JP7761161B2 - METHOD FOR DETERMINING QUALITY AND DESTINATION OF CONTINUOUSLY CAST BLADES, METHOD FOR DETERMINING CONTINUOUS CASTING CONDITIONS, AND METHOD FOR CONTINUOUSLY CASTING STEEL - Google Patents

Description

The present invention relates to a method for determining the quality of a product or continuously cast slab from the casting conditions and actual measurements taken during the continuous casting of steel, a method for determining the destination of a continuously cast slab, and a method for determining continuous casting conditions.In the following description, the unit of mass "t" represents ^10.sup.3 kg, and the unit of volume "L" represents 10.sup. ^{- 3 m.The} letter "N" attached to the unit of gas volume represents the volume of the gas under standard conditions, which are 0.degree. C. and 101,325 Pa.

During continuous steel casting, gas bubbles injected into the nozzle and non-metallic inclusions such as deoxidation products and sulfides can become trapped in the solidified shell and remain on the surface of the product. These bubbles and non-metallic inclusions degrade the quality of steel products, particularly thick steel plates. For example, in line pipes used to transport oil and natural gas, hydrogen-induced cracking can occur, originating from bubbles and non-metallic inclusions due to the action of sour gas. Similar problems also occur in offshore structures, storage tanks, and oil tanks. Moreover, in recent years, steel products are often required to be used in harsh environments, such as at lower temperatures or in more corrosive environments, making it increasingly important to reduce bubbles and non-metallic inclusions in cast slabs.

For this reason, sour-resistant linepipe steel undergoes HIC (Hydrogen Induced Cracking) testing before shipping, and only products that do not experience HIC are shipped as sour-resistant. However, it takes several weeks for the results of HIC testing to become known, and if HIC occurs, the product cannot be shipped as sour-resistant, resulting in a significant decrease in yield. Therefore, if HIC performance could be evaluated at the slab stage before plate rolling without conducting HIC testing, it would be possible to shorten production time and significantly improve yield.

Patent Document 1 discloses a method of measuring the opening thickness of horizontal cracks and the maximum segregation grain size on the cut surface of a slab, determining a threshold value from the measurement results and HIC measurement test results, and changing the destination. Patent Documents 2 and 3 disclose continuous steel casting methods that satisfy the Ca/S ratio and the relationship between Ca, S, and O, and further compensate for HIC by keeping the Ca reduction below a threshold. Furthermore, Patent Document 4 discloses an evaluation method that can detect center segregation with high accuracy by binarizing an etched print image of the cross section of the steel material.

JP 2015-58473 A JP 2016-125137 A JP 2016-125140 A JP 2017-181030 A

However, the above-mentioned conventional techniques have the following problems.
The technology disclosed in Patent Document 1 requires measuring segregated grains on the cut surface of a slab, which poses a problem in terms of shortening the manufacturing period. The technologies disclosed in Patent Documents 2 and 3 can handle HIC cracking originating from Ca-based inclusions, but cannot handle other non-metallic inclusions. Furthermore, the technology disclosed in Patent Document 4 simply evaluates center segregation, and does not clarify the correlation between bubbles and non-metallic inclusions and HIC cracking.

HIC cracking caused by non-metallic inclusions requires testing for seven days even after steel plate production. By the time a quality defect is discovered, a large number of products have already been manufactured, which can result in a large number of defective products being produced. Furthermore, if defects are predicted after casting, it is possible to avoid further processing and simply re-melt the product, but since it is not possible to determine whether there are defects at the casting stage, it is necessary to continue manufacturing the product all the way to the final product. Since evaluation is required after production of the final product, and any defects cannot be converted into good products, this leads to increased costs.

The present invention was made in consideration of the above circumstances, and its first objective is to propose a product quality assessment method that predicts the quality of the product produced by rolling a slab at the slab casting stage. It also aims to propose a method that can assess the quality of a slab cast by a continuous casting machine, particularly the HIC characteristics caused by bubbles and non-metallic inclusions, during or after casting. Additionally, it proposes a method for determining the destination of a continuously cast slab, a method for determining continuous casting conditions, and a method for continuously casting steel.

The inventors discovered that the area ratio of hydrogen-induced cracking (HIC) caused by bubbles and non-metallic inclusions can be predicted from actual casting data during casting, namely, parameters such as the cross-sectional size of the cast piece, chemical composition, casting speed, electromagnetic stirring conditions, lead time from secondary refining to the start of casting, amount of auxiliary raw materials added, flow rate of inert gas blown into the nozzle, and immersion depth of the submerged nozzle, and thus completed the present invention.

That is, it has been found that the above problems can be advantageously solved by the following invention.
[1] A method for assessing the quality of a product obtained by rolling a slab cast by a continuous casting machine, using a prediction model for hydrogen-induced cracking in the product surface layer, and predicting hydrogen-induced cracking in the product surface layer using one or more input variables selected from actual values of casting performance data measured during casting.
[2] A method for determining the quality of a slab cast by a continuous casting machine using the product quality determination method described in [1], wherein the prediction model links actual casting performance data with the area ratio of hydrogen-induced cracking in the surface layer of the product, and one or more selected from actual values of the casting performance data measured during casting are input into the prediction model to predict the area ratio of hydrogen-induced cracking in the surface layer of a product obtained from the slab during or after casting.
[3] In [2], the casting performance data is some or all of the following: cross-sectional size of the cast strand, component composition, casting speed, electromagnetic stirring conditions, lead time from secondary refining to the start of casting, amount of auxiliary raw materials added, flow rate of inert gas blown into the nozzle, and immersion depth of the submerged entry nozzle.
[4] The method for evaluating the quality of a continuously cast slab according to [3], wherein the component composition is at least one selected from a C concentration, a Mn concentration, a S concentration, and a C equivalent calculated in Ceq (mass%) by the following formula:
Ceq=[C]-0.0616[Al]+2.5275[S]-0.2652[P]+0.0023[Si]+0.0344[Mn]-1.525[S][ Mn]+0.021[Si][Mn]+0.02[Cu]-0.02[Mo]+0.06[Ni]+0.02[Cr]-0.04[V]-0.04[Nb]
Here, [M] in the formula is the content of element M expressed in mass percentage.
[5] The method for assessing the quality of a continuously cast slab according to any one of [2] to [4], wherein the prediction model uses principal component analysis and regression by the Random Forest method, and optionally performs machine learning on the prediction model using an actual measurement value of the area ratio of hydrogen-induced cracking in the surface layer of the product.
[6] A method for determining a destination of a continuously cast strand, which determines whether the strand can be used as a sour line pipe steel based on a quality prediction of the strand determined using the method for determining the quality of the continuously cast strand according to any one of [2] to [5].
[7] A method for determining continuous casting conditions, which determines casting conditions by reverse analysis based on the casting performance data and the prediction model, so that a predicted value of the area ratio of hydrogen-induced cracking in the product surface layer asymptotically approaches a predetermined value based on a quality prediction of a slab determined using the method for determining the quality of a continuously cast slab described in any one of [2] to [5].
[8] The method for determining continuous casting conditions according to [7], wherein the predetermined value is 2% or less.
[9] A method for continuous casting of steel, which comprises producing a cast piece according to the casting conditions determined by the method according to [7] or [8].

According to the present invention, actual values from measured casting performance data are input into a pre-prepared prediction model to predict the quality of the product or slab, particularly the area ratio of hydrogen-induced cracking in the product surface layer, during or after casting. Therefore, it is possible to accurately predict whether a slab is suitable for a specific product, enabling the production of products with a high yield. Furthermore, by determining casting conditions so that the obtained predicted value approaches a predetermined value and producing the slab under the determined casting conditions, products can be produced with a high yield, improving productivity and providing industrial utility. It is possible to determine whether a product is suitable for sour-resistant linepipe steel, for example, from its predicted HIC value without conducting time-consuming HIC testing, making it possible to quickly respond to demands for the production of steel products with diverse specifications, providing industrial utility.

1 is a schematic side view showing a slab continuous casting machine suitable for carrying out the present invention. 1 is a graph showing the relationship between the actual measured value and the predicted value of the HIC crack area ratio (CAR) of the surface layer portion. FIG. 1 is a flow chart showing an example of a method for predicting the quality of a continuously cast slab. This is a schematic flow diagram from continuous casting to shipping. 1 is a graph showing the magnitude of the influence of each variable on principal component 1 and principal component 2 in an example.

The following provides a detailed description of embodiments of the present invention. The following embodiments are intended to exemplify equipment and methods for embodying the technical concept of the present invention, and are not intended to limit the configuration to that described below. In other words, the technical concept of the present invention can be modified in various ways within the technical scope described in the claims.

Figure 1 is a schematic side view of a continuous slab casting machine suitable for use in a continuous steel casting method according to one embodiment of the present invention. As shown in Figure 1, the continuous slab casting machine 1 is equipped with a mold 5 into which molten steel 9 is poured, solidified, and used to form the outer shell shape of a slab 10. A tundish 2 is installed at a predetermined position above the mold 5 to relay molten steel 9 supplied from a ladle (not shown) to the mold 5. A sliding nozzle 3 is installed at the bottom of the tundish 2 to adjust the flow rate of the molten steel 9, and an immersion nozzle 4 is installed below the sliding nozzle 3. Meanwhile, multiple pairs of slab support rolls 6, each consisting of a support roll, a guide roll, and a pinch roll, are arranged below the mold 5. A secondary cooling zone is formed in the gaps between adjacent slab support rolls 6 in the casting direction (FD), where spray nozzles such as water spray nozzles or air mist spray nozzles (not shown) are installed. In the secondary cooling zone, the slab 10 is cooled as it is withdrawn by cooling water (also referred to as "secondary cooling water") sprayed from spray nozzles. Downstream of the final slab support roll 6 in the casting direction, a plurality of transport rolls 7 for transporting the cast slab 10 are installed, and above these transport rolls 7, a slab cutter 8 is disposed for cutting a slab 10a of a predetermined length from the cast slab 10. The cross-sectional size of the slab is expressed by the slab width Lw (mm) and the slab thickness Lt (mm).

As shown in Figure 1, a soft reduction zone 14 consisting of multiple pairs of strand support rolls is installed on both the upstream and downstream sides of the final solidification position (crater end: CE) 13 of the slab 10 in the casting direction. In the soft reduction zone 14, the distance between the opposing strand support rolls sandwiching the slab 10 (called the "roll gap") gradually narrows toward the downstream side in the casting direction. In other words, a reduction gradient (a roll gap that gradually narrows toward the downstream side in the casting direction) is set. In the soft reduction zone 14, the slab 10 can be soft reduced in its entirety or in selected areas. Spray nozzles for cooling the slab 10 are also located between each strand support roll in the soft reduction zone 14. The strand support rolls 6 arranged in the soft reduction zone 14 are also called reduction rolls.

An electromagnetic stirring device (not shown) is installed between the mold 5 and the slab support rolls 6 to cause the unsolidified molten steel 12 to flow and clean the inner surface of the solidified shell 11. In addition, an inert gas is blown into the molten steel 9 from the upper nozzle (not shown) and sliding nozzle 3 installed in the tundish 2 to prevent nozzle clogging.

The chemical composition of a slab can be determined using analytical values from samples taken from molten steel in a ladle or tundish. For example, C and Mn are known to be components that affect the toughness of a product. It is also known that the greater the C equivalent Ceq (mass%), expressed by the following formula, the greater the degree of decrease in toughness. The toughness of steel affects its HIC resistance. Furthermore, Mn and S form MnS-based non-metallic inclusions, which affect the HIC resistance of the surface layer.
Ceq=[C]-0.0616[Al]+2.5275[S]-0.2652[P]+0.0023[Si]+0.0344[Mn]-1.525[S][ Mn]+0.021[Si][Mn]+0.02[Cu]-0.02[Mo]+0.06[Ni]+0.02[Cr]-0.04[V]-0.04[Nb]
Here, [M] in the formula is the content of element M expressed in mass %.

The casting speed Vc (m/min) affects the flow rate of molten steel discharged from the submerged entry nozzle 4 and serves as an indicator of the depth to which bubbles and non-metallic inclusions penetrate into the solidification pool. The applied current value I (A) of the electromagnetic stirring system influences the cleaning power of the inner surface of the solidified shell 11, thereby affecting the capture of bubbles and non-metallic inclusions. The lead time from secondary refining to the start of casting affects the floating and separation of deoxidizing inclusions. The amount of auxiliary raw material added, for example, the CaSi consumption unit (kg/t-molten steel) and the FeSi consumption unit (kg/t-molten steel), affects the morphology control of S-based non-metallic inclusions and the floating and separation of inclusions. The inert gas blown into the upper nozzle or sliding nozzle, for example, the Ar gas flow rate QAr (NL/min), serves as an indicator of the bubbles captured in the slab. The immersion depth Ld (mm) of the immersion nozzle 4 affects the floating and separation of bubbles and non-metallic inclusions or the depth of their penetration into the solidification pool, depending on the direction of the molten steel discharge flow.

The above-mentioned casting performance data, which affect the amount of bubbles and non-metallic inclusions captured by the solidified shell, are used as variables to input into the prediction model for the hydrogen-induced crack area ratio (CAR) of the product surface layer. Here, the product surface layer refers to the area from the surface to 0.2 times the plate thickness in the plate thickness direction. For example, the prediction model can reduce the number of variables by performing dimensionality reduction using principal component analysis, and then perform regression using the Random Forest method, making it possible to accurately predict the HIC crack area ratio (CAR) of the surface layer.

Principal component analysis (PCA) is a method for compressing data with correlations between variables without reducing the information, thereby simplifying analysis by reducing the number of variables in complex data. In this embodiment, variables such as "slab width at casting Lw, slab thickness Lt, carbon concentration [C], manganese concentration [Mn], sulfur concentration [S], carbon equivalent Ceq, casting speed Vc, applied current value I for electromagnetic stirring, lead time (time) from secondary refining to the start of casting, amounts of auxiliary materials added (amount of FeSi added, amount of CaSi added), Ar gas flow rate QAr injected into the nozzle, and immersion depth Ld of the submerged entry nozzle" are compressed to, for example, five variables. Compressing to five variables allows the compressed variables to be expressed using variables such as principal component 1 through principal component 5, allowing data expressed with many variables to be represented with fewer variables without reducing the amount of information. Narrowing down the data variables without using PCA requires discarding some variables. This can sometimes require discarding important variables. In principal component analysis, principal components are generated in order from the first principal component so as to include as much information as possible about each variable, and therefore it is possible to reduce the number of variables more efficiently than usual.

The Random Forest method is a machine learning algorithm. It is an ensemble learning algorithm that integrates multiple weak learners using decision trees and performs cross-validation to improve generalization ability. In this embodiment, the regression involves calculating several hundred decision trees and integrating them using the average value. In other words, by compressing the many explanatory variables to around five using principal component analysis and then regressing using the Random Forest method, highly accurate regression is possible even with a small amount of data.

In this embodiment, an example is shown in which the hydrogen-induced crack area rate CAR of the product surface layer is predicted using all of the above-mentioned casting performance data, but even when only part of the casting performance data is used, it is possible to improve the accuracy of the regression by compressing variables using principal component analysis.

Figure 3 is a flow diagram showing an example of a method for predicting the HIC crack area ratio (CAR) in the product surface layer. Casting conditions and online measurements are input into a prediction model (S1), and variables are compressed using principal component analysis (S2). Regression is performed using the compressed variables using the random forest method (S3), and the HIC crack area ratio (CAR) in the product surface layer due to bubbles and non-metallic inclusions is predicted (S4). Furthermore, by using the actual measured values of the HIC crack area ratio (CAR) in the product surface layer as training data for principal component analysis (S5), it is possible to predict the HIC crack area ratio (CAR) with even greater accuracy. The obtained predicted CAR value can be used to determine whether or not to use the product in the next rolling process. Furthermore, the obtained predicted CAR value can be used to improve the quality of the slab by adjusting the immersion depth (Ld) of the submerged entry nozzle so that it approaches a predetermined value during casting, or by adjusting the applied current (I) of the electromagnetic stirring system (S6).

Figure 4 is a flow diagram showing the steps from continuous casting (S11) to rolling (S12) to shipping (S13). Typically, to determine the quality of slabs cast by a continuous casting machine, an EPMA is used to analyze the distribution of bubbles and nonmetallic inclusions (S14). This analysis takes one to two weeks. Furthermore, an HIC test is performed to determine whether the product should be shipped after rolling (S15). The HIC test involves immersing a test specimen in hydrogen sulfide and evaluating the crack area ratio (CAR) when hydrogen-induced cracks occur in the center of the thickness or on the surface of the plate (product). This test requires a minimum of one week. Product shipping requires that the CAR be below a threshold. Conventionally, by the time this test revealed a quality defect, a large quantity of products had already been manufactured, resulting in the production of a large number of defective products. This embodiment allows quality to be predicted during or immediately after casting without HIC testing, significantly shortening lead time and preventing mass nonconformity during this period.

Example 1
The present invention will be described in more detail below with reference to examples.
The continuous casting machine used in the test was the same as continuous casting machine 1 shown in Figure 1. Low-carbon aluminum-killed steel was cast using this continuous casting machine. Tables 1 to 3 show actual casting data, such as casting conditions, and measured and predicted values of the HIC crack area ratio (CAR) in the product surface layer using the continuous casting method according to the above embodiment. Here, the product surface layer refers to the area from the surface to 0.2 times the plate thickness in the plate thickness direction. Using the actual casting data shown in Tables 1 and 2 as input, a prediction model for the HIC crack area ratio (CAR) in the product surface layer was used to perform regression using principal component analysis and the Random Forest method. Figure 2 shows a graph of the relationship between the actual and predicted values of the HIC crack area ratio (CAR) in the product surface layer. In this prediction model, the explanatory variables were reduced to five variables using principal component analysis, and regression was performed using the Random Forest method. Figure 5 shows the relationship between principal component 1 and principal component 2 of the principal component analysis and the correlation coefficients of various operating conditions. The operating conditions with the largest sum of the correlation coefficients between principal component 1 and principal component 2 were then determined to be variables with a large influence on the HIC crack area ratio (CAR) in the product surface layer. For example, from Figure 5, the following variables (operating conditions) were extracted as having a large influence: "lead time" (the lead time from secondary refining to the start of casting), "amount of auxiliary raw materials added" (amount of FeSi added, amount of CaSi added), "immersion depth (Ld) of the submerged entry nozzle,""casting speed (Vc)," and "applied current (I) of electromagnetic stirring." Using this method, the actual measured and predicted values of the HIC crack area ratio (CAR) in the product surface layer showed good agreement, making it possible to predict HIC cracking in the product surface layer during or immediately after casting.
Table 3 shows an example in which the immersion depth Ld of the submerged entry nozzle and the applied current I of the electromagnetic stirring were changed during casting so that the predicted CAR value approached 0.00%. This control significantly reduced the area ratio of HIC cracks in the surface layer of the product.
Furthermore, the threshold value for the HIC crack occurrence area rate CAR in the product surface layer portion differs depending on the required quality. For example, in a steel material for which the target HIC crack occurrence area rate CAR in the product surface layer portion is 2% or less, the prediction model for the HIC crack occurrence area rate CAR in the product surface layer portion according to the above embodiment was used to redirect slabs for which the CAR prediction value was predicted to be greater than 2%, resulting in a 7% improvement in yield.

Example 2
An example of back-analysis of operating conditions will be described using Test No. 21 in Table 2 as an example. Under the initial operating conditions for Test No. 21, the predicted value of the HIC crack occurrence area ratio (CAR) of the product surface layer was 5.46%, and the actual value was 4.60%. As a result of the principal component analysis of Example 1, variables that significantly influence CAR and that can be changed during operation, such as the immersion depth Ld of the submerged entry nozzle and the applied current I of the electromagnetic stirring, were extracted. By changing these variables and using a prediction model, back-analysis was performed to search for operating conditions that would result in a predicted CAR value of 0.2%. The obtained conditions, the immersion depth Ld of the submerged entry nozzle, were then changed from 186 mm to 210 mm, and the applied current I of the electromagnetic stirring was changed from 400 A to 700 A. As a result, the actual measured value of the HIC crack occurrence area ratio (CAR) of the product surface layer achieved the target value of 2% or less.

REFERENCE SIGNS LIST 1 continuous casting machine 2 tundish 3 sliding nozzle 4 submerged nozzle 5 mold 6 strand support roll 7 transport roll 8 strand cutter 9 molten steel 10 strand 10a (cut) strand 11 solidified shell 12 molten steel in unsolidified phase 13 solidification completion position (crater end)
14 Light reduction zone FD Casting direction

Claims (8)

In determining the quality of a slab cast by a continuous casting machine, a method for determining product quality is used in which a prediction model for hydrogen-induced cracking in the surface layer of a product obtained by rolling a slab cast by a continuous casting machine is used, and one or more input variables are selected from actual values of casting performance data measured during casting to predict hydrogen-induced cracking in the surface layer of the product .
The prediction model links actual casting data with the area rate of hydrogen-induced cracking in the surface layer of the product,
A method for assessing the quality of a continuously cast slab, comprising inputting into the prediction model one or more values selected from the actual values of the casting performance data measured during casting, and predicting the area ratio of hydrogen-induced cracking in the surface layer of a product obtained from the slab during or after casting. 2. The method for determining the quality of a continuously cast slab according to claim 1, wherein the casting performance data is some or all of the following : cross-sectional size of the slab, component composition, casting speed, electromagnetic stirring conditions, lead time from secondary refining to the start of casting, amount of auxiliary raw materials added, flow rate of inert gas blown into the nozzle, and immersion depth of the submerged entry nozzle. 3. The method for evaluating the quality of a continuously cast slab according to claim 2 , wherein the component composition is at least one selected from a C concentration, a Mn concentration, a S concentration, and a C equivalent calculated in terms of Ceq (mass%) by the following formula:
Ceq=[C]-0.0616[Al]+2.5275[S]-0.2652[P]+0.0023[Si]+0.0344[Mn]-1.525[S][ Mn]+0.021[Si][Mn]+0.02[Cu]-0.02[Mo]+0.06[Ni]+0.02[Cr]-0.04[V]-0.04[Nb]
Here, [M] in the formula is the content of element M expressed in mass percentage. The prediction model uses principal component analysis and Random Forest regression,
The method for evaluating the quality of a continuously cast slab according to any one of claims 1 to 3 , optionally comprising machine learning the prediction model using an actual measurement value of an area ratio of hydrogen-induced cracking in a surface layer portion of the product. A method for determining a destination of a continuously cast slab, which determines whether the slab can be used as a sour-resistant line pipe steel based on a quality prediction of the slab determined using the method for determining the quality of the continuously cast slab according to any one of claims 1 to 3. A method for determining continuous casting conditions, comprising: determining casting conditions by reverse analysis based on the casting performance data and the prediction model, such that a predicted value of an area ratio of hydrogen-induced cracking in a surface layer of a product asymptotically approaches a predetermined value based on a quality prediction of a slab determined using the method for determining the quality of a continuously cast slab according to any one of claims 1 to 3. The method for determining continuous casting conditions according to claim 6 , wherein the predetermined value is set to 2% or less. A method for continuous casting of steel, comprising producing a cast piece in accordance with the casting conditions determined by the method according to claim 6 .