Quality factor analysis device and steelworks operation method

Abstract

To provide a quality factor analysis device capable of highly accurately estimating a quality factor and an iron mill operation method.SOLUTION: A quality factor analysis device (10) comprises: a database (110) which stores quality factors in products manufactured in the past and manufacturing data on operation factors and intermediate factors; an initialization unit (106) which calculates an initial probability distribution representing a value range of the quality factors, operation factors, and intermediate factors from the manufacturing data; a condition reading unit (107) which reads conditions determined by measurement or setting for some factors; a casual inference unit (108) which calculates a conditional probability on the basis of the determined conditions and a relation between the initial probability distribution and the factors and updates a probability distribution of factors on the basis of the conditional probability; and a display control unit (109) which displays the initial probability distribution and the updated probability distribution about at least one factor.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a quality factor analysis device and a method for operating a steel mill. [Background technology]

[0002] Conventionally, in manufacturing processes in which operational conditions affect quality, various methods have been proposed for predicting product quality from operational conditions, etc. For example, Patent Document 1 discloses an apparatus for predicting product quality from operational data, which predicts quality with high accuracy by having a functional formula that expresses the relationship between operation and quality for each local region of operational variables.

[0003] Furthermore, for example, Patent Document 2 discloses a method for supporting estimation of the cause of a quality abnormality, in which the correlation between a quality evaluation index and a manufacturing process, and numerical or textual information on temperature, pressure, and water volume selected by decision tree learning are displayed to the operator, thereby enabling the operator to estimate the cause of the quality abnormality regardless of the operator's level of experience.

[0004] Furthermore, for example, Patent Document 3 discloses a method for estimating an abnormal process for multiple manufacturing processes, which generates a correlation model for the features in a normal state for each process by machine learning, and determines the deviation from the normal state as the degree of abnormality, thereby enabling accurate estimation of an abnormal process. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2012-137813 A [Patent Document 2] JP 2014-179060 A [Patent Document 3] JP 2019-49940 A Summary of the Invention [Problem to be solved by the invention]

[0006] Here, in order to identify the cause of a quality abnormality, the above-mentioned conventional method estimates the cause by focusing on the relationship based on the correlation between quality and manufacturing conditions from product quality data and manufacturing condition data. However, a correlation does not necessarily mean that there is a causal relationship.

[0007] For example, when there is confounding between manufacturing conditions and quality, multicollinearity may occur between variables related to manufacturing conditions and quality, resulting in spurious correlation, and thus conventional methods may assume that a certain variable is a possible cause even when there is no causal relationship. Here, "there is confounding between manufacturing conditions and quality" means that there is an external variable that affects both the manufacturing condition variables and the quality variables. In such cases, even if the manufacturing conditions assumed to be the cause of the quality abnormality by conventional methods are changed, the quality defect or abnormal state may not improve or may worsen.

[0008] In order to solve the above problems, the present disclosure aims to provide a quality factor analysis device and a method for operating a steelworks that can estimate quality factors with high accuracy. Here, the quality factors are factors that cause quality changes such as quality abnormalities and quality improvements. [Means for solving the problem]

[0009] A quality factor analysis device according to an embodiment of the present disclosure includes: 1. A quality factor analysis device that analyzes factors of a quality characteristic using factors including a quality factor representing a quality characteristic of a product, an operation factor representing an operation amount of a manufacturing process for obtaining the quality characteristic, and an intermediate factor representing an intermediate factor that may affect a relationship between the operation amount and the quality characteristic, a database that accumulates manufacturing data relating to the quality factors, the operation factors, and the intermediate factors of the products that have been manufactured in the past; an initialization unit that calculates an initial probability distribution representing a range of values ​​of the quality factor, the operation factor, and the intermediate factor from the manufacturing data; a condition reading unit that reads conditions determined by measurement or setting for some of the factors; a causal inference unit that calculates a conditional probability based on the determined condition, the initial probability distribution, and a relationship between the factors, and updates the probability distribution of the factors based on the conditional probability; and a display control unit that displays the initial probability distribution and the updated probability distribution for at least one of the factors.

[0010] A method for operating a steel mill according to an embodiment of the present disclosure includes: The method includes determining manufacturing conditions for the steel product based on the probability distribution of the operational factors updated by the quality factor analysis device. Effect of the Invention

[0011] According to the present disclosure, it is possible to provide a quality factor analysis device and a steelworks operation method capable of estimating quality factors with high accuracy. [Brief description of the drawings]

[0012] [Figure 1] FIG. 1 is a diagram illustrating an outline of a quality factor analysis device according to an embodiment of the present disclosure. [Diagram 2] FIG. 2 is a flowchart showing the processing of the arithmetic processing unit. [Diagram 3] FIG. 3 is a flowchart showing the initialization process. [Figure 4] Figure 4 is an example of manufacturing data stored in the database. [Diagram 5] FIG. 5 is a diagram illustrating a causal connection table. [Figure 6] FIG. 6 is a diagram illustrating a causal diagram. [Figure 7] FIG. 7 is a diagram showing the probability distribution of factors. [Figure 8] FIG. 8 is a diagram illustrating a conditional probability table. [Figure 9] FIG. 9 is a graph of experimental data in the examples. [Figure 10] FIG. 10 is a graph showing the results of the conventional method in the examples. [Figure 11] FIG. 11 is a causal connection table in the embodiment. [Figure 12] FIG. 12 is a diagram showing a causal diagram and a probability distribution in an embodiment. [Figure 13] FIG. 13 is a conditional probability table in the embodiment. [Figure 14] FIG. 14 is a diagram showing the analysis results of the quality factor analysis device in the embodiment. [Figure 15] FIG. 15 is a diagram showing a causal diagram in the embodiment. [Figure 16] FIG. 16 is a diagram showing a case where the defective rate is set to 100% in the embodiment. [Figure 17] FIG. 17 is a diagram showing a probability distribution of Mn (manganese concentration) in the examples. [Figure 18] FIG. 18 is a diagram showing a case where the defect rate is set to zero in the embodiment. [Figure 19] FIG. 19 is a diagram showing the probability distribution of Mn when the defective rate is zero in the example. [Figure 20] FIG. 20 is a graph showing the distribution of manufacturing costs. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Hereinafter, a quality factor analysis device and a steelworks operation method according to an embodiment of the present disclosure will be described with reference to the drawings.

[0014] The quality factor analysis device according to this embodiment analyzes factors of quality characteristics using factors including quality factors representing the quality characteristics of a product, operation factors representing the operation amounts of a manufacturing process for obtaining the quality characteristics, and intermediate factors representing intermediate factors that may affect the relationship between the operation amounts and the quality characteristics. Although the products targeted by the quality factor analysis device are not limited, in this embodiment, steel materials manufactured by a steelmaking process are taken as an example. The processing executed by the quality factor analysis device is performed as a part of the operation method of a steelworks.

[0015] (composition) Fig. 1 is a diagram showing an example of the configuration of a quality factor analysis device 10 according to this embodiment. In addition to the quality factor analysis device 10, Fig. 1 shows a user 40 of the quality factor analysis device 10, a display device 20 that displays analysis results and the like to the user 40, and an input device 30 that accepts input from the user 40.

[0016] The quality factor analysis device 10 includes an arithmetic processing unit 101 that performs various arithmetic processing for executing a quality factor analysis, a memory unit (ROM 103) that stores a quality factor analysis program 102, a temporary memory unit (RAM 104) that temporarily stores results of various arithmetic processing and calculated values, a bus 105, and a database 110 that accumulates manufacturing data (manufacturing performance data) related to factors in products manufactured in the past.

[0017] The arithmetic processing unit 101 reads and executes the quality factor analysis program 102 stored in the ROM 103, and executes various arithmetic processing required for the quality factor analysis. The arithmetic processing unit 101 is one or more processors. The processor is, for example, a general-purpose processor or a dedicated processor specialized for a specific process, but is not limited to these and can be any processor. For example, the quality factor analysis device 10 may be a computer, and the arithmetic processing unit 101 may be realized by a CPU.

[0018] The arithmetic processing unit 101 includes an initialization unit 106, a condition reading unit 107, a causal inference unit 108, and a display control unit 109 as functional blocks that function when the quality factor analysis program 102 is executed.

[0019] (process) 2 is a flowchart showing the processing of the arithmetic processing unit 101. In summary, the arithmetic processing unit 101 executes the processes of initialization (step S201), condition reading (step S202), causal inference (step S203), result display (step S204), and end determination (step S205). The initialization process (step S201) is executed by the initialization unit 106. The condition reading process (step S202) is executed by the condition reading unit 107. The causal inference process (step S203) is executed by the causal inference unit 108. The result display (step S204) and end determination (step S205) are executed by the display control unit 109. When the arithmetic processing unit 101 receives a start instruction from the user 40 via the input device 30, it starts the quality factor analysis process.

[0020] (initialization) The initialization unit 106 executes the process (steps S201-1 to S201-4) based on the flowchart shown in FIG. 3, and calculates an initial probability distribution representing the range of the factor values.

[0021] The initialization unit 106 executes reading of the manufacturing data (step S201-1). The initialization unit 106 extracts factors such as heating conditions, rolling conditions, cooling conditions, chemical components of the material, feature quantities of the microstructure of the steel material imaged and read by a scanning electron microscope or the like, material requirement values ​​of the steel material such as strength, and material characteristic values ​​obtained by tensile tests or the like for a large number of steel materials manufactured in the past that are stored in the database 110. Here, in this embodiment, the material is a slab. Furthermore, the manufacturing data, which is data on a large number of steel materials manufactured in the past, is managed as manufacturing data that summarizes factors for each product, as shown in FIG. 4. Rows represent products manufactured in the past. Furthermore, columns represent factors.

[0022] As described above, the factors include quality factors that represent the quality characteristics of the product, operation factors that represent the operation amount of the manufacturing process to obtain the quality characteristics, and intermediate factors that represent intermediate factors that can affect the relationship between the operation amount and the quality characteristics. For example, the material requirement value of the steel material, such as strength, represents the quality characteristics of the steel material, which is the product, and corresponds to the quality factor. Also, for example, the heating condition, rolling condition, cooling condition, and chemical composition of the material are manufacturing conditions of the steel material, and correspond to the operation factor or intermediate factor (factor other than the quality factor). In the quality factor analysis performed as a part of the operation method of the steelworks such as this embodiment, the quality factor analysis device 10 may use, as the quality factor, a factor that represents at least one of the strength, ductility, toughness, surface defects, inclusions, and internal quality of the steel material. Here, the internal quality is, for example, center segregation, porosity, etc. Also, the quality factor analysis device 10 may use, as the operation factor, at least one of the heating condition, rolling condition, cooling condition, and chemical composition of the material among the manufacturing conditions of the steel material. However, in the following, for convenience of explanation and illustration, factors and factor values ​​may be abstractly expressed using numbers (1, 2, 3 . . . ) and letters (A, B, C).

[0023] The initialization unit 106 reads a causal connection table expressing causal relationships between factors (step S201-2). The causal connection table is prepared in advance by, for example, the user 40 and stored in the database 110. FIG. 5 is a diagram illustrating a causal connection table. In the causal connection table, the rows represent causal factors and the columns represent result factors, and if a 1 is displayed in a cell where the rows and columns intersect, it means that there is a causal relationship. In the example of FIG. 5, there are causal relationships from factor A to factor B, from factor A to factor C, and from factor B to factor C. The initialization unit 106 may generate a causal diagram illustrating the causal relationships based on the read causal connection table and store it in the database 110. FIG. 6 is a diagram illustrating a causal diagram. In the example of FIG. 6, the causal diagram shows factors with causal relationships connected by arrows, with the cause factor at the base of the arrow and the result factor at the tip of the arrow. The causal diagram may be read from the database 110 by the display control unit 109 and displayed on the display device 20.

[0024] The initialization unit 106 calculates a histogram for each factor (step S201-3). The initialization unit 106 also calculates the probability distribution (occurrence probability) of each factor. FIG. 7 is a diagram showing the probability distribution of factors. For example, the frequencies and probability distributions of factors A, B, and C in FIG. 4 are each expressed in a table format as shown in FIG. 7. Here, the first column shows the value range of each factor, the second column shows the frequency, and the third column shows the probability distribution.

[0025] The initialization unit 106 acquires a conditional probability (step S201-4). FIG. 8 is a diagram illustrating a conditional probability table. The initialization unit 106 can execute the process of step S201-4 by generating a conditional probability table based on the manufacturing data of FIG. 4, the causal relationship expressed in FIG. 5 or FIG. 6, and the probability distribution of each factor expressed in FIG. 7. Here, the conditional probability expresses the probability distribution of a factor when a condition for another factor is satisfied (a certain event occurs). The conditional probability table of FIG. 8 shows the probability that factor B can take the value a, b, or c when factor A is a factor for which a certain event occurs, factor B is another factor, and the value of factor A is in a specific range. As shown in FIG. 5, a causal relationship exists from factor A to factor B. For example, when factor A is 1.2 or more and less than 1.4, the conditional probability that factor B's value is a is 0.5, and the conditional probability that factor B's value is b is 0.5. Furthermore, the initialization unit 106 may execute the process of step S201-4 by reading the conditional probability table from the database 110. In other words, the conditional probability table may not be generated by the initialization unit 106, but may be prepared in advance and stored in the database 110. In this case, the values ​​in the conditional probability table may be set based on scientific theories such as physics, rules of thumb obtained from experiments, rules of thumb obtained from people's daily operations, and the like.

[0026] (Condition reading) The condition reading unit 107 reads conditions that have been determined by measurement or setting for some of the factors. In this embodiment, the condition reading unit 107 reads observation information of the factors input from the input device 30 as the condition reading process (step S202 in FIG. 2). The observation information (determined information) is information about the determined value of a factor, for example, that the value of factor A is 1.0. The condition reading unit 107 reads observation information about one or more factors. It is not necessary for the condition reading unit 107 to read observation information about all factors. Here, when observation information about multiple factors is read, the processes from step S202 to step S204 are executed for each factor.

[0027] (causal inference) The causal inference unit 108 calculates a conditional probability based on the determined condition read by the condition reading unit 107, the initial probability distribution calculated by the initialization unit 106, and the relationship between the factors, and updates the probability distribution of the factors based on the conditional probability. In this embodiment, the causal inference unit 108 updates the probability distribution calculated in S201 by Bayesian inference according to the formula described below, based on the observation information read in step S202, as a process corresponding to step S203 in Fig. 2.

[0028] When a condition (observation information) determined by a setting is given for a certain factor (here, referred to as factor U), the causal inference unit 108 updates the probability distribution of a factor (here, referred to as factor S) that causes factor U by Bayesian inference according to the following formula (1). The histogram of the updated factor S means the inferred value of the value of factor S that causes factor U when factor U is as shown in the observation information.

[0029]

number

[0030] Here, P(S|U=u) is the posterior probability of factor S when the value of factor U is determined to be u. P(U=u|S) is the likelihood function of factors S and U. P(S) is the prior probability of factor S. Also, P(U=u) is the observational information of factor U.

[0031] Furthermore, when a condition (observation information) determined by a setting is given for a certain factor (here, referred to as factor U), the causal inference unit 108 updates the probability distribution of a factor (here, referred to as factor T) that is a result of factor U by Bayesian inference according to the following formula (2). The histogram of the updated factor T indicates the inferred value of the value of factor T that is a result of factor U when factor U is as indicated by the observation information.

[0032]

number

[0033] Here, P(T|U=u) is the posterior probability of factor T when the value of factor U is determined to be u. P(T|U) is the likelihood function of factors T and U. Also, P(U=u) is the observation information of factor U.

[0034] Here, the causal inference unit 108 can identify a factor S that is the cause of a factor U and a factor T that is the result of the factor U by acquiring the causal relationships shown in Fig. 5 or 6 from the database 110. The causal inference unit 108 updates the probability distribution for all of the causal factors S and result factors T for the factor U for which observation information is given, according to the above formulas (1) and (2). Here, the causal inference unit 108 updates the probability distribution recursively (by tracing all cause and result factors) when the causal relationships between factors are set in multiple stages.

[0035] (Result display) The display control unit 109 displays the initial probability distribution and the updated probability distribution for at least one of the factors. In this embodiment, the display control unit 109 displays the probability distribution of each factor updated in step S203, together with a causal diagram, on the display device 20 as a process corresponding to step S204 in Fig. 2. That is, in this embodiment, the display control unit 109 visually shows the relationship between the factors by displaying the causal diagram, allowing the user 40 to easily grasp the relationship between the factors.

[0036] (End judgment) The display control unit 109 ends a series of processes when an end instruction is received from the user 40 via the input device 30, as a branching process corresponding to step S205 in Fig. 2. If not, the process of step S202 in Fig. 2 is performed again. For example, when there is no end instruction from the user 40 and observation information on a plurality of factors is read, observation information on another factor is read in the process of step S202 again.

[0037] (Operation method) The quality factor analysis device 10 can analyze the relationship between quality factors that represent at least one of the mechanical properties (e.g., strength, etc.) and internal and surface properties (e.g., the presence or absence of flaws and defects, etc.) of steel, operation factors that are manufacturing conditions of the steel, and intermediate factors. The quality factor analysis device 10 may output the analysis result (probability distribution of the updated operation factors) to a higher-level system. The higher-level system is configured, for example, with a process computer that manages the manufacturing of steel, and executes the operation method of the steelworks. For example, when a desired result is given for the quality factor as observation information, the quality factor analysis device 10 can identify the causative operation factor and update the probability distribution. In order to manufacture steel of desired quality, the higher-level system may determine the manufacturing conditions of the steel based on the probability distribution of the operation factors updated by the quality factor analysis device 10.

[0038] In the following, in order to explain the effects of the quality factor analysis device 10 according to this embodiment, an example in which the device is applied to the production of thick steel plates will be described.

[0039] Example 1 The strength of steel products is determined by the precipitation form of carbonitrides, the type of solid phase, and other characteristics of the steel microstructure. The characteristics of the steel microstructure are affected by the amount of steel temperature change [℃] during the cooling process in manufacturing. Here, experiments were conducted to analyze the relationship between the amount of steel temperature change during cooling, the size of precipitates, and strength, using experimental data with various conditions as manufacturing data.

[0040] A series of sample numbers, which are consecutive identification numbers, were given to the 500 samples under various conditions. Figure 9 is a graph of experimental data, with the sample number on the horizontal axis and the factor value of the sample with that identification number on the vertical axis. "A" is the temperature change amount. "B" is the inverse of the precipitate radius. "X" is the strength of the steel. These values ​​are normalized to be in the range of -1 to 1. Samples that show a strength equal to or greater than the standard value (-0.12) are deemed to pass the quality test, i.e., "good." In the example of Figure 9, out of the total number of samples of 500, 149 samples fell below the standard value, resulting in a defect rate of 29.8%.

[0041] In the conventional correlation-based method, first, in order to grasp the influence of "A" and "B" on "X", which is the strength, a linear regression model is created with "X" as the objective variable and "A" and "B" as explanatory variables. Based on the data in Figure 7, the regression equation X = 0.78 x A + 1.03 x B is obtained. In the conventional method, since the minimum value of "X" is -0.67, the explanatory variables are adjusted to increase the pass rate (the rate of "good"). In this case, since the value obtained by dividing 1.0 by the regression coefficient of "A", 0.78, is 1.28, the target value of "A" is increased by 1.28. Figure 10 shows the result after the adjustment (the target value of "A" is increased by 1.28). The average value of "X" in Figure 10 is 0.04, which is almost unchanged from the average value of "X" before adjustment, 0.05, in Figure 9. Also, in Figure 10, there were still 108 samples below the standard value, resulting in a defective rate of 21.6%.

[0042] The reason why the conventional correlation-based method cannot improve the defect rate as intended is thought to be because it does not take into account the phenomenon that factor B is affected by factor A. In other words, it is thought that a change in factor A affects factor B, and the improvement in factor X (strength) due to the change in factor A is canceled out by the deterioration of factor X due to the affected factor B, resulting in no improvement in "X."

[0043] Factor analysis was performed by the quality factor analysis device 10 for the same example. The initialization unit 106 read the causal connection table shown in Fig. 11 and the data in Fig. 9, and calculated the causal diagram for factors A, B, X, and Y, the probability distribution of each factor, and a conditional probability table. Fig. 12 shows the causal diagram and probability distribution in this embodiment. Fig. 13 shows the conditional probability table in this embodiment. Here, factor Y is a factor representing a pass / fail judgment.

[0044] Next, the condition reading unit 107 acquires the observation information that "the defective rate in factor Y is 0%." The causal inference unit 108 updates the probability distribution of each factor when "the defective rate in factor Y is 0%" is the observation information based on Bayesian inference. The display control unit 109 displays the probability distribution before and after the update, as shown in FIG. 14. According to FIG. 14, it is visually shown that in order to make the defective rate 0%, not only is it necessary to increase the average value of factor A by 0.15, but it is also necessary to narrow the distribution width of factor A and to decrease the average value of factor B by 0.2. Here, "OK" in FIG. 14 corresponds to "good" and "NG" corresponds to "bad."

[0045] As described above, the quality factor analysis device 10 can estimate quality factors with higher accuracy than the conventional techniques.

[0046] Example 2 In this embodiment, the risk of defects in the strength and toughness of thick steel plates is taken as an example to explain a method of analyzing and improving the cause of defects in the manufacturing process. As shown in the causal diagram of FIG. 15, the factors in this embodiment may include the finishing width [mm] of the steel material, the finishing thickness [mm], and the thickness [mm] of the rolled material during controlled rolling as factors of rolling conditions. In addition, the factors in this embodiment may include the heating furnace extraction temperature, the controlled rolling temperature, the finishing temperature, and the cooling temperature as factors of temperature conditions. In addition, the factors in this embodiment may include carbon, silicon, manganese, phosphorus, sulfur, copper, nickel, chromium, molybdenum, niobium, vanadium, titanium, boron, aluminum, nitrogen, hydrogen, oxygen, and calcium as factors of the chemical component concentrations of the slab. In addition, the factors in this embodiment may include the amount of carbide precipitation, the amount of element solid solution in the iron structure, and the martensite fraction as metallurgical factors of the product. In addition, the factors in this embodiment may include, as material factors, transition temperature calculated by Charvy test, which means tensile strength and toughness, and tensile strength and transition temperature excluding test piece variation and test condition variation. In addition, the factors in this embodiment may include, as material requirement value factors, an upper limit value of tensile strength, a lower limit value of tensile strength, and a lower limit value of transition temperature. In addition, the factors in this embodiment may include, as defect risk factors, a risk of tensile strength failing the required value and a risk of transition temperature failing the required value. Here, in FIG. 15, the black squares show a simplified probability distribution of each factor.

[0047] In this embodiment, quality factors are estimated based on causal relationships in three stages: (1) defect risk diagnosis, (2) defect factor estimation, and (3) decision of operation actions to achieve zero defect rate. In each of these three stages, the processes of condition reading (step S202), causal inference (step S203), and result display (step S204) in the above embodiment are executed.

[0048] (Defect risk diagnosis) The quality defect risk diagnosis is to calculate the risk of the occurrence of defects when all the conditions of rolling conditions, temperature conditions, and chemical component concentrations of the slab are given. After the initialization process in the above embodiment, the conditions of each factor of rolling conditions, temperature conditions, and chemical component concentrations of the slab are input as a condition reading process. The value of the factor is not a single definite value, but is given as a distribution considering the operational variation. By the causal inference process in the above embodiment, metallurgical factors, material factors, material requirement value factors, and defect risk factors are inferred, and the probability distribution of each factor is updated. FIG. 15 shows the state in which the probability of rejection (NG) and the probability of acceptance (OK) are displayed by such inference.

[0049] Figure 15 shows that the final defect rate can be 26.626% (result of quality defect risk diagnosis) due to the rolling conditions, temperature conditions, and variations (probability distribution) in the chemical component concentrations of the slabs in the current operation.

[0050] (Estimation of the cause of the defect) In estimating the cause of defects, in order to identify the cause of defects, the defect rate is set to 100% as a condition reading process. Fig. 16 is a diagram showing a case where such a setting is made. The process of causal inference in the above embodiment updates the probability distribution of each factor based on the inputted condition. The process of displaying the results displays the probability distribution of each factor before and after the update.

[0051] FIG. 17 shows, as an example, the probability distribution of Mn (manganese concentration), which is a factor in the chemical component concentration of a slab. The horizontal axis is the Mn concentration normalized to -5.5 to +5.5. The vertical axis shows the probability of each Mn concentration. The probability distribution of the Mn amount in the graph representing the state before renewal has a peak at 0.5 to 1.5. On the other hand, the probability distribution of the Mn amount in the graph representing the state after renewal has a peak at -2.5 to -1.5. Therefore, it can be inferred that one of the causes of the defects is that the Mn amount is about 3 units low. Here, 1 unit corresponds to one interval of the normalized concentration on the horizontal axis.

[0052] (Zero defect rate operational action) In an operation action with a zero defect rate, the pass rate is set to 100%, in other words, the defect rate is set to zero. Fig. 18 is a diagram showing the case where such a setting is made. The process of causal inference in the above embodiment updates the probability distribution of each factor based on the input conditions. The process of displaying the results displays the probability distribution of each factor before and after the update.

[0053] FIG. 19 is a diagram showing the probability distribution of Mn when the defect rate is zero. The horizontal axis is the Mn concentration normalized to -5.5 to +5.5. The vertical axis shows the probability of each Mn concentration. The probability distribution of the Mn amount in the graph representing before the update has an average of -0.25 and a standard deviation of 2.15. On the other hand, the probability distribution of the Mn amount in the graph representing after the update has an average of 0.63 and a standard deviation of 1.69. From this, it can be seen that in order to reduce the defect rate, it is necessary to increase the target Mn amount by 0.89 (= 0.63 - (-0.25)) from the current amount and further narrow the range by 0.46 (= 2.15 - 1.69).

[0054] In this embodiment, the manufacturing cost distribution is updated based on the chemical component concentrations of the slab and displayed together with other factors. The manufacturing cost distribution is displayed, for example, as shown in FIG. 20. The horizontal axis represents normalized costs expressed as values ​​from 0 to 10, and the vertical axis represents probability. Comparing the graph before the update (using the values ​​on the left axis) with the graph after the update (using the values ​​on the right axis), the values ​​after the update are larger than those before the update. Therefore, it can be estimated that the narrowing of the Mn range may increase costs.

[0055] As described above, the quality factor analysis device 10 can estimate quality factors with high accuracy. In addition, the quality factor analysis device 10 can easily execute both estimation of defect factors and operation actions for zero defect rate by similar operations, and can also analyze fluctuations in costs.

[0056] Although the embodiments of the present disclosure have been described based on the drawings and examples, it should be noted that those skilled in the art can easily make various modifications or corrections based on the present disclosure. Therefore, it should be noted that these modifications or corrections are included in the scope of the present disclosure. For example, the functions included in each component or each step can be rearranged so as not to be logically inconsistent, and multiple components or steps can be combined into one or divided. The embodiments of the present disclosure can also be realized as a program executed by a processor included in the device or a storage medium on which a program is recorded. It should be understood that these are also included in the scope of the present disclosure. [Explanation of symbols]

[0057] 10 Quality factor analysis device 20 Display device 30 Input Devices 40 User 101 Processing unit 102 Quality Factor Analysis Program 103 ROM 104 RAM 105 Bus 106 Initialization section 107 Condition reading section 108 Causal Reasoning Department 109 Display control section 110 Database

Claims

1. 1. A quality factor analysis device that analyzes factors of a quality characteristic using factors including quality factors that represent quality characteristics of a product, operation factors that represent operation amounts of a manufacturing process for obtaining the quality characteristics, and intermediate factors that represent intermediate factors that may affect a causal relationship between the operation amounts and the quality characteristics, comprising: a database for storing manufacturing data relating to the quality factors, the operation factors, and the intermediate factors in the products manufactured in the past, and a causal connection table expressing causal relationships between the factors; an initialization unit that reads the causal connection table and calculates an initial probability distribution representing a range of values ​​of the quality factors, the operation factors, and the intermediate factors from the manufacturing data; a condition reading unit that reads conditions determined by measurement or setting for some of the factors; a causal inference unit that calculates a conditional probability based on the determined condition, the initial probability distribution, and a causal relationship between the factors, and updates the probability distribution of the factors based on the conditional probability; a display control unit that displays the initial probability distribution and an updated probability distribution for at least one of the factors.

2. The quality factor analysis device according to claim 1 , wherein the display control unit visually displays the causal relationships between the factors.

3. The product is a steel material, 3. The quality factor analysis device according to claim 1 or 2, wherein the causal relationship between the factors is a causal relationship between the quality factor representing at least one of the mechanical properties and internal and surface properties of the steel material, the operational factor being a manufacturing condition of the steel material, and the intermediate factor.

4. The quality factor analysis device according to claim 3 , wherein the quality factor represents at least one of strength, ductility, toughness, surface defects, inclusions, and internal quality of the steel material.

5. 5. The quality factor analysis device according to claim 3, wherein the operational factor is at least one of heating conditions, rolling conditions, cooling conditions, and chemical components of the steel material, which are manufacturing conditions of the steel material.

6. A method for operating a steelworks, comprising determining manufacturing conditions for the steel product based on the probability distribution of the operation factors updated by the quality factor analysis device according to any one of claims 3 to 5.

Images