Circuit board production control method and system based on data monitoring

Abstract

Embodiments of the present application disclose a kind of line board production control method and system based on data monitoring, method includes: the real-time monitoring data set of target line board production line is collected, real-time monitoring data set includes multiple continuous time series process parameter monitoring data and equipment operation monitoring data;Real-time monitoring data set is executed multidimensional time series feature extraction processing, generates the production time sequence state feature and process fluctuation correlation feature of target line board production line;Call production decision model to carry out state fusion processing to production time sequence state feature and process fluctuation correlation feature, generate the production control decision information of target line board production line;According to production control decision information, generate equipment control instruction set, and equipment control instruction set is transmitted to the execution terminal of target line board production line to realize production state adjustment operation.Therefore, accurate control to line board production process can be realized, and the stability and efficiency of production are improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of data processing, in particular to a circuit board production control method and system based on data monitoring. BACKGROUND

[0002] In the field of circuit board production, the quality and production efficiency of circuit boards play a key role in the performance and cost of electronic products. With the diversification and high performance demand of electronic products, the production process of circuit boards is becoming increasingly complex, and the parameter control and equipment operation state monitoring in the production process are becoming increasingly important. However, the existing circuit board production control technology can only obtain single type of data in the data acquisition link, and cannot comprehensively and continuously monitor the process parameters and equipment operation state of the production line, resulting in incomplete understanding of the production process and difficulty in discovering potential problems; in the production decision-making process, the existing technology lacks effective multi-dimensional data analysis and fusion capability, and cannot fully mine the internal relationship between data, making the production decision-making not accurate enough to adapt to complex and changing production demands. In addition, the traditional technology lacks a dynamic adjustment mechanism based on real-time data analysis in the production state adjustment aspect, and it is difficult to make accurate adjustments in a timely manner according to the actual situation in the production process, affecting the stability and efficiency of the production line. SUMMARY

[0003] Embodiments of the present application provide a circuit board production control method and system based on data monitoring, which is used to realize accurate control of the circuit board production process by comprehensively collecting real-time monitoring data, multi-dimensional feature extraction, state fusion decision-making, and accurate device control instruction transmission, and to improve the stability and efficiency of production.

[0004] In a first aspect, embodiments of the present application provide a circuit board production control method based on data monitoring, applied to a circuit board production control system, the method comprising: collecting a set of real-time monitoring data of a target circuit board production line, the set of real-time monitoring data comprising a plurality of continuous time series of process parameter monitoring data and equipment operation monitoring data; performing multi-dimensional time series feature extraction processing on the set of real-time monitoring data to generate production time series state features and process fluctuation correlation features of the target circuit board production line; calling a production decision-making model to perform state fusion processing on the production time series state features and the process fluctuation correlation features to generate production control decision-making information of the target circuit board production line; generating a set of device control instructions according to the production control decision-making information, and transmitting the set of device control instructions to an execution terminal of the target circuit board production line to realize a production state adjustment operation.

[0005] In a second aspect, embodiments of the present application provide a circuit board production control system, comprising:

[0006] a processor;

[0007] a storage device having a computer program stored thereon,

[0008] When the computer program is executed by the processor, the processor implements any of the circuit board production control methods based on data monitoring.

[0009] An embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the circuit board production control method based on data monitoring are implemented.

[0010] It can be seen that the embodiments of the present invention have the following beneficial effects: the embodiments of the present invention perform multi-dimensional time series feature extraction processing on the real-time monitoring data set of the target circuit board production line collected, generate production time series state features and process fluctuation correlation features, can deeply analyze the operation characteristics of the production line from different dimensions, accurately extract key information, and provide a comprehensive and detailed basis for production decision-making. Subsequently, the production decision model is called to perform state fusion processing on the above two features, fully explore the potential connections between the features, and generate accurate production control decision information, which can comprehensively consider multiple factors and make the decision more scientific and reasonable. Finally, based on the production control decision information, a set of equipment control instructions is generated and transmitted to the execution terminal to achieve precise adjustment of the production status, effectively improve the stability and production efficiency of the production line, ensure the smooth progress of the circuit board production process, and comprehensively optimize the circuit board production control process.

[0011] In summary, the embodiments of the present invention aim to solve the problems of incomplete data collection, inaccurate decision-making, and untimely production status adjustment in the prior art. By comprehensively collecting real-time monitoring data, extracting multi-dimensional features, making state fusion decisions, and accurately transmitting equipment control instructions, precise control of the circuit board production process can be achieved, thereby improving production stability and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 The present invention provides a flow chart of a circuit board production control method based on data monitoring.

[0013] Figure 2 This is a schematic diagram of the basic structure of a circuit board production control system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0015] See also Figure 1As shown in FIG, this figure is a flow chart of a circuit board production control method based on data monitoring provided by an embodiment of the present invention, which can be applied to a circuit board production control system. Figure 1 As shown, the method includes steps 110 to 140.

[0016] Step 110: Collect a real-time monitoring data set of the target circuit board production line, wherein the real-time monitoring data set includes a plurality of continuous time series of process parameter monitoring data and equipment operation monitoring data.

[0017] In an embodiment of the present invention, a circuit board production line of an electronics factory is taken as an example. This production line produces various types of circuit boards, and various parameters need to be monitored in real time during the production process. For process parameter monitoring data, parameters such as etching solution concentration, electroplating current intensity, and drilling depth are covered. These parameters change continuously over time during the production process. For example, the etching solution concentration may be monitored every 10 seconds, and a numerical value is obtained each time it is monitored, forming a continuous time series monitoring data sequence. For equipment operation monitoring data, it includes the operating speed of the placement machine, the temperature change of the reflow oven, etc. The operation speed of the placement machine is measured by the number of placements per minute, and is also collected at certain time intervals to form time series data of the equipment operation status. By arranging sensors and data acquisition devices at key positions of the production line, these process parameters and equipment operation data are collected and summarized in real time, thereby obtaining a real-time monitoring data set containing multiple continuous time series of process parameter monitoring data and equipment operation monitoring data.

[0018] Step 120: Perform multi-dimensional time series feature extraction processing on the real-time monitoring data set to generate production time series state features and process fluctuation correlation features of the target circuit board production line.

[0019] Next, the collected real-time monitoring data sets are further processed to obtain key characteristic information of the production line. This process involves analyzing the data from multiple dimensions to generate production time series status characteristics and process fluctuation correlation characteristics that can reflect the production line operation status.

[0020] As an optional embodiment, step 120 includes:

[0021] Step 121: performing data cleaning processing on the real-time monitoring data set to obtain an optimized monitoring data set; the data cleaning processing includes: replacing abnormal data points in the process parameter monitoring data with adjacent time series data, and performing time window alignment on non-continuously collected data in the equipment operation monitoring data.

[0022] In an embodiment of the present invention, for process parameter monitoring data, such as etching solution concentration data sequence, if the concentration value monitored at a certain moment is too different from the value at the previous and next moments, and obviously does not conform to the normal production fluctuation range, the data point is determined to be an abnormal data point. At this time, the value of the previous or next normal data point adjacent to the abnormal data point is used to replace it. For equipment operation monitoring data, such as reflow oven temperature collection data, there may be inconsistent collection time intervals due to certain reasons. At this time, a time window alignment method is used to set a suitable time window, and the data collected at different times are aligned according to the time window to ensure that the data are consistent and comparable in the time dimension, thereby obtaining an optimized monitoring data set.

[0023] Step 122: Performing time series fluctuation analysis on the optimized monitoring data set to identify fluctuation characteristics of process parameters and equipment operation status of the target circuit board production line within multiple consecutive production cycles.

[0024] In this step, the optimized monitoring data set is further analyzed for time series fluctuations in order to gain a deeper understanding of the operational changes of the production line during different production cycles.

[0025] In one implementation, step 122 includes:

[0026] Step 1221: Divide the optimized monitoring data set into a process parameter time series sequence and an equipment operation status time series sequence.

[0027] In this embodiment of the present invention, the optimized monitoring data set is clearly divided into two sequences based on the data attributes. Data related to process parameters, such as etching solution concentration and electroplating current intensity, are composed of process parameter time series sequences; data related to equipment operating status, such as placement machine operating speed and reflow oven temperature, are composed of equipment operating status time series sequences, facilitating subsequent targeted analysis of each.

[0028] Step 1222: performing sliding window segmentation processing on the process parameter time series sequence to obtain multiple process parameter window subsequences, and performing trend fitting on each of the process parameter window subsequences to determine the fluctuation direction and fluctuation amplitude of the process parameter window subsequence.

[0029] For the process parameter time series, a fixed-size sliding window is set, for example, the window size is 20 time points. The window starts from the starting position of the sequence and moves one time point at a time, dividing the sequence in turn to obtain multiple process parameter window subsequences. For each window subsequence, a curve fitting algorithm, such as least squares fitting, is used to determine the change trend of the subsequence, thereby determining whether the fluctuation direction is rising, falling, or stable, and calculating the fluctuation amplitude. For example, for a window subsequence of a certain etching solution concentration, a straight line is obtained by fitting. If the slope of the straight line is positive, the fluctuation direction is rising. The fluctuation amplitude can be calculated based on the slope size combined with the data range.

[0030] Step 1223: performing state segmentation processing on the device operation state timing sequence to obtain multiple device operation state sub-intervals, and performing stability evaluation on each of the device operation state sub-intervals to determine the operation offset and offset duration of the device operation state sub-interval.

[0031] Optionally, for a timing sequence of equipment operating states, the states are segmented according to changes in the equipment operating states. For example, for a sequence of SMT machine operating speeds, when the speed remains relatively stable over a period of time, it is divided into a subinterval; when the speed changes significantly, a new subinterval is re-divided. A stability assessment is performed on each subinterval, and stability is determined by calculating statistics such as the variance of the data. If the variance is small, it indicates that the equipment operation in that subinterval is relatively stable; if the variance is large, there is an operating offset. The operating offset can be calculated by comparing the difference between the mean of the data in the subinterval and the normal operating value, and the duration of the offset is determined based on the time span of the subinterval.

[0032] Step 1224: Generate the process parameter fluctuation representation and the equipment operation status fluctuation representation according to the fluctuation direction, the fluctuation amplitude, the operation offset, and the offset duration.

[0033] Optionally, after obtaining the fluctuation direction and amplitude of the process parameter window subsequence, as well as the operating offset and offset duration of the equipment operating status subinterval, this information is integrated. For example, the fluctuation direction and amplitude of different process parameters can be summarized to form a process parameter fluctuation representation that can comprehensively reflect the process parameter fluctuation situation; the operating offset and offset duration of each equipment operating status subinterval can be sorted to generate an equipment operating status fluctuation representation, thereby clearly presenting the fluctuation characteristics of the process parameters and equipment operating status of the production line over multiple consecutive production cycles.

[0034] Step 123: Generate the process fluctuation correlation feature of the target circuit board production line based on the process parameter fluctuation representation, and generate the production timing state feature based on the equipment operation state fluctuation representation.

[0035] Optionally, based on the process parameter fluctuation characterization obtained above, the associated information is further mined to generate process fluctuation correlation features. For example, the correlation between different process parameter fluctuations, such as the relationship between etching solution concentration fluctuations and electroplating quality fluctuations, is analyzed. This associated information is quantified to form process fluctuation correlation features. For the characterization of equipment operating state fluctuations, the changes in equipment operating state at different time points are comprehensively considered to generate production timing state features that reflect the overall production timing state of the production line, providing a strong basis for subsequent production decisions.

[0036] Step 130: Calling a production decision model to perform state fusion processing on the production time sequence state characteristics and the process fluctuation correlation characteristics to generate production control decision information for the target circuit board production line.

[0037] This step aims to use the production decision model to deeply fuse the two features extracted previously to formulate production control decision information that meets the actual needs of the production line.

[0038] In one exemplary embodiment, step 130 includes:

[0039] Step 131: Map the production timing state feature into a first state vector, map the process fluctuation correlation feature into a second state vector, and perform vector splicing processing on the first state vector and the second state vector to obtain a fused state vector.

[0040] In this embodiment of the present invention, a feature mapping algorithm is used to convert production sequence state features into a vector, namely a first state vector, where each dimension represents a different aspect of the production sequence state. Similarly, process fluctuation-related features are mapped into a second state vector. These two vectors are then concatenated according to specific rules. For example, the dimensions of the first state vector are sequentially aligned with those of the second state vector to form a new fused state vector for subsequent model processing.

[0041] Step 132: Calling the multilayer perceptron in the production decision model to perform nonlinear transformation processing on the fusion state vector to generate a state prediction result of the target circuit board production line.

[0042] The multilayer perceptron in the production decision model consists of multiple layers of neurons. The fused state vector is input into the multilayer perceptron, where neurons process the input data through weighted summation and nonlinear activation functions. For example, neurons multiply the values ​​of each dimension of the fused state vector by their corresponding weights and add them together. These values ​​are then transformed using nonlinear activation functions such as ReLU. After processing through multiple layers of neurons, a prediction result for the target circuit board production line is generated, which contains information about the future operating status of the production line.

[0043] Step 133: Match the predefined production control strategy library according to the state prediction result to determine the production control decision information.

[0044] Optionally, the state prediction result is matched with a predefined production control strategy library to find the most appropriate production control strategy, thereby determining the production control decision information.

[0045] In one implementation, step 133 includes:

[0046] Step 1331: Analyze the process parameter fluctuation amplitude set, equipment operation offset set, and production line efficiency deviation index contained in the state prediction result to generate a current prediction parameter set.

[0047] Optionally, the state prediction results can be analyzed in detail to extract key information. For example, the fluctuation amplitudes of different process parameters in the prediction results can be organized into a process parameter fluctuation amplitude set, and the equipment operation offset values ​​can be organized into an equipment operation offset set. Furthermore, a production line efficiency deviation index can be calculated. This information is integrated to form the current prediction parameter set, which provides a basis for subsequent matching strategies.

[0048] Step 1332: Obtain the historical process parameter fluctuation amplitude set, historical equipment operation offset set and historical production line efficiency deviation index corresponding to each historical control strategy in the predefined production control strategy library, and perform dimensional alignment processing on the historical process parameter fluctuation amplitude set and the historical equipment operation offset set with the current prediction parameter set to generate a normalized historical parameter set; wherein, the dimensional alignment processing includes: performing a linear conversion from Celsius to a standard temperature scale on the temperature parameter in the historical process parameter fluctuation amplitude set, performing a decimal scaling from kilopascals to megapascals on the pressure parameter, and performing a unit unification from milliseconds to seconds on the time dimension parameter in the historical equipment operation offset set.

[0049] In this step, the relevant parameter sets corresponding to each historical control strategy are retrieved from the predefined production control strategy library. To make these historical parameters comparable with the current predicted parameters, dimensional alignment is performed. For temperature parameters in the historical process parameter fluctuation range set, degrees Celsius are converted to a standard temperature scale using a specific linear conversion formula. For pressure parameters, kilopascals are converted to megapascals using decimal scaling. For time dimension parameters in the historical equipment operation offset set, milliseconds are uniformly converted to seconds. After these processes, a normalized historical parameter set is generated.

[0050] Step 1333: Calculate the multi-dimensional similarity value between the current prediction parameter set and each of the normalized historical parameter sets, the multi-dimensional similarity value includes the process parameter fluctuation direction consistency coefficient, the equipment operation offset timing matching degree and the efficiency deviation weight distance; wherein, the calculation of the multi-dimensional similarity value meets the following conditions: the process parameter fluctuation direction consistency coefficient is determined by the slope sign matching degree of the corresponding process parameters in the current prediction parameter set and the normalized historical parameter set; the equipment operation offset timing matching degree calculates the minimum path distance of the two equipment operation offset sequences through the time warping algorithm; the efficiency deviation weight distance fuses the normalized difference of the production line efficiency deviation index through the weighted Euclidean distance formula.

[0051] In this step, the similarity values ​​between the current prediction parameter set and each normalized historical parameter set are calculated for multiple dimensions. For the process parameter fluctuation direction consistency coefficient, the slope signs of the corresponding process parameters in the current prediction parameter set and the normalized historical parameter set are compared. If the signs are the same, the consistency coefficient is high. For the equipment operation offset timing matching, a time warping algorithm is used to find the minimum path distance between the two equipment operation offset sequences. The smaller the distance, the higher the matching degree. For the efficiency deviation weighted distance, the weighted Euclidean distance formula is used to consider the weights of different factors and integrate the normalized difference of the production line efficiency deviation indicators to calculate the corresponding distance value. These similarity values ​​are combined to comprehensively measure the similarity between the two.

[0052] Step 1334: Sort the historical control strategies in the predefined production control strategy library according to the multi-dimensional similarity values, and select the target historical control strategy that meets the similarity threshold and has the highest priority.

[0053] Optionally, the historical control strategies in the production control strategy library are sorted based on the calculated multi-dimensional similarity values. A similarity threshold is set, and only strategies with similarity values ​​that meet or exceed this threshold are considered. Among the strategies that meet the threshold, the target historical control strategy with the highest priority is selected according to pre-set priority rules. This strategy is the one that best matches the predicted status of the current production line.

[0054] Step 1335: Extract the process adjustment parameter mapping table, equipment control instruction sequence and execution timing constraints in the target historical control strategy, perform parameter correction in combination with the real-time production line efficiency deviation index in the current prediction parameter set, and generate the production control decision information.

[0055] Optionally, key information such as the process adjustment parameter mapping table, equipment control instruction sequence, and execution timing constraints are extracted from the target historical control strategy. This information is then combined with the real-time production line efficiency deviation indicator in the current prediction parameter set to perform parameter corrections. For example, based on the efficiency deviation, certain parameters in the process adjustment parameter mapping table can be fine-tuned, and the instruction execution time in the equipment control instruction sequence can be appropriately adjusted. This generates the final production control decision information, ensuring that the decision effectively guides production line operations.

[0056] In an alternative embodiment, the training method of the production decision model includes:

[0057] Step 210: Obtain a historical monitoring data set of a historical production line and a corresponding historical control decision set; perform multi-dimensional time series feature extraction processing on the historical monitoring data set to generate historical production time series state features and historical process fluctuation correlation features.

[0058] In this step, historical production line monitoring data from the past period is collected. This data includes process parameter monitoring data and equipment operation monitoring data, forming a historical monitoring data set. Simultaneously, a set of historical control decisions corresponding to this historical data is obtained—that is, the control decisions made for the production line status at that time. Then, using the same method used for real-time monitoring data processing, multi-dimensional time series feature extraction is performed on this historical monitoring data set to generate historical production time series state features and historical process fluctuation correlation features, providing a data foundation for model training.

[0059] Step 220: Input the historical production time series state characteristics and the historical process fluctuation correlation characteristics as training samples into the initial production decision model to generate predictive control decision information.

[0060] Optionally, the generated historical production time series state features and historical process fluctuation correlation features are fed into the initial production decision model as input data. The initial production decision model processes these input features using its own structure and algorithm, and outputs predictive control decision information, which is the model's prediction of the control decision that should be made based on the historical data.

[0061] Step 230: Construct a loss function based on the difference between the predicted control decision information and the historical control decision set, and iteratively train the initial production decision model based on the loss function until convergence to obtain the production decision model.

[0062] In this step, a loss function is constructed by comparing the predicted control decision information with the set of historical control decisions. The loss function measures the gap between the model's predictions and the actual historical decisions. Based on this loss function, an optimization algorithm, such as stochastic gradient descent, is used to iteratively adjust the parameters of the initial production decision model. In each iteration, the model parameters are adjusted based on the value of the loss function, gradually bringing the model's predictions closer to the historical control decisions, until the loss function converges to a smaller value. This results in a trained production decision model.

[0063] Furthermore, constructing a loss function based on the difference between the predicted control decision information and the historical control decision set includes:

[0064] Step 231: performing vectorized coding processing on the predicted control decision information to obtain a predicted decision vector, and performing vectorized coding processing on the historical control decision set to obtain a historical decision vector.

[0065] Optionally, a feature encoding algorithm is used to convert the predictive control decision information into a vector, namely a predictive decision vector, where each dimension of the vector represents a different aspect of the decision information. Similarly, the set of historical control decisions is vectorized and encoded to obtain a historical decision vector for subsequent comparison and calculation.

[0066] Step 232: Calculate the cosine similarity between the predicted decision vector and the historical decision vector, determine the first loss component based on the cosine similarity, calculate the Euclidean distance between the predicted decision vector and the historical decision vector, and determine the second loss component based on the Euclidean distance.

[0067] Furthermore, using vector computation methods, we calculate the cosine similarity between the predicted decision vector and the historical decision vector. Cosine similarity reflects the degree of directional similarity between two vectors. This similarity value is used to determine the first loss component; higher similarity indicates a smaller first loss component. Simultaneously, we calculate the Euclidean distance between the two vectors. This distance measures the distance between the vectors in space. This is used to determine the second loss component; smaller distances indicate smaller second loss components.

[0068] Step 233: Perform a weighted summation on the first loss component and the second loss component to obtain the loss function.

[0069] In this step, different weights are assigned to the first and second loss components based on actual needs. For example, if consistency in decision direction is more important, a larger weight is assigned to the first loss component determined by cosine similarity; if the specific numerical differences in the decisions are more important, a larger weight is assigned to the second loss component determined by Euclidean distance. The two loss components are summed according to their weights to obtain the loss function used to measure the model's prediction error.

[0070] Step 140: Generate a set of equipment control instructions based on the production control decision information, and transmit the set of equipment control instructions to the execution terminal of the target circuit board production line to implement a production status adjustment operation.

[0071] Optionally, based on the generated production control decision information, a specific set of equipment control instructions is formulated and sent to the execution terminal of the production line, thereby adjusting the operating status of the production line.

[0072] In an optional embodiment, generating a set of equipment control instructions according to the production control decision information includes:

[0073] Step 141: parse the equipment adjustment type and equipment adjustment range in the production control decision information; match the corresponding execution equipment identifier in the target circuit board production line according to the equipment adjustment type, and generate the equipment control quantitative characteristics of the execution equipment identifier according to the equipment adjustment range; perform instruction encapsulation processing on the execution equipment identifier and the equipment control quantitative characteristics to generate the equipment control instruction set.

[0074] Production control decision information is thoroughly analyzed to extract the device adjustment type, such as device on / off, speed adjustment, and the specific numerical value of the device adjustment range. Based on the device adjustment type, the corresponding execution device identifier is found in the equipment list of the target circuit board production line. For example, if the decision information requires adjusting the operating speed of a placement machine, the unique identifier of the placement machine is determined by searching the equipment list. Based on the device adjustment range, a device control quantitative feature is generated for the execution device identifier, such as the specific numerical value of the speed adjustment. Finally, the execution device identifier and device control quantitative feature are packaged according to the set instruction format to form a device control instruction set.

[0075] In another optional embodiment, transmitting the device control instruction set to an execution terminal of the target circuit board production line to implement a production status adjustment operation includes:

[0076] Step 142: Prioritize the device control instruction set to determine the execution order of each device control instruction; transmit the device control instruction set to the execution terminal of the target circuit board production line in sequence according to the execution order; monitor the response status of the execution terminal to the device control instruction set, and when an abnormal response is detected, regenerate a backup device control instruction set and transmit it.

[0077] Optionally, the device control instruction set is prioritized based on factors such as the importance and urgency of the device control instruction. For example, device control instructions related to key links of the production line have a higher priority. After determining the execution order of each instruction, the device control instruction set is sent to the execution terminal of the target circuit board production line in sequence according to this order. During the transmission process, the response status of the execution terminal to the device control instruction set is monitored in real time. If the execution terminal fails to execute the instruction as expected, such as when an abnormal situation such as no response from the device or execution error occurs, a backup device control instruction set is regenerated based on the production control decision information and transmitted again to ensure that the production line can adjust the production status according to the correct control instructions.

[0078] As an alternative embodiment, the method further includes:

[0079] Step 310: Convert the production control decision information into visual chart data, where the visual chart data includes a process parameter adjustment trend chart, an equipment operation status comparison chart, and a production line efficiency prediction chart.

[0080] In an embodiment of the present invention, production control decision information needs to be converted to more intuitively present it and facilitate analysis and decision-making by production managers. For a process parameter adjustment trend chart, adjustment information for each process parameter, such as the change in the etching solution concentration adjustment value over time, is extracted from the production control decision information. A plotting algorithm is used to create a curve reflecting the concentration adjustment trend, with time as the horizontal axis and the etching solution concentration adjustment value as the vertical axis. The adjustment trends for multiple process parameters are plotted on the same graph to form a process parameter adjustment trend chart. For an equipment operating status comparison chart, operating status parameters for key equipment are selected, such as the operating speed of a placement machine and the temperature of a reflow oven. The current operating status parameters of the equipment in the decision information are compared with historical normal operating status parameters. Different graphical elements (such as bar charts and line graphs) are used to represent the current and historical statuses, respectively, to intuitively display the differences in equipment operating status and generate an equipment operating status comparison chart. For a production line efficiency prediction chart, changes in production line efficiency over a period of time are predicted based on the production line efficiency deviation indicator in the production control decision information and the relevant prediction model. With time as the horizontal axis and the predicted production line efficiency value as the vertical axis, a prediction curve is drawn to form a production line efficiency forecast chart. Through these visual charts and data, production managers can quickly understand the operating status and future development trends of the production line.

[0081] Step 320: Send the visualization chart data to the production management terminal for display, and when a decision correction instruction in response to the production management terminal is detected, re-execute the multi-dimensional time series feature extraction processing and the state fusion processing to update the production control decision information.

[0082] In an embodiment of the present invention, the generated visualization chart data is transmitted to a production management terminal via a network transmission method, such as a computer, tablet computer, etc. After viewing these visualization charts, the production manager may issue decision correction instructions based on actual experience and production needs.

[0083] In a preferred embodiment, when the decision correction instruction responded by the production management terminal is detected in step 320, re-execution of the multi-dimensional time series feature extraction process and the state fusion process to update the production control decision information includes:

[0084] Step 321: Analyze the process parameter adjustment amount set, the equipment operation offset compensation coefficient set, and the production cycle timing constraint condition set in the decision correction instruction to generate a correction parameter set.

[0085] Optionally, the decision modification instruction responded by the production management terminal is analyzed in detail. A process parameter adjustment amount set is extracted from the instruction, for example, specific values of different process parameters (etching liquid concentration, electroplating current intensity, etc.) that need to be adjusted; a device operation offset compensation coefficient set is extracted, that is, coefficients for compensating for device operation offset conditions; and a production cycle timing constraint condition set is extracted, such as time limit information of the production cycle. These information is integrated together to form a modification parameter set, which provides a basis for subsequent data processing.

[0086] Step 322: superimpose the process parameter adjustment amount set of the modification parameter set on the process parameter monitoring data of the optimized monitoring data set according to the corresponding timing window to generate a modified process parameter sequence; and perform sliding window filtering fusion on the device operation offset compensation coefficient set of the modification parameter set and the device operation monitoring data of the optimized monitoring data set to generate a modified device operation sequence.

[0087] Optionally, for the process parameter adjustment amount set in the modification parameter set, the adjustment amount is superimposed on the process parameter monitoring data of the optimized monitoring data set according to the corresponding time window. For example, if the decision modification instruction requires increasing the etching liquid concentration by a certain value within a certain time period, the increase is added to the etching liquid concentration monitoring data of the corresponding time period to generate a modified process parameter sequence. For the device operation offset compensation coefficient set, a sliding window filtering fusion method is used to process the device operation monitoring data of the optimized monitoring data set. A suitable sliding window size is set, and the window is slid on the device operation monitoring data sequence. In each window, the data is weighted and fused according to the compensation coefficient to generate a modified device operation sequence, so that the device operation data is more in line with the modified requirements.

[0088] Step 323: intercept the timing interval of the modified process parameter sequence and the modified device operation sequence according to the production cycle timing constraint condition set to generate a modified monitoring data set; perform the timing fluctuation analysis processing on the modified monitoring data set to generate updated process parameter fluctuation representation and device operation state fluctuation representation; generate an updated process fluctuation correlation feature based on the updated process parameter fluctuation representation, and generate an updated production timing state feature based on the updated device operation state fluctuation representation.

[0089] In an embodiment of the present invention, based on a set of production cycle timing constraints, time intervals that meet the time requirements are intercepted from the modified process parameter sequence and the modified equipment operation sequence to form a modified monitoring data set. The modified monitoring data set is then processed according to the previously described timing fluctuation analysis method to determine the fluctuation direction and amplitude of the process parameters, as well as the operating offset and offset duration of the equipment operating state, to generate updated process parameter fluctuation representations and equipment operating state fluctuation representations. Based on these new fluctuation representations, updated process fluctuation correlation features and updated production timing state features are further generated to provide new feature data for subsequent state fusion.

[0090] Step 324: Perform time-series dimension alignment processing on the updated process fluctuation correlation feature and the process fluctuation correlation feature. When the feature lengths are inconsistent, use endpoint replication or mean padding to expand them to the same time step. Perform feature channel matching processing on the updated production timing state feature and the production timing state feature. When the number of channels is inconsistent, map them to the same feature space through the fully connected layer.

[0091] In an embodiment of the present invention, the updated process fluctuation associated features and the process fluctuation associated features are aligned in time series dimension. If the time lengths of the two features are different, for example, the time step of the updated process fluctuation associated features is shorter, the endpoint replication method is used to repeat the endpoint value at the shorter end, or the mean filling method is used to fill the missing part with the mean of the previous and next data, so that they have the same time step. For the updated production timing status features and production timing status features, when their number of feature channels is inconsistent, a fully connected layer is used for mapping. Each neuron in the fully connected layer is connected to all neurons of the input features, and by learning the weight matrix, features with different channel numbers are mapped to the same feature space for subsequent vector splicing.

[0092] Step 325: perform vector concatenation on the updated process fluctuation correlation features after the timing dimension is aligned and the updated production timing state features after the feature channel is matched to generate a fused updated feature vector; call the production decision model to perform the state fusion processing on the fused updated feature vector to generate intermediate control decision information.

[0093] Optionally, the updated process fluctuation correlation feature vector, aligned with the time-series dimension, and the updated production time-series state feature vector, matched with the feature channels, are concatenated. The dimensions of the two vectors are arranged in a specific order to form a fused updated feature vector. This fused updated feature vector is then input into the production decision model, which performs state fusion processing on it. Using structures such as multilayer perceptrons within the model, the model performs nonlinear transformations and other operations to generate intermediate control decision information, which represents a preliminary decision result generated based on the new feature data.

[0094] Step 326: Calculate the absolute difference set of the corresponding process adjustment parameters in the intermediate control decision information and the production control decision information and the cosine similarity set of the equipment control instructions; filter out the conflicting control items in the absolute difference set that exceed the preset process threshold and the cosine similarity set that is lower than the preset equipment threshold; perform weighted sliding average correction on the process adjustment parameters in the conflicting control items, and adjust the weights according to the timing fluctuation amplitude of the corresponding process parameters.

[0095] In this step, the absolute differences between the corresponding process adjustment parameters in the intermediate control decision information and the production control decision information are calculated, and these differences are combined into an absolute difference set. At the same time, the cosine similarity of the equipment control instructions in the two decision information is calculated to form a cosine similarity set. A preset process threshold and a preset equipment threshold are set, and the process parameter differences greater than the preset process threshold are screened out from the absolute difference set, and the equipment control instruction similarities lower than the preset equipment threshold are screened out from the cosine similarity set to determine the conflicting control items. For the process adjustment parameters in the conflicting control items, a weighted sliding average correction method is adopted. The weight is determined according to the time series fluctuation amplitude of the corresponding process parameters. The parameters with large fluctuation amplitudes have larger weights. The process adjustment parameters are weighted averaged and corrected within the sliding window to make the corrected process parameters more reasonable.

[0096] Step 327: Prioritize the device control instructions in the conflicting control items, and determine the coverage order based on the time matching degree between the device response delay parameters and the production cycle timing constraints; merge the corrected process adjustment parameters and the device control instructions after priority coverage into the production control decision information to generate updated production control decision information.

[0097] Optionally, device control instructions in conflicting control items are prioritized based on the time matching between the device response delay parameter and the production cycle timing constraints. If a device has a long response delay but the production cycle requires tight time constraints on its operation, the device control instruction will be given a higher priority. After determining the priority, the device control instructions are overwritten in order of priority. The corrected process adjustment parameters and the overwritten device control instructions are merged with the original production control decision information to form updated production control decision information, ensuring that the decision information is more adaptable to the actual situation of the production line.

[0098] Step 328: Correct the timing matching between the temperature adjustment step of the process adjustment parameter in the updated production control decision information and the voltage change rate of the equipment control instruction so that the millisecond timestamp of the temperature adjustment window covers the voltage change interval.

[0099] Optionally, check the temporal matching of the temperature adjustment step size of the process adjustment parameter and the voltage change rate of the equipment control instruction in the updated production control decision information. If the temperature adjustment window and the voltage change interval do not match in time, for example, the temperature adjustment start time is later than the voltage change start time, or the temperature adjustment end time is earlier than the voltage change end time, correct the timestamp based on the specific situation. By adjusting the execution time of the temperature adjustment step size or the voltage change rate, ensure that the millisecond-level timestamp of the temperature adjustment window can fully cover the voltage change interval, so that the process adjustment and equipment control are coordinated in time, ensuring the stable operation of the production line.

[0100] In a non-limiting embodiment, after transmitting the equipment control instruction set to the execution terminal of the target circuit board production line to implement the production status adjustment operation, it also includes: collecting a response feedback data set from the execution terminal, the response feedback data set including a sequence of actual adjustment values ​​of process parameters and a sequence of equipment operation status updates; performing execution effect feature extraction on the response feedback data set to generate actual process fluctuation features and actual equipment status timing features; performing time-series alignment on the actual process fluctuation features and the process fluctuation correlation features to calculate the difference to generate a process adjustment deviation coefficient set; performing window sliding correlation analysis on the actual equipment status timing features and the production timing status features to generate an equipment control lag time parameter set; adjusting the control magnitude and execution timing interval in the equipment control instruction set according to the process adjustment deviation coefficient set and the equipment control lag time parameter set to generate a compensation control instruction set and additionally transmit it to the execution terminal.

[0101] In this embodiment, after a set of equipment control instructions is transmitted to an execution terminal and executed, data fed back from the execution terminal is collected. The actual process parameter adjustment value sequence records the actual adjusted values ​​of each process parameter, and the equipment operating status update sequence reflects the changes in the equipment's operating status after executing the instructions. Feature extraction is performed on these response feedback data sets to analyze the actual fluctuations in the process parameters and generate actual process fluctuation features. The temporal variations in the equipment operating status are analyzed to generate actual equipment status time series features. The actual process fluctuation features are time-series aligned with the previously generated process fluctuation correlation features, and the difference between the two is calculated to form a set of process adjustment deviation coefficients, which reflects the degree of deviation between the actual and expected process parameter adjustments. A window sliding correlation analysis is performed on the actual equipment status time series features and the production time series status features. For example, a sliding window is set, moved across the two feature sequences, and the correlation of the data within the window is calculated to determine the equipment control lag time parameters and generate a set of equipment control lag time parameters. Based on the information reflected by these two sets, the control levels in the equipment control instruction set are adjusted, such as increasing or decreasing the control strength of certain equipment, and the execution time intervals are adjusted, such as extending or shortening the interval between instruction executions. After the compensation control instruction set is generated, it is transmitted to the execution terminal to correct the deviations that occurred during the execution of the previous equipment control instructions, making the operation of the production line more in line with expectations.

[0102] In a non-limiting embodiment, after transmitting the device control instruction set to the execution terminal of the target circuit board production line to implement the production status adjustment operation, it also includes: acquiring the target monitoring data stream of the target circuit board production line in real time, the target monitoring data stream including the post-execution process parameter derivative sequence and the equipment operation status change rate sequence; performing abnormal fluctuation detection on the target monitoring data stream, identifying the abnormal parameter identifier and the corresponding abnormal time window that do not match the preset process threshold range in the production control decision information; extracting the device control instruction execution record corresponding to the abnormal parameter identifier, intercepting the historical operation offset sequence of the associated equipment according to the abnormal time window, and generating an abnormal traceability feature vector; inputting the abnormal traceability feature vector into a pre-trained compensation strategy generation model, and outputting a compensation control parameter set and priority adjustment coefficient for the abnormal parameter identifier; overwriting the parameter value of the corresponding instruction in the original device control instruction set according to the compensation control parameter set, and reordering the instruction transmission queue based on the priority adjustment coefficient to generate an optimized control instruction set and perform secondary transmission.

[0103] In this embodiment, the target monitoring data stream of the target circuit board production line is acquired in real time. The post-execution process parameter derivative sequence records the derivative data of the process parameters after the execution of the equipment control instructions, such as the etching accuracy of the circuit board after the etching process; the equipment operating state change rate sequence reflects the rate of change of the equipment operating state over time. This target monitoring data stream is subjected to abnormal fluctuation detection. Using a set detection algorithm, the monitoring data is compared with the process threshold range preset in the production control decision information. If the values ​​of certain process parameters are found to exceed the preset range, the identifiers of these abnormal parameters and the time window in which the abnormality occurred are determined. The equipment control instruction execution records corresponding to these abnormal parameter identifiers are found. Based on the abnormal time window, the historical operation offset sequence of the associated equipment within the time period is intercepted from the equipment operating data. These sequences are integrated to generate an abnormal tracing feature vector. The abnormal tracing feature vector is input into a pre-trained compensation strategy generation model. Based on previously learned knowledge and patterns, the model outputs a set of compensation control parameters and priority adjustment coefficients for these abnormal parameter identifiers. Based on the compensation control parameter set, the parameter values ​​of the corresponding instructions in the original equipment control instruction set are overwritten and updated; according to the priority adjustment coefficient, the instruction transmission queue is reordered, and the optimized control instruction set is generated and transmitted to the execution terminal again to solve the abnormal situations that occur during the operation of the production line and ensure the stable and efficient operation of the production line.

[0104] In detail, based on the above technical solution, in actual application, the dynamic time warping algorithm (DTW) in the existing technology can be combined to calculate the timing matching degree of equipment operation offset, and the minimum path distance can be found by constructing a cumulative distance matrix. At the same time, the historical process parameters are converted to units in combination with the equipment control parameter dimensional specifications in the industrial standard ISA-88. For example, when converting Celsius temperature to Kelvin temperature scale, the linear conversion formula of T (K) = T (℃) + 273.15 is used, and the pressure parameter is scaled from kilopascals to megapascals (1MPa=1000kPa).

[0105] In the actual application process of the embodiment of the present invention, for the sliding window segmentation processing, the window size can be determined according to the Nyquist-Shannon sampling theorem to ensure that the window can cover at least two complete process fluctuation cycles. In the design of the multi-layer perceptron, a three-layer network structure with a ReLU activation function is adopted, the number of neurons in the hidden layer is configured according to the 2 / 3 rule of the input feature dimension, and the parameters are updated based on the Adam optimizer. For the weight distribution of the weighted Euclidean distance, the index weight of each process parameter can be calculated in combination with the entropy weight method to ensure that the weight coefficient objectively reflects the importance of the parameter. The timing alignment processing can use a dynamic time bending algorithm for unequal length sequence matching, and anomaly detection is achieved by setting a dynamic threshold mechanism based on the 3σ principle.

[0106] As you can understand, in the visualization chart generation phase, the Matplotlib library's pyplot interface is used for multi-dimensional data visualization rendering, and the time series database InfluxDB is used to achieve precise millisecond-level timestamp alignment. Equipment control instructions are prioritized according to the equipment criticality classification system in the ANSI / ISA-95 standard, ensuring that critical equipment instructions are executed first. This design ensures accurate extraction of process parameter fluctuation representations and effective training of production decision models, enabling the precise generation and execution of production line control decisions.

[0107] It can be understood that the input and output of the aforementioned neural network model (e.g., the production decision model) construct a corresponding process control logic chain through a multi-level feature abstraction and state evolution mechanism. The input consists of real-time time series data of process parameters such as etching solution concentration and electroplating current intensity, as well as equipment status monitoring data such as placement machine operating speed and reflow oven temperature, collected from the target production line. After data cleaning, outlier point replacement and time window alignment are processed to form an optimized monitoring data set with temporal continuity. This data is then segmented through sliding windows and trend fitting to extract the fluctuation direction and amplitude characteristics of the process parameters. This is then combined with stability assessment of the equipment status segments to generate operational offset features, forming a primary feature layer that characterizes the microscopic fluctuation characteristics of the production line.

[0108] The intermediate processing layer analyzes the temporal correlation between the direction of process parameter fluctuations and the equipment operation offset, explores the meso-level correlation characteristics such as the coupling relationship between the changes in etching process parameters and the vibration of drilling equipment, and constructs a process fluctuation correlation matrix that reflects the collaborative working status of multiple devices; at the same time, the equipment operation status fluctuation representation is mapped into a production timing state vector that includes macro indicators such as production line beat synchronization and process connection efficiency.

[0109] In the decision-making stage, the multi-layer perceptron fuses the process fluctuation correlation matrix and the production timing state vector into a state prediction vector in a high-dimensional feature space through nonlinear transformation. This vector performs multi-dimensional similarity calculation with the historical control strategy through the strategy library matching mechanism. Finally, the abstract feature correlations such as the slope sign matching degree of the process parameter and the equipment offset timing path distance are converted into specific executable control outputs such as the etching liquid supply adjustment instructions and the acceleration parameters of the placement machine, forming a cross-time scale mapping from millisecond-level sensor data to minute-level control instructions. It not only maintains the real-time responsiveness of the process parameter adjustment, but also ensures the coordinated consistency of the control instructions of multiple devices within the production line cycle, and realizes the layer-by-layer semantic enhancement and logical connection of data features from the original monitoring values ​​to the equipment action signals.

[0110] The embodiment of the present invention performs multi-dimensional time series feature extraction processing on the real-time monitoring data set of the target circuit board production line collected, generates production time series state features and process fluctuation correlation features, can deeply analyze the operation characteristics of the production line from different dimensions, accurately extract key information, and provide a comprehensive and detailed basis for production decision-making. Subsequently, the production decision model is called to perform state fusion processing on the above two features, fully explore the potential connections between the features, and generate accurate production control decision information, which can comprehensively consider multiple factors and make the decision more scientific and reasonable. Finally, based on the production control decision information, a set of equipment control instructions is generated and transmitted to the execution terminal to achieve precise adjustment of the production status, effectively improve the stability and production efficiency of the production line, ensure the smooth progress of the circuit board production process, and comprehensively optimize the circuit board production control process.

[0111] In summary, the embodiments of the present invention aim to solve the problems of incomplete data collection, inaccurate decision-making, and untimely production status adjustment in the prior art. By comprehensively collecting real-time monitoring data, extracting multi-dimensional features, making state fusion decisions, and accurately transmitting equipment control instructions, precise control of the circuit board production process can be achieved, thereby improving production stability and efficiency.

[0112] See also Figure 2 As shown in FIG. 1 , this figure is a schematic diagram of the basic structure of a circuit board production control system 200 provided by an embodiment of the present invention. The circuit board production control system 200 includes:

[0113] Processor 201;

[0114] a storage device 202 having a computer program 2020 stored thereon;

[0115] When the computer program 2020 is executed by the processor 201, the processor 201 implements any of the circuit board production control methods based on data monitoring.

[0116] Based on the above, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0117] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems or devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

Claims

1. A circuit board production control method based on data monitoring, characterized in that: include: Collecting a real-time monitoring data set of a target circuit board production line, wherein the real-time monitoring data set includes a plurality of continuous time series of process parameter monitoring data and equipment operation monitoring data; Performing multi-dimensional time series feature extraction processing on the real-time monitoring data set to generate production time series state features and process fluctuation correlation features of the target circuit board production line; Calling a production decision model to perform state fusion processing on the production time sequence state characteristics and the process fluctuation correlation characteristics to generate production control decision information for the target circuit board production line; Generating a set of equipment control instructions according to the production control decision information, and transmitting the set of equipment control instructions to an execution terminal of the target circuit board production line to implement a production status adjustment operation; The method further comprises: Converting the production control decision information into visual chart data, the visual chart data including a process parameter adjustment trend chart, an equipment operation status comparison chart, and a production line efficiency prediction chart; sending the visual chart data to a production management terminal for display, and upon detecting a decision correction instruction responded by the production management terminal, re-executing the multi-dimensional time series feature extraction process and the state fusion process to update the production control decision information; The method further comprises: Collect a response feedback data set of the execution terminal, wherein the response feedback data set includes a sequence of actual adjustment values ​​of process parameters and a sequence of equipment operation status updates; perform execution effect feature extraction on the response feedback data set to generate actual process fluctuation features and actual equipment status timing features; perform time-series alignment on the actual process fluctuation features and the process fluctuation correlation features, and calculate the difference to generate a process adjustment deviation coefficient set; perform window sliding correlation analysis on the actual equipment status timing features and the production timing status features to generate an equipment control lag time parameter set; adjust the control magnitude and execution timing interval in the equipment control instruction set according to the process adjustment deviation coefficient set and the equipment control lag time parameter set, generate a compensation control instruction set, and transmit it additionally to the execution terminal.

2. The method according to claim 1, characterized in that The performing of multi-dimensional time series feature extraction processing on the real-time monitoring data set to generate production time series state features and process fluctuation correlation features of the target circuit board production line includes: Performing data cleaning on the real-time monitoring data set to obtain an optimized monitoring data set; the data cleaning includes: replacing abnormal data points in the process parameter monitoring data with adjacent time series data, and aligning time windows for discontinuously collected data in the equipment operation monitoring data; Performing time series fluctuation analysis on the optimized monitoring data set to identify fluctuation characteristics of process parameters and equipment operating status of the target circuit board production line over multiple consecutive production cycles; The process fluctuation correlation feature of the target circuit board production line is generated based on the process parameter fluctuation representation, and the production timing state feature is generated based on the equipment operation state fluctuation representation.

3. The method according to claim 2, characterized in that The performing of time series fluctuation analysis on the optimized monitoring data set to identify fluctuation characteristics of process parameters and equipment operation status of the target circuit board production line over multiple consecutive production cycles includes: Dividing the optimized monitoring data set into a process parameter time series sequence and an equipment operation status time series sequence; Performing sliding window segmentation processing on the process parameter time series to obtain multiple process parameter window subsequences, and performing trend fitting on each of the process parameter window subsequences to determine the fluctuation direction and fluctuation amplitude of the process parameter window subsequence; Performing state segmentation processing on the device operation state time series to obtain a plurality of device operation state subintervals, and performing stability evaluation on each of the device operation state subintervals to determine an operation offset and an offset duration of the device operation state subinterval; The process parameter fluctuation representation and the equipment operation status fluctuation representation are generated according to the fluctuation direction, the fluctuation amplitude, the operation offset and the offset duration.

4. The method according to claim 3, characterized in that The calling of the production decision model to perform state fusion processing on the production time sequence state characteristics and the process fluctuation correlation characteristics to generate production control decision information of the target circuit board production line includes: Mapping the production time series state feature into a first state vector, mapping the process fluctuation correlation feature into a second state vector, and performing vector splicing processing on the first state vector and the second state vector to obtain a fused state vector; Calling the multilayer perceptron in the production decision model to perform nonlinear transformation processing on the fused state vector to generate a state prediction result of the target circuit board production line; The production control decision information is determined by matching the state prediction result with a predefined production control strategy library.

5. The method according to claim 4, characterized in that The step of matching a predefined production control strategy library according to the state prediction result to determine the production control decision information includes: Analyzing the process parameter fluctuation amplitude set, equipment operation offset set, and production line efficiency deviation index included in the state prediction result to generate a current prediction parameter set; Obtaining a historical process parameter fluctuation amplitude set, a historical equipment operation offset set, and a historical production line efficiency deviation index corresponding to each historical control strategy in the predefined production control strategy library, and performing dimensional alignment processing on the historical process parameter fluctuation amplitude set and the historical equipment operation offset set with the current prediction parameter set to generate a normalized historical parameter set; Calculating a multi-dimensional similarity value between the current prediction parameter set and each of the normalized historical parameter sets, the multi-dimensional similarity value including a process parameter fluctuation direction consistency coefficient, a device operation offset timing matching degree, and an efficiency deviation weighted distance; Sort the historical control strategies in the predefined production control strategy library according to the multi-dimensional similarity values, and select the target historical control strategy that meets the similarity threshold and has the highest priority; Extracting the process adjustment parameter mapping table, equipment control instruction sequence, and execution timing constraints in the target historical control strategy, and performing parameter correction in combination with the real-time production line efficiency deviation index in the current prediction parameter set to generate the production control decision information; The dimensional alignment process includes: performing a linear conversion from Celsius to a standard temperature scale on the temperature parameters in the historical process parameter fluctuation amplitude set, performing a decimal scaling from kilopascals to megapascals on the pressure parameters, and performing a unit unification from milliseconds to seconds on the time dimension parameters in the historical equipment operation offset set; Among them, the calculation of the multi-dimensional similarity value meets the following conditions: the consistency coefficient of the process parameter fluctuation direction is determined by the slope sign matching degree of the corresponding process parameters in the current prediction parameter set and the normalized historical parameter set; the equipment operation offset timing matching degree is calculated by the time warping algorithm to calculate the minimum path distance of the two equipment operation offset sequences; the efficiency deviation weight distance is fused by the weighted Euclidean distance formula to obtain the normalized difference of the production line efficiency deviation index.

6. The method according to claim 4, characterized in that The training method of the production decision model includes: Obtain the historical monitoring data set of the historical production line and the corresponding historical control decision set; Performing multi-dimensional time series feature extraction processing on the historical monitoring data set to generate historical production time series state features and historical process fluctuation correlation features; Inputting the historical production time series state characteristics and the historical process fluctuation correlation characteristics as training samples into an initial production decision model to generate predictive control decision information; constructing a loss function according to the difference between the predicted control decision information and the historical control decision set, and iteratively training the initial production decision model based on the loss function until convergence, thereby obtaining the production decision model; The constructing a loss function according to the difference between the predicted control decision information and the historical control decision set includes: Performing vectorized coding processing on the prediction control decision information to obtain a prediction decision vector, and performing vectorized coding processing on the historical control decision set to obtain a historical decision vector; Calculating a cosine similarity between the predicted decision vector and the historical decision vector, determining a first loss component based on the cosine similarity, calculating a Euclidean distance between the predicted decision vector and the historical decision vector, and determining a second loss component based on the Euclidean distance; Performing a weighted summation on the first loss component and the second loss component to obtain the loss function.

7. The method according to claim 1, characterized in that Generating a set of equipment control instructions according to the production control decision information includes: parsing the equipment adjustment type and equipment adjustment range in the production control decision information; Matching the corresponding execution device identification in the target circuit board production line according to the device adjustment type, and generating the device control quantitative feature of the execution device identification according to the device adjustment range; Performing instruction encapsulation processing on the execution device identifier and the device control quantitative feature to generate the device control instruction set; The transmitting the device control instruction set to the execution terminal of the target circuit board production line to implement the production status adjustment operation includes: performing instruction priority sorting on the device control instruction set to determine the execution order of each of the device control instructions; Transmitting the device control instruction set in sequence to the execution terminal of the target circuit board production line according to the execution order; The response status of the execution terminal to the device control instruction set is monitored, and when a response abnormality is detected, a backup device control instruction set is regenerated and transmitted.

8. The method according to claim 1, characterized in that When the decision correction instruction responded by the production management terminal is detected, re-executing the multi-dimensional time series feature extraction process and the state fusion process to update the production control decision information includes: Parsing the process parameter adjustment amount set, the equipment operation offset compensation coefficient set, and the production cycle timing constraint condition set in the decision correction instruction to generate a correction parameter set; The process parameter adjustment amount set of the correction parameter set is superimposed on the process parameter monitoring data of the optimized monitoring data set according to the corresponding time series window to generate a correction process parameter sequence; the equipment operation offset compensation coefficient set of the correction parameter set is subjected to sliding window filtering and fusion with the equipment operation monitoring data of the optimized monitoring data set to generate a correction equipment operation sequence; According to the production cycle timing constraint condition set, the time intervals of the modified process parameter sequence and the modified equipment operation sequence are intercepted to generate a modified monitoring data set; the time series fluctuation analysis process is performed on the modified monitoring data set to generate an updated process parameter fluctuation representation and an equipment operation status fluctuation representation; based on the updated process parameter fluctuation representation, an updated process fluctuation correlation feature is generated; based on the updated equipment operation status fluctuation representation, an updated production time series status feature is generated; Perform time series dimension alignment processing on the updated process fluctuation correlation feature and the process fluctuation correlation feature. When the feature lengths are inconsistent, endpoint replication or mean padding is used to extend them to the same time step. Perform feature channel matching processing on the updated production time series status feature and the production time series status feature. When the number of channels is inconsistent, they are mapped to the same feature space through a fully connected layer. Performing vector concatenation on the updated process fluctuation correlation features after time series dimension alignment and the updated production time series state features after feature channel matching to generate a fused updated feature vector; calling the production decision model to perform the state fusion processing on the fused updated feature vector to generate intermediate control decision information; Calculating a set of absolute differences between the corresponding process adjustment parameters in the intermediate control decision information and the production control decision information and a set of cosine similarities of the equipment control instructions; screening conflicting control items whose absolute differences exceed a preset process threshold and whose cosine similarities are lower than a preset equipment threshold; performing a weighted sliding average correction on the process adjustment parameters in the conflicting control items, with the weights adjusted according to the time series fluctuation amplitude of the corresponding process parameters; Prioritizing the equipment control instructions in the conflicting control items, and determining a coverage order based on a time match between the equipment response delay parameter and the production cycle timing constraint; merging the corrected process adjustment parameters and the equipment control instructions after priority coverage into the production control decision information to generate updated production control decision information; The timing matching between the temperature adjustment step of the process adjustment parameter in the updated production control decision information and the voltage change rate of the equipment control instruction is corrected so that the millisecond timestamp of the temperature adjustment window covers the voltage change interval.

9. A circuit board production control system, characterized in that: include: processor; A storage device having a computer program stored thereon, wherein when the computer program is executed by the processor, the processor implements the circuit board production control method based on data monitoring as described in any one of claims 1-8.

Images