MES-based methods and related equipment for end-to-end traceability and root cause analysis of inkjet printing quality in PCBs.
By establishing end-to-end data association in PCB inkjet production through the MES system, and using data-driven analysis methods to quickly locate the root cause of inkjet defects, the problem of difficulty in quickly locating the cause of inkjet defects in existing technologies has been solved, thereby improving production efficiency and quality control.
Patent Information
- Application Number
- CN202611627952.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-21
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, it is difficult to quickly locate the cause of inkjet printing defects during PCB inkjet production, resulting in a cumbersome and inaccurate troubleshooting process. This makes it impossible to achieve end-to-end data correlation, affecting production efficiency and quality control.
By using MES-based end-to-end traceability and quality root cause analysis, the end-to-end data correlation of production-related data is established. By utilizing the correlation between defect information and production-related data, data-driven analysis methods are used to identify key influencing factors and root causes, generate root cause analysis results, and provide risk warnings or optimization measures.
It enables data integration in the PCB inkjet production process, shortens the troubleshooting cycle, improves the accuracy and efficiency of root cause identification, and reduces rework costs.
Smart Images

Figure CN123116199A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of digital inkjet and intelligent manufacturing technology, and in particular to a method and related apparatus for full-chain traceability and quality root cause analysis of PCB inkjet printing based on MES. Background Technology
[0002] PCB manufacturing typically utilizes Manufacturing Execution Systems (MES) to manage production batches and processes, and to record some production data and inspection results. For PCBs experiencing quality issues, corresponding production records can be retrieved based on batch information to achieve basic production traceability. However, in actual production, data such as product information, batch information, production equipment information, process parameters, production environment, production materials, and inspection results are usually generated and stored separately by different equipment or production stages. The lack of a unified correlation between these data makes it difficult to establish a complete end-to-end link between production data from different sources.
[0003] When inkjet printing defects occur on PCBs, existing traceability methods typically focus on querying production records based on batch information. This makes it difficult to correlate the detected defect information with corresponding equipment status, process parameters, environmental conditions, and material conditions. Therefore, determining the cause of the defect often requires engineers to systematically check multiple production factors based on their production experience. This process is not only cumbersome but also makes it difficult to quickly identify the primary factor influencing the defect when multiple factors are at play.
[0004] Therefore, there is a need for a method and related equipment for end-to-end traceability and root cause analysis of PCB inkjet printing based on MES. Summary of the Invention
[0005] To address the aforementioned technical issues, this application provides a method and related apparatus for PCB inkjet end-to-end traceability and root cause analysis based on MES, enabling end-to-end data binding, intelligent defect identification, and reduced rework costs.
[0006] The technical solution provided in this application is described below: Firstly, this application provides a method for PCB inkjet end-to-end traceability and quality root cause analysis based on MES, including: Acquire production-related data during the PCB manufacturing process, associate the production-related data based on batch identifiers, and establish a full-chain data association relationship; Obtain the detection data after inkjet printing on the PCB, and determine the defect information in the PCB inkjet printing process based on the detection data; Based on the aforementioned end-to-end data association relationship, the defect information is associated with the corresponding production-related data to obtain defect-related data; Based on the adverse correlation data, the production correlation data is analyzed to obtain at least one key influencing factor corresponding to the adverse information, and the root cause of the adverse is determined based on the key influencing factor. Based on the aforementioned root causes of problems, root cause analysis results are generated, and risk warnings or optimization measures are implemented for the PCB inkjet production process according to the root cause analysis results.
[0007] Optionally, the analysis of the production-related data based on the adverse correlation data includes: Establish a correlation model between defect types, process parameters, environmental conditions, and material properties; The process parameters, environmental conditions, and material properties in the production-related data are mapped to the association model to generate feature data corresponding to the defect information. Based on the aforementioned adverse information, adverse type labels are extracted; continuous features in the feature data are discretized, and a transaction database is constructed based on the discretized feature data and the corresponding adverse type labels; The transaction database is processed using a preset algorithm to obtain association rules corresponding to the defect type labels, and the key influencing factors are determined based on the association rules.
[0008] Optionally, the method further includes: The feature data is input into a pre-trained learning model to obtain the feature contribution value corresponding to the defect type label; The features are sorted in descending order according to the absolute value of their contribution values to obtain the root cause ranking list. Features that meet the first preset ranking criteria in the root cause ranking list are set as root cause candidate features.
[0009] Optionally, the feature data consists of a process parameter subspace, an environmental condition subspace, and a material property subspace; the determination of the root cause of the defect based on the key influencing factors includes: The corresponding root cause process source is determined based on the feature subspace to which the root cause candidate features belong. Calculate the root cause confidence score based on the feature contribution value corresponding to the root cause candidate features; Based on the root cause confidence level, determine whether the root cause candidate features meet the root cause reliability condition; When the root cause confidence level meets the root cause reliability condition, the root cause candidate features and the corresponding root cause process source are determined as the bad root cause. When the root cause confidence level does not meet the root cause reliability condition, a manual review process is triggered.
[0010] Optionally, the method further includes: The features in the association rules are matched with the root cause candidate features in the root cause ranking list to obtain the feature matching results; Based on the feature matching results, determine whether there is an intersection between the features included in the association rule and the sub-features in the root cause ranking list that satisfy the second preset ranking condition; If the intersection is determined to exist, the lift degree corresponding to the association rule is obtained, and it is determined whether the lift degree meets the preset lift degree condition. If the lift degree meets the preset lift degree condition, the association rule is determined as a reliable association rule corresponding to the bad information; Based on the reliable association rules, the root cause candidate features, and the adverse root causes, the final root cause analysis results are generated.
[0011] Optionally, the risk warning or optimization process includes: Real-time acquisition of key production parameters during PCB inkjet production; comparison of the key production parameters with the corresponding target parameter ranges; When the key production parameters deviate from the target parameter range, a risk warning message is output. Historical defect patterns are determined based on historical defect information and corresponding production-related data. Based on the historical defect patterns, the current PCB inkjet production process is risk-identified, and when a risk feature corresponding to the historical defect pattern is detected, a prevention prompt message is output.
[0012] Optionally, the method further includes: Perform forward or reverse tracing based on the aforementioned end-to-end data relationships; The forward traceability includes obtaining the corresponding batch information based on the order information, and obtaining at least one of the corresponding production parameters, production equipment, production materials and test results based on the batch information; The reverse tracing includes obtaining the corresponding batch information based on the defective information, and obtaining at least one of the corresponding production parameters, production equipment, production materials, and production environment based on the batch information.
[0013] Secondly, this application provides a MES-based PCB inkjet end-to-end traceability and quality root cause analysis device, comprising: The data association module is used to acquire production-related data during the PCB production process, associate the production-related data based on batch identifiers, and establish a full-link data association relationship; The defect determination module is used to acquire the inspection data after inkjet printing on the PCB and determine the defect information during the inkjet printing process based on the inspection data. The defect association module is used to associate the defect information with the corresponding production association data based on the full-link data association relationship to obtain defect association data; The root cause analysis module is used to analyze the production-related data based on the adverse correlation data, obtain at least one key influencing factor corresponding to the adverse information, and determine the root cause of the adverse based on the key influencing factor. The results output module is used to generate root cause analysis results based on the root causes of the defects, and to provide risk warnings or optimizations to the PCB inkjet production process based on the root cause analysis results.
[0014] Thirdly, this application provides a PCB inkjet end-to-end traceability and quality root cause analysis device based on MES. The device includes: a processor, a memory, an input / output unit, and a bus; the processor is connected to the memory, the input / output unit, and the bus; the memory stores a program, and the processor calls the program to execute the method of the first aspect and any one of the optional methods in the first aspect.
[0015] Fourthly, this application provides a computer-readable storage medium storing a program that, when executed on a computer, performs the methods of the first aspect and any one of the first aspects.
[0016] As can be seen from the above technical solutions, this application has the following beneficial effects: This application establishes a full-chain data association by linking production-related data in the PCB manufacturing process based on batch identifiers. This unifies previously scattered batch numbers, parameters, equipment, materials, and testing data into a unified whole, solving the problem of fragmented data that cannot be linked and queried in traditional solutions. It achieves seamless data flow from order to test results. By linking defect information with production-related data to obtain defect-related data, and analyzing this data to identify key influencing factors, root cause analysis no longer relies on engineer experience but is based on data-driven automated analysis. This shortens the troubleshooting cycle and improves the accuracy and efficiency of root cause analysis. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a schematic diagram of the PCB inkjet end-to-end traceability and quality root cause analysis method based on MES in this application; Figure 2 This is a schematic diagram illustrating the attribution analysis method for feature data using the association model of this application; Figure 3 This is a schematic diagram illustrating the method used in this application for attribution analysis of feature data using the machine learning model. Figure 4 This is a schematic diagram illustrating the specific method used in this application to determine the root causes of adverse events based on key influencing factors; Figure 5 This is a schematic diagram illustrating the method used in this application to fuse association rules with machine learning attribution results for decision-making. Figure 6 This is a schematic diagram illustrating the specific methods for risk warnings or optimization treatments in this application; Figure 7 This is a schematic diagram illustrating the method of performing forward or reverse tracing based on the end-to-end data association relationship in this application; Figure 8 This is a schematic diagram of the PCB inkjet end-to-end traceability and quality root cause analysis device based on MES in this application; Figure 9 This is another schematic diagram of the PCB inkjet end-to-end traceability and quality root cause analysis device based on MES in this application. Detailed Implementation
[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.
[0020] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.
[0021] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.
[0022] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."
[0023] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0024] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.
[0025] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0026] The method described in this application can be applied to servers, devices, terminals, or other devices with logical processing capabilities; therefore, this application does not limit its application. For ease of description, the following description uses a terminal as the executing entity.
[0027] refer to Figure 1 This embodiment provides a method for PCB inkjet end-to-end traceability and quality root cause analysis based on MES, including: S100: Acquire production-related data during the PCB manufacturing process, associate the production-related data based on batch identifiers, and establish a full-chain data association relationship.
[0028] In this step, the production-related data in the PCB manufacturing process comes from multiple data sources accessed by the MES system. Specifically, on the PCB inkjet production line, the MES system establishes data communication connections with inkjet equipment, AOI inspection equipment, environmental monitoring equipment, material management system, and operator terminals.
[0029] Before production begins, each PCB entering the inkjet process is assigned a unique batch identifier. The batch identifier can be a string generated from a combination of the production order number, product model, production date, and serial number, such as "ORD20250601-PCB-A-0001". When PCBs are produced in panels, a panel code is first assigned to each panel. Then, during the inkjet process, a corresponding Pcs code is assigned to each PCB within the panel, and the Pcs code is bound to the panel code, forming a hierarchical "Panel-PCB" correspondence.
[0030] During the inkjet printing process, the following production-related data are collected in real time and linked to the batch identifier: Product information: Model number, version number, and batch number; Equipment data: Inkjet printer ID, inkjet parameters, nozzle status, ink droplet velocity, and droplet volume; Environmental data: inkjet station temperature and humidity, cleanliness, printhead temperature, UV energy; Material data: ink batch, ink viscosity, ink expiration date; Operator information: Employee ID, shift; Timestamp: Inkjet start time, inkjet end time; AOI inspection results: yield, defect type, defect coordinates, defect image; PCS QR code.
[0031] The above data is stored in association according to batch identifiers, forming a full-link data association relationship with the batch identifier as the core index. This association relationship not only records the data of each PCB in the inkjet process, but can also be associated with data from other processes such as drilling, solder mask, developing, text, and AOI through the batch identifier, thus forming a full-link data link covering multiple processes.
[0032] S200: Acquire inspection data after inkjet printing on the PCB, and determine defect information during the inkjet printing process based on the inspection data.
[0033] In this step, after inkjet printing on the PCB, it enters the AOI (Automated Optical Inspection) process. The AOI inspection equipment performs optical scanning on the surface of the inkjet-printed PCB to acquire inspection data. The inspection data includes at least: inspection image, defect type, defect coordinates, defect area, and inspection time.
[0034] The AOI inspection equipment uploads the inspection data to the MES system. Based on the inspection data, defects in the PCB inkjet printing process are determined. These defects are structured descriptions of the inkjet quality issues reflected in the inspection data, and include at least: defect type, defect location, and defect severity.
[0035] In one implementation, defect types can be classified according to the manifestation of inkjet defects, including but not limited to: Off-center printing: There is a systematic offset between the actual inkjet pattern and the target position; Missing print: The target area is not covered with ink; Ghosting: The same image appears with two or more imprints; Blurry characters: The edges of the characters are not sharp enough, the strokes are stuck together, or the color difference exceeds the standard.
[0036] In one implementation, when multiple Pcs in the same Panel are found to be defective, the number of defective Pcs and the distribution of defect types in the Panel can be counted as input for subsequent judgment on whether it is a batch anomaly.
[0037] S300: Based on the full-link data association relationship, the defect information is associated with the corresponding production-related data to obtain defect-related data.
[0038] In this step, the batch identifier corresponding to the defective information is used as an index to search for the production-related data corresponding to the batch identifier in the full-link data association relationship, and the defective information is associated with the found production-related data to form defective-related data.
[0039] Specifically, based on the batch identifier in the adverse information, the following data corresponding to that batch is extracted from the end-to-end data association relationship: Equipment data: Inkjet printer ID, inkjet parameters, nozzle status, droplet velocity, droplet volume; Environmental data: workshop temperature and humidity, printhead temperature, ink temperature, ink viscosity, UV energy, cleanliness; Material data: ink batch, ink viscosity, and expiration date; Operator information, timestamp, and AOI detection results.
[0040] The defect information and the aforementioned production-related data are aggregated according to batch identifiers to form a structured defect-related data record. This record contains two types of information: "What is the defect?" and "What were the production conditions when the defect occurred?"
[0041] In one implementation, defective correlation data can also be aggregated according to time windows. For example, taking the inkjet printing time of defective Pcs as the center, all production correlation data corresponding to the same inkjet equipment, the same batch of ink, and the same operator within that time window can be extracted so that parameter fluctuations in the time dimension can be considered in subsequent analysis.
[0042] S400: Analyze production-related data based on adverse correlation data, obtain at least one key influencing factor corresponding to adverse information, and determine the root cause of adverse based on the key influencing factor.
[0043] In this step, the adverse correlation data is analyzed to identify at least one key influencing factor with a high degree of correlation with adverse information from a large amount of production-related data, and to determine the root cause of the adverseness.
[0044] Specifically, the production-related data in the defect correlation data are organized according to dimensions such as process parameters, environmental conditions, and material properties, and compared and analyzed with the corresponding defect types. The analysis methods include: comparing the production parameters of defective batches with those of normal batches to identify parameters that deviate from the normal range; statistically analyzing the recurring parameter combinations in defective batches; and analyzing the correlation between different parameters.
[0045] In one implementation approach, a data-driven analysis method can be used to extract parameters highly correlated with the defect type from defect correlation data as key influencing factors. For example, when a batch is found to have "ink bleeding" defects, the parameters such as ink viscosity, workshop temperature, printhead temperature, and inkjet speed corresponding to the "ink bleeding" defective batch are compared with those of normal batches. If it is found that the ink viscosity of the "ink bleeding" defective batch is generally lower than the normal range and the workshop temperature is generally lower, then "ink viscosity" and "workshop temperature" are determined to be the key influencing factors for this defect information.
[0046] In a specific example, after analyzing a batch of defective "ink bleeding," the key influencing factors were identified as follows: ink viscosity "17.5 mPa·s" (below the normal range of 18.0 to 22.0 mPa·s), workshop temperature "26.5℃" (below the normal range of 28.0 to 32.0℃), and printhead temperature "32.0℃" (below the normal range of 35.0 to 40.0℃). Based on these key influencing factors, the root cause of the defect was determined to be: the low ink viscosity led to excessive ink diffusion on the substrate surface, coupled with the low workshop temperature causing reduced ink fluidity; the combined effect of these two factors resulted in the "ink bleeding" defect.
[0047] In one implementation, key influencing factors can be ranked according to their contribution to the type of malfunction, and the ranking results can be output. These ranking results can be used to determine which parameters to prioritize for adjustment when generating a root cause analysis report.
[0048] In one implementation, when key influencing factors involve multiple processes, data can be traced across processes. For example, environmental temperature and humidity, and cleanroom cleanliness levels that affect inkjet printing performance may be passed on to the MES system from other processes or equipment, and this cross-process data can be called upon for comprehensive analysis.
[0049] S500: Generates root cause analysis results based on the root causes of defects, and provides risk warnings or optimizations to the PCB inkjet production process based on the root cause analysis results.
[0050] In this step, root cause analysis results are generated based on the identified root causes of defects, and risk warnings or optimization measures are taken for the PCB inkjet production process based on the root cause analysis results.
[0051] Root cause analysis results should include at least: defect type, key influencing factors, root cause of the defect, source process of the root cause, and optimization recommendations. Root cause analysis results can be output as a report for engineers to review, or as structured data for the MES system to automatically perform subsequent processing.
[0052] Risk warnings can include: real-time acquisition of key production parameters during the PCB inkjet production process; comparison of key production parameters with corresponding target parameter ranges; and output of risk warning information when key production parameters deviate from the target parameter range. For example, when the viscosity of the current batch of ink is detected to be lower than the normal range, a warning message is output: "Ink viscosity 17.5 mPa·s, lower than the target range of 18.0 to 22.0 mPa·s, please check the ink condition."
[0053] Optimization processes may include generating process parameter adjustment recommendations based on root cause analysis results and pushing these recommendations to inkjet equipment or operator terminals. For example, based on the root cause analysis result of "low ink viscosity causing ink bleeding," an optimization recommendation might be generated: "It is recommended to adjust the ink viscosity to 18.0 to 22.0 mPa·s and simultaneously adjust the workshop temperature to 28.0 to 32.0°C," and this recommendation would be pushed to the inkjet equipment control system and operator terminals.
[0054] In one implementation, historical defect patterns can be determined based on historical defect information and corresponding production-related data, and risks in the current PCB inkjet production process can be identified according to these patterns. When a risk characteristic corresponding to a historical defect pattern is detected, a preventative warning message is output. For example, based on historical data, a historical defect pattern is identified as "when the workshop temperature is below 28℃ and the ink viscosity is below 18mPa·s, the ink bleeding defect rate increases significantly." When the current production batch exhibits the same parameter combination, a preventative warning message is output in advance: "The current workshop temperature is 26.5℃ and the ink viscosity is 17.5mPa·s, which matches the historical ink bleeding defect pattern. It is recommended to adjust the parameters in advance."
[0055] In one implementation, maintenance work orders can also be triggered based on root cause analysis results. For example, when the root cause analysis results show that the key influencing factor is "high nozzle blockage rate", a nozzle maintenance work order is automatically generated and pushed to the equipment maintenance personnel, prompting them to clean or replace the corresponding nozzles.
[0056] refer to Figure 2 In one embodiment, the analysis of production-related data based on adverse correlation data specifically includes the following steps: S401. Establish a correlation model between defect types, process parameters, environmental conditions, and material properties.
[0057] In this step, a correlation model is established with defect type as the response variable and process parameters, environmental conditions, and material properties as explanatory variables. The correlation model is used to describe the potential relationships between defect types and various production factors.
[0058] PCB inkjet defects are not caused by a single factor, but are the result of the combined effects of process parameters, environmental conditions, and material properties. By establishing a four-dimensional correlation model, defect types and three types of explanatory variables can be unified into a single analytical framework, providing a structured data foundation for association rule mining.
[0059] Specifically, the data structure of the association model can be represented as: {defect type label, process parameter set, environmental condition set, material attribute set}.
[0060] The set of process parameters includes inkjet speed, inkjet spacing, number of inkjet passes, printhead temperature, nozzle ignition frequency, droplet volume, and UV curing strength. The set of environmental conditions includes workshop temperature, workshop humidity, ink temperature, ink viscosity, and nozzle clogging rate; The set of material properties includes substrate thickness, copper thickness, solder mask type, and incoming material warpage.
[0061] S402. Map the process parameters, environmental conditions and material properties in the production association data to the association model to generate feature data corresponding to the defect information.
[0062] In this step, process parameters, environmental conditions, and material properties are extracted from the adverse correlation data and mapped to the corresponding fields in the correlation model to generate feature data.
[0063] The mapping process includes: reading the process parameter values corresponding to the defect information from the defect association data and filling them into the process parameter set of the association model; reading the environmental condition values and filling them into the environmental condition set; and reading the material attribute values and filling them into the material attribute set. After the mapping is completed, each field in the association model is assigned a specific value, forming the feature data corresponding to the defect information.
[0064] S403. Extracting negative type tags based on negative information.
[0065] In this step, based on the type of defect in the defect information, the corresponding defect type label is extracted. The defect type label is a discrete categorical variable used to identify the defect category corresponding to the feature data.
[0066] Association rule mining algorithms require explicit response variables as mining targets. By extracting bad category labels, each feature data can be labeled with a clear category, providing supervision signals for subsequent association rule mining.
[0067] In one implementation, the defect type label can be defined as: missing print = 0, ink bleeding = 1, off-center print = 2, ghosting = 3, blurry characters = 4, normal = 5. Based on the defect type recorded in the defect information (such as "ink bleeding"), the corresponding label value "1" is extracted.
[0068] S404. Discretize the continuous features in the feature data, and construct a transaction database based on the discretized feature data and the corresponding defect type labels.
[0069] In this step, since association rule mining algorithms typically process discrete data, the continuous features in the feature data are discretized.
[0070] Specifically, the equal-frequency binning method is used to divide each continuous feature into 5 intervals, labeled as extremely low, low, medium, high, and extremely high. For example, the ink viscosity range [5.0, 30.0] mPa·s is divided into 5 intervals: extremely low (5.0~10.0), low (10.0~15.0), medium (15.0~20.0), high (20.0~25.0), and extremely high (25.0~30.0). For an ink viscosity of 17.5 mPa·s, the discretization result is "medium".
[0071] After discretization, the discretized feature data is combined with defect type labels to construct a transaction database. Each transaction in the transaction database corresponds to a defect-related data record, containing all discretized feature values and defect type labels for that record.
[0072] In a specific example, a transaction can be represented as: {Inkjet speed = low, printhead temperature = medium, UV curing strength = medium, workshop temperature = low, workshop humidity = medium, ink viscosity = medium, substrate thickness = medium, copper thickness = medium, solder mask type = LPI, defect type = ink bleeding}.
[0073] S405. Use a preset algorithm to process the transaction database, obtain the association rules corresponding to the defective type labels, and determine the key influencing factors based on the association rules.
[0074] In this step, a preset association rule mining algorithm is used to process the transaction database and mine the association rules between feature items and bad type labels.
[0075] Transaction databases contain numerous combinations of feature terms and defect type labels. It is impossible for humans to quickly identify which feature combinations have strong correlations with defect types from thousands of transactions. Association rule mining algorithms can automatically traverse all feature combinations, calculate their frequency and confidence, and then filter out statistically significant strong association rules.
[0076] The preset algorithm can be the Apriori algorithm, the FP-Growth algorithm, or other association rule mining algorithms. In one embodiment, the FP-Growth algorithm is used to process the transaction database.
[0077] In one implementation, a minimum support threshold is set. =0.03, minimum confidence threshold =0.60, retain the condition that support(A)≥ Frequent itemsets, and generate those that satisfy confidence(A B)≥ =and lift(A) B) Strong association rules with a value ≥ 1.5.
[0078] In a specific example, after performing association rule mining on the transaction database, the following strong association rules were obtained: Rule 1: Ink viscosity = extremely low, therefore workshop temperature = low, therefore ink bleeding (confidence level 0.78, elevation 2.3). Rule 2: Nozzle clogging rate = extremely high, implied printhead temperature = low, implied print leakage (confidence level 0.85, lift 3.1) Rule 3: Inkjet speed = very high, which implies inkjet pitch = low, which implies ghosting (confidence 0.72, lift 1.9). According to Rule 1, "ink viscosity" and "workshop temperature" are identified as the key factors affecting poor ink penetration. According to Rule 2, "nozzle clogging ratio" and "printer temperature" are identified as key factors affecting printing defects. According to Rule 3, "inkjet speed" and "inkjet pitch" are identified as the key factors affecting ghosting defects.
[0079] In one implementation, association rules can also be arranged in descending order of lift, forming an association rule base R={ , , ..., Each rule r_j is represented as a quadruple ( ). , , , ),in As a triggering condition, To predict defect types, For rule confidence, To improve the rule level.
[0080] refer to Figure 3 In one embodiment, analyzing the production-related data based on the adverse correlation data further includes: performing attribution analysis on the feature data using a machine learning model, specifically including the following steps: S501. Establish a correlation model between defect types, process parameters, environmental conditions, and material properties; S502. Map the process parameters, environmental conditions and material properties in the production association data to the association model to generate feature data corresponding to the defect information; S503. Extract the defect type label based on the defect information; S504. Discretize the continuous features in the feature data, and construct a transaction database based on the discretized feature data and the corresponding defect type labels; S505. The transaction database is processed using a preset algorithm to obtain association rules corresponding to the defect type labels, and the key influencing factors are determined based on the association rules.
[0081] S506. Input the feature data into the pre-trained learning model to obtain the feature contribution value corresponding to the bad type label.
[0082] In this step, the feature data generated in step S502 is input into a pre-trained learning model, and the feature contribution value of each feature in the feature data to the prediction result of the bad type label is calculated by the learning model.
[0083] While association rule mining methods can uncover strong associations between feature combinations and defect types from historical data, they are essentially based on frequent itemset statistics using discretized interval labels. This makes them insensitive to specific numerical changes in continuous features and unable to provide the relative contribution of each feature to defect type prediction. Introducing a learning model for attribution analysis allows for the quantification and ranking of feature importance, thus overcoming the shortcomings of association rule mining in quantifying feature contribution.
[0084] In one embodiment, the pre-trained learning model is an XGBoost gradient boosting decision tree multi-class classification model. The training process of the learning model is as follows: collect feature data from historical batches, with each feature data corresponding to a known defect type label, forming a training sample set. Use the 29-dimensional feature vector from the feature data as input and the defect type label as the output target to train the XGBoost multi-class classification model.
[0085] The hyperparameters of the XGBoost multi-class model are configured as follows: n_estimators=200, max_depth=8, learning_rate=0.05, min_child_weight=5, subsample=0.8, colsample_bytree=0.8, objective=multi:softprob, num_class=6, eval_metric=mlogloss.
[0086] In one embodiment, the feature data is preprocessed before being input into the learning model. Preprocessing may include: imputing missing values in the feature data; and normalizing the feature data to scale the values of each feature to the [0,1] range to avoid model training bias due to different feature units.
[0087] S507. Sort each feature in descending order according to the absolute value of its contribution value to obtain the root cause ranking list.
[0088] In this step, after obtaining the feature contribution value of each feature, the features are sorted in descending order according to the absolute value of the feature contribution value to obtain the root cause ranking list.
[0089] Feature contribution values can be positive or negative. Positive values indicate that changes in the feature's value drive the model to predict the current defective type, while negative values indicate that changes in the feature's value inhibit the model from predicting the current defective type. Regardless of whether it is positive or negative, the larger the absolute value, the greater the influence of the feature on the model's prediction result. Using absolute value sorting can capture important features in both positive and negative directions simultaneously.
[0090] In one embodiment, the feature contribution value is the SHAP value. For the XGBoost model, the SHAP value of each feature is calculated using SHAPTreeExplainer.
[0091] In a specific example, after performing SHAP attribution analysis on the characteristic data of a batch of "ink bleeding" defects, the SHAP values of each characteristic were as follows: ink viscosity φ = +0.23, UV curing strength φ = -0.18, workshop temperature φ = -0.12, inkjet speed φ = +0.08, printhead temperature φ = -0.06, workshop humidity φ = +0.04, and the absolute values of the remaining 23 characteristics were all less than 0.04. Arranged in descending order of the absolute values of the SHAP values, the root cause ranking list was: ink viscosity (0.23), UV curing strength (0.18), workshop temperature (0.12), inkjet speed (0.08), printhead temperature (0.06), and workshop humidity (0.04).
[0092] S508. Set the features in the root cause ranking list that meet the first preset ranking condition as root cause candidate features.
[0093] In this step, features that meet the first preset ranking criteria are set as root cause candidate features according to the root cause ranking list.
[0094] The root cause ranking list contains all 29 features. Without filtering, subsequent root cause confidence calculations and process source identification will involve too many features, resulting in scattered and unfocused analysis results. By setting ranking criteria, a few features with the greatest adverse impact can be selected from the 29 features as root cause candidates.
[0095] In one embodiment, the first preset ranking criterion is the top 5. The top 5 features in the root cause ranking list are set as root cause candidate features. For example, if the root cause ranking list is: ink viscosity (0.23), UV curing strength (0.18), workshop temperature (0.12), inkjet speed (0.08), and printhead temperature (0.06), then the above 5 features are set as root cause candidate features.
[0096] In one embodiment, after determining the root cause candidate features, the corresponding root cause process source can be determined based on the feature subspace to which the root cause candidate features belong. The feature subspace includes a process parameter subspace, an environmental condition subspace, and a material attribute subspace. If the root cause candidate feature belongs to the process parameter subspace, it is located to the inkjet printing process or the curing process; if the root cause candidate feature belongs to the environmental condition subspace, it is located to environmental anomalies or equipment aging; if the root cause candidate feature belongs to the material attribute subspace, it is located to the incoming material or warehousing process.
[0097] refer to Figure 4 In one embodiment, the feature data consists of a process parameter subspace, an environmental condition subspace, and a material property subspace; determining the root cause of defects based on key influencing factors specifically includes the following steps: S601. Determine the corresponding root cause process source based on the feature subspace to which the root cause candidate features belong.
[0098] In this step, the subspace to which each root cause candidate feature belongs in the feature space is first determined, and then the root cause candidate feature is mapped to the corresponding root cause process source according to the preset subspace-process mapping rules.
[0099] In PCB inkjet printing, different subspaces correspond to different production stages and responsible departments. The characteristics of the process parameter subspace are directly controlled by inkjet and curing equipment; the characteristics of the environmental condition subspace are determined by the workshop environment and auxiliary equipment; and the characteristics of the material property subspace are determined by incoming material quality and storage conditions. By mapping the characteristic subspaces to specific process sources, the root cause can be automatically located to the specific production stage.
[0100] Specifically, the mapping rules between the feature subspace and the source of the root cause process are as follows: If the root cause candidate features belong to the process parameter subspace If so, then it is located in the inkjet process or the curing process; If the root cause candidate features belong to the environmental condition subspace If so, the cause is determined to be an abnormal environment or aging equipment; If the root cause candidate features belong to the material property subspace Then it is located at the incoming material or warehousing stage.
[0101] In a specific example, the top 5 candidate root cause features in the root cause ranking list are: ink viscosity (φ=+0.23, belonging to...). UV curing strength (φ=-0.18, belonging to) ), workshop temperature (φ=-0.12, belonging to Inkjet speed (φ=+0.08, belonging to) ), nozzle temperature (φ=-0.06, belonging to The above root cause candidate features are statistically analyzed by subspace: those belonging to the environmental condition subspace. There are three characteristics, which belong to the process parameter subspace. It has two characteristics and belongs to the material attribute subspace. There are 0 features. When the root cause candidate features are distributed across multiple feature subspaces, the weighted calculation is performed based on the absolute value of the SHAP marginal contribution value of each feature, and the process source corresponding to the subspace with the highest weight is taken as the localization result. The weight is 0.41. The weight is 0.26. This accounted for 61.2%, indicating that the root cause was "environmental anomaly".
[0102] S602. Calculate the root cause confidence based on the feature contribution values corresponding to the root cause candidate features.
[0103] In this step, the root cause confidence C is calculated based on the feature contribution value corresponding to the root cause candidate features, which is used to measure the overall explanatory power of the root cause candidate features for the prediction results of adverse types.
[0104] If the sum of the contribution values of the root cause candidate features accounts for a high proportion of the total contribution values of all features, it indicates that these features collectively explain the causes of defects, and the root cause localization is reliable; if the proportion is low, it indicates that the causes of defects are relatively scattered, and the reliability of root cause localization is insufficient.
[0105] In one embodiment, the root cause confidence level C is calculated using the following formula: .in, The SHAP value of the k-th ranked candidate feature of the root cause in the root cause ranking list, | | is its absolute value; For the SHAP values of all 29 features, | | is its absolute value.
[0106] In a specific example, the top 3 features in the root cause ranking list have the following SAP values: ink viscosity φ = +0.23, UV curing strength φ = -0.18, and workshop temperature φ = -0.12.
[0107] The sum of the absolute values of the SHAP values for all 29 features is 0.89. Therefore, the root cause confidence C = (|+0.23|+|-0.18|+|-0.12|) / 0.89 = 0.53 / 0.89 ≈ 0.60.
[0108] S603. Determine whether the root cause candidate features meet the root cause reliability conditions based on the root cause confidence level.
[0109] In this step, the calculated root cause confidence C is compared with the preset root cause reliability conditions to determine whether the root cause candidate features meet the root cause reliability conditions.
[0110] In one embodiment, the root cause reliability condition is a root cause confidence level C ≥ 0.6. When C ≥ 0.6, it means that the sum of the contribution values of the top 3 root cause candidate features accounts for more than 60% of the total feature contribution values, indicating that the root cause candidate features collectively explain the main causes of the defect, and the root cause localization is reliable; when C < 0.6, it means that the sum of the contribution values of the root cause candidate features accounts for a low percentage, indicating that the causes of the defect are relatively dispersed, and the root cause localization is not reliable enough.
[0111] In a specific example, the calculated root cause confidence C = 0.60, which satisfies the root cause reliability condition of C ≥ 0.6. Therefore, the root cause candidate feature is determined to satisfy the root cause reliability condition.
[0112] S604. When the root cause confidence level meets the root cause reliability condition, the root cause candidate characteristics and the corresponding root cause process source are determined as bad root causes.
[0113] In this step, different processing procedures are executed based on the judgment result of step S603.
[0114] When the root cause confidence level meets the root cause reliability condition, the root cause candidate features and the corresponding root cause process source are identified as undesirable root causes. For example, when C=0.60 meets the condition C≥0.6, the root cause candidate features "ink viscosity, UV curing strength, workshop temperature" and the corresponding root cause process source "environmental anomaly" are identified as undesirable root causes of the "ink bleeding" defect.
[0115] S605. When the root cause confidence level does not meet the root cause reliability condition, the manual review process is triggered.
[0116] When the root cause confidence level does not meet the root cause reliability condition, a manual review process is triggered. For example, when C=0.48 does not meet the condition of C≥0.6, a manual review work order is generated and pushed to the engineer's terminal. The manual review work order includes at least: defect type, root cause ranking list, root cause confidence level C value, and root cause candidate features. After receiving the work order, the engineer manually analyzes the features in the root cause ranking list to determine whether there are other important factors not included in the root cause candidate features, and gives the final root cause judgment.
[0117] In one embodiment, the manual review process further includes: recording the results of the manual review, and storing the final root cause determined by the manual review along with the corresponding feature data and defect type label into a historical database for incremental training of the subsequent learning model.
[0118] refer to Figure 5In one embodiment, the method further includes fusing association rules with machine learning attribution results to make a decision, specifically including the following steps: S701. Match the features in the association rules with the root cause candidate features in the root cause ranking list to obtain the feature matching results.
[0119] Association rule mining and machine learning attribution are two independent analytical paths. By matching the two, the consistency of the analysis results can be cross-validated, and root cause features that are supported by historical data and contribute significantly to the current batch can be selected, thereby improving the credibility of the root cause analysis results.
[0120] Specifically, iterate through each rule in the association rule base R. Extracting the antecedent of the rule The included feature names, and these feature names are compared with the root cause ranking list. The names of each feature are compared. If a feature in the antecedent of the rule appears in the root cause ranking list, then the feature is recorded as a matching feature, and its ranking in the root cause ranking list is recorded.
[0121] In a specific example, the association rule base contains rules "Ink viscosity = very low ∧ Oven temperature = low, leading to ink bleeding (confidence 0.78, elevation 2.3)". The root cause ranking list is: ink viscosity (rank=1), UV curing strength (rank=2), oven temperature (rank=3), inkjet speed (rank=4), printhead temperature (rank=5). Matching the antecedent features "ink viscosity" and "oven temperature" of rule r_1 with the root cause ranking list yields the following matching results: ink viscosity matched successfully (rank=1), oven temperature matched successfully (rank=3).
[0122] S702. Based on the feature matching results, determine whether there is an intersection between the features included in the association rule and the sub-features in the root cause ranking list that meet the second preset ranking conditions.
[0123] In this step, it is further determined whether there is any intersection between the features contained in the association rule and the sub-features in the root cause ranking list that meet the second preset ranking condition.
[0124] The root cause ranking list contains all 29 features. If the features in the antecedent of the association rule are matched with all 29 features, it is possible that a feature in the antecedent of the association rule happens to be at the end of the root cause ranking list. By setting a second preset ranking condition, sub-features with higher contribution values can be filtered from the root cause ranking list. Only when the features in the antecedent of the association rule intersect with these high-contribution sub-features is the association rule considered to be relevant to the actual situation of the current batch.
[0125] In one embodiment, the second preset ranking condition is the top 5. The top 5 features are extracted from the root cause ranking list as sub-features, and it is determined whether the features in the antecedent of the association rule intersect with this set of sub-features.
[0126] In a specific example, association rules The antecedent features are "ink viscosity" and "room temperature". The top 5 sub-features in the root cause ranking list are: ink viscosity (rank=1), UV curing strength (rank=2), room temperature (rank=3), inkjet speed (rank=4), and printhead temperature (rank=5). The antecedent features of judgment rule r_1 have an intersection with the above set of sub-features, and the intersection elements are "ink viscosity" and "room temperature".
[0127] S703. If an intersection is found, obtain the lift of the association rule and determine whether the lift meets the preset lift condition.
[0128] In this step, if it is determined that the features in the antecedent of the association rule have an intersection with the top 5 sub-features in the root cause ranking list, the lift of the association rule is further obtained, and it is determined whether the lift meets the preset lift condition.
[0129] Intersection judgment only indicates that the features in the antecedents of the association rule overlap with the high-contribution features of the current batch, but it cannot indicate the statistical strength of the association rule itself. By setting a preset lift condition, association rules with insufficient statistical strength can be further filtered out, and only strong association rules with high lift can be retained as reliable association rules.
[0130] In one embodiment, the preset lift condition is a lift greater than or equal to 2.0. The lift of association rules that have intersections is obtained, and it is determined whether they satisfy the condition lift ≥ 2.0.
[0131] In a specific example, association rule r_1, "Ink viscosity = very low ∧ Shop temperature = low, leading to ink bleeding," has a lift of 2.3, satisfying the condition lift ≥ 2.0, and is therefore marked as a reliable association rule. Association rule r_5, "Inkjet speed = high ∧ Shop humidity = medium, leading to offset printing," has a lift of 1.6, which does not satisfy the condition lift ≥ 2.0, and is therefore not marked as a reliable association rule.
[0132] S704. The association rules that satisfy the preset lift condition are determined as the reliable association rules corresponding to the bad information.
[0133] In this step, when step S703 determines that the lift of the association rule meets the preset lift condition, the association rule is determined as a reliable association rule corresponding to the bad information.
[0134] The reason for identifying association rules that meet the lift condition as reliable association rules is that lift is a core indicator for measuring the true strength of the association between the antecedent and consequent of an association rule. When the lift is greater than or equal to a preset threshold, it indicates that the probability of the consequent occurring when the antecedent occurs is significantly higher than the probability of the consequent occurring alone, meaning there is a positive correlation between the antecedent and consequent. By marking association rules that meet the lift condition as reliable association rules, highly reliable rules that are supported by both feature intersection and statistical strength can be selected from the association rule base, providing a reliable rule basis for generating the final root cause analysis results.
[0135] Specifically, the algorithm iterates through all intersecting association rules in the association rule base, obtains the lift of each association rule, and determines whether it meets the preset lift condition. If it does, the association rule is marked as a reliable association rule, and the antecedent features, consequent undesirable type, confidence level, and lift of the rule are recorded.
[0136] In a specific example, association rules The lift of "Ink viscosity = extremely low ∧ workshop temperature = low, leading to ink bleeding" is 2.3, satisfying the condition lift≥2.0. This will change the rule. Marked as a trusted association rule. Association rule The lift of "Inkjet speed = high ∧ workshop humidity = medium" is 1.6, which does not meet the condition lift≥2.0, so it is not marked as a reliable association rule.
[0137] In one embodiment, the threshold in the preset lift condition can be dynamically adjusted according to actual production needs. For example, for PCB products with high quality requirements, the threshold can be set to 2.5 to improve the screening criteria for reliable association rules; for products with relatively relaxed quality requirements, the threshold can be set to 1.8 to include more potential association rules.
[0138] In one embodiment, when multiple reliable association rules satisfy a preset lift condition exist, these rules can be sorted in descending order of lift, and the rule with the highest lift can be selected as the content of the association rule field in the final root cause analysis result. For example, if multiple rules exist simultaneously... (Lifting degree 2.3) and rules If all conditions (lifting degree 2.1) are met, then the rule will be prioritized. As a trusted association rule output.
[0139] This embodiment achieves the final screening of association rules by identifying those that meet preset lift conditions as reliable association rules. Compared to methods that determine reliable association rules solely based on intersection, the lift determination in this embodiment can filter out weak association rules with insufficient statistical strength, ensuring that the association rules included in the final root cause analysis results have sufficient statistical significance, thus improving the reliability of the root cause analysis results.
[0140] S705. Based on the trusted association rules, the root cause candidate features, and the adverse root causes, generate the final root cause analysis results.
[0141] The reason for merging the credible association rules, root cause candidate features, and root causes of defects to generate the final root cause analysis result is that they describe the causes of defects from different perspectives: credible association rules provide strong correlation patterns in historical data as evidence; root cause candidate features provide a quantitative ranking of the contribution values of each feature in the current batch; and root causes of defects provide process source location based on feature subspace mapping. Merging the three outputs ensures that the final root cause analysis result is supported by historical data, validated by current batch data, and clearly indicates the process, facilitating engineers' comprehensive understanding of the causes of defects and the development of targeted optimization measures.
[0142] Specifically, the credible association rules, root cause candidate features, and defect root causes are structured and encapsulated according to a preset report format to generate the final root cause analysis results. The final root cause analysis results include at least the following fields: defect type, Top-3 root cause parameters and their SHAP contribution values, root cause process source, root cause confidence level, credible association rules matching the root cause, and process parameter adjustment suggestions based on the root cause.
[0143] In a specific example, the final root cause analysis results are as follows: Defect type: "Ink bleeding"; Top-3 root cause parameters: "Ink viscosity (φ=+0.23), UV curing strength (φ=-0.18), workshop temperature (φ=-0.12)"; Root cause process source: "Abnormal environment"; Root cause confidence level C=0.60; Confidential association rule: "Ink viscosity = very low ∧ Workshop temperature = low leads to ink bleeding (confidence level 0.78, elevation 2.3)"; Process parameter adjustment suggestion: "It is recommended to adjust the ink viscosity to 18.0 to 22.0 mPa·s, and at the same time adjust the workshop temperature to 28.0 to 32.0℃".
[0144] In one embodiment, the final root cause analysis results can also be pushed to the quality control interface of the MES system for engineers to review and confirm. After confirmation by the engineers, based on the process parameter adjustment suggestions in the root cause analysis results, equipment parameter adjustment instructions are automatically generated and sent to the inkjet equipment control system, realizing a closed loop from root cause analysis to parameter adjustment.
[0145] In one embodiment, when the root cause analysis results contain multiple credible association rules, the credible association rules can be sorted in descending order of lift, and the top-ranked credible association rules can be prioritized for display in the root cause analysis report. For example, if multiple rules exist simultaneously... (Lifting degree 2.3) and rules (Lifting degree 2.1), then the rule Prioritize display.
[0146] In one embodiment, the process parameter adjustment suggestions in the final root cause analysis results can also be converted into specific equipment control instructions. For example, when the root cause analysis result is "low ink viscosity leads to ink bleeding," a temperature adjustment instruction for the ink heating system is generated to adjust the ink temperature from the current value to the target range, and is automatically sent to the ink heating controller to achieve automatic closed-loop regulation.
[0147] This embodiment achieves a fusion decision-making process combining association rule mining and machine learning attribution by merging reliable association rules, root cause candidate features, and adverse root causes to generate the final root cause analysis result. Compared to using association rules or machine learning attribution alone, the fusion decision-making mechanism in this embodiment can cross-validate the results of the two analysis paths, selecting root cause features that are supported by historical data and contribute significantly to the current batch. This avoids misjudgments that may be caused by a single analysis path and improves the reliability and practicality of the final root cause analysis result.
[0148] refer to Figure 6 In one embodiment, the risk warning or optimization process specifically includes the following steps: S801: Real-time acquisition of key production parameters during the PCB inkjet production process.
[0149] In this step, key production parameters in the PCB inkjet production process are acquired in real time through a data acquisition interface. Key production parameters refer to parameters that have a direct impact on inkjet quality and are identified as key influencing factors in the analysis of step S400, including but not limited to: inkjet speed, inkjet spacing, printhead temperature, nozzle ignition frequency, droplet volume, UV curing strength, workshop temperature, workshop humidity, ink temperature, ink viscosity, and nozzle clogging ratio.
[0150] Specifically, the MES system establishes a real-time data communication connection with the control system of the inkjet equipment, acquiring the current values of key production parameters at a preset acquisition frequency (e.g., once every 100 milliseconds). The acquired parameter values are accompanied by timestamps and bound to the batch identifier currently in production.
[0151] In a specific example, the key production parameters acquired in real time during a certain batch of production are: inkjet speed "2.6m / min", inkjet pitch "3.0mm", printhead temperature "31.5℃", nozzle ignition frequency "18kHz", droplet volume "12pL", UV curing strength "275mW / cm²", workshop temperature "26.3℃", workshop humidity "54%RH", ink temperature "28.0℃", ink viscosity "17.3mPa·s", and nozzle clogging ratio "0.08".
[0152] S802. Compare the key production parameters with the corresponding target parameter ranges.
[0153] In this step, the key production parameters acquired in real time are compared with the preset target parameter range to determine whether each parameter is within the normal range.
[0154] The target parameter range is determined based on historical good product data statistics. In one embodiment, key production parameter data of all good product batches over a recent period are collected, the mean and standard deviation of each parameter are calculated, and the range of mean ± 3 times the standard deviation is set as the target parameter range. For example, if the mean viscosity of ink in good product batches is 20.0 mPa·s and the standard deviation is 0.67 mPa·s, then the target parameter range is 18.0 to 22.0 mPa·s.
[0155] In a specific example, the key production parameters acquired in real time were compared with the target parameter range. The comparison results are as follows: inkjet speed "2.6 m / min" (target range 2.0 to 3.0 m / min, normal), printhead temperature "31.5℃" (target range 30.0 to 35.0℃, normal), workshop temperature "26.3℃" (target range 28.0 to 32.0℃, low), ink viscosity "17.3 mPa·s" (target range 18.0 to 22.0 mPa·s, low), and nozzle clogging ratio "0.08" (target range 0.0 to 0.05, high).
[0156] S803. When key production parameters deviate from the target parameter range, output risk warning information.
[0157] In this step, when the comparison result of step S802 shows that the key production parameters deviate from the target parameter range, a risk warning message is output.
[0158] Parameter deviations are often a precursor to batch defects. By promptly issuing early warnings, operators or equipment control systems can be alerted to intervene before large-scale defects occur, thereby preventing batch defects.
[0159] Specifically, based on the type and degree of deviation of the parameter, as well as historical adverse data, corresponding risk warning information is generated. The risk warning information includes at least: the name of the parameter exceeding the limit, its current value, the target parameter range, the direction of deviation, and suggested handling measures.
[0160] In a specific example, if the workshop temperature of 26.3℃ is detected as lower than the lower limit of the target range of 28.0℃, the ink viscosity of 17.3mPa·s is lower than the lower limit of the target range of 18.0mPa·s, and the nozzle clogging ratio of 0.08 is higher than the upper limit of the target range of 0.05, then a risk warning message will be output: "Warning: Workshop temperature 26.3℃, lower than the target range of 28.0 to 32.0℃; ink viscosity 17.3mPa·s, lower than the target range of 18.0 to 22.0mPa·s; nozzle clogging ratio 0.08, higher than the target range of 0.0 to 0.05. Recommendation: Check the workshop air conditioning operation status, check the ink heating system, and arrange nozzle cleaning." In one embodiment, the risk warning information can also be graded according to the degree of deviation. For example, when the degree of deviation is less than 5%, a "general warning" is output; when the degree of deviation is between 5% and 15%, a "critical warning" is output; and when the degree of deviation is greater than 15%, an "emergency warning" is output.
[0161] S804. Determine historical defect patterns based on historical defect information and corresponding production-related data.
[0162] In this step, historical defect patterns are determined based on historical defect information and corresponding production-related data. Historical defect patterns refer to the regularity of parameter combinations or trends in parameter changes that have repeatedly led to defects throughout history.
[0163] A single parameter deviation may be merely an occasional fluctuation and not necessarily lead to defects. However, if a certain combination of parameters has repeatedly caused defects historically, then that parameter combination is considered a high-risk pattern. By analyzing historical defect patterns, early warnings can be issued when the same or similar parameter combinations appear in current production, enabling preventative control based on historical experience.
[0164] Specifically, historical defect information and corresponding production-related data are collected, grouped according to defect type, and the parameter combination characteristics corresponding to each group of defects are analyzed. For example, the distribution patterns of parameters such as workshop temperature, ink viscosity, and inkjet spacing corresponding to all batches of "ink bleeding" defects are statistically analyzed to identify the parameter combinations that occur most frequently.
[0165] In a specific example, after analyzing historical "ink bleeding" defect data, the historical defect pattern was determined to be: "When the workshop temperature is below 28℃ and the ink viscosity is below 18mPa·s, the incidence of ink bleeding defects increases significantly. Historical data shows that the ink bleeding defect rate under this parameter combination is 4.2 times that under normal conditions."
[0166] In one embodiment, parameter drift patterns can also be determined based on historical defect information. For example, if statistics show that "inkjet speed increases by an average of 0.3 m / min after 4 hours of continuous production," this drift pattern may lead to an increased ghosting defect rate in subsequent batches.
[0167] S805. Based on historical defect patterns, identify risks in the current PCB inkjet production process and output prevention alerts when risk characteristics corresponding to historical defect patterns are detected.
[0168] In this step, risks are identified in the current PCB inkjet production process based on the historical defect patterns determined in step S804. When a risk characteristic corresponding to a historical defect pattern is detected in the current production process, a prevention alert is output.
[0169] Risk warning information is only output when the parameters have deviated from the target range, while historical adverse patterns can help provide early warnings when the parameters have not yet deviated from the target range but have shown a trend similar to historical adverse patterns.
[0170] Specifically, the key production parameters in the current production process are matched with parameter combinations from historical defect patterns. If the similarity between the current parameter combination and the parameter combinations from historical defect patterns exceeds a preset threshold, it is determined that a risk characteristic corresponding to the historical defect pattern has been detected, and a prevention alert is output.
[0171] In a specific example, the current production parameters are: workshop temperature "28.5℃" and ink viscosity "17.8 mPa·s". The historical defect pattern is "the ink bleeding defect rate increases significantly when the workshop temperature is below 28℃ and the ink viscosity is below 18 mPa·s". It is determined that the current ink viscosity "17.8 mPa·s" already meets the condition of "below 18 mPa·s". Although the workshop temperature "28.5℃" is not below 28℃, it is close to the threshold. The similarity exceeds the preset threshold, so a preventative warning message is output: "Prevention warning: The current ink viscosity of 17.8 mPa·s is below 18.0 mPa·s, and the workshop temperature of 28.5℃ is close to the 28.0℃ threshold, matching the historical ink bleeding defect pattern. It is recommended to check the ink heating system and workshop air conditioning operation status in advance to avoid further deviation of the parameters." In one embodiment, parameter change trends can also be predicted based on historical defect patterns. For example, based on historical data, when the inkjet speed continuously increases at a rate of 0.1 m / min per hour, it is predicted that the high-risk range for ghosting defects will be reached after 4 hours. Based on this, a preventative warning message is output: "Prevention Warning: The current inkjet speed is 2.5 m / min. Based on the current drift trend, it is predicted to reach 2.9 m / min in 4 hours, approaching the high-risk range for ghosting defects (≥2.9 m / min). It is recommended to check the inkjet speed control system in advance." refer to Figure 7 In one embodiment, after establishing the end-to-end data association and identifying adverse information, the method further includes performing forward or reverse tracing based on the end-to-end data association, specifically including the following steps: S901. Obtain production-related data during the PCB production process, associate the production-related data based on batch identifiers, and establish a full-link data association relationship.
[0172] S902. Obtain the inspection data after inkjet printing on the PCB, and determine the defect information in the inkjet printing process based on the inspection data.
[0173] S903. Perform forward or reverse tracing based on the data association relationship across the entire chain.
[0174] In this step, based on the query conditions entered by the user, forward or reverse tracing is performed based on the data relationship across the entire chain. Forward tracing refers to starting from the production order and tracing layer by layer to the specific production batch and its corresponding production parameters, production equipment, production materials, and test results; reverse tracing refers to starting from the defect information and tracing back to the specific production batch and its corresponding production parameters, production equipment, production materials, and production environment.
[0175] Forward tracing is used to "find which parameters, equipment, and materials were used in the production process of a given order or batch", which is suitable for quality confirmation or customer audit scenarios before order delivery; reverse tracing is used to "find which parameters, equipment, and materials may have caused the defect in the production process of a given PCB", which is suitable for root cause investigation scenarios after the defect occurs.
[0176] In one embodiment, the specific process of forward traceability is as follows: starting from the production order, obtain the corresponding Panel code (master board), read the Panel code and print the PCS code (daughter board) through the text inkjet process, complete the Panel-PCS binding and upload, bind the PCS code to the full process data (drilling / solder resist / development / text / AOI), and finally, the full process history can be queried by entering the PCS code.
[0177] Specifically, forward traceability includes: obtaining the corresponding batch information based on the order information, and obtaining at least one of the corresponding production parameters, production equipment, production materials, and test results based on the batch information.
[0178] In one embodiment, the specific process of reverse tracing is as follows: starting from the defective board detected by AOI, scan the PCS code of the defective board, and use the PCS code to query the Panel code and the data of the entire process. Associate the inkjet parameters, equipment, time, and operator information of the text process, and finally realize the cross-process location of the root cause (drilling / solder resist / development / text / AOI).
[0179] Specifically, reverse tracing includes: obtaining the corresponding batch information based on the defective information, and obtaining at least one of the corresponding production parameters, production equipment, production materials, and production environment based on the batch information.
[0180] In one embodiment, forward and reverse tracing also support multi-condition combined queries. For example, a user can simultaneously input three conditions: "Defect type = ink seepage", "Device ID = PRT-003", and "Time range = 2025-06-01 to 2025-06-07". The system will then retrieve batch records that meet all of the above conditions from the end-to-end data association and output the corresponding tracing results.
[0181] In one embodiment, when a defective Pcs is detected, the production parameters of other Pcs in the same panel can be automatically traced. Specifically, the panel code to which the defective Pcs belong is retrieved based on the Pcs code. Then, using the panel code as an index, the production parameters and detection results of all Pcs in the same panel are obtained from the end-to-end data association relationship. If the number of defective Pcs in the same panel is greater than or equal to a preset threshold (e.g., 3), it is determined to be a batch anomaly, a batch anomaly prompt message is output, and the production parameters of all Pcs in the panel are compared and analyzed with those of the normal panel to identify the common factors causing the batch anomaly.
[0182] S904. Based on the full-link data association relationship, the defect information is associated with the corresponding production-related data to obtain defect-related data.
[0183] S905. Analyze production-related data based on adverse correlation data, obtain at least one key influencing factor corresponding to adverse information, and determine the root cause of adverse based on the key influencing factor.
[0184] S906. Generate root cause analysis results based on the root causes of defects, and provide risk warnings or optimizations to the PCB inkjet production process based on the root cause analysis results.
[0185] refer to Figure 8This application also provides an embodiment of a MES-based PCB inkjet end-to-end traceability and quality root cause acquisition device, the device comprising: The data association module 201 is used to acquire production-related data in the PCB production process, associate the production-related data based on batch identifiers, and establish a full-link data association relationship. The defect determination module 202 is used to acquire the inspection data after inkjet printing on the PCB and determine the defect information during the inkjet printing process based on the inspection data. The defect association module 203 is used to associate defect information with corresponding production association data based on the full-link data association relationship to obtain defect association data; The root cause analysis module 204 is used to analyze production-related data based on adverse correlation data, obtain at least one key influencing factor corresponding to the adverse information, and determine the root cause of the adverse based on the key influencing factor. The result output module 205 is used to generate root cause analysis results based on the root causes of defects, and to provide risk warnings or optimizations to the PCB inkjet production process based on the root cause analysis results.
[0186] Please see Figure 9 This application also provides an embodiment of a MES-based PCB inkjet end-to-end traceability and quality root cause analysis device, the device comprising: Processor 301, memory 302, input / output unit 303, bus 304; The processor 301 is connected to the memory 302, the input / output unit 303, and the bus 304; The memory 302 stores the program, and the processor 301 calls the program to execute any of the above-mentioned MES-based PCB inkjet end-to-end traceability and quality root cause analysis methods.
[0187] This application also relates to a computer-readable storage medium on which a program is stored, which, when run on a computer, causes the computer to execute any of the above-mentioned MES-based PCB inkjet end-to-end traceability and quality root cause analysis methods.
[0188] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0189] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another, 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 units may be electrical, mechanical, or other forms.
[0190] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0191] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0192] If the integrated unit 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, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part 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 described 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.
Claims
1. A method for end-to-end traceability and root cause analysis of PCB inkjet printing based on MES, characterized in that, include: Acquire production-related data during the PCB manufacturing process, associate the production-related data based on batch identifiers, and establish a full-chain data association relationship; Obtain the inspection data after inkjet printing on the PCB, and determine the defect information in the PCB inkjet printing process based on the inspection data; Based on the aforementioned end-to-end data association relationship, the defect information is associated with the corresponding production-related data to obtain defect-related data; Based on the aforementioned adverse correlation data, the production correlation data is analyzed to obtain at least one key influencing factor corresponding to the adverse information, and the root cause of the adverse is determined based on the key influencing factor, including: Establish a correlation model between defect types, process parameters, environmental conditions, and material attributes; map the process parameters, environmental conditions, and material attributes in the production correlation data to the correlation model to generate feature data corresponding to the defect information; extract defect type tags based on the defect information; discretize the continuous features in the feature data, and construct a transaction database based on the discretized feature data and the corresponding defect type tags; process the transaction database using a preset algorithm to obtain the correlation rules corresponding to the defect type tags, and determine the key influencing factors according to the correlation rules; The feature data is input into a pre-trained learning model to obtain the feature contribution value corresponding to the bad type label; the features are sorted in descending order according to the absolute value of the feature contribution value to obtain the root cause ranking list; the features in the root cause ranking list that meet the first preset ranking condition are set as root cause candidate features. The features in the association rule are matched with the root factor candidate features in the root cause ranking list to obtain feature matching results. Based on the feature matching results, it is determined whether there is an intersection between the features contained in the association rule and the sub-features in the root cause ranking list that meet the second preset ranking condition. If the intersection is determined, the lift degree corresponding to the association rule is obtained, and it is determined whether the lift degree meets the preset lift degree condition. If the lift degree meets the preset lift degree condition, the association rule is determined as a reliable association rule corresponding to the adverse information. Based on the reliable association rule, the root cause candidate features, and the adverse root cause, the final root cause analysis result is generated. Based on the aforementioned root causes of defects, root cause analysis results are generated, and risk warnings or optimization measures are implemented for the PCB inkjet production process according to the root cause analysis results.
2. The MES-based PCB inkjet end-to-end traceability and quality root cause analysis method according to claim 1, characterized in that, The feature data consists of a process parameter subspace, an environmental condition subspace, and a material property subspace; the determination of the root cause of defects based on the key influencing factors includes: The corresponding root cause process source is determined based on the feature subspace to which the root cause candidate features belong. Calculate the root cause confidence score based on the feature contribution value corresponding to the root cause candidate features; Based on the root cause confidence level, determine whether the root cause candidate features meet the root cause reliability condition; When the root cause confidence level meets the root cause reliability condition, the root cause candidate features and the corresponding root cause process source are determined as the bad root cause. When the root cause confidence level does not meet the root cause reliability condition, a manual review process is triggered.
3. The MES-based PCB inkjet end-to-end traceability and quality root cause analysis method according to claim 1, characterized in that, The risk warning or optimization measures include: Real-time acquisition of key production parameters during the PCB inkjet production process; Compare the key production parameters with the corresponding target parameter ranges; When the key production parameters deviate from the target parameter range, a risk warning message is output. Historical defect patterns are determined based on historical defect information and corresponding production-related data. The risk identification process for the current PCB inkjet production process is performed based on the historical defect patterns, and prevention prompts are output when risk characteristics corresponding to the historical defect patterns are detected.
4. The MES-based PCB inkjet end-to-end traceability and quality root cause analysis method according to claim 1, characterized in that, Also includes: Perform forward or reverse tracing based on the aforementioned end-to-end data relationships; The forward traceability includes obtaining the corresponding batch information based on the order information, and obtaining at least one of the corresponding production parameters, production equipment, production materials and test results based on the batch information; The reverse tracing includes obtaining the corresponding batch information based on the defective information, and obtaining at least one of the corresponding production parameters, production equipment, production materials, and production environment based on the batch information.
5. A MES-based PCB inkjet end-to-end traceability and quality root cause analysis device, applied to the MES-based PCB inkjet end-to-end traceability and quality root cause analysis method according to any one of claims 1 to 4, characterized in that, include: The data association module is used to acquire production-related data during the PCB production process, associate the production-related data based on batch identifiers, and establish a full-link data association relationship; The defect determination module is used to acquire the inspection data after inkjet printing on the PCB and determine the defect information during the inkjet printing process based on the inspection data. The defect association module is used to associate the defect information with the corresponding production association data based on the full-link data association relationship to obtain defect association data; The root cause analysis module is used to analyze the production-related data based on the adverse correlation data, obtain at least one key influencing factor corresponding to the adverse information, and determine the root cause of the adverse based on the key influencing factor. The results output module is used to generate root cause analysis results based on the root causes of the defects, and to provide risk warnings or optimizations to the PCB inkjet production process based on the root cause analysis results.
6. A PCB inkjet end-to-end traceability and quality root cause analysis device based on MES, characterized in that, include: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The memory stores a program, and the processor calls the program to execute the MES-based PCB inkjet end-to-end traceability and quality root cause analysis method as described in any one of claims 1 to 4.
7. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains a program that, when executed on a computer, performs the MES-based PCB inkjet end-to-end traceability and quality root cause analysis method as described in any one of claims 1 to 4.
