Circuit board manufacturing method and electronic device

Abstract

The application discloses a circuit board manufacturing method and electronic equipment, relates to the technical field of production data collaborative analysis, and first establishes a process influence coupling model based on single-board time sequence data sequences corresponding to a whole production line in a historical production process. The process influence coupling model can quantize the dynamic influence capability of each process on a subsequent quality state. Thus, the fusion weight expressing the influence and causality of each process and the reliability of data can be determined in combination with the process influence coupling model and the reliability of each process data source equipment. In this way, the fusion weight is used to fuse the quality characteristics of each process of the current batch, so that a comprehensive quality characterization vector corresponding to the collaborative state of the whole process is obtained, and a defect propagation path diagram matched with the comprehensive quality characterization vector is constructed. With the aid of the defect propagation path diagram, a target process causing defects of the circuit board can be determined from the production line, and the target process is optimized, so that the overall process performance can be effectively improved.

Description

Technical Field

[0001] This application relates to the field of collaborative analysis technology for production data, and in particular to a circuit board manufacturing method and an electronic device. Background Technology

[0002] In the field of collaborative analysis of production data, collaborative analysis technology for production data is mostly based on monitoring independent KPIs (Key Performance Indicators) for each process.

[0003] The lack of a comprehensive indicator that can uniformly reflect the cumulative quality status of the entire board from material input to final inspection makes it difficult for management to judge the overall risk level of a certain batch, to make horizontal comparisons between different production lines or shifts, and to determine which process is evolving into a new dynamic bottleneck, thus making it difficult to optimize the production process.

[0004] Therefore, how to effectively optimize processes precisely during collaborative analysis of production data has become a pressing technical problem that needs to be solved. Summary of the Invention

[0005] This application provides a circuit board manufacturing method and an electronic device to solve the problems in circuit board manufacturing.

[0006] A circuit board manufacturing method includes: acquiring a single-board time-series data sequence corresponding to the entire production line process during historical production; determining the quality correlation characteristics between preceding and following processes using the single-board time-series data sequence, and determining the causal verification results between preceding and following processes based on the quality correlation characteristics; establishing a process influence coupling model using the causal verification results, and determining the fusion weight of each process by combining the process influence coupling model and the credibility of the data source equipment of each process; fusing the quality characteristics of each process in the current batch using the fusion weight to obtain a comprehensive quality characterization vector corresponding to the overall process collaborative state, and constructing a defect propagation path diagram matching the comprehensive quality characterization vector; using the defect propagation path diagram to determine the target process currently causing defects in the circuit board from the production line, and using the production line after optimizing the target process to produce the circuit board.

[0007] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the circuit board manufacturing methods described above.

[0008] This application also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the circuit board manufacturing methods described above.

[0009] This application also provides a computer program circuit board, including a computer program, which, when executed by a processor, implements the steps of any of the circuit board manufacturing methods described above.

[0010] In this application, firstly, based on the single-board time-series data sequence corresponding to the entire production line process in historical production, the quality correlation characteristics between preceding and following processes are determined, and the causal verification results between preceding and following processes are determined based on the quality correlation characteristics. A process influence coupling model can be established based on the causal verification results. The process influence coupling model can quantify the dynamic influence capability of each process on the subsequent quality state. Therefore, after obtaining the process influence coupling model, the fusion weight of each process can be determined by combining the process influence coupling model and the reliability of the data source equipment for each process. In other words, the causal relationship of each process's influence and the reliability of the data can be expressed based on the fusion weight. Thus, by fusing the quality characteristics of each process in the current batch using the fusion weight, a comprehensive quality characterization vector corresponding to the overall process collaborative state can be obtained, and a defect propagation path diagram matching the comprehensive quality characterization vector can be constructed. Finally, with the help of the defect propagation path diagram, the target process causing the circuit board defect can be identified from the production line, the target process can be optimized, and then the production line after optimizing the target process can be used to produce circuit boards, thereby effectively improving the circuit board production capacity.

[0011] This application achieves a systematic quantification of the impact on the process through multi-dimensional contribution assessment, avoiding the assessment bias caused by relying on a single indicator. The contribution of the impact integrates structural impact and process stability, and can identify target processes that are both critical and defective. Optimizing the target process can effectively improve the overall process performance. Attached Figure Description

[0012] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating an embodiment of a circuit board manufacturing method provided in this application;

[0014] Figure 2 This is a schematic diagram illustrating a specific closed-loop optimization of a circuit board manufacturing method provided in an embodiment of this application;

[0015] Figure 3 This is a schematic diagram of the structure of a circuit board manufacturing apparatus provided in an embodiment of this application;

[0016] Figure 4This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0017] Figure 5 This is a schematic diagram of the specific structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. 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 protection scope of this application.

[0019] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0020] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Please refer to Figure 1 The circuit board manufacturing method includes the following steps.

[0022] S101. Obtain the single-board timing data sequence corresponding to the entire production line process during the historical production process.

[0023] In this application, the production line is a combination of a series of equipment, workstations, and personnel arranged in a standardized and sequential manner. Its core function is to efficiently transform raw materials / components into finished or semi-finished products through continuous and clearly defined processes. This production line can be any production line requiring comprehensive analysis and optimization of its processes; for example, it can be used for the production of printed circuit boards (PCBs).

[0024] A single-board timing data sequence refers to the data generated, collected, and evaluated during the production of a specific circuit board on a production line, arranged in a time-series manner. It's important to note that this single-board timing data sequence is not just data from a single process, but rather data from multiple processes throughout the entire production flow, including process data and quality characteristic data.

[0025] In one specific embodiment of this application, obtaining the single-board timing data sequence corresponding to the entire production line process during historical production includes: Step 1, collecting the original dataset generated by each process equipment in the production line during the production process; Step 2, based on the unique identifier of a single circuit board and the material entry and exit time events of each process, performing time alignment and structuring processing on the original dataset to obtain the single-board timing data sequence.

[0026] Specifically, in practical applications, raw data generated by equipment at each stage of the production line can be collected during the production process and placed into a raw dataset. Then, based on the unique identifier of each circuit board and the material input / output time events of each stage, these raw datasets can be time-aligned and structured to obtain a single-board timing data sequence.

[0027] For example, if the production line corresponds to PCBA processing in the electronics manufacturing industry, multi-source heterogeneous data generated by the equipment in each process of the PCBA production line during processing can be collected. The processes include solder paste printing, chip mounting, reflow soldering, AOI inspection, DIP insertion, wave soldering, ICT / FCT testing, and rework, forming a raw dataset at the single-board level. Then, based on the unique identifier of each PCB board and the material entry and exit time events of each process, the raw dataset is time-aligned and structured to generate a single-board level time-series data sequence containing the quality characteristics of the entire process (single-board time-series data sequence when the circuit board is a PCB board).

[0028] Quality characteristics are encompassed within the diverse and heterogeneous data generated by each process's equipment during manufacturing. Raw data includes operating parameters (such as temperature profiles and pressure values) from equipment like solder paste printers, pick-and-place machines, reflow ovens, AOI, and ICT / FCT, sensor readings (such as thickness and positional offset), image recognition results (such as defect type and quantity), and operation logs (such as material change records). All of this data is multi-source and heterogeneous. Quality characteristics are key indicators extracted or calculated from this raw data, used to characterize the degree of impact of each process on the final circuit board quality.

[0029] S102. Use the single-board timing data sequence to determine the quality correlation characteristics between the preceding and following processes, and determine the causal verification results between the preceding and following processes based on the quality correlation characteristics.

[0030] On a production line, different processes, especially those preceding and following processes, often have a causal impact on the latter. In this embodiment, the quality correlation characteristics between preceding and following processes can be determined using single-board timing data sequences, and then the causal verification results between the preceding and following processes can be determined based on the quality correlation characteristics. In other words, for processes with quality correlation, we then examine whether a causal relationship exists between them.

[0031] In one specific embodiment of this application, the quality correlation characteristics between preceding and subsequent processes are determined using a single-board time-series data sequence, and the causal verification results between preceding and subsequent processes are determined based on the quality correlation characteristics. This includes: pairing key process parameters and subsequent quality inspection results in the single-board time-series data sequence to obtain data pairs of process input and output quality; calculating the correlation coefficient within the data pairs and using the correlation coefficient to filter out statistically significant target data pairs from the data pairs; performing causal verification on the variable pairs of the target data pairs under controlled environmental temperature and humidity, equipment aging degree, and material batch fluctuations to determine whether there is a unidirectional causal drive from the preceding process to the subsequent result; if a unidirectional causal drive exists, recording the significant index corresponding to the target data pair, determining the significant index as the initial causal strength, and marking the causal direction; and determining the causal verification results for process pairs with causal relationships, the initial causal strength corresponding to the process pair, and the causal direction corresponding to the process pair.

[0032] Specifically, key process parameters and subsequent quality inspection results in the historical production single-board time-series data are paired to form process input-output data pairs. Based on these data pairs, the Pearson correlation coefficient between preceding process parameters and subsequent quality indicators is calculated, and statistically significant correlations are screened through significance testing. Under controlled conditions of ambient temperature and humidity, equipment aging, and material batch fluctuations, Granger causality tests are performed on the variable pairs that pass the correlation test to determine whether a unidirectional causal drive exists from the preceding process to the subsequent result. For process pairs that pass the causality test, their F-statistics are recorded as an initial causal strength reference, and the causal direction is marked. In other words, the causality test results include process pairs with causal relationships, the initial causal strength of the process pairs, and the causal direction of the process pairs.

[0033] Among them, the Granger causality test's F-statistic is the core statistical tool for determining whether one variable can significantly predict another variable. Essentially, it quantifies whether the improvement in prediction accuracy is significant by comparing the fitting effects of two regression models that include and do not include the lagged term of the variable being tested.

[0034] After obtaining the causal test results, a process influence coupling model can be established based on the causal test results.

[0035] In one specific embodiment of this application, a process influence coupling model is established using causal test results, including: creating nodes using processes in a process pair; connecting nodes using causal directions and directed edges; and assigning values ​​to the directed edges using initial causal strength and correlation coefficients to obtain the process influence coupling model. That is, all process pairs with causal relationships are organized into a directed graph structure, where nodes represent processes and directed edges represent causal influence paths. Then, a weight is assigned to each directed edge, which is jointly determined by causal strength and correlation coefficient, constituting the process influence coupling model. This process influence coupling model is used to quantify the dynamic influence capability of each process on the subsequent quality status.

[0036] S103. After establishing the process influence coupling model using the causal test results, determine the fusion weight of each process by combining the process influence coupling model and the credibility of the data source equipment of each process.

[0037] After constructing the process influence coupling model, the fusion weight of each process can be determined by combining the process influence coupling model with the credibility of the data source equipment for each process. The credibility of the data source equipment for each process can be evaluated using multi-dimensional information.

[0038] In one specific embodiment of this application, the fusion weight of each process is determined by combining the process influence coupling model and the credibility of the data source equipment for each process. This includes: extracting the influence intensity value of each process pointing to all subsequent processes from the process influence coupling model, and summing all the influence intensity values ​​corresponding to a single process to obtain the comprehensive influence assessment value of that single process; obtaining the historical operation data integrity, calibration cycle compliance, and sensor stability information of the production equipment connected to each process; determining the equipment credibility level using the historical operation data integrity, calibration cycle compliance, and sensor stability information; mapping the equipment credibility level to a data credibility factor within a preset numerical range; multiplying the comprehensive influence assessment value of each process by the corresponding data credibility factor to obtain the original fusion score; and normalizing the original fusion scores of all processes to obtain the fusion weight corresponding to each process.

[0039] The influence intensity value of each process pointing to all subsequent processes is extracted from the process influence coupling model and summed to obtain the comprehensive influence assessment value of the process. Historical operating data integrity, calibration cycle compliance, and sensor stability information of the production equipment connected to each process are obtained, and equipment reliability levels are generated based on preset scoring rules. The equipment reliability level is mapped to a data reliability factor between 0.8 and 1.2 (other values ​​can be set in practical applications) to reflect the reliability level of the process data. The comprehensive influence assessment value of each process is multiplied by its corresponding data reliability factor to obtain the original fusion score. The original fusion scores of all processes are normalized so that the sum is 1, obtaining the fusion weight of each process in the current production batch. The fusion weight is used to reflect the differences in the importance of different process data in the subsequent weighted integration process.

[0040] S104. Use fusion weights to fuse the quality characteristics of each process in the current batch to obtain a comprehensive quality characterization vector corresponding to the overall collaborative state of the process, and construct a defect propagation path diagram that matches the comprehensive quality characterization vector.

[0041] In this embodiment, the quality characteristics of each process in the current batch can be obtained by referring to the method of obtaining the single-board time-series data sequence during the historical generation process. Then, the quality characteristics of each process in the current batch are fused using fusion weights to obtain a comprehensive quality characterization vector corresponding to the overall collaborative state of the process.

[0042] To identify the processes that cause defects, a defect propagation path diagram that matches the comprehensive quality characterization vector can be constructed.

[0043] In one specific embodiment of this application, the quality characteristics of each process in the current batch are fused using fusion weights to obtain a comprehensive quality representation vector corresponding to the overall collaborative state of the process. This includes: obtaining the current single-board time-series data sequence corresponding to the current batch of circuit boards; extracting key quality feature values ​​of each process from the current single-board time-series data sequence; multiplying the key quality feature values ​​by the fusion weights of the corresponding processes to obtain weighted feature components; and linearly summing the weighted feature components of all processes to generate a comprehensive quality representation vector for the circuit board.

[0044] The following explanation uses a PCB board as an example to illustrate the above steps.

[0045] PCB (Printed Circuit Board) refers to a printed circuit board; PCBA (Printed Circuit Board Assembly) refers to a printed circuit board.

[0046] Extract key quality characteristic values ​​for each process from the single-board level timing data sequence of each PCB board in the current batch. This includes solder paste thickness deviation, mounting offset, number of AOI defects, and electrical test parameter offset values; quality characteristic values. Data fusion weights with corresponding processes Multiplying them together yields the weighted eigencomponents. Weighted characteristic components for all processes Perform linear summation to generate the comprehensive quality characterization vector Z of the PCB board. Its calculation expression is as follows: Where n is the total number of processes involved in the analysis, and Z reflects the cumulative quality deviation of the board in the entire production process; the above weighted fusion operation is performed on all PCB boards in the current batch to obtain a batch-level comprehensive quality characterization vector set; the comprehensive quality characterization vector set is used for subsequent defect propagation path modeling and trend identification.

[0047] In one specific embodiment of this application, constructing a defect propagation path graph matching the comprehensive quality representation vector includes: obtaining the comprehensive quality representation vector set of the current batch of circuit boards; the comprehensive quality representation vector set includes the comprehensive quality representation vectors corresponding to each circuit board in the current batch; selecting quality variables from the comprehensive quality representation vector set whose correlation with the final functional test failure rate is higher than a preset threshold, and adding the selected quality variables to the key variable set; using the key variable set as input, performing unsupervised structure learning under the condition of potential confounding factors to identify directed causal edges between variables; and calculating the propagation strength of each identified directed causal edge. The formula for calculating conduction strength is as follows: ; For variables i and j in the control parent node set p-value for conditional independence test under given conditions The partial correlation coefficient is the residual; using directed causal edges and their strength values. Construct a defect propagation path graph; the nodes of the defect propagation path graph correspond to quality variables, and the directed edges correspond to the causal propagation direction.

[0048] The following explanation uses a PCB board as an example to illustrate the above steps.

[0049] Quality variables with a correlation higher than a preset threshold to the final functional test failure rate are selected from the comprehensive quality representation vector set to form a key variable set. Using this key variable set as input, the FCI causal discovery algorithm is used for unsupervised structure learning under the condition of potential confounding factors to identify directed causal edges between variables. The propagation strength of each identified directed causal edge is calculated. This strength is determined jointly based on the p-value of the conditional independence test and the residual partial correlation coefficient, and its expression is: ;in, For variables i and j in the control parent node set p-value for conditional independence test under given conditions The partial correlation coefficient is the residual; all directed causal edges and their strength values ​​are considered. The organization is represented by a defect propagation path diagram, where nodes represent quality variables and directed edges represent the direction of causal transmission.

[0050] The defect propagation path diagram can also be visualized, making it easier for engineers to understand the defect evolution mechanism.

[0051] S105. Using the defect propagation path diagram, identify the target process in the production line that is causing the circuit board defect, and use the production line after optimizing the target process to produce the circuit board.

[0052] Specifically, one can start from the final functional failure point and trace back along the path of maximum cumulative intensity to identify the earliest starting point of abnormal fluctuations.

[0053] By mapping the starting node back to its corresponding physical process, the root cause process leading to the defect is identified, i.e., the target process is determined. This target process can then be optimized specifically. Finally, circuit boards can be produced on the optimized production line, thereby increasing capacity, such as by accelerating production or improving the performance of the circuit boards themselves.

[0054] In one specific embodiment of this application, when new key quality variables are added (such as introducing new test items); or when the original causal relationship between certain variables becomes invalid due to process adjustments; or when it is found that the existing path cannot explain the current defect pattern, indicating the existence of a new, unidentified transmission mechanism, the defect propagation path diagram can be updated. The update method can refer to the method used to construct the defect propagation path diagram, that is, to reconstruct a new defect propagation path diagram based on the current situation. In this way, the defect propagation path diagram can adapt to the new situation, thereby identifying the target process in the production line that is currently causing defects in the circuit board, and optimizing the target process.

[0055] The circuit board manufacturing method provided in this application first determines the quality correlation characteristics between preceding and following processes based on the single-board time-series data sequence corresponding to the entire production line process in historical production. Then, based on these quality correlation characteristics, the causal verification results between preceding and following processes are determined. A process influence coupling model can be established based on the causal verification results. This model quantifies the dynamic influence of each process on the subsequent quality state. Therefore, after obtaining the process influence coupling model, the fusion weight of each process can be determined by combining the model with the reliability of the data source equipment for each process. In other words, the causal relationship of each process's influence and the reliability of the data can be expressed based on the fusion weight. Thus, by fusing the quality characteristics of each process in the current batch using the fusion weight, a comprehensive quality characterization vector corresponding to the overall process coordination state can be obtained, and a defect propagation path diagram matching the comprehensive quality characterization vector can be constructed. Finally, using the defect propagation path diagram, the target process causing the circuit board defect can be identified from the production line. This target process can be optimized, and then the production line after optimizing the target process can be used to produce circuit boards, effectively improving the circuit board production capacity.

[0056] This application achieves a systematic quantification of the impact on the process through multi-dimensional contribution assessment, avoiding the assessment bias caused by relying on a single indicator. The contribution of the impact integrates structural impact and process stability, and can identify target processes that are both critical and defective. Optimizing the target process can effectively improve the overall process performance.

[0057] It should be noted that, based on the above embodiments, the embodiments of this application also provide corresponding improvement schemes. In the preferred / improved embodiments, the same or corresponding steps as in the above embodiments can be referred to each other, and the corresponding beneficial effects can also be referred to each other; however, these will not be elaborated upon in the preferred / improved embodiments herein.

[0058] In one specific embodiment of this application, the method further includes: enumerating all complete transmission paths from initial deviation to final failure in the defect propagation path diagram to obtain a path set; counting the frequency of each process occurrence in all transmission paths, and determining the frequency as its path coverage metric; and calculating the overall reliability of each path P. The formula for calculating overall credibility is: ; Let P be the number of edges. Let the strength value of the edges in the path be denoted by ; for each path P containing a certain process i, calculate its local contribution value; the formula for calculating the local contribution value is path frequency and path reliability. Reciprocal of path length The product of the values; summing over all local contribution values ​​yields the process. Path weights The formula for calculating the action path weight is as follows: ; Let i be the set of all paths that contain process i. Let P be the frequency of the path; calculate the standard deviation of key indicators for each process in recent consecutive production batches to obtain the quality stability index. ; Weight the action path With quality stability indicators Multiply to obtain the influence contribution. Trend analysis was conducted on the contribution of the process to the impact, and the process with a high contribution and an upward trend was identified as the dynamic bottleneck process.

[0059] Specifically, all complete transmission paths from initial deviation to final failure in the defect propagation path diagram can be enumerated to form a path set; the frequency of each process appearing in all transmission paths can be counted as its path coverage metric; and for each path P, its overall reliability can be calculated. This confidence level is determined by the strength values ​​of all edges in the path. The geometric mean is determined by the following expression: ;in, Let be the number of edges in path P; for each path P containing a certain process i, calculate its local contribution value, which is determined by the path frequency and path reliability. Reciprocal of path length Multiply to obtain; sum over all local contribution values ​​to obtain the action path weight of process i. Its expression is: ;in, Given the set of all paths containing process i; calculate the standard deviation of key performance indicators for each process in recent consecutive production batches to obtain the quality stability index. ; Weight the action path With quality stability indicators Multiply to obtain the influence contribution. Its expression is: Impact contribution It comprehensively reflects the structural impact of each process on the final yield and the stability of the process; it performs trend analysis on the contribution sequence of the impact and identifies processes with high contribution and an upward trend as dynamic bottleneck processes.

[0060] In one specific embodiment of this application, the method further includes: searching for matching cases from historical cases in the process optimization knowledge base based on the process type and the changing trend of the impact contribution of the dynamic bottleneck process; extracting effective parameter adjustment strategies from the matching cases; converting the parameter adjustment strategies into control instruction packages; sending the control instruction packages to the production equipment controller of the corresponding process using the manufacturing execution system; collecting key quality indicators and final function test pass rates of the process in several subsequent continuous production batches to form an optimized dataset; comparing the optimized dataset with the unoptimized dataset to obtain an evaluation of the optimization effect; and determining that the optimization is effective if the evaluation of the optimization effect passes the significance assessment and the improvement is greater than a preset threshold.

[0061] The process optimization knowledge base is constructed by collecting historical production records of effective anomaly handling that have been verified through engineering processes, forming an original case set. Each case undergoes structured extraction, including the anomaly pattern, involved processes, adjusted parameters, adjustment direction, implementation batch, and verification results. Anomaly patterns are uniformly coded to establish a standardized anomaly code system. The mapping relationship between anomaly patterns and effective countermeasures in each case is stored as structured table entries. The mapping relationship table is periodically reviewed and updated. An access interface is provided for the analysis system to support real-time querying and matching. When performing pattern matching, a weighted cosine similarity method is used to calculate the matching degree between the current anomaly pattern vector a and the historical pattern vector b in the knowledge base. Its expression is: ;in, The weights for the k-th dimension features are set by expert experience; the process optimization knowledge base provides strategy support for closed-loop control, improving optimization response speed and success rate.

[0062] Specifically, the identified dynamic bottleneck process types and their changing trends in impact contribution can be matched with historical cases in the process optimization knowledge base; effective parameter adjustment strategies can be extracted from the successfully matched cases, including the parameter names, adjustment directions, and target value ranges; the adjustment strategies can be converted into control instruction packages conforming to the OPC UA protocol specification; the control instruction packages can be sent to the production equipment controllers of the corresponding processes through the manufacturing execution system; key quality indicators and final functional test pass rates of the process can be collected in at least three consecutive production batches to form an optimized dataset; the optimized dataset can be compared with the unoptimized dataset to evaluate the optimization effect; a two-sample t-test can be used to assess the significance of the difference between the means of the two sets of yield data, and the t-value can be calculated, the expression of which is: ;in, The average yield before and after optimization are shown below. Standard deviation, The sample size is used; if the t-test result is significant and the improvement is greater than the preset threshold, the optimization is confirmed to be effective, and the model update process is initiated.

[0063] To facilitate those skilled in the art to better understand and implement the circuit board manufacturing method provided in the embodiments of this application, the circuit board manufacturing method will be described in detail below using a production line corresponding to a PCB board as an example.

[0064] Please refer to Figure 2 The circuit board manufacturing method provided in this application embodiment can collaboratively analyze production data for PCBA processing in the electronics manufacturing industry, specifically including the following steps.

[0065] S1. Collect multi-source heterogeneous data generated by equipment in each process of the PCBA production line during processing to obtain the original dataset in units of single boards.

[0066] Among them, multi-source heterogeneous data acquisition covers processes such as solder paste printing, chip mounting, reflow soldering, AOI inspection, DIP insertion, wave soldering, ICT / FCT testing, and rework.

[0067] Furthermore, the multi-source heterogeneous data includes equipment operating parameters, real-time sensor data, process settings, quality inspection results, and operation log information. The data is collected in real time from the control systems of equipment in each process through the OPC UA protocol or Modbus TCP interface, and is initially cleaned and cached by edge computing nodes.

[0068] The original dataset uses the unique serial number of each PCB board as the index unit, recording all the original data entries generated during each process, ensuring that the data source is traceable, the timestamp is complete, and the field semantics are consistent, providing a data foundation with high integrity and low noise for subsequent cross-process correlation analysis.

[0069] By constructing the original dataset using the unique serial number of each board as the core index, the precise binding of physical circuit boards and digital data is achieved, ensuring the accuracy of subsequent cross-process data association. The initial cleaning at the edge effectively removes communication noise, duplicate reporting, and obvious outliers, reducing the processing load of the central system. The unified collection and semantic alignment of multi-source data provides basic support for breaking down data silos at the device and system levels, ensuring the data integrity and traceability of the entire process analysis.

[0070] S2. Based on the unique identifier of each PCB board and the material entry and exit time events of each process, the original dataset is time-aligned and structured to generate a single-board level time-series data sequence containing the quality characteristics of the entire process.

[0071] Furthermore, the time alignment process is based on the material feeding and discharging time event sequence of each process. The Dynamic Time Warping (DTW) algorithm is used to align data streams with different sampling frequencies and time offsets to eliminate time misalignment caused by equipment response delays or communication jitter.

[0072] Structured processing maps unstructured logs, image detection results, and numerical process parameters into standardized fields, constructing a structured data table containing timestamps, process codes, equipment numbers, process parameter values, detection result codes, and quality feature labels.

[0073] The single-board level time sequence data is aggregated according to the unique identifier of the PCB board to form a multi-dimensional time sequence data structure with the board as the unit, the process as the stage, and time as the axis. It completely preserves the quality evolution trajectory of the entire process from material feeding to final inspection, and supports subsequent causal modeling and collaborative analysis.

[0074] The Dynamic Time Warping (DTW) algorithm is used for time alignment, which can effectively address real-world industrial problems such as inconsistent sampling frequencies of equipment in different processes and uneven data upload delays. It avoids mis-associations caused by time misalignment. The structured processing transforms heterogeneous data (such as image defect labels, log text, and numerical parameters) into a unified semantic format, realizing the fusion expression of multimodal information. The generated single-board-level time series data sequence not only retains the time evolution characteristics but also explicitly expresses the logical sequence of processes, providing a structured and high-fidelity analytical foundation for subsequent causal modeling and quality traceability.

[0075] S3. Utilize historical production single-board level time-series data to analyze the quality correlation characteristics between preceding and subsequent processes, and establish a process influence coupling model based on causal test results.

[0076] Furthermore, key process parameters and subsequent quality inspection results in the single-board level time sequence data of historical production are paired to form process input-output data pairs.

[0077] Based on the data, the Pearson correlation coefficient between the parameters of the preceding process and the quality indicators of the following process was calculated, and statistically significant correlations were selected through significance testing.

[0078] Under controlled conditions of ambient temperature and humidity, equipment aging degree and material batch fluctuation, Granger causality test is performed on variable pairs that pass the correlation test to determine whether there is a unidirectional causal drive from the preceding process to the subsequent result.

[0079] For process pairs that pass the causality test, record their F-statistics as an initial causal strength reference and mark the causal direction.

[0080] Organize all process pairs with causal relationships into a directed graph structure, where nodes represent processes and directed edges represent causal influence paths.

[0081] Each directed edge is assigned a weight, which is determined by the causal strength and the correlation coefficient, thus forming a process influence coupling model.

[0082] The process influence coupling model is used to quantify the dynamic influence of each process on the subsequent quality status.

[0083] By combining Pearson correlation screening and Granger causality test, statistically significant and directional inter-process influence relationships can be identified under the premise of controlling for external interference factors, avoiding misjudging correlation as causality. The process influence coupling model not only reflects direct influences, but also transmits indirect influences through graph structure, enhancing the ability to characterize multi-level coupling effects in complex manufacturing systems. As the core input for subsequent weight calculation and collaborative analysis, this model provides key support for realizing the paradigm shift from data-driven to mechanism-enhanced analysis.

[0084] S4. Based on the output of the process influence coupling model and the credibility level of the data source equipment for each process, calculate the data fusion weight of each process in the current production batch.

[0085] Furthermore, the influence intensity value of each process pointing to all subsequent processes is extracted from the process influence coupling model, and these values ​​are summed to obtain the comprehensive influence assessment value of the process.

[0086] The system acquires historical operating data integrity, calibration cycle compliance, and sensor stability information of the production equipment connected to each process, and generates equipment reliability levels based on preset scoring rules.

[0087] The equipment reliability level is mapped to a data reliability factor between 0.8 and 1.2, reflecting the reliability level of the process data.

[0088] The original fusion score is obtained by multiplying the comprehensive impact assessment value of each process with its corresponding data credibility factor.

[0089] The original fusion scores of all processes are normalized so that the sum is 1, thus obtaining the data fusion weight of each process in the current production batch.

[0090] Data fusion weights are used to reflect the differences in the importance of data from different processes during the subsequent weighted integration process.

[0091] By introducing equipment reliability levels to quantitatively assess the quality of data sources, the interference of low-reliability equipment data on the fusion results is avoided, and the rationality of weight allocation is improved. The comprehensive impact assessment value reflects the structural role of the process in the process flow, while the data reliability factor reflects the accuracy of the data itself. The fusion of the two realizes a dual consideration of capability and reliability. The calculated data fusion weights are dynamically adaptable and can be adjusted according to changes in equipment status, thereby enhancing the system's responsiveness to real production fluctuations.

[0092] S5. Based on the data fusion weight, the quality characteristics of each process in the current batch are weighted and integrated to generate a comprehensive quality characterization vector that reflects the collaborative status of the entire process.

[0093] Furthermore, key quality characteristic values ​​for each process are extracted from the single-board-level timing data sequence of each PCB board in the current batch. This includes solder paste thickness deviation, mounting offset, number of AOI defects, and electrical test parameter offset values.

[0094] quality characteristic value Data fusion weights with corresponding processes Multiplying them together yields the weighted eigencomponents. .

[0095] Weighted characteristic components of all processes Perform linear summation to generate the overall quality characterization value Z of the PCB board. The calculation expression is as follows: ;in Z represents the total number of processes involved in the analysis, reflecting the cumulative quality deviation of the board throughout the entire production process.

[0096] Perform the above weighted fusion operation on all PCBs in the current batch to obtain a batch-level comprehensive quality characterization vector set.

[0097] The comprehensive quality characterization vector set is used for subsequent defect propagation path modeling and trend identification.

[0098] By multiplying and summing the quality characteristics of each process with their corresponding fusion weights, a quantitative integration of the overall process quality status is achieved. This overcomes the limitations of isolated analysis of process indicators in traditional methods, resulting in a comprehensive quality characterization value. As a single scalar indicator, it can reflect the overall quality level of a single board while retaining the interpretability of the contribution of each process, which facilitates subsequent batch-level statistical analysis and anomaly detection. This vector set provides a high-quality, low-dimensional set of input variables for defect propagation path modeling, improving the efficiency and accuracy of causal discovery.

[0099] S6. Using the key quality variables in the comprehensive quality characterization vector as nodes, construct a defect propagation path diagram through the causal structure learning method, identify the cross-process transmission path from initial process deviation to final functional failure, and locate the root cause process that leads to the defect.

[0100] Furthermore, quality variables with a correlation higher than a preset threshold with the final functional test failure rate are selected from the comprehensive quality characterization vector set to form a key variable set.

[0101] Using a set of key variables as input, the FCI causal discovery algorithm is employed to perform unsupervised structural learning under the condition of potential confounding factors, and to identify directed causal edges between variables.

[0102] Calculate the conduction strength for each identified directed causal edge. This strength is determined jointly based on the p-value of the conditional independence test and the residual partial correlation coefficient, and its expression is: ;in, For variables and Controlling the parent node set p-value for conditional independence test under given conditions This is the partial correlation coefficient of the residuals.

[0103] All directed causal edges and their strength values The organization is represented by a defect propagation path diagram, where nodes represent quality variables and directed edges represent the direction of causal transmission.

[0104] Starting from the final functional failure point, trace back along the path of maximum cumulative intensity to identify the earliest starting point where abnormal fluctuations occurred.

[0105] Map the starting node back to its corresponding physical process to identify the process that caused the defect.

[0106] The defect propagation path diagram supports visualization, making it easier for engineers to understand the defect evolution mechanism.

[0107] The FCI algorithm is used to construct a defect propagation path map, which can maintain robust identification of the true causal structure even under real-world conditions with unobserved confounding factors (such as environmental fluctuations and material differences), and avoid the generation of false paths. The edge strength formula, which is jointly defined by p-value and partial correlation coefficient, can effectively distinguish between strong and weak causal relationships, improve the accuracy of path credibility ranking, and the reverse tracing mechanism combined with the maximum cumulative strength path strategy can accurately locate the source process of defects, provide engineers with clear root cause analysis conclusions, and significantly shorten the problem investigation cycle.

[0108] S7. Based on the impact path weights of each process in the defect propagation path diagram and combined with its quality fluctuation characteristics, evaluate the contribution of each process to the final yield, identify the dynamic bottleneck process in the current production state, generate process optimization instructions for the process, and feed them back to the production equipment through the manufacturing execution system to achieve closed-loop collaborative control of the production process.

[0109] Furthermore, all complete propagation paths from initial deviation to final failure are enumerated in the defect propagation path diagram, forming a path set.

[0110] The frequency of each process in all transmission paths is counted and used as a measure of its path coverage.

[0111] For each path P, calculate its overall credibility. This confidence level is determined by the strength values ​​of all edges in the path. The geometric mean is determined by the following expression: ;in, Let P be the number of edges in path P.

[0112] For each path P containing a certain process i, calculate its local contribution value, which is composed of path frequency and path reliability. Reciprocal of path length Multiply them to get the result.

[0113] The path weight of process i is obtained by summing all local contribution values. Its expression is: ;in, Let i be the set of all paths that contain process i.

[0114] Calculate the standard deviation of key indicators for each process in recent consecutive production batches to obtain the quality stability index. .

[0115] Apply path weights With quality stability indicators Multiply to obtain the influence contribution. Its expression is: .

[0116] Impact contribution It comprehensively reflects the structural impact of each process on the final yield and process stability.

[0117] Trend analysis was performed on the impact contribution sequence, and processes with high contribution and an upward trend were identified as dynamic bottleneck processes.

[0118] The identified dynamic bottleneck process types and their changing trends in impact contribution are matched with historical cases in the process optimization knowledge base.

[0119] Extract effective parameter adjustment strategies from successfully matched cases, including adjusting parameter names, adjustment directions, and target value ranges.

[0120] The adjustment strategy is converted into a control instruction package that conforms to the OPC UA protocol specification.

[0121] The manufacturing execution system sends control instruction packages to the production equipment controllers of the corresponding processes.

[0122] Key quality indicators and final functional test pass rates for this process are collected in at least three subsequent consecutive production batches to form an optimized dataset.

[0123] The optimized dataset is compared with the unoptimized dataset to evaluate the optimization effect.

[0124] The significance of the difference between the means of the two groups of yield data was assessed using a two-sample t-test. The t-value was calculated and expressed as follows: ;in, The average yield before and after optimization are shown below. Standard deviation, This represents the sample size.

[0125] If the t-test result is significant and the improvement is greater than the preset threshold, the optimization is confirmed to be effective, and the model update process is initiated.

[0126] The process optimization knowledge base is constructed as follows: Collect historical, engineering-verified records of effective anomaly handling during production processes to form an initial case set; extract structured data from each case, including the anomaly pattern, involved processes, adjusted parameters, adjustment direction, implementation batch, and verification results; uniformly encode the anomaly patterns to establish a standardized anomaly code system; store the mapping relationship between the anomaly patterns and effective countermeasures in each case as structured table entries; periodically review and update the mapping relationship table; provide an access interface for the analysis system to support real-time querying and matching; and when performing pattern matching, use a weighted cosine similarity method to calculate the matching degree between the current anomaly pattern vector a and the historical pattern vector b in the knowledge base. Its expression is: ;in, The weights for the k-th dimension features are set by expert experience; the process optimization knowledge base provides strategy support for closed-loop control, improving optimization response speed and success rate.

[0127] By using a multi-dimensional contribution assessment weighted by path frequency, credibility, and length, the impact on processes is systematically quantified, avoiding assessment bias caused by relying on a single indicator. The contribution assessment integrates structural impact and process stability, enabling the identification of truly bottleneck processes that are both critical and subject to fluctuation risks. The closed-loop control mechanism, combined with t-test verification and knowledge base matching, ensures that the optimization strategy is scientific, verifiable, and reusable, forming a complete intelligent closed loop of analysis-decision-execution-verification-learning, and possessing continuous optimization capabilities.

[0128] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0129] The embodiments of this application also provide a circuit board manufacturing apparatus, which can be referred to in conjunction with the circuit board manufacturing method provided in the embodiments of this application.

[0130] Please refer to Figure 3 The device includes the following modules.

[0131] The data acquisition module 101 is used to acquire the single-board time sequence data sequence corresponding to the entire production line process in the historical production process.

[0132] The causal correlation determination module 102 is used to determine the quality correlation characteristics between the preceding and following processes using the single-board time series data sequence, and to determine the causal verification results between the preceding and following processes based on the quality correlation characteristics.

[0133] The weight determination module 103 is used to determine the fusion weight of each process after establishing the process influence coupling model using the causal test results and combining the process influence coupling model with the credibility of the data source equipment of each process.

[0134] The path graph generation module 104 is used to fuse the quality characteristics of each process in the current batch using fusion weights to obtain a comprehensive quality characterization vector corresponding to the overall process collaborative state, and to construct a defect propagation path graph that matches the comprehensive quality characterization vector.

[0135] Product manufacturing module 105 is used to use a defect propagation path diagram to identify the target process in the production line that is currently causing defects in the circuit board, and to use the production line that optimizes the target process to produce the circuit board.

[0136] The circuit board manufacturing apparatus provided in this application first determines the quality correlation characteristics between preceding and following processes based on the single-board time-series data sequence corresponding to the entire production line during historical production. Then, it determines the causal verification results between these processes based on the quality correlation characteristics. A process influence coupling model can be established based on the causal verification results. This model quantifies the dynamic influence of each process on the subsequent quality state. Therefore, after obtaining the process influence coupling model, the fusion weight of each process can be determined by combining the model with the reliability of the data source equipment for each process. In other words, the causal relationship of each process's influence and the reliability of the data can be expressed based on the fusion weight. Thus, by fusing the quality characteristics of each process in the current batch using the fusion weight, a comprehensive quality characterization vector corresponding to the overall process coordination state can be obtained, and a defect propagation path diagram matching the comprehensive quality characterization vector can be constructed. Finally, using the defect propagation path diagram, the target process causing the circuit board defect can be identified from the production line. This target process can be optimized, and then the production line after optimizing the target process can be used to produce circuit boards, effectively improving the circuit board production capacity.

[0137] This application achieves a systematic quantification of the impact on the process through multi-dimensional contribution assessment, avoiding the assessment bias caused by relying on a single indicator. The contribution of the impact integrates structural impact and process stability, and can identify target processes that are both critical and defective. Optimizing the target process can effectively improve the overall process performance.

[0138] In one specific embodiment of this application, the causal relationship determination module is specifically used to pair key process parameters and subsequent quality inspection results in the single-board time-series data sequence to obtain data pairs of process input and output quality; calculate the correlation coefficient within the data pairs, and use the correlation coefficient to screen out statistically significant target data pairs from the data pairs; under the conditions of controlling environmental temperature and humidity, equipment aging degree, and material batch fluctuations, perform causal tests on the variable pairs of the target data pairs to determine whether there is a unidirectional causal drive from the preceding process to the subsequent result; if there is a unidirectional causal drive, record the significance index corresponding to the target data pair, determine the significance index as the initial causal strength, and mark the causal direction; determine the process pairs with causal relationship, the initial causal strength corresponding to the process pairs, and the causal direction corresponding to the process pairs as the causal test results.

[0139] In one specific embodiment of this application, the weight determination module is specifically used to create nodes using the processes in the process pair; connect the nodes using causal direction and directed edges; and assign values ​​to the directed edges using initial causal strength and correlation coefficient to obtain the process influence coupling model.

[0140] In one specific embodiment of this application, the weight determination module is specifically used to extract the influence intensity values ​​of each process pointing to all subsequent processes from the process influence coupling model, and sum all the influence intensity values ​​corresponding to a single process to obtain the comprehensive influence assessment value of that single process; obtain the historical operation data integrity, calibration cycle compliance, and sensor stability information of the production equipment connected to each process; determine the equipment reliability level using the historical operation data integrity, calibration cycle compliance, and sensor stability information; map the equipment reliability level to a data reliability factor within a preset numerical range; multiply the comprehensive influence assessment value of each process by the corresponding data reliability factor to obtain the original fusion score; and normalize the original fusion scores of all processes to obtain the fusion weight corresponding to each process.

[0141] In one specific embodiment of this application, the path diagram generation module is specifically used to obtain the current single-board timing data sequence corresponding to the current batch of circuit boards; extract the key quality feature values ​​of each process from the current single-board timing data sequence; multiply the key quality feature values ​​with the fusion weight of the corresponding process to obtain the weighted feature components; and linearly sum the weighted feature components of all processes to generate the comprehensive quality characterization vector of the circuit board.

[0142] In one specific embodiment of this application, the path graph generation module is specifically used to obtain the comprehensive quality representation vector set of the current batch of circuit boards; the comprehensive quality representation vector set includes the comprehensive quality representation vectors corresponding to each circuit board in the current batch; quality variables with a correlation higher than a preset threshold with the final functional test failure rate are selected from the comprehensive quality representation vector set, and the selected quality variables are added to the key variable set; using the key variable set as input, unsupervised structure learning is performed under the condition of potential confounding factors to identify directed causal edges between variables; and the conduction strength of each identified directed causal edge is calculated. The formula for calculating conduction strength is as follows: ; For variables i and j in the control parent node set p-value for conditional independence test under given conditions The partial correlation coefficient is the residual; using directed causal edges and their strength values. Construct a defect propagation path graph; the nodes of the defect propagation path graph correspond to quality variables, and the directed edges correspond to the causal propagation direction.

[0143] In one specific embodiment of this application, the data acquisition module is specifically used to collect the original datasets generated by the equipment of each process in the production line during the production process; based on the unique identifier of a single circuit board and the material feeding and discharging time events of each process, the original datasets are time-aligned and structured to obtain a single-board time sequence data sequence.

[0144] In one specific embodiment of this application, the method further includes: a bottleneck process identification module, used to enumerate all complete transmission paths from initial deviation to final failure in the defect propagation path diagram to obtain a path set; count the frequency of each process in all transmission paths, and determine the frequency as its path coverage metric; and calculate the overall reliability of each path P. The formula for calculating the overall credibility is: ; Let P be the number of edges. Let the strength value of the edge in the path be denoted by ; for each path P containing a certain process i, calculate its local contribution value; the formula for calculating the local contribution value is path frequency and path reliability. Reciprocal of path length The product of the values; summing all local contribution values ​​yields the action path weight of process i. The formula for calculating the action path weight is as follows: ; Given the set of all paths containing process i; calculate the standard deviation of key performance indicators for each process in recent consecutive production batches to obtain the quality stability index. ; weight the action path With the aforementioned quality stability index Multiply to obtain the influence contribution. Trend analysis is performed on the contribution of the impact, and the processes with high contribution and an upward trend are identified as dynamic bottleneck processes.

[0145] In one specific embodiment of this application, the system further includes: a process optimization closed-loop module, used to search for matching cases from historical cases in the process optimization knowledge base based on the process type and the changing trend of the impact contribution of the dynamic bottleneck process; extract effective parameter adjustment strategies from the matching cases; convert the parameter adjustment strategies into control instruction packages; send the control instruction packages to the production equipment controller of the corresponding process using the manufacturing execution system; collect key quality indicators and final function test pass rates of the process in several subsequent continuous production batches to form an optimized dataset; compare the optimized dataset with the unoptimized dataset to obtain an evaluation of the optimization effect; and determine that the optimization is effective if the evaluation of the optimization effect passes the significance assessment and the improvement is greater than a preset threshold.

[0146] For a description of the features in the embodiment corresponding to the circuit board manufacturing apparatus, please refer to the relevant description in the embodiment corresponding to the circuit board manufacturing method, which will not be repeated here.

[0147] Corresponding to the above method embodiments, this application also provides an electronic device. The electronic device described below and the circuit board manufacturing method described above can be referred to each other.

[0148] See Figure 4 As shown, the electronic device includes:

[0149] Memory 332 is used to store computer programs;

[0150] The processor 322 is used to execute a computer program to implement the steps of the circuit board manufacturing method described in the above method embodiments.

[0151] For details, please refer to Figure 5 , Figure 5 This is a schematic diagram of the specific structure of an electronic device provided in this embodiment. The electronic device can vary significantly due to differences in configuration or performance. It may include one or more central processing units (CPUs) (e.g., one or more processors) and a memory 332. The memory 332 stores one or more computer programs 342 or data 344. The memory 332 can be temporary or permanent storage. The program stored in the memory 332 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the data processing device. Furthermore, the processor 322 may be configured to communicate with the memory 332 and execute the series of instruction operations stored in the memory 332 on the electronic device 301.

[0152] Electronic device 301 may also include one or more power supplies 326, one or more wired or wireless network interfaces 350, one or more input / output interfaces 358, and / or one or more operating systems 341.

[0153] The steps in the circuit board manufacturing method described above can be implemented by the structure of the electronic device.

[0154] Corresponding to the above method embodiments, this application also provides a readable storage medium. The readable storage medium described below corresponds to the circuit board manufacturing method described above. This application also provides a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps of any of the circuit board manufacturing method embodiments described above when run.

[0155] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0156] An embodiment of this application also provides a computer program circuit board, which includes a computer program that, when executed by a processor, implements the steps in any of the circuit board manufacturing method embodiments described above.

[0157] Embodiments of this application also provide another computer program circuit board, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the circuit board manufacturing method embodiments described above.

[0158] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0159] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only intended to help understand the methods and core ideas of this application. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for manufacturing a circuit board, characterized in that, include: Obtain the single-board timing data sequence corresponding to the entire production line process during the historical production process; The single-board timing data sequence is the data generated, collected, and evaluated by multiple processes throughout the entire production process when the circuit board is produced on the production line. The quality correlation characteristics between the preceding and following processes are determined using the single-board timing data sequence, and the causal verification results between the preceding and following processes are determined based on the quality correlation characteristics. After establishing a process influence coupling model using the causal test results, the fusion weight of each process is determined by combining the process influence coupling model with the credibility of the data source equipment for each process. The quality characteristics of each process in the current batch are fused using the fusion weights to obtain a comprehensive quality characterization vector corresponding to the overall collaborative state of the process, and a defect propagation path diagram matching the comprehensive quality characterization vector is constructed. Using the defect propagation path diagram, the target process currently causing the circuit board defect is identified from the production line, and the circuit board is produced using the production line after optimizing the target process. The process of determining the quality correlation characteristics between preceding and following processes using the single-board timing data sequence, and determining the causal verification results between preceding and following processes based on the quality correlation characteristics, includes: The key process parameters and subsequent quality inspection results in the single-board timing data sequence are paired to obtain data pairs of process input and output quality. Calculate the correlation coefficient within the data pairs, and use the correlation coefficient to filter out statistically significant target data pairs from the data pairs; Under controlled environmental temperature and humidity, equipment aging degree and material batch fluctuations, causal tests are performed on the variable pairs of the target data pair to determine whether there is a unidirectional causal drive from the preceding process to the subsequent result. If a one-way causal drive exists, record the significant index corresponding to the target data pair, determine the significant index as the initial causal strength, and mark the causal direction; The causal test result is determined by the initial causal strength and the causal direction corresponding to the process pair that have a causal relationship.

2. The method according to claim 1, characterized in that, A process influence coupling model is established using the causal test results, including: Create nodes using the processes in the process pair; The nodes are connected using the causal direction and directed edges. The directed edges are assigned values ​​using the initial causal strength and the correlation coefficient to obtain the process influence coupling model.

3. The method according to claim 1, characterized in that, Based on the process influence coupling model and the reliability of the data source equipment for each process, the fusion weights of each process are determined, including: The influence intensity values ​​of each process pointing to all subsequent processes are extracted from the process influence coupling model, and all the influence intensity values ​​corresponding to a single process are summed to obtain the comprehensive influence assessment value of the single process. Obtain information on the integrity of historical operating data, compliance of calibration cycles, and stability of sensors from the production equipment connected to each process. The device reliability level is determined by utilizing the integrity of the historical operating data, the compliance of the calibration cycle, and the sensor stability information. Map the device credibility level to a data credibility factor within a preset numerical range; The overall impact assessment value of each process is multiplied by the corresponding data credibility factor to obtain the original fusion score; The original fusion scores of all processes are normalized to obtain the fusion weights corresponding to each process.

4. The method according to claim 1, characterized in that, The quality characteristics of each process in the current batch are fused using the fusion weights to obtain a comprehensive quality representation vector corresponding to the overall collaborative state of the process, including: Obtain the timing data sequence of the current single board corresponding to the current batch of circuit boards; Extract key quality characteristic values ​​for each process from the current single-board timing data sequence; The key quality feature values ​​are multiplied by the fusion weights of the corresponding processes to obtain weighted feature components; The weighted feature components of all processes are linearly summed to generate the comprehensive quality characterization vector of the circuit board.

5. The method according to claim 1, characterized in that, Constructing a defect propagation path graph that matches the comprehensive quality characterization vector includes: Obtain the comprehensive quality characterization vector set of the current batch of circuit boards; the comprehensive quality characterization vector set includes the comprehensive quality characterization vector corresponding to each circuit board in the current batch. Quality variables that are correlated with the final functional test failure rate higher than a preset threshold are selected from the comprehensive quality characterization vector set, and the selected quality variables are added to the key variable set. Using the set of key variables as input, unsupervised structural learning is performed under the condition of potential confounding factors to identify directed causal edges between variables; Calculate the conduction strength for each identified directed causal edge. The formula for calculating conduction strength is as follows: ; For variables and Controlling the parent node set p-value for conditional independence test under given conditions This is the partial correlation coefficient of the residuals; Using directed causal edges and their strength values Construct the defect propagation path graph; the nodes of the defect propagation path graph correspond to quality variables, and the directed edges correspond to the causal propagation direction.

6. The method according to claim 1, characterized in that, Obtain the single-board time sequence data corresponding to the entire production line process during historical production, including: Collect the raw datasets generated by the equipment in each process of the production line during the production process; Based on the unique identifier of a single circuit board and the material input / output time events of each process, the original dataset is time-aligned and structured to obtain the single-board timing data sequence.

7. The method according to any one of claims 1 to 6, characterized in that, Also includes: Enumerate all complete propagation paths from initial deviation to final failure in the defect propagation path diagram to obtain the path set; The frequency of each process appearing in all transmission paths is counted, and the frequency is determined as its path coverage metric. For each path P, calculate its overall credibility. The formula for calculating the overall credibility is: ; Let P be the number of edges. This represents the strength value of the edges in the path; For each path P containing a certain process i, calculate its local contribution value; the formula for calculating the local contribution value is path frequency and path reliability. Reciprocal of path length The product; The path weight of process i is obtained by summing all local contribution values. The formula for calculating the action path weight is as follows: ; It is the set of all paths that contain process i; Calculate the standard deviation of key indicators for each process in recent consecutive production batches to obtain the quality stability index. ; The action path weight With the aforementioned quality stability index Multiply to obtain the influence contribution. ; A trend analysis is performed on the contribution of the impact, and the processes with high contribution and an upward trend are identified as dynamic bottleneck processes.

8. The method according to claim 7, characterized in that, Also includes: Based on the process type and the changing trend of its impact contribution of the dynamic bottleneck process, matching cases are searched from historical cases in the process optimization knowledge base; Extract effective parameter adjustment strategies from the matching cases; The parameter adjustment strategy is converted into a control instruction package; The control instruction package is sent to the production equipment controller of the corresponding process using the manufacturing execution system; Key quality indicators and final functional test pass rates for this process were collected in several subsequent consecutive production batches to form an optimized dataset; The optimized dataset is compared with the unoptimized dataset to evaluate the optimization effect; If the optimization effect passes the significance assessment and the improvement is greater than the preset threshold, then the optimization is determined to be effective.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the circuit board manufacturing method as described in any one of claims 1 to 8.

Images