Screen support production data management method

Abstract

The invention relates to the technical field of data processing, and particularly discloses a screen bracket production data management method, which comprises the following steps of: obtaining production data of each batch of screen brackets; clustering the production data of each batch to obtain a plurality of clusters; the Box-Cox algorithm is used to transform the production data in each cluster, and transformed production data is obtained; and taking the normalized value of the mahalanobis distance of the converted production data in the corresponding cluster as the abnormal degree of the screen bracket of the corresponding batch, and judging that the batch of which the abnormal degree is greater than a threshold value is an abnormal batch. According to the scheme provided by the invention, the anomaly detection accuracy of the screen bracket can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for managing screen bracket production data. Background Art

[0002] The Internet of Things (IoT) refers to the technology that connects any object to the internet through various information sensing devices and agreed-upon protocols, enabling information exchange and communication to achieve intelligent identification, positioning, tracking, monitoring, and management. With the rapid development of the electronics, smart home, and industrial equipment markets, IoT technology is gradually being applied to production processes.

[0003] A screen bracket is a device used to support and secure monitor or TV screens. Its production involves multiple processes, including stamping, welding, and surface treatment. Parameters for screen brackets include stamping pressure, welding temperature, welding voltage, and welding current. Abnormal production data for screen brackets will inevitably affect the quality of the final product. Therefore, the Internet of Things (IoT) can be used to collect, transmit, and manage screen bracket production data, facilitating timely detection of abnormal production data and preventing the release of substandard screen brackets.

[0004] In existing technology, screen bracket anomaly detection often involves comparing parameters in each dimension with corresponding thresholds. For example, the stamping pressure, welding temperature, welding voltage, and welding current are compared with corresponding thresholds. When a parameter in a particular dimension exceeds the corresponding threshold, the screen bracket is determined to be abnormal. However, this anomaly detection method only considers the parameters of each dimension, but does not consider the combined impact of the parameters in each dimension on anomaly detection, resulting in errors in the anomaly detection results.

[0005] The existing technology also uses Euclidean distance to realize anomaly detection of screen brackets, which can take into account the comprehensive impact of parameters in various dimensions on anomaly detection. Specifically, for each piece of production data, the Euclidean distance between the production data and the center point of all production data is calculated, and the screen bracket corresponding to the production data with a Euclidean distance greater than the threshold is determined to be abnormal. However, since the screen bracket includes parameters in multiple dimensions, and there is a certain correlation between the parameters of each dimension, for example, the greater the welding voltage and welding current, the higher the welding temperature. The Euclidean distance measures the direct distance between two pieces of production data, without considering the distribution characteristics of the production data. Therefore, using the Euclidean distance to detect anomalies in screen brackets will still result in errors. Summary of the Invention

[0006] The present invention provides a screen bracket production data management method, which aims to solve the technical problem of errors in the prior art in using Euclidean distance to perform abnormality detection on screen brackets.

[0007] A screen bracket production data management method of the present invention comprises the following steps:

[0008] Obtaining production data of each batch of screen brackets; the production data includes parameter means of the screen brackets in different dimensions in the same batch;

[0009] Clustering the production data of each batch to obtain multiple clusters;

[0010] The Box-Cox algorithm is used to transform the production data in each cluster to obtain the transformed production data;

[0011] Among them, the Box-Cox algorithm includes an improved transformation coefficient, which is the ratio of the initial transformation coefficient to the influence factor; the influence factor is the product of the aggregation degree of the corresponding cluster cluster and the distribution similarity; the aggregation degree represents the degree of aggregation of the production data in the corresponding cluster cluster; the distribution similarity is the product of the fit and stationarity of the corresponding cluster cluster; the fit is inversely correlated with the absolute value of the difference between the probability of the parameters of each dimension in the production data within different setting ranges and the setting value, and the stationarity is inversely correlated with the absolute value of the difference between the probability of the parameters of each dimension in the setting range and the mean probability of the parameters of all dimensions in the corresponding setting range; wherein the setting range includes [μ-σ, μ+σ], [μ-2σ, μ+2σ] and [μ-3σ, μ+3σ], μ and σ are the mean and standard deviation of the parameters of the corresponding dimensions, respectively;

[0012] The value of the transformed production data after normalization of the Mahalanobis distance of the corresponding cluster is used as the abnormality degree of the corresponding batch of screen brackets, and the batches with abnormality degrees greater than the threshold are determined to be abnormal batches.

[0013] In the above solution, the Mahalanobis distance is used to calculate the degree of abnormality for each batch of screen brackets. Because the Mahalanobis distance can adapt to production data with correlations between parameters in various dimensions, the calculation results are more accurate and anomaly detection is more reliable. Furthermore, when using the Box-Cox algorithm to transform production data, an improved transformation coefficient is used, further improving the accuracy of the calculation results.

[0014] Preferably, the clustering degree is inversely correlated with the mean of the discreteness of all production data in the corresponding cluster, and inversely correlated with the ratio of the discreteness of each production data to the mean of the discreteness of all production data; the discreteness is the Euclidean distance between the production data and the cluster center of the corresponding cluster.

[0015] Preferably, the aggregation degree F of the mth cluster is m for:

[0016]

[0017] Where, is the mean of the dispersion of all production data in the mth cluster, is the mean of the dispersion of all production data in each cluster, L m,k is the dispersion of the kth production data in the mth cluster, C m is the total number of production data in the mth cluster.

[0018] In the above scheme, random errors can be reduced by comparing the mean of the discreteness of all clusters with the discreteness of the corresponding cluster. The accuracy of the calculation results is further improved by comparing the discreteness of each production data in the corresponding cluster with the mean of the discreteness of all production data.

[0019] Preferably, the fit Q of the mth cluster is m for:

[0020]

[0021] Where, P 1,i,m is the probability that the parameter of the i-th dimension of all production data in the m-th cluster is within the range of [μ-σ,μ+σ], P 2,i,m is the probability that the parameter of the i-th dimension of all production data in the m-th cluster is within the range of [μ-2σ,μ+2σ], P 3,i,m is the probability that the parameter of the i-th dimension in all production data in the m-th cluster is in the range of [μ-3σ,μ+3σ], and N is the total number of dimensions of the production data in the m-th cluster.

[0022] In the above scheme, the degree to which the parameters conform to the Gaussian distribution is represented by the probability of the parameters of each dimension within different set ranges and the absolute value of the difference between the set values, thereby representing the fit of the cluster clusters, making the calculation results more reasonable and accurate.

[0023] Preferably, the stationarity R of the mth cluster is m for:

[0024]

[0025] Where, P α,i,m is the probability that the parameter of the i-th dimension in all production data in the m-th cluster is within the α-th set range, is the mean probability of the parameters of all dimensions in all production data in the mth cluster within the αth set range, and N is the total number of dimensions of the production data in the mth cluster.

[0026] In the above scheme, the stability of the data's fit to the Gaussian distribution is represented by the absolute value of the difference between the probability of the parameter of each dimension within the set range and the mean probability of the parameters of all dimensions within the corresponding set range, thereby representing the stability of the clustering cluster, making the calculation results more reasonable and accurate.

[0027] Preferably, the calculation formula of the Box-Cox algorithm is:

[0028]

[0029] Where y is the production data before transformation, is the transformed production data, λ is the improved transformation coefficient, and ln() is the logarithmic function with the natural constant e as the base.

[0030] Preferably, the Mahalanobis distance D of the transformed production data x in the mth cluster is m,x for:

[0031]

[0032] Where μ m is the mean of all transformed production data in the mth cluster, is the inverse matrix of the covariance matrix of all transformed production data in the mth cluster, (x-μ m ) T is the matrix x-μ m The transposed matrix of .

[0033] Preferably, the clustering of the production data of each batch is achieved by using a K-means clustering algorithm.

[0034] In the above scheme, the K-means clustering algorithm has the advantages of being simple and easy to implement, high computational efficiency, strong interpretability, and strong adaptability.

[0035] Preferably, the parameters of the screen bracket include stamping pressure, welding temperature, welding voltage and welding current.

[0036] Preferably, the initial transformation coefficients are obtained by a maximum likelihood estimation method.

[0037] Among the above schemes, the maximum likelihood estimation method has the advantages of simplicity and high efficiency.

[0038] The beneficial effects are:

[0039] The solution of the present invention uses the Mahalanobis distance to measure the degree of abnormality of each batch of screen brackets in the corresponding cluster, which can fully consider the similarity between parameters of different dimensions, thereby making the calculation result of the abnormality more accurate and realizing efficient management of the production data of the screen bracket. Moreover, before calculating the Mahalanobis distance, the transformed production data is obtained through the Box-Cox algorithm, so that the transformed production data is more consistent with the Gaussian distribution, thereby improving the accuracy of the Mahalanobis distance calculation. Furthermore, in the Box-Cox algorithm, the improved transformation coefficient is obtained by the aggregation degree and distribution similarity of the corresponding clusters, making the transformation of the production data more reasonable and improving the accuracy of the calculation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flowchart of the steps of a screen bracket production data management method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention, but should not be construed as limiting the present invention.

[0042] like Figure 1 As shown, the present invention provides a screen bracket production data management method, comprising the following steps:

[0043] S1. Obtain production data of each batch of screen brackets.

[0044] In this step, the production of the screen bracket involves multiple processes such as stamping, welding, and surface treatment. Therefore, the production of the screen bracket involves multiple parameters, each of which will affect the production quality of the screen bracket. Therefore, the production data includes the average values ​​of the parameters of the screen brackets in different dimensions in the same batch. Specifically, the parameters of the screen bracket include stamping pressure, welding temperature, welding voltage, and welding current. The production data includes the average stamping pressure, welding temperature, welding voltage, and welding current of the screen brackets in the same batch.

[0045] The stamping pressure can be collected using a pressure sensor, the welding temperature can be collected using a temperature sensor, and the welding voltage and welding current can be directly obtained through the welding equipment.

[0046] It's understandable that during the production of screen brackets, multiple screen brackets are produced in each batch. Screen brackets from the same batch share similar production conditions, such as the same production equipment. Therefore, the production data for multiple screen brackets from the same batch is more similar. Therefore, the average of the production data for multiple screen brackets from the same batch can be used as the production data for the corresponding batch of screen brackets. Using the degree of abnormality in this production data to represent the degree of abnormality for all screen brackets in the batch allows for anomaly detection of each batch of production brackets, resulting in faster calculations.

[0047] In some embodiments, a sampling detection method can also be used to obtain the production data of each batch of screen brackets, which can reduce the amount of calculation and speed up the calculation.

[0048] S2. Cluster the production data of each batch to obtain multiple clusters.

[0049] In this step, the production data for each batch exhibits clustering, meaning that some batches have high similarity. If anomalies in each production data item are determined based on the entire production data set, the calculation results will be inaccurate. Therefore, this step clusters the production data for each batch into multiple clusters. Then, anomalies in each batch of production data are calculated within each cluster, resulting in more accurate calculation results and reducing errors.

[0050] The K-means clustering algorithm can be used to cluster production data for each batch. The K-means clustering algorithm includes the following steps: Step 1: Determine the total number of clusters K and select an initial center point for each cluster; Step 2: Assign each production data item to the cluster with the initial center point closest to it; Step 3: Calculate the new center point of each cluster and replace the initial center point with the new center point; Step 4: Repeat steps 2 and 3 until the cluster division no longer changes or the maximum number of iterations is reached, and then terminate the algorithm. The advantages of the K-means clustering algorithm are clear algorithm structure, simple ideas, simple implementation, easy explanation, and high accuracy.

[0051] In some alternative embodiments, the DBSCAN clustering algorithm and the dispersion peak clustering algorithm may also be used to cluster the production data of each batch.

[0052] S3. Use the Box-Cox algorithm to transform the production data in each cluster to obtain the transformed production data.

[0053] As can be seen from step S1, production data includes multiple parameters, and the parameters in each dimension have a certain correlation. For example, the higher the welding voltage and welding current, the higher the welding temperature. However, the existing use of Euclidean distance to characterize the degree of anomaly of each batch of production data within the corresponding cluster can lead to errors. This is because the Euclidean distance measures the direct distance between two pieces of production data, without considering the distribution characteristics of the production data. Therefore, the present invention uses Mahalanobis distance to characterize the degree of anomaly of each batch of production data within the corresponding cluster. Taking into account the distribution characteristics of production data, this can reduce errors and achieve good anomaly detection results.

[0054] When calculating the Mahalanobis distance, the production data should also satisfy the Gaussian distribution, otherwise the calculation results will be inaccurate. Therefore, the Box-Cox algorithm can be used to transform the production data so that the production data satisfy the Gaussian distribution.

[0055] As an existing technology, the Box-Cox algorithm has the following calculation formula:

[0056]

[0057] Where y is the production data before transformation, is the transformed production data, λ0 is the initial transformation coefficient, and ln() is the logarithmic function with the natural constant e as the base.

[0058] The Box-Cox algorithm's calculation formula shows that the initial transformation coefficients determine how production data is transformed and the resulting output data. These initial transformation coefficients can be obtained using the maximum likelihood estimation method. This method can be used to calculate the optimal initial transformation coefficients, maximizing the correlation coefficient between the transformed production data and the initial transformation coefficients. In other words, the initial transformation coefficients calculated using the maximum likelihood estimation method ensure that the transformed production data closely approximate a Gaussian distribution, thereby improving the analyzability of the production data and ensuring the accuracy of the Mahalanobis distance calculation in subsequent steps.

[0059] However, the degree to which production data in different clusters deviates from the Gaussian distribution varies. The initial transformation coefficients derived directly from the maximum likelihood estimation method cannot meet the transformation requirements of production data in different clusters. Furthermore, the initial transformation coefficients are also affected by abnormal production data during calculation, resulting in inaccurate production data transformation. Therefore, it is necessary to modify the initial transformation coefficients of each cluster using the influencing factor to obtain improved transformation coefficients.

[0060] Therefore, step S3 includes the following steps:

[0061] S31. Calculate the initial transformation coefficients of each cluster.

[0062] In this step, the initial transformation coefficients of each cluster are calculated by the maximum likelihood estimation method. As an existing technology, the maximum likelihood estimation method generally includes the following steps: First, define the likelihood function: the likelihood function is a function of the joint probability distribution of the production data in each cluster with respect to the parameter; second, construct the log-likelihood function: in order to simplify the calculation, the likelihood function is usually taken logarithm to obtain the log-likelihood function; then, derive and solve: calculate the derivative of the log-likelihood function with respect to the parameter, set the derivative to zero, and solve for the value of the parameter; finally, verify the rationality of the solution: verify the rationality of the solution through second-order derivatives or numerical methods to ensure that the maximum value is found. Among them, the value of the parameter obtained by the solution is the value of the initial transformation coefficient.

[0063] In some alternative embodiments, the least square method and Bayesian estimation method may also be used to calculate the initial transformation coefficients.

[0064] S32. Calculate the impact factor corresponding to the initial transformation coefficient of each cluster.

[0065] The impact factor represents the degree to which the production data in each cluster conforms to a Gaussian distribution. The impact factor is the product of the corresponding cluster's clustering degree and distribution similarity. The clustering degree of a cluster represents the degree of clustering of the production data within the cluster. Since production data that conforms to a Gaussian distribution is close to the mean of the production data, meaning that the production data are closer together, the greater the clustering degree of a cluster, the greater the likelihood that the production data in that cluster conforms to a Gaussian distribution, and thus the greater the impact factor. Distribution similarity represents the degree of proximity between the probability distribution of the production data in the cluster and the probability distribution of production data that conforms to a Gaussian distribution. Therefore, the greater the distribution similarity, the greater the impact factor.

[0066] Therefore, step S32 includes the following steps:

[0067] S321. Calculate the degree of aggregation of each cluster.

[0068] The degree of aggregation is inversely correlated with the mean of the discreteness of all production data in the corresponding cluster, and is inversely correlated with the ratio of the discreteness of each production data to the mean of the discreteness of all production data. The discreteness is the Euclidean distance between the production data and the cluster center of the corresponding cluster. This is because the smaller the mean of the discreteness of all production data in the cluster, the closer all production data in the cluster are to the cluster center, and the higher the degree of aggregation of the cluster. The smaller the ratio of the discreteness of each production data to the mean of the discreteness of all production data, the more production data has a discreteness smaller than the mean of all discreteness, and the more production data is close to the cluster center, the higher the degree of aggregation of the cluster.

[0069] In one embodiment, the clustering degree F of the mth cluster is m for:

[0070]

[0071] Where, is the mean of the dispersion of all production data in the mth cluster, is the mean of the dispersion of all production data in each cluster, L m,k is the dispersion of the kth production data in the mth cluster, C m is the total number of production data in the mth cluster.

[0072] In this embodiment, the degree of clustering is represented by the ratio of the mean of the dispersion of all production data in each cluster to the mean of the dispersion of all production data in the corresponding cluster. Comparison with the dispersion of all clusters can avoid accidental errors. Furthermore, the degree of clustering is represented by the product of the ratio of the dispersion of each production data in the cluster to the mean of the dispersion of all production data in the corresponding cluster. The smaller the product, the more production data in the cluster have dispersions less than the mean of the dispersions. This indicates a higher degree of clustering of the production data and a higher degree of clustering, resulting in a more accurate calculation result.

[0073] S322. Calculate the distribution similarity of each cluster.

[0074] The distribution similarity is the product of the fit and stability of the corresponding clusters. Therefore, step S322 includes the following steps:

[0075] S3221. Calculate the degree of fit of each cluster.

[0076] The fit is inversely correlated with the probability of each dimension's parameter in the production data falling within different set ranges and the absolute value of the difference between the set values. The set ranges include [μ - σ, μ + σ], [μ - 2σ, μ + 2σ], and [μ - 3σ, μ + 3σ], where μ and σ are the mean and standard deviation of the parameter in the corresponding dimension, respectively. The set values ​​include 0.68, 0.95, and 0.99.

[0077] In one embodiment, the fitness Q of the mth cluster is m for:

[0078]

[0079] Where, P 1,i,m is the probability that the parameter of the i-th dimension of all production data in the m-th cluster is within the range of [μ-σ,μ+σ], P 2,i,m is the probability that the parameter of the i-th dimension of all production data in the m-th cluster is within the range of [μ-2σ,μ+2σ], P 3,i,m is the probability that the parameter of the i-th dimension in all production data in the m-th cluster is in the range of [μ-3σ,μ+3σ], and N is the total number of dimensions of the production data in the m-th cluster.

[0080] According to the characteristics of Gaussian distribution, when the production data in the cluster conform to Gaussian distribution, the probability of production data in the range of [μ-σ, μ+σ] is 0.68, the probability of production data in the range of [μ-2σ, μ+2σ] is 0.95, and the probability of production data in the range of [μ-3σ, μ+3σ] is 0.99.

[0081] In this embodiment, the degree of fit of the cluster is represented by the absolute value of the difference between the probability of the parameters of each dimension in the production data being within different set ranges and the set values. The smaller the absolute value of the difference between the two, the more the parameters of each dimension in the cluster conform to the Gaussian distribution, and the greater the degree of fit of the cluster.

[0082] S3222. Calculate the stationarity of each cluster.

[0083] The degree of stationarity is inversely correlated with the absolute value of the difference between the probability of the parameter of each dimension being in the set range and the mean probability of the parameters of all dimensions in the corresponding set range.

[0084] In one embodiment, the stationarity R of the mth cluster is m for:

[0085]

[0086] Where, P α,i,m is the probability that the parameter of the i-th dimension in all production data in the m-th cluster is within the α-th set range, is the mean probability of the parameters of all dimensions in all production data in the mth cluster within the αth set range, and N is the total number of dimensions of the production data in the mth cluster.

[0087] In this embodiment, the absolute value of the difference between the probability of the parameter of each dimension being within the set range and the mean probability of the parameters of all dimensions being within the corresponding set range is used to characterize the stability of the cluster. The smaller the absolute value of the difference between the two, the more stable the fit of the parameters of each dimension to the Gaussian distribution, and therefore the higher the stability of the cluster.

[0088] S3223. Obtain the distribution similarity of each cluster.

[0089] Then the distribution similarity G of the mth cluster is m for:

[0090] G m =Q m ×R m ;

[0091] Where Q m is the fit of the mth cluster, R m is the stationarity of the mth cluster.

[0092] S323: Obtain the impact factor corresponding to the initial transformation coefficient of each cluster.

[0093] Then the influence factor u of the mth cluster is m for:

[0094] U m =Fm ×G m ;

[0095] Where, F m is the aggregation degree of the mth cluster, G m is the distribution similarity of the mth cluster.

[0096] S33. Calculate the improved transformation coefficient of each cluster.

[0097] The improved transformation coefficient is the ratio of the initial transformation coefficient to the impact factor. This is because the impact factor represents the degree to which the production data in each cluster conforms to the Gaussian distribution. The more the production data in the cluster conforms to the Gaussian distribution, the less adjustment is needed to the initial transformation coefficient. Therefore, the ratio of the initial transformation coefficient to the impact factor is used as the improved transformation coefficient.

[0098] Therefore, the improved transformation coefficient λ of the mth cluster is m for:

[0099]

[0100] Where λ0 is the initial transformation coefficient, U m is the impact factor of the mth cluster.

[0101] S34. Obtain the transformed production data.

[0102] Substitute the improved transformation coefficients into the Box-Cox algorithm calculation formula in step S1 to obtain the transformed production data. Specifically, first calculate the unit vector of each dimension of the corresponding production data and the Euclidean distance between the production data and the cluster center of the corresponding cluster. Then substitute the Euclidean distance into the Box-Cox algorithm calculation formula to obtain the transformed Euclidean distance. Finally, obtain the component of the unit vector of the transformed Euclidean distance in each dimension and use it as the value of the transformed production data in each dimension, thereby obtaining the transformed production data.

[0103] S4. The value of the transformed production data after normalization in the corresponding cluster Mahalanobis distance is used as the abnormality degree of the corresponding batch of screen brackets, and the batch with an abnormality degree greater than a threshold is determined to be an abnormal batch.

[0104] The Mahalanobis distance D of the transformed production data x in the mth cluster m,x for:

[0105]

[0106] Where μ m is the mean of all transformed production data in the mth cluster, is the inverse matrix of the covariance matrix of all transformed production data in the mth cluster, (x-μ m ) T is the matrix x-μ m The transposed matrix of .

[0107] The calculated Mahalanobis distance is normalized to obtain the abnormality degree of the corresponding batch of screen brackets. When the abnormality degree is greater than the threshold, it means that there is an abnormality in the batch of screen brackets, and the production personnel are reminded to pay special attention to the batch of screen brackets to prevent abnormal screen brackets from flowing into the market.

[0108] The threshold value range is [0.6, 0.8]. Preferably, the threshold value is 0.7. Of course, the threshold value can be adjusted as needed.

[0109] In the screen bracket production data management method of the present invention, the production data of each batch of screen brackets is first clustered to obtain multiple clusters. The production data in each cluster is then transformed using the Box-Cox algorithm to obtain the transformed production data. Finally, the Mahalanobis distance of each production data in the corresponding cluster is calculated to determine the degree of abnormality of the corresponding batch of screen brackets, thereby implementing the management of the screen bracket production data. Because the Mahalanobis distance is used to detect anomalies in the screen brackets, it can be applied to the calculation of production data with correlations between parameters in various dimensions, making the calculation results more accurate and reducing the error of anomaly detection.

[0110] Furthermore, when transforming the production data within each cluster using the Box-Cox algorithm, the influence factors are calculated by calculating the clustering degree and distribution similarity of each cluster, and an improved transformation coefficient is obtained based on the influence factors. Because the production data within each cluster deviates from the Gaussian distribution to varying degrees, the transformed production data is more consistent with the Gaussian distribution, which improves the analyzability of the transformed production data and makes the calculation of the Mahalanobis distance in subsequent steps more accurate.

[0111] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. A screen bracket production data management method, characterized in that: The method comprises the following steps: obtaining production data of each batch of screen brackets; the production data comprises parameter means of the screen brackets in different dimensions in the same batch; Clustering the production data of each batch to obtain multiple clusters; The production data in each cluster is transformed using the Box-Cox algorithm to obtain the transformed production data; wherein the Box-Cox algorithm includes an improved transformation coefficient, which is the ratio of the initial transformation coefficient to the influence factor; the influence factor is the product of the clustering degree of the corresponding cluster and the distribution similarity; the clustering degree represents the degree of clustering of the production data in the corresponding cluster; the distribution similarity is the product of the fit and the stationarity of the corresponding cluster; the fit is inversely correlated with the absolute value of the difference between the probability of the parameter of each dimension in the production data within different set ranges and the set value, and the stationarity is inversely correlated with the absolute value of the difference between the probability of the parameter of each dimension within the set range and the mean probability of the parameters of all dimensions within the corresponding set range; wherein the set range includes [μ-σ, μ+σ], [μ-2σ, μ+2σ] and [μ-3σ, μ+3σ], and μ and σ are the mean and standard deviation of the parameters of the corresponding dimension, respectively; The value of the transformed production data after normalization of the Mahalanobis distance of the corresponding cluster is used as the abnormality degree of the corresponding batch of screen brackets, and the batches with abnormality degrees greater than the threshold are determined to be abnormal batches.

2. The screen bracket production data management method according to claim 1, characterized in that: The clustering degree is inversely correlated with the mean of the dispersion of all production data in the corresponding cluster, and is inversely correlated with the ratio of the dispersion of each production data to the mean of the dispersion of all production data; the dispersion is the Euclidean distance between the production data and the cluster center of the corresponding cluster.

3. The screen bracket production data management method according to claim 2, characterized in that: The clustering degree F of the mth cluster m for: Where, is the mean of the dispersion of all production data in the mth cluster, is the mean of the dispersion of all production data in each cluster, L m,k is the dispersion of the kth production data in the mth cluster, C m is the total number of production data in the mth cluster.

4. The screen bracket production data management method according to claim 1, characterized in that: The fitness Q of the mth cluster m for: Where, P 1,i,m is the probability that the parameter of the i-th dimension of all production data in the m-th cluster is within the range of [μ-σ,μ+σ], P 2,i,m is the probability that the parameter of the i-th dimension of all production data in the m-th cluster is within the range of [μ-2σ,μ+2σ], P 3,i,m is the probability that the parameter of the i-th dimension in all production data in the m-th cluster is in the range of [μ-3σ,μ+3σ], and N is the total number of dimensions of the production data in the m-th cluster.

5. The screen bracket production data management method according to claim 1, characterized in that: The stationarity R of the mth cluster m for: Where, P α,i,m is the probability that the parameter of the i-th dimension in all production data in the m-th cluster is within the α-th set range, is the mean probability of the parameters of all dimensions in all production data in the mth cluster within the αth set range, and N is the total number of dimensions of the production data in the mth cluster.

6. The screen bracket production data management method according to claim 1, characterized in that: The calculation formula of the Box-Cox algorithm is: Where y is the production data before transformation, is the transformed production data, λ is the improved transformation coefficient, and ln() is the logarithmic function with the natural constant e as the base.

7. The screen bracket production data management method according to claim 1, characterized in that: The Mahalanobis distance D of the transformed production data x in the mth cluster m,x for: Where μ m is the mean of all transformed production data in the mth cluster, is the inverse matrix of the covariance matrix of all transformed production data in the mth cluster, (x-μ m ) T is the matrix x-μ m The transposed matrix of .

8. The screen bracket production data management method according to claim 1, characterized in that: The clustering of the production data of each batch is achieved by using the K-means clustering algorithm.

9. The screen bracket production data management method according to claim 1, characterized in that: The parameters of the screen bracket include stamping pressure, welding temperature, welding voltage and welding current.

10. The screen bracket production data management method according to claim 1, characterized in that: The initial transformation coefficients are obtained by maximum likelihood estimation method.

Images