Circuit board production data management method and system based on big data

Abstract

The application relates to the technical field of electronic data processing, and discloses a circuit board production data management method and system based on big data, which comprises the following steps: obtaining original production data of a circuit board; constructing multi-dimensional feature vectors according to timestamps, belonging processes and data physical properties of each data point; clustering the multi-dimensional feature vectors into multiple process feature data layers by using a clustering algorithm and a distance measurement function; calculating cross-layer transmission entropy; when the cross-layer transmission entropy is greater than a preset threshold value and the direction is consistent with a process flow, establishing a directed dependence relationship edge and constructing a non-cyclic directed graph; extracting to-be-monitored circuit board production data and mapping the to-be-monitored circuit board production data to the process feature data layers; calculating mutual information between real-time data; and calculating a judgment threshold value; when the mutual information is lower than the judgment threshold value, it is determined that data correlation is abnormal. The application can structure and layer production data, quantify causal dependence between layers, and realize early warning of potential quality problems.

Description

Technical Field

[0001] This invention relates to the field of electronic data processing technology, and more specifically to a method and system for managing circuit board production data based on big data. Background Technology

[0002] The manufacturing process of printed circuit boards is extremely complex, encompassing dozens of precise physical and chemical processes such as drilling, electroplating, etching, and lamination. With the development of industrial automation and intelligence, systems such as Manufacturing Execution Systems (MES), equipment sensors, and Automated Optical Inspection (AOI) on the production line generate massive amounts of multi-dimensional data in real time. This data meticulously records key information such as equipment status, process parameters, material characteristics, and product quality.

[0003] Traditional data management often employs linear storage methods based on time series or production batches, simply stacking data from different processes and with different physical attributes. This makes it difficult to identify the inherent structure and correlations of the data across time and process dimensions. During analysis, methods such as statistical process control are typically used to independently monitor individual parameters of a single process. However, even minor fluctuations in upstream processes can affect downstream processes through complex transmission effects, leading to product defects.

[0004] During anomaly detection, if the parameters of two processes are both within specifications, but the historical strong correlation between them suddenly weakens, there may be a potential process anomaly. Furthermore, while simple correlation analysis can identify relationships between variables, it cannot determine the direction and intensity of the influence, making it difficult to construct a data model that reflects the causal transmission chain in the process flow.

[0005] Therefore, it is necessary to design a production data management method that can automatically structure and layer production data, quantify the causal dependencies between levels, and make anomaly detection based on this, so as to improve the level of intelligent control of the PCB production process. Summary of the Invention

[0006] This invention provides a method and system for managing circuit board production data based on big data to solve the problem in the prior art that it is impossible to structure and layer production data or quantify the causal dependencies between levels.

[0007] In a first aspect, the circuit board production data management method based on big data of the present invention includes the following steps: Obtain raw production data during the circuit board manufacturing process, and construct a multi-dimensional feature vector for each data point based on timestamp, process, and physical attributes. A density peak-based clustering algorithm is adopted, and a distance metric function that represents time proximity and upstream and downstream relationships of processes as nonlinear weights is used to cluster multidimensional feature vectors into multiple process feature data layers. Calculate the cross-layer transfer entropy between any two process feature data layers. When the cross-layer transfer entropy is greater than a preset threshold and the direction is consistent with the process flow, establish a directed dependency edge between the two process feature data layers, and construct an acyclic directed graph representing the inherent causal relationship of the data using the cross-layer transfer entropy as the edge weight. Production data of the production batch of the circuit board to be monitored is extracted and mapped to the corresponding process feature data layer. For any pair of process feature data layers connected by directed dependency edges in the acyclic directed graph, the mutual information between real-time data in the process feature data layer is calculated, and a judgment threshold positively correlated with the historical edge weight of the directed dependency edge is calculated. When the mutual information is lower than the judgment threshold, it is determined that the data association of the production batch of the circuit board to be monitored is abnormal.

[0008] Preferably, the step of constructing a multi-dimensional feature vector for each data point based on its timestamp, the process it belongs to, and its physical attributes includes: The timestamp is converted into the cumulative number of seconds since the start of the production batch; the process is converted into a binary vector using one-hot encoding to obtain the process vector; the physical attributes of the data are Z-score standardized to obtain standardized physical attribute values, including temperature, pressure, and chemical concentration; the cumulative number of seconds, the process vector, and the standardized physical attribute values ​​are concatenated in sequence to obtain the multidimensional feature vector.

[0009] Preferably, the distance metric function that represents time proximity and upstream / downstream process relationships as nonlinear weights is: ; in, Multidimensional feature vectors With multidimensional feature vectors The weighted distance between them Multidimensional feature vectors With multidimensional feature vectors The Euclidean distance between them Multidimensional feature vectors With multidimensional feature vectors Temporal proximity weight, Multidimensional feature vectors With multidimensional feature vectors The weight of process relationships.

[0010] Preferred, multidimensional feature vector With multidimensional feature vectors Temporal proximity weight Calculated using the following formula: ;in, The preset time decay coefficient, Multidimensional feature vectors With multidimensional feature vectors The difference in timestamps; Multidimensional feature vectors With multidimensional feature vectors Process relationship weight Calculated in the following way: Multidimensional feature vectors The process to be described is Multidimensional feature vectors The process to be described is ,when and When it is the same process, It is 1.0; when and When they are adjacent processes, It is 1.5; when and When they are not adjacent processes, It is version 3.0.

[0011] Preferably, the step of clustering multidimensional feature vectors into multiple process feature data layers includes: Set cutoff distance The median of the set of distance values ​​between all pairs of data points; for each multidimensional feature vector ,calculate Local density The local density To be among all other multidimensional feature vectors Distance less The number of vectors; Calculate multidimensional feature vectors minimum distance The minimum distance For all local densities greater than In the multidimensional feature vector, with The minimum distance between them; Select and The top 5% of the multidimensional feature vectors in the product are used as cluster centers, and the remaining multidimensional feature vectors are assigned to the process feature data layer to which the nearest cluster center belongs.

[0012] Preferably, the calculation of the cross-layer transfer entropy between any two process feature data layers includes: The two process feature data layers are the source data layer and the target data layer, respectively. The data from the source and target data layers are processed by extracting the mean of a preset physical attribute at each time step to form two continuous time series. These two continuous time series are then binned and discretized to obtain the discrete time series of the source data layer. and the discrete time series of the target data layer Calculate the cross-layer transfer entropy from X to Y using the following formula. ; in, For the cross-layer transfer entropy from X to Y, the joint probability Conditional probability Conditional probability All of these were obtained through frequency statistics of historical production data.

[0013] Preferably, the preset threshold is calculated in the following way: Extract data from the most recent 30 normal production batches in the historical database; calculate the cross-layer transfer entropy between all data layers to form a cross-layer transfer entropy sample set; sort all cross-layer transfer entropies greater than zero in the cross-layer transfer entropy sample set in ascending order; select the cross-layer transfer entropy at the 95th percentile after sorting as the preset threshold.

[0014] Preferably, the threshold for determining whether the weights of historical edges in a directed dependency relationship are positively correlated includes: Extract the historical edge weights W of directed dependency edges. The historical edge weights are the cross-layer transfer entropy calculated when constructing the acyclic directed graph. The determination threshold is calculated using the following formula: ;in, To determine the threshold, This is a preset proportional coefficient. , Historical edge weights.

[0015] Preferably, the mutual information between real-time data in the feature data layer of the calculation process includes: The real-time production data corresponding to two process feature data layers connected by directed dependency edges are used to form two real-time data sequences A and B by extracting the mean of a preset physical attribute of each process feature data layer at each time step; the mutual information between A and B is calculated according to the following formula. ; in, The mutual information between A and B is represented by the joint probability p(a, b), which is obtained by performing two-dimensional histogram statistics on real-time data sequences A and B. The marginal probability p(a) is obtained by performing one-dimensional histogram statistics on real-time data sequence A, and the marginal probability p(b) is obtained by performing one-dimensional histogram statistics on real-time data sequence B.

[0016] Secondly, the circuit board production data management system based on big data of the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the above-mentioned circuit board production data management method based on big data.

[0017] The beneficial effects of this invention are as follows: This invention can organize heterogeneous raw production data into a structured process feature data layer based on the inherent connections between the raw production data in terms of process flow and time. By constructing an acyclic directed graph through cross-layer transfer entropy calculation, the fuzzy and implicit influence relationships between various processes can be transformed into a causal transmission network. In anomaly detection, anomalies are identified by monitoring changes in the strength of data correlation on the constructed key causal paths. This allows for the detection of early risks where the parameter values ​​themselves are within the specification range, but their inherent correlation has weakened, thus achieving early warning of potential quality problems. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a big data-based circuit board production data management method provided in an embodiment of the present invention. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0020] like Figure 1 As shown, an embodiment of the circuit board production data management method based on big data provided by the present invention includes the following steps: S1: Obtain the raw production data during the circuit board manufacturing process, and construct a multi-dimensional feature vector for each data point based on the timestamp, the process to which it belongs, and the physical attributes of the data.

[0021] Raw production data such as temperature, pressure, current, speed, and chemical concentration are collected in real time from the Manufacturing Execution System (MES), Programmable Logic Controllers (PLCs) of equipment, and various sensors. Each data point is associated with a unique production batch number, equipment number, process code, and collection timestamp. For example, a data point of spindle speed of 30,000 revolutions per minute collected at 10:01:30 AM during the drilling process, with the process represented by the number 1 and the physical attribute spindle speed represented by the number 5, can be constructed as a four-dimensional feature vector containing the timestamp, process code 1, attribute code 5, and the value 30,000.

[0022] S2 employs a density peak-based clustering algorithm and utilizes a distance metric function that represents temporal proximity and upstream / downstream process relationships as nonlinear weights to cluster multidimensional feature vectors into multiple process feature data layers.

[0023] Specifically, for any two multidimensional feature vectors, their Euclidean distance in terms of time, process, and physical attribute values ​​is calculated. The Euclidean distance is then corrected using a distance metric function with nonlinear weights. This nonlinear weight distance metric function includes a time proximity weight function and a process relationship weight function. The time proximity weight function makes the weight between vectors whose timestamps differ by a few seconds approach 1, while the weight between vectors differing by several hours increases exponentially. The process relationship weight function makes the weight between vectors belonging to the same or adjacent processes smaller, while increasing the weight between vectors belonging to processes that are far apart in the process flow. A density peak clustering algorithm is used to calculate the local density of the multidimensional feature vectors (i.e., the number of neighbors whose corrected distance is less than a certain cutoff distance) and the minimum distance to the high-density point. Multidimensional feature vectors with both large local density and large minimum distance are selected as cluster centers, and the remaining multidimensional feature vectors are assigned to the process feature data layer to which the nearest cluster center belongs.

[0024] S3. Calculate the cross-layer transfer entropy between any two process feature data layers. When the cross-layer transfer entropy is greater than a preset threshold and the direction is consistent with the process flow, establish a directed dependency edge between the two process feature data layers, and construct an acyclic directed graph representing the inherent causal relationship of the data using the cross-layer transfer entropy as the edge weight.

[0025] Specifically, first, all data points contained in any two process feature data layers A and B are extracted and arranged into two continuous time series in chronological order. Then, the cross-layer transfer entropy from A to B is calculated. This value quantifies the degree to which the historical information of data layer A reduces the uncertainty of the future state of data layer B. For example, the cross-layer transfer entropy of lamination process temperature data layer A to subsequent curing process temperature data layer B is calculated. Simultaneously, the system determines whether A is an upstream process of B based on a preset process flow sequence. Only when the calculated cross-layer transfer entropy is greater than a threshold set based on historical experience, and data layer A is indeed upstream of B in the process flow, will the system establish a directed dependency edge from A to B between the nodes representing A and B, using the cross-layer transfer entropy value as the edge weight. By traversing all data layer pairs and repeating the above operations, a non-cyclic directed graph with process feature data layers as nodes and inherent causal relationships as edges can be constructed.

[0026] S4. Extract the production data of the production batch of the circuit board to be monitored and map it to the corresponding process feature data layer. For any pair of process feature data layers connected by directed dependency edges in the acyclic directed graph, calculate the mutual information between real-time data in the process feature data layer and calculate the judgment threshold that is positively correlated with the historical edge weight of the directed dependency edge. When the mutual information is lower than the judgment threshold, it is determined that the data association of the production batch of the circuit board to be monitored is abnormal.

[0027] Specifically, after a new production batch starts, the system calculates the corrected distance between the generated real-time data vector and all cluster centers, and assigns it to the process feature data layer represented by the nearest cluster center. For any pair of data layers A and B connected by dependency edges in an acyclic directed graph, the system sets a judgment threshold that is positively correlated with their historical edge weights. Subsequently, the system collects all real-time data mapped to layers A and B for the current batch and calculates the mutual information between these data. If the calculated mutual information value is lower than the set judgment threshold, it is determined that the data association strength between processes A and B of the production batch is weaker than the historical normal level, indicating a potential process association anomaly, and an early warning is issued.

[0028] In an optional embodiment, constructing a multi-dimensional feature vector for each data point based on its timestamp, the process it belongs to, and its physical attributes includes: The timestamp is converted into the cumulative number of seconds since the start of the production batch; the process is converted into a binary vector using one-hot encoding to obtain the process vector; the physical attributes of the data are Z-score standardized to obtain standardized physical attribute values, including temperature, pressure, and chemical concentration; the cumulative number of seconds, the process vector, and the standardized physical attribute values ​​are concatenated in sequence to obtain the multidimensional feature vector.

[0029] Specifically, taking a single data point as an example, the construction process of the multidimensional feature vector is explained in detail. First, timestamps and process information are processed: the timestamp is converted into the cumulative number of seconds since the start of the production batch, serving as the first numerical element of the vector (e.g., 100s). The process (e.g., exposure) is converted into a binary vector using one-hot encoding. If the entire production process includes three processes: cleaning, exposure, and etching, the exposure process is represented as a three-dimensional vector (0, 1, 0), immediately following the cumulative number of seconds. Second, Z-score standardization is performed on the physical properties: assuming the original physical properties of the data point are: temperature 25℃, pressure 101kPa, and chemical concentration 0.5mol / L, Z-score standardization is performed using historical statistical data (e.g., mean temperature 23℃, standard deviation 2℃; mean pressure 100kPa, standard deviation 0.5kPa; mean concentration 0.4mol / L, standard deviation 0.05mol / L). The standardized results are: temperature 1.0; pressure 2.0; concentration 2.0. Finally, the processed cumulative seconds, process vector, and standardized physical attribute values ​​are concatenated in sequence to obtain a multidimensional feature vector (100, 0, 1, 0, 1.0, 2.0, 2.0).

[0030] In an optional embodiment, the distance metric function that represents time proximity and upstream / downstream process relationships as nonlinear weights is: ; in, Multidimensional feature vectors With multidimensional feature vectors The weighted distance between them Multidimensional feature vectors With multidimensional feature vectors The Euclidean distance between them Multidimensional feature vectors With multidimensional feature vectors Temporal proximity weight, Multidimensional feature vectors With multidimensional feature vectors The weight of process relationships.

[0031] Multidimensional feature vectors With multidimensional feature vectors Temporal proximity weight Calculated using the following formula: ;in, The preset time decay coefficient, Multidimensional feature vectors With multidimensional feature vectors The difference in timestamps; Multidimensional feature vectors With multidimensional feature vectors Process relationship weight Calculated in the following way: Multidimensional feature vectors The process to be described is Multidimensional feature vectors The process to be described is ,when and When it is the same process, It is 1.0; when and When they are adjacent processes, It is 1.5; when and When they are not adjacent processes, It is version 3.0.

[0032] For example, let's calculate two multidimensional feature vectors using an example. and The weighted distance. Assuming vectors Generated during the exposure process at 100s, vector The etching process occurs at 120s in adjacent etching operations. By calculating the difference in characteristic values ​​between the two processes, the Euclidean distance d(i,j) is found to be 5.8. This Euclidean distance is then adjusted using temporal proximity weights and process relationship weights. The timestamp difference is 20s. If the preset time decay coefficient α is 0.01, then the temporal proximity weight... The value is approximately 0.82, reflecting the closer the data points are in time, the stronger the relationship. Since exposure and etching are adjacent steps in the process flow, the process relationship weight... The weight is set to 1.5, which is greater than the weight of 1.0 for processes in the same workflow and less than the weight of 3.0 for non-adjacent processes, reflecting the strength of the correlation between processes. The calculated weighted distance D(i,j) is approximately 7.13. The weighted distance takes into account the differences in the data themselves as well as their relative positions in time and process flow.

[0033] In an optional embodiment, clustering the multidimensional feature vectors into multiple process feature data layers includes: Set cutoff distance The median of the set of distance values ​​between all pairs of data points; for each multidimensional feature vector ,calculate Local density The local density To be among all other multidimensional feature vectors Distance less The number of vectors; Calculate multidimensional feature vectors minimum distance The minimum distance For all local densities greater than In the multidimensional feature vector, with The minimum distance between them; Select and The top 5% of the multidimensional feature vectors in the product are used as cluster centers, and the remaining multidimensional feature vectors are assigned to the process feature data layer to which the nearest cluster center belongs.

[0034] For example, in a dataset containing 1000 feature vectors, calculating the weighted distance between all vector pairs will produce nearly half a million distance values. After sorting these distance values, the value at the middle position is taken as the cutoff distance. For example, 15.2. Then, calculate the local density of each vector (e.g., vector A) in the dataset. Local density The calculation method is as follows: count how many vectors have a distance to vector A less than 15.2. Assume there are 50 such vectors. Then calculate the local density of vector A. It is 50.

[0035] Then, calculate the minimum distance of vector A. The specific method is as follows: Among all vectors with a local density greater than 50, find the one closest to vector A and record this distance value, let's say 25.6. If there is no point with a higher density than vector A, then the minimum distance is the farthest distance between vector A and all other points. Then, multiply the local density by the minimum distance to obtain a decision value. Repeat this calculation for all 1000 vectors and sort them according to the decision values, selecting the top 5% of 50 vectors as cluster centers. For each of the remaining 950 vectors, calculate the corrected distance to these 50 cluster centers and then assign it to the process feature data layer represented by the nearest cluster center.

[0036] In an optional embodiment, calculating the cross-layer transfer entropy between any two process feature data layers includes: The two process feature data layers are the source data layer and the target data layer, respectively. The data from the source and target data layers are processed by extracting the mean of a preset physical attribute at each time step to form two continuous time series. These two continuous time series are then binned and discretized to obtain the discrete time series of the source data layer. and the discrete time series of the target data layer Calculate the cross-layer transfer entropy from X to Y using the following formula. ; in, For the cross-layer transfer entropy from X to Y, the joint probability Conditional probability Conditional probability All of these were obtained through frequency statistics of historical production data.

[0037] For example, suppose we need to calculate the transfer entropy from the exposure data layer to the etching data layer, and choose temperature as the key physical property. First, extract the average temperature values ​​of these two data layers at each time step from historical data, thus forming two continuous time series, denoted as sequence X (exposure layer) and sequence Y (etching layer). For example, sequence X is (25.1, 25.3, 25.2, ...), and sequence Y is (35.5, 35.4, 35.6, ...). Next, these two continuous temperature sequences are binned and discretized. The entire temperature range is divided into finite discrete intervals, for example, 20℃ to 30℃ is divided into state 1, and 30℃ to 40℃ is divided into state 2. After discretization, the original continuous sequence is transformed into a discrete sequence composed of integer states. Then, by statistically analyzing the frequency of occurrence of the discrete sequence in a large amount of historical data, the various probabilities required for the transfer entropy can be calculated. For example, the frequency of combinations where the target layer Y is in state 2 at the next time step and in state 2 at the current time step, and the source layer X is in state 1 at the current time step, and the frequency of combinations where only the target layer Y is in state 2 at the next time step and in state 2 at the current time step, are calculated. Finally, the probability values ​​obtained through frequency statistics are substituted into the transfer entropy calculation formula and summed to obtain a numerical value that quantifies the amount of information transferred from the temperature change of the exposure layer to the temperature change of the etching layer.

[0038] In an optional embodiment, the preset threshold is calculated in the following manner: Extract data from the most recent 30 normal production batches in the historical database; calculate the cross-layer transfer entropy between all data layers to form a cross-layer transfer entropy sample set; sort all cross-layer transfer entropies greater than zero in the cross-layer transfer entropy sample set in ascending order; select the cross-layer transfer entropy at the 95th percentile after sorting as the preset threshold.

[0039] For example, the system first selects the 30 most recent qualified production batches from the production history. For each of these 30 qualified batches, the system performs data stratification and cross-stratum transfer entropy calculations. Assuming there are 10 process feature data layers, each batch requires calculating the transfer entropy value between 90 directed data layer pairs. Then, all the calculation results from the 30 batches are aggregated to obtain a sample set containing 2700 transfer entropy values. This sample set is then cleaned and a threshold is determined. Data cleaning removes all transfer entropy values ​​less than or equal to zero, as these values ​​do not represent effective information flow. Assuming 2500 positive values ​​remain after removal, these 2500 positive values ​​are sorted in ascending order, and the 95th percentile (i.e., the 2375th position) is determined. The transfer entropy value at this position is selected as the preset threshold. In subsequent causal relationship modeling, only when the calculated cross-stratum transfer entropy value exceeds this preset threshold is it considered a strong association requiring the establishment of a dependency edge.

[0040] In an optional embodiment, the determination threshold for determining a positive correlation between the weights of historical edges and directed dependency edges includes: Extract the historical edge weights W of directed dependency edges. The historical edge weights are the cross-layer transfer entropy calculated when constructing the acyclic directed graph. The determination threshold is calculated using the following formula: ;in, To determine the threshold, This is a preset proportional coefficient. , Historical edge weights.

[0041] For example, when monitoring the production process in real time, for a known dependency, such as a directed edge from the cleaning data layer to the exposure data layer, the system queries the historical weight W of that directed edge from an acyclic directed graph constructed based on historical normal data. This weight W is the cross-layer transfer entropy calculated historically, representing the average strength of the cleaning process's impact on the exposure process under normal production conditions. Subsequently, the system introduces a scaling factor β to adjust the monitoring sensitivity. A larger β value means that even a small decrease in information flow intensity will be detected by the system. A judgment threshold is calculated by multiplying the historical edge weight W by the scaling factor β. This judgment threshold is used to determine whether there has been an abnormal weakening of the real-time information intensity between the current cleaning layer and the exposure layer.

[0042] In an optional embodiment, the mutual information between real-time data in the computation process feature data layer includes: The real-time production data corresponding to two process feature data layers connected by directed dependency edges are used to form two real-time data sequences A and B by extracting the mean of a preset physical attribute of each process feature data layer at each time step; the mutual information between A and B is calculated according to the following formula. ; in, The mutual information between A and B is represented by the joint probability p(a, b), which is obtained by performing two-dimensional histogram statistics on real-time data sequences A and B. The marginal probability p(a) is obtained by performing one-dimensional histogram statistics on real-time data sequence A, and the marginal probability p(b) is obtained by performing one-dimensional histogram statistics on real-time data sequence B.

[0043] For example, to calculate the mutual information between the cleaning layer and the exposure layer in real time, the system collects data within a continuous time window (e.g., 10 minutes). First, pressure is selected as the representative attribute of the cleaning layer, and temperature as the representative attribute of the exposure layer. By calculating the mean of the corresponding attribute in these two data layers at each time step (e.g., every 5 seconds), two real-time data sequences, each with 120 data points, are obtained, denoted as Sequence A (pressure) and Sequence B (temperature), respectively. Next, a histogram method is used to estimate the probability distribution: the pressure value range of Sequence A is divided into 5 intervals, and the temperature value range of Sequence B is divided into 4 intervals. The probability distribution is then calculated by counting how many of the 120 data pairs fall within the range. In each cell of the two-dimensional grid, the estimated joint probability p(a, b) can be obtained. The frequencies of data from sequence A and sequence B falling into their respective one-dimensional intervals are counted to obtain estimated marginal probabilities p(a) and p(b). Finally, the probability values ​​obtained through histogram statistics are substituted into the mutual information calculation formula. The summation of the calculation results for all 20 cells yields the mutual information between the cleaning layer pressure and the exposure layer temperature within the current 10 minutes, representing the correlation strength between the two under the current production state.

[0044] The implementation principle of the big data-based circuit board production data management method in this invention is as follows: This invention can integrate diverse and structurally varied raw production data, and reconstruct it into a structured process feature data layer based on its inherent logical relationships in the process flow and time dimensions. Furthermore, by calculating the transfer entropy to construct an acyclic directed graph, it can effectively reveal the originally ambiguous and implicit mutual influences between processes and visualize them as a clear causal relationship network. Moreover, in terms of anomaly detection, this invention identifies risks by real-time monitoring changes in the strength of data associations along key causal paths. Even if all parameter readings are within the normal range, it can keenly capture early signs of weakening correlations between them, thereby achieving early warning of potential quality problems.

[0045] An embodiment of the circuit board production data management system based on big data provided by the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the circuit board production data management method based on big data in the above embodiment.

[0046] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for managing circuit board production data based on big data, characterized in that, Includes the following steps: Obtain raw production data during the circuit board manufacturing process, and construct a multi-dimensional feature vector for each data point based on timestamp, process, and physical attributes. A density-peak-based clustering algorithm is adopted, and a distance metric function that represents temporal proximity and upstream / downstream process relationships as nonlinear weights is used, as follows: ； , , , Multidimensional feature vectors With multidimensional feature vectors Weighted distance, Euclidean distance, temporal proximity weight, and process relationship weight between them; ; The preset time decay coefficient, Multidimensional feature vectors With multidimensional feature vectors The difference in timestamps; Multidimensional feature vectors , The process to be described is , ,when and When the processes are the same, adjacent, or non-adjacent, The values ​​are 1.0, 1.5, and 3.

0. Cluster multidimensional feature vectors into multiple process feature data layers; Calculate the cross-layer transfer entropy between any two process feature data layers, including: The two process feature data layers are the source data layer and the target data layer, respectively. The data from the source and target data layers are processed by extracting the mean of a preset physical attribute at each time step to form two continuous time series. These two continuous time series are then binned and discretized to obtain the discrete time series of the source data layer. and the discrete time series of the target data layer Cross-layer transfer entropy from X to Y for: ; Joint probability Conditional probability and All data were obtained through frequency statistics of historical production data; When the cross-layer transfer entropy is greater than the preset threshold and the direction is consistent with the process flow, a directed dependency edge is established between the two process feature data layers, and an acyclic directed graph representing the inherent causal relationship of the data is constructed with the cross-layer transfer entropy as the edge weight. Extract production data of the production batch of the circuit board to be monitored and map it to the corresponding process feature data layer. For any pair of process feature data layers connected by directed dependency edges in the acyclic directed graph, calculate the mutual information between real-time data in the process feature data layer. And calculate the threshold for determining whether the weights of historical edges in directed dependency relationships are positively correlated, including: Extract the historical edge weights W of directed dependency edges. The historical edge weights are the cross-layer transfer entropy calculated when constructing the acyclic directed graph. Determination threshold for: , This is a preset proportional coefficient. , Historical edge weights; When the mutual information is lower than the judgment threshold, it is determined that there is an anomaly in the data association of the production batch of the circuit board to be monitored.

2. The circuit board production data management method based on big data according to claim 1, characterized in that, The process of constructing a multi-dimensional feature vector for each data point based on its timestamp, the process it belongs to, and its physical attributes includes: The timestamp is converted into the cumulative number of seconds since the start of the production batch; the process is converted into a binary vector using one-hot encoding to obtain the process vector; the physical attributes of the data are Z-score standardized to obtain standardized physical attribute values, including temperature, pressure, and chemical concentration; the cumulative number of seconds, the process vector, and the standardized physical attribute values ​​are concatenated in sequence to obtain the multidimensional feature vector.

3. The circuit board production data management method based on big data according to claim 1, characterized in that, The process of clustering multidimensional feature vectors into multiple process feature data layers includes: Set cutoff distance The median of the set of distance values ​​between all pairs of data points; for each multidimensional feature vector ,calculate Local density The local density To be among all other multidimensional feature vectors Distance less The number of vectors; Calculate multidimensional feature vectors minimum distance The minimum distance For all local densities greater than In the multidimensional feature vector, with The minimum distance between them; Select and The top 5% of the multidimensional feature vectors in the product are used as cluster centers, and the remaining multidimensional feature vectors are assigned to the process feature data layer to which the nearest cluster center belongs.

4. The circuit board production data management method based on big data according to claim 1, characterized in that, The preset threshold is calculated in the following way: Extract data from the most recent 30 normal production batches in the historical database; calculate the cross-layer transfer entropy between all data layers to form a cross-layer transfer entropy sample set; sort all cross-layer transfer entropies greater than zero in the cross-layer transfer entropy sample set in ascending order; The cross-layer transfer entropy located at the 95th percentile after arrangement is selected as the preset threshold.

5. The circuit board production data management method based on big data according to claim 1, characterized in that, The mutual information between real-time data in the feature data layer of the calculation process includes: The real-time production data corresponding to two process feature data layers connected by directed dependency edges are used to form two real-time data sequences A and B by extracting the mean of a preset physical attribute of each process feature data layer at each time step; the mutual information between A and B is calculated according to the following formula. ； in, The mutual information between A and B is represented by the joint probability p(a, b), which is obtained by performing two-dimensional histogram statistics on real-time data sequences A and B. The marginal probability p(a) is obtained by performing one-dimensional histogram statistics on real-time data sequence A, and the marginal probability p(b) is obtained by performing one-dimensional histogram statistics on real-time data sequence B.

6. A circuit board production data management system based on big data, characterized in that, It includes a memory and a processor. The memory stores computer instructions. When the processor executes the computer instructions, it implements the circuit board production data management method based on big data as described in any one of claims 1-5.

Images