A real-time monitoring method and system for a circuit board intelligent production line

Abstract

The application discloses a kind of real-time monitoring method and system of circuit board intelligent production line, it is related to production monitoring technical field.A kind of real-time monitoring system of circuit board intelligent production line, it includes: production line monitoring module and production line optimization adjustment module.The application is integrated by multimodal production data and time series analysis, captures each key index of circuit board production line, so that the state of production line can be dynamically monitored, and potential problems and deviations are identified in time, ensure the stability and efficiency of production process;By longitudinal time axis analysis and horizontal parallel line analysis, the trend change in production process, periodic fluctuation and the performance difference between different production lines can be identified, help to quickly locate weak links in production, help to more accurately identify factors affecting production performance, and realize cross-production line optimization and comparison.

Description

Technical Field

[0001] This invention relates to the field of production monitoring technology, and in particular to a real-time monitoring method and system for intelligent circuit board production lines. Background Technology

[0002] With the continuous improvement of the level of intelligence in manufacturing, the monitoring and optimization of circuit board production lines has become a key link in improving production efficiency and product quality. Fluctuations in production quality during circuit board production are usually the result of the interplay of multiple factors, including equipment performance, environmental conditions, raw material quality, and human factors. However, traditional circuit board production monitoring methods typically rely on a single data source and lack the ability to perform time-series comparative analysis of multiple factors in the production process. Due to the lack of effective time-series comparative analysis, these methods struggle to identify potential problems in the production process in a timely manner, resulting in the inability to detect and resolve abnormal fluctuations in production at an early stage, thereby affecting production efficiency and product quality.

[0003] Therefore, there is an urgent need for a monitoring method for circuit board production lines that can compare and analyze multimodal circuit board production line data under different time series, monitor the performance fluctuations of circuit board production lines in real time, and perform intelligent optimization. Summary of the Invention

[0004] This invention aims to provide a real-time monitoring method and system for intelligent circuit board production lines, which compares and analyzes multimodal circuit board production line data under different time sequences, monitors the performance fluctuations of the circuit board production line in real time, and performs intelligent optimization.

[0005] A method for real-time monitoring of a smart circuit board production line includes the following steps:

[0006] The multimodal production data package for circuit boards is configured to include circuit board fingerprint data, production auxiliary material data, circuit board production environment data, and production manpower data.

[0007] For the same production line, longitudinal time axis analysis is performed. Corresponding multimodal production data of circuit boards are collected at different time points within a preset time period to obtain a longitudinal circuit board production dataset. Based on the longitudinal circuit board production dataset, analysis is performed to identify the performance of the circuit board production line and the auxiliary performance of circuit board production.

[0008] A horizontal parallel line analysis is performed on different production lines to obtain multimodal production data of multiple circuit boards at the same point in time, resulting in a horizontal circuit board production dataset. Based on the horizontal circuit board production dataset, analysis is conducted to identify circuit board production difference indicators and circuit board production line abnormal indicators.

[0009] A dynamic digital ecosystem model for the circuit board production line is established based on the performance of the circuit board production line, the auxiliary performance of circuit board production, the difference indicators of circuit board production, and the abnormal indicators of the circuit board production line. The circuit board production line is monitored and adjusted based on the dynamic digital ecosystem model to intelligently optimize the real-time status of the circuit board production line.

[0010] As a preferred technical solution of the present invention, the specific steps for analyzing and identifying the performance of the circuit board production line and the auxiliary performance of circuit board production based on the longitudinal circuit board production dataset include:

[0011] Establish a vertical time axis model within a preset time period;

[0012] Based on the longitudinal time axis model, circuit board feature identification is performed on all circuit board fingerprint data in the longitudinal circuit board production dataset to obtain the circuit board production performance trend of the same production line at different time points.

[0013] Simultaneously, abnormal fluctuation areas of all production auxiliary material data, circuit board production environment data, and production manpower data in the longitudinal circuit board production dataset are extracted on the longitudinal time axis model to obtain the circuit board production abnormal fusion data sequence.

[0014] Based on the longitudinal time axis model, the circuit board production performance trend and circuit board production anomaly fusion data sequence are correlated and analyzed to obtain the circuit board-performance drift prediction model; based on the circuit board-performance drift prediction model, the circuit board production line performance and circuit board production auxiliary performance are output.

[0015] As a preferred technical solution of the present invention, the specific steps for analyzing horizontal circuit board production datasets and identifying circuit board production variation indicators and circuit board production line anomaly indicators include:

[0016] Establish a horizontal production line benchmark model based on the production line numbers corresponding to different production lines;

[0017] Based on the horizontal production line benchmark model, circuit board feature identification is performed on all circuit board fingerprint data in the horizontal circuit board production dataset to obtain differences in circuit board production performance.

[0018] Meanwhile, the cross-difference values ​​of all production auxiliary material data, circuit board production environment data and production manpower data in the horizontal circuit board production dataset are extracted in the horizontal production line benchmark model to obtain the differences in circuit board auxiliary performance.

[0019] Machine learning clustering analysis was performed based on differences in circuit board production performance and differences in circuit board auxiliary performance to obtain circuit board production difference indicators and circuit board production line anomaly indicators.

[0020] As a preferred embodiment of the present invention, the specific steps for establishing a dynamic digital ecosystem model of a circuit board production line include:

[0021] Based on the timestamps and production line numbers of the time points, the performance of the circuit board production line, the auxiliary performance of circuit board production, the difference indicators of circuit board production, and the abnormal indicators of circuit board production line are aligned and preprocessed to obtain two-dimensional circuit board production line monitoring data.

[0022] Entity extraction is performed on the monitoring data of a two-dimensional circuit board production line using a digital twin graph network, and a circuit board production line vector containing production line unit edges and production line unit nodes is constructed.

[0023] A dynamic digital ecosystem model of the circuit board production line is established by dynamically updating the feature representations of the unit edges and unit nodes of the production line using a dynamic graph neural network.

[0024] As a preferred technical solution of the present invention, the specific steps for monitoring and adjusting the circuit board production line based on the dynamic digital ecosystem model include:

[0025] Based on the dynamic digital ecosystem model, causal reasoning is performed on the pre-trained knowledge graph of circuit board production to obtain causal pairs of production factors and circuit board performance.

[0026] By identifying production factors with significant differences in production performance and optimizing the causal relationship between circuit board performance, intelligent optimization of the circuit board production line can be achieved.

[0027] As a preferred technical solution of the present invention, the basic model for constructing the dynamic digital ecosystem model is a digital twin network and a graph neural network model.

[0028] A real-time monitoring system for a smart circuit board production line includes:

[0029] The production line monitoring module includes a monitoring data unit, a vertical production line monitoring unit, and a horizontal production line monitoring unit. The monitoring data unit is used to set up multimodal production data packages for circuit boards, containing circuit board fingerprint data, production auxiliary material data, circuit board production environment data, and production manpower data. The vertical production line monitoring unit performs vertical time-axis analysis on the same production line, collecting corresponding multimodal production data of circuit boards at different time points within a preset time period to obtain a vertical circuit board production dataset. Analysis based on the vertical circuit board production dataset identifies the performance of the circuit board production line and the auxiliary performance of circuit board production. The horizontal production line monitoring unit performs horizontal parallel line analysis on different production lines, acquiring multiple circuit board multimodal production data points corresponding to the same time point to obtain a horizontal circuit board production dataset. Analysis based on the horizontal circuit board production dataset identifies circuit board production difference indicators and circuit board production line abnormal indicators.

[0030] The production line optimization and adjustment module includes a monitoring and intelligent optimization unit. The monitoring and intelligent optimization unit is used to establish a dynamic digital ecosystem model of the circuit board production line based on the performance of the circuit board production line, the auxiliary performance of circuit board production, the difference indicators of circuit board production, and the abnormal indicators of the circuit board production line. Based on the dynamic digital ecosystem model, the circuit board production line is monitored and adjusted to intelligently optimize the real-time status of the circuit board production line.

[0031] The present invention has the following advantages:

[0032] 1. This invention, through the integration and time-series analysis of multimodal production data, can comprehensively capture various key indicators of the circuit board production line, such as production performance, auxiliary performance, production variation indicators, and abnormal indicators. The fusion and real-time updates of this data enable dynamic monitoring of the production line status, timely identification of potential problems and deviations, and ensure the stability and efficiency of the production process. Through longitudinal time axis analysis and horizontal parallel line analysis, it can identify trend changes, periodic fluctuations, and performance differences between different production lines in the production process, helping to quickly locate weak links in production. This multi-dimensional analysis method helps to more accurately identify factors affecting production performance and achieve cross-production line optimization and comparison.

[0033] 2. This invention constructs a dynamic digital ecosystem model that deeply correlates the performance of the production line with other elements of the production process. It can update the characteristic representation of the production line unit in real time, thereby dynamically adjusting the production process, equipment configuration and operation strategy to achieve intelligent optimization. Combining causal reasoning and machine learning technology, this model can not only identify abnormal factors in production, but also predict possible performance changes, providing a scientific basis for intelligent scheduling and resource optimization of the production line. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the structure of a real-time monitoring system for an intelligent production line of circuit boards used in an embodiment of the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0036] Example 1: A method for real-time monitoring of a smart circuit board production line, comprising the following steps:

[0037] The multimodal production data package for circuit boards is configured to include circuit board fingerprint data, production auxiliary material data, circuit board production environment data, and production manpower data.

[0038] Circuit board fingerprint data refers to the unique identifier obtained by collecting the microstructural features of a circuit board. It includes solder joint quality, component installation status, electrical performance test results, and the type and location of circuit board defects. This data is usually acquired through high-precision image recognition, online inspection equipment (such as automated optical inspection equipment), and electrical testing equipment. It can reflect the manufacturing quality and functionality of the circuit board and help identify potential production defects or performance deviations.

[0039] Production auxiliary material data involves data on various consumables used in the circuit board production process, including the quality and usage of solder paste, glue, flux, cleaning agents, etc. In particular, information such as the decay of the physicochemical activity of these materials, concentration changes, and impurity accumulation directly affect the production quality of the circuit board. Production auxiliary material data usually comes from quality control records in the material supply database and real-time monitoring of the environment and material usage by various sensors (such as temperature, humidity, and pressure sensors) during the production process.

[0040] Circuit board production environment data describes the changes in the environment surrounding the production line, mainly including data on temperature, humidity, air flow, electromagnetic interference, vibration frequency, etc. These data are acquired by sensors installed in the production environment and reflect the impact of the external environment on the manufacturing quality of circuit boards during the production process. For example, fluctuations in temperature and humidity may lead to changes in soldering quality, and electromagnetic interference in the air may affect the accuracy of electronic components.

[0041] Production human resources data refers to data related to operators, including but not limited to operators' cognitive load, concentration, fatigue status, working hours, non-standardized operating behaviors, and verbal communication content. This data can be collected through employee wearable devices, voice recognition systems, or human resource management systems to help assess the impact of human factors on production quality. Through this data, production deviations caused by improper operation or fatigue can be identified, thereby enabling corresponding optimization measures to be taken.

[0042] Since the aforementioned circuit board fingerprint data, production auxiliary material data, circuit board production environment data, and production manpower data are multimodal data with different data structures, in order to address the heterogeneity of the data, before analyzing and processing the above data, data preprocessing and standardization are used to convert different data types into a unified format before feature analysis and feature recognition are performed.

[0043] For the same production line, a longitudinal time axis analysis is performed. Corresponding multimodal production data of circuit boards are collected at different time points within a preset time period to obtain a longitudinal circuit board production dataset. The longitudinal circuit board production dataset is analyzed to identify the performance of the circuit board production line and the auxiliary performance of circuit board production. The preset time period is set manually.

[0044] A longitudinal timeline is a model constructed by arranging the operational processes and event data of a production line at different points in time. It aims to display and track changes in data over time during the production process. A longitudinal timeline clearly presents the long-term trends, periodic fluctuations, and unexpected events of the production line, thereby helping to analyze the regularity of production activities and performance changes within different time periods.

[0045] Based on the analysis of longitudinal PCB production datasets, specific steps were identified to determine the performance of PCB production lines and auxiliary performance aspects of PCB production, including:

[0046] A vertical time axis model is established within a preset time period. The purpose of establishing the vertical time axis model is to connect all time points in the production process and provide a continuous time frame for the data at each time point. Through this model, the state changes of the circuit board production line at different time points can be accurately described, and the foundation for subsequent data analysis can be laid.

[0047] Based on the longitudinal time axis model, circuit board feature identification is performed on all circuit board fingerprint data in the longitudinal circuit board production dataset to obtain the circuit board production performance trend of the same production line at different time points. The circuit board fingerprint data includes information such as the solder joint quality, component status, and electrical performance of the circuit board. By analyzing these features, the production performance trend of the circuit board at different time points can be revealed. This analysis helps us identify the quality fluctuation trend in the production process and the time periods in which the production performance of the circuit board has changed significantly.

[0048] The specific steps for feature recognition are as follows: First, use a convolutional neural network to extract image features from the image data in the circuit board fingerprint data, thereby identifying information such as solder joint quality, component mounting location, and surface defects. Second, use statistical feature extraction methods (such as mean, standard deviation, kurtosis, skewness, etc.) and frequency domain analysis (such as Fourier transform) to extract electrical performance characteristics of the circuit board, such as resistance, capacitance, and frequency response, from the electrical test data or other numerical data in the image data of the circuit board fingerprint data. Third, use these extracted features to perform time series analysis to reveal the performance trends during the circuit board production process. Fourth, by performing time series analysis on the circuit board fingerprint features at each time point, quality fluctuations and trend changes during the production process can be discovered. For example, a sliding window method can be used to capture feature changes within a specific time period, or deep learning methods such as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs) can be used to model time series data, predict the circuit board performance over a future period, and principal component analysis can be used to select the most representative and predictive features from all extracted features to obtain the circuit board production performance trend.

[0049] Simultaneously, abnormal fluctuation areas of all production auxiliary material data, circuit board production environment data, and production manpower data in the longitudinal circuit board production dataset are extracted on the longitudinal time axis model to obtain the circuit board production abnormal fusion data sequence.

[0050] Historical data on circuit board production is collected. Based on this data and the normal fluctuation range during production, a baseline range for normal circuit board production is established. Based on this baseline range, criteria for identifying anomalies are set, resulting in an abnormal fluctuation zone. This abnormal fluctuation zone includes both beneficial and detrimental anomalies. Beneficial anomalies typically refer to beneficial changes or trends occurring during production, which may lead to quality improvements, efficiency optimization, or new performance breakthroughs. Specific criteria include: when abnormal fluctuations during circuit board production are related to improvements in production efficiency or quality, such as improved solder joint quality, faster production speed, or higher electrical performance standards, these fluctuations represent innovation or optimization in the production process, such as new material formulations or the introduction of new production processes. Slight fluctuations in the production environment (such as temperature and humidity) that improve the product's adaptability or stability can also be considered beneficial anomalies; for example, under specific temperature conditions, the soldering quality of the circuit board may actually improve.

[0051] Unproductive production anomalies refer to fluctuations that may negatively impact the production process, product quality, or production efficiency. These anomalies typically lead to production interruptions, quality problems, or resource waste. Specific criteria include: any abnormal fluctuations that cause a decline in circuit board quality, such as soldering defects, component damage, or substandard electrical performance, are considered unproductive production anomalies. These fluctuations affect product functionality and reliability, and may even lead to significant rework or scrap. If abnormal fluctuations are related to a decline in production efficiency, such as extended equipment downtime or reduced production speed, these fluctuations usually indicate production line instability and may result in reduced output.

[0052] By extracting these abnormal data, a circuit board production anomaly fusion data sequence can be formed, which helps to deeply understand the impact of different factors on the production line status.

[0053] Based on the longitudinal time axis model, the circuit board production performance trend and circuit board production anomaly fusion data sequence are correlated and analyzed to obtain the circuit board-performance drift prediction model; based on the circuit board-performance drift prediction model, the circuit board production line performance and circuit board production auxiliary performance are output;

[0054] When conducting analysis based on a longitudinal time-axis model, the first step is to perform correlation analysis between the circuit board production performance trend and the circuit board production anomaly fusion data sequence. The core of this step is to synchronously compare the circuit board production performance data collected at different time points with the abnormal fluctuation data detected during the production process, analyzing the correlation between the two. Through in-depth analysis of this data, it is possible to identify which changes in production performance are directly related to abnormal fluctuations, such as whether a quality decline at a specific time point is related to equipment failure, environmental fluctuations, or material quality issues. This correlation analysis can reveal the performance drift trend during the production process, i.e., the pattern of performance change in the time series, thus providing data support for subsequent predictive models. Based on the correlation analysis between circuit boards and performance drift... Based on the analysis results, a circuit board performance drift prediction model can be established. This model, through learning from historical data, can accurately predict the future performance and auxiliary performance trends of the circuit board production line. By comprehensively analyzing historical performance data and abnormal fluctuation sequences, the model can identify potential performance degradation or abnormal fluctuations and predict when quality problems or production efficiency declines may occur. The model's output includes the performance status and auxiliary performance trends of the circuit board production line. The performance of the circuit board production line refers to key indicators reflecting the operating status of equipment and processes during circuit board production, typically including data such as production efficiency, yield rate, defect rate, and throughput. The actual data comes from real-time data collected during the production process, including equipment status, production rate, and finished product inspection results.

[0055] Circuit board production auxiliary performance focuses on the impact of auxiliary materials and environmental conditions used in production on the production process. This mainly includes the consumption of auxiliary materials, ambient temperature and humidity, airflow, and equipment cleanliness. These prediction results can help production managers identify problems in advance and take corresponding measures to make adjustments, thereby avoiding unnecessary downtime or quality problems during production and ensuring the smooth and efficient operation of the production line.

[0056] A horizontal parallel line analysis is performed on different production lines to obtain multimodal production data of multiple circuit boards at the same point in time, resulting in a horizontal circuit board production dataset. Based on the horizontal circuit board production dataset, analysis is conducted to identify circuit board production difference indicators and circuit board production line abnormal indicators.

[0057] Based on the analysis of a horizontal PCB production dataset, the specific steps for identifying PCB production variation indicators and PCB production line anomaly indicators include:

[0058] A horizontal production line benchmark model is established based on the production line numbers corresponding to different production lines. This benchmark model serves to set a standard reference for each production line, helping to analyze the performance differences between different production lines. By analyzing the data of different production lines, they are unified under a common framework, facilitating horizontal comparison. Based on this benchmark model, the data of all production lines can be compared under the same standard, forming the basis for comparative analysis.

[0059] Based on the horizontal production line benchmark model, circuit board feature identification is performed on all circuit board fingerprint data in the horizontal circuit board production dataset to obtain differences in circuit board production performance.

[0060] Circuit board feature identification is performed on all circuit board fingerprint data. The purpose of this step is to identify the differences in production performance between production lines by comparing the circuit board fingerprint data of different production lines. Circuit board fingerprint data reflects the quality characteristics of circuit boards during the production process, such as solder joint quality and component status. These characteristics can help us identify the differences between different production lines under the same process conditions, and thus analyze the differences in production performance.

[0061] Differences in circuit board manufacturing performance refer to variations or inconsistencies in various quality and efficiency indicators of circuit boards across different production lines or production cycles. These differences typically reflect changes in factors such as equipment, processes, materials, environment, or operators during the production process, which may lead to fluctuations in product quality and production efficiency. For example, the proportion of qualified products may differ across different production lines within a given timeframe; or the number or type of defects occurring during the same production process on different production lines may vary, potentially due to differences in equipment condition, operators, or environmental factors; and within the same production cycle, differences in production speed or output may exist across different production lines.

[0062] Meanwhile, the cross-difference values ​​of all production auxiliary material data, circuit board production environment data and production manpower data in the horizontal circuit board production dataset are extracted in the horizontal production line benchmark model to obtain the differences in circuit board auxiliary performance.

[0063] Cross-difference values ​​(CDRs) represent the differences between certain key parameters (such as production auxiliary materials, environmental data, and human resource data) between different production lines or production cycles in a cross-sectional analysis. These differences can reveal variations in the behavior and performance of different production lines under the same or similar conditions. In other words, CDRs identify discrepancies or inconsistencies between different production lines by comparing similar data points at the same point in time or within a time period. For example, in production auxiliary material data, if two production lines use the same type of solder paste, but one production line's solder paste consumption and quality stability are significantly higher than the other, the difference in consumption or quality between the two can be considered a CDR. These CDRs can help identify potential problems affecting production line performance.

[0064] This process involves extracting all auxiliary material data, environmental data, and human resource data from a horizontal PCB production dataset. These data are then cross-integrated within the framework of a horizontal production line benchmark model to identify differences in auxiliary performance during the production process. Analyzing these differences typically involves cross-comparing auxiliary material data, environmental data, and human resource data to identify key differences affecting production line performance. An unsupervised learning clustering algorithm is used to categorize the production lines based on these differences. Finally, by analyzing the differences between these categories, the key factors influencing production line performance—the auxiliary performance differences of the PCBs—are identified.

[0065] Machine learning cluster analysis is performed based on differences in circuit board production performance and auxiliary performance to obtain circuit board production difference indicators and circuit board production line anomaly indicators. Different production differences are classified through machine learning cluster analysis. The purpose of this step is to extract production difference indicators and anomaly indicators through data clustering, thereby identifying potential problems that may affect the normal operation of the production line.

[0066] Circuit board production variation indicators refer to the changes or differences in various performance, quality, or efficiency indicators of circuit boards in different production lines, production cycles, or production processes. These indicators include, but are not limited to, yield rate, defect rate, production efficiency, and soldering quality. By analyzing these variation indicators, the performance differences between different production lines or production cycles can be revealed, thereby identifying potential production problems or optimization opportunities. For example, a low yield rate on a certain production line may be due to equipment problems, environmental factors, or differences in operator skills.

[0067] Circuit board production line anomaly indicators refer to indicators that reflect abnormalities in the production process. They are usually used to identify abnormal states or faults in the production line. These anomaly indicators may involve equipment failure, process deviation, quality instability, etc., which are specifically manifested as a sudden drop in performance or unpredictable fluctuations in the production process. By monitoring these anomaly indicators, problems in production can be detected and eliminated in a timely manner, ensuring the stability and reliability of the production line. For example, if the defect rate of a certain production line suddenly increases, it may be due to equipment failure or changes in environmental conditions.

[0068] Specifically, a clustering algorithm is selected and appropriate parameters are set to determine the K value (the number of clusters). Data is then input into the algorithm for cluster analysis. By calculating the distance between each data point and the cluster center, the algorithm assigns data points to the nearest cluster. After clustering, data within each cluster exhibits high similarity, while significant differences are observed between different clusters. Finally, by analyzing the characteristics of each cluster, patterns in production performance and anomalies across different production lines can be discovered, thereby identifying poorly performing production lines or abnormal production states. Cluster analysis not only helps identify differences in production lines but also reveals potential problems hidden in the production process, providing data support for subsequent production optimization.

[0069] A dynamic digital ecosystem model for a circuit board production line is established based on the performance of the circuit board production line, the auxiliary performance of circuit board production, the difference indicators of circuit board production, and the abnormal indicators of the circuit board production line. The circuit board production line is monitored and adjusted based on the dynamic digital ecosystem model to intelligently optimize the real-time status of the circuit board production line. The basic models for constructing the dynamic digital ecosystem model are digital twin network and graph neural network model.

[0070] The specific steps for establishing a dynamic digital ecosystem model for a circuit board production line include:

[0071] Based on the timestamps and production line numbers of the time points, the performance of the circuit board production line, the auxiliary performance of circuit board production, the difference indicators of circuit board production, and the abnormal indicators of circuit board production line are aligned and preprocessed to obtain two-dimensional circuit board production line monitoring data.

[0072] It should be noted that the input data for this step includes various types of data, such as production performance, auxiliary performance, difference indicators, and anomaly indicators, collected from different production lines at multiple time points. This data is sorted and aligned using timestamps to ensure accurate comparison of data from different production lines at the same time point. The output is a structured two-dimensional dataset, where rows represent different production cycles or time points, and columns represent different production lines and their corresponding performance indicators, auxiliary performance data, and anomaly data. The preprocessing stage may include data cleaning, noise reduction, missing value imputation, and standardization to ensure that the data is processed on the same scale and in the same format. The beneficial effect of this step is that it provides a standardized, aligned two-dimensional dataset for subsequent in-depth analysis, ensuring data consistency and comparability.

[0073] Entity extraction is performed on the monitoring data of a two-dimensional circuit board production line using a digital twin graph network, and a circuit board production line vector containing production line unit edges and production line unit nodes is constructed.

[0074] In this step, the input data is pre-processed and aligned 2D circuit board production line monitoring data, and the output is a graph-structured data representation, including production line unit nodes and production line unit edges. Specifically, production line unit nodes can represent individual production lines, equipment, processes, or production environments, while production line unit edges represent the relationships and interactions between these nodes, such as the dependencies between equipment and processes or the mutual influence between production lines and the production environment. Through entity extraction from the digital twin graph network, the complex relationships between different production lines and production elements can be captured, enabling each element in the production process to be effectively represented in the graph. The beneficial effect of this step is that the graph structure can more intuitively and accurately express the relationships between various elements of the production line, and lay the foundation for subsequent dynamic analysis.

[0075] A dynamic graph neural network is used to dynamically update the feature representations of unit edges and unit nodes of the circuit board production line vector, establishing a dynamic digital ecosystem model of the circuit board production line. In this step, the input is graph structure data extracted through a digital twin graph network, i.e., the circuit board production line vector containing nodes and edges. The graph neural network dynamically updates the feature representations of nodes and edges through a message passing mechanism, comprehensively analyzing production line data (such as production performance, auxiliary material usage, environmental changes, etc.) combined with the interaction relationships between production lines. As the state and characteristics of the production line change over time, the dynamic graph neural network can update these changes in real time, capturing the dynamic changes in performance during the production process. The output is a continuously updated dynamic digital ecosystem model representing the state of the circuit board production line, containing the latest feature representations of unit nodes and unit edges. The beneficial effect of this step is that the dynamic graph neural network can reflect the state changes of the production line in real time, dynamically optimize the production process, and promptly detect potential anomalies or performance problems, thus providing strong support for intelligent monitoring and optimization of the production line.

[0076] It should be noted that the dynamic digital ecosystem model can integrate multi-source data from the production line, such as equipment status, environmental conditions, production efficiency, and quality indicators, and reflect the production line's operating status in real time. Through dynamic analysis of this data, the model can identify potential anomalies or performance deviations in the production line. For example, when the failure rate of a piece of equipment increases, production efficiency decreases, or environmental conditions change, the model can issue timely warnings, reminding operators or managers to take measures to prevent the problem from escalating further. Furthermore, the dynamic digital ecosystem model can digitally map each link of the production line (equipment, personnel, environment, etc.) and describe their interactions, enabling different departments (such as production, quality management, equipment maintenance, etc.) to share data and collaborate based on the same model. When production problems occur, relevant departments can quickly access model data for coordination and resource allocation, thereby improving overall response efficiency. For example, the maintenance team can directly understand the health status of the equipment based on the real-time status of equipment nodes and promptly arrange maintenance or replacement of parts.

[0077] The specific steps for monitoring and adjusting the circuit board production line based on the dynamic digital ecosystem model include:

[0078] Based on the dynamic digital ecosystem model, causal reasoning is performed on the pre-trained knowledge graph of circuit board production to obtain causal pairs of production factors and circuit board performance.

[0079] In this step, the input data includes multimodal data obtained from the production line (such as equipment status, production quality, environmental data, and human resource data), as well as known relationships between production factors and circuit board performance obtained from a pre-trained circuit board production knowledge graph. This knowledge graph integrates the relationships between various factors in the production process and the production results, and forms a knowledge base through patterns learned from historical data. Using this knowledge graph, causal reasoning methods can be used to infer the impact of production factors on circuit board performance. Specifically, using causal reasoning models (such as structural equation modeling, Bayesian networks, etc.), by calculating the causal relationships between different production factors and circuit board performance, inference causal pairs of production factors and circuit board performance are obtained. These inference causal pairs reveal which production factors (such as equipment status, environmental factors, operator behavior, etc.) have a significant impact on circuit board performance under different production conditions. The output is a causal reasoning graph containing causal relationship chains between various production factors and circuit board performance. Through this method, it is possible to accurately identify which factors have a major impact on circuit board production performance, providing a scientific basis for the next step of optimization.

[0080] By identifying production factors that exhibit significant differences in production performance and optimizing the causal relationship between them and circuit board performance, intelligent optimization of the circuit board production line can be achieved.

[0081] In this step, the input data consists of causal pairs between production factors and circuit board performance obtained through causal inference, as well as performance differences and anomalies identified in historical production data. First, based on the causal pairs identified in the causal inference graph, and combined with a comparative analysis of performance differences across different production lines or production cycles, production factors exhibiting significant performance variations under different conditions are identified. For example, equipment failures, material fluctuations, or environmental factors may cause the production performance of a particular production line to be significantly lower than that of other production lines. This method identifies key factors affecting production performance. Based on these factors, optimization algorithms (such as machine learning-based optimization models, heuristic algorithms, or artificial intelligence techniques) are applied to adjust the production process and optimize those production factors affecting performance. For example, if a piece of equipment has a high failure rate, the failure rate can be reduced by optimizing equipment maintenance strategies or adjusting usage methods in the production process, thereby improving the overall performance of the production line. The output is the optimized production line operating parameters and optimization scheme, aimed at improving production performance and efficiency. The beneficial effect of this step is that it enables accurate identification and optimization of problems in the production process based on the relationship between production factors and performance obtained through causal inference, achieving intelligent optimization of the production line and improving the overall operating efficiency and product quality of the circuit board production line.

[0082] In this embodiment, for example, in a circuit board production line, multimodal data was collected through IoT technology and a production management system, including circuit board fingerprint data, production auxiliary material data, environmental data, and production manpower data. The circuit board fingerprint data of production line 1 showed that the soldering defect rate increased from 0.5% to 1.5% during the production cycle; the production auxiliary material data indicated that the effectiveness of the solder paste decreased; the environmental data showed that the temperature fluctuation exceeded 27°C; and the manpower data reflected that after an operator worked for more than 8 hours, the concentration of attention decreased, resulting in an increase in the defect rate.

[0083] Based on this data, a longitudinal timeline model was established within a preset time period, and correlation analysis was conducted, revealing a significant correlation between temperature fluctuations and the increase in soldering defect rate. Then, through horizontal production line analysis, comparisons were made with production lines 2 and 3, revealing a significant performance difference in production line 1, primarily affected by solder paste consumption and ambient temperature fluctuations. Machine learning cluster analysis identified the performance deficiencies of production line 1, and causal reasoning identified temperature, solder paste consumption, and operator fatigue as the main factors affecting production performance. Finally, by adjusting temperature control, optimizing solder paste management, and adjusting operator shifts, the soldering defect rate of production line 1 decreased from 1.5% to 0.7%, significantly improving production line efficiency and product quality.

[0084] Example 2: A real-time monitoring system for an intelligent circuit board production line, see [link / reference] Figure 1 As shown, it includes:

[0085] The production line monitoring module includes a monitoring data unit, a vertical production line monitoring unit, and a horizontal production line monitoring unit. The monitoring data unit is used to set up multimodal production data packages for circuit boards, containing circuit board fingerprint data, production auxiliary material data, circuit board production environment data, and production manpower data. The vertical production line monitoring unit performs vertical time-axis analysis on the same production line, collecting corresponding multimodal production data of circuit boards at different time points within a preset time period to obtain a vertical circuit board production dataset. Analysis based on the vertical circuit board production dataset identifies the performance of the circuit board production line and the auxiliary performance of circuit board production. The horizontal production line monitoring unit performs horizontal parallel line analysis on different production lines, acquiring multiple circuit board multimodal production data points corresponding to the same time point to obtain a horizontal circuit board production dataset. Analysis based on the horizontal circuit board production dataset identifies circuit board production difference indicators and circuit board production line abnormal indicators.

[0086] The production line optimization and adjustment module includes a monitoring and intelligent optimization unit. The monitoring and intelligent optimization unit is used to establish a dynamic digital ecosystem model of the circuit board production line based on the performance of the circuit board production line, the auxiliary performance of circuit board production, the difference indicators of circuit board production, and the abnormal indicators of the circuit board production line. Based on the dynamic digital ecosystem model, the circuit board production line is monitored and adjusted to intelligently optimize the real-time status of the circuit board production line.

[0087] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for real-time monitoring of a smart circuit board production line, characterized in that, Includes the following steps: The multimodal production data package for circuit boards includes circuit board fingerprint data, production auxiliary material data, circuit board production environment data, and production manpower data. Circuit board fingerprint data refers to the unique identifier obtained by collecting the microstructural features of the circuit board. It includes solder joint quality, component installation status, electrical performance test results, and the type and location of defects in the circuit board. It is obtained through image recognition, online inspection equipment, and electrical testing equipment. For the same production line, longitudinal time axis analysis is performed. Corresponding multimodal production data of circuit boards are collected at different time points within a preset time period to obtain a longitudinal circuit board production dataset. Based on the longitudinal circuit board production dataset, analysis is performed to identify the performance of the circuit board production line and the auxiliary performance of circuit board production. A horizontal parallel line analysis is performed on different production lines to obtain multimodal production data of multiple circuit boards at the same point in time, resulting in a horizontal circuit board production dataset. Based on the horizontal circuit board production dataset, analysis is conducted to identify circuit board production difference indicators and circuit board production line abnormal indicators. A dynamic digital ecosystem model of the circuit board production line is established based on its performance, auxiliary performance, production variance indicators, and anomaly indicators. This model is used to monitor and adjust the production line, intelligently optimizing its real-time status. Analysis of the longitudinal circuit board production dataset identifies specific steps for improving production line performance and auxiliary performance, including: Establish a vertical time axis model within a preset time period; Based on the longitudinal time axis model, circuit board feature identification is performed on all circuit board fingerprint data in the longitudinal circuit board production dataset to obtain the circuit board production performance trend of the same production line at different time points. Simultaneously, abnormal fluctuation areas of all production auxiliary material data, circuit board production environment data, and production manpower data in the longitudinal circuit board production dataset are extracted on the longitudinal time axis model to obtain the circuit board production abnormal fusion data sequence. Based on the longitudinal time axis model, the circuit board production performance trend and circuit board production anomaly fusion data sequence are correlated and analyzed to obtain the circuit board-performance drift prediction model; based on the circuit board-performance drift prediction model, the circuit board production line performance and circuit board production auxiliary performance are output; Based on the analysis of a horizontal PCB production dataset, the specific steps for identifying PCB production variation indicators and PCB production line anomaly indicators include: Establish a horizontal production line benchmark model based on the production line numbers corresponding to different production lines; Based on the horizontal production line benchmark model, circuit board feature identification is performed on all circuit board fingerprint data in the horizontal circuit board production dataset to obtain differences in circuit board production performance. Meanwhile, the cross-difference values ​​of all production auxiliary material data, circuit board production environment data and production manpower data in the horizontal circuit board production dataset are extracted in the horizontal production line benchmark model to obtain the differences in circuit board auxiliary performance. Machine learning clustering analysis was performed based on differences in circuit board production performance and auxiliary performance to obtain circuit board production difference indicators and circuit board production line anomaly indicators.

2. The method for real-time monitoring of a circuit board intelligent production line according to claim 1, characterized in that, The specific steps for establishing a dynamic digital ecosystem model for a circuit board production line include: Based on the timestamps and production line numbers of the time points, the performance of the circuit board production line, the auxiliary performance of circuit board production, the difference indicators of circuit board production, and the abnormal indicators of circuit board production line are aligned and preprocessed to obtain two-dimensional circuit board production line monitoring data. Entity extraction is performed on the monitoring data of a two-dimensional circuit board production line using a digital twin graph network, and a circuit board production line vector containing production line unit edges and production line unit nodes is constructed. A dynamic digital ecosystem model of the circuit board production line is established by dynamically updating the feature representations of the unit edges and unit nodes of the production line using a dynamic graph neural network.

3. The method for real-time monitoring of a circuit board intelligent production line according to claim 2, characterized in that, The specific steps for monitoring and adjusting the circuit board production line based on the dynamic digital ecosystem model include: Based on the dynamic digital ecosystem model, causal reasoning is performed on the pre-trained knowledge graph of circuit board production to obtain causal pairs of production factors and circuit board performance. By identifying production factors with significant differences in production performance and optimizing the causal relationship between circuit board performance, intelligent optimization of the circuit board production line can be achieved.

4. The method for real-time monitoring of a circuit board intelligent production line according to claim 3, characterized in that, The basic models for constructing a dynamic digital ecosystem model are digital twin networks and graph neural network models.

5. A real-time monitoring system for an intelligent circuit board production line, characterized in that, The system employs a real-time monitoring method for a smart circuit board production line according to any one of claims 1-4, comprising: The production line monitoring module includes a monitoring data unit, a vertical production line monitoring unit, and a horizontal production line monitoring unit. The monitoring data unit is used to set up multimodal production data packages for circuit boards, containing circuit board fingerprint data, production auxiliary material data, circuit board production environment data, and production manpower data. The vertical production line monitoring unit performs vertical time-axis analysis on the same production line, collecting corresponding multimodal production data of circuit boards at different time points within a preset time period to obtain a vertical circuit board production dataset. Analysis based on the vertical circuit board production dataset identifies the performance of the circuit board production line and the auxiliary performance of circuit board production. The horizontal production line monitoring unit performs horizontal parallel line analysis on different production lines, acquiring multiple circuit board multimodal production data points corresponding to the same time point to obtain a horizontal circuit board production dataset. Analysis based on the horizontal circuit board production dataset identifies circuit board production difference indicators and circuit board production line abnormal indicators. The production line optimization and adjustment module includes a monitoring and intelligent optimization unit. The monitoring and intelligent optimization unit is used to establish a dynamic digital ecosystem model of the circuit board production line based on the performance of the circuit board production line, the auxiliary performance of circuit board production, the difference indicators of circuit board production, and the abnormal indicators of the circuit board production line. Based on the dynamic digital ecosystem model, the circuit board production line is monitored and adjusted to intelligently optimize the real-time status of the circuit board production line.

Images