The final step in forming the inspection data chain is to integrate and serialize the correlated data. The integration process includes data cleaning to remove outliers that clearly exceed the process specifications. For example, if the internal temperature data of the exposure machine in the array process exceeds the set temperature threshold range, it is marked as invalid data. The set temperature threshold range is determined according to the process specifications as a reference temperature ±5 degrees Celsius. Serialization processing follows the actual flow sequence of the product unit in the manufacturing process—that is, the chronological order from the array process to the cell assembly process and then to the module process—organizing all environmental parameter data, equipment status data, and process variable data corresponding to each product unit into a complete data record chain. The final inspection data chain treats each product unit as an independent data object, containing all inspection data items for that product unit in the three major manufacturing processes. The data storage format is a set of key-value pairs with timestamps, where the key is the inspection item name, and the value is the corresponding measurement value and unit, uniformly using the International System of Units (SI).

[0025] S2. Perform correlation analysis on the multi-source data in the detection data chain to obtain a set of variable pairs related to the final product defect statistics. Specifically, this is implemented as follows: In practice, final product defect statistics are achieved by calculating the overall yield rate of each production batch as the batch yield rate. The batch yield rate is calculated based on the ratio of the number of good products in each production batch recorded by the quality inspection system to the total production quantity. For example, the number of good products divided by the total production quantity and then multiplied by 100 yields the percentage value. Simultaneously, the frequency of occurrence of various defect types per unit area is calculated as a defect density index. The defect density index is calculated by dividing the frequency of each defect type by the total area of ​​the LCD panel, with the unit being, for example, the number of defects per square meter. The total area of ​​the LCD panel is determined according to product specifications, for example, 0.5 square meters or 1.0 square meters. Finally, the batch yield rate and defect density index are combined to form the final product defect statistics. The combination method uses the batch yield rate and the defect density index of each defect type as different dimensions of a multi-dimensional vector. Each production batch corresponds to a final product defect statistics vector, and the data in each dimension of the vector is standardized to eliminate the influence of dimensions.

[0026] The correlation coefficients between environmental parameter data, equipment status data, and process variable data and the final product defect statistics are calculated using statistical analysis methods. The correlation coefficient is calculated using the Pearson correlation coefficient formula, which measures the degree of linear correlation between two variables. Input data includes environmental parameter data, equipment status data, process variable data, and the final product defect statistics vector extracted from the inspection data chain. Data preprocessing is required before calculation, including aligning time series data to ensure each data point corresponds to the same production batch and time period, handling missing values ​​through interpolation or removal methods, and standardizing the data to ensure all variables have zero mean and unit variance. For example, for the exposure machine internal temperature data in the environmental parameter data, the correlation coefficient is calculated with the batch yield dimension in the final product defect statistics vector. The calculation involves the ratio of covariance to standard deviation, resulting in a correlation coefficient value ranging from -1 to +1.

[0027] The selection of variable combinations with absolute correlation coefficients greater than a preset correlation threshold is based on calculated correlation coefficients. The preset correlation threshold is set according to the correlation distribution between variables and defect statistics in historical production data. For example, by analyzing the percentile distribution of all correlation coefficients in the past year's production data, the preset correlation threshold is set to the 75th percentile, representing a moderate to high correlation strength. The selection process involves iterating through all calculated correlation coefficients, comparing the absolute value of each correlation coefficient with the preset correlation threshold, and retaining combinations of variables with corresponding defect statistics dimensions that have an absolute correlation coefficient greater than the preset threshold as candidate variable combinations. For example, if the absolute correlation coefficient between the internal temperature data of the exposure machine and the batch yield rate is 0.6, which is greater than the preset correlation threshold of 0.5, then this variable combination is retained. Meanwhile, variable combinations with absolute correlation coefficients less than or equal to the preset correlation threshold are excluded.

[0028] The selection of variable combinations and their corresponding final product defect statistics are combined to form a variable pair set through data integration. Each variable pair in the set includes a variable identifier from environmental parameter data, equipment status data, or process variable data, and a dimension indicator identifier from the final product defect statistics. The process involves storing all selected variable combinations in a structured format, such as using a database table or list structure. Each record contains a variable name, a defect statistics dimension name, and the corresponding correlation coefficient value. For example, a record in the variable pair set might contain exposure machine internal temperature data as a variable and batch yield rate as a defect statistics dimension, forming a correlation between the two. The correlation coefficient between them is recorded for subsequent analysis. The final variable pair set is used as output for subsequent process analysis steps.

[0029] S3. Analyze the set of variable pairs to determine if there is an indirect mediating path between each variable pair through at least one third variable. Based on the existence of the mediating path, identify and exclude spurious related variable pairs caused by the mediating effect. Specifically, this is implemented as follows: In practical implementation, the potential mediating relationships between predefined variables based on the process knowledge base are achieved through the systematic extraction of causal relationship knowledge from the manufacturing process. The process knowledge base stores verified process principles in the LCD manufacturing field, derived from historical production records, process experimental data, and expert experience summaries. Known physical and chemical interaction mechanisms between variables are extracted from the process knowledge base. Specifically, this extraction process involves querying the process principle rule table stored in the knowledge base, which includes fields such as process name, variable type, and mechanism description. For example, extracting the chemical corrosion mechanism between cleaning solution concentration and metal line etching rate in the array process; this mechanism describes how changes in cleaning solution concentration affect the metal line etching rate through chemical reactions. Variable pairs with direct process associations are identified based on the interaction mechanism. This identification process involves parsing the interaction mechanism description text to identify combinations of variables with direct causal relationships. For example, identifying a direct chemical interaction between the cleaning solution concentration variable and the metal line etching rate variable; the strength of the association is determined based on the degree of interaction recorded in the process knowledge base. Variable pairs with direct process correlations are recorded as potential mediation relationships. The recording is done in a structured storage manner, including variable pair identification, mechanism type, direction of action, and intensity coefficient, forming a set of potential mediation relationships for subsequent analysis.

[0030] For variable pairs in the variable pair set that do not satisfy a predefined potential mediation relationship, a search is conducted in the detection data chain for candidate third variables that are statistically correlated with both variables in the variable pair. The search process first filters variable pairs not included in the potential mediation relationship set from the variable pair set. The filtering criteria are implemented by comparing the variable pair identifiers with records in the potential mediation relationship set. Then, all process variables in the detection data chain are traversed, and the correlation coefficient between each process variable and the two variables in the current variable pair is calculated. The correlation coefficient calculation uses Pearson correlation analysis, with the input data being time series data of the process variables and the variables in the variable pair. The filtering criteria for candidate third variables require that the absolute value of the correlation coefficient between the variable and the two variables in the variable pair is greater than a preset correlation threshold. The preset correlation threshold is set based on the distribution characteristics of correlations between variables in historical production data. For example, by analyzing the distribution of correlation coefficients of all variable pairs in the past year's production data, the preset correlation threshold is set to the 25th percentile value. The search process considers the temporal order of variables, requiring that the candidate third variable be located between the two variables in the variable pair in terms of process timing. Timing verification is achieved by comparing process timestamps. For example, in the array process, the process time point of the candidate third variable is required to be between the process time points of the two variables in the variable pair.

[0031] By examining the changes in the correlation coefficients between variable pairs before and after controlling for the candidate third variable, it is determined whether the candidate third variable constitutes a complete or partial mediation path. The correlation coefficient between variable pairs before controlling for the candidate third variable is calculated using the Pearson correlation coefficient method. The input data consists of the original observation sequences of the two variables in the variable pair. Data preprocessing includes outlier removal and standardization. The partial correlation coefficient between variable pairs after controlling for the candidate third variable is calculated using statistical control methods. Under the condition of keeping the candidate third variable's values ​​fixed, the conditional correlation between the two variables in the variable pair is calculated. The partial correlation coefficient calculation is based on residual analysis of a linear regression model. A mediation effect judgment threshold is set when comparing the numerical differences between the correlation coefficient and the partial correlation coefficient. The mediation effect judgment threshold is determined comprehensively based on statistical significance theory and practical process requirements. For example, when the ratio of the partial correlation coefficient to the correlation coefficient is less than 0.1, it is judged as a complete mediation path; when the ratio is between 0.1 and 0.7, it is judged as a partial mediation path. The judgment process also considers the statistical significance level, requiring the change in the partial correlation coefficient to pass a significance test. The significance level is set at 0.05 based on the conventional requirements of statistical testing.

[0032] Identifying variable pairs forming complete mediation paths as spurious associations due to mediation effects is based on mediation path analysis results. The identification process first filters out all variable pairs identified as complete mediation paths by comparing the mediation path identification results with preset complete mediation path criteria. Then, it verifies whether a direct process interaction mechanism exists between the variable pairs. Verification methods include querying the process knowledge base to see if a direct interaction relationship between the variable pairs is recorded; this query is achieved by searching for variable pair identifiers in the process knowledge base. It also involves checking whether historical process data supports a direct causal relationship between the variable pairs; this check is achieved by analyzing the correlation of time-series data between the variable pairs. Variable pairs that form complete mediation paths but lack a direct interaction mechanism in the process knowledge base are identified as spurious associations due to mediation effects. For example, when exposure energy and liquid crystal molecule alignment angle form a complete mediation path through alignment film thickness, and the process knowledge base does not record a direct interaction mechanism between exposure energy and liquid crystal molecule alignment angle, this variable pair is identified as a spurious association.

[0033] The removal of all spurious association pairs identified as being caused by mediation effects from the original set of variable pairs, forming the remaining set of variable pairs, is accomplished through data filtering. The removal operation first establishes a list of identifiers for spurious association pairs by collecting all such identifiers. Then, it iterates through each variable pair record in the original set. For each record, it checks if it exists in the list of spurious association pair identifiers using a string matching algorithm. If it does, the record is removed from the set. The remaining variable pairs after the removal operation constitute the remaining set, where each pair satisfies the condition of either no complete mediation path or a direct process mechanism. This final set of remaining variable pairs serves as the filtered variable association result after mediation effect purification and is used in subsequent causal analysis steps.

[0034] S4. For the set of remaining variable pairs after identifying and eliminating spurious association pairs, apply a causal discovery algorithm based on conditional independence to determine the true causal direction. Specifically, the implementation is as follows: In practical implementation, the initial causal structure graph for the causal discovery algorithm, based on the time sequence constraints of the manufacturing process, is achieved through a systematic analysis of the process sequence in the LCD manufacturing flow. The time sequence constraints of the manufacturing process originate from the inherent order of the production process: the array process precedes the cell assembly process, and the cell assembly process precedes the module assembly process. Each process contains specific process variables. The construction of the initial causal structure graph first maps all variables in the remaining variable pair set to their corresponding manufacturing processes. This mapping process is completed by querying the association table between process variables and processes, which records the process name and process time point for each process variable. Then, based on the time sequence constraints, possible causal relationships between variables are established. Variables in processes that appear earlier in time can only point to variables in processes that appear later; reverse pointing or cyclic pointing is not allowed. A directed acyclic graph data structure is used during construction, where nodes represent process variables and edges represent possible causal relationships. Initially, the direction of all edges is set to undetermined. For example, the exposure energy variable in the array process precedes the alignment film thickness variable in the cell assembly process; therefore, an edge from the exposure energy variable to the alignment film thickness variable is added to the initial causal structure graph, but the direction is not yet determined.

[0035] Conditional independence tests are performed on the set of residual variable pairs. The edge orientation of the initial causal structure graph is determined based on the test results using statistical hypothesis testing. The conditional independence test employs a chi-square test-based method. The input data is a sequence of historical observations of the variables in the set of residual variable pairs. Data preprocessing includes missing value imputation and outlier removal. The testing process examines whether each pair of adjacent nodes in the initial causal structure graph is conditionally independent, given other relevant variables. The choice of the condition set is based on the correlation strength and process-relatedness between the variables. The significance level threshold for the conditional independence test is set to 0.05, determined by a combination of standard statistical testing criteria and historical data analysis. Based on the test results, if two variables are conditionally independent under the given condition set, their edges are removed; if they are not conditionally independent, the edge direction is determined based on the time sequence constraint and the p-value of the conditional independence test. For example, for the exposure energy variable and the alignment film thickness variable, their independence under a given cleaning solution concentration variable is tested. If the p value is less than 0.05, the edge is retained and the edge direction is changed from the exposure energy variable to the alignment film thickness variable according to the time sequence.

[0036] Verifying the consistency of the orienting results by combining the physical interactions between variables recorded in the process knowledge base is achieved through a systematic comparison of causal directions with process knowledge. The process knowledge base stores known physical interactions between variables, derived from historical process experimental data, engineering principles, and expert experience. The verification process first extracts the physical interactions related to the orienting causal direction from the process knowledge base by querying the causal relationship table, which contains causal variables, outcome variables, and descriptions of the action mechanism. Then, it compares whether the orienting causal direction is consistent with the physical interactions. The consistency criterion requires that the causal variables and outcome variables have a direct interaction relationship in the process knowledge base and that the interaction directions match. If the orienting result is consistent with the process knowledge base record, the causal direction is retained; if inconsistent, the edge direction is adjusted according to the record in the process knowledge base, or the causal direction is marked as pending verification. For example, if the orienting result indicates that the cleaning solution concentration variable points to the metal wire etching rate variable, and the process knowledge base records that the cleaning solution concentration affects the metal wire etching rate through chemical corrosion, then the verification passes.

[0037] Outputting validated causal directions as true causal directions is accomplished through data transformation and formatting operations. The output process first collects all causal directions that have passed consistency validation, including causal variables, outcome variables, and corresponding process information. Then, the causal directions are converted into a structured data format, such as storing them as a list of causal relationships or a matrix. Each record contains a causal variable identifier, an outcome variable identifier, a causal strength value, and a validation status. The causal strength value is calculated based on the statistic of the conditional independence test, for example, using the absolute value of the test statistic as the strength index. The final output of true causal directions serves as the foundational data for quality root cause analysis, used in subsequent defect analysis and process optimization steps.

[0038] Figure 2 A flowchart for determining whether the true causal direction remains consistent is provided. S5: Based on multiple characteristic indicators of the final product defect, unsupervised clustering analysis is performed on the detection data chain to divide the data into subsets, and it is determined whether the true causal direction remains consistent in the data subsets corresponding to different defect feature clusters. The specific implementation is as follows: In practice, extracting various characteristic indicators of final product defects to form a defect feature vector is accomplished through the systematic collection and integration of quality inspection data. These indicators include batch yield, defect density, defect type distribution, and defect spatial distribution characteristics. All indicators originate from the quality data of each production batch recorded by the quality inspection system. The batch yield is calculated by dividing the number of good products in the batch by the total production quantity; defect density is calculated by dividing the total number of defects by the inspection area; defect type distribution is the percentage of each type of defect in the total defects; and defect spatial distribution characteristics are obtained by dividing the LCD panel into multiple regions and calculating the defect density in each region. The defect feature vector is constructed by combining all characteristic indicators in a fixed order into a multi-dimensional vector, with each dimension corresponding to one characteristic indicator. The vector values ​​are standardized to eliminate dimensional differences. The standardization process uses a min-max standardization method, which linearly transforms the original value of each characteristic indicator to between 0 and 1. This transformation involves subtracting the minimum value from the current value and then dividing by the difference between the maximum and minimum values. The minimum and maximum values ​​are determined based on the distribution range of the feature index in historical data. For example, the minimum and maximum values ​​of batch yield in the past year's production data can be used as standardization parameters. For instance, a defect feature vector can include five dimensions: batch yield, bright spot defect density, dark spot defect density, edge area defect density, and center area defect density. The values ​​of each dimension are processed using the standardization method described above.

[0039] Unsupervised clustering analysis of defect feature vectors to form multiple defect feature clusters is achieved through a clustering algorithm. The K-means clustering algorithm is used, which groups similar defect feature vectors into the same cluster based on the Euclidean distance between vectors. The Euclidean distance is calculated as the square root of the sum of squares of the differences in each dimension of two vectors. The number of clusters is determined using the elbow rule, which calculates the sum of squares within each cluster under different numbers of clusters and selects the inflection point where the sum of squares within each cluster decreases with increasing cluster size as the optimal number of clusters. The sum of squares within each cluster is calculated as the sum of the squares of the distances from all vectors in each cluster to the cluster center. The inflection point is identified by visualizing the curve of the sum of squares within each cluster changing with the number of clusters, selecting the number of clusters corresponding to the point with the greatest curvature of the curve. During the clustering process, the maximum number of iterations is set to 300, and the convergence threshold is 0.001. All parameters are determined based on the results of historical clustering experiments; for example, specific parameter values ​​are set after observing the iterative convergence through multiple clustering experiments. Before clustering analysis, principal component analysis is performed on the defect feature vectors to reduce dimensionality, retaining principal components with a cumulative contribution rate of 85%. The cumulative contribution rate is calculated as the proportion of the variance of the top k principal components to the total variance. For example, by clustering the defect feature vectors of 1000 production batches, three defect feature clusters may be formed, representing a high yield rate and low defect density cluster, a medium yield rate and medium defect density cluster, and a low yield rate and high defect density cluster.

[0040] Dividing data belonging to the same defect feature cluster in the inspection data chain into the same data subset is accomplished through data mapping operations. The partitioning process first establishes a mapping relationship between each production batch and its corresponding defect feature cluster. This mapping relationship is determined by cluster labels obtained through cluster analysis. Each cluster label is a unique identifier assigned to each defect feature vector during the clustering process. Then, based on the mapping relationship, the process data in the inspection data chain is grouped, and the process data of all production batches corresponding to the same defect feature cluster are divided into the same data subset. The data subset contains all process variable data for all production batches in that cluster across the array, box-building, and module processes. The data subset is stored using a cluster-based independent storage approach. For example, all process data from 150 production batches belonging to the high-yield, low-defect-density cluster is divided into data subset A, and all process data from 120 production batches belonging to the low-yield, high-defect-density cluster is divided into data subset B.

[0041] Verifying whether the true causal directions identified from the set of remaining variable pairs maintain the same causal orientation across each data subset is achieved by re-executing the causal discovery process on each subset. The verification process first reapplies the conditional independence-based causal discovery algorithm to each subset, in the same manner as its application on the main dataset. This includes constructing an initial causal structure graph, performing conditional independence tests, and verifying with process knowledge. The initial causal structure graph is constructed based on the time sequence constraints of the manufacturing process. The conditional independence test employs a chi-square test-based method, and the verification with process knowledge involves querying physical interaction relationships in the process knowledge base. Then, the consistency of the causal directions discovered across different data subsets is compared. This comparison involves checking the occurrence and directional consistency of each true causal direction across all data subsets. A consistency threshold is set based on the stability analysis of causal directions in historical verification data. For example, by analyzing the consistency ratio of causal directions across different subsets in past project data, the consistency threshold is set to 80%. A causal direction is considered consistent when it exists in more than 80% of the data subsets and is consistent in direction. True causal directions that maintain the same causal orientation across all data subsets are considered consistent true causal directions. During recording, the support and consistency index of this causal direction in each data subset are simultaneously saved. Support is calculated as the proportion of subsets containing this causal direction out of the total number of subsets. For example, if the causal direction of cleaning fluid concentration pointing to metal line etching rate is verified to exist and be consistent in data subsets A, B, and C, then this causal direction is recorded as a consistent causal direction.

[0042] S6. The output is determined to be the consistent true causal direction as the final quality root cause analysis result. The specific implementation is as follows: In practice, associating the consistent true causal directions with the location information of the involved process variables in the manufacturing process is accomplished through a systematic data mapping operation. The association process first extracts all involved process variables, including causal and consequential variables, from the consistent true causal directions. The consistent true causal directions originate from the set of causal directions validated through multiple data subsets in step S5. Then, the manufacturing process location information database is queried. This database stores the mapping relationship between each process variable and array, cassette, or module processes, as well as the specific step information of the variable within the process. The query operation uses a structured query language to retrieve the corresponding process name and location description based on the process variable identifier. The process variable identifier follows a unified naming convention and consists of a process code, equipment code, and parameter type code. The association operation uses a key-value pair mapping method to bind each true causal direction with its corresponding process location information, forming causal direction data with location tags. For example, in the causal relationship between cleaning solution concentration and metal line etching rate, the cleaning solution concentration variable maps to the cleaning step of the array process, and the metal line etching rate variable maps to the etching step of the array process, thus associating this causal relationship with a specific location within the array process. Location information includes the process name, process step number, and equipment location information, which originates from the process route data recorded in the manufacturing execution system.

[0043] The generation of a quality root cause analysis report, identifying key control processes and process parameters based on the correlation results, is achieved through a report generation algorithm. This algorithm identifies key control processes and key process parameters based on the correlated causal direction data. Key control processes are identified by statistically analyzing the frequency of each process involved in the causal direction, selecting processes with frequencies exceeding a preset key process threshold. This preset threshold is set based on the significance of the process's impact on defects in historical quality data; for example, by analyzing the proportion of processes involved in causal relationships in quality incidents over the past year, the preset threshold is set to 50%. Key process parameters are identified by extracting all process variables associated with key control processes and ranking them based on the strength of the causal direction, selecting the parameters with the highest strength as key process parameters. The strength of the causal direction is calculated based on statistical measures of conditional independence tests, such as using the absolute value of the partial correlation coefficient as the strength indicator. The report content is organized in a structured format, including a list of key control processes, a list of key process parameters, a causal direction diagram, and suggested improvement measures. During report generation, a template engine is used to populate a preset report template with data, which includes chapter titles, data tables, and visualization chart areas. For example, a quality root cause analysis report may identify the array process as a critical control process, the cleaning solution concentration and exposure energy as critical process parameters, and provide specific suggestions for adjusting the cleaning solution concentration and exposure energy. The suggestions are based on optimization schemes stored in the process knowledge base.

[0044] Outputting the root cause analysis report to the quality management terminal to guide process improvements is accomplished through a data output interface. The output process first converts the generated root cause analysis report into a specified format, such as Extensible Markup Language (XML) or Portable Document Format (PDF). Format conversion is achieved using a document processing library, ensuring the report's structural integrity and consistent style. Then, the report data is transmitted to the quality management terminal via an Application Programming Interface (API), a computer system used to monitor and improve manufacturing processes. The API uses Hypertext Transfer Protocol (HTTP), and transmission parameters include the terminal address, port number, and authentication credentials. Transmission protocols and parameters are set during output to ensure data integrity and security, such as encrypting the data stream using Transport Layer Security (TLS). Upon receiving the report, the quality management terminal displays the report content on its user interface and provides interactive functions for process engineers to analyze and implement improvement measures. The user interface includes a report viewer, data filtering tools, and an operation log module. For example, based on the key process parameter adjustment suggestions in the report, process engineers can modify the cleaning fluid concentration setpoint for the array process, thereby optimizing the production process and improving product yield. The effectiveness of the improvement measures is verified by changes in the yield of subsequent production batches, forming a closed-loop quality improvement process.

[0045] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0046] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0047] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. Computer-readable storage media can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0048] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0049] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0050] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0051] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0052] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0054] In conclusion, the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A collaborative quality control method for LCD display production based on a detection data link, characterized in that: include: S1. Obtain the detection data chain covering the array, cell assembly and module processes in the manufacturing process of the LCD display; S2. Perform correlation analysis on the multi-source data in the detection data chain to obtain a set of variable pairs related to the final product defect statistics; S3. Analyze the set of variable pairs to see if there is an indirect mediating path between each variable pair through at least one third variable, and identify and exclude spurious related variable pairs caused by the mediating effect based on the existence of the mediating path. S4. For the set of remaining variable pairs after identifying and eliminating spurious association pairs, apply a causal discovery algorithm based on conditional independence to determine the true causal direction. S5. Based on multiple characteristic indicators of final product defects, perform unsupervised clustering analysis on the detection data chain to divide the data into subsets and determine whether the true causal direction is consistent in the data subsets corresponding to different defect feature clusters. S6. The output is determined to be the consistent true causal direction as the final quality root cause analysis result.

2. The collaborative quality control method for LCD display production based on a detection data link according to claim 1, characterized in that, Acquire the inspection data chain covering the array, cell assembly, and module processes in the LCD manufacturing process, including: Obtain environmental parameter data from the manufacturing execution system; obtain equipment status data from the equipment monitoring system; obtain process variable data from the quality inspection system; Environmental parameter data, equipment status data, and process variable data are linked through product unit identifiers to form a detection data chain.

3. The collaborative quality control method for LCD display production based on a detection data link according to claim 2, characterized in that, Correlation analysis was performed on multi-source data in the inspection data chain to obtain a set of variable pairs related to the final product defect statistics, including: Calculate the correlation coefficients between environmental parameter data, equipment status data, and process variable data and the final product defect statistics; The implementation of final product defect statistics includes: calculating the total yield rate of each production batch as the batch yield rate; calculating the frequency of occurrence of various defect types per unit area as the defect density index; and combining the batch yield rate and the defect density index to form the final product defect statistics. Select variable combinations whose absolute correlation coefficient is greater than a preset correlation threshold; The selected variable combinations are combined with the corresponding final product defect statistics to form a variable pair set.

4. The collaborative quality control method for LCD display production based on a detection data link according to claim 3, characterized in that, The analysis of the variable pair set examines whether each variable pair has an indirect mediating path through at least one third variable, and identifies and excludes spurious association variable pairs caused by mediation effects based on the existence of such mediating paths, including: Based on the potential mediation relationships between predefined variables in the process knowledge base; For variable pairs in the set that do not satisfy a predefined potential mediation relationship, search the detection data chain for a candidate third variable that is statistically correlated with both variables in the variable pair. By examining the changes in the correlation coefficients between variable pairs before and after controlling for the candidate third variable, it can be determined whether the candidate third variable constitutes a complete or partial mediation path. Variable pairs that form a complete mediation path will be identified as spurious associations caused by the mediation effect. Remove all spurious association pairs identified as being caused by mediation effects from the set of variable pairs, resulting in the set of remaining variable pairs.

5. The collaborative quality control method for LCD display production based on a detection data link according to claim 4, characterized in that, Potential mediation relationships between predefined variables based on the process knowledge base include: extracting known physical and chemical interaction mechanisms between variables from the process knowledge base; identifying variable pairs with direct process associations based on the interaction mechanisms; and recording variable pairs with direct process associations as potential mediation relationships.

6. The collaborative quality control method for LCD display production based on a detection data link according to claim 4, characterized in that, Determining whether a candidate third variable constitutes a complete or partial mediation path by examining the changes in the correlation coefficients between variable pairs before and after controlling for the candidate third variable includes: calculating the correlation coefficients between variable pairs before controlling for the candidate third variable; calculating the partial correlation coefficients between variable pairs after controlling for the candidate third variable; comparing the numerical differences between the correlation coefficients and the partial correlation coefficients; determining a complete mediation path when the partial correlation coefficient approaches zero; and determining a partial mediation path when the partial correlation coefficient significantly decreases but remains significantly greater than zero.

7. The collaborative quality control method for LCD display production based on a detection data link according to claim 4, characterized in that, For the set of remaining variable pairs after identifying and eliminating spurious association pairs, a causal discovery algorithm based on conditional independence is applied to determine the true causal direction, including: The initial causal structure graph of the causal discovery algorithm is constructed based on the time sequence constraints of the manufacturing process; Perform a conditional independence test on the set of remaining variable pairs, and orient the edges of the initial causal structure graph based on the test results; The consistency of the orientation results is verified by combining the physical interaction relationships between variables recorded in the process knowledge base; Output the verified causal direction as the true causal direction.

8. The collaborative quality control method for LCD display production based on a detection data link according to claim 7, characterized in that, Unsupervised clustering analysis is performed on the detection data chain based on multiple characteristic indicators of final product defects to divide the data into subsets and determine whether the true causal direction remains consistent in the data subsets corresponding to different defect feature clusters, including: Multiple characteristic indicators of defects in the final product are extracted to form a defect feature vector; Unsupervised clustering analysis is performed on the defect feature vectors to form multiple defect feature clusters; Data belonging to the same defect feature cluster in the detection data chain are divided into the same data subset; On each subset of data, verify whether the true causal direction determined from the set of remaining variable pairs maintains the same causal orientation; True causal directions that maintain the same causal orientation across all data subsets are considered to be consistent true causal directions.

9. The collaborative quality control method for LCD display production based on a detection data link according to claim 8, characterized in that, Verify on each data subset whether the true causal directions determined from the set of remaining variable pairs maintain the same causal orientation, including: reapplying the causal discovery algorithm based on conditional independence on each data subset; comparing whether the causal directions found on different data subsets are consistent; and recording the causal directions that point consistently across all data subsets.

10. The collaborative quality control method for LCD display production based on a detection data link according to claim 8, characterized in that, The output is determined to maintain a consistent true causal direction as the final quality root cause analysis result, including: The true causal direction that will be determined to be consistent will be correlated with the location information of the process variables involved in the manufacturing process; Generate a quality root cause analysis report based on the correlation results, identifying key control processes and parameters. The root cause analysis report is output to the quality management terminal to guide process improvement.

Images