Physical quantity distribution estimation method of an object, manufacturing method of an object, method for setting manufacturing conditions, method for developing a manufacturing process, method for generating a model, physical quantity distribution estimation program of an object, and physical quantity distribution estimation device of an object

Abstract

The method for estimating the physical quantity distribution of an object is a method for estimating the distribution of physical quantities of an object when a coolant is sprayed from a spray nozzle to cool the object, and includes a heat dissipation amount acquisition step of acquiring information about the amount of heat dissipated from the object by the coolant by inputting multiple flow velocities in a normal direction and multiple flow velocities in a tangential direction of the coolant to a machine learning model, and a physical quantity distribution calculation step of calculating the physical quantity distribution of at least one of the surface and interior of the object using the heat dissipation amount as a boundary condition.

Description

[Technical field]

[0001] The present invention relates to a method for estimating a physical quantity distribution of an object, a method for manufacturing an object, a method for setting manufacturing conditions, a method for developing a manufacturing process, a method for generating a machine learning model, a program for estimating a physical quantity distribution of an object, and an apparatus for estimating a physical quantity distribution of an object. [Background technology]

[0002] In the cooling process of metal materials, the cooling liquid is often sprayed using a spray nozzle. However, this method has the characteristic of directly spraying the cooling liquid, making it difficult to uniformly cool the object to be cooled. Since uneven cooling increases the risk of defects such as cracks in the metal material, it is necessary to accurately predict the temperature distribution (distribution of heat extraction) and optimize the cooling method.

[0003] For example, Non-Patent Document 1 discloses a technique for calculating temperature distribution from a relational equation obtained in a cooling experiment using the distribution of water flow density, which represents the flow rate per unit area directly below the spray. Non-Patent Document 2 discloses a technique for analyzing the coolant flow by the MPS method, which is a type of particle method, and calculating the temperature distribution using the results. Patent Document 1 also discloses a technique for analyzing the coolant flow on the surface of a metal material by fluid analysis, and calculating the temperature distribution by substituting the obtained flow-related parameters into an estimation equation for the heat transfer coefficient. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2013-134110 A [Non-patent literature]

[0005] [Non-Patent Document 1] "Cooling Technology in the Steel Manufacturing Process", The Iron and Steel Institute of Japan, Report of the Cooling Technology Research Subcommittee, 1988 [Non-Patent Document 2] Hakuko Yamazaki et al., "Analysis of Spray Water Behavior in Secondary Cooling of Continuous Casting Using Particle Method," Iron and Steel Institute of Japan, Tetsu-to-Hagane, Vol. 99, 2013 Summary of the Invention [Problem to be solved by the invention]

[0006] The temperature distribution calculated by the method of Non-Patent Document 1 has good accuracy directly below the spray, but it cannot take into account the amount of heat loss caused by the flow of coolant on the surface of the metal material, such as interference between multiple sprays, dripping water, and pooling water. In particular, when the flow rate per spray or the number of sprays is large, the amount of coolant also increases, and the amount of heat loss associated with that flow also increases, resulting in a deviation from the actual phenomenon.

[0007] The method of Non-Patent Document 2 can also evaluate the amount of heat dissipation associated with the flow of the coolant, and can therefore obtain a temperature distribution closer to the actual phenomenon than Non-Patent Document 1. However, the method of Non-Patent Document 2 uses only the water flow density distribution obtained by analysis to calculate the temperature distribution from an estimation formula similar to that of Non-Patent Document 1. In reality, there are factors other than the water flow density distribution that affect the amount of heat dissipation, so there is still room for improvement in the accuracy of prediction of the amount of heat dissipation.

[0008] The method of Patent Document 1, unlike Non-Patent Document 2, uses flow parameters other than the water volume density, making it possible to make predictions that take into account the boiling mode. However, since the flow parameters used in Patent Document 1 are highly non-linear, it is difficult to formulate an accurate estimation equation that corresponds to various situations, and the implementation cost is high.

[0009] The present invention has been made in consideration of the above, and has an object to provide a method for estimating the physical quantity distribution of an object, which is capable of estimating with high accuracy the temperature distribution when an object is cooled using a spray nozzle, a method for manufacturing an object, a method for setting manufacturing conditions, a method for developing a manufacturing process, a method for generating a machine learning model, a program for estimating the physical quantity distribution of an object, and a device for estimating the physical quantity distribution of an object. [Means for solving the problem]

[0010] (1) A method for estimating a physical quantity distribution of an object according to the present invention is a method for estimating a distribution of a physical quantity of an object when the object is cooled by spraying a coolant from a spray nozzle, the method comprising the steps of: a heat dissipation amount acquisition step of acquiring information regarding the amount of heat dissipated by the coolant from the object by inputting a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of the coolant to the object into a machine learning model; a physical quantity distribution calculation step of calculating a physical quantity distribution of at least one of a surface and an interior of the object using the amount of heat dissipation as a boundary condition; It includes.

[0011] (2) A method for estimating a physical quantity distribution of an object according to the present invention is the method for estimating a physical quantity distribution of an object according to the above (1), comprising: The heat dissipation amount acquisition step causes the machine learning model to output a heat transfer coefficient as information regarding the heat dissipation amount, and calculates the heat dissipation amount from the heat transfer coefficient.

[0012] (3) A method for estimating a physical quantity distribution of an object according to the present invention is the method for estimating a physical quantity distribution of an object according to the above (1) or (2), further comprising: The machine learning model has been subjected to machine learning using training data so that when multiple flow velocities of the coolant in the normal direction and multiple flow velocities in the tangential direction relative to the object are input as input values, information regarding the amount of heat removed from the object by the coolant is output as an output value.

[0013] (4) A method for estimating a physical quantity distribution of an object according to the present invention is the method for estimating a physical quantity distribution of an object according to any one of (1) to (3) above, further comprising: The physical quantity is temperature or solidification.

[0014] (5) A method for manufacturing an object according to the present invention uses the method for estimating physical quantity distribution of an object described in any one of (1) to (4) above to estimate a physical quantity distribution of the object in a specified process, and controls manufacturing parameters for manufacturing the object based on the estimated physical quantity distribution of the object.

[0015] (6) A manufacturing condition setting method according to the present invention is a method for setting manufacturing conditions for one or more predetermined steps included in a manufacturing method for manufacturing an object, the method comprising the steps of: The physical quantity distribution of the object in the specified process is estimated by using the method for estimating physical quantity distribution of the object according to any one of (1) to (4) above, and manufacturing conditions in the specified process are set based on the estimated physical quantity distribution of the object.

[0016] (7) A manufacturing process development method according to the present invention is a method for developing a manufacturing process for one or more predetermined steps included in a manufacturing method for manufacturing an object, the method comprising the steps of: The physical quantity distribution of the object in the specified process is estimated by using the method for estimating physical quantity distribution of the object according to any one of (1) to (4) above, and a manufacturing process in the specified process is developed based on the estimated physical quantity distribution of the object.

[0017] (8) A method for generating a machine learning model according to the present invention uses training data in which a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of a coolant for cooling an object are input values, and information on the amount of heat removed by the coolant from the object is output values, A model is generated using machine learning, with input values ​​being multiple flow velocities of the coolant in the normal direction and multiple flow velocities in the tangential direction relative to the object, and output values ​​being information regarding the amount of heat removed from the object by the coolant.

[0018] (9) A physical quantity distribution estimation program for an object according to the present invention is a program for estimating a distribution of physical quantities of an object when the object is cooled by spraying a coolant from a spray nozzle, the program comprising: Computer, a heat transfer amount acquisition means for acquiring information regarding the amount of heat transfer from the object by the coolant by inputting a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of the coolant to a machine learning model; a physical quantity distribution calculation means for calculating a physical quantity distribution of at least one of a surface and an interior of the object using the amount of heat dissipation as a boundary condition; It is intended to function as a

[0019] (10) A physical quantity distribution estimation device of an object according to the present invention is a device for estimating a distribution of physical quantities of an object when the object is cooled by spraying a coolant from a spray nozzle, the device comprising: a heat transfer amount acquisition unit that includes a machine learning model and acquires information about an amount of heat transfer from the object by the coolant by inputting a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of the coolant to the object into the machine learning model; a physical quantity distribution calculation unit that calculates a physical quantity distribution of at least one of a surface and an interior of the object using the amount of heat dissipation as a boundary condition; It is equipped with the following.

[0020] (11) The present invention provides a physical quantity distribution estimation device for an object according to the above (10), comprising: The physical quantity is temperature or solidification.

[0021] (12) A method for setting manufacturing conditions according to the present invention is the method for setting manufacturing conditions described in (6) above, further comprising the steps of: the object is a steel material, the physical quantity is temperature, The predetermined step is secondary cooling during continuous casting or rapid cooling during hot rolling.

[0022] (13) A method for developing a manufacturing process according to the present invention is the method for developing a manufacturing process according to the above (7), further comprising the steps of: the object is a steel material, the physical quantity is temperature, The predetermined step is secondary cooling during continuous casting or rapid cooling during hot rolling.

[0023] (14) A recording medium according to the present invention is a recording medium having a program for estimating a physical quantity distribution of an object recorded thereon, Computer, a heat transfer amount acquisition means for acquiring information regarding the amount of heat transfer from the object by the coolant by inputting a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of the coolant to a machine learning model; a physical quantity distribution calculation means for calculating a physical quantity distribution of at least one of a surface and an interior of the object using the amount of heat dissipation as a boundary condition; This is a recording of a program that functions as a

[0024] (15) A recording medium according to the present invention is a recording medium having a machine learning model generation program recorded thereon, Computer, Using training data in which a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of a cooling liquid for cooling an object are input values ​​and information on the amount of heat removed by the cooling liquid from the object is output values, a means for generating a model by machine learning, the model having a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of the cooling liquid as input values ​​and information on an amount of heat removed from the object by the cooling liquid as an output value; This is a recording of a program that functions as a Effect of the Invention

[0025] According to the method for estimating a physical quantity distribution of an object, the method for manufacturing an object, the method for setting manufacturing conditions, the method for developing a manufacturing process, the method for generating a machine learning model, the program for estimating a physical quantity distribution of an object, and the device for estimating a physical quantity distribution of an object, which relate to the present invention, it is possible to estimate the temperature distribution during cooling of an object using a spray nozzle with high accuracy. [Brief description of the drawings]

[0026] [Figure 1] FIG. 1 is a schematic diagram showing a configuration for realizing an apparatus for estimating temperature distribution on a metallic material and an apparatus for generating a machine learning model according to an embodiment. [Diagram 2] FIG. 2 is a schematic diagram showing a spray nozzle used in the cooling process of a metal material. [Diagram 3] Figure 3 is a graph comparing the prediction results of the heat transfer coefficient when the normal flow velocity of the coolant is used as training data for the machine learning model and when it is not used. [Figure 4] FIG. 4 is a flowchart showing the flow of a machine learning model generation method according to the embodiment. [Diagram 5] FIG. 5 is a flowchart showing the flow of the teacher data creation step in the machine learning model generation method according to the embodiment. [Figure 6] FIG. 6 is a flowchart showing the flow of the method for estimating temperature distribution on a metallic material according to the embodiment. [Figure 7] FIG. 7 is a flowchart showing a first example of the flow of the heat release amount acquiring step in the method for estimating temperature distribution of a metallic material according to the embodiment. [Figure 8] FIG. 8 is a flowchart showing a flow of a second example of the heat release amount acquiring step in the method for estimating temperature distribution of a metallic material according to the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0027] A method for estimating a physical quantity distribution of an object, a method for manufacturing an object, a method for setting manufacturing conditions, a method for developing a manufacturing process, a method for generating a machine learning model, a program for estimating a physical quantity distribution of an object, and an apparatus for estimating a physical quantity distribution of an object according to embodiments of the present invention will be described with reference to the drawings.

[0028] In addition, the components in the following embodiments include those that are replaceable and easy for a person skilled in the art, or those that are substantially the same. In the following description, the description of the same or overlapping parts will be omitted or simplified as appropriate. In each drawing referred to below, the same reference numerals are used for the same or overlapping parts.

[0029] [First embodiment] (Temperature distribution estimation device and model generation device) The metallic material temperature distribution estimation device according to the embodiment estimates the temperature distribution on the surface and / or inside of the metallic material during the cooling process of the metallic material. Examples of the "metallic material cooling process" include a process of secondary cooling of a slab in continuous casting of steel material, and a process of rapidly cooling a steel plate during hot rolling of the steel plate. Furthermore, the machine learning model generation device according to the embodiment (hereinafter, "model generation device") generates a machine learning model (trained model) used when estimating the above temperature distribution.

[0030] The temperature distribution estimation device and model generation device for a metallic material according to the embodiment can be realized by, for example, a temperature distribution estimation device 1 and a model generation device 2 as shown in Fig. 1. This temperature distribution estimation device 1 includes an input unit 11, a calculation unit 12, and an output unit 13.

[0031] 1 illustrates the temperature distribution estimation device 1 and the model generating device 2 as separate devices, it is also possible to assign the function of the model generating device 2 to a calculation unit 12 of the temperature distribution estimation device 1 and provide a "model generating unit" within the calculation unit 12. Below, the model generating device 2 will be described first, and then each component of the temperature distribution estimation device 1 will be described.

[0032] The model generation device 2 generates a machine learning model used to estimate the temperature distribution. In generating this machine learning model, parameters that indicate the flow state of the coolant on the surface of the metal material when the coolant is sprayed from a spray nozzle are prepared in advance, for example, as shown in Fig. 2.

[0033] The "flow state of the coolant" includes, for example, the flow rate of the coolant on the surface of the metal material. In addition to the flow rate of the coolant, the flow state of the coolant on the surface of the metal material may also include at least one of the density, pressure, temperature, water flow density, and measured temperature of the metal material, or physical quantities calculated using these.

[0034] The term "coolant flow rate" specifically refers to multiple flow rates in the normal direction to the surface of the metal material and multiple flow rates in the tangential direction to the surface of the metal material. Furthermore, the term "multiple flow rates" specifically refers to flow rates at multiple coordinates (positions) on the surface of the metal material.

[0035] Here, the flow velocity of the coolant is difficult to measure experimentally because it is in a nonlinear flow field formed by the coolant sprayed from the spray nozzle colliding with and interfering with the surface of the metal material or with coolant sprayed from other spray nozzles. Therefore, it is desirable to perform a fluid analysis of the coolant in advance to obtain the flow velocity of each coordinate on the surface of the steel plate from the flow field of the coolant. The method of this fluid analysis is not particularly limited, but it is desirable to use a method that can analyze the flow of the coolant including the gas-liquid interface, such as the VOF (Volume Of Fluid) method, the particle method, the phase field method, etc.

[0036] The model generation device 2 uses previously prepared training data to generate a model through machine learning, in which the flow state of the coolant on the surface of the metal material is used as an input value, and information regarding the amount of heat removed from the metal material by the coolant is used as an output value.

[0037] "Teacher data" refers to a data set in which input values ​​are multiple flow velocities of the coolant in the normal direction to the surface of the metal material and multiple flow velocities in the tangential direction to the surface of the metal material, and output values ​​are information regarding the amount of heat removed from the metal material by the coolant.

[0038] Furthermore, the "information on the amount of heat dissipation" includes, for example, a heat transfer coefficient and an amount of heat dissipation. That is, the model generation device 2 may apply machine learning to the model so that, by inputting the flow state of the coolant on the surface of the metal material, the model generation device 2 outputs the heat transfer coefficient on the surface of the metal material. Furthermore, the model generation device 2 may apply machine learning to the model so that, by inputting the flow state of the coolant on the surface of the metal material, the model generation device 2 outputs the amount of heat dissipation itself on the surface of the metal material. Details of the process (model generation step) by the model generation device 2 and details of the creation of the teacher data will be described later (see FIG. 4 and FIG. 5).

[0039] Here, in the conventional technology, the heat transfer coefficient was estimated by substituting the flow rate (water volume density) of the coolant analyzed by the particle method or the like into the estimation formula for the heat transfer coefficient. In other words, in the conventional technology, the focus was on accurately analyzing the flow rate of the coolant, assuming the use of the estimation formula for the heat transfer coefficient, and no attention was paid to using two types of flow velocities of the coolant (normal direction, tangential direction) as parameters.

[0040] On the other hand, in this embodiment, two types of flow velocities of the coolant that have not been focused on in the past are used as input values, and a machine learning model is generated using training data in which information on the corresponding heat transfer amount is used as output values. This makes it possible to accurately calculate the heat transfer coefficient without using a conventional estimation formula for the heat transfer coefficient.

[0041] Figure 3 shows a graph comparing the prediction results of the heat transfer coefficient when the normal flow velocity of the coolant is used as training data for the machine learning model and when it is not used, with all parameters other than the flow velocity remaining the same.

[0042] As shown in Figure 3, the root mean square error (RMSE) of predicting the heat transfer coefficient using the normal flow velocity of the coolant is 316.12486 W / m 2 K". On the other hand, the RMSE for predicting the heat transfer coefficient without using the normal flow velocity of the coolant is 477.49837 W / m 2K". In this way, it can be seen that the prediction accuracy of the heat transfer coefficient is greatly improved by using the flow velocity in the normal direction of the coolant as training data.

[0043] This is presumably because the normal flow velocity is also likely to be large in the area directly below the spray, where the heat transfer coefficient is large. Therefore, by adding the normal flow velocity of the coolant to the training data, it is possible to improve the accuracy of the cooling prediction directly below the spray, which has a large effect on the overall heat transfer amount.

[0044] The input unit 11 is an input means for the calculation unit 12, and is realized by an input device such as a keyboard, a mouse pointer, a numeric keypad, etc. The input unit 11 inputs information used for various processes in the calculation unit 12.

[0045] The calculation unit 12 is realized by a processor such as a CPU (Central Processing Unit) and a memory (main storage unit) such as a RAM (Random Access Memory) or a ROM (Read Only Memory).

[0046] The calculation unit 12 loads a program into a working area of ​​the main storage unit, executes the program, and realizes a function that meets a predetermined purpose by controlling each component through the execution of the program. The calculation unit 12 functions as a heat dissipation amount acquisition unit 121 and a physical quantity distribution calculation unit 122 through the execution of the program.

[0047] The heat dissipation amount acquisition unit 121 acquires information on the amount of heat dissipated from the metal material by the coolant (heat transfer coefficient, amount of heat dissipation) by inputting the flow state of the coolant on the surface of the metal material to the machine learning model. The heat dissipation amount acquisition unit 121 may, for example, cause the machine learning model to output a heat transfer coefficient as information on the amount of heat dissipation, and calculate the amount of heat dissipation from the heat transfer coefficient. The heat dissipation amount acquisition unit 121 may also, for example, cause the machine learning model to output the amount of heat dissipation itself as information on the amount of heat dissipation. Details of the process (heat dissipation amount acquisition step) by the heat dissipation amount acquisition unit 121 will be described later (see Figs. 6 to 8).

[0048] The physical quantity distribution calculation unit 122 calculates the temperature distribution on the surface and / or inside of the metal material using as a boundary condition the amount of dissipated heat acquired by the dissipated heat amount acquisition unit 121. Details of the process (temperature distribution calculation step) by the physical quantity distribution calculation unit 122 will be described later (see FIG. 6).

[0049] The output unit 13 is realized by an output device such as a display, etc. The output unit 13 outputs the calculation result by the calculation unit 12.

[0050] (Model generation method) A process of a machine learning model generation method (hereinafter, referred to as a "model generation method") according to an embodiment will be described with reference to Fig. 4 and Fig. 5. The model generation method includes a teacher data creation step and a model generation step.

[0051] <Steps for creating training data> In the teacher data creation step, a data set for teacher data is prepared (step S1). The teacher data creation step will be described in detail with reference to FIG.

[0052] First, the flow state of the coolant on the surface of the metal material, i.e., multiple flow velocities in the normal direction and multiple flow velocities in the tangential direction to the surface of the metal material, are acquired (step S11). Next, the heat transfer coefficient is calculated from the temperature actually measured during cooling of the metal material (step S12). Next, the amount of heat removed is calculated from the heat transfer coefficient (step S13). Next, a data set is created in which the two types of flow velocities are input values ​​and the amount of heat removed is output value (step S14).

[0053] Here, the flow state of the coolant, which is the explanatory variable, is obtained for each of a plurality of coordinates on the surface of the metal material. The flow state of the coolant is obtained in advance, for example, by using the fluid analysis method described above. The heat transfer coefficient, which is the objective variable, is obtained for each of the same coordinates on the surface of the metal material as those obtained by the explanatory variables. The heat transfer coefficient can be calculated from the temperature of the metal material actually measured during cooling. When measuring the actual temperature, it is preferable to use materials that are as similar as possible in composition, structure, viscosity, physical properties, etc. to the metal material and coolant to be analyzed, and it is most preferable to use the same materials.

[0054] As described above, the flow state of the coolant as the explanatory variable and the flow state of the coolant as the explanatory variable are obtained for the same coordinates on the surface of the metal material. Therefore, the number of data input to the machine learning model and the number of data output from the machine learning model are basically the same, but the number of input and output data may be different. For example, by inputting the flow state of the coolant for multiple coordinates on the surface of the metal material to the machine learning model, the heat transfer coefficient for one area on the surface of the metal material, which is an integration of these multiple coordinates, may be output from the machine learning model. In this way, it is also possible to appropriately change the number of input and output data in the machine learning model depending on the reduction of the calculation load and the required accuracy. Returning to FIG. 4, the explanation will be continued.

[0055] <Model generation step> In the model generation step, machine learning is performed using the teaching data so that when the flow state of the coolant on the surface of the metal material is input as an input value, information regarding the amount of heat removed from the metal material by the coolant (heat transfer coefficient, amount of heat removed) is output as an output value. This generates a machine learning model (step S2). By performing the above-mentioned processing, a machine learning model can be constructed by combining the temperature measurement results of the metal material that has actually been cooled or is being cooled with a fluid analysis that reproduces the flow state of the coolant during this cooling process.

[0056] Here, the machine learning method in the model generation step is not particularly limited, but it is preferable that the method be one that can take into account the nonlinearity of the flow state and boiling form of the coolant. Examples of such methods include neural networks and gradient boosting decision trees. On the other hand, linear prediction methods include simple regression, multiple regression, and simple relational expressions, but these are not used in the model generation step.

[0057] In addition, in the present embodiment, a case where a machine learning model is used is described, but a model using a database (hereinafter, referred to as a "database model") may also be used. In this case, a database model is constructed in the model generation step after sufficient data is prepared in advance so that the flow state and boiling form of the coolant can be considered as nonlinear.

[0058] (Temperature distribution estimation method) The process of the method for estimating temperature distribution of a metallic material according to the embodiment will be described with reference to Figures 6 to 8. The method for estimating temperature distribution includes a heat transfer amount acquisition step and a temperature distribution calculation step.

[0059] <Heat extraction amount acquisition step> In the heat dissipation amount acquisition step, a machine learning model created in advance is used to acquire the amount of heat dissipated when the metal material is cooled by the coolant (step S21). In the heat dissipation amount acquisition step, the amount of heat dissipated can be acquired by a number of methods depending on how the machine learning model is trained. Details of the heat dissipation amount acquisition step will be described below with reference to Figs. 7 and 8.

[0060] In the method shown in FIG. 7, first, the flow state of the coolant on the surface of the metal material (e.g., multiple flow velocities in the normal direction and multiple flow velocities in the tangential direction) is input to the machine learning model (step S31). This causes the machine learning model to output a heat transfer coefficient (step S32). Then, from this heat transfer coefficient, the amount of heat removed when the metal material is cooled by the coolant is calculated (step S33). In this way, in this method, a machine learning model that has been subjected to machine learning using training data is used so that when the flow state of the coolant on the surface of the metal material is input as an input value, the heat transfer coefficient of the metal material by the coolant is output as an output value.

[0061] In the method shown in FIG. 8, first, the flow state of the coolant on the surface of the metal material (e.g., multiple flow velocities in the normal direction and multiple flow velocities in the tangential direction) is input to the machine learning model (step S41). This causes the machine learning model to output the amount of heat removed from the metal material (step S42). In this way, in this method, a machine learning model that has been subjected to machine learning using training data is used so that when the flow state of the coolant on the surface of the metal material is input as an input value, the amount of heat removed from the metal material by the coolant is output as an output value. Returning to FIG. 6, the explanation will be continued.

[0062] <Temperature distribution calculation step> In the temperature distribution calculation step, the temperature distribution on the surface and / or inside of the metal material is calculated using the amount of heat transfer as a boundary condition (step S22). In the temperature distribution calculation step, for example, the surface and / or inside of the metal material for which the temperature distribution is to be estimated is discretized by a lattice, and the temperature distribution is calculated by solving the heat conduction equation (i.e., performing heat transfer analysis) at each calculation point. In this case, instead of directly solving the heat conduction equation, a relational equation created using the result of solving the heat conduction equation, a machine learning model, etc. may be used.

[0063] In addition, there are no particular limitations on the method of calculating the temperature distribution, such as the explicit method or the implicit method, or the steady-state calculation or the unsteady calculation. Since the amount of heat dissipation is calculated at each coordinate on the surface of the metal material, the amount of heat dissipation at any surface coordinate can be obtained by linear interpolation. By performing the above-mentioned processing, the temperature distribution on the surface and / or inside the metal material can be estimated.

[0064] According to the method for estimating the temperature distribution of a metallic material of the embodiment described above, it is possible to estimate the temperature distribution of a metallic material when it is cooled using a spray nozzle with high accuracy. This is because the amount of heat dissipation used in the temperature distribution calculation step can be accurately calculated by obtaining information on the amount of heat dissipation corresponding to two types of flow velocities (normal direction and tangential direction) of the coolant, which have not been focused on in the past, as parameters. Furthermore, according to the method for generating a machine learning model of the embodiment, it is possible to construct a machine learning model for estimating the temperature distribution of a metallic material when it is cooled using a spray nozzle with high accuracy.

[0065] (Example 1: How to set continuous casting conditions) The method for estimating the temperature distribution of a metallic material according to the embodiment can be used for setting the continuous casting conditions. In this case, the method for estimating the temperature distribution of a metallic material according to the embodiment is used to estimate the temperature distribution of a slab in secondary cooling during continuous casting of a steel material, and the continuous casting conditions are set based on the estimated temperature distribution of the slab. A specific example of the method for setting the continuous casting conditions according to the embodiment will be described below.

[0066] First, a machine learning model is created in advance that outputs the amount of heat removed from metal materials by the coolant as an output value when the flow state of the coolant obtained by fluid analysis is entered as an input value. Here, the heat transfer coefficient, which is the objective variable, is collected from a cooling experiment on a steel plate, and the flow state of the coolant, which is the explanatory variable, is collected by a flow analysis of the coolant that reproduces this.

[0067] The flow of the steel plate cooling experiment is explained below. First, a steel plate with multiple embedded thermocouples is heated to 950°C in a furnace. Next, the heated steel plate is removed from the furnace and moved to the front of the cooling spray nozzles. At this time, a simulation roll is placed on the steel plate to approximate the actual cooling conditions. Once the movement of the steel plate is complete and it is fixed in place, cooling liquid is sprayed from the spray nozzles to cool the steel plate. During cooling, the temperature of the steel plate is obtained from the thermocouples every 0.1 seconds. This experiment is performed for each arrangement of multiple spray nozzles and for each flow rate.

[0068] This makes it possible to obtain the temperature history at each coordinate on the steel plate during cooling for each spray nozzle arrangement and flow rate. Next, this temperature history of the steel plate is converted into a heat transfer coefficient history. This conversion to a heat transfer coefficient history was performed using a one-dimensional unsteady heat transfer analysis in the thickness direction of the steel plate. The steel plate temperatures obtained by the analysis and measurement are then compared every 0.1 seconds, and the heat transfer coefficient due to cooling by the spray nozzle is determined so that the difference is sufficiently small.

[0069] The flow analysis of the coolant was carried out using the SPH method, a type of particle method. The steel plate used in the cooling experiment was used as the wall boundary condition, and the coolant discretized with SPH particles was sprayed from the spray nozzle position. Unsteady analysis was then carried out until the coolant flow on the steel plate became steady, and the average value of the flow state over a 10-second period was obtained. The SPH method analysis makes it possible to obtain the density, flow velocity, pressure, number of nearby particles, etc. of each particle. These were ensemble averaged for multiple particles near the temperature measurement position in the cooling experiment, and used as parameters of the flow state.

[0070] A learning dataset (teacher data) was created using the heat transfer coefficient obtained in the cooling experiment as the objective variable and the flow state parameters obtained from the coolant flow analysis plus the steel plate temperature obtained in the cooling experiment as the explanatory variable. 10% of the learning dataset was used as validation data, and the remaining 90% was used as learning data. However, if the dataset was simply divided using random numbers, data from the same measurement points would exist in both the validation data and the learning data, leading to an overestimate of the prediction accuracy. Therefore, in order to be able to handle unknown data as well, the data was grouped by temperature measurement point, and data from the same group was not divided.

[0071] The gradient boosting decision tree was used as the machine learning method. Parameter tuning was also performed using Bayesian optimization. Next, the flow state of the coolant on the surface of the metal material obtained by fluid analysis was input into the constructed machine learning model to obtain the corresponding heat transfer coefficient, which was then converted to calculate the amount of heat dissipation. Note that the amount of heat dissipation may be calculated directly without using the heat transfer coefficient by setting the objective variable of the machine learning model to the amount of heat dissipation. The above steps complete the pre-creation of the machine learning model.

[0072] Next, the flow state of the coolant on the surface of the target slab is analyzed. The flow state parameters output as a result here are input to a machine learning model created in advance, so they are the same as those used in the learning. Therefore, in this case as well, the flow state of the coolant on the surface of the slab is analyzed using the SPH method, just like when creating the learning data.

[0073] The analysis flow using the SPH method was the same as when creating the learning data, and the coolant discretized with SPH particles was sprayed from the spray nozzle position of the actual machine, and a non-steady analysis was performed until the coolant flow on the slab became steady, and the average value of the flow state was obtained for 10 seconds. In addition, multiple coordinates for obtaining the coolant flow state were set on the surface of the slab, so that the numerical value of any coordinate could be obtained by linear interpolation.

[0074] Next, the amount of heat removed from the slab by the coolant is calculated. The flow state of each coordinate of the slab obtained in the previous step and arbitrary temperature data are input into a machine learning model created in advance to obtain the heat transfer coefficient of each coordinate of the slab. The temperature data is divided into equal intervals within the range observed in actual operation. Here, the range of 100°C to 1500°C is input in 5°C increments. Then, by associating the obtained heat transfer coefficient data with the coordinates and temperature of the input data, a table (heat transfer coefficient prediction table) is created that can predict the heat transfer coefficient by linear interpolation from the x, y coordinates and temperature of the slab surface.

[0075] Finally, the temperature distribution on the surface and / or inside of the slab is estimated by heat transfer analysis using a heat transfer coefficient prediction table. The heat transfer analysis method is an implicit steady-state analysis, and the pouring speed is modeled as advection. The boundary conditions of the cooled surface are the heat transfer coefficient obtained from the heat transfer coefficient prediction table, and the non-cooled surface is a natural radiation condition based on the Stefan-Boltzmann law. By performing this analysis until convergence, the temperature distribution on the surface and / or inside of the slab can be estimated. The temperature distribution obtained by the above is a highly accurate value close to the actual phenomenon, because it uses the two types of flow velocities of the coolant (normal direction, tangential direction) as parameters to precisely represent the amount of heat removed by the nonlinear fluid behavior of the coolant on the surface of the slab.

[0076] (Application example 2: Development method for continuous casting process) The method for estimating temperature distribution of a metallic material according to the embodiment can be used in a method for developing a continuous casting process. In this case, the method for estimating temperature distribution of a metallic material is used to estimate the temperature distribution of a slab during secondary cooling during continuous casting of a steel material, and the continuous casting process is developed based on the estimated temperature distribution of the slab.

[0077] In the method for developing a continuous casting process, the conditions of various equipment when developing a new continuous casting machine can be determined based on the temperature distribution of the slab estimated by, for example, the method for estimating the temperature distribution of the metallic material. The conditions of various equipment include, for example, the position of the spray nozzle, the necessity of spray width cutting, the flow rate and velocity of the cooling liquid, the casting speed (transport speed) of the slab, the position of the roll chocks, etc.

[0078] (Application example 3: Continuous casting process control method and casting manufacturing method) The method for estimating temperature distribution of a metallic material according to the embodiment can be used in a method for controlling a continuous casting process. In this case, the method for estimating temperature distribution of a metallic material is used to estimate the temperature distribution of a slab in secondary cooling during continuous casting of a steel material, and feedback control of control parameters in continuous casting is performed based on the estimated temperature distribution of the slab to produce the slab.

[0079] In the control method of the continuous casting process, the flow rate of the cooling liquid and other operating conditions can be monitored and controlled in real time or with a short time lag when producing a slab. For example, in the case of secondary cooling during continuous casting of a slab, the flow rate of the cooling liquid by the spray nozzle, the presence or absence of width cutting, the casting speed, etc. are control parameters in order to always maintain a desired distribution of the temperature of the slab.

[0080] Therefore, feedback control is performed with these parameters based on the temperature distribution of the slab estimated by the method for estimating the temperature distribution of the metal material. However, since the flow state of the coolant on the surface of the slab cannot be obtained in real time or with a short time lag, it is preferable to obtain in advance the flow rate of each coolant and the flow state under each operating condition by numerical analysis or the like. Then, input to the machine learning model is performed by referring to the flow state of the coolant calculated in advance based on the flow rate of the coolant in the actual machine and other operating conditions. By performing such processing, it is possible to estimate the uneven temperature distribution in the width direction in the continuous casting of the slab, and therefore it is possible to suppress the occurrence of defects such as short side cracks by changing the flow rate of the coolant and the position of the spray based on the estimation result.

[0081] (Example 4: How to set hot rolling conditions) The method for estimating the temperature distribution of a metallic material according to the embodiment can be used to set hot rolling conditions. In this case, the method for estimating the temperature distribution of a metallic material is used to estimate the temperature distribution of a steel sheet during rapid cooling during hot rolling of the steel sheet, and the hot rolling conditions are set based on the estimated temperature distribution of the steel sheet.

[0082] (Application example 5: Hot rolling process development method) The method for estimating temperature distribution of a metallic material according to the embodiment can be used in a method for developing a hot rolling process. In this case, the method for estimating temperature distribution of a metallic material is used to estimate the temperature distribution of a steel plate during rapid cooling during hot rolling of the steel plate, and the hot rolling process is developed based on the estimated temperature distribution of the steel plate.

[0083] (Application Example 6: Method for estimating temperature distribution of slag) Slag, a by-product of steelmaking, is sometimes reused as roadbed material, etc. However, the slag immediately after being discharged is in a high temperature state of over 1,000°C, and is cooled with water sprays, etc. to prevent spontaneous combustion, etc. The temperature distribution of the slag pile during this cooling process affects the quality of the slag after cooling, so estimation using trained models and numerical analysis, etc., is very useful for understanding this temperature distribution.

[0084] The method for estimating the temperature distribution of a metallic material according to the embodiment can be used as a method for estimating the temperature distribution when cooling a slag pile. A slag pile is different from general metallic materials in that the surface shape is not flat (it has projections and recesses), making it difficult to calculate two types of flow velocities of the coolant (normal direction, tangential direction). Therefore, a normal vector is calculated from the curvature of each curved surface of the slag pile surface, and the flow velocity of the coolant is decomposed into the normal direction and tangential direction of the slag pile surface. By using these as parameters of the flow state, the temperature distribution of the slag can be estimated.

[0085] [Second embodiment] (Coagulation distribution estimation method) The process of the method for estimating solidification distribution of a metallic material according to the embodiment will be described. In the above-mentioned temperature distribution estimation method, the distribution of temperature is estimated as a physical quantity, but in the solidification distribution estimation method, the distribution of solidification is estimated as a physical quantity. In the solidification distribution estimation method, a heat transfer amount acquisition step and a solidification distribution calculation step are performed.

[0086] <Heat extraction amount acquisition step> In the heat dissipation amount acquisition step, the heat dissipation amount acquisition unit 121 acquires the amount of heat dissipated when the metal material is cooled by the coolant, using a machine learning model created in advance. In the heat dissipation amount acquisition step, a heat transfer coefficient may be output from the machine learning model, as in Fig. 7, and the amount of heat dissipation may be calculated from the heat transfer coefficient. In addition, in the heat dissipation amount acquisition step, the amount of heat dissipation may be directly output from the machine learning model, as in Fig. 8.

[0087] <Coagulation distribution calculation step> In the solidification distribution calculation step, the physical quantity distribution calculation unit 122 calculates the solidification distribution (three-dimensional solidification distribution) on the surface and / or inside of the metallic material with the heat transfer amount as the boundary condition. In the solidification distribution calculation step, for example, the surface and / or inside of the metallic material for which the solidification distribution is to be estimated is discretized by a lattice, and the solidification distribution is calculated by solving the heat conduction equation (performing heat transfer analysis) at each calculation point. In this case, instead of directly solving the heat conduction equation, a relational equation created using the result of solving the heat conduction equation, a machine-learned model, or the like may be used. In addition, in the heat transfer analysis, the latent heat when the metal changes from solid to liquid is taken into consideration, and the solidification distribution is calculated by flagging the calculation points determined to be solidified.

[0088] In the heat transfer analysis of the solidification distribution calculation step, for example, the following physical quantities are input. Boundary condition: Heat transfer distribution on metal surface (2D) Initial condition: Initial temperature distribution inside the metal (3D) Initial condition: Initial solidification distribution inside the metal (3D) -Physical properties of metals (e.g. thermal conductivity, density, specific heat, latent heat of solidification, etc.) ·Analysis time T

[0089] In the heat transfer analysis in the solidification distribution calculation step, for example, the following physical quantities are output: Temperature distribution inside the metal after T (3D) Solidification distribution inside the metal after T (3D)

[0090] In the heat transfer analysis of the solidification distribution calculation step, the solidification distribution inside the metallic material is calculated, for example, in the following manner. (1) Set the above boundary conditions, initial conditions, physical properties, and analysis time. (2) The heat conduction equation is calculated at each calculation point of the metal material, and the temperature distribution and solidification distribution inside the metal are sequentially updated (loop processing is performed until T has elapsed). (3) When the calculation is complete, the temperature distribution inside the metal and the solidification distribution inside the metal are output.

[0091] In the solidification distribution calculation step, the solidification distribution is calculated together with the temperature distribution of the metallic material. In the solidification distribution calculation step, the value indicating the solidification at each calculation point is output as a value of, for example, 0 to 1, and the solidification distribution is output as a map or the like showing each value by color.

[0092] According to the method for estimating solidification distribution of a metallic material according to the embodiment described above, the solidification distribution of the metallic material during cooling can be estimated with high accuracy. Furthermore, the method for estimating solidification distribution of a metallic material according to the embodiment can be effectively used for estimating the solidification distribution during the production of a slab in particular.

[0093] The object physical quantity distribution estimation method, object manufacturing method, manufacturing condition setting method, manufacturing process development method, machine learning model generation method, object physical quantity distribution estimation program, and object physical quantity distribution estimation device according to the present invention have been specifically described above using the form and examples for carrying out the invention, but the gist of the present invention is not limited to these descriptions and must be broadly interpreted based on the claims. Furthermore, it goes without saying that various changes, modifications, etc. based on these descriptions are also included in the gist of the present invention. [Explanation of symbols]

[0094] 1 Temperature distribution estimation device 11 Input section 12 Arithmetic section 121 Heat extraction unit 122 Physical quantity distribution calculation section 13 Output section 2. Model generation device

Claims

1. A method for estimating a distribution of a physical quantity of an object when the object is cooled by spraying a cooling liquid from a spray nozzle, comprising the steps of: a heat dissipation amount acquisition step of acquiring information regarding the amount of heat dissipated by the coolant from the object by inputting a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of the coolant to the object into a machine learning model; a physical quantity distribution calculation step of calculating a physical quantity distribution of at least one of a surface and an interior of the object using the amount of heat dissipation as a boundary condition; A method for estimating the distribution of physical quantities of an object, comprising:

2. The method for estimating a physical quantity distribution of an object according to claim 1 , wherein the heat dissipation amount acquisition step causes the machine learning model to output a heat transfer coefficient as information regarding the heat dissipation amount, and calculates the heat dissipation amount from the heat transfer coefficient.

3. 2. The method for estimating a physical quantity distribution of an object according to claim 1, wherein the machine learning model is subjected to machine learning using training data so that, when multiple flow velocities in a normal direction and multiple flow velocities in a tangential direction of the coolant relative to the object are input as input values, information regarding the amount of heat removed from the object by the coolant is output as an output value.

4. The method for estimating a physical quantity distribution of an object according to claim 1 , wherein the physical quantity is a temperature or a solidification.

5. 5. A method for manufacturing an object, comprising: estimating a physical quantity distribution of the object in a predetermined process by using the method for estimating a physical quantity distribution of the object according to claim 1; and controlling manufacturing parameters related to the manufacture of the object based on the estimated physical quantity distribution of the object.

6. A method for setting manufacturing conditions for one or more predetermined steps included in a manufacturing method for manufacturing an object, comprising the steps of:

5. A manufacturing condition setting method comprising: estimating a physical quantity distribution of the object in the predetermined process by using the method for estimating physical quantity distribution of the object according to claim 1; and setting manufacturing conditions in the predetermined process based on the estimated physical quantity distribution of the object.

7. 1. A method for developing a manufacturing process for one or more predetermined steps in a manufacturing method for producing an object, comprising the steps of:

5. A manufacturing process development method comprising: estimating a physical quantity distribution of the object in the specified process by using the method for estimating physical quantity distribution of the object according to any one of claims 1 to 4; and developing a manufacturing process in the specified process based on the estimated physical quantity distribution of the object.

8. Using training data in which a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of a cooling liquid for cooling an object are input values ​​and information on the amount of heat removed by the cooling liquid from the object is output values, A model is generated by machine learning, in which a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of the cooling liquid with respect to the object are input values, and information on the amount of heat removed from the object by the cooling liquid is output values. How to generate machine learning models.

9. A program for estimating a distribution of a physical quantity of an object when the object is cooled by spraying a cooling liquid from a spray nozzle, comprising: Computer, a heat transfer amount acquisition means for acquiring information regarding the amount of heat transfer from the object by the coolant by inputting a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of the coolant to a machine learning model; a physical quantity distribution calculation means for calculating a physical quantity distribution of at least one of a surface and an interior of the object using the amount of heat dissipation as a boundary condition; A program for estimating the distribution of physical quantities of an object.

10. 1. An apparatus for estimating a distribution of a physical quantity of an object when the object is cooled by spraying a cooling liquid from a spray nozzle, comprising: a heat transfer amount acquisition unit that acquires information regarding an amount of heat transfer from the object by the coolant by inputting a plurality of flow velocities in a normal direction and a plurality of flow velocities in a tangential direction of the coolant to the object into the machine learning model; a physical quantity distribution calculation unit that calculates a physical quantity distribution of at least one of a surface and an interior of the object using the amount of heat dissipation as a boundary condition; An object physical quantity distribution estimation device comprising:

11. The device for estimating a physical quantity distribution of an object according to claim 10 , wherein the physical quantity is a temperature or a solidification.

Images