Material data processing device, material data processing method, program, and manufacturing method of sintered magnet

Abstract

To generate the crystal structure data of a material with quality characteristics satisfying a target, that is, XRD data obtained by an X-ray diffraction method, and manufacturing conditions for the material.SOLUTION: A material data processing device 1 comprises a generation model construction part 21 to generates the generation model for generating material XRD data using XRD data 101 obtained by analyzing the material using an X-ray diffraction method as training data, a characteristic prediction model generating part 22 to create a regression model 32 for predicting the quality characteristic of arbitrary XRD data by learning the data associating the material XRD data 101 with the quality characteristic 110, and an XRD data generating part 24 to generate the XRD data 120 with the target quality characteristic using the generating model 31 created by the generation model construction part 21 and using the regression model 32 created by the characteristic prediction model generating part 22.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a technique for predicting the quality characteristics of materials, as well as manufacturing conditions such as composition conditions, blending conditions, and process conditions, from X-ray diffraction (XRD) data obtained from materials such as alloys, magnets, ceramics, and resins. [Background technology]

[0002] In recent years, the technological field of materials informatics, which effectively utilizes information science, particularly data science, to develop new materials, has been attracting attention. In materials informatics, data on various experimental conditions and results is associated and stored in a database, and information useful for developing new materials is extracted using statistical analysis, machine learning, simulation, etc.

[0003] For example, in Patent Documents 1 and 2, a regression model (mathematical model / machine learning model) is created using supervised machine learning, with multiple manufacturing conditions such as composition conditions and heat treatment conditions as explanatory variables and material properties such as tensile strength and elongation as target variables. Then, using the created regression model, the quality characteristics of the material are virtually predicted for multiple manufacturing conditions such as composition conditions and heat treatment conditions. From the predicted quality characteristics of the material, manufacturing conditions that satisfy the requirements for the target quality characteristics are extracted and used as the design values ​​for the material. The techniques for learning and creating regression models such as those disclosed in Patent Documents 1 and 2 are considered to be easily applicable in the field of materials design. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6617842 [Patent Document 2] Patent No. 6950119 Summary of the Invention [Problem to be solved by the invention]

[0005] However, there is a limit to further improving the prediction accuracy in the methods of predicting the quality characteristics of a material using a regression model trained using only the manufacturing conditions of the material as explanatory variables, as in Patent Documents 1 and 2. Therefore, in order to achieve more accurate predictions, it is conceivable to consider information other than the manufacturing conditions, such as information on the structure of the material.

[0006] The present invention has been made in view of these problems, and aims to provide a materials data processing device and the like that can generate crystal structure data of materials that have quality characteristics that meet targets, i.e., XRD data obtained by X-ray diffraction, and manufacturing conditions for those materials. [Means for solving the problem]

[0007] The first invention for solving the above-mentioned problems is a material data processing device comprising: a generative model creation unit that creates a generative model for generating XRD data of the material using XRD data obtained by analyzing the material using X-ray diffraction as training data; a property prediction model creation unit that creates a regression model that predicts the quality characteristics of any XRD data by learning data linking the XRD data of the material with quality characteristics; and an XRD data generation unit that generates XRD data having target quality characteristics using the generative model created by the generative model creation unit and the regression model created by the property prediction model creation unit.

[0008] In the first invention, the system may further include a manufacturing condition prediction model creation unit that creates a regression model that predicts manufacturing conditions for any XRD data by learning data linking the XRD data of the material with manufacturing conditions, and a manufacturing condition prediction unit that predicts manufacturing conditions that will achieve target quality characteristics using the regression model created by the manufacturing condition prediction model creation unit and the XRD data generated by the XRD data generation unit.

[0009] Furthermore, from the multiple XRD data generated by the generation model, the quality characteristics and manufacturing conditions of each XRD data may be predicted using a regression model created by the characteristic prediction model creation unit and a regression model created by the manufacturing condition prediction model, and the prediction results may be plotted as a heat map.

[0010] The second invention is a material data processing device comprising: a generative model creation unit that creates a generative model for generating XRD data, quality characteristics, and manufacturing conditions for the material using training data that includes XRD data obtained by analyzing a material using X-ray diffraction and data that links the quality characteristics of the material with the manufacturing conditions of the material; and a data generation unit that uses the generative model created by the generative model creation unit to generate multiple pieces of data including the XRD data, quality characteristics, and manufacturing conditions of the material, and output manufacturing conditions that achieve target quality characteristics.

[0011] In the second invention, a heat map may be drawn based on quality characteristics and manufacturing conditions included in each piece of data generated by the generative model.

[0012] A third invention is a material data processing method including: a generative model creation step in which a computer creates a generative model for generating XRD data of the material using XRD data obtained by analyzing the material using X-ray diffraction as training data; a property prediction model creation step in which a regression model for predicting quality characteristics of any XRD data is created by learning data linking the XRD data of the material with quality characteristics; and an XRD data generation step in which the generative model created in the generative model creation step and the regression model created in the property prediction model creation step are used to generate XRD data having target quality characteristics.

[0013] In the third invention, the method may further include a manufacturing condition prediction model creation step in which the computer creates a regression model that predicts manufacturing conditions for any XRD data by learning data linking the XRD data of the material with manufacturing conditions, and a manufacturing condition prediction step in which the computer predicts manufacturing conditions that will achieve target quality characteristics using the regression model created in the manufacturing condition prediction model creation step and the XRD data generated in the XRD data generation step.

[0014] A fourth invention is a material data processing method including: a generative model creation step in which a computer creates a generative model for generating XRD data, quality characteristics, and manufacturing conditions for the material using training data that links XRD data obtained by analyzing a material using X-ray diffraction and data that links the quality characteristics of the material with the manufacturing conditions of the material; and a data generation step in which the computer uses the generative model created by the generative model creation step to generate multiple pieces of data including the XRD data, quality characteristics, and manufacturing conditions of the material, and output manufacturing conditions that achieve target quality characteristics.

[0015] A fifth invention provides a computer including a generative model creation unit that creates a generative model for generating XRD data of a material using XRD data obtained by analyzing the material using X-ray diffraction as training data, a property prediction model creation unit that creates a regression model that predicts quality characteristics of any XRD data by learning data that links the XRD data of the material with quality characteristics, and an XRD data generation unit that generates XRD data having target quality characteristics using the generative model created by the generative model creation unit and the regression model created by the property prediction model creation unit. It is a program that functions as a

[0016] A sixth invention is a program that causes a computer to function as a generative model creation unit that creates a generative model for generating data including XRD data, quality characteristics, and manufacturing conditions of the material using training data that links XRD data obtained by analyzing a material using X-ray diffraction, the quality characteristics of the material, and the manufacturing conditions of the material, and a data generation unit that uses the generative model created by the generative model creation unit to generate multiple pieces of data including XRD data, quality characteristics, and manufacturing conditions of the material and output manufacturing conditions that achieve target quality characteristics.

[0017] A seventh aspect of the present invention is a method for producing a sintered magnet, which produces a sintered magnet based on the production conditions predicted by the material data processing method of the third aspect of the present invention.

[0018] An eighth aspect of the present invention is a method for producing a sintered magnet, which produces a sintered magnet based on the production conditions output by the material data processing method of the fourth aspect of the present invention. [Effects of the Invention]

[0019] The present invention makes it possible to provide a materials data processing device and the like that can generate crystal structure data of materials having quality characteristics that meet targets, i.e., XRD data obtained by X-ray diffraction, and manufacturing conditions for those materials. [Brief explanation of the drawings]

[0020] [Figure 1] FIG. 2 is a diagram illustrating an example of a hardware configuration of a material data processing device. [Figure 2] FIG. 2 is a block diagram showing an example of the functional configuration of a material data processing device. [Figure 3] FIG. 1 is a diagram showing an example of XRD data. [Figure 4] FIG. 1 is a diagram showing an example of a table in which XRD data of a plurality of materials are normalized for each material and compiled in a tabular format. [Figure 5] 1 is a flowchart showing an example of the overall flow of a material data processing method. [Figure 6]10 is a flowchart illustrating an example of a process for creating a generative model. [Figure 7] 10 is a flowchart illustrating an example of a process for creating a regression model for predicting a quality characteristic. [Figure 8] FIG. 1 is a diagram showing an example of a table in which XRD data and quality characteristics are linked together in a tabular format. [Figure 9] FIG. 10 is a diagram illustrating an example of the results of verifying the prediction accuracy of a regression model that predicts quality characteristics. [Figure 10] 10 is a flowchart illustrating an example of a process for creating a regression model for predicting manufacturing conditions. [Figure 11] FIG. 10 is a diagram showing an example of a table in which XRD data and manufacturing conditions are linked together in a tabular format. [Figure 12] FIG. 10 is a diagram showing an example of the results of verifying the prediction accuracy of a regression model that predicts manufacturing conditions. [Figure 13] 1 is a flowchart illustrating an example of a process for generating XRD data having target quality characteristics. [Figure 14] FIG. 10 is a diagram showing an example of a table in which XRD data having target quality characteristics is compiled in a tabular format. [Figure 15] FIG. 10 is a diagram showing an example of a table summarizing manufacturing conditions for realizing target quality characteristics in a tabular format. [Figure 16] FIG. 10 is a diagram showing an example of a heat map drawn based on predicted quality characteristics and manufacturing conditions. [Figure 17] FIG. 10 is a block diagram showing an example of a functional configuration of a material data processing apparatus according to another embodiment. [Figure 18] 10 is a flowchart illustrating an example of a process for creating a generative model. [Figure 19] FIG. 1 is a diagram showing an example of a table in which data linking XRD data, quality characteristics, and manufacturing conditions is compiled in a tabular format. [Figure 20] FIG. 10 is a diagram showing an example of the accuracy verification results of the quality characteristics and manufacturing conditions generated by the generative model. [Figure 21]10 is a flowchart showing an example of a process for outputting XRD data having target quality characteristics and manufacturing conditions for realizing the target quality characteristics. [Figure 22] 1 is a flowchart showing an example of a manufacturing process for a ferrite magnet based on manufacturing conditions predicted or generated by a materials data processing device. [Figure 23] 10 is a flowchart showing an example of a processing flow of a material data processing apparatus according to another embodiment. [Figure 24] 10 is a flowchart showing an example of a process for outputting XRD data having target quality characteristics and manufacturing conditions for realizing the target quality characteristics. DETAILED DESCRIPTION OF THE INVENTION

[0021] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In this embodiment of the present invention, an example will be described in which XRD data is generated for a calcined body of a ferrite magnet, which is one type of sintered magnet having target quality characteristics, using a materials data processing device 1. Note that the material is not limited to sintered magnets, and the present invention can also be applied to various materials including, for example, alloys, ceramics, resins, etc.

[0022] [First embodiment] First, a first embodiment of the present invention will be described. FIG. 1 is a diagram showing an example of the hardware configuration of a materials data processing apparatus 1 according to a first embodiment of the present invention. For example, when a general personal computer is used as the materials data processing apparatus 1, the materials data processing apparatus 1 includes a control unit 11, a memory unit 12, a communication unit 13, an input unit 14, a display unit 15, a peripheral device interface (I / F) unit 16, and other components connected via a bus 17, as shown in FIG. 1. Note that the hardware configuration shown in FIG. 1 is merely an example, and various configurations can be adopted depending on the application and purpose. The materials data processing apparatus 1 may also be a multi-computer including multiple computers. It may also be a server-client system, a cloud server, or a virtual machine virtually constructed using software. In the following explanation, the materials data processing apparatus 1 will be described as a single computer.

[0023] The control unit 11 has a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The CPU loads a program stored in the storage unit 12 or a recording medium such as the ROM into a work memory area on the RAM and executes it, and drives and controls each unit connected via a bus 17 to realize each process of the material data processing device 1, which will be described later.

[0024] The ROM is a non-volatile memory that permanently stores the computer's boot program, programs such as the BIOS, and data, etc. The RAM is a volatile memory that temporarily stores programs and data loaded from the storage unit 12 or a recording medium such as a ROM, and also has a work area that the control unit 11 uses to perform various processes.

[0025] The storage unit 12 is a hard disk drive, a solid state drive (SSD), a flash memory, or the like, and stores programs executed by the control unit 11, data necessary for executing the programs, an operating system (OS), etc. For example, the storage unit 12 stores application programs for causing the materials data processing device 1 to execute the processes described below, XRD data which is structural information of the material, trained machine learning models (generative models and regression models), etc.

[0026] The communication unit 13 includes a communication interface and a communication control circuit that mediate communication of the material data processing device 1, and controls communication via a network. The network may be a local area network (LAN), a wide area network (WAN), the Internet, or the like, and may be wired or wireless.

[0027] The input unit 14 includes input devices such as a keyboard, a mouse, or a touch panel, and various operation buttons, etc. The input unit 14 transmits input data and operation instructions to the control unit 11.

[0028] The peripheral device interface (I / F) unit 16 is a port for connecting peripheral devices, and includes short-range wireless communication such as USB and Bluetooth (registered trademark). The control unit 11 transmits and receives data to and from peripheral devices via the peripheral device interface unit 16. For example, a printer or the like is connected to the peripheral device interface unit 16.

[0029] Next, the functional configuration of the material data processing device 1 will be described. 2 is a block diagram showing an example of the functional configuration of a materials data processing device 1 according to this embodiment. As shown in the figure, the materials data processing device 1 includes a generation model creation unit 21, a property prediction model creation unit 22, a manufacturing condition prediction model creation unit 23, an XRD data generation unit 24, and a manufacturing condition prediction unit 25.

[0030] The generative model creation unit 21 creates a generative model 31 for generating XRD data of a material using XRD data 101 obtained by analyzing the material using X-ray diffraction as training data. The property prediction model creation unit 22 creates a regression model 32 for predicting the quality characteristics of any XRD data by learning data linking the XRD data 101 of the material with quality characteristics 110.

[0031] The XRD data generation unit 24 generates XRD data 120 having the target quality characteristics using the generation model 31 created by the generation model creation unit 21 and the regression model 32 created by the characteristic prediction model creation unit 22.

[0032] The XRD data generation unit 24 is composed of a property prediction unit 241 and an alignment / determination unit 242. The property prediction unit 241 predicts quality properties for each of the multiple XRD data 201 generated by inputting random values ​​into the generation model 31, using a regression model 32. The alignment / determination unit 242 aligns the multiple XRD data 201 based on the values ​​of the quality properties, based on the quality property prediction results by the property prediction unit 241, and outputs XRD data 120 having the target quality properties.

[0033] The manufacturing condition prediction model creation unit 23 creates a regression model 33 that predicts manufacturing conditions for any XRD data by learning data linking the XRD data 101 of the material with the manufacturing conditions 111. The manufacturing condition prediction unit 25 uses the regression model 33 created by the manufacturing condition prediction model creation unit 23 and the XRD data 120 generated by the XRD data generation unit 24 to predict and output manufacturing conditions 130 that will achieve target quality characteristics.

[0034] FIG. 3 is a diagram showing an example of XRD data targeted by this embodiment. The data format of devices that measure crystal structures using X-ray diffraction varies slightly depending on the manufacturer and model. A data sheet 101 (XRD data 101) output as measurement results by the device used in this embodiment describes 2-theta values ​​and intensity values. In the example of FIG. 3, 2-theta values ​​are described in the first column and intensity values ​​in the second column. Graph 102 is a waveform plotted from the measurement results of data sheet 101 (XRD data 101), with 2-theta values ​​on the horizontal axis and intensity values ​​on the vertical axis. With XRD data 101, the absolute intensity values ​​are not very meaningful, but the relative values, i.e., the waveform shape, are meaningful.

[0035] For this reason, it is common to normalize the measured XRD data 101 by setting the minimum intensity value to zero and the maximum intensity value to one before using it for data analysis. In this embodiment, the XRD data 101 is also first normalized. Table 103 is an example of the measurement results of the data sheet 101 (XRD data 101) normalized according to Equation 61 and summarized in tabular format. For example, the column name A0500 is obtained by multiplying the 2-theta value 5.0000 by 100 to obtain the integer 500, and then adding a leading zero to make it 0500 to unify it to four digits, and then adding the letter A before it to form the column name. Furthermore, table 103 assigns material IDs to the first column so that multiple materials can be identified. For ease of explanation, both the XRD data before normalization and the XRD data after normalization will be referred to as XRD data 101.

[0036] Figure 4 is an example of a table in which XRD data 101 for multiple materials is normalized for each material and compiled in tabular format. Table 104 is created by vertically combining table 103, one for each material. The first column is the material ID. The material ID is identification information for linking the XRD data 101, quality characteristics 110, and manufacturing conditions 111. As with table 103, the second and subsequent columns are named based on the 2-theta value and contain real numbers between 0 and 1, with the intensity normalized for each material.

[0037] Next, a material data processing method executed by the material data processing device 1 will be described with reference to FIGS. 5 is a flowchart showing an example of the overall processing flow of the materials data processing method. In step S1, the materials data processing device 1 creates a generation model 31 for generating XRD data of a material.

[0038] Fig. 6 is a flowchart showing an example of the process for creating the generative model 31 executed in step S1 of Fig. 5. This process is executed by the generative model creation unit 21 of the materials data processing device 1. As described above, the generative model creation unit 21 is a functional unit that creates the generative model 31 for generating XRD data for the material using the XRD data 101 obtained by analyzing the material using X-ray diffraction as training data, and Fig. 6 specifically shows the process of this functional unit.

[0039] First, the materials data processing device 1 (generative model creation unit 21) reads XRD data 101 for multiple materials from the storage unit 12 into the RAM of the control unit 11 and normalizes the XRD data 101 for each material (step S11). An example of normalized XRD data 101 for multiple materials is the table 104 (FIG. 4). Next, the materials data processing device 1 (generative model creation unit 21) learns and creates a generative model 31 using a variational autoencoder or a Gaussian mixture model using neural network technology (step S12).

[0040] The variational autoencoder assigns column names A0500 to A9000 in table 104 to each node in the input and output layers of the neural network, respectively, and trains the latent space, i.e., generative model 31, so that the input and output have as similar values ​​as possible. If random values ​​are input into the latent space trained by the variational autoencoder, i.e., generative model 31, it is possible to generate a large amount of XRD data that is similar but not the same as the data in table 104 used for training.

[0041] On the other hand, the Gaussian mixture model does not use a neural network but instead uses the EM algorithm to train a generative model 31 consisting of a combination of multiple normal distributions. Like the variational autoencoder, when random values ​​are input into the trained generative model 31 of the Gaussian mixture model, it is possible to generate a large amount of XRD data that is similar but not the same as the data in table 104 used for training.

[0042] Next, in step S2 of FIG. 5, the materials data processing device 1 creates a regression model 32 that predicts the quality characteristics of any XRD data. Fig. 7 is a flowchart showing an example of the process for creating the regression model 32 executed in step S2 of Fig. 5. This process is executed by the property prediction model creation unit 22 of the materials data processing device 1. As described above, the property prediction model creation unit 22 is a functional unit that creates a regression model 32 that predicts the quality characteristics of any XRD data by learning data that links the XRD data 101 and quality characteristics 110 of a material, and Fig. 7 specifically shows the process of this functional unit.

[0043] First, the material data processing device 1 (property prediction model creation unit 22) reads the XRD data 101 of multiple materials from the storage unit 12 into the RAM of the control unit 11 and normalizes the XRD data 101 for each material (step S21). The above-mentioned table 104 (FIG. 4) is an example of normalized XRD data 101 of multiple materials. Note that if the same XRD data 101 as in FIG. 6 is used, the processing of step S21 can be omitted. Next, the material data processing device 1 (property prediction model creation unit 22) associates quality characteristics 110 with the normalized XRD data 101 based on the material ID (step S22). In the case of a sintered magnet, the quality characteristics 110 are measurement data such as saturation magnetization, residual magnetic flux density, coercive force, and squareness.

[0044] 8 is an example of a table in which XRD data 101 and quality characteristics 110 are linked together in tabular form. Table 105 is the result of linking saturation magnetization as a quality characteristic 110 based on the material ID to table 104 created in step S21 in step S22. The first column of table 105 is the material ID, columns A0500 to A9000 are normalized XRD data 101, and the rightmost column Y01 is data on saturation magnetization, one of the quality characteristics 110. In this example, only saturation magnetization is linked as the quality characteristic 110, but it is also possible to link multiple variables such as remanence, coercivity, and squareness.

[0045] Then, the material data processing device 1 (property prediction model creation unit 22) learns and creates a regression model 32 for predicting quality characteristics from the XRD data for each linked quality characteristic 110 (step S23 in FIG. 7). When a neural network is used in step S23, column names A0500 to A9000 of table 104 are assigned to each node in the input layer of the neural network, and each quality characteristic 110 is assigned to the output layer for learning. In other words, when the quality characteristic 110 has multiple variables, such as residual magnetic flux density and coercive force, a regression model 32 is learned and created separately for each variable.

[0046] In step S23, various regression models can be used, including not only neural networks but also random forests, Gaussian process regression, multiple regression analysis, and lasso regression. However, when using a regression model that assumes linearity, such as multiple regression analysis or lasso regression, inputting more than several dozen explanatory variables theoretically makes it impossible to create a correct regression model. In such cases, a dimensionality reduction process such as principal component analysis is performed in advance, and then the regression model 32 is trained. Specifically, principal component regression or PLS regression can be used.

[0047] Figure 9 shows an example of the results of verifying the predictive accuracy of regression model 32. Scatter plot 71 in Figure 9(a) shows the verification results for a regression model using a fully connected neural network, while scatter plot 72 in Figure 9(b) shows the verification results for a regression model using principal component regression, i.e., multiple regression analysis after dimensionality reduction using principal component analysis. Both scatter plots 71 and 72 show the results of verifying the predictive accuracy of the dependent variable using 10-fold cross-validation (cross-validation) of table 105, with normalized XRD data 101 from column A0500 to column A9000 as the explanatory variable and column Y01, i.e., saturation magnetization, one of the quality characteristics 110, as the dependent variable. In both scatter plots 71 and 72, the horizontal axis represents the measured values, i.e., the values ​​written in column Y01 of table 105, and the vertical axis represents the predicted values ​​calculated using the regression model. Both regression models achieved prediction accuracy with a coefficient of determination exceeding 0.95. Although this prediction accuracy cannot be said to be sufficient, it was confirmed that the performance of sintered magnets, that is, the saturation magnetization, which is one of the quality characteristics, can be predicted solely from XRD data, which is the measurement result of the crystal structure.

[0048] Next, in step S3 of FIG. 5, the materials data processing device 1 creates a regression model 33 for predicting the manufacturing conditions of any XRD data. Fig. 10 is a flowchart showing an example of the process for creating the regression model 33, which is executed in step S3 of Fig. 5. This process is executed by the manufacturing condition prediction model creation unit 23 of the materials data processing device 1. As described above, the manufacturing condition prediction model creation unit 23 is a functional unit that creates the regression model 33 for predicting manufacturing conditions for any XRD data by learning data linking the XRD data 101 of the material with the manufacturing conditions 111, and Fig. 10 specifically shows the process of this functional unit.

[0049] First, the material data processing device 1 (manufacturing condition prediction model creation unit 23) reads the XRD data 101 for multiple materials from the storage unit 12 into the RAM of the control unit 11 and normalizes the XRD data 101 for each material (step S31). The table 104 (FIG. 4) shows an example of normalized XRD data 101 for multiple materials. Note that if the same XRD data 101 as in FIG. 6 or 7 is used, step S31 can be omitted. Next, the material data processing device 1 (manufacturing condition prediction model creation unit 23) associates the manufacturing conditions 111 with the normalized XRD data 101 based on the material ID (step S32). For a sintered magnet, the manufacturing conditions 111 include variables such as the amounts and ratios of each composition mixed as raw materials, the amounts and ratios of additives, the temperature and time of heat treatment, and the particle size after pulverization. Variables in the manufacturing conditions 111 often include both set values ​​and measured values. It is best to use measured values ​​whenever possible, but set values ​​are used for values ​​that have not been measured or cannot be measured.

[0050] FIG. 11 is an example of a table in which XRD data 101 and manufacturing conditions 111 are linked together in a tabular format. Table 106 is the result of linking the amounts and ratios of calcium, lanthanum, iron, and other components for each composition as manufacturing conditions 111 in step S32 to table 104 created in step S31. In table 106, the first column is the material ID, columns A0500 to A9000 are normalized XRD data 101, and columns X01 to X04 are data on manufacturing conditions 111. Columns X01 and X02 are composition ratios, and the sum of the two columns is 1. On the other hand, columns X03 and X04 are independent data.

[0051] Then, the material data processing device 1 (manufacturing condition prediction model creation unit 23) learns and creates a regression model 33 for predicting manufacturing conditions from the XRD data 101 for each linked manufacturing condition 111 (step S33 in FIG. 10). When a neural network is used in step S3, column names A0500 to 9000 of table 104 are assigned to each node in the input layer of the neural network, and each manufacturing condition 111 is assigned to the output layer for learning. In other words, when the manufacturing conditions 111 have multiple variables, a regression model 33 is learned and created separately for each variable.

[0052] In step S33, various regression models can be used, including not only neural networks but also random forests, Gaussian process regression, multiple regression analysis, and lasso regression. However, when using a regression model that assumes linearity, such as multiple regression analysis or lasso regression, inputting more than several dozen explanatory variables theoretically makes it impossible to create a correct regression model. In such cases, a dimensionality reduction process such as principal component analysis is performed in advance, and then the regression model 33 is trained. Specifically, principal component regression or PLS regression can be used.

[0053] Figure 12 shows an example of the results of verifying the predictive accuracy of regression model 33. Scatter plot 73 in Figure 12(a) shows the results of a regression model using a fully connected neural network, while scatter plot 74 in Figure 12(b) shows the results of a regression model using principal component regression, i.e., multiple regression analysis after dimensionality reduction using principal component analysis. Both scatter plots 73 and 74 show the results of verifying the predictive accuracy of the dependent variable using 10-fold cross-validation (cross-validation) of table 106, with normalized XRD data 101 from column A0500 to column A9000 as the explanatory variable and column X01, i.e., the calcium ratio, as the dependent variable. Furthermore, in both scatter plots 73 and 74, the horizontal axis represents the set value defined as the manufacturing condition 111, i.e., the value written in column X01 of table 106, and the vertical axis represents the predicted value calculated by the regression model. Both regression models achieved prediction accuracy with a coefficient of determination exceeding 0.87. Although this prediction accuracy cannot be said to be sufficient, it was confirmed that the calcium ratio, one of the manufacturing conditions that determines the performance of sintered magnets, can be predicted from only the XRD data, which is the measurement result of the crystal structure.

[0054] The machine learning models (generative model 31, regression model 32, and regression model 33) used in this embodiment are created by the processing of steps S1 to S3 in Fig. 5 described above. Note that the order in which steps S1 to S3 are executed is not limited to the example in Fig. 5, and steps S1 to S3 may be executed in any order.

[0055] Subsequently, in step S4 of FIG. 5, the material data processing apparatus 1 generates XRD data 120 having the target quality characteristics. 13 is a flowchart showing an example of the process for generating XRD data 120 having target quality characteristics, which is executed in step S4 of FIG. 5. This process is executed by the property prediction unit 241 and sorting / determining unit 242 of the materials data processing device 1. As described above, the property prediction unit 241 is a functional unit that predicts quality characteristics for each of the multiple XRD data 201 generated by inputting random values ​​into the generation model 31, using the regression model 32. The sorting / determining unit 242 is a functional unit that sorts the multiple XRD data 201 based on the quality characteristic values, based on the quality characteristic prediction results by the property prediction unit 241, and outputs XRD data 120 having the target quality characteristics. FIG. 13 specifically shows the processing of these functional units.

[0056] First, random values ​​are input into the generation model 31 to generate multiple XRD data 201 that are similar but different. Next, the materials data processing device 1 (property prediction unit 241) inputs the generated multiple XRD data 201 into the regression model 32 and predicts quality characteristics for each XRD data 201 (step S41). The materials data processing device 1 (property prediction unit 241) also associates the predicted quality characteristics with each XRD data 201. Then, the materials data processing device 1 (sorting / determining unit 242) sorts the multiple XRD data 201 in descending order based on the quality characteristic values, and selects and outputs XRD data 120 that have the target quality characteristics (step S42). This selection is performed, for example, by selecting XRD data 201 whose quality characteristic values ​​are equal to or greater than a predetermined threshold as XRD data 120 that have the target quality characteristics.

[0057] 14 is an example of a table summarizing the XRD data 120 having the target quality characteristics output in step S42. Table 122 summarizes the output XRD data 120 having the target quality characteristics and their quality characteristics in tabular form. The first column of table 122 contains an ID for identifying the output XRD data 120 having the target quality characteristics, columns A0500 to A9000 contain the output XRD data 120, and column Y01 contains the predicted quality characteristics. Table 122 is created by sorting the generated multiple XRD data 201 in descending order based on the quality characteristic values ​​and selecting XRD data 201 having quality characteristic values ​​equal to or greater than a predetermined threshold.

[0058] Then, the material data processing device 1 (manufacturing condition prediction unit 25) predicts manufacturing conditions 130 that will achieve the target quality characteristics in step S5 of Fig. 5. Specifically, the material data processing device 1 (manufacturing condition prediction unit 25) inputs the XRD data 120 having the target quality characteristics output in the process of Fig. 13 into the regression model 33 created in the process of Fig. 10, and predicts manufacturing conditions 130 that will achieve the target quality characteristics.

[0059] FIG. 15 is an example of a table summarizing, in tabular form, manufacturing conditions 130 that will achieve the predicted target quality characteristics. Table 132 is table 122 of FIG. 14 to which the predicted manufacturing conditions 130 have been added. The predicted manufacturing conditions 130 correspond to columns named X01 to X04. These manufacturing conditions 130 are predicted for each material ID by inputting the XRD data 120 from column named A0500 to column named A9000 shown in table 122 of FIG. 14 as explanatory variables of the regression model 33.

[0060] 15, manufacturing conditions 130 that achieve the target quality characteristics are determined. However, since the quality characteristics and manufacturing conditions are prediction results based on the regression models 32 and 35, they contain prediction errors. In other words, the table 132 also contains singular values. Therefore, the materials data processing device 1 may predict a large number of quality characteristics and a large number of manufacturing conditions from the large amount of generated XRD data 201 using the regression models 32 and 35, and plot the prediction results as a heat map.

[0061] FIG. 16 is an example of a heat map 80 drawn based on predicted quality characteristics and manufacturing conditions. The horizontal axis of the heat map 80 represents the ratio of X01 to X02, i.e., the ratio of calcium to lanthanum, predicted as a manufacturing condition. The vertical axis represents the ratio of X03 to X04, i.e., the ratio of iron among transition metal elements, predicted as a manufacturing condition. The horizontal and vertical axes are discretized into a grid, and the average value of the predicted quality characteristics assigned to each grid is represented by a shaded area. Whiter grids represent higher average values ​​of the quality characteristics, and darker grids represent lower average values ​​of the quality characteristics. The heat map 80 may be drawn based on all predicted quality characteristics and manufacturing conditions, or it may be drawn based only on the top predicted quality characteristics (e.g., quality characteristics whose values ​​are equal to or greater than a predetermined value and the corresponding manufacturing conditions).

[0062] From the heat map 80, it is visually apparent that the manufacturing conditions that maximize the quality characteristics are when the ratio of X01 to X02 is approximately 50% and when the ratio of X03 to X04 is approximately 95%, making it possible to determine the optimal manufacturing conditions. In this embodiment, since there are few manufacturing conditions to be predicted, such as X01 to X04, the heat map 80 can be drawn with the relationship between X01 and X02 on the horizontal axis and the relationship between X03 and X04 on the vertical axis. However, in order to visualize more manufacturing conditions simultaneously, it is effective to draw the heat map 80 after compressing the dimensions using a statistical method such as principal component analysis or UMAP.

[0063] This completes the description of this embodiment. The materials data processing device 1 of this embodiment uses XRD data 101 acquired by analyzing a material using X-ray diffraction as training data to create a generation model 31 for generating XRD data for the material. Furthermore, by learning data linking the XRD data 101 with quality characteristics 110, a regression model 32 is created that predicts the quality characteristics of any XRD data. Furthermore, by learning data linking the XRD data 101 with manufacturing conditions 111, a regression model 33 is created that predicts the manufacturing conditions for any XRD data.

[0064] The material data processing device 1 then generates XRD data 120 having the target quality characteristics using the generation model 31 and the regression model 32, and further predicts manufacturing conditions 130 that will achieve the target quality characteristics using the regression model 33 and the generated XRD data 120 having the target quality characteristics. This makes it possible to obtain crystal structure data of a material having quality characteristics that satisfy the targets, i.e., XRD data obtained by X-ray diffraction, and suitable manufacturing conditions for that material.

[0065] [Second embodiment] A second embodiment of the present invention will be described. The material data processing device 1a according to the second embodiment does not create a regression model. A regression model must be trained for each quality characteristic and manufacturing condition, which requires time and effort. Although the prediction accuracy of quality characteristics and manufacturing conditions is somewhat reduced in the second embodiment, it is highly convenient because all data including quality characteristics and manufacturing conditions can be generated using a single generation model. The hardware configuration of the material data processing device 1a is the same as that of the first embodiment (FIG. 1).

[0066] 17 is a block diagram showing an example of the functional configuration of a material data processing device 1a according to the second embodiment. As shown in the figure, the material data processing device 1a includes a generative model creation unit 41 and a data generation unit .

[0067] The generative model creation unit 41 uses as training data XRD data 101 obtained by analyzing a material using X-ray diffraction, data linking the quality characteristics 110 of the material, and the manufacturing conditions 111 of the material, to create a generative model 51 for generating data including the XRD data, quality characteristics, and manufacturing conditions of the material.

[0068] The data generation unit 42 generates and outputs XRD data 140 having target quality characteristics and manufacturing conditions 150 that realize the target quality characteristics, using the generative model 51 created by the generative model creation unit 41. The data generation unit 42 is composed of a generation unit 421 and an alignment / determination unit 422.

[0069] The generation unit 421 inputs random values ​​into the generation model 51 to generate multiple pieces of data (hereinafter referred to as "data 301") including XRD data, quality characteristics, and manufacturing conditions. The sorting and determination unit 422 sorts the multiple pieces of data 301 generated by the generation unit 421 based on the values ​​of the quality characteristics, and outputs XRD data 140 having target quality characteristics and manufacturing conditions 150 that realize the target quality characteristics.

[0070] A material data processing method executed by the material data processing device 1a will be described with reference to FIGS. Fig. 18 is a flowchart showing an example of a process for creating a generative model 51. This process is executed by the generative model creation unit 41 of the material data processing device 1a. As described above, the generative model creation unit 41 is a functional unit that creates a generative model 51 for generating data including the XRD data, quality characteristics, and manufacturing conditions of a material using, as training data, data linking XRD data 101 obtained by analyzing a material using X-ray diffraction, quality characteristics 110 of the material, and manufacturing conditions 111 of the material. Fig. 18 specifically shows the processing of this functional unit.

[0071] First, the material data processing device 1a (generative model creation unit 41) reads XRD data 101 for multiple materials from the storage unit 12 into the RAM of the control unit 11, and normalizes the XRD data 101 for each material (step S61). Next, the material data processing device 1a (generative model creation unit 41) links the normalized XRD data 101 to quality characteristics 110 based on the material ID (step S62). Furthermore, the material data processing device 1a (generative model creation unit 41) further links the manufacturing conditions 111 to the XRD data 101 and quality characteristics 110 linked in step S62 (step S63).

[0072] Next, the material data processing device 1a (generative model creation unit 41) normalizes the quality characteristics and manufacturing conditions for each variable if necessary (step S64). Unlike the normalization of the XRD data 101 in step S61, step S64 normalizes the quality characteristics 110 and manufacturing conditions 111 for each variable so that the minimum value of all materials is 0 and the maximum value is 1. Then, the material data processing device 1a (generative model creation unit 41) learns and creates the generative model 51 (step S65). Like the generative model 31, the generative model 51 may use a variational autoencoder or a Gaussian mixture model.

[0073] 19 is an example of a table in tabular form that associates data of XRD data 101, quality characteristics 110, and manufacturing conditions 111. In table 107, the first column is the material ID, columns A0500 to A9000 are data in which the XRD data 101 is normalized for each material, columns X01 and X02 are manufacturing conditions 111 that do not require normalization, columns X03N and X04N are normalized manufacturing conditions 111, and Y01N is normalized quality characteristics 110.

[0074] Then, the materials data processing device 1a (generative model creation unit 41) learns and creates the generative model 51 (step S65 in FIG. 18). The generative model 51 may use a variational autoencoder or a Gaussian mixture model, similar to the generative model 31. The materials data processing device 1a (generative model creation unit 41) learns each column of the table 107 in FIG. 19 without distinguishing between XRD data, manufacturing conditions, and quality characteristics, and creates the generative model 51.

[0075] Figure 20 shows an example of the results of verifying the accuracy of the quality characteristics and manufacturing conditions generated by the generation model 51. 10-hold cross-validation was used to verify the accuracy. Scatter plot 91 in Figure 20(a) shows the results of verifying the generation accuracy of saturation magnetization, one of the quality characteristics. The horizontal axis shows the measured value of saturation magnetization, and the vertical axis shows the generated value of saturation magnetization. The coefficient of determination was 0.97. Scatter plot 92 in Figure 20(b) shows the results of verifying the generation accuracy of calcium ratio, one of the manufacturing conditions. The horizontal axis shows the measured value of calcium ratio, and the vertical axis shows the generated value of calcium ratio. The coefficient of determination was 0.93. While neither accuracy can be said to be sufficient, it was confirmed that the accuracy is practical.

[0076] FIG. 21 is a flowchart showing an example of a process for outputting XRD data 140 having target quality characteristics and manufacturing conditions 150 that realize the target quality characteristics. First, the material data processing device 1a (generation unit 421) inputs random values ​​into the generation model 51 to generate multiple pieces of data 301 including XRD data, quality characteristics, and manufacturing conditions (step S71). Then, the material data processing device 1a (sorting / determination unit 422) sorts the multiple pieces of data 301 generated in step S71 based on the values ​​of the quality characteristics, and selects and outputs XRD data 140 having target quality characteristics and manufacturing conditions 150 that realize the target quality characteristics (step S72). Specifically, the multiple pieces of data 301 generated in step S71 are sorted in descending order based on the values ​​of the quality characteristics, and XRD data of data 301 whose quality characteristic values ​​are equal to or greater than a predetermined threshold is selected and output as XRD data 140 having the target quality characteristics. Furthermore, manufacturing conditions of data 301 whose quality characteristic values ​​are equal to or greater than a predetermined threshold are selected and output as manufacturing conditions 150 that realize the target quality characteristics.

[0077] As in the first embodiment (FIG. 16), the material data processing apparatus 1a may plot a heat map using the quality characteristics and manufacturing conditions included in the large amount of data 301 generated in step S71.

[0078] (Method of manufacturing ferrite magnets) Finally, we will explain a method for manufacturing a ferrite magnet based on the manufacturing conditions 130 predicted by the material data processing device 1 according to the first embodiment and the manufacturing conditions 150 generated and output by the material data processing device 1a according to the second embodiment.

[0079] FIG. 22 is a flowchart showing an example of a manufacturing process for a ferrite magnet based on manufacturing conditions (manufacturing conditions 130 or 150) predicted or generated by the material data processing device 1 or 1a. According to the manufacturing conditions 130 or 150 that achieve target quality characteristics predicted or generated by the material data processing device 1 or 1a, appropriate amounts of raw materials are prepared and mixed (step S81). Next, calcination is performed in a rotary kiln to produce clinker (step S82). The clinker, mixed with water and additives, is then placed in a grinding device such as a ball mill and pulverized to a particle size of approximately 1 micrometer to produce slurry (step S83). The slurry is then poured into a mold and compacted in a press under a magnetic field to produce a green body (step S84). The green body is then sintered in a heat treatment furnace to produce a sintered body (step S85). The sintered body is then ground with a grinding wheel to produce a completed ferrite magnet (step S86). Finally, the completed ferrite magnet is inspected (step S87). Inspection items include measuring the magnetic properties of the ferrite magnets and checking for chips or scratches on the exterior.

[0080] In this embodiment, an example of the manufacturing conditions for the composition ratio of the raw materials mixed in step S81 is shown, which are determined based on the table 132 and the heat map 80. However, the manufacturing conditions determined by the material data processing device 1 or 1a include not only the composition ratio but also the oxygen concentration and temperature in the rotary kiln used in step S82, the calcination time, the type and amount of additives mixed with the clinker in step S83, particle size, the magnetizing current and press pressure for applying a magnetic field in step S84, the temperature and time for sintering in step S85, and so on.

[0081] While the preferred embodiments of the present invention have been described above with reference to the accompanying drawings, the present invention is not limited to these examples. It is clear that those skilled in the art can conceive of various modifications or alterations within the scope of the technical ideas disclosed herein, and it is understood that these modifications also fall within the technical scope of the present invention.

[0082] For example, it is possible to create a generative model (hereinafter referred to as "generative model 52") that generates XRD data of a material and data including the quality characteristics of the material. In this case, a regression model 33 that predicts manufacturing conditions from the XRD data of the material is also created. That is, the generative model 52 is used to generate multiple pieces of data including XRD data of the material and the quality characteristics of the material, and XRD data having the desired quality characteristics is selected from that data. Furthermore, the regression model 33 is used to predict the manufacturing conditions for the XRD data, thereby identifying the manufacturing conditions that will achieve the desired quality characteristics.

[0083] FIG. 23 is a flowchart showing an example of the processing flow of the materials data processing device 1b of the above example. First, the materials data processing device 1b reads XRD data 101 of multiple materials from the storage unit 12 into the RAM of the control unit 11 and normalizes the XRD data 101 for each material (step S91). Next, the materials data processing device 1b associates the normalized XRD data 101 with quality characteristics 110 based on the material ID (step S92). Next, the materials data processing device 1b normalizes the quality characteristics 110 for each variable if necessary (step S93). Unlike the normalization of the XRD data 101 in step S91, step S93 normalizes the quality characteristics 110 for each variable so that the minimum value of all materials is 0 and the maximum value is 1. Then, the materials data processing device 1b trains and creates a generative model 52 (step S94). Like the generative models 31 and 51, the generative model 52 may use a variational autoencoder or a Gaussian mixture model.

[0084] Next, the materials data processing device 1b links the manufacturing conditions 111 with the normalized XRD data 101 based on the material ID (step S95). Then, the materials data processing device 1b learns and creates a regression model 33 for predicting the manufacturing conditions 111 from the XRD data 101 for each linked manufacturing condition 111 (step S96). The processing in steps S95 to S96 is the same as steps S32 to S33 in FIG. 10.

[0085] 24 is a flowchart showing an example of a process for outputting XRD data 160 having target quality characteristics and manufacturing conditions 170 that realize the target quality characteristics. First, the material data processing device 1b inputs random values ​​into the generative model 52 to generate multiple pieces of data 401 including XRD data of the material and quality characteristics (step S101). Then, the material data processing device 1b sorts the multiple pieces of data 401 generated in step S101 based on the values ​​of the quality characteristics, and selects and outputs XRD data 160 having the target quality characteristics (step S102). Specifically, the multiple pieces of data 401 generated in step S101 are sorted in descending order based on the values ​​of the quality characteristics, and XRD data 401 whose quality characteristic values ​​are equal to or greater than a predetermined threshold are selected and output as XRD data 160 having the target quality characteristics. Then, the material data processing device 1b inputs the output XRD data 160 having the target quality characteristics into the regression model 33 created in step S96 of FIG. 23, thereby predicting and outputting manufacturing conditions 170 that will realize the target quality characteristics (step S102).

[0086] Although detailed description is omitted, it is also possible to create a generative model that generates data including XRD data of a material and manufacturing conditions of the material. In this case, a regression model 32 that predicts quality characteristics from the XRD data of the material is also created. Using the created generative model, multiple pieces of data including XRD data of the material and manufacturing conditions of the material are generated, and the XRD data contained in each piece of data is input into the regression model 32 to predict quality characteristics. Then, data containing XRD data having target quality characteristics is identified from the generated data, and the XRD data and manufacturing conditions contained in the identified data can be determined as XRD data having the target quality characteristics and manufacturing conditions that will achieve the target quality characteristics. [Explanation of symbols]

[0087] 1, 1a, 1b: Material data processing device 11: Control unit 12: Storage section 13: Communications Department 14: Input section 15: Display section 16: Peripheral device interface (I / F) section 21: Generative model creation unit 22: Characteristic prediction model creation section 23: Manufacturing condition prediction model creation section 24: XRD data generation unit 25: Manufacturing condition prediction section 41: Generative model creation unit 42: Data generation unit 241: Characteristics prediction unit 242, 422: Alignment and judgment section 421: Generation part 31, 51, 52: Generative models 32: Regression model for predicting quality characteristics 33: Regression model for predicting manufacturing conditions 71-75, 91-92: Scatter plot 80: Heatmap 101: XRD data 102:Graph of XRD data 103: Table of normalized XRD data 104: Table of normalized XRD data 105: A table linking XRD data and quality characteristics 106: A table linking XRD data and manufacturing conditions 107: Table of XRD data, normalized quality characteristics, and manufacturing conditions 110:Quality characteristics 111: Manufacturing conditions 120, 140, 160: XRD data with target quality characteristics 122: Table of XRD data with target quality characteristics 130, 150, 170: Manufacturing conditions to achieve the target quality characteristics 132: Table of manufacturing conditions to achieve target quality characteristics 201: Generated XRD data 301: Data including generated XRD data, quality characteristics, and manufacturing conditions 401: Generated XRD data and data including quality characteristics

Claims

1. a generation model creation unit that creates a generation model for generating XRD data of the material using XRD data acquired by analyzing the material using an X-ray diffraction method as training data; a characteristic prediction model creation unit that creates a regression model for predicting quality characteristics of any XRD data by learning data linking the XRD data and quality characteristics of the material; an XRD data generation unit that generates XRD data having target quality characteristics using the generative model created by the generative model creation unit and the regression model created by the characteristic prediction model creation unit; A material data processing device comprising:

2. a manufacturing condition prediction model creation unit that creates a regression model that predicts manufacturing conditions for any XRD data by learning data linking the XRD data of the material with manufacturing conditions; a manufacturing condition prediction unit that predicts manufacturing conditions that will achieve target quality characteristics using the regression model created by the manufacturing condition prediction model creation unit and the XRD data generated by the XRD data generation unit; The material data processing apparatus of claim 1 further comprising:

3. 3. The materials data processing device according to claim 2, wherein quality characteristics and manufacturing conditions of each XRD data are predicted from the plurality of XRD data generated by the generation model using a regression model created by the property prediction model creation unit and a regression model created by the manufacturing condition prediction model creation unit, and the prediction results are plotted as a heat map.

4. a generative model creation unit that creates a generative model for generating data including the XRD data, quality characteristics, and manufacturing conditions of the material using training data that links XRD data obtained by analyzing the material using an X-ray diffraction method, quality characteristics of the material, and manufacturing conditions of the material; and a data generation unit that uses the generative model created by the generative model creation unit to generate multiple pieces of data including XRD data, quality characteristics, and manufacturing conditions of the material, and outputs manufacturing conditions that realize target quality characteristics; A material data processing device comprising:

5. The material data processing device according to claim 4 , wherein a heat map is drawn based on quality characteristics and manufacturing conditions included in each piece of data generated by the generative model.

6. The computer a generation model creation step of creating a generation model for generating XRD data of the material using XRD data obtained by analyzing the material by an X-ray diffraction method as training data; a characteristic prediction model creation step of creating a regression model that predicts quality characteristics of any XRD data by learning data linking the XRD data and quality characteristics of the material; an XRD data generation step of generating XRD data having target quality characteristics using the generative model created in the generative model creation step and the regression model created in the characteristic prediction model creation step; A material data processing method comprising:

7. The computer a manufacturing condition prediction model creation step of creating a regression model that predicts manufacturing conditions for any XRD data by learning data linking the XRD data of the material with manufacturing conditions; a manufacturing condition prediction step of predicting manufacturing conditions that realize target quality characteristics using the regression model created in the manufacturing condition prediction model creation step and the XRD data generated in the XRD data generation step; 7. The materials data processing method of claim 6, further comprising:

8. The computer a generation model creation step of creating a generation model for generating the XRD data, quality characteristics, and manufacturing conditions of the material using training data that links XRD data obtained by analyzing the material using an X-ray diffraction method, quality characteristics of the material, and manufacturing conditions of the material; a data generation step of generating a plurality of pieces of data including XRD data, quality characteristics, and manufacturing conditions of the material using the generative model created by the generative model creation step, and outputting manufacturing conditions that realize target quality characteristics; A material data processing method comprising:

9. Computer, a generation model creation unit that creates a generation model for generating XRD data of the material using XRD data acquired by analyzing the material using an X-ray diffraction method as training data; a characteristic prediction model creation unit that creates a regression model that predicts quality characteristics of any XRD data by learning data linking the XRD data and quality characteristics of the material; an XRD data generation unit that generates XRD data having target quality characteristics using the generative model created by the generative model creation unit and the regression model created by the characteristic prediction model creation unit; A program that functions as a

10. Computer, a generative model creation unit that creates a generative model for generating data including the XRD data, quality characteristics, and manufacturing conditions of the material, using training data that links XRD data obtained by analyzing the material using an X-ray diffraction method, quality characteristics of the material, and manufacturing conditions of the material; a data generation unit that generates a plurality of pieces of data including XRD data, quality characteristics, and manufacturing conditions of the material using the generative model created by the generative model creation unit, and outputs manufacturing conditions that realize target quality characteristics; A program that functions as a

11. A method for producing a sintered magnet, comprising producing a sintered magnet based on the production conditions predicted by the material data processing method according to claim 7.

12. A method for manufacturing a sintered magnet, comprising manufacturing a sintered magnet based on the manufacturing conditions output by the material data processing method according to claim 8.

Images