A method for process control of longitudinal cracks in high manganese steel based on isolation forest and extreme random tree

Abstract

This invention discloses a method for controlling longitudinal cracks in high-manganese steel based on isolated forest and extreme random tree methods. The method collects process parameters and crack defect data during the continuous casting process of high-manganese steel to construct a longitudinal crack database, and uses the isolated forest algorithm to remove abnormal operating condition data. Then, the extreme random tree algorithm is used to establish the nonlinear relationship between process parameters and longitudinal crack occurrence, and the model hyperparameters are optimized using a Bayesian optimization algorithm. After training, the model's built-in feature_importance module is used to rank the process parameters by importance, and the SHAP method is used to perform feature dependency analysis on the high-ranking and controllable process parameters to quantify the influence of key process parameters on crack formation. This invention enables data-driven process optimization, improves the quality of high-manganese steel billets, reduces the incidence of longitudinal cracks, and provides an efficient and interpretable intelligent process control method for the metallurgical industry.

Description

Technical Field

[0001] This invention relates to the field of metallurgical continuous casting technology, and in particular to a method for controlling longitudinal cracks in high manganese steel based on isolated forests and extreme random trees. Background Technology

[0002] High-manganese steel is widely used in high-strength structural materials due to its excellent mechanical properties. However, during continuous casting, longitudinal cracks are prone to form in the billets, affecting product quality and production stability. Currently, crack control mainly relies on empirical adjustments to process parameters such as casting speed, cooling intensity, and superheat, but this method is difficult to accurately and quickly reduce the risk of longitudinal cracks in the target steel grade.

[0003] In recent years, most researchers have used numerical simulation and experimental studies to analyze the crack formation mechanism in continuous casting and to explore the influence of process parameters on crack defects. For example, patent application CN202411634313.2 reduces the occurrence of billet cracks by adjusting the narrow taper of the chamfered crystallizer, the cooling water flow rate, and the casting superheat. However, such methods are often costly to experiment with and difficult to update in real time. Currently, with the widespread application of machine learning technology in the metallurgical field, data-driven methods provide new solutions for crack prediction and optimization. Using machine learning methods, key factors affecting crack formation can be mined from historical production data, and process parameters can be optimized through data-driven approaches. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extreme random trees. This method combines isolated forest and extreme random tree algorithms for visual analysis, accurately identifies key process parameters affecting crack formation, and optimizes process parameters based on on-site production experience to reduce the crack incidence rate.

[0005] The technical solution of this invention is implemented as follows:

[0006] This invention provides a method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extreme random trees. The method includes:

[0007] S1. Collect process parameters and crack defect data of high manganese steel from the production line;

[0008] S2. After deleting missing data and removing abnormal working condition data from the high manganese steel process parameters and crack defect data, construct a high manganese steel longitudinal crack database.

[0009] S3. Divide the data in the database into a training set and a test set;

[0010] S4. Use the training set to construct an extreme random tree ensemble model, and perform hyperparameter optimization on the trained model to establish a nonlinear relationship between process parameters and the occurrence of longitudinal cracks.

[0011] S5. The trained model is tested on the test set to obtain the model's predicted value for whether longitudinal cracks occur under different process parameters, and the predicted value is compared with the actual value. The area under the curve is used as the evaluation criterion. If the area under the curve is within the preset range, the trained model is considered qualified and the model is used to predict longitudinal cracks in high manganese steel. Otherwise, the trained model is considered unqualified, and steps S2 to S5 are repeated.

[0012] S6. Perform feature importance analysis and rank the trained models.

[0013] S7. Select a preset number of continuous casting process parameters that are ranked high in importance and are easy to adjust on-site. Use a visualization algorithm to perform feature dependency analysis on the selected parameters to obtain the parameter value range that is conducive to reducing the occurrence of billet cracks.

[0014] S8. Based on the parameter value range obtained in step S7 and combined with prior knowledge, adjust the on-site process parameters to control the occurrence of longitudinal cracks in the billet.

[0015] In one embodiment, the process parameters in step S1 include casting speed, nozzle insertion depth, superheat, and type of protective slag.

[0016] In one embodiment, in step S2, an isolated forest machine learning algorithm is used to remove outlier data:

[0017] Assume the dataset is ,sample The average path length in a forest is:

[0018]

[0019] The standardized outlier score is:

[0020]

[0021] in, Representative sample In the Path length in a tree The total number of trees in the forest. For sample size The expected path length is used for normalization to make The value ranges from 0 to 1; when If the value exceeds the set threshold, the sample is considered an outlier and is removed.

[0022] In one embodiment, in step S3, 75% to 85% of the data is selected as training set data, and the remaining data is used as test set data.

[0023] In one embodiment, in step S4, four hyperparameters—max_leaf_nodes, max_features, max_depth, and n_estimators—are selected from the model for hyperparameter optimization using the Bayesian optimization algorithm. The range of hyperparameter max_leaf_nodes is 200–2000; the range of hyperparameter max_features is 0.2–1; the range of hyperparameter max_depth is 1–500; and the range of hyperparameter n_estimators is 50–2000. Bayesian optimization is then used to optimize the model's hyperparameters.

[0024] In one embodiment, in step S5, features are ranked by importance based on gain or weight values ​​according to the interpretability of the trained model itself.

[0025] In one embodiment, in step S6, based on the preset number of continuous casting process parameters selected in step S5 that are prioritized in importance and easily adjustable on-site, SHAP is used to perform feature dependency analysis on them. The feature dependency analysis using SHAP includes: calculating the marginal contribution of each process parameter in the model to the occurrence of longitudinal cracks, evaluating the contribution of different feature values ​​of the feature to the result, and quantitatively analyzing the correlation between process parameters and longitudinal cracks from a machine learning perspective. When the SHAP value corresponding to the range of a certain process parameter is greater than 0, it is considered that the range of the process parameter is conducive to the occurrence of longitudinal cracks; when the SHAP value corresponding to the range of a certain process parameter is less than 0, it is considered that the range of the process parameter is unfavorable to the occurrence of longitudinal cracks.

[0026] The method in this embodiment has the following beneficial effects:

[0027] This embodiment's method, based on accurately predicting whether longitudinal cracks will appear in the cast billet, quantitatively analyzes the impact of key process parameters on the occurrence of longitudinal cracks in the cast billet, and obtains the optimal range of process parameters. This allows on-site engineers to quickly and accurately adjust the process to reduce the occurrence of longitudinal cracks in high-manganese steel. This embodiment's method can significantly reduce the blind spots for smelting engineers in optimizing process parameters, while reducing time and economic losses caused by trial and error, providing a data-driven solution for optimizing the continuous casting process of high-manganese steel. Furthermore, this embodiment's method has universality and can also provide a reference for reducing longitudinal cracks in other steel grades, promoting the application of intelligent manufacturing in the metallurgical field. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extreme random trees, as described in an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram illustrating the implementation process of an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram showing the importance ranking of process parameters affecting longitudinal cracks in high-manganese steel, obtained based on the interpretability of the extreme random tree model in an embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram of the characteristic dependency graph analysis of the protective slag parameters based on SHAP values ​​in an embodiment of the present invention;

[0032] Figure 5 This is a schematic diagram of the superheat parameter feature dependency graph analysis based on SHAP values ​​in an embodiment of the present invention;

[0033] Figure 6 This is a schematic diagram of the pull speed parameter feature dependency graph analysis based on SHAP value in an embodiment of the present invention. Detailed Implementation

[0034] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0035] This invention provides a method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extremely random trees. Figure 1 As shown, the method includes:

[0036] S1. Collect process parameters and crack defect data of high manganese steel from the production line;

[0037] S2. After deleting missing data and removing abnormal working condition data from the high manganese steel process parameters and crack defect data, construct a high manganese steel longitudinal crack database.

[0038] S3. Divide the data in the database into a training set and a test set;

[0039] S4. Use the training set to construct an extreme random tree ensemble model, and perform hyperparameter optimization on the trained model to establish a nonlinear relationship between process parameters and the occurrence of longitudinal cracks.

[0040] S5. The trained model is tested on the test set to obtain the model's predicted value for whether longitudinal cracks occur under different process parameters, and the predicted value is compared with the actual value. The area under the curve is used as the evaluation criterion. If the area under the curve is within the preset range, the trained model is considered qualified and the model is used to predict longitudinal cracks in high manganese steel. Otherwise, the trained model is considered unqualified, and steps S2 to S5 are repeated.

[0041] S6. Perform feature importance analysis and rank the trained models.

[0042] S7. Select a preset number of continuous casting process parameters that are ranked high in importance and are easy to adjust on-site. Use a visualization algorithm to perform feature dependency analysis on the selected parameters to obtain the parameter value range that is conducive to reducing the occurrence of billet cracks.

[0043] S8. Based on the parameter value range obtained in step S7 and combined with prior knowledge, adjust the on-site process parameters to control the occurrence of longitudinal cracks in the billet.

[0044] This embodiment's method, based on accurately predicting whether longitudinal cracks will appear in the cast billet, quantitatively analyzes the impact of key process parameters on the occurrence of longitudinal cracks in the cast billet, and obtains the optimal range of process parameters. This allows on-site engineers to quickly and accurately adjust the process to reduce the occurrence of longitudinal cracks in high-manganese steel. This embodiment's method can significantly reduce the blind spots for metallurgical engineers in optimizing process parameters, while also reducing time and economic losses caused by trial and error, providing a data-driven solution for optimizing the continuous casting process of high-manganese steel. Furthermore, this embodiment's method has universality and can also provide a reference for reducing longitudinal cracks in other steel grades, promoting the application of intelligent manufacturing in the metallurgical field.

[0045] Specifically, see Figure 2 In practical applications, the implementation process of this embodiment can be completed as follows:

[0046] S1: Collect process parameters and crack defect data for high-manganese steel from the production line. This may include: specific steel composition, casting speed, nozzle insertion depth, superheat, and type of protective slag.

[0047] S2: The specific composition of the steel grade, casting speed, nozzle insertion depth, superheat, and type of protective slag are used as input features. The presence or absence of longitudinal cracks is binary encoded and used as tags to construct a database. Duplicate data is deleted, and the sample is calculated using formulas (1)-(2). Outlier data is removed by using the average path length and standardized outlier score in the forest.

[0048] Here, the Isolation Forest machine learning algorithm can be used to remove outlier data. This involves constructing a forest by randomly selecting features and cut points, and calculating the path length of each sample point within the tree. Samples with shorter paths are more likely to be outliers.

[0049] Assume the dataset is Then the sample The average path length in the forest is shown in Equation (1), and the standardized outlier score is shown in Equation (2):

[0050] (1)

[0051] (2)

[0052] in, Representative sample In the Path length in a tree The total number of trees in the forest. For sample size The expected path length is used for normalization to make The value ranges from 0 to 1; when If the value exceeds the set threshold, the sample is considered an outlier and is removed.

[0053] The isolated forest algorithm described above can effectively remove abnormal process parameter data, improve the quality of the dataset, and thus enhance the accuracy and stability of the extreme random tree prediction model.

[0054] S3: Divide the high-manganese steel longitudinal crack database into 85% training set and 15% test set.

[0055] S4: The training set of high-manganese steel longitudinal cracks divided in S3 is used to train the extreme random tree model.

[0056] S41: Based on the construction principle of extreme random trees, select the hyperparameters with significant impact for parameter value adjustment. The selection and range of hyperparameters for extreme random trees are shown in Table 1:

[0057]

[0058] S42: Using AUC as the evaluation metric, Bayesian optimization is employed to optimize the hyperparameters of an extreme random tree. Ten-fold cross-validation is used during each optimization iteration to improve the model's generalization ability. Bayesian optimization constructs a probabilistic surrogate model and utilizes the expected improvement criterion to search for the optimal hyperparameters. The optimization process stops when the number of iterations meets a preset stopping condition, and the optimal hyperparameter combination and the corresponding optimal objective function value are output.

[0059] S43: The optimal hyperparameter combination is used for extreme random tree training. The optimal extreme random tree hyperparameter combination is shown in Table 2:

[0060]

[0061] S5: Compare the predicted results of the optimized extreme random tree on the test set with the actual results, and use AUC as the evaluation metric to evaluate the model performance. If AUC ≥ 0.8, the constructed extreme random tree model is considered qualified and can be used for subsequent feature importance ranking and feature dependency analysis.

[0062] S6: Perform feature importance analysis and ranking on the qualified extreme random tree model trained by S5, such as... Figure 3 As shown.

[0063] When ranking, features can be ranked based on their importance according to the interpretability of the trained model itself, using either the gain value or the weight value.

[0064] S7: Based on the importance ranking results obtained in S6, the composition of the protective slag, superheat, and casting speed were identified as the most important continuous casting process parameters that are also easily adjustable on-site. A visualization algorithm was used to perform feature dependency analysis on the selected parameters to obtain the range of parameter values ​​that are beneficial in reducing billet cracking. For example... Figures 4-6 As shown.

[0065] Here, the marginal contribution of each process parameter in the extreme random tree to the occurrence of longitudinal cracks can be calculated to evaluate the contribution of different feature values ​​to the results, and the correlation between process parameters and longitudinal cracks can be quantitatively analyzed from a machine learning perspective. When the SHAP value corresponding to the range of a process parameter is greater than 0, it is considered that the range of the process parameter is conducive to the occurrence of longitudinal cracks; when the SHAP value corresponding to the range of a process parameter is less than 0, it is considered that the range of the process parameter is unfavorable to the occurrence of longitudinal cracks.

[0066] S8: Adjust the on-site process parameters based on the results obtained in S7 and in combination with prior knowledge to control the occurrence of longitudinal cracks in the billet.

[0067] Comparative Example 1

[0068] For a period of time, three different types of protective slag (No. 1, No. 2, and No. 3) were used successively in the production of high-manganese steel on a continuous casting machine at this steel plant. Because the process parameters were constantly adjusted during the slag replacement, it was impossible to directly determine the optimal slag selection from the produced billets. In a survey of 30 billets along the casting direction, an average of 29 longitudinal cracks were found per billet.

[0069] Example 1

[0070] Based on the importance ranking results of extreme random tree analysis, the type of flux was found to be the most significant parameter affecting the occurrence of longitudinal cracks in high-manganese steel. Characteristic dependency analysis using SHAP revealed that even with continuously changing process parameters, the SHAP value of flux No. 3 used in high-manganese steel production was mostly less than 0 and generally lower than that of fluxes No. 1 and No. 2. This indicates that this parameter range has an inhibitory effect on the occurrence of longitudinal cracks in high-manganese steel, which is beneficial for producing high-quality cast billets. Fluid No. 3 was selected for continuous casting production. In a survey of 30 cast billets along the casting direction, an average of 18 longitudinal cracks were observed per billet. The reduced number of longitudinal cracks, mostly micro-cracks, improved the quality of the cast billets.

[0071] Comparative Example 2

[0072] During a period of time, the superheated temperature of a continuous casting machine at this steel plant ranged from 12℃ to 35℃, and the casting speed ranged from 0.75m / min to 0.95m / min during the production of high-manganese steel. Due to the large fluctuations in process parameters, the quality stability of the cast billets was poor, exhibiting significant variability. When No. 3 protective slag was used in the production of high-manganese steel, an average of 18 longitudinal cracks were found in each of the 30 cast billets surveyed along the casting direction.

[0073] Example 2

[0074] Based on the importance ranking results of extreme random tree analysis, superheat and casting speed were found to be the main parameters affecting longitudinal cracks in high-manganese steel, besides the protective slag. Characteristic dependency analysis using SHAP revealed that when the superheat was around 15℃ to 19℃ and the casting speed was between 0.8 m / min and 0.86 m / min, the corresponding SHAP values ​​were mostly less than 0. This indicates that this parameter range has an inhibitory effect on the formation of longitudinal cracks in high-manganese steel, which is beneficial for producing high-quality cast billets. Slag type 3 was selected, with the superheat controlled at approximately 17℃ and the casting speed at approximately 0.83 m / min. In a total of 30 cast billets, an average of 13 longitudinal cracks occurred per billet. The reduction in the number of longitudinal cracks improved the quality of the cast billets.

[0075] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0076] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extremely random trees, characterized in that, The method includes: S1. Collect process parameters and crack defect data of high manganese steel from the production line; S2. After deleting missing data and removing abnormal working condition data from the high manganese steel process parameters and crack defect data, construct a high manganese steel longitudinal crack database. S3. Divide the data in the database into a training set and a test set; S4. Use the training set to construct an extreme random tree ensemble model, and perform hyperparameter optimization on the trained model to establish a nonlinear relationship between process parameters and the occurrence of longitudinal cracks. S5. The trained model is tested on the test set to obtain the model's predicted value for whether longitudinal cracks occur under different process parameters, and the predicted value is compared with the actual value. The area under the curve is used as the evaluation criterion. If the area under the curve is within the preset range, the trained model is considered qualified and the model is used to predict longitudinal cracks in high manganese steel. Otherwise, the trained model is considered unqualified, and steps S2 to S5 are repeated. S6. Perform feature importance analysis and rank the trained models. S7. Select a preset number of continuous casting process parameters that are ranked high in importance and are easy to adjust on-site. Use a visualization algorithm to perform feature dependency analysis on the selected parameters to obtain the parameter value range that is conducive to reducing the occurrence of billet cracks. S8. Based on the parameter value range obtained in step S7 and combined with prior knowledge, adjust the on-site process parameters to control the occurrence of longitudinal cracks in the billet. In step S7, based on the preset number of continuous casting process parameters selected in step S6 that are prioritized in importance and easily adjustable on-site, SHAP is used to perform feature dependency analysis on them. The feature dependency analysis using SHAP includes: calculating the marginal contribution of each process parameter in the model to the occurrence of longitudinal cracks, evaluating the contribution of different feature values ​​to the results, and quantitatively analyzing the correlation between process parameters and longitudinal cracks from a machine learning perspective. When the SHAP value corresponding to the range of a certain process parameter is greater than 0, it is considered that the range of the process parameter is conducive to the occurrence of longitudinal cracks; when the SHAP value corresponding to the range of a certain process parameter is less than 0, it is considered that the range of the process parameter is unfavorable to the occurrence of longitudinal cracks.

2. The method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extremely random trees according to claim 1, characterized in that, In step S1, the process parameters include casting speed, nozzle insertion depth, superheat, and type of protective slag.

3. The method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extremely random trees according to claim 1, characterized in that, In step S2, the Isolation Forest machine learning algorithm is used to remove outlier data: Assume the dataset is ,sample The average path length in a forest is: The standardized outlier score is: in, Representative sample In the Path length in a tree The total number of trees in the forest. For sample size The expected path length is used for normalization to make The value ranges from 0 to 1; when If the value exceeds the set threshold, the sample is considered an outlier and is removed.

4. The method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extremely random trees according to claim 1, characterized in that, In step S3, 75% to 85% of the data is selected as the training set data, and the remaining data is used as the test set data.

5. The method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extremely random trees according to claim 1, characterized in that, In step S4, four hyperparameters—max_leaf_nodes, max_features, max_depth, and n_estimators—are selected from the model for hyperparameter optimization using the Bayesian optimization algorithm. The range of hyperparameter max_leaf_nodes is 200–2000; the range of hyperparameter max_features is 0.2–1; the range of hyperparameter max_depth is 1–500; and the range of hyperparameter n_estimators is 50–2000. Bayesian optimization is then used to optimize the model's hyperparameters.

6. The method for controlling longitudinal cracks in high-manganese steel based on isolated forests and extremely random trees according to claim 1, characterized in that, In step S5, features are ranked in order of importance based on gain or weight values, according to the interpretability of the trained model itself.

Images