A method for optimizing the longitudinal crack process on the surface of continuously cast billets based on K-Means clustering

By using a K-Means clustering method combined with Spearman correlation coefficient to optimize the longitudinal cracking process on the surface of continuously cast billets, the problem of weak multidimensional complex data processing capability and poor adaptability of optimization strategies in existing technologies is solved, realizing intelligent optimization of longitudinal cracks on the surface of continuously cast billets and reducing the incidence of longitudinal cracks.

CN118296420BActive Publication Date: 2026-11-10DALIAN UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410424177.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-10
Publication Date
2026-11-10
Estimated Expiration
2044-04-10

AI Technical Summary

Technical Problem

Existing technologies for dealing with longitudinal cracks on the surface of continuously cast billets rely heavily on manual data statistics. They have a weak ability to process multi-dimensional and complex data and their optimization strategies are not highly adaptable, making it difficult to meet the needs of high efficiency and intelligence in continuous casting technology.

Method used

The process of longitudinal cracking on the surface of continuously cast billets was optimized by using K-Means clustering and Spearman correlation coefficient. A sample library was established by online detection and recording of key process parameters, attribute selection and dimensionality reduction were performed, and K-Means clustering was used to screen clusters with low longitudinal cracking rate and extract the corresponding process parameter ranges.

Benefits of technology

It achieves intelligent optimization of longitudinal cracks on the surface of continuously cast billets, significantly reduces the incidence of longitudinal cracks, improves billet quality and casting machine output, and has high efficiency and adaptability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118296420B_ABST
    Figure CN118296420B_ABST
Patent Text Reader

Abstract

A method for optimizing the longitudinal crack process of continuously cast billets based on K-Means clustering is presented, belonging to the field of continuous casting process technology in iron and steel metallurgy. First, a sample library with different longitudinal crack incidence rates is established. Second, Spearman correlation coefficient is used to select attributes and reduce the dimensionality of samples in the sample library. Then, K-Means clustering is used to iteratively cluster the dimensionality-reduced data samples, filtering and locking the clusters corresponding to low longitudinal crack incidence rates. Finally, the process parameter ranges corresponding to the low longitudinal crack incidence rate clusters are extracted, thereby optimizing various process parameters. The integrated method for optimizing the longitudinal crack process of continuously cast billets by combining Spearman correlation analysis and K-Means clustering demonstrates high efficiency and good adaptability, providing a technical means for intelligent optimization of longitudinal cracks in cast billets.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of continuous casting process technology in iron and steel metallurgy, and relates to a method for optimizing the process of longitudinal cracks on the surface of continuously cast billets based on K-Means clustering. Background Technology

[0002] Longitudinal cracks in continuously cast billets are a common surface defect that seriously affects billet quality and casting machine output. Surface longitudinal cracks typically form near the meniscus of the crystallizer, with uneven heat transfer, solidification, and shrinkage during the formation of the initial billet shell being the main causes. Numerous factors influence longitudinal crack formation; steel composition, casting process, casting machine condition, and the properties of the protective slag all directly or indirectly affect the uniformity of heat flow, slag film, and billet thickness distribution. The interactions between these factors are highly complex. Therefore, controlling and optimizing the casting process to reduce the incidence of surface longitudinal cracks in billets is of great significance for improving billet quality and casting machine output.

[0003] Patent CN104889358A, already granted, proposes a method for controlling surface cracks in continuously cast slabs. This method controls the heat flux density below a critical value by adjusting parameters such as casting speed, width, and cooling water flow rate at each stage, effectively reducing the incidence of longitudinal cracks on the surface of continuously cast slabs. Patent CN107630172A, also granted, proposes a method for preventing surface cracks in low-carbon boron-containing steel. This method effectively reduces the incidence of surface cracks after hot rolling of low-carbon boron-containing steel by controlling factors such as steel composition, protective slag basicity, secondary cooling water ratio, amplitude, and frequency. In summary, current methods for optimizing crack defects mostly focus on manual data statistics. While these methods can reduce the incidence of longitudinal cracks to some extent, they suffer from limitations such as weak ability to handle multi-dimensional and complex data and poor adaptability of optimization strategies. To meet the demands of efficient and intelligent continuous casting technology, there is an urgent need to seek intelligent methods for optimizing longitudinal cracks in continuously cast slabs. Summary of the Invention

[0004] The purpose of this invention is to optimize the longitudinal crack defect on the surface of the billet. A process optimization method for longitudinal crack on the surface of continuously cast billets based on K-Means clustering is proposed, which can effectively reduce the occurrence rate of longitudinal cracks and provide an effective method for optimizing longitudinal cracks in continuously cast billets for quality control.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] A method for optimizing the longitudinal cracking process on the surface of continuously cast billets based on K-Means clustering is disclosed. The optimization method first involves online detection and recording of key process parameters and heat transfer parameters that may induce longitudinal cracking in continuous casting production, based on the main factors influencing longitudinal cracking. Combined with production quality reports, a sample library is established for different longitudinal cracking rates. Second, Spearman correlation coefficients are used to select attributes and reduce the dimensionality of samples in the sample library. Then, K-Means clustering is used to iteratively cluster the dimensionality-reduced data samples, filtering and locking down clusters corresponding to low longitudinal cracking rates. Finally, the process parameter ranges corresponding to the low longitudinal cracking rate clusters are extracted, thereby optimizing various process parameters. Specifically, the method includes the following steps:

[0007] Step 1: Building the attribute data sample library

[0008] For the same steel grade and protective slag production process, historical data from steel mills were reviewed to construct a historical sample database containing both normal and longitudinal crack conditions, starting from the main factors affecting longitudinal cracking in continuous casting. The meaning of each sample is as follows: For billets under normal conditions, each attribute of the sample represents the average value of each process parameter during the entire time the billet passes through the crystallizer; for longitudinal crack samples, each attribute of the sample represents the average value of each process parameter during the time period in which the longitudinal crack occurred in the billet. The acquired sample database is defined as D. r×c Where r is the number of rows, representing the number of samples, including normal samples and longitudinally fractured samples; c is the number of attribute columns, where columns 1 to c-1 represent c-1 attributes, i.e., the mean of c-1 parameters, and column c is the longitudinally fracture label corresponding to the sample, where 1 represents a longitudinally fractured sample and 0 represents a normal sample. To improve the accuracy of clustering results in subsequent models, the total number of samples in the data sample library is guaranteed to be no less than 500, and the number of longitudinally fractured samples is guaranteed to be no less than 25.

[0009] Step 2: Attribute Selection and Dimensionality Reduction

[0010] The Spearman correlation coefficient is a nonparametric indicator that measures the dependence between two variables and can be used to analyze nonlinear and monotonic relationships between variables. Using the Spearman correlation coefficient analysis method, for N original sample data (X, Y) of variables X and Y... i ,Y i ), (i = 1, 2, 3, ..., N), x i For X i The rank of variable X in the sample values, that is, the order of the sample values ​​of variable X after arranging them from smallest to largest. i ordinal number; y i For Y i The Spearman correlation coefficient between variables X and Y, based on the rank of variable Y in the sample values, can be obtained using the following formula:

[0011]

[0012] In the formula, ρ is the Spearman correlation coefficient between variables X and Y; d i For X i Y i The difference in rank x i -y i N is the number of samples.

[0013] The Spearman correlation coefficient ranges from -1 to 1 and reflects the sign and strength of the correlation between variables. A positive value represents a positive correlation, and a negative value represents a negative correlation. The larger the absolute value of the coefficient, the stronger the correlation between the two variables.

[0014] Spearman correlation coefficient was used to analyze the data sample database D obtained in the first step. r×c Perform correlation analysis to achieve the analysis of the data sample database D. r×c Attribute selection and dimensionality reduction. When the absolute value of the correlation coefficient between any two attributes exceeds 0.4, these two attributes are considered a highly correlated combination. Strong correlation between attributes indicates cross-coupling and information redundancy, which will affect subsequent process optimization. Therefore, highly correlated attribute combinations are processed, retaining only one attribute to ensure that the correlation coefficient between any two attributes is less than 0.4, thus achieving the purpose of attribute selection and dimensionality reduction. Correlation analysis is used to optimize the data sample database D. r×c The c-1 attributes in the dataset are reduced to d-1 attributes, resulting in the dimensionality-reduced data sample database D. r×d After dimensionality reduction, the d-th column of the sample library still contains the sample label, where 1 represents a longitudinally fractured sample and 0 represents a normal sample.

[0015] Step 3: Constructing the K-Means clustering model

[0016] Clustering can be used to find the inherent distribution structure of data, where each cluster can be viewed as a different data distribution structure. K-Means, as a hard clustering method, divides data samples into K non-overlapping clusters by pre-specifying the value of K and iteratively updating the clusters and their centroids. The specific implementation steps of K-Means clustering are as follows:

[0017] (3.1) Normalize the data samples

[0018] For sample database D r×d Retain the label records in column d, and normalize each of the (d-1) attributes to obtain a new data sample database X. r×d The normalization process is as follows:

[0019]

[0020] In the formula, D ij For sample database D r×d The value of the j-th attribute in the i-th sample, where i = 1, 2, ..., r, j = 1, 2, ..., d-1; D jmin For the data sample database D r×d The minimum value of the j-th attribute in D; jmax This represents the maximum value of the j-th attribute.

[0021] (3.2) Perform K-Means clustering on the normalized results.

[0022] First, from the new data sample library X r×(d-1) First, randomly select K samples as initial centroids. Then, calculate the distance between each sample and the initial centroid. Next, assign each sample to the cluster represented by its nearest centroid. Then, recalculate the mean of the samples in the K clusters as the new cluster centroids. Repeat this process iteratively, iterating through the cluster centroids, until all mean vectors are no longer updated. The specific process is as follows:

[0023] (3.2.1) Initialize the cluster centroid

[0024] Data sample library X r×(d-1) ={x1,x2,...,x r Randomly select K samples as K initial centroids {μ1, μ2, ... μ} k}, and initialize the cluster collection, let

[0025] (3.2.2) Divide the sample into clusters

[0026] Examine the data sample database X respectively r×(d-1) ={x1,x2,...,x r Each sample in the array is associated with K initial centroids {μ1, μ2, ... μ}. k Distance between}: T ij =||x i -μ j ||2, where i = 1, 2, ..., r, j = 1, 2, ..., k, the subscript i represents the i-th sample, and j represents the j-th initial centroid; the current sample x is determined based on the nearest centroid. i Cluster label: λ i =argmin j∈{1,2,...,k} T ij ; take sample x i Assign to the corresponding cluster:

[0027] (3.2.3) Update the cluster centroid

[0028] Calculate new mean vectors for each cluster and use them as the centroids of the new clusters:

[0029]

[0030] In the formula, μ' j Let |C| be the centroid of the new cluster, where j = 1, 2, ..., k; j | represents the number of samples in the j-th cluster, and x represents the vector in the j-th cluster.

[0031] (3.2.4) Determine the termination condition

[0032] For all cluster centroids, if the mean vector of the iterations no longer changes, i.e., μ' appears... j =μ j This indicates that the clustering process is complete, resulting in k clusters {C1, C2, ... C}. k The corresponding cluster centroids are {μ1, μ2, ... μ}. k Otherwise, continue with steps (3.2.1) to (3.2.4) until the clustering results meet the termination condition.

[0033] Step 4: Optimization of longitudinal cracking process

[0034] Through the K-means clustering described above, K non-overlapping clusters are obtained, each of which can be viewed as a different data distribution structure. Each cluster represents the data distribution structure under different process combinations in continuous casting production, and the probability of longitudinal cracking varies significantly under different process combinations. The purpose of longitudinal cracking process optimization is to find the process range where longitudinal cracking does not occur or the probability of longitudinal cracking is extremely low, that is, to identify the cluster with the lowest longitudinal cracking rate, and then extract the process information contained in this cluster to optimize various process parameters. The specific process is as follows:

[0035] (4.1) Locking in low longitudinal fracture clusters

[0036] Before performing K-means clustering, the number of clusters needs to be predetermined, and the determination of the number of clusters usually depends on the specific problem. The current process optimization aims to distinguish between clusters with high longitudinal crack incidence and clusters with low longitudinal crack incidence, but the specific number of clusters cannot be determined before clustering. Therefore, a cyclic clustering approach is needed to continuously change the number of clusters, thereby identifying the target cluster with low longitudinal crack incidence.

[0037] By pre-setting different numbers of clusters (2≤K≤10), the data sample database X is iteratively processed. r×(d-1)Perform K-means clustering until the cyclic clustering termination condition is met. The cyclic clustering process terminates when a cluster with a longitudinal breakage rate of 0% is found, and the current cluster with a longitudinal breakage rate of 0% is the target. If no cluster with a longitudinal breakage rate of 0% is found, then the cluster with the lowest longitudinal breakage rate among the results of 10 cyclic clustering iterations is locked.

[0038] A certain type of cluster C x The corresponding longitudinal crack incidence rate l is derived from the data sample database X. r×d The label record corresponding to column d is calculated using the following formula:

[0039]

[0040] In the formula, α is C x The number of samples with a label of 1 in the cluster, i.e., the number of longitudinally split samples; β is C x The number of samples with a label of 0 in the cluster, i.e. the number of normal samples.

[0041] Through the above cyclic clustering, when the number of clusters is set to K s When the loop termination condition is met, the resulting low longitudinal splitting cluster is designated as C. s .

[0042] (4.2) Extracting the process parameter range for low longitudinal crack clusters

[0043] Step (4.1) identified the low longitudinal splitting cluster C. s The low longitudinal crack sample library corresponding to this cluster is Z. t×(d-1) This includes t samples, each containing d-1 attributes. The data is obtained from a low longitudinal crack sample library Z. t×(d-1) Inverse normalization is performed to restore the d-1 normalized attributes to the initial d-1 attributes representing process parameters, resulting in the restored low longitudinal crack sample library O. t×(d-1) The inverse normalization calculation formula is as follows:

[0044] O ij =Z ij ×(D jmax -D jmin )+D jmin (5)

[0045] In the formula, O ij Z represents the value of the j-th attribute in the i-th sample, where i = 1, 2, ..., t, j = 1, 2, ..., d-1; ij For the low longitudinal crack sample library before inverse normalization; D jmin The second step is to reduce the dimensionality to obtain the database D. r×d The minimum value of the j-th attribute in D; jmax For database D r×dThe maximum value of the j-th attribute.

[0046] Finally, a low longitudinal crack sample library O was extracted. t×(d-1) The value ranges corresponding to the d-1 attributes, i.e. the value intervals corresponding to each process parameter, are used to optimize the process for longitudinal cracks on the surface of continuously cast billets.

[0047] Furthermore, the above-mentioned method for optimizing the longitudinal crack process on the surface of continuously cast billets is applicable to the optimization of the longitudinal crack process for continuously cast billets such as slabs, square billets, round billets, and irregular-shaped billets.

[0048] The beneficial effects of this invention are as follows:

[0049] This invention selects and reduces the dimensionality of the main factors influencing longitudinal crack occurrence through Spearman correlation analysis. Based on this, a process optimization model for longitudinal cracks on the surface of continuously cast billets based on K-Means clustering is established and applied to continuous casting production practice. The integrated method for optimizing the longitudinal crack process of continuously cast billets using Spearman correlation analysis and K-Means clustering demonstrates high efficiency and good adaptability, providing a technical means for intelligent optimization of longitudinal cracks in cast billets. Attached Figure Description

[0050] Figure 1 This is a flowchart of the process optimization for longitudinal cracks on the surface of continuously cast billets based on K-Means clustering;

[0051] Figure 2 This is a heatmap of the Spearman correlation coefficients between various process factors;

[0052] Figure 3 This is a schematic diagram of clustering results under different numbers of clusters. Detailed Implementation

[0053] The present invention will be further described below through specific embodiments and in conjunction with the accompanying drawings, but this does not limit the present invention.

[0054] A method for optimizing the longitudinal crack process on the surface of continuously cast billets based on K-Means clustering includes the following steps:

[0055] Step 1: Building the attribute data sample library

[0056] like Figure 1The diagram shows the process optimization flowchart for longitudinal cracks on the surface of continuously cast billets based on K-Means clustering. The main process is as follows: First, based on the main factors affecting longitudinal crack occurrence in continuous casting production and combined with production quality reports, a feature sample library of longitudinal crack influencing factors is established. Second, Spearman correlation coefficient is used to select attributes and reduce the dimensionality of samples in the sample library. Third, K-Means clustering is used to iteratively cluster the dimensionality-reduced data samples, filtering and locking down clusters corresponding to low longitudinal crack incidence rates. Fourth, the process parameter ranges corresponding to the low longitudinal crack incidence rate clusters are extracted, thereby completing the optimization of various process parameters.

[0057] First, for the same steel grade and protective slag production process, starting from the mechanism of longitudinal crack formation on the surface of the continuously cast billet, 19 process factors affecting the occurrence of longitudinal cracks were selected. These mainly include casting speed, superheat, maximum liquid level fluctuation, sprue immersion depth, copper plate inlet water temperature, billet width, taper, cooling water volume on all four sides of the copper plate, average cooling water volume on the wide side, average cooling water volume on the narrow side, average inlet and outlet water temperature difference on all four sides, average inlet and outlet water temperature difference on the wide side, average inlet and outlet water temperature difference on the narrow side, average heat flow on all four sides, average heat flow on the wide side, average heat flow on the narrow side, heat flow difference between the outer and inner arcs, heat flow difference between the right and left sides, and heat flow difference between the narrow and wide sides.

[0058] Taking a semi-peritectic steel grade with high susceptibility to longitudinal cracking in a steel plant as an example, a slab data sample library containing the aforementioned 19 process factors was obtained by retrospectively analyzing historical casting data. This library includes 1116 normal slabs and 123 longitudinally cracked slabs. Each slab represents a sample. For normal samples, the 19-dimensional attributes represent the average value of each process factor during the entire time the slab passes through the crystallizer; for longitudinally cracked samples, the 19-dimensional attributes represent the average value of each process factor during the time period when the longitudinal crack occurred. Based on production report records, the label for each sample was determined, ultimately resulting in the sample library D. 1239×20 The distribution of data for each attribute is shown in Table 1.

[0059] Table 1. Distribution of data for each attribute

[0060]

[0061]

[0062] Step 2: Attribute Selection and Dimensionality Reduction

[0063] Spearman correlation coefficient was used to analyze the data sample database D obtained in the first step. 1239×20 Correlation analysis was performed on the attributes, and the correlation assessment results are as follows: Figure 2 As shown.

[0064] For highly correlated attribute combinations, retain only one attribute to ensure that the correlation coefficient between any two attributes is less than 0.4, thus achieving optimal performance for the data sample database D.1239×20 Attribute selection and dimensionality reduction. The retained attributes mainly include superheat, maximum liquid level fluctuation, immersion depth of the nozzle, copper plate inlet temperature, taper, total water volume on all four sides, average heat flux on all four sides, heat flux difference between the outer and inner arcs, heat flux difference between the right and left sides, and heat flux difference between the narrow and wide sides—a total of 10 attributes. Correlation analysis was used to optimize the data sample database D. 1239×20 The dimensionality of the data sample database D was reduced from 19 attributes to 10 attributes. 1239×11 After dimensionality reduction, the 11th column of the sample library still contains the sample label, where 1 represents a longitudinally fractured sample and 0 represents a normal sample.

[0065] The third step is to perform K-Means clustering on the dimensionality-reduced data samples.

[0066] (3.1) Normalize the data samples

[0067] For sample database D 1239×11 The label records in column 11 are retained, and the 10 attributes are normalized to obtain a new data sample library X. 1239×11 The normalization results are shown in Table 2.

[0068] Table 2. Details of the data sample library after normalization.

[0069]

[0070] (3.2) Perform K-Means clustering on the normalized data.

[0071] Preset different numbers of clusters (2≤K≤10) for the normalized data sample database X. 1239×10 Perform cyclic clustering. Taking K=3 clusters as an example, the specific implementation of the K-Means clustering process is as follows:

[0072] (3.2.1) Initialize the cluster centroid

[0073] Data sample library X 1239×10 ={x1,x2,...,x 1239 Three samples were randomly selected as three initial centroids {μ1, μ2, μ3}.

[0074] And initialize the class cluster collection, let

[0075] (3.2.2) Sample clustering

[0076] Examine the data sample database X respectively 1239×10 ={x1,x2,...,x 1239 The distance between each sample and the three initial centroids {μ1,μ2,μ3} in the matrix is: T ij=||x i -μ j ||2, where i = 1, 2, ..., 1239, j = 1, 2, 3; determine the current sample x based on the nearest centroid. i Cluster label: λ i =argmin j∈{1,2,3} T ij ; take sample x i Assign to the corresponding cluster:

[0077] (3.2.3) Update the cluster centroid

[0078] Calculate the new mean vector for each of the three clusters, and use it as the centroid of the new cluster:

[0079]

[0080] In the formula, μ' j Let |C be the centroid of the new cluster, where j = 1, 2, 3; j | represents the number of samples in the j-th cluster, and x represents the vector in the j-th cluster.

[0081] (3.2.4) Termination condition judgment

[0082] For the three centroids of the clusters, if the mean vectors of the previous and next iterations no longer change, i.e., μ' appears... j =μ j If the result is positive, it indicates that the clustering process is complete and three clusters {C1, C2, C3} have been obtained, with the corresponding centroids {μ1, μ2, μ3}. Otherwise, continue to execute steps (3.2.1) to (3.2.4) until the clustering result meets the conditions.

[0083] By performing the above steps, the K-Means clustering process is finally completed when the number of clusters K=3. The calculated cluster centroids are as follows:

[0084] μ1=[0.4664,0.0403,0.4201,0.4860,0.6386,0.0091,0.7171,0.4887,0.5364,0.5449]

[0085] μ2=[0.4826,0.0470,0.4391,0.4755,0.6121,0.7624,0.7351,0.5314,0.5531,0.5618]

[0086] μ3=[0.4276,0.0431,0.4412,0.5272,0.5359,0.5112,0.6667,0.4265,0.2348,0.4788]

[0087] Step 4: Optimize the process for longitudinal cracks.

[0088] (4.1) Locking in low longitudinal fracture clusters

[0089] The above iterative clustering process terminates when a cluster with a longitudinal cracking rate of 0% is found. If no cluster with a longitudinal cracking rate of 0% is found, the cluster with the lowest longitudinal cracking rate in each iterative clustering process is locked. The longitudinal cracking rate l corresponding to each cluster is determined by the data sample database X. 1239×11 The label record corresponding to column 11 is calculated.

[0090] Cyclic clustering results with different numbers of clusters are as follows Figure 3 As shown in (a) to (f), when the number of clusters is set to 7, a cluster with a longitudinal cracking rate of 0% appears, which satisfies the loop termination condition. The low longitudinal cracking cluster obtained at this time is designated as C5.

[0091] (4.2) Extraction of process parameter range for low longitudinal crack clusters

[0092] Step (4.1) identified the low longitudinal crack cluster C5, which contains 126 samples, each with 10 attributes, thus obtaining the low longitudinal crack sample library Z corresponding to the low longitudinal crack cluster. 126×10 By analyzing the low longitudinal crack sample library Z 126×10 Inverse normalization is performed to restore the 10 normalized attributes to the initial 10 attributes representing process parameters, resulting in the restored low longitudinal crack sample library O. 126×10 The inverse normalization calculation results are shown in Table 3.

[0093] Table 3. Details of the low longitudinal crack sample library after denormalization.

[0094]

[0095] Finally, a low longitudinal crack sample library O was extracted. 126×10The value ranges corresponding to the 10 attributes, i.e., the value intervals corresponding to each process parameter, can be regarded as the targets for longitudinal cracking process optimization. The optimization results are shown in Table 4. Taking superheat and nozzle immersion depth as examples, before process optimization, their variation ranges in production were 9–40℃ and 118–180mm, respectively. After optimization, their recommended value intervals were 13–33℃ and 122–150mm, respectively. For other parameters, the recommended value intervals can also be determined by referring to the same method. Before process optimization, longitudinal cracks were found in 123 out of 1239 cast billets, with an incidence rate of 9.93%. After optimizing the process using this method, for this longitudinal crack-sensitive steel grade, longitudinal cracks were found in 17 out of 500 cast billets, with an incidence rate of 3.4%. Before and after process optimization, the incidence rate of longitudinal cracks decreased significantly, and the quality of continuously cast billets was significantly improved.

[0096] Table 4 Comparison of process optimization before and after.

[0097]

[0098]

[0099] The above-described embodiments are merely illustrative of the implementation methods of the present invention, but should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for optimizing the process of longitudinal cracks on the surface of continuously cast billets based on K-Means clustering, characterized in that, The process optimization method first establishes a sample library with different longitudinal crack incidence rates; second, it uses Spearman correlation coefficient to perform attribute selection and dimensionality reduction on the samples in the sample library; then, it uses K-Means clustering to perform iterative clustering on the dimensionality-reduced data samples, filtering and locking the clusters corresponding to low longitudinal crack incidence rates; finally, it extracts the process parameter ranges corresponding to the low longitudinal crack incidence rate clusters, thereby optimizing various process parameters. Includes the following steps: Step 1: Building the attribute data sample library For the same steel grade and protective slag production process, historical data from steel mills were reviewed to construct a historical sample database containing both normal and longitudinal crack conditions, starting from the main factors affecting longitudinal cracking in continuous casting. The meaning of each sample is as follows: for normal condition slabs, the attributes of the sample are the average values ​​of each process parameter during the entire time the slab passes through the crystallizer; for longitudinal crack samples, the attributes of the sample are the average values ​​of each process parameter during the time period in which the longitudinal crack occurred in the slab. The acquired sample database is defined as follows: ,in r The row number represents the number of samples, including normal samples and longitudinally fractured samples; c The number of attribute columns, from column 1 to... c -1 column is c -1 attribute, that is c -1 is the mean of the parameters, the th c The column is the longitudinal crack label corresponding to the sample, where 1 represents a longitudinal crack sample and 0 represents a normal sample. Step 2: Attribute Selection and Dimensionality Reduction Using the Spearman correlation coefficient analysis method, for variables X and Y of N Original sample data ( X i , Y i ),( i = 1,2,3,…,N), x i for X i In variables X The rank of the sample values, i.e., the variable X The sample values ​​are arranged in ascending order. X i ordinal number; y i for Y i In variables Y The rank of the sample values, the variable X and Y The Spearman correlation coefficient between them can be obtained by the following formula: (1) In the formula, ρ For variables X and Y Spearman correlation coefficient; d i for X i , Y i rank difference x i - y i ; N The number of samples; The Spearman correlation coefficient ranges from -1 to 1 and reflects the sign and strength of the correlation between variables. A positive value represents a positive correlation, and a negative value represents a negative correlation. The larger the absolute value of the coefficient, the stronger the correlation between the two variables. Spearman correlation coefficient was used to analyze the data sample database obtained in the first step. Perform correlation analysis to achieve the analysis of the data sample database. The selection and dimensionality reduction of each attribute; when the absolute value of the correlation coefficient between any two attributes exceeds 0.4, these two attributes are identified as a highly correlated combination; the highly correlated attribute combinations are processed, retaining only one attribute, ensuring that the correlation coefficient between any two attributes is less than 0.4, thus achieving the purpose of attribute selection and dimensionality reduction; the data sample database is then analyzed through correlation analysis. middle c -1 attribute reduced to dimensionality d -1 attribute, to obtain the dimensionality-reduced data sample library. After dimensionality reduction, the d-th column of the sample library still contains the sample label, where 1 represents a longitudinally fractured sample and 0 represents a normal sample. Step 3: Constructing the K-Means clustering model K-Means is pre-defined K Values, update iterative clusters and their centroids, and divide the data samples into... K K-Means clustering consists of non-overlapping clusters; the specific implementation steps are as follows: (3.1) Normalize the data samples. For sample database Retain the first d The column's label record will d -1 attribute is normalized separately to obtain a new data sample database. ; (3.2) Perform K-Means clustering on the normalized results. First, from the new data sample library Random selection K First, select one sample as the initial centroid; second, calculate the distance between each sample and the initial centroid; then, assign each sample to the cluster represented by its nearest centroid; finally, recalculate the distance between each sample and the initial centroid. K The sample mean in each cluster is used as the new cluster centroid; the above process is repeated to iterate over the cluster centroids until all mean vectors are no longer updated. Step 4: Optimization of longitudinal cracking process The purpose of longitudinal cracking process optimization is to find the process range where longitudinal cracking does not occur or the probability of longitudinal cracking is extremely low, that is, to identify the cluster with the lowest longitudinal cracking rate, extract the process information contained in this cluster, and optimize various process parameters. The specific process is as follows: (4.1) Locking in low longitudinal crack clusters Before performing K-means clustering, the number of clusters needs to be preset in advance, and the determination of the number of clusters usually needs to be combined with the actual problem. The purpose of the current process optimization is to distinguish between clusters with high longitudinal crack incidence and clusters with low longitudinal crack incidence. The specific number of clusters cannot be determined before clustering. The cyclic clustering method is used to continuously change the number of clusters, thereby locking in the target cluster with low longitudinal crack incidence. By presetting different numbers of clusters ( ), loop through the data sample database Perform K-means clustering until the cyclic clustering termination condition is met; the cyclic clustering process ends when a cluster with a longitudinal splitting rate of 0% is found, and the current cluster with a longitudinal splitting rate of 0% is the target; if a cluster with a longitudinal splitting rate of 0% is not found, then the cluster with the lowest longitudinal splitting rate in the 10 cyclic clustering results is locked. A certain type of cluster C x Corresponding longitudinal crack incidence l From the data sample library The label record corresponding to column d is calculated using the following formula: (4) In the formula, for C x The number of samples in the cluster whose label is 1, i.e. the number of longitudinally split samples; for C x The number of samples in a cluster whose label is 0, i.e., the number of normal samples; Through the above iterative clustering, when the number of clusters is set to... K s When the loop termination condition is met, the resulting low longitudinal split cluster is designated as follows: C s ; (4.2) Extracting the process parameter range for low longitudinal crack clusters Step (4.1) identified low longitudinal splitting clusters. C s The low longitudinal crack sample library corresponding to this cluster is The dataset contains t samples, each with d-1 attributes; this is achieved through a low longitudinal crack sample library. Perform inverse normalization, and d -1 normalized attribute is restored to its initial state. d -1 attribute representing process parameters, resulting in the restored low longitudinal crack sample library. The inverse normalization calculation formula is as follows: (5) In the formula, For the first i In the nth sample j The value of the attribute, where ; This is a sample library of low longitudinal cracks before inverse normalization; To obtain the database in the second step of dimensionality reduction The Middle j The minimum value of the attribute; For database The Middle j The maximum value of the attribute; Finally, a low longitudinal crack sample library was extracted. middle d -1 The value range corresponding to the attribute, that is, the value range corresponding to each process parameter, realizes the process optimization of longitudinal cracks on the surface of continuous casting billet.

2. The method for optimizing the longitudinal crack process on the surface of continuously cast billets based on K-Means clustering according to claim 1, characterized in that, In the first step, to improve the accuracy of clustering results in subsequent models, it is ensured that the total number of samples in the data sample library is no less than 500, and at the same time, the number of longitudinal crack samples is no less than 25.

3. The method for optimizing the longitudinal crack process on the surface of continuously cast billets based on K-Means clustering according to claim 1, characterized in that, The normalization process in step (3.1) is as follows: (2) In the formula, For sample database The Middle i In the nth sample j The value of the attribute, where ; For data sample library The Middle j The minimum value of the attribute; For the first j The maximum value of the attribute.

4. The method for optimizing the longitudinal crack process on the surface of continuously cast billets based on K-Means clustering according to claim 1, characterized in that, The specific process of step (3.2) is as follows: (3.2.1) Initialize the cluster centroid From the data sample library Random selection K Each sample is used as K An initial centroid and initialize the cluster set, let ; (3.2.2) Divide the sample into clusters Examine the data sample database separately Each sample in K An initial centroid Distance between: ,in The subscript i represents the i-th sample, and j represents the j-th initial centroid; The current sample is determined based on the nearest centroid. x i Cluster label: ; take the sample x i Assign to the corresponding cluster: ; (3.2.3) Update the cluster centroid Calculate new mean vectors for each cluster and use them as the centroids of the new clusters: (3) In the formula, For the centroid of the new cluster, where ; For the first j The number of samples in a cluster, x For the first j Vectors in a cluster; (3.2.4) Determine the termination condition For all cluster centroids, if the mean vector of the iterations no longer changes, i.e., when... This indicates that the clustering process is complete and the results are obtained. k Clusters The corresponding cluster centroids are respectively Otherwise, continue with steps (3.2.1) to (3.2.4) until the clustering results meet the termination conditions.

5. A method for optimizing the longitudinal crack process on the surface of continuously cast billets based on K-Means clustering as described in any one of claims 1-4, characterized in that, The method for optimizing the longitudinal crack process on the surface of continuously cast billets is applicable to the optimization of the longitudinal crack process for slabs, square billets, round billets, irregular-shaped billets, and other continuously cast billets.

Citation Information

Patent Citations

  • Method for controlling surface crack of continuous cast slab

    CN104889358A

  • Method for preventing surface cracks on low-carbon boron-containing steel

    CN107630172A