Method for establishing continuous casting process-structure-performance relation based on interpretable machine learning

Abstract

The invention provides an interpretable machine learning-based continuous casting process-structure-performance relationship establishment method, which comprises the following steps of: 1, carrying out a laboratory scaling experiment, and producing continuous casting blank samples under different process conditions; 2, sampling is conducted on the center position of the continuous casting blank, an EBSD sample is prepared, and the structure morphology of the EBSD sample is observed; 3, extracting different structure characteristics, and measuring Vickers hardness values of the sample under different process conditions; step 4, analyzing the correlation between the structure characteristics and the hardness; and 5, dividing the organization characteristic cloud picture by using a grid method, selecting four items with the highest correlation degree for contributive superposition, and establishing a continuous casting process-organization-performance relationship. According to the method, the experimental data demand can be greatly reduced, the relation between the continuous casting process and the performance can be broken through by means of the microstructure morphology, the relation between the continuous casting process and the structure and the performance is established, and the cost is greatly reduced compared with industrial experiments.

Description

Technical Field

[0001] The present invention relates to the technical field of steel casting, and in particular to a method for establishing a continuous casting process-structure-performance relationship based on explainable machine learning. Background Art

[0002] The continuous casting process is an important part of the steel production process. The slabs, billets, and round billets produced by the continuous casting process can be used for further deep processing, and the products can be applied to manufacturing industries such as shipbuilding, aerospace, and bridges. As a key step in the cooling and forming of molten steel, the quality stability control and performance of the continuous casting billet play a key role in affecting the performance of subsequent deep-processing products.

[0003] Smart factories and smart manufacturing technologies are the main focus of my country in the next decade. With the implementation of a series of major national scientific and technological projects, my country's traditional metallurgical industry is developing rapidly in the direction of digitalization and intelligence. The scientific and technological innovation of continuous casting technology has begun to shift from tracking to a new stage of tracking, running side by side, and leading. However, due to the weak forward-looking basic theoretical research on continuous casting and the lack of material / process knowledge base, the formulation of continuous casting process relies on experience accumulation and simple "trial and error", and lacks process design theory and methods for the quality control needs of continuous casting billets. The process-organization-performance evolution law of the continuous casting process is unclear, resulting in low yield and qualified rates of key metal components for high-end equipment in my country, poor batch stability, high cost, low self-sufficiency rate, and high-end equipment and key components are still controlled by others.

[0004] CN202311463827.1 discloses an integrated optimization method for continuous casting process of niobium-containing austenitic stainless steel based on phase field method. The method obtains the solidification structure characterization results of stainless steel continuous casting billets through experiments, and simulates and predicts the evolution law of dendrite growth and solute distribution at the solid-liquid interface under different cooling rates. The solute concentration and grain boundary data are extracted as the initial conditions of the model, and the grain growth, grain boundary diffusion, and precipitation phase evolution process are described according to the solute concentration, precipitation phase size, grain size, and grain boundary parameters. At the same time, the elastic strain energy is coupled to obtain the equivalent stress-strain distribution of the organization over time. Based on the equivalent stress-strain distribution, the damage factor parameters under different continuous casting process conditions are obtained in combination with the damage criterion model to achieve continuous casting process optimization. The method establishes the relationship between the continuous casting billet and the organization, and realizes the rapid prediction and process optimization of the organizational morphology and crack sensitivity of the billet when different continuous casting processes are actually produced.

[0005] CN202310610803.8 discloses a continuous casting process simulation and prediction method. This method integrates parameters such as storage device parameters, steel grade parameters, process parameters, and model parameters, combines temperature simulation models, segregation prediction models, and solidification structure prediction models for simulation, and monitors the quality of the cast slab, calibrates temperature control, and optimizes process parameters based on the results of the simulation calculations. Based on the actual production process parameters and combined with the database method, this method systematically and comprehensively analyzes the relationship between the parameters in the continuous casting process and the solute distribution of the continuous casting slab, providing analysis means and data support for many fields such as process parameter optimization, internal quality monitoring, casting machine capacity development, and new product development in the continuous casting process.

[0006] CN202310610798.0 discloses a method for temperature simulation and macrostructure prediction of continuous casting processes. Based on the combination of continuous casting process parameters and the physical properties of the steel grade, this method applies a temperature simulation sub-model and a macrostructure prediction sub-model to predict the macrostructure, and stores the temperature simulation and macrostructure prediction results in a MongoDB database. This method visualizes the relationship between the continuous casting process temperature and the macrostructure as a cloud map, providing effective data result support for temperature control calibration, macrostructure control, and casting machine working capacity development in the continuous casting process, providing analysis means and multi-sample data.

[0007] CN202311630918.X discloses a method for adjusting continuous casting process parameters. This method constructs a continuous casting process parameter training set; trains an initial network model based on the continuous casting process parameter training set to obtain a continuous casting quality prediction model; and generates an adaptive optimization model for the continuous casting process parameters based on the continuous casting process evaluation criteria and the continuous casting quality prediction model. Based on the large-scale data collation and analysis of the relationship between the actual continuous casting production process data and the quality of the continuous casting slab, this method obtains an adaptive adjustment model for continuous casting process parameters, making the adjusted continuous casting process parameters closer to the continuous casting quality requirements, ensuring the quality of continuous casting products, and effectively improving the yield of continuous casting product quality.

[0008] The main defects of the above-mentioned prior arts are that they do not establish a connection between the continuous casting process - microstructure - properties, and secondly, they have a large demand for and dependence on continuous casting process parameters and performance data in the actual production process. The above prior arts are all quite different from the present invention and fail to solve the technical problems we want to solve. Therefore, we have invented a new method for establishing the relationship between continuous casting process - microstructure - properties based on interpretable machine learning. Summary of the Invention

[0009] The objective of the present invention is to provide a method for establishing the relationship between continuous casting process - microstructure - properties based on interpretable machine learning that optimizes the combination of continuous casting process parameters according to the performance of the target product during the steelmaking process, greatly improves the consistency of product performance, and establishes the relationship between continuous casting process - microstructure - properties.

[0010] The object of the present invention can be achieved by the following technical measures: a method for establishing the relationship between continuous casting process, microstructure and properties of interpretable machine learning, and the method for establishing the relationship between continuous casting process, microstructure and properties of interpretable machine learning includes: Step 1, conduct laboratory scaled-down experiments to produce continuous casting billet samples under different process conditions; Step 2, take samples from the central position of the continuous casting billet and prepare EBSD samples to observe their microstructural morphology; Step 3, extract different microstructural features and measure the Vickers hardness values of samples under different process conditions; Step 4, analyze the correlation relationship between microstructural features and hardness; Step 5, apply the grid method to divide the microstructural feature cloud map and select the four items with the highest degree of correlation for contribution superposition to establish the relationship between continuous casting process, microstructure and properties.

[0011] The object of the present invention can also be achieved by the following technical measures: In Step 1, when producing continuous casting billet samples under different process conditions, the casting speed is 0.5 - 2 m / min, the step interval is 0.3 mm / min, and the cooling water flow rate in the secondary cooling zone is 30 - 50 L / (t·min).

[0012] In Step 1, the range of molten steel composition is: C accounts for 350 - 550×10 -6 wt%, Si accounts for 45 - 70×10 -6 wt%, Mn accounts for 1550 - 2350×10 -6 wt%, P accounts for 130 - 190×10 -6 wt%, T.S accounts for 70 - 110×10 -6 wt%, Al accounts for 260 - 400×10 -6 wt%, Cr accounts for 140 - 210×10 -6 wt%, Ni accounts for 75 - 110×10 -6 wt%, Cu accounts for 75 - 110×10 -6 wt%, Mo accounts for 30 - 50×10 -6 wt%.

[0013] In Step 2, take small squares from the central area of the continuous casting billet under different process conditions, and use the method of electrolytic polishing to prepare EBSD samples and observe their microstructural morphology.

[0014] In Step 2, the electrolytic solution used for electrolytic polishing is a perchloric acid-alcohol mixed solution with a volume ratio of 1:3, the electrolytic polishing voltage is 20 - 21 V, the electrolytic time is 10 - 12 s, and the temperature of electrolytic polishing is 0 - 1℃.

[0015] In step 2, the small cube taken from the central region of the continuous casting billet is 10 10 a 10 mm small cube.

[0016] In step 3, the extracted tissue features include 8 characteristic indexes: average grain area size, average grain boundary angle mean value, proportion of large-angle grain boundaries, proportion of dynamically recrystallized grain sizes, average geometrically necessary dislocation density, proportion of dynamically recovered grains, aspect ratio, and proportion of deformed grain sizes.

[0017] In step 4, the SHAP analysis method is applied to analyze the correlation between tissue features and hardness.

[0018] The method for establishing the continuous casting process - microstructure - property relationship of interpretable machine learning further includes, after step 4, using the interpolation method to draw the single tissue feature cloud map of the sample under different process conditions. In step 5, the grid method is used to divide the area of the single tissue feature cloud map, and grids are superimposed according to the contribution of the single tissue feature to the performance. Finally, the correctness of the SHAP analysis method is verified, and the optimal continuous casting process parameter combination under the target performance in the continuous casting process of the corresponding grade steel is obtained.

[0019] The method for establishing the continuous casting process - microstructure - property relationship of interpretable machine learning in the present invention can give the optimal interval of continuous casting process parameters for specific types of steel products, and give the specific continuous casting process parameter combination according to the performance index requirements. The present invention adopts a scaled-down experiment under laboratory conditions, conducts a controlled variable experiment on two key indexes of the casting speed and the cooling water flow rate in the secondary cooling zone during the continuous casting process, produces continuous casting billets under different process parameter combinations, and conducts microstructure observation and hardness detection on the continuous casting billets. Combining the machine learning SHAP (SHapley Additive exPlanations) analysis method, the types of tissue parameters that have a greater impact on the performance are found, and the grid method is applied to analyze and verify the corresponding tissue parameters under different process parameters. This method requires less data volume and clearly elaborates the continuous casting process - microstructure - property relationship in combination with interpretable machine learning. Compared with the prior art, the present invention has the following beneficial effects: (1) The method of the present invention can greatly reduce the demand for experimental data.

[0020] (2) The method of the present invention can establish the relationship between the continuous casting process and the performance by means of the microstructure morphology, and establish the relationship between the continuous casting process - microstructure - property.

[0021] (3) The cost of the method of the present invention is greatly reduced compared with industrial experiments. Description of the Drawings

[0022] Figure 1 Schematic diagram of the correlation analysis of the screening results of key tissue characteristic factors in a specific embodiment of the present invention; Figure 2 Schematic diagram of the SHAP values of key tissue characteristic factors affecting the performance of continuous casting billets in a specific embodiment of the present invention; Figure 3 Tissue characteristic nephogram obtained by the grid division difference method in a specific embodiment of the present invention; Figure 4 Schematic diagram of the contribution analysis of a single tissue characteristic to performance in a specific embodiment of the present invention; Figure 5 Schematic diagram of the grid superposition of the contribution of a single tissue characteristic to performance in a specific embodiment of the present invention; Figure 6 Comparison diagram of the theoretically excellent performance area and the actual performance nephogram under the synergistic influence of multiple tissue characteristics in a specific embodiment of the present invention; Figure 7 Flowchart of a specific embodiment of the method for establishing the continuous casting process - tissue - performance relationship of interpretable machine learning of the present invention. Detailed implementation manners

[0023] It should be noted that the following detailed description is exemplary and is intended to provide further illustration of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0024] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they specify the presence of features, steps, operations, and / or combinations thereof.

[0025] As Figure 7 shown, Figure 7 Flowchart of the method for establishing the continuous casting process - tissue - performance relationship of interpretable machine learning of the present invention. The method for establishing the continuous casting process - tissue - performance relationship of interpretable machine learning includes: Step 1, conduct laboratory scaled - down continuous casting experiments to obtain continuous casting billet samples under different process conditions; The casting speed is 0.5 - 2 m / min, and its step interval is 0.3 mm / min. The cooling water flow rate in the secondary cooling zone is 30 - 50 L / (t·min), and the range of molten steel composition is: C accounts for 350 - 550×10 -6 wt%, Si accounts for 45 - 70×10-6 wt%, Mn accounts for 1550 - 2350×10 -6 wt%, P accounts for 130 - 190×10 -6 wt%, T.S. accounts for 70 - 110×10 -6 wt%, Al accounts for 260 - 400×10 -6 wt%, Cr accounts for 140 - 210×10 -6 wt%, Ni accounts for 75 - 110×10 -6 wt%, Cu accounts for 75 - 110×10 -6 wt%, Mo accounts for 30 - 50×10 -6 wt%.

[0026] Step 2: Take small cubes of 10 10 10 mm from the central region of the continuous casting billets under different process conditions, and use the electrolytic polishing method to prepare EBSD (Electron Back Scatter Diffraction) samples and observe their microstructural morphology; The electrolytic solution for electrolytic polishing is a perchloric acid - alcohol mixed solution with a volume ratio of 1:3. The electrolytic polishing voltage is 20 - 21 V, the electrolytic time is 10 - 12 s, and the temperature of electrolytic polishing is 0 - 1°C.

[0027] Step 3: Use Aztec software to extract different microstructural features and measure the Vickers hardness values of the samples under different process conditions.

[0028] The extraction of microstructural features includes 8 characteristic indexes such as the average grain area size, the average grain boundary angle mean value, the proportion of large - angle grain boundaries, the proportion of dynamically recrystallized grain sizes, the average geometrically necessary dislocation density, the proportion of dynamically recovered grains, the aspect ratio, and the proportion of deformed grain sizes.

[0029] Step 4: Use the SHAP analysis method to analyze the correlation between microstructural features and hardness; Step 5: Use the interpolation method to draw the contour maps of single microstructural features of the samples under different process conditions; Step 6: Use the grid method to divide the contour maps of microstructural features and select the four with the highest degree of correlation for contribution superposition to establish the continuous casting process - microstructure - property relationship.

[0030] Use the grid method to divide the region of the contour map and superpose the grid according to the contribution of single microstructural features to the performance. Finally, verify the correctness of the SHAP analysis method and obtain the optimal continuous casting process parameter combination for the corresponding grade of steel under the target performance during continuous casting.

[0031] In a specific embodiment of applying the present invention, the method for establishing the continuous casting process-structure-property relationship of interpretable machine learning of the present invention includes the following steps: Step 1: Conduct laboratory-scale continuous casting experiments to obtain continuous casting billet samples under different process conditions.

[0032] The liquid steel in the tundish during the continuous casting process used comes from laboratory preparation, and its raw materials come from a steel mill in Jiangsu. The elemental composition (unit: 10 -6 ×wt%) is shown in the following table: Table 1 Elemental composition table of tundish liquid steel

[0033] Experiments are carried out with the casting speed ranging from 0.5 to 2 m / min and the step interval of 0.3 mm / min, and the cooling water flow rate in the secondary cooling zone ranging from 30 to 50 L / (t·min) and the step interval of 5 L / (t·min). A total of 30 samples are obtained through the experiments.

[0034] Step 2: Take small cubes with a size of 10 10 10 mm from the central region of the continuous casting billets under different process conditions, and prepare EBSD samples by electrolytic polishing. The electrolytic solution for electrolytic polishing is a perchloric acid-alcohol mixed solution with a volume ratio of 1:3. The electrolytic polishing voltage is 20 - 21 V, the electrolytic time is 10 - 12 s, and the temperature of electrolytic polishing is 0 - 1 °C.

[0035] Step 3: Apply Aztec software to extract eight characteristic indexes such as the average grain area size, average grain boundary angle mean value, proportion of large-angle grain boundaries, proportion of dynamically recrystallized grain sizes, average geometrically necessary dislocation density, proportion of dynamically recovered grains, aspect ratio, and proportion of deformed grain sizes.

[0036] Step 4: Measure the Vickers hardness values of the samples under different process conditions.

[0037] Apply the SHAP analysis method to analyze the correlation relationship between the tissue characteristics and hardness. The correlation analysis of the screening results of the key tissue characteristic factors is shown in the appendix Figure 1 As shown, the accuracy of the experimental set is 95%, and the accuracy of the test set is 85%, with relatively high accuracy.

[0038] The key tissue characteristic factors affecting the performance of the continuous casting billets obtained by applying the SHAP analysis method are shown in the appendix Figure 2 As shown, among the eight tissue characteristic factors analyzed, the first four key tissue characteristic factors with relatively large correlations are screened out. Among them, the proportion of large-angle grain boundaries is positively correlated with the performance of the continuous casting billets, and the three characteristics of average grain size, proportion of dynamically recrystallized grains, and average grain boundary angle mean value are negatively correlated with the performance of the continuous casting billets.

[0039] Step 5: Apply the difference method to draw the cloud diagrams of four key tissue characteristic factors, and use the grid method for division, as shown in the appendix Figure 3 as follows

[0040] Step 6: According to the results of the correlation between single tissue characteristics and performance obtained in Step 4, screen out the grid numbers that are beneficial to performance improvement. Assume that among three tissue characteristics under a certain grid, all are beneficial to performance, then it is determined that the performance of the continuous casting billet is better under the process conditions corresponding to this grid

[0041] The grid numbers of the three tissue characteristics that are all beneficial to performance are boxed with a red frame, and the contribution analysis results are as shown in the appendix Figure 4 as follows

[0042] Sort the grid numbers into a 10×10 grid for comparison with the hardness field distribution. The superimposed contribution grids are as shown in the appendix Figure 5 as follows

[0043] Apply the interpolation method to draw the schematic diagram of the hardness field distribution under all process conditions, and superimpose and compare it with the grid area in the appendix Figure 5 The comparison results are as shown in the appendix Figure 6 as follows

[0044] It can be seen from the comparison that the region with excellent theoretical performance under the synergistic influence of multiple tissue characteristics is almost close to the region of the actual performance cloud diagram, indicating the establishment of the continuous casting process - tissue - performance relationship based on interpretable machine learning

[0045] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention

[0046] Except for the technical features described in the specification, the rest are the known technologies of those skilled in the art

Claims

1. A method for establishing the relationship between continuous casting process, structure and performance based on interpretable machine learning, characterized in that: The method for establishing the relationship between continuous casting process, structure and performance by interpretable machine learning includes: Step 1, conducting laboratory scale-down experiments to produce continuous casting billet samples under different process conditions; Step 2, sampling the center position of the continuous casting billet and preparing an EBSD sample to observe its microstructure; Step 3, extracting different tissue features and measuring the Vickers hardness values ​​of samples under different process conditions; Step 4, analyzing the correlation between tissue characteristics and hardness; Step 5: Use the grid method to divide the organizational characteristic cloud map and select the four items with the highest correlation for contribution superposition to establish the continuous casting process-organization-performance relationship.

2. The method for establishing the relationship between continuous casting process, structure and performance based on interpretable machine learning according to claim 1, characterized in that: In step 1, when producing continuous casting samples under different process conditions, the billet drawing speed is 0.5 to 2 m / min, the step interval is 0.3 mm / min, and the cooling water flow rate in the secondary cooling zone is 30 to 50 L / (t·min).

3. The method for establishing the relationship between continuous casting process, structure and performance based on interpretable machine learning according to claim 2, characterized in that: In step 1, the composition range of the molten steel is: C accounts for 350~550×10 -6 wt%, Si accounts for 45~70×10 -6 wt%, Mn accounts for 1550~2350×10 -6 wt%,P accounts for 130~190×10 -6 wt%,TS accounts for 70~110×10 -6 wt%,Al accounts for 260~400×10 -6 wt%,Cr accounts for 140~210×10 -6 wt%,Ni accounts for 75~110×10 -6 wt%,Cu accounts for 75~110×10 -6 wt%, Mo accounts for 30~50×10 -6 wt%.

4. The method for establishing the relationship between continuous casting process, structure and performance based on interpretable machine learning according to claim 1, characterized in that: In step 2, small squares were taken from the center area of ​​the continuous casting billet under different process conditions, and the EBSD samples were prepared by electrolytic polishing method to observe their microstructure.

5. The method for establishing the relationship between continuous casting process, structure and performance based on interpretable machine learning according to claim 4, characterized in that: In step 2, the electrolytic solution used for electropolishing is a mixed solution of perchloric acid alcohol with a volume ratio of 1:3, the electropolishing voltage is 20-21V, the electrolysis time is 10-12s, and the electropolishing temperature is 0-1°C.

6. The method for establishing the relationship between continuous casting process, structure and performance based on interpretable machine learning according to claim 1, characterized in that: In step 3, the extracted organizational features include eight characteristic indicators: average grain area size, average grain boundary angle, high-angle grain boundary ratio, dynamic recrystallized grain size ratio, average geometric required dislocation density, dynamic recovery grain ratio, aspect ratio, and deformed grain size ratio.

7. The method for establishing the relationship between continuous casting process, structure and performance based on interpretable machine learning according to claim 1, characterized in that: In step 4, the SHAP analysis method is applied to analyze the correlation between tissue characteristics and hardness.

8. The method for establishing the relationship between continuous casting process, structure and performance based on interpretable machine learning according to claim 1, characterized in that: The method for establishing the continuous casting process-structure-performance relationship based on interpretable machine learning also includes, after step 4, using an interpolation method to draw a single structure characteristic cloud map of the sample under different process conditions.

9. The method for establishing the relationship between continuous casting process, structure and performance based on interpretable machine learning according to claim 8, characterized in that: In step 5, the grid method is used to divide the single organizational characteristic cloud map into regions, and the grids are superimposed according to the contribution of the single organizational characteristics to the performance. Finally, the correctness of the SHAP analysis method is verified, and the optimal continuous casting process parameter combination for the corresponding steel grade under the target performance in the continuous casting process is obtained.

Images