Method for controlling wall thickness of mother pipe in stretch reducer rolling

Abstract

To provide a thickness control method of a stretch reducer which highly stabilizes thickness accuracy in stretch reducer rolling, and thereby contributes to improvement of a profit.SOLUTION: A thickness control method of a tube stock in stretch reducer rolling includes: a prediction evaluation step of expressing sensibility of master data (MD) and a taper of a rolling condition, and calculating a prediction taper value from a taper value prediction model using a numeric value as an explanatory variable without including a disturbance element, when determining a taper value of stretch reducer rolling; a validity evaluation step of calculating a taper initial value from a model of evaluating validity of the taper value prediction model, further setting up the prediction taper value and calculating steel pipe thickness when the stretch reducer rolling is performed, and evaluating the validity of the steel pipe thickness on the calculation; a taper value correction step of determining a correction value W (including 0) of the prediction taper value, according to the result of the evaluation of the validity; and a final taper value determination step of determining a final taper value set up to the stretch reducer rolling from the result of the taper value correction step.SELECTED DRAWING: Figure 7

Description

[Technical Field]

[0001] The present invention relates to a method for obtaining a hot seamless steel pipe of a predetermined thickness by controlling the wall thickness of a mother pipe carried out in the stretch reducer rolling process, which is one of the processes for producing a hot seamless steel pipe. [Background technology]

[0002] FIG. 1 shows a typical manufacturing process for a hot seamless steel pipe according to the present invention. First, a cylindrical steel slab 1 called a billet is heated to a predetermined temperature in a heating furnace 2, and then pierced and rolled in a piercer 3 to obtain a hollow mother pipe. This mother pipe is then subjected to elongation rolling in a mandrel mill 4 to further reduce the diameter and wall thickness to approach the desired mother pipe dimensions. After this elongation rolling, the temperature of the mother pipe has dropped due to heat dissipation and extraction, so it is heated to a predetermined temperature in a heating furnace 5 as necessary.

[0003] Next, in order to obtain a steel pipe of the final target dimensions, a stretch reducer 6 is used to reduce the diameter by tandem rolling using multiple rolling stands, and the difference in roll rotation speed (taper value) is controlled to reduce the thickness of the raw pipe.

[0004] The taper value of the stretch reducer will be explained in detail below. That is, Figure 2 shows a graph showing the ratio of the rotation speed of each stand to the rotation speed of stand number 4. In Figure 2, the taper value is the difference in the rotation speed ratio of stand number 1, which has the lowest rotation speed, to the rotation speed of stand number 4, and the final stand, which has the highest rotation speed.

[0005] Qualitatively, a large taper value increases the tension and reduces the wall thickness of the steel pipe. On the other hand, a small taper value decreases the tension and increases the wall thickness of the steel pipe. In this way, the wall thickness can be controlled by controlling the taper value using a stretch reducer.

[0006] In conventional rolling methods, the first material in a lot is rolled using a physics-based regression equation, and after checking the deviation from the target thickness, the second and subsequent rolled materials are rolled with a correction amount at the operator's discretion, and this procedure is repeated until the target thickness is achieved.

[0007] However, the dimensions of the mother pipe before stretch reducer rolling vary even within the same lot due to variations during mandrel mill rolling, and the amount of correction varies greatly from mother pipe to mother pipe. Therefore, adjusting the wall thickness has relied heavily on experience and intuition.

[0008] In response to this, for example, Patent Document 1 discloses a technique for controlling wall thickness in such stretch reducer rolling, in which the tension of the mother pipe between stands is adjusted based on a physical formula and a regression formula to correct the deviation between the target wall thickness of the finished pipe after rolling.

[0009] Furthermore, Patent Document 2 discloses a technology for predicting the taper using a neural network and controlling the wall thickness of a stretch reducer. [Prior art documents] [Patent documents]

[0010] [Patent Document 1] Japanese Patent Application Publication No. 4-238608 [Patent Document 2] Japanese Patent Application Publication No. 9-108721 Summary of the Invention [Problem to be solved by the invention]

[0011] However, in the wall thickness control method in a stretch reducer such as that in Patent Document 1, the influence of the rolling temperature of the mother pipe and the characteristics of the stands on the tension of the mother pipe cannot be ignored, and therefore it is desirable to use a correction formula that takes these into consideration. However, in stretch reducer rolling, which is performed using an arrangement of rolls in an arrangement of up to 28 stands, but at least 7 stands, the number of variables required is the same as the number of stands, resulting in a problem that it is difficult to define a multivariate function with high accuracy.

[0012] Furthermore, there is a problem in that the number of constant terms, such as the initial values of various variables, becomes enormous and it becomes extremely difficult to manage them.

[0013] On the other hand, as in the above-mentioned Patent Document 2, a method has been proposed in which a neural network, which is a type of deep learning technology that simulates the neural circuits of the human brain, is used to model the complex rolling form and control the wall thickness of a stretch reducer.

[0014] However, although the thickness control of a stretch reducer involves many items that contain a large amount of variation due to external factors, the learning process of such a neural network is a black box, and if learning is performed only on items that contain a large amount of variation, unexpected items will be detected as features.

[0015] Therefore, in Patent Document 2, the idea of the cumulative number of rolls rolled is adopted in order to take into consideration the amount of uneven wear of the roll and the properties of the surface texture, which may be disturbance factors.

[0016] However, the characteristics of such rolls are also significantly affected by factors such as the accumulated deformation resistance of the rolled material and the rolling time. Therefore, even if the idea of the accumulated number of rolled pieces is adopted, there will be large variations in factors related to wall thickness control, just like with the length of the mother pipe and the temperature. Therefore, in order to perform general-purpose wall thickness control in a stretch reducer, it is necessary to ensure stable accuracy.

[0017] In contrast to a neural network, prediction of a setup value using machine learning has excellent prediction accuracy within the data used for learning, i.e., for rolling conditions that are interpolated and have been previously experienced. On the other hand, there is a problem that the accuracy deteriorates for rolling conditions for which there is no learning data, i.e., for rolling conditions that are extrapolated and have not been previously experienced.

[0018] The present invention has been made in consideration of the above circumstances, and aims to contribute to improving profits by stabilizing the wall thickness accuracy at a high level in the setup of tandem groove rolling in hot seamless steel pipe rolling, i.e., in stretch reducer rolling. In addition, the present invention aims to reduce the safety risks associated with operational troubles caused by dimensional deviations from target values and non-routine work such as dealing with such operational troubles by stabilizing the dimensional accuracy at a high level. [Means for solving the problem]

[0019] The present invention controls the wall thickness of a mother pipe in a stretch reducer rolling process, which is one of the processes for manufacturing hot seamless steel pipes and in which rolling is carried out using multiple stands, and in which many influencing factors are intricately intertwined. To control the thickness, a means is used to control the taper value, which is the difference in the rotation speed ratio between the stand with the lowest rotation speed and the stand with the highest rotation speed relative to the rotation speed of a reference stand.

[0020] The inventors focused on using machine learning to accurately formulate the relationships between a huge number of variables in order to determine such taper values. Furthermore, the inventors focused on the fact that stretch reducer rolling involves taper sensitivity, which is a constant that expresses rolling characteristics, in addition to variables that are practically impossible to avoid due to variations in measuring instruments, such as the outer diameter and wall thickness before rolling, and master data (MD), which are the basic conditions for rolling, such as the target wall thickness and length. Then, the inventors conducted various studies, regarding such taper sensitivity as a taper characteristic value (also referred to as the A value or constant A in the present invention).

[0021] As a result, we found that by adding this A value to a trained machine learning model that includes it as an explanatory variable, the prediction accuracy of the taper value, which is a setup item during stretch reducer rolling, can be significantly stabilized. The A value is a number such that X=A when it is considered that a change of 1 in the taper value, which is an explanatory variable, results in a change of X% in the thickness, which is a response variable.

[0022] Furthermore, even if the taper value is set according to this A value, since A value is merely a theoretical value, the accuracy of taper prediction will not improve due to variations in operating conditions and measuring equipment. Therefore, the inventors attempted to add this A value as an explanatory variable in machine learning.

[0023] That is, rather than predicting the taper value under the target rolling conditions using master data (MD) of rolling conditions such as various materials, diameter reduction amounts, and wall thickness reduction amounts, we first grouped the rolling conditions under which the theoretical A value was similar. We then considered that within each group, taper predictions could be made using a machine learning model that had been trained using the above MD, which would be effective in improving prediction accuracy.

[0024] However, when it comes to predicting taper under rolling conditions where no group with similar A values exists, that is, under rolling conditions that require extrapolation, it has been difficult to obtain excellent taper prediction accuracy using methods that use machine learning, due to the principles of machine learning.

[0025] Therefore, the inventors noticed that there is a linear relationship between the taper value in stretch-reducer rolling and the change in wall thickness (%) before and after stretch-reducer rolling, as shown in Figure 3. They attempted to learn the x-axis intercept of this line (also referred to as the initial taper value B value in the present invention) and apply this to the material to be rolled. That is, the B value is predicted using a machine learning model, and then the wall thickness after stretch-reducer rolling is predicted using the predicted taper value (also referred to as the predicted taper value in the present invention) and the wall thickness ratio (= actual wall thickness / target wall thickness) (%) from the equation of the line shown in Figure 3. It was found that this method produced highly accurate predictions, with some exceptions.

[0026] The inventors have conducted extensive research to further improve the prediction accuracy and have found that if this predicted calculated value is within a certain control range, the wall thickness after stretch reducer rolling will fall within the predicted range. In other words, if this predicted calculated value is outside the specified control range, the inventors have found that the wall thickness after stretch reducer rolling will fall within the predicted range by correcting it using taper correction using the A value or manual intervention by an operator as necessary.

[0027] That is, it has been found that the above procedure is essential for the process of manufacturing hot seamless steel pipes with excellent wall thickness accuracy in a stretch reducer.

[0028] The present invention has been completed based on the above findings. That is, the gist of the present invention is as follows. 1. In stretch reducer rolling, a plurality of rolling stands using grooved rolls are arranged in the rolling direction, and a difference in the rotation speed between each roll from the first stand to the last stand of the rolling stands is made to apply tension to a mother pipe passing through the plurality of rolling stands to perform wall reduction rolling and obtain a predetermined steel pipe wall thickness. In determining the difference in rotation speed, a prediction evaluation step is performed in which a predicted taper value is calculated using a taper value prediction model that is created in advance as a model for predicting an optimal taper value from master data (MD) of rolling conditions and numerical values that represent the sensitivity of taper and do not include external disturbance elements and are used as explanatory variables, and the validity of the taper value prediction model is evaluated. a validity evaluation step of calculating an initial taper value, which is the taper value when the wall thickness ratio of the mother pipe does not change before and after stretch reducer rolling, from a model (validity evaluation model) that calculates the predicted taper value and calculates the wall thickness of the steel pipe when stretch reducer rolling is performed, thereby evaluating the validity of the calculated wall thickness of the steel pipe; a taper value correction step of determining a correction value W (including 0) for the predicted taper value based on the result of the validity evaluation; and a final taper value determination step of determining a final taper value to be set up for stretch reducer rolling based on the result of the taper value correction step.

[0029] 2. The sensitivity of the taper is calculated by multiplying the number of rotations N of each stand set to reduce the wall thickness of the mother tube by 1% relative to the reference number of rotations N0 of the rolling rolls in a reference stand selected in advance from the plurality of rolling stands. i (i=1, , end) is the ratio (N i / N0), the rotation speed ratio of the final stand (N end 2. The method for controlling the wall thickness of a mother pipe in stretch reducer rolling according to 1 above, wherein the constant A is calculated by the following formula (1) using the rotation speed ratio of the first stand (N1 / N0) and the rotation speed ratio of the first stand (N1 / N0): A(%)={1 / ((N end / N0)-(N1 / N0))}×100 (1)

[0030] 3. A method for controlling the wall thickness of a mother pipe in stretch reducer rolling according to 2 above, wherein, when evaluating the validity of the steel pipe wall thickness from the validity evaluation model, the constant A, the predicted taper value Tp, the initial taper value B, and the actual wall thickness ratio S (%) to the target wall thickness before stretch reducer rolling are used to calculate the ratio T% of the predicted wall thickness to the target wall thickness after stretch reducer rolling by the following equation (2), and it is confirmed before stretch reducer rolling whether T satisfies the following equation (3). T(%)=S(%)+A(%)×(Tp(%)-B(%)) ···(2) -5.0% ≦ T(%) ≦ +5.0% ···(3)

[0031] 4. The method for controlling the wall thickness of a mother pipe in stretch-reducer rolling according to 3 above, wherein the taper value correction step is a step of setting the correction value W to 0 when the predicted wall thickness ratio T satisfies the formula (3), setting the correction value W to be greater than 0 when the predicted wall thickness ratio T does not satisfy the formula (3), adding or subtracting the correction value W to the predicted taper value Tp to obtain Tp', and setting T to T' that satisfies the formula (3).

[0032] 5. The method for controlling the wall thickness of a mother pipe in stretch reducer rolling according to 4 above, wherein the final taper value determination step is a step of determining the predicted taper value Tp as Ts, which is the final taper value to be set up, when the correction value W is 0, and determining Tp' as the final taper value Ts' to be set up when the correction value W is greater than 0. [Effects of the Invention]

[0033] According to the present invention, the taper value, which is the most important setting value for rolling in a stretch reducer, can be predicted with high accuracy. Furthermore, according to the present invention, it is possible to stabilize the wall thickness accuracy in stretch reducer rolling at a high level even under rolling conditions that are extrapolated. Furthermore, according to the present invention, it is possible to reduce the amount of scrapped products due to out-of-size dimensions and the amount of absorption of non-steady parts given during manufacturing, thereby contributing to improved profits. In addition, the improved dimensional accuracy can reduce safety risks associated with non-routine work, such as operational problems caused by dimensions deviating from the target values and the need to deal with such operational problems. [Brief explanation of the drawings]

[0034] [Figure 1] FIG. 1 is a diagram showing an example of a manufacturing process for a hot seamless steel pipe. [Figure 2] FIG. 10 is a diagram illustrating a taper. [Figure 3] FIG. 10 is a diagram illustrating the relationship between a taper characteristic value and an initial taper value. [Figure 4] FIG. 1 is a diagram showing a flow of implementation when creating a model according to the present invention. [Figure 5] FIG. 2 is a diagram showing a flow of the implementation before stretch reducer rolling of the present invention. [Figure 6] FIG. 10 is a diagram showing a change in taper prediction accuracy due to the addition of a taper characteristic value. [Figure 7] 1 is a diagram showing a configuration of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0035] The present invention provides a method for effectively controlling the wall thickness of a mother pipe and the length of a steel pipe in a stretch reducer rolling process, which is one step in the process of producing a hot seamless steel pipe, in which rolling is carried out using multiple stands and many influencing factors are intricately intertwined. That is, the present invention is a method for determining the value (taper value) of the difference in rotational speed ratio between rolls in a stretch reducer rolling process in which a plurality of grooved rolls are arranged in the rolling direction, and a difference in rotational speed ratio (taper) is provided between these rolls to apply tension to a mother pipe and reduce its wall thickness, thereby obtaining a target steel pipe wall thickness, as shown in FIG. 1 .

[0036] As shown in Figure 4, this method uses explanatory variables that represent master data (MD) (to be described later) and the sensitivity of the taper value, such as the material of the mother pipe, the final target dimensions of the steel pipe, and the actual dimensions of the mother pipe before stretch reducer rolling, and that do not include any disturbance factors.These explanatory variables are then used to create in advance a model that predicts the taper value (taper value prediction model) and a model that evaluates its validity (validity evaluation model). The process of creating such a taper value prediction model and validity evaluation model may be referred to as a model creation step.

[0037] The method further includes a prediction evaluation step of calculating a predicted taper value Ts from the taper value prediction model, as shown in FIG.

[0038] As shown in Fig. 5, the method also includes a validity evaluation step of calculating an initial taper value B, which is the taper value when the wall thickness ratio of the mother pipe before and after stretch-reducer rolling remains unchanged, from the validity evaluation model, and then setting up the predicted taper value to calculate the wall thickness of the steel pipe on the assumption that the stretch-reducer has been performed, and evaluating the validity of the calculated wall thickness of the steel pipe.

[0039] Furthermore, as shown in FIG. 5, the method includes a taper value correction step of determining a correction value W (including 0) for the predicted taper value depending on the result of the validity evaluation, and a final taper value determination step of determining a final taper value to be set up in stretch reducer rolling based on the result of the taper value correction step.

[0040] Next, the equipment configuration used in the present invention will be described using the flow chart shown in FIG. First, a cylindrical steel billet 1, which is the raw material for a hot rolled seamless steel pipe, is heated to a predetermined temperature in a heating furnace 2, and then a hollow blank is produced using hot piercing equipment utilizing the Mannesmann effect, known as a piercer 3. Next, a solid rolling tool known as a mandrel mill bar is inserted into the hollow blank.

[0041] In this state, the blank tube passes through a rolling mill called a mandrel mill 4, which has multiple grooved rolls arranged at 90° angles, and is elongated by the gap between the rolls and the bar. Thereafter, if necessary, the cooled rolled material is heated to a specified temperature in a heating furnace 5. Finally, tension is applied in a stretch reducer 6, which is a rolling mill that is the subject of the present invention, to further stretch the raw pipe to obtain dimensions equivalent to those of the final product (steel pipe).

[0042] In rolling with such a stretch reducer, tandem rolling is carried out using 7 to a maximum of 28 rolling stands, each of which is made up of two or three grooved rolls.

[0043] The outer diameter is determined by the caliber diameter of the grooved roll in the final stand, but as for the thickness, the rotation speed is allocated to each stand so that the ratio of the rotation speed to the reference stand (number 4 in this case) increases as the stand moves to the later stage (right side of the graph) shown in Figure 2, in other words, in order to apply tension and reduce the thickness, the rotation speed is allocated to each stand so that it forms a gentle S-shaped curve as shown in Figure 2. The standard stand can be determined based on the total number of stands and the tension balance in each stand. The gentle S-shaped curve is determined by allocating a taper to each stand based on the cross-sectional area of the final stand and the cross-sectional area ratio of each stand.

[0044] The difference in the rotational speed ratio between the rolls at the first and last stands is called the taper, and the target thickness and length can be achieved by controlling the value of this taper (taper value).

[0045] The explanatory variables used in the present invention will be explained below. In this invention, a total of 56 items (when 28 rolling stands are used) are used as explanatory variables, and a list of these is shown in Table 1. Among these, X1 to X2, such as material, product outer diameter, product wall thickness, product length, actual mandrel mill outer diameter, target mandrel mill wall thickness, actual mandrel mill length, number of stretch reducer stands, roll shape, and rotation speed of each stand, are used. 27 and X when 28 rolling stands are used 29 ~X 56 The 55 items (in the present invention, these are referred to as master data (MD) which are conditions that are the basis of rolling, or simply as master data or MD) are not particularly new and are also used in conventional techniques.

[0046] In the present invention, the taper characteristic value A (X in Table 1) that indicates the sensitivity of the taper is added to the master data. 28 ) improves the taper prediction accuracy of machine learning model X, as shown in Figure 6.

[0047] [Table 1]

[0048] The present invention will be further explained with reference to FIG. In the present invention, first, items (master data) that serve as explanatory variables for the machine learning model, such as the billet 1, which is the material managed by the information management data server 7, and measurement results such as the outer diameter, thickness, length, and mass of the steel pipe after rolling by various rolling mills, as well as the specifications of the target final product, are extracted (see Table 1 above). In addition, in the present invention, it is important to create a machine learning model X8 that predicts the taper and a machine learning model Y12 that predicts the taper initial value B using 56 types of explanatory variables, including the master data (MD) and a constant A that represents the taper sensitivity described below.

[0049] In the present invention, the number of rotations N of the rolling rolls in each stand, which is set to reduce the wall thickness of the mother pipe by 1%, is set to N0, which is the number of rotations N of the rolling rolls in each stand, which is set to reduce the wall thickness of the mother pipe by 1%. i (i=1, 2, , end) is expressed as (N i / N0), it takes (N i The rotation speed ratio of the final stand (N end Using the rotation speed ratio (N1 / N0) of the leading stand (N1 / N0) and the rotation speed ratio (N1 / N0) of the leading stand (N1 / N0) that is the smallest value, the sensitivity of the taper is defined as the taper characteristic value A calculated by the following formula (1). Note that in the explanation of Figure 2, N0 = N4. A(%)={1 / ((N end / N0)-(N1 / N0))}×100 (1)

[0050] In the present invention, when creating the above-mentioned machine learning models X and Y, the difference between the predicted steel pipe length ratio calculated using the output of the machine learning models X and Y and the actual steel pipe length ratio can be considered to be the accuracy of the model, so as long as the 3σ of the standard deviation of that value satisfies ±3%, the type of learning model does not matter. Specifically, it is desirable to use gradient boosting, but logistic regression, a binary classification regression model, a decision tree, or a random forest may also be used.

[0051] The pre-trained machine learning models X and Y created in this way are loaded onto the calculation server 9. Machine learning model X constitutes the prediction calculation part, and machine learning model Y constitutes the validity evaluation part, and they automatically output the taper to be set up before stretch reducer rolling and control the wall thickness of the stretch reducer.

[0052] In this manner, when rolling begins, 55 items that serve as explanatory variables in the prediction calculation section are collected and stored in the calculation server. Next, the calculation server calculates the A value using the formula (1), and adds the A value to the collected explanatory variables to create a data set 10 consisting of a total of 56 explanatory variables. After that, this data set 10 is loaded into machine learning model X, and the taper value is predicted (taper value 11) before stretch reducer rolling.

[0053] <Validity verification> A new data set 10' is created by adding the taper value 11 derived from the prediction calculation unit to the data set, and the validity evaluation unit reads this into a machine learning model Y installed in the calculation server to calculate the B value. After that, the predicted post-rolling predicted wall thickness ratio T when rolling with a stretch reducer is determined using the predicted calculated taper value by the calculation process of equation (2) described later, and its validity is evaluated.

[0054] That is, when evaluating the validity of the steel pipe wall thickness using the validity evaluation model, the A value, the predicted taper value Tp, the initial taper value B, and the actual wall thickness ratio S (%) to the target wall thickness before stretch reducer rolling are used and substituted into the following equation (2) to calculate the predicted wall thickness ratio T (%) to the target wall thickness after stretch reducer rolling, and it is confirmed before stretch reducer rolling whether or not the T satisfies the following equation (3). T(%)=S(%)+A(%)×(Tp(%)-B(%)) ···(2) -5.0% ≦ T(%) ≦ +5.0% ···(3)

[0055] <Taper handling after validation> After the validity is evaluated, the taper value is corrected. When T calculated by the formula (2) satisfies the formula (3), the correction value W is set to 0, and the predicted taper value Tp is determined as the taper value Ts to be set up. On the other hand, when T calculated by the formula (2) does not satisfy the formula (3), the correction value W is set to greater than 0, and the predicted taper value Tp is adjusted by adding or subtracting the correction value W to set the predicted taper value Tp'. Next, T' is calculated by substituting this Tp' for Tp in the formula (2), and when T in the formula (3) is set to T', T' that satisfies the formula (3) is calculated. Then, this taper value Tp' is set as the taper value Ts' to be set up.

[0056] That is, the predicted taper value Tp, calculated by the validity evaluation unit, for which the wall thickness ratio T after stretch-reducer rolling satisfies the control range formula (3), is set up in the stretch-reducer as is (correction value W is 0). On the other hand, taper values for which the wall thickness ratio T does not satisfy the control range are extrapolated data with no prior rolling experience. Therefore, the correction unit performs correction 13 using a constant A that can derive a taper value for a 1% change in wall thickness, or manual correction 14 by an operator (in both cases, the correction value W is set to be greater than 0, and this correction value W is added or subtracted), to determine the taper value Tp' that results in the wall thickness ratio T' that satisfies the formula (3), and sets this as the final taper value Ts'. This Ts' is then set up in the stretch-reducer.

[0057] <Control range of formula (3)> Here, the control range of T (%) is preferably ±3.0%, but since the steel pipe wall thickness tolerance is ±8.0%, it may be ±5.0% taking into account the variation in the dimension measuring device. Furthermore, if the aim is to reduce the dimensional variation of the product and improve the yield, it is even more preferable to set it to ±1.0%. -5.0% ≦ T(%) ≦ +5.0% ···(3)

[0058] <Learning data> Rolling results are accumulated in an information management data server each time, and after a certain period of time, the machine learning model can be recreated as a learning dataset with new data added, reducing the amount of extrapolated data and improving the performance of the machine learning model. The calculation of machine learning models X and Y can be performed on an information management data server. Also, there is no need to separate the functions into two types, an information management data server and a calculation server; both information management and machine learning model calculations can be performed on a single server.

[0059] In addition, in the method for manufacturing the steel pipe described in the present invention, for any items not described in this specification, conventional methods (for example, the method described in JP 2021-171806 A) can be used. [Example]

[0060] In this example, first, 55 items managed by the information management data server 7, such as material, product (steel pipe) outer diameter, product (steel pipe) wall thickness, product (steel pipe) length, actual mandrel mill outer diameter, mandrel mill target wall thickness, actual mandrel mill length, number of stretch reducer stands, roll shape, and rotation speed of each stand, were extracted for approximately 315,000 pieces, and the taper characteristic value A was calculated using the above formula (1) for each rolling result, creating a total of 56 explanatory variables. These explanatory variables are listed in Table 2.

[0061] [Table 2]

[0062] Next, we created a machine learning model using these explanatory variables to predict taper using gradient boosting, and a machine learning model using random forest to predict the initial taper value B.

[0063] These trained models were loaded onto a calculation server, and approximately 86,000 tubes (lot number: 6,000) with rolling conditions that were extrapolated were rolled using a stretch reducer. The length variation of the hot seamless steel pipes after rolling was divided into three stages of product outer diameter: large, medium, and small, and the results were compared with those obtained using feedback control based on physical equations. Table 3 shows the results. Note that steel pipes with a large product outer diameter refer to steel pipes rolled using a billet diameter of 210 mm. Steel pipes with a medium product outer diameter refer to steel pipes rolled using a billet diameter of 190 mm, and steel pipes with a small product outer diameter refer to steel pipes rolled using a billet diameter of 110 mm.

[0064] [Table 3]

[0065] By following the present invention, even if the rolling conditions are extrapolated conditions, the length variation is generally reduced, and the rate of deviation from the control range of ±5.0% in length is reduced to a maximum of 0.42%.

[0066] Furthermore, Table 4 shows the results of this example, narrowed down to 6000 pieces at the top of the lot, which have particularly large variations in length. As shown in Table 4, the rate of deviation from the control range of ±5.0% in length was significantly reduced. This can be attributed to the effect of adding rolling characteristics as an explanatory variable, improving prediction accuracy, and adopting a machine learning model with a low calculation load, which made feedforward control possible.

[0067] [Table 4] [Explanation of symbols]

[0068] 1 billet 2 Furnace 3 Piercer 4. Mandrel Mill 5 Heating furnace 6 Stretch Reducer 7. Data server for information management 8. Machine Learning Model X 9 Computational Server Dataset using 10 A values Dataset using 10´ Tp values 11 Predictive Taper 12 Machine Learning Model Y 13 Correction using A value 14 Manual Correction

Claims

1. In stretch reducer rolling, a plurality of rolling stands using grooved rolls are arranged in the rolling direction, and a difference in the rotation speed of each roll is provided between the first stand and the last stand of the rolling stands, thereby applying tension to a mother pipe passing through the plurality of rolling stands to perform wall reduction rolling and obtain a predetermined steel pipe wall thickness, When determining the difference in rotation speed, a prediction evaluation step of calculating a predicted taper value using a taper value prediction model that is created in advance as a model for predicting an optimal taper value from master data (MD) of rolling conditions and numerical values that represent the sensitivity of the taper and do not include disturbance factors, and that are used as explanatory variables; a validity evaluation step of calculating an initial taper value, which is a taper value when the wall thickness ratio of the mother pipe does not change before and after stretch reducer rolling, from a model (validity evaluation model) for evaluating the validity of the taper value prediction model, and further calculating the wall thickness of the steel pipe when stretch reducer rolling is performed using the predicted taper value, and evaluating the validity of the calculated wall thickness of the steel pipe; a taper value correction step of determining a correction value W (including 0) of the predicted taper value according to the result of the validity evaluation; a final taper value determination step for determining a final taper value to be set up in stretch reducer rolling from the result of the taper value correction step; A method for controlling the wall thickness of a mother pipe in stretch reducer rolling having the above structure.

2. The sensitivity of the taper is A reference rotation speed N of the rolling rolls in a reference stand selected in advance from the plurality of rolling stands ０ The number of rotations N of each stand set to reduce the wall thickness of the blank tube by 1% ｉ (i = 1, ..., end) is the ratio (N ｉ / N ０ ), the rotation speed ratio of the final stand (N end / N ０ ) and the rotation speed ratio of the leading stand (N １ / N ０ 2. The method for controlling the wall thickness of a mother pipe in stretch reducer rolling according to claim 1, wherein the constant A is calculated by the following formula (1): A(%)={1 / ((N end / N ０ )-(N １ / N ０ ))}×100・・・(1)

3. When evaluating the validity of the steel pipe wall thickness using the validity evaluation model, Using the constant A, the predicted taper value Tp, the initial taper value B, and the actual thickness ratio S (%) to the target thickness before stretch reducer rolling, The ratio T% of the predicted thickness to the target thickness after stretch reducer rolling is calculated using the following formula (2):

3. The method for controlling the wall thickness of a mother pipe in stretch reducer rolling according to claim 2, further comprising confirming, before stretch reducer rolling, whether T satisfies the following formula (3): T (%) = S (%) + A (%) × (Tp (%) - B (%)) ... (2) -5.0% ≦ T (%) ≦ +5.0% ... (3)

4. The taper value correction step 4. The method for controlling the wall thickness of a mother pipe in stretch reducer rolling according to claim 3, further comprising the steps of: setting a correction value W to 0 when the predicted wall thickness ratio T satisfies the formula (3); and setting the correction value W to be greater than 0 when the predicted wall thickness ratio T does not satisfy the formula (3); adding or subtracting the correction value W from the predicted taper value Tp to obtain Tp'; and setting T to T' that satisfies the formula (3).

5. The final taper value determining step 5. The method for controlling the wall thickness of a mother pipe in stretch reducer rolling according to claim 4, further comprising the step of determining, when the correction value W is 0, the predicted taper value Tp as Ts, which is the final taper value to be set up, and, when the correction value W is greater than 0, determining Tp' as the final taper value Ts' to be set up.

Images