A production data anomaly identification method for production process control

Abstract

The present application relates to the technical field of industrial big data, and particularly relates to a production data abnormality identification method for production process control, comprising: obtaining each dimension of to-be-analyzed data segments; classifying each dimension of to-be-analyzed data segments to obtain a plurality of sets of each dimension; obtaining abnormal factors of the sets according to the similarity between to-be-analyzed data segments in the sets; obtaining the coupling degree of hidden dangers between different dimensions according to to-be-analyzed data segments of corresponding time periods in different dimensions and defect types corresponding to the to-be-analyzed data segments; obtaining the overall abnormality degree of each dimension in the current time period in combination with production data of each dimension at the current time, to-be-analyzed data segments and abnormal factors of the corresponding sets of the to-be-analyzed data segments; and judging whether each dimension in the current time period is abnormal. The present application analyzes the coupling relationship of multi-dimensional production parameters through historical defect data, thereby improving the accuracy of real-time identification of abnormal data in injection molding production.

Description

Technical Field

[0001] This invention relates to the field of industrial big data technology, specifically to a method for identifying anomalies in production data for production process control. Background Technology

[0002] During injection molding production, the quality of molded products is influenced by a variety of production parameters, such as melt temperature, holding pressure, cooling time, and screw speed. These parameters are typically collected and recorded in real time by the equipment control system and correlated with the final product's pass rate during production quality inspection. However, existing methods for identifying anomalies in production data often rely on threshold judgments for single parameters or statistical methods to analyze fluctuations in single-dimensional data. While these methods can detect significant deviations from normal ranges to some extent, they struggle to identify hidden combinations of anomalies between different parameters within the normal range. Summary of the Invention

[0003] This invention provides a production data anomaly identification method for production process control to solve existing problems: traditional methods that rely on threshold judgment of a single parameter or fluctuation analysis of single-dimensional data cannot accurately monitor whether anomalies occur in various production data of injection molding production.

[0004] The present invention provides a method for identifying production data anomalies in production process control, which employs the following technical solution:

[0005] Includes the following steps:

[0006] Based on the production time period corresponding to the historical defective products, the data segments to be analyzed in each dimension are selected from the historical data of each dimension;

[0007] The data segments to be analyzed in each dimension are classified to obtain several sets for each dimension; the homogeneity of the sets is obtained based on the similarity between the data segments to be analyzed within each set; and the outlier factors of the sets are obtained based on the homogeneity of each set and the differences in homogeneity between different sets in the same dimension.

[0008] Based on the data segments to be analyzed corresponding to the same time period in different dimensions, and the defect types corresponding to the data segments to be analyzed, the data segments to be analyzed in different dimensions are divided into several synchronous groups and synchronous clusters are constructed based on the synchronous groups. Based on the similarity between the data segments to be analyzed within the synchronous groups of different synchronous clusters in different dimensions and the abnormal factors of the corresponding sets, the non-coupling coefficients between different synchronous groups are obtained, and then the degree of coupling of hidden dangers between different dimensions is obtained.

[0009] Based on the similarity between each dimension in the current time period and the data segment to be analyzed, and combined with the abnormal factors of the corresponding set of the data segment to be analyzed for each dimension, the degree of single-dimensional anomaly of each dimension in the current time period is obtained; based on the degree of single-dimensional anomaly of each dimension in the current time period and the degree of potential coupling between different dimensions, the overall degree of anomaly of each dimension in the current time period is obtained, and then it is determined whether each dimension in the current time period is abnormal.

[0010] Preferably, the specific method for classifying the data segments to be analyzed in each dimension to obtain several sets for each dimension includes:

[0011] For the i-th dimension, according to the defect type corresponding to each data segment to be analyzed in the i-th dimension, several data segments to be analyzed in the i-th dimension with the same defect type are grouped into the same set, resulting in several sets of the i-th dimension.

[0012] Preferably, the method for obtaining the homogeneity of the set based on the similarity between the data segments to be analyzed within the set includes:

[0013] For the u-th set in the i-th dimension, perform pairwise matching on all data segments to be analyzed in the u-th set in the i-th dimension to obtain several matching pairs;

[0014] For any matching pair, the DTW algorithm is used to obtain the DTW distance between the two data segments to be analyzed in the matching pair; the DTW distance between the two data segments to be analyzed in the matching pair is negatively correlated and normalized, and the result of the negative correlation normalization is used as the similarity between the two data segments to be analyzed in the matching pair.

[0015] The average similarity between the two data segments to be analyzed within all matching pairs is used as the homogeneity of the u-th set in the i-th dimension.

[0016] Preferably, the method for obtaining the outlier factor of a set based on the homogeneity of each set and the homogeneity differences between different sets of the same dimension includes:

[0017]

[0018] In the formula, Q i,u Represents the outlier factor of the u-th set in the i-th dimension; n i D represents the number of sets in the i-th dimension; i,u D represents the homogeneity of the u-th set in the i-th dimension; i,v represents the homogeneity of the v-th set in the i-th dimension; sigmoid() represents the sigmoid function.

[0019] Preferably, the specific method for dividing the data segments to be analyzed in different dimensions into several synchronous groups and constructing synchronous clusters based on the synchronous groups, according to the data segments to be analyzed in different dimensions corresponding to the same time period and the defect type corresponding to the data segments to be analyzed, includes:

[0020] For any two dimensions, according to the time period corresponding to the data segments to be analyzed in the two dimensions, the data segments to be analyzed in the two dimensions with the same corresponding time period are grouped into the same synchronous group, resulting in several synchronous groups of the two dimensions; according to the defect type corresponding to the data segments to be analyzed in each synchronous group of the two dimensions, the synchronous groups with the same defect type are grouped into the same synchronous cluster, resulting in several synchronous clusters of the two dimensions.

[0021] Preferably, the method for obtaining the non-coupling coefficients between different synchronic groups based on the similarity between the data segments to be analyzed within the synchronic groups of different synchronic clusters in different dimensions and the anomaly factors of the corresponding sets includes:

[0022] For any two dimensions, the b-th synchronic group in the a-th synchronic cluster of the two dimensions is denoted as the target group, and all synchronic groups in all synchronic clusters of the two dimensions except for the a-th synchronic cluster are denoted as the comparison group;

[0023] For a target group and any comparison group, obtain the similarity between two data segments to be analyzed in any dimension of the target group and the comparison group; obtain the similarity between two data segments to be analyzed in that dimension of the target group and the comparison group, and take the product of the similarity between two data segments to be analyzed in each of the two dimensions of the target group and the comparison group as the similarity between the target group and the comparison group.

[0024] The product of the abnormal factors of the corresponding sets of the data segments to be analyzed in the two dimensions of the comparison group is used as the abnormality coefficient of the comparison group, and the ratio of the similarity between the target group and the comparison group to the abnormality coefficient of the comparison group is used as the decoupling coefficient between the target group and the comparison group.

[0025] Preferably, the specific method for obtaining the degree of coupling between potential hazards in different dimensions is as follows:

[0026] The mean of the uncoupling coefficients between the target group and all comparison groups is negatively correlated and normalized. The result of the negative correlation normalization of the mean of the uncoupling coefficients is used as the hidden danger coupling factor of the target group.

[0027] Obtain the hazard coupling factor for all synchronous groups, and use the mean of the hazard coupling factors for all synchronous groups as the degree of hazard coupling between the two dimensions.

[0028] Preferably, the specific method for obtaining the single-dimensional anomaly degree of each dimension in the current time period based on the similarity between each dimension and the data segment to be analyzed, combined with the anomaly factors of the corresponding sets of the data segments to be analyzed in each dimension, includes:

[0029] The average time length of all data segments to be analyzed is recorded as the baseline time length. The current time period is composed of all times within the previous baseline time length and the current time.

[0030] For any data segment to be analyzed in the i-th dimension, the product of the similarity between the i-th dimension of the current time period and the i-th dimension of the data segment to be analyzed, multiplied by the anomaly factor of the set corresponding to the i-th dimension of the data segment to be analyzed, is used as the anomalous correlation coefficient between the i-th dimension of the data segment to be analyzed and the i-th dimension of the current time period.

[0031] The average of the abnormal correlation coefficients between all data segments to be analyzed in the i-th dimension and the i-th dimension in the current time period is taken as the degree of unidimensional abnormality in the i-th dimension in the current time period.

[0032] Preferably, the method for obtaining the overall anomaly level of each dimension in the current time period based on the single-dimensional anomaly level of each dimension and the degree of coupling of potential risks between different dimensions includes:

[0033]

[0034] In the formula, G i This represents the overall anomaly level of the i-th dimension in the current time period; N represents the number of dimensions; F i F represents the degree of single-dimensional anomaly in the i-th dimension of the current time period; j E represents the degree of unidimensional anomaly in the j-th dimension of the current time period; i,j This indicates the degree of coupling between the i-th and j-th dimensions; sigmoid() represents the sigmoid function.

[0035] Preferably, the specific method for determining whether each dimension of the current time period is abnormal includes:

[0036] A preset overall anomaly threshold γ is defined. If the overall anomaly level of the i-th dimension in the current time period is greater than or equal to γ, then the production data of the i-th dimension in the current time period is considered abnormal production data.

[0037] The beneficial effects of the technical solution of the present invention are as follows: Based on the production time period corresponding to the historical defect injection molded products, the data segments to be analyzed in each dimension are selected from historical data of each dimension; the data segments to be analyzed in each dimension are classified to obtain several sets for each dimension. During the injection molding production process, when a certain production data is abnormal, the resulting defects in the injection molded products are stable. Therefore, the data segments to be analyzed in each dimension are divided into several sets according to the defect type to accurately extract the unique abnormal features of the production data corresponding to various types of defects, thereby obtaining the abnormal factors of each set in all dimensions. This better facilitates subsequent analysis of the impact of each dimension on production, accurately identifying abnormal production data during the injection molding production process, and obtaining the abnormal factors of the set based on the similarity between the data segments to be analyzed within the set.

[0038] Furthermore, since the data segments to be analyzed in each dimension of injection molding production all originate from production data of various dimensions during the time periods in historical injection molding production where defective injection molded products occurred, it is possible that not all production data segments of different dimensions will exhibit anomalies within the same time period. Moreover, production data of different dimensions may produce injection molded products with the same defect type under certain combinations. Therefore, it is necessary to classify data segments with the same defect type before analysis to ensure consistent defect types during comparison. At the same time, the defect types caused by production data of different dimensions under certain combinations are stable. Thus, based on the data segments to be analyzed in the same time period in different dimensions and the defect types corresponding to the data segments to be analyzed, the degree of potential coupling between different dimensions is obtained. Combining the production data of each dimension at the current moment, the data segments to be analyzed, and the anomaly factors of their corresponding sets, the overall anomaly degree of each dimension in the current time period is obtained, and it is determined whether each dimension in the current time period is abnormal, thereby improving the accuracy of real-time identification of abnormal data in injection molding production. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating the steps of a production data anomaly identification method for production process control according to the present invention. Detailed Implementation

[0041] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a production data anomaly identification method for production process control proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0042] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0043] The following description, in conjunction with the accompanying drawings, details a specific scheme for a production data anomaly identification method for production process control provided by the present invention.

[0044] Please see Figure 1 The diagram illustrates a flowchart of a method for identifying production data anomalies in production process control according to an embodiment of the present invention. The method includes the following steps:

[0045] Step S001: Based on the production time period corresponding to the historical defective products, select the data segments to be analyzed from the historical data of each dimension.

[0046] It should be noted that this embodiment is a production data anomaly identification method for production process control. Its purpose is to identify abnormal production data in the injection molding process. The production data in the injection molding process includes, but is not limited to, the temperature of each section of the barrel, the mold temperature, the injection pressure, the holding pressure, and the hydraulic oil temperature. If abnormal production data occurs during the injection molding process, it will cause defects such as shrinkage marks, flash, warping, and bubbles in the produced injection molded products. In order to avoid defects in the produced injection molded products, it is necessary to monitor the production data for anomalies in real time during the injection molding process to ensure the quality of the produced injection molded products.

[0047] Specifically, production data for each dimension is obtained within the time period corresponding to the production of defective products in the history of injection molding. Several production data points for any dimension within a time period corresponding to the production of defective products are arranged in chronological order to form a data segment to be analyzed for that dimension, resulting in several data segments to be analyzed for each dimension.

[0048] It should be noted that the data segments to be analyzed in each dimension are the production data of each dimension in the time period corresponding to the historical injection molding production of defective products. An anomaly in the production data of one dimension can cause the injection molded product to be defective. That is, in all the data segments to be analyzed in the time period corresponding to a defective product, not all data segments to be analyzed are abnormal, but at least one data segment to be analyzed is abnormal.

[0049] At this point, we have obtained several data segments to be analyzed for each dimension.

[0050] Step S002: Classify the data segments to be analyzed in each dimension to obtain several sets in each dimension; obtain the homogeneity of the sets based on the similarity between the data segments to be analyzed within each set; obtain the outlier factors of the sets based on the homogeneity of each set and the homogeneity differences between different sets in the same dimension.

[0051] It should be noted that this embodiment is a production data anomaly identification method for production process control. When a certain production data is abnormal during the injection molding process, the resulting product defects are stable. For example, abnormal injection pressure will likely cause flash defects in injection molded products. Therefore, the data segments to be analyzed in each dimension can be divided into several sets by defect type to accurately extract the exclusive production data anomaly features corresponding to each type of defect, and then obtain the anomaly factors of each set in all dimensions. This will better facilitate the subsequent analysis of the impact of various dimensions on production and accurately identify abnormal production data in the injection molding process.

[0052] Preferably, in a specific embodiment of the present invention, for the i-th dimension, according to the defect type corresponding to each data segment to be analyzed in the i-th dimension, several data segments to be analyzed in the i-th dimension of the same defect type are grouped into the same set to obtain several sets in the i-th dimension; in particular, for any set in the i-th dimension, this embodiment takes the time period corresponding to the data segment to be analyzed that is farthest from the current time in the set in the i-th dimension as the base time period of the set in the i-th dimension, sorts each set in the i-th dimension in time sequence according to the base time period of each set in the i-th dimension, and assigns an index label to each set in the i-th dimension accordingly.

[0053] For the u-th set in the i-th dimension, perform pairwise matching on all data segments to be analyzed in the u-th set in the i-th dimension to obtain several matching pairs;

[0054] For any matching pair, the DTW algorithm is used to obtain the DTW distance between the two data segments to be analyzed within the matching pair. Since the DTW clustering algorithm is a well-known existing technology, it will not be described in detail in this embodiment. The DTW distance between the two data segments to be analyzed within the matching pair is negatively correlated and normalized (in this embodiment, the exp(-x) model is used to present the negative correlation normalization process, where x is the input of the model. The implementer can set the negative correlation normalization function according to the actual situation). The result of the negative correlation normalization is used as the similarity between the two data segments to be analyzed within the matching pair.

[0055] Furthermore, the mean similarity between the two data segments to be analyzed within all matching pairs is used as the homogeneity of the u-th set in the i-th dimension.

[0056] It should be noted that the data segments to be analyzed in each dimension are production data for each dimension within the time period corresponding to the historical injection molding production of defective products. That is, not all data segments to be analyzed within the same time period are abnormal. The homogeneity of the set represents the stability of the data segments to be analyzed within the set. Since production data should be stable under normal circumstances, the smaller the homogeneity of the set, the more likely the production data is to be abnormal. At the same time, the smaller the homogeneity of a set is compared with the homogeneity of other sets, the more concentrated the characteristics of the defect type corresponding to that set are in the dimension corresponding to that set. Therefore, this can be used as a basis to obtain the abnormality factor of the set.

[0057] Preferably, in a specific embodiment of the present invention, for the u-th set in the i-th dimension, based on the homogeneity of the u-th set in the i-th dimension, and combined with the difference between the homogeneity of the u-th set in the i-th dimension and the homogeneity of other sets in the i-th dimension, the anomaly factor of the u-th set in the i-th dimension is obtained, and the specific calculation formula is as follows:

[0058]

[0059] In the formula, Q i,u Represents the outlier factor of the u-th set in the i-th dimension; n i D represents the number of sets in the i-th dimension; i,u D represents the homogeneity of the u-th set in the i-th dimension; i,v The i-th dimension represents the homogeneity of the v-th set; sigmoid() represents the sigmoid function, which is used for normalization in this embodiment.

[0060] It should be noted that the greater the homogeneity of a set, the more stable the data segment to be analyzed within the set; however, under normal circumstances, production data should be stable during the production process, therefore D i,uThe smaller the value, the more likely the data segment to be analyzed in the u-th set of the i-th dimension is abnormal production data; and the smaller the homogeneity of the u-th set of the i-th dimension is compared to the homogeneity of other sets in the i-th dimension, the more concentrated the characteristics of the defect type corresponding to the u-th set of the i-th dimension are on the dimension corresponding to that set, meaning the defect type corresponding to the u-th set is more likely to be caused by anomalies in the data segment to be analyzed within the u-th set of the i-th dimension; therefore The larger the value, the more anomalous the data segment to be analyzed in the u-th set of the i-th dimension is.

[0061] Thus, the anomaly factors for each set in each dimension are obtained.

[0062] Step S003: Based on the data segments to be analyzed corresponding to the same time period in different dimensions, and the defect type corresponding to the data segments to be analyzed, divide the data segments to be analyzed in different dimensions into several synchronous groups and construct synchronous clusters based on the synchronous groups. Based on the similarity between the data segments to be analyzed within the synchronous groups of different synchronous clusters in different dimensions and the abnormal factors of the corresponding sets, obtain the non-coupling coefficients between different synchronous groups, and then obtain the degree of coupling of hidden dangers between different dimensions.

[0063] It should be noted that in the actual injection molding production process, the various production data are not independent of each other, and there are certain correlations between them. For example, when the melt temperature is 195℃ and the holding pressure is 61MPa, shrinkage defects will always occur. This embodiment is a production data anomaly identification method for production process control. Its purpose is to identify abnormal production data in the injection molding process. In order to accurately identify abnormal production data in the injection molding process, it is necessary to analyze the impact of the correlation between multi-dimensional production data on product defects and obtain the degree of coupling of potential problems between dimensions.

[0064] It should be further clarified that, since the data segments to be analyzed for each dimension are all derived from production data of each dimension during the time period in historical injection molding production where defective products occurred, not all production data segments of different dimensions may exhibit anomalies within the same time period. Because production data from different dimensions produces injection molded products with the same defect type under specific combinations, it is necessary to classify data segments with the same defect type before analysis to ensure consistent defect types for comparison. Furthermore, since the defect types caused by production data from different dimensions are stable under specific combinations—for example, when the melt temperature is 195℃ and the holding pressure is 61MPa, shrinkage defects always appear, rather than flash defects—this allows for accurate identification of potential implicit coupling relationships between dimensions.

[0065] For any two dimensions (all dimensions are combined in pairs), according to the time periods corresponding to the data segments to be analyzed in the two dimensions, the data segments to be analyzed in the two dimensions with the same corresponding time periods are grouped into the same synchronous group, resulting in several synchronous groups of the two dimensions (the time periods corresponding to the two data segments to be analyzed in the synchronous group are the same, so the product defect types corresponding to the two data segments to be analyzed in the synchronous group are the same); according to the defect types corresponding to the data segments to be analyzed in each synchronous group of the two dimensions, the synchronous groups with the same defect types are grouped into the same synchronous cluster, resulting in several synchronous clusters of the two dimensions;

[0066] The b-th synchronic group in the a-th synchronic cluster of the two dimensions is denoted as the target group, and all synchronic groups in all synchronic clusters of the two dimensions except for the a-th synchronic cluster are denoted as the comparison group.

[0067] For a target group and any comparison group, obtain the similarity between two data segments to be analyzed in any dimension of the target group and the comparison group (the specific process of obtaining the similarity between the data segments to be analyzed is the same as obtaining the similarity between two data segments to be analyzed within the matching pair, so it will not be described again in this embodiment); obtain the similarity between two data segments to be analyzed in that dimension of the target group and the comparison group, and take the product of the similarity between the two data segments to be analyzed in each of the two dimensions of the target group and the comparison group as the similarity between the target group and the comparison group;

[0068] The product of the abnormal factors of the corresponding sets of the data segments to be analyzed in the two dimensions of the comparison group is used as the abnormality coefficient of the comparison group, and the ratio of the similarity between the target group and the comparison group to the abnormality coefficient of the comparison group is used as the non-coupling coefficient between the target group and the comparison group.

[0069] The mean of the uncoupling coefficients between the target group and all comparison groups is negatively correlated and normalized (in this embodiment, the exp(-x) model is used to present the negative correlation normalization process, where x is the input of the model, and the implementer can set the negative correlation normalization function according to the actual situation). The result of the negative correlation normalization of the mean of the uncoupling coefficients between them is used as the hidden danger coupling factor of the target group.

[0070] As an example, the specific formula for calculating the hazard coupling factor of the target group is as follows:

[0071]

[0072] In the formula, W represents the hazard coupling factor of the target group; M represents the number of comparison groups; A q Q′ represents the similarity between the target group and the q-th comparison group. qrepresents the outlier coefficient of the q-th comparison group; exp() represents the exponential function with the natural constant as the base.

[0073] Furthermore, the hazard coupling factors of all synchronous groups are obtained, and the mean of the hazard coupling factors of all synchronous groups is used as the degree of hazard coupling between the two dimensions.

[0074] It should be noted that the degree of hazard coupling between the dimensions represents the probability that production data from different dimensions, under a specific combination, will lead to defects in injection-molded products. The decoupling coefficient represents the similarity between synchronic groups in different synchronic clusters (different defect types) on the corresponding dimension. Since the defect types caused by production data from different dimensions under a specific combination are stable, the similarity between synchronic groups in different synchronic clusters on the corresponding dimension indicates that the defect types caused by production data from different dimensions under a specific combination are unstable. Therefore, the larger the decoupling coefficient between synchronic groups in different synchronic clusters of different dimensions, the lower the degree of hazard coupling between different dimensions. At the same time, to avoid defects in injection-molded products being caused by anomalies in a single dimension, when calculating the decoupling coefficient, a negatively correlated calculation weight is constructed based on the anomaly factors of the corresponding sets of the two data segments to be analyzed within the synchronic group to accurately quantify the decoupling coefficient. This embodiment measures the potential linkage risk under abnormal conditions by analyzing the historical data similarity of two dimensions under different defect types and combining the anomaly factors of these defect types in their respective dimensions. First, the similarity between the target data segment and other defect type data segments in their respective dimensions is calculated, and the similarity of the two dimensions is multiplied to reflect the overall cross-dimensional similarity. Then, the product of the two set anomaly factors is used as the risk weight to perform exponential decay processing on the overall similarity to quantify the degree of coupling between the two dimensions and provide a basis for cross-dimensional anomaly identification.

[0075] Thus, the degree of coupling of potential risks between different dimensions is obtained.

[0076] Step S004: Based on the similarity between each dimension in the current time period and the data segment to be analyzed, and combined with the abnormal factors of the corresponding set of the data segment to be analyzed for each dimension, obtain the degree of single-dimensional anomaly of each dimension in the current time period; based on the degree of single-dimensional anomaly of each dimension in the current time period and the degree of coupling of potential risks between different dimensions, obtain the overall degree of anomaly of each dimension in the current time period, and then determine whether each dimension in the current time period is abnormal.

[0077] It should be noted that after obtaining the abnormal factors of each set in each dimension and the degree of coupling of potential risks between different dimensions through steps S002 and S003 respectively, the overall abnormality of each dimension data in the local range at the current moment can be obtained based on the obtained abnormal factors of each set in each dimension and the degree of coupling of potential risks between different dimensions, combined with the similarity between the data of each dimension in the local range at the current moment and the data segments to be analyzed in each dimension (the production data of each dimension corresponding to the production time period of the historical defective product).

[0078] It should be further explained that if the similarity between the data of a certain dimension in the local range at the current moment and the data segment to be analyzed in that dimension is greater, and the anomaly factor of the set corresponding to the data segment to be analyzed in that dimension is greater, then it means that the data of that dimension in the local range at the current moment is more similar to the data segment to be analyzed in that dimension that has anomalous characteristics. Therefore, the data of that dimension in the local range at the current moment is more likely to be anomalous data. Thus, the degree of single-dimensional anomaly of the data of that dimension in the local range at the current moment can be obtained in this way.

[0079] Preferably, in a specific embodiment of the present invention, the average time length corresponding to all data segments to be analyzed is recorded as the baseline time length, and all times within the previous baseline time length and the current time are used to constitute the current time period;

[0080] For any data segment to be analyzed in the i-th dimension, the similarity between the i-th dimension and the i-th dimension of the current time period (the process of obtaining the similarity between the i-th dimension and the i-th dimension of the current time period is the same as obtaining the similarity between two data segments to be analyzed within the matching pair, so it will not be described again in this embodiment) is multiplied by the abnormal factor of the set corresponding to the i-th dimension of the data segment to be analyzed, and the product is used as the abnormal correlation coefficient between the i-th dimension of the data segment to be analyzed and the i-th dimension of the current time period.

[0081] Furthermore, the average of the abnormal correlation coefficients between all data segments to be analyzed in the i-th dimension and the i-th dimension in the current time period is taken as the unidimensional abnormality degree of the i-th dimension in the current time period.

[0082] As an example, the specific formula for calculating the single-dimensional anomaly level of the i-th dimension in the current time period is:

[0083]

[0084] In the formula, F i This represents the degree of single-dimensional anomaly in the i-th dimension of the current time period; m i A′ represents the number of data segments to be analyzed in the i-th dimension. i,pThis represents the similarity between the i-th dimension and the p-th data segment to be analyzed in the current time period; Q″ i,p This represents the anomaly factor of the set corresponding to the data segment to be analyzed in the i-th dimension.

[0085] It should be noted that A ′ i,p The larger the value of Q″, the greater the similarity between the i-th dimension and the data segment to be analyzed in the current time period. i，p This indicates that the more anomalous the data segment to be analyzed in the i-th dimension, the higher F becomes. i The larger the value, the more abnormal the i-th dimension is in the current time period; after obtaining the single-dimensional abnormality of the i-th dimension in the current time period, we can combine the single-dimensional abnormality of other dimensions and the degree of hidden danger coupling between the i-th dimension and other dimensions to accurately assess the overall abnormality of each dimension.

[0086] Preferably, in a specific embodiment of the present invention, for the i-th dimension, based on the single-dimensional anomaly degree of the i-th dimension and other dimensions in the current time period, and combined with the degree of potential coupling between the i-th dimension and other dimensions, the overall anomaly degree of the i-th dimension in the current time period is obtained, and the specific calculation formula is as follows:

[0087]

[0088] In the formula, G i This represents the overall anomaly level of the i-th dimension in the current time period; N represents the number of dimensions; F i F represents the degree of single-dimensional anomaly in the i-th dimension of the current time period; j E represents the degree of unidimensional anomaly in the j-th dimension of the current time period; i,j This indicates the degree of coupling between the i-th and j-th dimensions; sigmoid() represents the sigmoid function, which is used for normalization in this embodiment.

[0089] It should be noted that the degree of anomaly in the current time period reflects the degree of matching between that dimension and its historical defect data in terms of anomalous features. Combined with the degree of coupling between dimensions, the potential risk of anomaly linkage between different dimensions is quantified, thereby measuring the impact of anomalies in other dimensions on the probability of anomalies in this dimension. By weighted summing of the degree of anomaly in the current time period and the degree of coupling, the overall degree of anomaly in the previous time period can be accurately assessed. This can then be used as a basis to accurately identify whether anomalies have occurred in the current production data of each dimension.

[0090] Specifically, a threshold value γ for overall anomaly is preset. The specific value of γ can be set according to the actual situation (it can be set based on industry standards). This embodiment does not make a hard requirement. In this embodiment, γ = 0.6 is used as an example. If the overall anomaly of the i-th dimension in the current time period is greater than or equal to γ, then the production data of the i-th dimension in the current time period is abnormal production data.

[0091] This concludes the embodiment.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A production data anomaly identification method for production process control, characterized by, The method comprises the following steps: According to the production time period corresponding to the historical defective product, the dimensional data segments to be analyzed are screened from the historical data of each dimension; Classifying the dimensional data segments to be analyzed to obtain a plurality of sets of each dimension; obtaining the homogeneity of the set according to the similarity between the data segments to be analyzed in the set; obtaining the abnormal factor of the set according to the homogeneity of each set and the homogeneity difference between different sets of the same dimension; According to the data segments to be analyzed corresponding to the same time period in different dimensions, and the defect type corresponding to the data segments to be analyzed, the data segments to be analyzed in different dimensions are divided into a plurality of co-time groups and co-time clusters are constructed based on the co-time groups; The specific method for dividing the data segments to be analyzed in different dimensions into a plurality of co-time groups and constructing co-time clusters based on the co-time groups according to the data segments to be analyzed corresponding to the same time period in different dimensions, and the defect type corresponding to the data segments to be analyzed, comprises: For any two dimensions, the data segments to be analyzed of the two dimensions corresponding to the same time period are classified into the same co-time group according to the time period corresponding to the data segments to be analyzed of the two dimensions, to obtain a plurality of co-time groups of the two dimensions; the co-time groups with the same defect type are classified into the same co-time cluster according to the defect type corresponding to the data segments to be analyzed in each co-time group of the two dimensions, to obtain a plurality of co-time clusters of the two dimensions; According to the similarity between the data segments to be analyzed in the co-time groups of different co-time clusters in different dimensions and the abnormal factor of the corresponding set, the uncoupling coefficient between different co-time groups is obtained, and then the hidden danger coupling degree between different dimensions is obtained; The specific method for obtaining the uncoupling coefficient between different co-time groups according to the similarity between the data segments to be analyzed in the co-time groups of different co-time clusters in different dimensions and the abnormal factor of the corresponding set, comprises: For any two dimensions, the first co-temporal group in the first co-temporal cluster of the two dimensions is recorded as a target group, and all co-temporal groups in all co-temporal clusters of the two dimensions except the first co-temporal cluster are recorded as contrast groups. ​​​ For the target group and any comparison group, the similarity between the two data segments to be analyzed of any dimension in the target group and the comparison group is obtained; the similarity between the two data segments to be analyzed of the two dimensions in the target group and the comparison group is obtained, and the product of the similarity between the two data segments to be analyzed of the two dimensions in the target group and the comparison group is taken as the similarity between the target group and the comparison group; The product of the abnormal factors of the data segments to be analyzed of the two dimensions in the comparison group is taken as the abnormal coefficient of the comparison group, and the ratio of the similarity between the target group and the comparison group to the abnormal coefficient of the comparison group is taken as the uncoupling coefficient between the target group and the comparison group; The specific method for obtaining the hidden danger coupling degree between different dimensions, comprises: The mean of the uncoupling coefficients between the target group and all comparison groups is negatively correlated, and the result of negatively correlating the mean of the uncoupling coefficients between the target group and all comparison groups is taken as the hidden danger coupling factor of the target group; The hidden danger coupling factors of all co-time groups are obtained, and the mean of the hidden danger coupling factors of all co-time groups is taken as the hidden danger coupling degree between the two dimensions. According to the similarity between each dimension and its to-be-analyzed data segment in the current time period, and in combination with the abnormal factor of the to-be-analyzed data segment corresponding set of each dimension, a single-dimensional abnormality degree of each dimension in the current time period is obtained; according to the single-dimensional abnormality degree of each dimension in the current time period and the hidden danger coupling degree between different dimensions, an overall abnormality degree of each dimension in the current time period is obtained, and then whether each dimension in the current time period is abnormal is judged.

2. The production data anomaly identification method for production process control according to claim 1, characterized by, The method for classifying the to-be-analyzed data segments of each dimension to obtain a plurality of sets of each dimension comprises the following specific method: For the The dimension, according to the first For each data segment to be analyzed in each dimension, the corresponding defect type will be determined by the first defect type. Several data segments to be analyzed in each dimension are grouped into the same set to obtain the first... Several sets of dimensions.

3. The production data anomaly identification method for production process control according to claim 1, characterized by, The method for obtaining the homogeneity of the set according to the similarity between the to-be-analyzed data segments in the set comprises the following specific method: For the The first dimension The set will be the first set. The first dimension All data segments to be analyzed in each set are matched pairwise to obtain several matching pairs; For any matching pair, the DTW distance between the two to-be-analyzed data segments in the matching pair is obtained by using the DTW algorithm; the DTW distance between the two to-be-analyzed data segments in the matching pair is negatively correlated and normalized, and the result of the negative correlation and normalization is taken as the similarity between the two to-be-analyzed data segments in the matching pair; The average of the similarity between all matching pairs of two data segments to be analyzed is taken as the homogeneity of the first set of the first dimension. The average of the similarity between all matching pairs of two data segments to be analyzed is taken as the homogeneity of the first set of the first dimension. The average of the similarity between all matching pairs of two data segments to be analyzed is taken as the 4. The production data anomaly identification method for production process control according to claim 1, characterized by, The method for obtaining the abnormal factor of the set according to the homogeneity of each set and the homogeneity difference between different sets of the same dimension comprises the following specific method: ； wherein represents an anomaly factor of the th set of the th dimension; represents a number of sets of the th dimension; represents homogeneity of the th set of the th dimension; represents homogeneity of the th set of the th dimension; represents a sigmoid function.

5. The production data anomaly identification method for production process control according to claim 3, characterized by, The method for obtaining the single-dimensional abnormality degree of each dimension in the current time period according to the similarity between each dimension and its to-be-analyzed data segment in the current time period, and in combination with the abnormal factor of the to-be-analyzed data segment corresponding set of each dimension comprises the following specific method: The mean of the time lengths of all to-be-analyzed data segments is taken as a reference duration, and all time points in the reference duration before the current time and the current time point constitute a current time period; For the For any data segment to be analyzed in any dimension, the first segment of the current time period will be... The dimension and the first The similarity between the data segments to be analyzed in the first dimension is multiplied by the first dimension. The product of the outlier factors of the set corresponding to the data segment to be analyzed in the nth dimension is used as the product of the outlier factors of the set in the nth dimension. The data segment to be analyzed in the current time period is the first dimension. The abnormal correlation coefficients of each dimension; The first All data segments to be analyzed in each dimension and the current time period. The mean of the abnormal correlation coefficients in the current dimension is used as the first value in the current time period. The degree of anomaly in each dimension.

6. The production data anomaly identification method for production process control according to claim 1, characterized by, The method for obtaining the overall abnormality degree of each dimension in the current time period according to the single-dimensional abnormality degree of each dimension in the current time period and the hidden danger coupling degree between different dimensions comprises the following specific method: ； In the formula, represents the overall anomaly degree of the i-th dimension of the current time period; represents the number of dimensions; represents the single-dimension anomaly degree of the i-th dimension of the current time period; represents the number of dimensions; represents the single-dimension anomaly degree of the i-th dimension of the current time period; represents the number of dimensions; represents the single-dimension anomaly degree of the i-th dimension of the current time period; represents the number of dimensions; represents the number of dimensions; represents the hazard coupling degree between the i-th dimension and the j-th dimension; represents a sigmoid function.

7. The production data anomaly identification method for production process control according to claim 1, characterized by, The method for judging whether each dimension in the current time period is abnormal comprises the following specific method: A preset overall abnormality degree threshold is set If the overall abnormality degree of the first dimension in the current time period is greater than or equal to , the production data of the first dimension in the current time period is abnormal production data.

Images