Quality prediction model generation method, metal material quality prediction method, metal material manufacturing method, metal material manufacturing condition presentation method, quality prediction model generation device, metal material quality prediction device, metal material manufacturing condition presentation device, and metal material manufacturing system

Abstract

The quality prediction model generation method comprises: an acquisition step (S1) for acquiring an explanatory variable selected from the manufacturing conditions of each step and a target variable, which is the state of a quality defect of the manufactured metal material; a storage step (S3) in which the explanatory variable and the target variable are stored as learning data in association with each other; calculation steps (S4-S6) for grouping the learning data and checking whether or not there is a significant difference in the state of the quality defect; a searching step (S7) of searching the group with the highest significance; and a generation step (S8) in which a quality prediction model is generated by machine learning from among the searched groups having the highest significance.

Description

Technical Field

[0001] This disclosure relates to a method for generating a quality prediction model, a method for predicting the quality of metallic materials, a method for manufacturing metallic materials, a method for indicating manufacturing conditions for metallic materials, a device for generating a quality prediction model, a device for predicting the quality of metallic materials, a device for indicating manufacturing conditions for metallic materials, and a manufacturing system for metallic materials. Background Technology

[0002] As a method for predicting quality for any required conditions, one known method is to calculate the distance between multiple past observation conditions stored in a performance database and the desired required conditions, and then use the calculated distances to predict the quality. For example, according to Patent Document 1, a method is known to calculate the weights of the observation data (performance data) based on the calculated distances, create a function that fits to the vicinity of the required conditions based on the calculated weights, and use the created function to predict the quality for the required conditions.

[0003] Patent Document 1: Japanese Patent No. 7207547

[0004] Patent Document 1's quality prediction method generates a quality prediction model that accumulates actual data in a correlated manner with the manufacturing conditions of each process and the quality of the metal material manufactured under those conditions for each specified range, predicting the quality of the metal material for each specified range. The specified range is determined, for example, by considering the cutting position of the metal material. While Patent Document 1's quality prediction method can predict the quality for any manufacturing conditions with high accuracy, further improvements in prediction accuracy are desired. For example, objectively dividing the data in a manner corresponding to changes in the quality of the metal material, and generating a quality prediction model in each division, would be expected to further improve prediction accuracy. Summary of the Invention

[0005] The purpose of this disclosure, made in view of this situation, is to provide a method for generating a quality prediction model, a method for predicting the quality of metallic materials, a method for manufacturing metallic materials, an apparatus for generating a quality prediction model, and an apparatus for predicting the quality of metallic materials, all capable of making high-precision predictions of the quality of metallic materials.

[0006] (1) A quality prediction model generation method according to one embodiment of the present disclosure is a method for generating a quality prediction model of a metal material manufactured through one or more processes, wherein the method includes:

[0007] The acquisition steps involve acquiring the explanatory variables selected from the manufacturing conditions of each process, and the state of quality defects of the aforementioned metal material being manufactured, i.e., the target variables.

[0008] The saving step involves establishing a connection between the above-described variables and the above-described target variables and saving them as learning data;

[0009] The calculation steps involve dividing the above-mentioned learning data into groups and testing whether there are significant differences in the state of the above-mentioned quality defects.

[0010] The search steps involve searching for the group with the highest significance; and

[0011] The generation step involves using machine learning to generate the aforementioned quality prediction model from the groups with the highest significance found in the search.

[0012] (2) As one embodiment of this disclosure, based on (1),

[0013] The above description indicates that the variables are obtained as numerical or categorical variables.

[0014] (3) As one embodiment of this disclosure, based on (2),

[0015] When the aforementioned explanatory variables are obtained as categorical variables, the above calculation steps perform numerical transformation by assigning different binary values ​​to the aforementioned categorical variables for classification, and calculate an index to evaluate the aforementioned significant differences based on the assigned values ​​and the status of the aforementioned quality defects.

[0016] (4) As one embodiment of this disclosure, based on (3),

[0017] The above calculation steps calculate the above indicators for each of the multiple groups.

[0018] (5) As one embodiment of this disclosure, based on any one of (1) to (4),

[0019] After performing the above search steps, the above calculation steps further group the searched groups according to the highest significance group and perform the above tests.

[0020] (6) As one embodiment of this disclosure, based on any one of (1) to (5),

[0021] The aforementioned machine learning methods include at least linear regression, local regression, principal component regression, PLS regression, neural networks, regression trees, random forests, LightGBM, and XGBoost.

[0022] (7) As one embodiment of this disclosure, based on any one of (1) to (6),

[0023] If the prediction accuracy of the above quality prediction model deviates from the specified range, or at specified intervals, the above quality prediction model shall be regenerated.

[0024] (8) The quality prediction method for metallic materials according to one embodiment of the present disclosure uses a quality prediction model generated by any one of the quality prediction model generation methods (1) to (7) to predict the state of quality defects of the above-mentioned metallic materials.

[0025] (9) In one embodiment of this disclosure, the method for manufacturing a metallic material predicts the state of quality defects of the metallic material using the metallic material quality prediction method of (8).

[0026] If a quality defect in the aforementioned metallic material is predicted, at least a portion of the aforementioned manufacturing conditions shall be modified.

[0027] (10) The manufacturing condition prompting method for metal materials according to one embodiment of the present disclosure predicts the state of quality defects of the metal material by the metal material quality prediction method of (8), and if the metal material quality defects are predicted to exist, searches for and prompts changes to at least a portion of the manufacturing conditions to eliminate the quality defects.

[0028] (11) The method for manufacturing a metal material according to one embodiment of the present disclosure manufactures a metal material based on the change of the manufacturing conditions indicated by the metal material manufacturing condition indication method of (10).

[0029] (12) One embodiment of the present disclosure relates to a quality prediction model generation apparatus that generates a quality prediction model for a metal material manufactured through one or more processes, wherein the apparatus comprises:

[0030] The acquisition department acquires the explanatory variables selected from the manufacturing conditions of each process, and the state of the quality defects of the aforementioned metal material being manufactured, i.e., the target variables.

[0031] The storage department saves the above-described variables and the above-described target variables as learning data by establishing a correlation between them;

[0032] The computing department divides the aforementioned learning data into groups and examines whether there are significant differences in the status of the aforementioned quality defects.

[0033] The search department focuses on the group with the highest search significance; and

[0034] The generation department generates the aforementioned quality prediction model using machine learning from the groups with the highest significance found in the search.

[0035] (13) In one embodiment of the present disclosure, the quality prediction device for metal materials uses a quality prediction model generated by the quality prediction model generation device of (12) to predict the state of quality defects of the metal materials.

[0036] (14) The manufacturing condition prompting device for metal materials according to one embodiment of the present disclosure prompts at least a portion of the changes in the manufacturing conditions of the metal material based on the presence or absence of quality defects of the metal material predicted by the metal material quality prediction device of (13).

[0037] (15) A manufacturing apparatus for manufacturing metal materials according to an embodiment of the present disclosure controls each process of manufacturing metal materials based on changes in the manufacturing conditions output from the output unit of the manufacturing condition prompting device for metal materials in (14).

[0038] According to this disclosure, a method for generating a quality prediction model, a method for predicting the quality of metallic materials, a method for manufacturing metallic materials, a device for generating a quality prediction model, and a device for predicting the quality of metallic materials are provided, all capable of providing high-precision predictions of the quality of metallic materials. Attached Figure Description

[0039] Figure 1 This is a block diagram illustrating a structural example of a quality prediction model generation apparatus according to one embodiment of the present disclosure.

[0040] Figure 2 This is a flowchart illustrating the process of a quality prediction model generation method according to one embodiment of this disclosure.

[0041] Figure 3 It is a diagram used to illustrate grouping.

[0042] Figure 4 It is a diagram representing the combination of two groups and the presence or absence of quality defects.

[0043] Figure 5 This diagram illustrates the casting process of a slab. Detailed Implementation

[0044] Hereinafter, with reference to the accompanying drawings, a method for generating a quality prediction model, a method for predicting the quality of metallic materials, a method for manufacturing metallic materials, a method for indicating manufacturing conditions for metallic materials, and a quality prediction model generating apparatus 10 (see accompanying drawings) are described in one embodiment of the present disclosure. Figure 1This embodiment describes a metal material quality prediction device, a metal material manufacturing condition prompting device, and a metal material manufacturing system. In summary, the quality prediction model generation method and the quality prediction model generation device 10 involved in this embodiment acquire manufacturing conditions and quality data for each process in manufacturing metal materials, examine the acquired data, and divide it into groups with significant differences in quality. Then, the quality prediction model generation method and the quality prediction model generation device 10 generate a quality prediction model for each group to predict quality. The metal material quality prediction method involved in this embodiment uses the quality prediction model generated by the quality prediction model generation method or the quality prediction model generation device 10 involved in this embodiment to predict the state of quality defects in the metal material. The state of quality defects is divided into a first state and a second state. For example, the first state can correspond to "present," "above a specified amount," or "distributed in a specified area," etc., and the second state can correspond to "absent," "less than a specified amount," or "distributed in a less than specified area," etc. In the following description, the first state is assumed to be "present" and the second state to be "absent." The "present or absent" of quality defects is an example of the "state" of quality defects, and "present or absent" can be replaced with "state." The metal material manufacturing method according to this embodiment is a method for manufacturing metal materials. The metal material quality prediction method according to this embodiment predicts the presence or absence of quality defects in the metal material. If a quality defect is predicted to exist, at least a portion of the manufacturing conditions are modified. The metal material quality prediction method according to this embodiment uses a quality prediction model generated by the quality prediction model generation method or the quality prediction model generation device 10 according to this embodiment to predict the presence or absence of quality defects in the metal material. Furthermore, the metal material manufacturing condition indication method according to this embodiment, based on the result of the presence or absence of quality defects predicted by the metal material quality prediction method or the quality prediction device according to this embodiment, indicates a modification to at least a portion of the metal material manufacturing conditions.

[0045] Figure 1 This is a block diagram illustrating a structural example of the quality prediction model generation apparatus 10 according to this embodiment. The quality prediction model generation apparatus 10 is used in the manufacturing process of metal materials. The metal materials are, for example, steel products, and may include semi-finished products such as slabs, and products such as steel plates manufactured by rolling slabs. The quality prediction model generation apparatus 10 generates a quality prediction model for the metal material manufactured through one or more processes. The quality prediction model generation apparatus 10 executes a quality prediction model generation method (see [reference]). Figure 2 Additionally, in Figure 1In the example, the quality prediction model generation device 10 also functions as a quality prediction device for metal materials, predicting the presence or absence of quality defects. The metal material quality prediction device executes a metal material quality prediction method (see reference). Figure 2 Here, the quality prediction model generation device 10 may only have the function of generating a quality prediction model. In this case, the quality prediction device for metallic materials may be configured as a device different from the quality prediction model generation device 10, capable of obtaining a quality prediction model from the quality prediction model generation device 10. Furthermore, in Figure 1 In this example, the quality prediction model generation device 10 also functions as a metal material manufacturing condition prompting device, which prompts for changes in at least a portion of the metal material's manufacturing conditions based on the presence or absence of quality defects in the metal material predicted by the metal material prediction method. The metal material manufacturing condition prompting device executes the metal material manufacturing condition prompting method. The manufacturing condition prompting device can be configured as a different device from the quality prediction model generation device 10.

[0046] like Figure 1 As shown, the quality prediction model generation apparatus 10 includes a communication unit 11, a storage unit 12, and a control unit 13. The control unit 13 includes an acquisition unit 131, a storage unit 132, a calculation unit 133, a search unit 134, a generation unit 135, a prediction unit 136, a manufacturing condition prompting unit 137, and an output unit 138. As a hardware structure, the quality prediction model generation apparatus 10 can be implemented using a general-purpose computer (information processing device) such as a personal computer or a workstation. The computer, for example, includes a processor such as a CPU (Central Processing Unit). In addition, the computer includes memory such as RAM (Random Access Memory) and ROM (Read Only Memory) as main components. Details regarding the constituent elements of the quality prediction model generation apparatus 10 will be described later.

[0047] The quality prediction model generation device 10 obtains quality data and operational data from the operational data server 20. The operational data includes the manufacturing conditions for each process in the production of metal materials. Figure 1 In this example, the quality prediction model generation device 10 obtains quality data and operational data stored in the performance database of the operational data server 20. The operational data server 20 can communicate with the quality prediction model generation device 10 via a network, for example, a computer that manages the manufacturing process. The network is, for example, the Internet. Furthermore, in this embodiment, the display unit 30 displays data from the quality prediction device for metal materials (in... Figure 1The example shows the prediction results output by the integrated quality prediction model generation device 10.

[0048] Here, the manufacturing conditions accumulated in the performance database can include at least one of the following: composition of the metal material, temperature, pressure, plate thickness, and plate throughput speed in each process. Additionally, the quality data accumulated in the performance database can include at least one of the following: material properties such as tensile strength, presence or absence of surface defects, presence or absence of internal defects, quantity of defects, and location of defects. Manufacturing conditions and quality data are obtained, for example, from sensors, manufacturing equipment, control equipment, or quality management devices installed in the factory, etc., as operational performance data, operational target values, analytical values, or quality judgment results. For example, the presence or absence of internal defects in a steel plate (an example of a metal material) can be obtained as quality data using an ultrasonic flaw detection device (an example of a sensor and quality management device). The performance database is a database obtained by establishing and database-izing the quality data of the manufactured metal material with the corresponding manufacturing conditions, operational performance data, and various analytical values ​​for each process.

[0049] Here, the quality prediction model generation device 10 is in Figure 1 In this example, the computer is different from the running data server 20, but it could also be the same computer as the running data server 20. The display unit 30 can be a device for operators or others performing tasks during the manufacturing process to obtain information. Alternatively, the display unit 30 can be implemented using the display of a terminal device such as a smartphone or tablet. If a quality defect in the metal material is predicted, a manufacturing method for the metal material with at least a change in manufacturing conditions can be executed based on the operator's instructions. The display unit 30 can be, for example, a display device such as a liquid crystal display (LCD) or an organic electroluminescence panel (EL panel).

[0050] The following details the components of the quality prediction model generation apparatus 10. The communication unit 11 is configured to include one or more communication modules connected to a network. The communication unit 11 may, for example, include communication modules corresponding to mobile communication standards such as 4G (4th Generation) and 5G (5th Generation). The communication unit 11 may, for example, include communication modules corresponding to wired or wireless LAN standards.

[0051] The storage unit 12 is one or more memories. The memory may be, for example, a semiconductor memory, a magnetic memory, or an optical memory, but is not limited to these, and can be any type of memory. The storage unit 12 may be built into the quality prediction model generation device 10, but it may also be configured to be accessed externally by the quality prediction model generation device 10 via any interface.

[0052] The storage unit 12 stores various data used in the various calculations performed by the control unit 13. Additionally, the storage unit 12 can store the results and intermediate data of the various calculations performed by the control unit 13. In this embodiment, the storage unit 12 stores learning data and the generated model.

[0053] The control unit 13 is one or more processors. The processor may be a general-purpose processor or a dedicated processor for a specific process, but is not limited to these; any processor can be used. The control unit 13 controls the overall operation of the quality prediction model generation device 10.

[0054] Here, the quality prediction model generation device 10 may have the following software structure. One or more programs for controlling the operation of the quality prediction model generation device 10 are stored in the storage unit 12. If the programs stored in the storage unit 12 are read by the processor of the control unit 13, the control unit 13 functions as the acquisition unit 131, the storage unit 132, the calculation unit 133, the search unit 134, the generation unit 135, and the prediction unit 136.

[0055] The acquisition unit 131 obtains manufacturing conditions and quality data from the performance database. Here, the performance database contains a large amount of quality data, such as time series data, including manufacturing conditions for each process and measured values ​​of the presence or absence of quality defects in the manufactured metal material under those conditions. The acquisition unit 131 can selectively obtain explanatory variables and target variables from the performance database that can be used as learning data for machine learning, as described later. In this embodiment, the acquisition unit 131 obtains explanatory variables selected from the manufacturing conditions of each process, and the presence or absence of quality defects in the manufactured metal material, i.e., the target variable.

[0056] The storage unit 132 stores the data acquired by the acquisition unit 131, i.e., the explanatory variables and the target variables, as learning data in a correlated manner. That is, the storage unit 132 correlates the manufacturing conditions of each process, which are obtained as explanatory variables, with quality data, including measurement values ​​of the presence or absence of quality defects in the metal material manufactured under those manufacturing conditions, and stores this data as learning data. The destination for storing the learning data could be, for example, the storage unit 12. Here, if the manufacturing conditions of each process involve a large amount of data acquired over a long period or at a high frequency, the storage unit 132 can perform interval filtering or conversion to representative values ​​such as averages or median values ​​to make it an appropriate amount of data before storing the learning data. Furthermore, as... Figure 3 As shown, the storage unit 132 can establish a correlation between "the manufacturing conditions of each process and the presence or absence of quality defects (data set)" for each predetermined range of the metal material. Figure 3 In the example, the learning data is associated with each specified range starting from the beginning of the metal material's transport direction, and is sequentially divided into data group 1, data group 2, data group 3, data group 4, ..., data group N. Here, N is an integer greater than 2. Additionally, for example, if the metal material is cut during manufacturing, the specified range can be determined based on the cutting position, etc., but is not limited to this method of determination.

[0057] The calculation unit 133 divides the learning data into groups and performs a test to determine whether there are significant differences in the presence or absence of quality defects (significant difference test). For example, the calculation unit 133... Figure 3 Similar to patterns 1 to M, the aforementioned data groups 1 to N are divided into group X and group Y. Here, M is an integer of 2 or higher. The calculation unit 133 performs multiple groupings and conducts a significant difference test for each of the multiple groupings. The calculation unit 133 can perform grouping on all combinations, or it can group only a portion of the combinations. For example, due to limitations such as upper limits on computation time, the calculation unit 133 can group the combinations that can be computed within the limits. In addition, based on manufacturing insights, the priority order of grouping can be determined in advance.

[0058] In this embodiment, the calculation unit 133 uses Fisher's exact probability test. However, the method for testing significant differences is not limited. Furthermore, in this embodiment, the calculation unit 133 uses the p-value as an indicator of significant difference. The p-value represents the probability that a combination exhibits an observed difference or greater among all possible combinations. Therefore, the smaller the p-value, the higher the significance (the higher the probability of a significant difference). Figure 4 As shown, let a, b, c, and d be the number of combinations corresponding to the presence or absence of two groups and quality defects, respectively. In this case, the P-value is calculated using the following formula. n is the total number of data points used in the training.

[0059] [Formula 1]

[0060]

[0061] Here, grouping is based on explanatory variables. Furthermore, the explanatory variables are obtained as either numerical or categorical variables. For example, if the explanatory variable used for grouping is a numerical variable, the calculation unit 133 can assign a group to group X if the explanatory variable is above a threshold, and assign it to group Y if the explanatory variable is below the threshold.

[0062] For example, when the explanatory variable used for grouping is obtained as a categorical variable, the calculation unit 133 can perform a numerical transformation to classify the categorical variable by assigning different binary values. Different binary values ​​can be, for example, 0 or 1; in this embodiment, 0 or 1 is used for explanation. Alternatively, one-hot encoding can be used as a numerical transformation, but it is not limited to this. Then, the calculation unit 133 can calculate an index (P-value) to evaluate significant differences based on the assigned 0 or 1 and the presence or absence of quality defects. That is, the calculation unit 133 sets the group X when the explanatory variable after numerical transformation is 1 (i.e., if it is a certain category), and sets it to group Y when it is 0 (i.e., if it is not a certain category).

[0063] Here, the explanatory variables are selected from the manufacturing conditions of each process. Furthermore, manufacturing conditions include many variables such as the composition of the metal materials and temperature in various processes. Therefore, the selection of explanatory variables for grouping is diverse. For example, if n explanatory variables are used for grouping, then for 2... n The p-values ​​for all patterns can be calculated, or p-values ​​can be calculated for a subset. Additionally, candidate explanatory variables for grouping can be identified in advance based on manufacturing insights.

[0064] The search unit 134 searches for the group with the highest significance. In this embodiment, the search unit 134 determines the group with the lowest p-value. Figure 3 In this example, the calculation unit 133 calculates the P-value for each of patterns 1 to M, and the search unit 134 determines the pattern with the lowest P-value (pattern 2 as an example) as the group with the highest significance. In this example, the search unit 134 determines that group X, which includes data groups 1 and 2, and group Y, which includes data groups 3 to N, have the most significant difference.

[0065] Here, the calculation unit 133 and the search unit 134 can perform multi-level grouping starting from the group with the lowest p-value. That is, after the search unit 134 finds the group with the highest significance, the calculation unit 133 can further group and perform a significance test (p-value calculation) on one of the groups according to that group. Figure 3 In the example, after the search unit 134 determines that pattern 2 is the group with the highest significance, the calculation unit 133 can further perform grouping and p-value calculations for group Y, which includes data groups 3 to N. Then, the search unit 134 can search for the group with the highest significance for group Y, for example, determining that group Y1, which includes data groups 3 to i, and group Y2, which includes data groups i to N, have the most significant difference. Here, i is an integer greater than or equal to 3. In addition, there is no limit to the number of times multi-level grouping (repetition count), but in order not to reduce the accuracy of the generated quality prediction model, the number of training data included in a group should not be too small.

[0066] Here, for multiple explanatory variables, explanatory variables with p-values ​​below a predetermined value (e.g., 1 / n relative to the total number of training data points n) can be designated as important variables and grouped using important variables. Alternatively, only one important variable that readily explains the relationship with product quality can be selected, or more than two important variables can be chosen. Furthermore, if the accuracy of predictions using previously generated quality prediction models is achieved, this accuracy can be considered when selecting important variables.

[0067] The generation unit 135 generates quality prediction models using machine learning from the groups with the highest significance found by the search unit 134. The machine learning method is not particularly limited; for example, it can be linear regression, local regression, principal component regression, PLS regression, neural networks, regression trees, random forests, LightGBM, or XGBoost. In other words, the machine learning methods that the generation unit 135 can use include at least linear regression, local regression, principal component regression, PLS regression, neural networks, regression trees, random forests, LightGBM, and XGBoost. For example, in the above example where the calculation unit 133 and the search unit 134 have performed multi-level grouping, the generation unit 135 uses the learning data from each of group X, group Y1, and group Y2 to generate three quality prediction models. The generation unit 135 stores the generated quality prediction models, for example, in the storage unit 12.

[0068] The prediction unit 136 uses the generated quality prediction model to predict the presence or absence of quality defects in the manufactured metal material under arbitrary manufacturing conditions. The prediction results are displayed, for example, on the display unit 30. If a quality defect in the metal material is predicted to exist, at least a portion of the manufacturing conditions is changed based on operator instructions to improve the quality of the manufactured metal material.

[0069] To maintain the accuracy of predicting the presence or absence of quality defects in metallic materials, the prediction unit 136 can output instructions to the calculation unit 133, the search unit 134, and the generation unit 135 to regenerate the quality prediction model for the metallic materials. Specifically, the prediction unit 136 can instruct the calculation unit 133 to re-implement the grouping of the learning data and re-implement the test for significant differences in the state of quality defects. The prediction unit 136 can instruct the search unit 134 to re-search for the group with the highest superiority and output instructions to the generation unit 135 to regenerate the quality prediction model using machine learning within the group with the highest superiority found. For example, if the accuracy of predicting the presence or absence of quality defects in the metallic materials deviates from a specified range (not within the specified range) or at specified intervals (e.g., every 3 months), the prediction unit 136 can output instructions to regenerate the quality prediction model for the metallic materials. Then, the regeneration of the quality prediction model can be implemented based on these instructions.

[0070] The manufacturing condition prompting unit 137, based on the presence or absence of quality defects in the metal material predicted by the prediction unit 136, searches the prediction results for at least a portion of the manufacturing conditions for each process that eliminate quality defects, and prompts the display unit 30 with changes to the searched manufacturing conditions. The manufacturing condition prompting unit 137 changes at least a portion of the operable manufacturing conditions for each process from their current values ​​to other values, while maintaining the current values ​​for other manufacturing conditions, as input as explanatory variables for the quality prediction model generated by the generation unit 135. Then, the manufacturing condition prompting unit 137 predicts the presence or absence of quality defects in the metal material. If the prediction result is "no" quality defects, the display unit 30 prompts the changed manufacturing conditions. If the prediction result is "yes" quality defects, the values ​​of the operable manufacturing conditions for each process are further changed to other values ​​until the prediction result of the quality prediction model becomes "no" quality defects. For example, optimization methods such as iterative methods can be used. The manufacturing condition prompting unit 137 can use the quality prediction model generated by the generation unit 135 through machine learning to calculate the variable importance (feature importance) of the target variable, i.e., the presence or absence of quality defects in the manufactured metal material. Then, based on the variable importance, the manufacturing condition prompting unit 137 can select the manufacturing conditions that contribute significantly to the target variable among the manufacturing conditions of each process, which are used as explanatory variables, as the manufacturing conditions to be changed. The manufacturing condition prompting unit 137 can apply mathematical methods such as SHAP (Shapley Additive exPlanations), Gini impurity, frequency, and gain (the reduction in the objective function based on the presence or absence of the feature quantity) to calculate the variable importance (feature importance).

[0071] Furthermore, the quality prediction model generation device 10 may include an output unit 138, which outputs (sends) via a network to a control computer that controls the manufacturing apparatus for each process of manufacturing metal materials, showing changes to the manufacturing conditions prompted by the manufacturing condition prompt unit 137. The control computer can control the manufacturing apparatus for each process of manufacturing metal materials based on the sent changes to the manufacturing conditions. The quality prediction model generation device 10 and the control computer constitute a metal material manufacturing system. Additionally, the control computer can accept modifications to the manufacturing conditions from the operator to control the manufacturing apparatus for each process.

[0072] The output unit 138 can display the changes in manufacturing conditions prompted by the manufacturing condition prompt unit 137 via a display device (such as a liquid crystal display) connected via a network.

[0073] Figure 2This is a flowchart illustrating the processing of the quality prediction model generation method executed by the quality prediction model generation apparatus 10 according to this embodiment. In this embodiment, the quality prediction model generation apparatus 10 also functions as a quality prediction device for metallic materials, predicting the presence or absence of quality defects in the metallic materials. Figure 2 This is also a flowchart illustrating the process of a method for predicting the quality of metallic materials. In this embodiment, the quality prediction model generation device 10 also functions as a device for indicating the manufacturing conditions of metallic materials, indicating changes in at least a portion of the manufacturing conditions of the metallic materials. Figure 2 This is also a flowchart illustrating the processing of a method for indicating manufacturing conditions for metallic materials. In this embodiment, the quality prediction model generation device 10 and the control computer for controlling the manufacturing apparatus for manufacturing metallic materials together constitute a manufacturing system for metallic materials. Figure 2 This is also a flowchart illustrating the manufacturing process of metallic materials. Here, the prediction of the presence or absence of quality defects in the metallic materials (step S9), the regeneration of the quality prediction model (step S10), the indication of manufacturing conditions for the metallic materials (step S11), and the manufacturing of the metallic materials (step S12) can be executed consecutively with the generation of the quality prediction model (step S8). Furthermore, steps S9 to S12 can be executed after a certain period of time following the generation of the quality prediction model.

[0074] The acquisition unit 131 acquires manufacturing conditions and quality data from the performance database (step S1). Step S1 corresponds to the acquisition step.

[0075] The storage unit 132 performs multicollinearity exclusion on the data acquired by the acquisition unit 131 (step S2), and saves the explanatory variables and target variables as learning data in a correlated manner (step S3). Multicollinearity exclusion can be performed using known methods. Alternatively, step S2 can be omitted. Step S3 corresponds to the storage step.

[0076] If the explanatory variable used for grouping is obtained as a categorical variable (yes in step S4), the calculation unit 133 performs numerical transformations such as one-hot encoding (step S5). If the explanatory variable used for grouping is obtained as a numerical variable (no in step S4), the calculation unit 133 proceeds to step S6. The calculation unit 133 groups the learning data and performs a significant difference test (step S6). Steps S4 to S6 correspond to the calculation steps.

[0077] Search unit 134 searches for the group with the highest significance (step S7). Step S7 corresponds to the search step.

[0078] The generation unit 135 generates a quality prediction model in the groups according to the searched groups (step S8). Step S8 corresponds to the generation step.

[0079] The prediction unit 136 uses the generated quality prediction model to predict whether there are quality defects in the metal material (step S9).

[0080] To maintain the accuracy of predicting the presence or absence of quality defects in metallic materials, the prediction unit 136 can output an instruction to regenerate the quality prediction model for metallic materials (step S10). For example, if the accuracy of predicting the presence or absence of quality defects in metallic materials deviates from a specified range (outside the specified range), or at predetermined intervals, the prediction unit 136 can output an instruction to regenerate the quality prediction model for metallic materials. If the quality prediction model for metallic materials is regenerated (yes in step S10), the process returns to step S6. If the quality prediction model for metallic materials is not regenerated (no in step S10), the process proceeds to step S11.

[0081] The manufacturing condition prompting unit 137 searches for at least a portion of the manufacturing conditions for each process that predict the presence or absence of quality defects in the metal material, such as "no quality defects," and prompts for changes to the found manufacturing conditions (step 11). The found changes to the manufacturing conditions are sent as manufacturing condition change instructions via the output unit 138 to the control computer of the manufacturing apparatus that controls each process of manufacturing the metal material, and are displayed via the display unit 30.

[0082] The metal material is manufactured based on the detected changes to the manufacturing conditions (step S12). The changes to the manufacturing conditions are sent to the control computer, and the metal material is manufactured according to the sent changes. The operator can set the changes to the manufacturing conditions on the control computer based on the display on the display unit 30.

[0083] The effects of this disclosure will be specifically described below based on the embodiments, but this disclosure is not limited to these embodiments.

[0084] In this embodiment, the method for predicting the presence or absence of defects in the structural steel plate is applied to the metal material quality prediction method described in the above-described embodiments. The structural steel plate in this embodiment is formed by... Figure 5 The intermediate product manufactured by the casting process (continuous casting process) shown is the slab.

[0085] exist Figure 5In the casting process shown, molten steel in a ladle (steel pot) is poured into a mold (casting mold) via an tundish. The molten steel in the mold solidifies from the surface through primary cooling, forming a casting sheet. The casting sheet, pulled from the mold, is then subjected to secondary cooling by cooling water beneath the mold, eventually solidifying completely. The casting sheet is then cut by cutting tools at predetermined lengths to form slabs. These slabs are rolled in units in the next process, undergoing further processing to become the final product. Here, each steel pot corresponding to the manufactured slabs is numbered from 1 to N from top to bottom according to the cutting sequence, determining the slab cutting order.

[0086] In this embodiment, the target variable (quality) for quality prediction is the presence or absence of defects based on ultrasonic testing of the product (structural steel plate, pressure vessel steel, marine structure steel). The variables (manufacturing conditions) include the chemical composition of the metal materials in the smelting process. Additionally, the variables include the temperature of the metal materials, casting speed, cooling conditions, and slab cutting sequence in the casting process. Furthermore, the variables include the temperature of the metal materials in the heating process, the temperature of the metal materials in the hot rolling process, and the temperature of the metal materials in the cooling process. Finally, the variables include the temperature of the metal materials in the cold rolling process and the cutting dimensions.

[0087] As described in the quality prediction model generation method of the above implementation, a quality prediction model, which serves as a machine learning model, is generated by obtaining data from a performance database. LightGBM (Light Gradient Boosting Machine) is used as the machine learning method. Furthermore, Fisher's exact probability test is used as the method for testing significant differences. For a continuous variable, any split point is determined by dividing the difference between the maximum and minimum values ​​of the variable ("maximum value - minimum value") into 25 parts and searching for the smallest p-value; the split point corresponding to the smallest p-value is used as the split point for that variable.

[0088] Furthermore, the slab cutting order is a categorical variable. Data corresponding to cutting order 1 is classified into data group 1, data corresponding to cutting orders 2 through N-1 are classified into data groups 2 through N-1 respectively, and data corresponding to the last (N) of the cutting order is classified into data group N. Each data group is classified using binarization with one-hot encoding. In this embodiment, the most significant difference is determined to be in the segmentation of the first group including data groups 1 and 2, the second group including data groups N and N-1, and the third group including the other data groups.

[0089] For each of the aforementioned groups, a quality prediction model was generated using LightGBM. Furthermore, these quality prediction models were used to predict the presence or absence of quality defects. Here, the sample size in the performance database was 307, and the number of variables was 260. Regarding the prediction results, the correct answer rate for "quality defects exist" based on this method was 89%. The correct answer rate for "quality defects exist" based on the existing method was 68%, confirming a significant increase in the correct answer rate. Here, the existing method does not perform grouping using a significant difference test, but instead uses all training data to generate a single quality prediction model.

[0090] As described above, the quality prediction model generation method, the metal material quality prediction method, the metal material manufacturing method, the quality prediction model generation apparatus 10, the metal material quality prediction device, and the metal material manufacturing system involved in this embodiment perform appropriate grouping based on significant differences during the generation of the quality prediction model. Therefore, by using the generated quality prediction model, it is possible to predict the quality of the metal material with high accuracy.

[0091] In the model generation of this embodiment, the operating conditions during each process of manufacturing a previously manufactured product and the quality of the product manufactured under those operating conditions are obtained as performance data, and the corresponding operating condition data and quality data are stored in an associated manner. Based on this, multiple explanatory variables are grouped, and the group with the highest significance relative to the quality data is determined. Previously, such grouping was not performed. With the method of this disclosure, even operators who do not have deep knowledge or sufficient experience related to the manufacturing process can perform objectively meaningful grouping. The grouping based on the method of this disclosure is performed through a significance difference test, dividing the manufacturing conditions and quality data in a manner corresponding to changes in the quality of the metal material (in a way that reflects factors contributing to quality). Therefore, the quality prediction model generated based on such grouping can make high-precision predictions.

[0092] Furthermore, using the manufacturing condition prompting method and device for metal materials described in this embodiment, manufacturing conditions for improving internal defects (defects based on ultrasonic testing) within structural steel plates are searched and prompted. In this embodiment, a quality prediction model generated using the aforementioned LightGBM is used. Among the operable manufacturing conditions in the refining, casting, heating, hot rolling, cooling, and cold rolling processes, the heating temperature of the heating process with the highest gain under LightGBM and the highest contribution to the presence or absence of internal defects as the target variable is selected as the change condition. Using the quality prediction model, the presence or absence of quality defects in the metal material is predicted. Within the range of the maximum to minimum heating temperature obtained from the running data server 20, an iterative method (golden section method) is used to search for the heating temperature at which the predicted result is "no" quality defects. Then, the change of the heating temperature of the searched heating process is set in the control computer, and the metal material is manufactured. The result is that with an average heating temperature increase of +30°C, the occurrence rate of quality defects is 9%. Before the application of the manufacturing condition reminder method for metal materials involved in this embodiment, the occurrence rate of quality defects was 10%, and the occurrence rate of quality defects was improved.

[0093] The embodiments described herein are based on the accompanying drawings and examples; however, it should be noted that those skilled in the art can readily make various modifications or alterations based on this disclosure. Therefore, it should be understood that such modifications or alterations are included within the scope of this disclosure. For example, the functions contained in each structural component or step can be reconfigured in a logically consistent manner, and multiple structural components or steps can be combined into one or divided. The embodiments described herein can also be implemented as a program executed by a processor of a device or a storage medium containing a program. These should also be understood to be included within the scope of this disclosure.

[0094] Here, as described above, the quality prediction model generation device 10 may not have the function of a quality prediction device for metallic materials. Specifically, it may be configured such that the quality prediction model generation device 10 only has the function of generating quality prediction models, and the prediction unit 136 is included in a computer (information processing device) that is different from the quality prediction model generation device 10, i.e., a quality prediction device for metallic materials. In this case, the quality prediction device for metallic materials obtains the model generated by the quality prediction model generation device 10 and stored in the storage unit 12 via a network, and predicts whether there are quality defects in the metallic materials.

[0095] Alternatively, as another example, the display unit 30 may be included in the quality prediction model generation device 10.

[0096] Explanation of reference numerals in the attached figures

[0097] 10…Quality prediction model generation device (quality prediction device for metallic materials); 11…Communication unit; 12…Storage unit; 13…Control unit; 20…Running data server; 30…Display unit; 131…Acquisition unit; 132…Storage unit; 133…Calculation unit; 134…Search unit; 135…Generation unit; 136…Prediction unit; 137…Manufacturing condition prompting unit; 138…Output unit.

Claims

1. A method for generating a quality prediction model, characterized in that, a method for generating a quality prediction model for metal materials manufactured through one or more processes, wherein... include: The acquisition step involves acquiring the explanatory variables selected from the manufacturing conditions of each process, and the state of quality defects of the manufactured metal material, i.e., the target variables. The saving step involves establishing an association between the explanatory variables and the target variables and saving them as learning data; The calculation step involves dividing the learning data into groups and testing whether there are significant differences in the state of the quality defects. The search steps involve searching for the group with the highest significance. as well as The generation step involves generating the quality prediction model using machine learning from the groups with the highest saliency found in the search.

2. The method for generating a quality prediction model according to claim 1, characterized in that, The explanatory variables are obtained as numerical variables or categorical variables.

3. The method for generating a quality prediction model according to claim 2, characterized in that, When the explanatory variable is obtained as a categorical variable, the calculation step performs a numerical transformation to assign different binary values ​​to the categorical variable for classification, and calculates an index to evaluate the significant difference based on the assigned values ​​and the state of the quality defect.

4. The method for generating a quality prediction model according to claim 3, characterized in that, The calculation steps calculate the index for each of the multiple groups.

5. The method for generating a quality prediction model according to any one of claims 1 to 4, characterized in that, After performing the search step, the calculation step further groups and performs the test on one of the groups with the highest significance found in the search.

6. The method for generating a quality prediction model according to any one of claims 1 to 5, characterized in that, The machine learning methods mentioned include at least linear regression, local regression, principal component regression, PLS regression, neural networks, regression trees, random forests, LightGBM, and XGBoost.

7. The method for generating a quality prediction model according to any one of claims 1 to 6, characterized in that, If the prediction accuracy of the quality prediction model deviates from the specified range, or at specified intervals, the quality prediction model will be regenerated.

8. A method for predicting the quality of metallic materials, characterized in that, The quality prediction model generated by the quality prediction model generation method according to any one of claims 1 to 7 is used to predict the state of quality defects in the metallic material.

9. A method for manufacturing a metallic material, characterized in that, The quality prediction method for metallic materials as described in claim 8 is used to predict the state of quality defects in the metallic material. If a quality defect in the metal material is predicted, at least a portion of the manufacturing conditions are modified.

10. A method for indicating manufacturing conditions of a metallic material, characterized in that, The method for predicting the quality of metal materials according to claim 8 is used to predict the state of quality defects in the metal material. If the presence of quality defects in the metal material is predicted, changes to at least a portion of the manufacturing conditions that would eliminate the quality defects are searched and suggested.

11. A method for manufacturing a metallic material, characterized in that, The metal material is manufactured based on the changes in the manufacturing conditions indicated by the manufacturing condition indication method for the metal material according to claim 10.

12. A quality prediction model generation device, which generates a quality prediction model for a metal material manufactured through one or more processes, characterized in that... have: The acquisition department acquires the explanatory variables selected from the manufacturing conditions of each process, and the state of quality defects of the manufactured metal material, i.e., the target variables. The storage unit establishes an association between the explanatory variables and the target variables and saves them as learning data; The computing unit divides the learning data into groups and checks whether there are significant differences in the state of the quality defects; Search department, searching for the group with the highest salience; as well as The generation unit generates the quality prediction model using machine learning from the groups with the highest saliency found in the search.

13. A device for predicting the quality of metallic materials, characterized in that, The quality prediction model generated by the quality prediction model generation device according to claim 12 is used to predict the state of quality defects of the metallic material.

14. A device for indicating manufacturing conditions of a metallic material, characterized in that, Based on the presence or absence of quality defects in the metal material predicted by the metal material quality prediction device of claim 13, changes to at least a portion of the manufacturing conditions of the metal material are indicated.

15. A manufacturing system for metallic materials, characterized in that... ， A manufacturing apparatus that controls each step of manufacturing a metal material based on changes in the manufacturing conditions output from the output unit of the manufacturing condition prompting device for the metal material as described in claim 14.

Images