Material property prediction device, operation condition determination device, material property prediction method, operation condition determination method, and program

Abstract

To provide a material characteristic prediction device and the like capable of accurately predicting a material characteristic value of a steel material and interpreting a prediction result of a prediction model.SOLUTION: A material characteristic prediction device 100 according to the present invention is a device that predicts a material characteristic value of a steel material, and inputs an operation condition in a manufacturing process of the steel material and outputs a feature amount obtained by quantifying a change in a structure of the steel material due to a metallurgical phenomenon in the manufacturing process characterized by the operation condition. And a material characteristic value prediction model MM that receives at least the feature quantity as an input and outputs an expected value or a probability distribution of the material characteristic value of the steel material, in which the material characteristic value of the target steel material is predicted by inputting the feature quantity output from the phenomenology model by inputting the operation result in the manufacturing process of the target steel material to the phenomenology model to the material characteristic value prediction model and outputting the expected value or the probability distribution of the material characteristic value of the target steel material.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a material property prediction device, a material property prediction method, and a program for causing a computer to function as a material property prediction device, which are provided with a prediction model for predicting material property values ​​such as tensile stress of steel material based on operating conditions in a manufacturing process of steel material such as steel plate, and are capable of accurately predicting material property values ​​and interpreting the prediction results of the prediction model.The present invention also relates to an operating condition determination device, an operating condition determination method, and a program for causing a computer to function as an operating condition determination device, which are provided with a prediction model for predicting material property values ​​of steel material based on operating conditions in the manufacturing process of steel material, are capable of determining appropriate operating conditions in the manufacturing process of steel material according to order information, and are capable of interpreting the prediction results of the prediction model. [Background technology]

[0002] Prediction models for predicting the material property values ​​of steel based on the operating conditions in the steel manufacturing process are used in various operations in the steel industry. For example, prediction models are used in quality assurance of steel products. In other words, to guarantee the required quality of steel products, after all manufacturing processes of the steel product are completed, the actual operating conditions (operational results) of the manufacturing processes are input into the prediction model to predict the material property values ​​over the entire length of the steel product. Furthermore, the predictive model is utilized in quality design work (the work of determining appropriate operating conditions in the steel manufacturing process so that the material property values ​​of the manufactured steel satisfy the order specifications). In other words, appropriate operating conditions are determined by back-analyzing the predictive model so that the material property values ​​of the steel predicted by the predictive model satisfy the order specifications. Furthermore, in order to prevent a decrease in product yield due to the material property values ​​of the manufactured steel not satisfying the order specifications (the material property values ​​of the steel are rejected), when the operational results of the upstream process in a manufacturing process having upstream and downstream processes deviate from the appropriate operating conditions (operating conditions with a high probability of the material property values ​​of the steel being acceptable), the predictive model is utilized in the task of appropriately determining the operating conditions of the downstream process to be executed in the future based on the operational results of the upstream process so as to increase the probability of the material property values ​​of the steel being acceptable (in this specification, this is referred to as "material property value FF (feedforward) technology"). In other words, appropriate operating conditions for the downstream process are determined by back-analyzing the predictive model so that the material property values ​​of the steel predicted by the predictive model satisfy the order specifications.

[0003] As the above-mentioned predictive models, machine learning models constructed by machine learning such as neural networks are often used. However, because machine learning models are black boxes, it is difficult to interpret the prediction results of the predictive models. Another problem with machine learning models is that they have poor prediction accuracy for operating conditions that fall outside the range of past operating results (operating conditions that fall outside the range of training data).

[0004] Conventionally, as a technique for predicting material property values ​​of steel materials using a prediction model, for example, the technique described in Patent Document 1 has been proposed. Also, as a technique for determining operating conditions in a steel material manufacturing process using a prediction model, for example, the technique described in Patent Document 2 has been proposed. However, the techniques described in Patent Documents 1 and 2 cannot solve the problems mentioned above, such as the difficulty in interpreting the prediction results of the prediction model and the poor prediction accuracy for operating conditions outside the range of past operating performance.

[0005] Non-Patent Documents 1 and 2 describe heteroscedastic Gaussian processes (Gaussian process regression taking heteroscedasticity into consideration). Furthermore, Non-Patent Document 3 describes matters relating to an equation for calculating the critical cooling rate, which is the minimum cooling rate at which martensite is generated in the quenched structure during cooling during quenching of steel material. Furthermore, Non-Patent Document 4 describes matters related to a formula for calculating a tempering parameter, which is an index showing the effect of the tempering temperature (holding temperature) and holding time on the hardness of the steel material after tempering. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2022-48037 [Patent Document 2] Japanese Patent Application Publication No. 2020-114597 [Non-patent literature]

[0007] [Non-Patent Document 1] Alan D. Saul, James Hensman, Aki Vehtari, Neil D. Lawrence, "Chained Gaussian Processes", Proceedings of the 19th International Conference on Artificial Intelligence and Statistics, PMLR 51:1431-1440, 2016 [Non-patent document 2] "Heteroskedastic Likelihood and Multi-Latent GP", [online], [Retrieved October 23, 2023], Internet<URL:https: / / gpflow.github.io / GPflow / develop / notebooks / advanced / heteroskedastic.html> [Non-patent document 3] Masakatsu Ueno and Kametaro Ito, "A new prediction formula for hardenability of steel to replace the Grossmann formula", Iron and Steel, 74th year (1988), No. 6, pp. 1073-1080 [Non-patent document 4] Tsuyoshi Inoue, "New Tempering Parameters and Their Application to the Integration Method of Tempering Effects along Continuous Heating Curves," Iron and Steel, Vol. 66 (1980), No. 10, pp. 1532-1541 Summary of the Invention [Problem to be solved by the invention]

[0008] The present invention has been made to solve the problems of the prior art, and aims to provide a material property prediction device, a material property prediction method, and a program for causing a computer to function as a material property prediction device, which are equipped with a prediction model for predicting material property values ​​such as tensile stress of steel based on operating conditions in a steel manufacturing process, and are capable of accurately predicting material property values ​​of steel and capable of interpreting the prediction results of the prediction model. Another aim of the present invention is to provide an operating condition determination device, an operating condition determination method, and a program for causing a computer to function as an operating condition determination device, which are equipped with a prediction model for predicting material property values ​​of steel based on operating conditions in the steel manufacturing process, are capable of determining appropriate operating conditions in the steel manufacturing process according to order information, and are capable of interpreting the prediction results of the prediction model. [Means for solving the problem]

[0009] In order to solve the above-mentioned problems, the present invention provides a material property prediction device that predicts the material property values ​​of steel material, comprising: a metallurgically formulated phenomenological model that receives as input operating conditions in a manufacturing process of the steel material and outputs feature quantities that quantify changes in the structure of the steel material due to metallurgical phenomena during the manufacturing process characterized by the operating conditions; and a material property value prediction model that receives as input at least the feature quantities and outputs expected values ​​or probability distributions of the material property values ​​of the steel material, and that predicts the material property values ​​of the target steel material by inputting operational performance in the manufacturing process of a target steel material into the phenomenological model, and inputting the feature quantities output from the phenomenological model into the material property value prediction model and outputting expected values ​​or probability distributions of the material property values ​​of the target steel material.

[0010] In the material property prediction device according to the present invention, the "manufacturing process" refers to a process up to the process of manufacturing a steel material into a product (such as a hot rolling process, cold rolling process, or annealing process) by changing the thickness, surface condition, or internal metal structure of the steel material, and examples thereof include a steelmaking process, a hot rolling process, a cold rolling process, and an annealing process. The manufacturing process may not include a finishing process that only involves cutting and inspecting the steel material. The same applies to the operating condition determination device according to the present invention, which will be described later. Furthermore, in the material property prediction device according to the present invention, "operating conditions" are conditions that affect the quality of the material property values ​​of the steel material in the manufacturing process, and examples thereof include component values ​​of carbon (C), manganese (Mn), silicon (Si), etc. in the steelmaking process, coiling temperature in the hot rolling process, size such as the thickness of the steel material after cold rolling in the cold rolling process, heating temperature, soaking temperature, holding temperature and holding time in the annealing process, etc. The same applies to the operating condition determination device according to the present invention described below. Furthermore, in the material property prediction device according to the present invention, "operational results" refer to actual values ​​of operational conditions, and are data that affect the quality of the material property values ​​of the steel material. The operational results may include not only measured values ​​and set values ​​obtained in the steel manufacturing process, but also data obtained by processing these values ​​using physical formulas, etc. The same applies to the operational condition determination device according to the present invention, which will be described later. Furthermore, in the material property prediction device according to the present invention, the "material property value" refers to a mechanical property value of a steel material, such as tensile strength (TS), yield point (YP), elongation, surface hardness, bending strength, toughness, and fatigue life. The same applies to the operating condition determination device according to the present invention, which will be described later. Furthermore, in the material property prediction device according to the present invention, examples of "feature quantities that quantify changes in the structure of a steel material due to metallurgical phenomena during a manufacturing process characterized by operational conditions" include, for example, a feature quantity that correlates with the structure of a steel material characterized by component values ​​in the steelmaking process, such as a critical cooling rate, which is the minimum cooling rate at which martensite forms in the quenched structure during cooling of the steel material during quenching in the annealing process. Also, examples of feature quantities that correlate with the hardness of a steel material characterized by the holding temperature and holding time in the annealing process, such as a tempering parameter, are examples of feature quantities that correlate with the hardness of a steel material after tempering in the annealing process. The same applies to the operational condition determination device according to the present invention, which will be described later. Furthermore, in the material property prediction device according to the present invention, the term "target steel material" refers to a steel material whose material property values ​​are to be predicted.

[0011] The material property prediction device according to the present invention comprises a metallurgically formulated phenomenological model that takes as input the operating conditions in a steel manufacturing process and outputs feature quantities that quantify changes in the structure of the steel due to metallurgical phenomena during the manufacturing process characterized by the operating conditions, and a material property value prediction model that takes as input at least the feature quantities output from the phenomenological model and outputs expected values ​​or probability distributions of the material property values ​​of the steel. The material property value prediction model included in the material property prediction device according to the present invention outputs an expected value or a probability distribution of the material property value of a steel material. When the material property value prediction model outputs an expected value of the material property value of a steel material, the material property value prediction model is similar to a conventional machine learning model. On the other hand, when the material property value prediction model outputs a probability distribution of the material property value of a steel material, in other words, when the material property value prediction model outputs the expected value and variability of the material property value as prediction results, it is expected that the material property value (especially the variability) can be accurately predicted even when the feature values ​​output by inputting operational results from the manufacturing process of the target steel material, which have no past operational results, are input into the phenomenological model. Furthermore, it is expected that the prediction accuracy of the material property value can be improved even for operational results from the target steel material, which are difficult to control and prone to variability in the material property value. Furthermore, according to the findings of the inventors, even if the operational performance of the target steel material is outside the range of past operational performance, the features input to the material property value prediction model (feature values ​​output from the phenomenological model) often fall within the range of feature values ​​obtained for the past operational performance. In this respect, it is expected that the prediction accuracy of the material property value will be improved regardless of whether the material property value prediction model outputs an expected value or a probability distribution of the material property value of the steel material. The phenomenological model provided in the material property prediction device according to the present invention is a metallurgically formulated model that outputs feature quantities that quantify changes in the structure of steel due to metallurgical phenomena during the manufacturing process characterized by the operating conditions. Therefore, it has the advantage that it becomes possible to interpret the prediction results by visualizing the relationship between the feature quantities such as the critical cooling rate and tempering parameters output from the phenomenological model and the prediction results of the material property values ​​output from the material property value prediction model. As described above, the material property prediction device according to the present invention can accurately predict the material property values ​​of steel materials, and also makes it possible to interpret the prediction results of the prediction model.

[0012] The material property value prediction model provided in the material property prediction device according to the present invention is constructed by using known data, for example, in which at least the feature quantities output when the operational performance of a manufacturing process of a steel material manufactured in the past (also referred to as a "reference steel material" in this specification) is input to a phenomenological model, and the material property values ​​measured on a test piece obtained from the reference steel material are output. For example, the model may be constructed by machine learning using such known data as training data. The same applies to the material property value prediction model provided in the operating condition determination device according to the present invention, which will be described later. Furthermore, the material property value prediction model may be constructed by the material property prediction device according to the present invention, or a material property value prediction model constructed by another device may be stored in the material property prediction device according to the present invention.

[0013] In the material property prediction device of the present invention, preferably, the material property value prediction model takes as input the feature values ​​and the operating conditions in the manufacturing process of the steel material, and outputs an expected value or probability distribution of the material property values ​​of the steel material; and by inputting the operational performance in the manufacturing process of the target steel material into the phenomenological model, the feature values ​​output from the phenomenological model and the operational performance in the manufacturing process of the target steel material are input into the material property value prediction model, and the expected value or probability distribution of the material property values ​​of the target steel material is output, thereby predicting the material property values ​​of the target steel material.

[0014] According to the above-described preferred configuration, by using the operating conditions in the steel manufacturing process in addition to the feature values ​​output from the phenomenological model as input to the material property value prediction model, it is possible to compensate for phenomena related to material property values ​​that cannot be taken into account by the phenomenological model alone, and it is expected that the prediction accuracy of material property values ​​will be improved.

[0015] Furthermore, in order to solve the above-mentioned problems, the present invention provides an operational condition determination device that determines operational conditions in a steel manufacturing process in accordance with order information, the operational condition determination device comprising: a metallurgically formulated phenomenological model that receives as input the operational conditions in the steel manufacturing process and outputs feature quantities that quantify changes in the structure of the steel due to metallurgical phenomena during the manufacturing process characterized by the operational conditions; and a material property value prediction model that receives as input at least the feature quantities and outputs a probability distribution of material property values ​​of the steel, and the operational condition determination device determines the operational conditions in the manufacturing process of the target steel by inversely analyzing the phenomenological model and the material property value prediction model so that the probability of acceptance of the material property values ​​of the target steel, which is calculated by inputting the operational conditions in the manufacturing process of the target steel into the phenomenological model, and inputting the feature quantities that are output from the phenomenological model into the material property value prediction model, and the specifications of the material property values ​​of the target steel included in the order information, is equal to or greater than a predetermined threshold value.

[0016] In the operating condition determination device according to the present invention, the term "target steel material" refers to a steel material for which operating conditions in a manufacturing process are to be determined. Furthermore, in the operating condition determination device according to the present invention, the "specifications of the material property values ​​of the target steel material included in the order information" may include, for example, the required upper limit value of the material property value, or the required lower limit value, or both. Furthermore, in the operating condition determination device according to the present invention, "inversely analyzing the phenomenological model and the material property value prediction model" means inversely analyzing a composite function of a function represented by the phenomenological model and a function represented by the operating condition prediction model.

[0017] The operational condition determination device according to the present invention includes a metallurgically formulated phenomenological model that receives as input operational conditions in a steel manufacturing process and outputs feature values ​​that quantify changes in the structure of the steel due to metallurgical phenomena during the manufacturing process characterized by the operational conditions, and a material property value prediction model that receives as input at least the feature values ​​output from the phenomenological model and outputs a probability distribution of material property values ​​of the steel. The material property value prediction model may be constructed by the operational condition determination device according to the present invention, or a material property value prediction model constructed by another device may be stored in the operational condition determination device according to the present invention. The material property value prediction model included in the operating condition determination device according to the present invention outputs a probability distribution of material property values ​​of a steel material, in other words, the expected value and variability of material property values, as prediction results. Therefore, even when feature values ​​output from the phenomenological model are input into the material property value prediction model under operating conditions in the manufacturing process of the target steel material that have no past operating history, the material property values ​​(especially variability) can be predicted with high accuracy. Furthermore, the prediction accuracy of material property values ​​can be improved even for operating conditions of the target steel material that are difficult to control and prone to variability in material property values. Furthermore, according to the inventors' findings, even when the operating conditions of the target steel material are outside the range of past operating history, the feature values ​​input into the material property value prediction model (feature values ​​output from the phenomenological model) often fall within the range of feature values ​​obtained from past operating history, which can also be expected to improve the prediction accuracy of material property values. The phenomenological model provided in the operating condition determination device according to the present invention is a metallurgically formulated model that outputs feature quantities that quantify changes in the structure of steel due to metallurgical phenomena during the manufacturing process characterized by the operating conditions. Therefore, it has the advantage that it becomes possible to interpret the prediction results by visualizing the relationship between the feature quantities such as the critical cooling rate and tempering parameters output from the phenomenological model and the prediction results of the material property values ​​output from the material property value prediction model.

[0018] In the operating condition determination device according to the present invention, the pass probability of the material property value of the steel is calculated based on the probability distribution of the material property value of the steel output by inputting the feature values ​​output from the phenomenological model into the material property value prediction model and the specifications of the material property value of the steel included in the order information. For example, if the specifications of the material property value of the steel included in the order information are the required upper limit and lower limit of the material property value, the pass probability is calculated as the ratio of the area within the range from the required lower limit to the required upper limit to the area of ​​the entire probability distribution of the material property value. Therefore, by performing reverse analysis on the phenomenological model and the material property value prediction model so that the probability of passing the material property values ​​of the steel material is above a predetermined threshold, it is possible to determine appropriate operating conditions in the manufacturing process according to the order information, which makes it possible, for example, to improve the efficiency of quality design work.

[0019] In the operating condition determination device of the present invention, preferably, the material property value prediction model takes as input the feature values ​​and the operating conditions in the steel manufacturing process, and outputs a probability distribution of the material property values ​​of the steel, and the operating conditions in the manufacturing process of the target steel are input into the phenomenological model, and the feature values ​​are output from the phenomenological model, and the operating conditions in the manufacturing process of the target steel are input into the material property value prediction model, and the probability distribution of the material property values ​​of the target steel is output by inputting the feature values ​​and the operating conditions in the manufacturing process of the target steel into the material property value prediction model, and the specifications of the material property values ​​of the target steel included in the order information.The operating conditions in the manufacturing process of the target steel are preferably determined by inverse analysis of the phenomenological model and the material property value prediction model so that the pass probability of the material property values ​​of the target steel is equal to or greater than a predetermined threshold value.

[0020] According to the above-described preferred configuration, by using the operating conditions in the steel manufacturing process in addition to the feature values ​​output from the phenomenological model as inputs to the material property value prediction model, it is possible to compensate for phenomena related to material property values ​​that cannot be taken into account by the phenomenological model alone, which is expected to improve the prediction accuracy of material property values ​​and ultimately make it possible to determine even more appropriate operating conditions.

[0021] The operating condition determination device according to the present invention can be used not only for quality design work but also for material characteristic value FF technology work. In other words, a preferred operating condition determination device of the present invention is an operating condition determination device in which the steel manufacturing process has a pre-process and a post-process, and after the pre-process of the steel is executed, the operating conditions in the post-process of the steel are determined according to the order information, and the operating conditions in the post-process of the target steel are determined by inversely analyzing the phenomenological model and the material property value prediction model so that the probability of passing the material property values ​​of the target steel, calculated by inputting the operational performance of the target steel in the pre-process and the operating conditions in the post-process of the target steel into the phenomenological model, and inputting the feature values ​​output from the phenomenological model into the material property value prediction model, and the specifications of the material property values ​​of the target steel included in the order information, is greater than or equal to a predetermined threshold value. Alternatively, a preferred operating condition determination device according to the present invention is an operating condition determination device in which the steel manufacturing process has a pre-process and a post-process, and after the pre-process of the steel is carried out, the operating conditions for the post-process of the steel are determined according to the order information, and the operating conditions for the post-process of the target steel are determined by inversely analyzing the phenomenological model and the material property value prediction model so that the pass probability of the material property values ​​of the target steel calculated using the feature values ​​output from the phenomenological model by inputting the operational performance of the target steel in the pre-process and the operational conditions in the post-process into the phenomenological model, the operational performance of the target steel in the pre-process and the operational conditions in the post-process into the material property value prediction model, and the specifications of the material property values ​​of the target steel included in the order information, is greater than or equal to a predetermined threshold.

[0022] According to the above-described preferred configuration, it is possible to determine appropriate operating conditions for downstream processes in accordance with the order information, and to suppress a decrease in the product yield of the target steel material.

[0023] Furthermore, a preferred operating condition determining device according to the present invention determines operating conditions for all manufacturing processes of the target steel material. According to the above-described preferred configuration, it is possible to improve the efficiency of quality design work.

[0024] Furthermore, in order to solve the above-mentioned problems, the present invention is also provided as a material property prediction method for predicting material property values ​​of steel material, which uses a metallurgically formulated phenomenological model that takes as input operating conditions in a manufacturing process of the steel material and outputs feature quantities that quantify changes in the structure of the steel material due to metallurgical phenomena during the manufacturing process characterized by the operating conditions, and a material property value prediction model that takes as input at least the feature quantities and outputs expected values ​​or probability distributions of the material property values ​​of the steel material, and which predicts the material property values ​​of the target steel material by inputting operational performance in the manufacturing process of the target steel material into the phenomenological model, and inputting the feature quantities output from the phenomenological model into the material property value prediction model and outputting expected values ​​or probability distributions of the material property values ​​of the target steel material.

[0025] Furthermore, in order to solve the above-mentioned problems, the present invention is also provided as an operational condition determination method for determining operational conditions in a steel manufacturing process in accordance with order information, the operational condition determination method using a metallurgically formulated phenomenological model that takes as input the operational conditions in the steel manufacturing process and outputs feature quantities that quantify changes in the structure of the steel due to metallurgical phenomena during the manufacturing process characterized by the operational conditions, and a material property value prediction model that takes at least the feature quantities as input and outputs a probability distribution of material property values ​​of the steel, and determining the operational conditions in the manufacturing process of the target steel by reverse-analyzing the phenomenological model and the material property value prediction model so that the probability of acceptance of the material property values ​​of the target steel, which is calculated by inputting the operational conditions in the manufacturing process of the target steel into the phenomenological model, and inputting the feature quantities output from the phenomenological model into the material property value prediction model, and the specifications of the material property values ​​of the target steel included in the order information, is equal to or greater than a predetermined threshold value.

[0026] Furthermore, in order to solve the above-mentioned problems, the present invention also provides a program for causing a computer to function as the material property prediction device, and a program for causing a computer to function as the operating condition determination device. [Effects of the Invention]

[0027] According to the present invention, it is possible to accurately predict the material property values ​​of steel products and to interpret the prediction results of the prediction model. Furthermore, according to the present invention, it is possible to determine appropriate operating conditions in the steel product manufacturing process according to order information and to interpret the prediction results of the prediction model. [Brief explanation of the drawings]

[0028] [Figure 1] 1 is a block diagram showing a schematic configuration of a material property prediction device according to a first embodiment. [Figure 2] 2 is a flowchart showing an outline of the procedure of a material property prediction method executed using the material property prediction device 100 shown in FIG. [Figure 3] 2 is a diagram showing an example of operation results acquired by the operation results acquisition unit 10 shown in FIG. 1. FIG. [Figure 4] FIG. 2 is a diagram for schematically explaining an overview of a material characteristic value predicting unit 60 shown in FIG. [Figure 5] FIG. 5 is a diagram showing an example of input for a material characteristic value prediction model MM shown in FIG. [Figure 6] 2 is a diagram showing an example of the content output from the material characteristic value output unit 70 shown in FIG. [Figure 7] FIG. 2 is a diagram illustrating an example of an advantage of providing the material property prediction apparatus 100 according to the first embodiment with the phenomenological model PM. [Figure 8] FIG. 10 is a diagram illustrating another example of the advantage of the material property prediction apparatus 100 according to the first embodiment being provided with the phenomenological model PM. [Figure 9] FIG. 10 is a diagram for schematically explaining an overview of an operating condition determination device according to a second embodiment. [Figure 10]FIG. 10 is a block diagram showing a schematic configuration of an operation condition determination device according to a second embodiment. [Figure 11] 11 is a flowchart showing an outline of the procedure of an operating condition determination method executed using the operating condition determination device 200 shown in FIG. [Figure 12] 11 is a diagram showing an example of order information acquired by an order information acquiring unit 60A shown in FIG. 10 and operation results in a previous process acquired by a previous process operation results acquiring unit 10A. FIG. [Figure 13] 11 is a diagram showing an example of the content output from the post-process operating condition output unit 90 shown in FIG. 10. FIG. [Figure 14] FIG. 1 is a diagram summarizing the operating conditions input to various phenomenological models. DETAILED DESCRIPTION OF THE INVENTION

[0029] Hereinafter, with reference to the accompanying drawings as appropriate, the material property prediction device according to the first embodiment of the present invention and the operating condition determination device according to the second and third embodiments will be described using an example in which the steel material is a steel plate.

[0030] [First embodiment] Fig. 1 is a block diagram showing a schematic configuration of a material property prediction device according to the first embodiment. Fig. 2 is a flow chart showing a schematic procedure of a material property prediction method executed using the material property prediction device shown in Fig. 1. The material property prediction device 100 according to the first embodiment is a device for predicting material property values ​​of steel materials, and is used in quality assurance work for steel materials. As shown in Fig. 1, the material property prediction device 100 includes an operation performance acquisition unit 10, a phenomenological model storage unit 20, a phenomenological model acquisition unit 30, a material property value prediction model storage unit 40, a material property value prediction model acquisition unit 50, a material property value prediction unit 60, and a material property value output unit 70.

[0031] The material property prediction apparatus 100 is configured, for example, by a computer having one or more hardware processors, such as a central processing unit (CPU), and one or more memories, such as a random access memory (RAM) and a read-only memory (ROM). One or more programs stored in the memory are executed by the one or more hardware processors to perform various calculations. As a result, the material property prediction apparatus 100 functions as an operation performance acquisition unit 10, a phenomenological model storage unit 20, a phenomenological model acquisition unit 30, a material property value prediction model storage unit 40, a material property value prediction model acquisition unit 50, a material property value prediction unit 60, and a material property value output unit 70. The material property prediction apparatus 100 may be implemented as a programmable logic controller (PLC) or dedicated hardware such as an application-specific integrated circuit (ASIC). Each of the units 10 to 70 included in the material property prediction apparatus 100 will be described below.

[0032] <Operational Performance Acquisition Department 10> The operation history obtaining unit 10 executes step ST11 shown in Fig. 2. Specifically, the operation history obtaining unit 10 obtains, from a predetermined database (not shown), operation history in the manufacturing process of the target steel material, which is obtained after the manufacturing process (in the first embodiment, the steelmaking process, the hot rolling process, the cold rolling process, and the annealing process) of the target steel material whose material characteristic value is to be predicted. FIG. 3 is a diagram showing an example of operational performance acquired by the operational performance acquisition unit 10. In the example shown in FIG. 3, the material characteristic value prediction unit 60 calculates the pass probability of the material characteristic value of the target steel material, as will be described later, and therefore also acquires order information for the target steel material in addition to operational performance. However, if the material characteristic value prediction unit 60 does not calculate the pass probability of the material characteristic value of the target steel material (only predicts the material characteristic value), it is not necessarily necessary to acquire order information for the target steel material. In the example shown in FIG. 3, the operational performance acquisition unit 10 acquires, as operational performance in the manufacturing process of the target steel material, the carbon (C) component value x C , manganese (Mn) component value x MNand silicon (Si) component value x SI and the coiling temperature x in the hot rolling process CT and the heating temperature x in the annealing process (continuous annealing process) HF , soaking temperature x S , holding temperature x OAtemp and retention time x OAtime The operation record acquiring unit 10 acquires the size of the target steel material (product thickness x t ) is treated as the operational performance in the cold rolling process (the actual thickness of the target steel material after cold rolling). This is because in many cases the thickness of the target steel material after cold rolling is manufactured without any error from the product thickness, which is the order information. However, this is not limited to this, and it is also possible to actually measure the thickness of the target steel material after cold rolling using a known thickness gauge, and use the measured value as the operational performance.

[0033] <Phenomenological Model Storage Unit 20> The phenomenological model storage unit 20 stores (memorizes) a phenomenological model formulated in metallurgy, which takes as input the operating conditions in the steel manufacturing process and outputs feature quantities that quantify changes in the steel structure due to metallurgical phenomena during the manufacturing process characterized by the operating conditions. The phenomenological model storage unit 20 of the first embodiment stores the component value x C , x MN and x SI The phenomenological model storage unit 20 stores a phenomenological model that outputs a critical cooling rate z1, which is a feature quantity correlated with the structure of a steel material characterized by the following formula (1) and formula (1)', which can be derived from the description of Non-Patent Document 3. z1=g1(x C ,x MN ,x SI )=10 xx ···(1) xx=2.94-0.75(2.7x C +0.4x SI +x MN ) ···(1)' The phenomenological model storage unit 20 also stores the holding temperature x OAtemp and retention time x OAtime The phenomenological model storage unit 20 stores a phenomenological model that outputs a tempering parameter z2, which is a feature quantity correlated with the hardness of a steel material characterized by the following equation: z2 is an index that represents the influence on the hardness of the steel material after tempering in the annealing process. Specifically, the phenomenological model storage unit 20 stores a phenomenological model expressed by the following equation (2), which can be derived from the description of Non-Patent Document 4. z2=g2(x OAtemp ,x OAtime )=(x OAtemp +273)(20+log(x OAtime )) / 1000 ···(2) In the above equation (2), the coefficient multiplied by the entire right-hand side is 1, and the value of the first term in the second parentheses on the right-hand side is 20. These values ​​are adjusted according to the component values ​​in the steelmaking process. However, the present invention is not limited to this, and various types of phenomenological models can be stored in the phenomenological model storage unit 20 and used in the material characteristic value prediction unit 60 described below, as long as they are metallurgically formulated phenomenological models that take operating conditions in a steel manufacturing process as input and output feature amounts that quantify changes in the structure of the steel due to metallurgical phenomena during the manufacturing process characterized by the operating conditions. Other examples of phenomenological models will be described later.

[0034] <Phenomenological Model Acquisition Unit 30> The phenomenological model acquisition unit 30 executes step ST12 shown in Fig. 2. Specifically, the phenomenological model acquisition unit 30 acquires the phenomenological model z1 = g1(x C ,x MN ,x SI ) and z2=g2(x OAtemp ,x OAtime ) to get the

[0035] <Material characteristic value prediction model storage unit 40> The material property value prediction model storage unit 40 stores (memorizes) a material property value prediction model that receives as input at least the critical cooling rate z1 and tempering parameter z2, which are feature quantities output from the phenomenological model, and outputs an expected value or probability distribution of the material property values ​​of the steel (probability distribution in the first embodiment). In a preferred embodiment, the material property value prediction model of the first embodiment receives as input operating conditions in the manufacturing process of the steel, in addition to the critical cooling rate z1 and tempering parameter z2.

[0036] The material property value prediction model stored in the material property value prediction model storage unit 40 is constructed using known data, which takes as input the critical cooling rate z1 and tempering parameter z2 output when the operational performance of the manufacturing process of a reference steel material produced in the past is input to the phenomenological model, and the operational performance of the manufacturing process of the reference steel material, and outputs material property values ​​measured on a test piece obtained from the reference steel material. For example, the material property value prediction model may be constructed by machine learning using such known data as training data. In the first embodiment, the material property value prediction model is constructed using Gaussian process regression, a type of Bayesian estimation, to quantify the uncertainty of prediction due to the lack of past operational performance. However, the present invention is not limited to this. For example, the material property value prediction model may be constructed using a statistical method such as quantile regression using known data, or by other machine learning methods such as a Bayesian neural network. The following provides an overview of the Gaussian process regression used in the first embodiment. Note that in the following equations, variables written in bold represent vectors or matrices.

[0037] When the output vector f, expressed by the following equation (4) corresponding to the input vector x, expressed by the following equation (3), follows a Gaussian distribution N(0,K) with a mean of 0 and a covariance matrix K, expressed by the following equation (5), the output f is said to follow a Gaussian process, and is written as shown in the following equation (6).

number

number

[0038] Although it is possible to use ordinary Gaussian process regression as shown in the above formula (7) as a material property value prediction model, ordinary Gaussian process regression cannot take into account changes in the variation that occurs in material property values ​​due to differences in operational performance. For this reason, in the first embodiment, Gaussian processes gP(0,K1) and gP(0,K2) that individually estimate the expected value (average value) and variance (standard deviation) expressed by the following formulas (8) to (12) are defined, and a heteroscedastic Gaussian process (a Gaussian process that takes heteroscedasticity into account) is used in which a Gaussian distribution with the output of each Gaussian process as a parameter is used as the final prediction result.

number

[0039] In the first embodiment, a material characteristic value prediction model using the above-described non-homogeneous variance Gaussian process is constructed and stored in the material characteristic value prediction model storage unit 40. The material property value prediction model may be constructed by having the material property prediction device 100 equipped with a material property value prediction model construction unit (not shown), which constructs the material property value prediction model and stores it in the material property value prediction model storage unit 40, or a material property value prediction model constructed by another device may be stored in the material property value prediction model storage unit 40.

[0040] <Material characteristic value prediction model acquisition unit 50> The material characteristic value prediction model acquisition unit 50 executes step ST13 shown in Fig. 2. Specifically, the material characteristic value prediction model acquisition unit 50 acquires the material characteristic value prediction model stored in the material characteristic value prediction model storage unit 40.

[0041] <Material characteristic value prediction unit 60> The material property value prediction unit 60 executes step ST14 shown in Fig. 2. Specifically, the material property value prediction unit 60 predicts the material property values ​​of the target steel material using the phenomenological model acquired by the phenomenological model acquisition unit 30 and the material property value prediction model acquired by the material property value prediction model acquisition unit 50. In the first embodiment, the yield point (YP) and tensile strength (TS) are predicted as the material property values. 4 is a diagram for schematically explaining the outline of the material characteristic value prediction unit 60. As shown in FIG. 4, the material characteristic value prediction unit 60 predicts the operating conditions x in the manufacturing process of the steel material. i is input, and the operating conditions x i The feature z is characterized by i The phenomenological model PM outputs (=z1, z2), and the feature z output from the phenomenological model PM i and operating conditions in the steel manufacturing process x i The material property value of the target steel material is predicted using a material property value prediction model MM that takes as input the operation performance x in the manufacturing process of the target steel material acquired by the operation performance acquisition unit 10 and outputs the probability distribution of the material property value of the steel material. i is input to the phenomenological model PM, the feature value z i Then, the material characteristic value prediction unit 60 outputs the feature value z iand operational performance in the manufacturing process of the target steel material x i are input to the material characteristic value prediction model MM, and the probability distribution of the material characteristic values ​​of the target steel material is output from the material characteristic value prediction model MM.

[0042] 5 is a diagram showing an example of an input of the material property value prediction model MM. In the example shown in FIG. 5, the feature value z i In addition to the critical cooling rate z1 and tempering parameter z2, the operational performance x i The coiling temperature x in the hot rolling process CT and the product thickness x as the thickness after cold rolling in the cold rolling process t and the heating temperature x in the annealing process HF , soaking temperature x S , holding temperature x OAtemp and retention time x OAtime and are input to the material property value prediction model MM. That is, the vector x = [x t ,x CT ,x HF ,x S ,x OAtemp ,x OAtime , z1, z2] are used as inputs to the material property value prediction model MM. In the example shown in Fig. 5, the operational performance x of the steelmaking process is used as the input for the material property value prediction model MM. C , x MN and x SI (see FIG. 3) are excluded because these operational results are used as inputs to the phenomenological model PM expressed by equations (1) and (1)' as described above. However, it is not necessary to exclude the operational results used as inputs to the phenomenological model PM, and these operational results may be used as inputs to the material characteristic value prediction model MM. In the example shown in FIG. 5, the holding temperature x in the annealing process for calculating the tempering parameter z2, which is also used as inputs to the phenomenological model PM, is used as inputs to the material characteristic value prediction model MM. OAtemp and retention time x OAtime However, these may be excluded from the input. In other words, the input of the material property value prediction model MM is the operational performance data x used as the input of the phenomenological model PM.i may or may not include. The material characteristic value prediction unit 60 calculates the average value μ of the material characteristic values ​​S∈{YP,TS} of the target steel material from the material characteristic value prediction model MM for the input vector x. S (x) and standard deviation σ S (x) is output as the predicted result. In addition, in a preferred embodiment, the material characteristic value prediction unit 60 of the first embodiment calculates the probability distribution (mean value μ S (x) and standard deviation σ S (x)) and the specification of the material property value of the target steel material included in the order information acquired by the operation record acquisition unit 10 (in the example shown in FIG. 3, the lower limit value y YP lower , the upper limit of YP y YP upper , the lower limit of TS y TS lower ) to calculate the pass probability of the material characteristic value of the target steel material. The specific method for calculating the pass probability is the same as that described in the second embodiment below, and therefore will not be described here.

[0043] <Material characteristic value output unit 70> The material characteristic value output unit 70 executes step ST15 shown in Fig. 2. Specifically, the material characteristic value output unit 70 outputs the average values ​​μ of the material characteristic values ​​YP and TS of the target steel material predicted by the material characteristic value prediction unit 60. S (x) and standard deviation σ S In the first embodiment, as a preferred mode, the pass probability of the material characteristic value of the target steel material calculated by the material characteristic value prediction unit 60 is also output. Fig. 6 is a diagram showing an example of the content output from the material characteristic value output unit 70. In Fig. 6, for the target steel material No. 001, the average value and standard deviation of YP and TS are calculated as the predicted results of the material characteristic values, and it is shown that the pass probability is a high value of 0.983 for the specifications of the material characteristic values ​​of the target steel material included in the order information shown in Fig. 3.

[0044] In the first embodiment, as shown in Fig. 3, representative values ​​(such as average values) over the entire length and width of the target steel are used as inputs to the phenomenological model PM and the material property value prediction model MM as operational performance data for the target steel, and therefore the predicted results of the material property values ​​shown in Fig. 6 are also representative values ​​over the entire length and width of the target steel. However, the present invention is not limited to this. When operational performance data are acquired for each predetermined longitudinal position and each predetermined width position of the target steel (in the case of operational performance data for which only representative values ​​can be acquired, such as component values ​​in the steelmaking process, this also includes the case where a value equivalent to this representative value is assigned to each predetermined longitudinal position and each predetermined width position of the target steel), this can be used as input to the phenomenological model PM and the material property value prediction model MM to predict material property values ​​for each longitudinal position and each width position of the target steel. In addition, when operational results are obtained for each specified longitudinal position of the target steel (in the case where only representative values ​​can be obtained, this also includes the case where the same value as this representative value is assigned to each specified longitudinal position of the target steel), this can be used as input for the phenomenological model PM and the material property value prediction model MM, making it possible to predict material property values ​​for each longitudinal position of the target steel. Furthermore, in the first embodiment, an example was given in which the material characteristic value prediction model MM outputs the probability distribution of the material characteristic values ​​of steel material, but the present invention is not limited to this, and it is also possible to adopt a configuration in which the material characteristic value prediction model MM outputs the expected value of the material characteristic value of steel material, similar to conventional machine learning models.

[0045] 7 is a diagram illustrating an example of an advantage of the material property prediction device 100 according to the first embodiment being provided with a phenomenological model PM. Specifically, FIG. 7 shows the relationship between the manganese (Mn) component value x in the steelmaking process and the carbon (C) component value x in the steelmaking process for a plurality of steel materials whose carbon (C) component value is in the range of 0.148 to 0.152 (which is a substantially constant value). MN and the component value of silicon (Si) x SI The graph plots the relationship between the Mn component value x and the critical cooling rate z1 calculated by the phenomenological model PM using these component values ​​(calculated by the above-mentioned formula (1) and formula (1)'). MNand the component value x of Si SI The relationship between the component value x of Si and the SI and the critical cooling rate z1, and Fig. 7(c) shows the relationship between the Mn component value x MN and the critical cooling rate z1. In Figure 7, the data plotted with "●" are actual values ​​for multiple reference steel materials manufactured in the past. Therefore, the material property value prediction model MM is constructed using this data as input for known data (training data in the case of machine learning). The data plotted with "X" in Figure 7 are hypothetical data for the target steel material whose material property values ​​will be predicted from this. If the material property prediction device 100 does not have a phenomenological model PM, the actual values ​​of the reference steel plotted with "●" in FIG. 7(a) are used as inputs of known data to construct the material property value prediction model MM. Therefore, the data of the target steel plotted with "×" is within the range of known data (the component value x of Mn). MN Range of Si and component value x SI In this case, the operating conditions may fall outside the range of the critical cooling rate z1 (range of the critical cooling rate z1), which may result in poor prediction accuracy. In contrast, the material property prediction device 100 according to the first embodiment includes the phenomenological model PM. Therefore, the actual values ​​of the critical cooling rate z1 output from the phenomenological model PM, plotted with a black circle in FIGS. 7(b) and 7(c), can be used as inputs to the material property value prediction model MM. As shown in FIGS. 7(b) and 7(c), the data for the target steel material plotted with an x ​​has a critical cooling rate z1 that falls within the range of the known data. Therefore, the accuracy of the prediction of the material property value by the material property prediction device 100 is expected to be improved compared to the case shown in FIG. 7(a).

[0046] 8 is a diagram illustrating another example of the advantage of including the phenomenological model PM in the material property prediction device 100 according to the first embodiment. Specifically, FIG. 8 is a partial dependency graph showing the relationship between the feature quantities output from the phenomenological model PM and the prediction results of the material property values ​​(average values ​​of the material property values) output from the material property value prediction model MM for a plurality of reference steel materials. FIG. 8(a) shows the relationship between the critical cooling rate z1 and the prediction results of YP, FIG. 8(b) shows the relationship between the critical cooling rate z1 and the prediction results of TS, FIG. 8(c) shows the relationship between the tempering parameter z2 and the prediction results of YP, and FIG. 8(d) shows the relationship between the tempering parameter z2 and the prediction results of TS. Each partial dependency graph shown in Fig. 8 can be created as follows: Among the inputs of the material property value prediction model MM, the input of interest is designated as x target Then, x target If the predicted results of the material property values ​​(average values ​​of the material property values), which are the vertical axes of each partial dependency graph corresponding to the above, are expressed by the partial function shown in the following formula (13), this partial function can be calculated as the average value of the actual values ​​of the predicted results of the material property values ​​for multiple reference steel materials based on the Monte Carlo method. Specifically, it can be calculated by the following formula (14). The vector x in formula (14) C (n) is the input of interest x for multiple reference steels. target The inputs other than the reference steel are used as the actual values. N represents the number of reference steels. For example, in the case of the partial dependency graphs shown in Figures 8(a) and 8(b), the input of interest is z1 = g1(x C ,x MN ,x SI ), the partial function can be calculated by the following equation (15).

number

[0047] As shown in Figure 8, the material property values ​​YP and TS tend to increase as the critical cooling rate z1 and tempering parameter z2 decrease. A smaller critical cooling rate z1 facilitates the formation of martensite in the quenched structure even at a slow cooling rate, resulting in larger material property values ​​YP and TS. Therefore, the trends shown in Figures 8(a) and 8(b) are consistent with metallurgical knowledge. Furthermore, the tempering parameter z2 correlates with the softening behavior of martensite; the smaller this parameter, the harder the tempered steel becomes. Therefore, the trends shown in Figures 8(c) and 8(d) are also consistent with metallurgical knowledge. While the material property prediction model MM, constructed using machine learning, alone cannot interpret the prediction results in light of metallurgical knowledge, the phenomenological model PM can be used to interpret the behavior of the prediction model from a metallurgical perspective, making the prediction results more persuasive.

[0048] In the first embodiment described above, the material property prediction apparatus 100 may be implemented using a plurality of computers. For example, the material property prediction apparatus 100 may be implemented using an apparatus such as a cloud server. Furthermore, the hardware processor and memory of the material property prediction apparatus 100 may be distributed and implemented on a plurality of computers. The same applies to the operating condition determination apparatuses according to the second and third embodiments described below.

[0049] [Second embodiment] First, an overview of the operating condition determination device according to the second embodiment will be described. FIG. 9 is a diagram for schematically explaining the outline of the operating condition determination device according to the second embodiment. The operating condition determination device according to the second embodiment is a device that determines the operating conditions for a downstream process in accordance with order information after the upstream process has been executed for a steel material produced through a manufacturing process having upstream and downstream processes, and is utilized in material property value FF operations. Specifically, as shown in Fig. 9(a), the operating condition determination device according to the second embodiment includes a phenomenological model PM formulated in metallurgical terms, which receives as input the operating conditions for the steel material production process (upstream and downstream processes) and outputs feature quantities that quantify changes in the steel material's structure due to metallurgical phenomena during the production process characterized by the operating conditions, and a material property value prediction model MM that receives as input at least the feature quantities (in the example shown in Fig. 9, the feature quantities and the operating conditions for the steel material production process are input) and outputs a probability distribution of the material property values ​​of the steel material. Based on the probability distribution of the material property values ​​of the steel product output from this material property value prediction model MM and the specifications of the material property values ​​of the steel product included in the order information (for example, the required upper and lower limits of the material property values), the ratio of the area within the range from the required lower limit to the required upper limit (the area of ​​the hatched region shown in Figure 9(a)) to the area of ​​the entire probability distribution of the material property values ​​can be calculated as the pass probability. As shown in Figure 9(b), the operating condition determination device of the second embodiment inputs the operational performance of the target steel in the previous process after the previous process has been executed and the operating conditions in the subsequent process (i.e., the subsequent process before execution) of the target steel to be determined into the phenomenological model PM, and then inputs the feature quantities output from the phenomenological model PM (in the example shown in Figure 9, these feature quantities, the operational performance of the target steel in the previous process and the operating conditions in the subsequent process) into the material property value prediction model MM.The feature quantities output from the phenomenological model PM are input into the material property value prediction model MM.The feature quantities are then output using the probability distribution of the material property values ​​of the target steel, and the specifications of the material property values ​​of the target steel included in the order information (e.g., the required upper and lower limits of the material property values).The operating condition determination device determines the operating conditions in the subsequent process of the target steel by inversely analyzing the phenomenological model PM and the material property value prediction model MM (solving an optimization problem) so that the pass probability (the area of ​​the hatched area shown in Figure 9(b) relative to the area of ​​the entire probability distribution) is greater than or equal to a predetermined threshold value. Hereinafter, regarding the specific configuration of the operating condition determination device according to the second embodiment, the upstream processes are the steelmaking process, the hot rolling process, and the cold rolling process, and the downstream process is the annealing process (continuous annealing process). The operating conditions in the downstream process are the heating temperature (heating plate temperature) of the target steel material, the soaking temperature (soaking plate temperature), the holding temperature x OAtemp and retention time x OAtime The following describes an example in which the above-mentioned case is determined.

[0050] FIG. 10 is a block diagram showing a schematic configuration of an operating condition determination apparatus according to a second embodiment. FIG. 11 is a flowchart showing a schematic procedure of an operating condition determination method executed using the operating condition determination apparatus 200 shown in FIG. 10. As shown in FIG. 10, the operating condition determination apparatus 200 according to the second embodiment includes a pre-process operation performance acquisition unit 10A, a phenomenological model storage unit 20, a phenomenological model acquisition unit 30, a material property value prediction model storage unit 40, a material property value prediction model acquisition unit 50, an order information acquisition unit 60A, a constraint condition setting unit 70A, a post-process operating condition determination unit 80, and a post-process operating condition output unit 90. Like the material property prediction apparatus 100 according to the first embodiment, the operating condition determination apparatus 200 is configured, for example, by a computer, and performs various calculations by executing one or more programs stored in memory by one or more hardware processors. As a result, the operating condition determination device 200 functions as a front-end operation performance acquisition unit 10A, a phenomenological model storage unit 20, a phenomenological model acquisition unit 30, a material characteristic value prediction model storage unit 40, a material characteristic value prediction model acquisition unit 50, an order information acquisition unit 60A, a constraint condition setting unit 70A, a rear-end operation condition determination unit 80, and a rear-end operation condition output unit 90. Of the various units 10A to 90 included in the operating condition determination device 200, the configurations and operations (steps ST23 and ST24 shown in FIG. 11) of the phenomenological model storage unit 20, the phenomenological model acquisition unit 30, the material property value prediction model storage unit 40, and the material property value prediction model acquisition unit 50 are the same as those of the phenomenological model storage unit 20, the phenomenological model acquisition unit 30, the material property value prediction model storage unit 40, and the material property value prediction model acquisition unit 50 included in the material property prediction device 100 according to the first embodiment, and therefore will not be described here. The order information acquisition unit 60A, the previous process operation result acquisition unit 10A, the constraint condition setting unit 70A, the next process operation condition determination unit 80, and the next process operation condition output unit 90 will be described below in order.

[0051] <Order information acquisition unit 60A> The order information acquisition unit 60A executes step ST21 shown in Fig. 11. Specifically, the order information acquisition unit 60A acquires, as order information, the product size of the target steel material and the specifications of the material characteristic values ​​of the target steel material from a predetermined database (not shown). The order information acquisition unit 60A of the second embodiment acquires, as the product size of the target steel material, the product thickness x t is set as the required upper limit value y of the material characteristic value (yield point (YP) and tensile strength (TS) in the second embodiment) of the target steel material. YP upper , y TS upper or the required lower limit y YP lower , y TS lower Or get both.

[0052] <Pre-process operation record acquisition unit 10A> The upstream process operation record acquisition unit 10A executes step ST22 shown in Fig. 11. Specifically, the upstream process operation record acquisition unit 10A acquires, from a predetermined database (not shown), operation records of the upstream process of the target steel material, which are obtained after the upstream process of the target steel material has been executed. Specifically, the upstream process operation record acquisition unit 10A of the second embodiment acquires, as operation records of the upstream process, the product thickness included in the product size of the target steel material (the size of the target steel material after the cold rolling process), the component value x of carbon (C) in the steelmaking process, C , manganese (Mn) component value x MN , silicon (Si) component value x SI Operational results such as the coiling temperature x in the hot rolling process CT Obtain operational results such as:

[0053] 12 is a diagram showing an example of order information acquired by the order information acquiring unit 60A and operation results in the previous process acquired by the previous process operation results acquiring unit 10A. In the example shown in FIG. 12, the order information, which is the product thickness x t However, it is also used as the operational performance of the upstream process (operational performance in the cold rolling process) as it is. This is because in many cases the thickness of the target steel material after cold rolling is manufactured without any error from the product thickness in the order information. However, this is not limited to this, and it is also possible to actually measure the thickness of the target steel material after cold rolling using a known thickness gauge, and use the measured value as the operational performance of the upstream process.

[0054] <Constraint condition setting unit 70A> The phenomenological model acquisition unit 30 performs step ST23 shown in FIG. 11 (the phenomenological model PM(z1=g1(x C ,x MN ,x SI ) and z2=g2(x OAtemp ,x OAtime 11 (acquisition of the material characteristic value prediction model MM stored in the material characteristic value prediction model storage unit 40), and the material characteristic value prediction model acquisition unit 50 executes step ST24 shown in FIG. 11 (acquisition of the material characteristic value prediction model MM stored in the material characteristic value prediction model storage unit 40), and then the constraint condition setting unit 70A executes step ST25 shown in FIG. 11. Specifically, the constraint condition setting unit 70A sets constraint conditions for the inverse analysis to be executed by the later-described post-process operating condition determination unit 80 as needed. The constraint condition setting unit 70A of the second embodiment sets parameters constituting the operating conditions in the post-process to be determined (in the second embodiment, the heating temperature x of the target steel material in the annealing process) HF and soaking temperature x S ) is set as necessary. The constraint setting unit 70A may also set constraints between parameters. The constraint condition setting unit 70A of the second embodiment may set, for example, the heating temperature x HF The upper limit of the heating temperature x HF and soaking temperature x SThe constraint is set that the difference between the temperature and the soaking temperature must be 10°C or less. This is because the heating and soaking processes that make up the annealing process are carried out in adjacent equipment, and it is impossible to issue operational instructions that would cause a sudden change in strip temperature between the two processes.

[0055] <Post-process operation condition determination unit 80> The downstream process operational condition determination unit 80 executes step ST26 shown in Fig. 11. Specifically, as outlined with reference to Fig. 9, the downstream process operational condition determination unit 80 determines operational conditions (heating temperature x HF , soaking temperature x S , holding temperature x OAtemp and retention time x OAtime ) is determined. The following will specifically explain the content of the inverse analysis executed by the post-process operating condition determining unit 80.

[0056] If a model expressed by a composite function of the function expressed by the phenomenological model PM and the function expressed by the material property value prediction model MM is called a composite model, this composite model is expressed by the following equations (16) to (18). org =[x t ,x C ,x MN ,x SI ,x CT ,x HF ,x S ,x OAtemp ,x OAtime ], the probability distribution y of the material property value S∈{YP,TS} of the target steel material is obtained. S (Average value μ S '(x org ) and standard deviation σ S '(x org )) will be output as the prediction result from the synthesis model.

number

[0057] Therefore, the post-process operation condition determination unit 80 fixes the operation results of the previous process as the operation conditions, and then determines the operation conditions x that maximize the pass probability. org That is, the post-process operating condition determination unit 80 searches for and determines the operating conditions u=[x HF ,x S ,x OAtemp ,x OAtime ] (determine the operating conditions u that maximize the pass probability P(u)).

number

[0058] <Post-process operation condition output unit 90> The post-process operating condition output unit 90 executes step ST27 shown in Fig. 11. Specifically, the post-process operating condition output unit 90 outputs the operating conditions (heating temperature x HF , soaking temperature x S , holding temperature x OAtemp and retention time x OAtime The post-process operating condition output unit 90 not only outputs the determined operating conditions for the post-process, but also preferably outputs the probability distribution (average value, standard deviation) of the material characteristic values ​​output from the composite model when the determined operating conditions (including the operational results for the pre-process) are input to the composite model, and the pass probability of the material characteristic values ​​calculated from this probability distribution and the order information. 13 is a diagram showing an example of the contents output from the post-process operating condition output unit 90. In FIG. 13, for the target steel material No. 001, the operating results in the previous process shown in FIG. 12 and the heating temperature x determined by the post-process operating condition determination unit 80 are output. HF =795℃, soaking temperature x S =825℃, holding temperature x OAtemp =297℃, holding time x OAtime This shows that when manufactured under the condition of =196 seconds, the pass probability for yield point (YP) and tensile strength (TS) is a high value of 0.998.

[0059] The material property value prediction model MM included in the operating condition determination device 200 according to the second embodiment described above outputs a probability distribution of material property values ​​of a steel material, in other words, the expected value and variability of material property values, as prediction results. Therefore, even when the feature values ​​output by the phenomenological model PM are input into the material property value prediction model MM for operating conditions in the manufacturing process of the target steel material that have no past operating history, the material property values ​​(especially the variability) can be predicted with high accuracy. Furthermore, the prediction accuracy of material property values ​​can be expected to be improved even for operating conditions of the target steel material that are difficult to control and prone to variability in material property values. Furthermore, even when the operating conditions of the target steel material are outside the range of past operating history, the feature values ​​input into the material property value prediction model MM (the feature values ​​output from the phenomenological model PM) often fall within the range of feature values ​​obtained from past operating history. This also contributes to the improvement of prediction accuracy of material property values. Furthermore, the phenomenological model PM provided in the operating condition determination device 200 according to the second embodiment is a metallurgically formulated model that outputs feature values ​​that quantify changes in the structure of a steel material due to metallurgical phenomena during a manufacturing process characterized by operating conditions. Therefore, as in the case shown in FIG. 8 described above, for example, by visualizing the relationship between feature values ​​such as the critical cooling rate z1 and the tempering parameter z2 output from the phenomenological model PM and the prediction results of the material property values ​​output from the material property value prediction model MM, it is possible to advantageously interpret the prediction results.

[0060] [Third embodiment] The operating condition determination device according to the third embodiment is a modified version of the operating condition determination device 200 according to the second embodiment, and has the same configuration and operation as the operating condition determination device 200, except that it does not include the upstream process operation record acquisition unit 10A shown in Fig. 10 and does not execute step ST22 shown in Fig. 11. Unlike the operating condition determination device 200, the operating condition determination device according to the third embodiment is a device that determines operating conditions in all manufacturing processes of a target steel material (steelmaking process, hot rolling process, cold rolling process, and annealing process), and is used in quality design work. Specifically, the operating condition determination device 200 fixes the operational results of the upstream processes (steelmaking process, hot rolling process, and cold rolling process) as operating conditions, and then searches for and determines the operating conditions that maximize the pass probability. However, the operating condition determination device according to the third embodiment does not fix the operating conditions of the upstream processes, but instead searches for and determines the operating conditions for all manufacturing processes that maximize the pass probability by performing an inverse analysis (solving an optimization problem) on the phenomenological model PM and the material property value prediction model MM. In the operating condition determination device according to the third embodiment, the downstream process operating condition determination unit 80 shown in FIG. 10 determines the operating conditions for not only the downstream process but also the upstream process, and the downstream process operating condition output unit 90 outputs the operating conditions for not only the downstream process but also the upstream process. Except for the above differences, the operating condition determination device according to the third embodiment has the same configuration and operation as the operating condition determination device 200 according to the second embodiment, and therefore a detailed description thereof will be omitted here.

[0061] In each embodiment, a phenomenological model that outputs a critical cooling rate and a phenomenological model that outputs tempering parameters have been described as examples of metallurgically formulated phenomenological models that take operating conditions in a steel manufacturing process as input and output feature quantities that quantify changes in the structure of the steel due to metallurgical phenomena during the manufacturing process characterized by the operating conditions.However, as mentioned above, the phenomenological models used in the present invention are not limited to these. For example, it is possible to adopt a phenomenological model that outputs any one of the following (1) to (28) as a feature. (1) Amount of inclusions: An index for evaluating the amount of oxides, sulfides, and nitrides (2) Segregation: An index for evaluating the distribution of component values ​​during solidification (3) Solidification temperature: an index for evaluating segregation during solidification (4) Homogenization: an index for evaluating segregation relaxation (5) Reduction ratio: An index for evaluating the amount of strain (6) Dynamic recrystallization: An index for evaluating the structure during hot working (7) Static recrystallization: an index for evaluating the structure after processing (8) Texture: An index for evaluating crystalline anisotropy (9) Grain growth rate: an index for evaluating grain size (10) Work hardening: An index for evaluating the amount of rolling strain (11) Precipitation rate: An index for evaluating the precipitation behavior of alloys. (12) Amount of precipitates: an index for evaluating the amount of precipitation strengthening (13) Spheroidization ratio (spheroidization ratio of cementite): an index for evaluating the amount of precipitation strengthening (14) Lamellar spacing: an index for evaluating the hardness of pearlite (15) Decarburization, carburization, and nitriding: Indicators for evaluating surface carbides (16) Internal oxide thickness: an index for evaluating surface hardness (17) Strain aging: an index for evaluating aging properties (18) Secondary hardening amount: Secondary hardening to evaluate the amount of precipitation during tempering (19) Transformation driving force: an index for evaluating the phase transformation rate (20) Transformation temperature: An index for evaluating the hardness of a phase (21) Age hardening: an index for evaluating the amount of solute carbon and nitrogen after coiling (22) Dislocation density: an index for evaluating the hardness of a phase (23) Grain size: an index for evaluating yield strength (24) Residual stress: An index for evaluating internal stress (25) Twin and grain boundary orientation: Indicators for evaluating the strength of grain boundaries (26) Stability: An index for evaluating the amount of retained austenite (27) Packet and block particle sizes: Indicators for evaluating the hardness of the hard phase (28) Amount of reverse transformation: An index for evaluating the structure during annealing FIG. 14 is a diagram summarizing the operating conditions input to various phenomenological models (phenomenological models that output the above (1) to (28) as feature quantities and phenomenological models that output the critical cooling rate or tempering parameters).

[0062] Each embodiment of the present invention has been described above in detail with reference to the drawings, but the specific configuration is not limited to this embodiment, and includes designs within the scope that do not deviate from the gist of the present invention. [Explanation of symbols]

[0063] 10. Operational Performance Acquisition Department 10A Front-end operation performance acquisition department 20. Phenomenological model storage 30. Phenomenological model acquisition section 40 Material property value prediction model storage section 50. Material property value prediction model acquisition section 60. Material property value prediction section 60A···Order Information Acquisition Unit 70 Material characteristic value output section 70A···Constraint condition setting section 80 Post-process operation condition determination section 90 Post-process operation condition output section 100. Material property prediction device 200 Operating condition determination device MM: Material property value prediction model PM: Phenomenological Model

Claims

1. A material property prediction device for predicting material property values ​​of steel materials, a phenomenological model formulated metallurgically, which inputs operational conditions in a manufacturing process of the steel material and outputs feature quantities that quantify changes in the structure of the steel material due to metallurgical phenomena during the manufacturing process characterized by the operational conditions; a material property value prediction model that receives at least the feature amount as an input and outputs an expected value or a probability distribution of the material property value of the steel material, inputting operational results in a manufacturing process of a target steel material into the phenomenological model, and inputting the feature values ​​output from the phenomenological model into the material property value prediction model, and outputting expected values ​​or probability distributions of the material property values ​​of the target steel material, thereby predicting the material property values ​​of the target steel material; Material property prediction device.

2. the material property value prediction model receives the feature amount and an operating condition in a manufacturing process of the steel material as input, and outputs an expected value or a probability distribution of the material property value of the steel material, 2. A material property prediction device as described in claim 1, which predicts the material property values ​​of the target steel material by inputting the operational performance in the manufacturing process of the target steel material into the phenomenological model, inputting the feature values ​​output from the phenomenological model and the operational performance in the manufacturing process of the target steel material into the material property value prediction model, and outputting an expected value or probability distribution of the material property values ​​of the target steel material.

3. An operating condition determination device that determines operating conditions in a steel manufacturing process according to order information, a phenomenological model formulated metallurgically, which inputs operational conditions in a manufacturing process of the steel material and outputs feature quantities that quantify changes in the structure of the steel material due to metallurgical phenomena during the manufacturing process characterized by the operational conditions; a material property value prediction model that receives at least the feature amount as an input and outputs a probability distribution of the material property value of the steel material, The operating conditions in the manufacturing process of the target steel material are input into the phenomenological model, and the feature values ​​output from the phenomenological model are input into the material property value prediction model, and the probability of passing the material property values ​​of the target steel material calculated using the probability distribution of the material property values ​​of the target steel material output by inputting the feature values ​​into the material property value prediction model and the specifications of the material property values ​​of the target steel material included in the order information is determined by reverse analysis of the phenomenological model and the material property value prediction model so that the operating conditions in the manufacturing process of the target steel material are equal to or greater than a predetermined threshold value. Operating condition determination device.

4. the material property value prediction model receives the feature amount and an operating condition in a manufacturing process of the steel material as input, and outputs a probability distribution of the material property value of the steel material; 4. An operating condition determination device as described in claim 3, wherein the operating conditions in the manufacturing process of the target steel are determined by inversely analyzing the phenomenological model and the material property value prediction model so that the pass probability of the material property values ​​of the target steel calculated by inputting the operating conditions in the manufacturing process of the target steel into the phenomenological model, the feature values ​​output from the phenomenological model, and the operating conditions in the manufacturing process of the target steel into the material property value prediction model, and the specifications of the material property values ​​of the target steel included in the order information, is greater than or equal to a predetermined threshold.

5. 4. An operating condition determination device according to claim 3, wherein the manufacturing process of the steel material has a pre-process and a post-process, and after the pre-process of the steel material is executed, the operating conditions of the post-process of the steel material are determined in accordance with the order information, The operational results of the target steel material in the preceding process and the operational conditions in the subsequent process are input into the phenomenological model, and the feature quantities output from the phenomenological model are input into the material property value prediction model, and the feature quantities are output from the phenomenological model, and the feature quantities are input into the material property value prediction model, and the probability distribution of the material property values ​​of the target steel material output from the phenomenological model and the material property value prediction model are input, and the specification of the material property values ​​of the target steel material included in the order information are used to calculate the probability of acceptance of the material property values ​​of the target steel material, and the operating conditions in the subsequent process of the target steel material are determined by reverse analysis of the phenomenological model and the material property value prediction model, so that the probability of acceptance of the material property values ​​of the target steel material is equal to or greater than a predetermined threshold value. Operating condition determination device.

6. 5. An operating condition determination device according to claim 4, wherein the manufacturing process of the steel material has a pre-process and a post-process, and after the pre-process of the steel material is executed, the operating conditions of the post-process of the steel material are determined in accordance with the order information, the feature quantities output from the phenomenological model by inputting the operational performance of the target steel material in the upstream process and the operational conditions in the downstream process into the phenomenological model, the operational performance of the target steel material in the upstream process and the operational conditions in the downstream process into the material property value prediction model, the probability distribution of the material property values ​​of the target steel material output by inputting the operational performance of the target steel material in the upstream process and the operational conditions in the downstream process into the material property value prediction model, and the specifications of the material property values ​​of the target steel material included in the order information, are used to determine the operational conditions in the downstream process of the target steel material by reverse-analyzing the phenomenological model and the material property value prediction model so that the pass probability of the material property values ​​of the target steel material is equal to or greater than a predetermined threshold value; Operating condition determination device.

7. determining the operating conditions for all manufacturing processes of the target steel material; 5. The operating condition determining device according to claim 3 or 4.

8. A material property prediction method for predicting a material property value of a steel material, comprising: a phenomenological model formulated metallurgically, which inputs operational conditions in a manufacturing process of the steel material and outputs feature quantities that quantify changes in the structure of the steel material due to metallurgical phenomena during the manufacturing process characterized by the operational conditions; a material property value prediction model that receives at least the feature amount as an input and outputs an expected value or a probability distribution of the material property value of the steel material, inputting operational results in a manufacturing process of a target steel material into the phenomenological model, and inputting the feature values ​​output from the phenomenological model into the material property value prediction model, and outputting expected values ​​or probability distributions of the material property values ​​of the target steel material, thereby predicting the material property values ​​of the target steel material; Material property prediction methods.

9. An operating condition determination method for determining operating conditions in a steel manufacturing process according to order information, comprising: a phenomenological model formulated metallurgically, which inputs operational conditions in a manufacturing process of the steel material and outputs feature quantities that quantify changes in the structure of the steel material due to metallurgical phenomena during the manufacturing process characterized by the operational conditions; a material property value prediction model that receives at least the feature amount as an input and outputs a probability distribution of the material property value of the steel material, The operating conditions in the manufacturing process of the target steel material are input into the phenomenological model, and the feature values ​​output from the phenomenological model are input into the material property value prediction model, and the probability of passing the material property values ​​of the target steel material calculated using the probability distribution of the material property values ​​of the target steel material output by inputting the feature values ​​into the material property value prediction model and the specifications of the material property values ​​of the target steel material included in the order information is determined by reverse analysis of the phenomenological model and the material property value prediction model so that the operating conditions in the manufacturing process of the target steel material are equal to or greater than a predetermined threshold value. Method for determining operating conditions.

10. A program for causing a computer to function as the material property prediction device according to claim 1 or 2.

11. A program for causing a computer to function as the operating condition determining device according to claim 3 or 4.

Images