System, method and program that assist in material creation

Abstract

To contribute to proposing a preparation recipe that assists in developing a material that has desired material properties.SOLUTION: In accordance with the accuracy of at least one model among first to a third models, a system carries out either an inference using the first model or an inference using the second and third models, and generates or updates recipe property data in said inference, that indicates the association of a material preparation recipe with material properties. The first model accepts as input a preparation recipe dataset that represents the preparation recipe and outputs a material property dataset that represents material properties. The second model accepts as input a material characteristics dataset that represents material characteristics and outputs the material property dataset. The third model accepts as input the preparation recipe dataset and outputs the material characteristics dataset.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention generally relates to technology that supports materials creation. [Background technology]

[0002] Materials informatics (MI) is known as one of the methods for supporting material creation. MI is a method for supporting material creation using informatics techniques, and generally involves aggregating data on material properties and using machine learning to search for materials with new material properties (e.g., Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-043959 Summary of the Invention [Problem to be solved by the invention]

[0004] For example, for organic materials, it is easy to simulate the molecular properties from the chemical composition (e.g., first-principles calculations), and therefore it is possible to obtain a large amount of highly accurate simulation data on the properties, making it easy to apply MI.

[0005] However, depending on the type of material, it can be difficult to apply MI. For example, in the case of inorganic materials (ceramics), the complex higher-order structure (e.g., grain size, grain size distribution, grain boundary phase, etc.) of a large number of crystal particles affects the characteristics, making it difficult to simulate the characteristics. For this reason, it is difficult to obtain highly accurate simulation data on the characteristics, and therefore it is difficult to apply MI. [Means for solving the problem]

[0006] The material creation support system includes first to third models and a processing unit. The first model is a model that inputs a recipe data set representing a recipe and outputs a material property data set representing material properties. The second model is a model that inputs a material property data set representing material properties and outputs a material property data set. The third model is a model that inputs a recipe data set and outputs a material property data set. The processing unit performs inference using the first model or inference using the second model and the third model depending on the accuracy of at least one of the first to third models, and generates or updates recipe property data that is data representing an association between a recipe for a material and material properties in the inference. Effect of the Invention

[0007] The present invention can contribute to the proposal of manufacturing recipes that support the development of materials having desired material properties. [Brief description of the drawings]

[0008] [Figure 1] 1 illustrates an example of a system configuration according to an embodiment of the present invention. [Diagram 2] 1 shows an example of the configuration of an MI platform system 100. [Diagram 3] 2 shows an example of a functional block of the AI ​​unit 108. [Figure 4] 13 shows an example of a flow of processing performed by the learning unit 351. [Diagram 5] 13 shows an example of a flow of processing performed by the inference unit 352. [Figure 6] 1 illustrates a schematic diagram of recipe provision. [Figure 7] 3 shows a variant of the training of the first model 301. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] In the following description, an "interface unit" may be one or more interface devices. The one or more interface devices may be at least one of the following: One or more I / O (Input / Output) interface devices. The I / O (Input / Output) interface devices are interface devices to at least one of the I / O devices and a remote display computer. The I / O interface device to the display computer may be a communications interface device. The at least one I / O device may be a user interface device, e.g., either an input device such as a keyboard and pointing device, or an output device such as a display device. One or more communication interface devices. The one or more communication interface devices may be one or more homogeneous communication interface devices (e.g., one or more NICs (Network Interface Cards)) or two or more heterogeneous communication interface devices (e.g., a NIC and an HBA (Host Bus Adapter)).

[0010] In the following description, a "memory" refers to one or more memory devices, which are an example of one or more storage devices, and may typically be a primary storage device. At least one memory device in the memory may be a volatile memory device or a non-volatile memory device.

[0011] In the following description, a "persistent storage device" may be one or more persistent storage devices, which are an example of one or more persistent storage devices. A persistent storage device may typically be a non-volatile storage device (e.g., an auxiliary storage device), and more specifically, may be, for example, a hard disk drive (HDD), a solid state drive (SSD), a non-volatile memory express (NVME) drive, or a storage class memory (SCM).

[0012] Also, in the following description, "storage device" may be at least memory, including memory and persistent storage device.

[0013] Furthermore, in the following description, a "processor" may be one or more processor devices. The at least one processor device may typically be a microprocessor device such as a CPU (Central Processing Unit), but may also be other types of processor devices such as a GPU (Graphics Processing Unit). The at least one processor device may be a single-core or multi-core. The at least one processor device may be a processor core. The at least one processor device may also be a processor device in a broader sense, such as a circuit that is a collection of gate arrays written in a hardware description language that performs part or all of the processing (e.g., a Field-Programmable Gate Array (FPGA), a Complex Programmable Logic Device (CPLD), or an Application Specific Integrated Circuit (ASIC)).

[0014] In the following description, functions may be described using the expression "yyy unit", but the functions may be realized by one or more computer programs being executed by a processor, or by one or more hardware circuits (e.g., FPGA or ASIC), or by a combination thereof. When a function is realized by a program being executed by a processor, the function may be at least a part of the processor, since the specified processing is performed using a storage device and / or an interface device, etc., as appropriate. Processing described with a function as the subject may be processing performed by a processor or a device having the processor. A program may be installed from a program source. The program source may be, for example, a program distribution computer or a computer-readable recording medium (e.g., a non-transitory recording medium). The description of each function is an example, and multiple functions may be combined into one function, or one function may be divided into multiple functions.

[0015] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the following description, "MI" is an abbreviation for materials informatics, and "DB" is an abbreviation for database.

[0016] FIG. 1 shows an example of a system configuration according to an embodiment of the present invention.

[0017] A material creation support system 10 that supports material creation is constructed. The material creation support system 10 may be a physical computer system, a logical computer system based on a physical computer system, or a combination of at least a part of a physical computer system and at least a part of a logical computer system. The physical computer system may be composed of one or more physical computers, and may include an interface device, a storage device, and a processor connected thereto. The logical computer system may include a virtual machine, or may include a system as a cloud computing service.

[0018] The material creation support system 10 includes a data presentation unit 50, a data acquisition unit 60, and a data conversion unit 70. These functions 50, 60, and 70 may be realized by causing a processor to execute a computer program.

[0019] The data presentation unit 50 periodically or whenever an inquiry is received, acquires data from the data mart 107, and presents the target data. Hereinafter, the data in the data mart 107 is referred to as "organized data." The target data is at least one of the acquired organized data and data based on an inference result obtained by inputting the organized data into a machine learning model. The data mart 107 is an example of a first data source. The data preferably includes data on ceramic materials.

[0020] The data acquisition unit 60 stores the experimental data, which is data obtained by an experiment based on the target data, in the electronic experiment notebook 131, the production DB 121, and the evaluation DB 122. The electronic experiment notebook 131, the production DB 121, and the evaluation DB 122 are examples of a second data source. In this embodiment, the "experimental data" may be data related to an experiment, such as a summary, details, result, or evaluation of the experiment.

[0021] The data conversion unit 70 acquires experimental data from the electronic experiment notebook 131, the production DB 121, and the evaluation DB 122, and converts the experimental data into data contained in an existing or new data mart 107.

[0022] After the organized data based on the experimental data is stored in the existing data mart 107, or after a new data mart 107 including the organized data based on the experimental data is created, the target data is based on the updated or newly created data mart 107, i.e., the organized data of the experimental data of the experiment performed based on the previously presented target data. Therefore, regardless of whether the type of material to be created is a material type for which highly accurate simulation data regarding the material properties can be obtained, the possibility of data useful for material creation can be increased. As a result, the types of materials that can be supported for material creation by the informatics method are expanded. Specifically, for example, even if the type of material to be created is an inorganic material (ceramic), data useful for material creation can be presented.

[0023] This embodiment will be described in detail below.

[0024] With respect to the material creation support system 10, the interface device may be communicably connected to at least one of the experiment system 110, the data acquisition unit 60, and the researcher terminal 11A. The storage device may store at least some of the data in the electronic experiment notebook 131, the production DB 121, the evaluation DB 122, the various data 150, the data lake 102, and the data mart 107. The processor may realize the above-mentioned functions 50, 60, and 70 by executing a computer program.

[0025] The data presentation unit 50 and the data conversion unit 70 are provided in the MI platform system 100. The data acquisition unit 60 is provided outside (or inside) the MI platform system 100. The MI platform system 100 may be a physical computer system, a logical computer system, or a combination of at least a part of a physical computer system and at least a part of a logical computer system. In this embodiment, as shown in FIG. 2, the MI platform system 100 is a physical computer system, and has an interface device 201, a storage device 202, and a processor 203 connected thereto. Communication with a researcher terminal 11 (at least a trainee terminal 11A) may be performed through the interface device 201. A data lake 102 and a data mart 107 may be provided in the storage device 202 and data may be stored therein. In addition, a program may be stored in the storage device 202. At least a part of the functions of the data presentation unit 50 and the data conversion unit 70 in the MI platform system 100 may be realized by the processor 203 reading and executing the program from the storage device 202.

[0026] The data presentation unit 50 includes an AI (Artificial Intelligence) unit 108 and an IF (Interface) unit 109.

[0027] The AI ​​unit 108 performs learning of the machine learning model and inference using the machine learning model. For example, in response to an instruction from the IF unit 109, the AI ​​unit 108 outputs to the IF unit 109 an inference result obtained by inputting the organized data acquired from the data mart 107 to the machine learning model.

[0028] The IF unit 109 receives an inquiry from the researcher terminal 11A and presents the target data to the researcher terminal 11A in response to the inquiry. The researcher terminal 11A is an information processing terminal (for example, a personal computer or smartphone) of the material researcher 5A (an example of a user). The researcher terminal 11A is an example of a sender of an inquiry for target data, and is also an example of a recipient of the target data. When the IF unit 109 receives an inquiry from the researcher terminal 11A (or periodically), it presents the target represented by the target data (organized data acquired from the data mart 107 and / or data based on an inference result obtained by instructing the AI ​​unit 108). The presented target may be, for example, a manufacturing recipe or material properties. In this embodiment, the "manufacturing recipe" means a method of creating a material, and may typically include a material composition and / or a synthesis process. The "material composition" may be, for example, a blending composition, and the "synthesis process" may be a raw material type (for example, different particle size) or process conditions. The manufacturing recipe may include a material development strategy.

[0029] The materials researcher 5A creates new materials and conducts experiments based on the presented target data. The presentation destination (transmission destination) of the target data may be a system such as an experiment system 110 instead of or in addition to the researcher terminal 11A.

[0030] The material researcher 5A performs an experiment using the experimental system 110 based on the presented subject. For example, the experiment using the experimental system 110 may be a combinatorial experiment. The experimental system 110 may include one or more devices, such as a ceramics production device 111 (e.g., an air-sintering furnace) and a ceramics evaluation device 112 (e.g., a device for evaluating the thermal expansion coefficient). Experimental data is output from the experimental devices such as the production device 111 and the evaluation device 112. The output experimental data is transmitted to and stored in the data server 120. The combinatorial experiment is an example of an experiment, and at least one of a high-throughput experiment, an automated experiment using a robot, and an experiment mainly performed by human labor may be adopted instead of or in addition to the combinatorial experiment.

[0031] The data acquisition section 60 includes a data server 120 and a lab notebook section 130 .

[0032] The data server 120 manages a DB in which the experimental data from the experimental system 110 is stored. A DB managed by the data server 120 may exist for each type of experimental equipment. For example, a creation DB 121 in which the experimental data from the production equipment 111 is stored, and an evaluation DB 122 in which the experimental data from the evaluation equipment 112 is stored may be managed. The experimental data from the experimental system 110 is structured by the data server 120 and stored in the DB.

[0033] The experiment notebook section 130 manages the experiment data acquired from the DB managed by the data server 120 as an electronic experiment notebook 131. The electronic experiment notebook 131 is structured data, for example a DB.

[0034] The data conversion unit 70 includes a collection unit 101, a feature amount calculation unit 103, an image analysis unit 104, a natural language analysis unit 105, and an organization unit 106.

[0035] The collection unit 101 collects experimental data from the electronic experiment notebook 131 and the DB of the data server 120, formats the collected experimental data, and stores the formatted experimental data in the data lake 102. "Forming" here means adjusting the structure of the experimental data to a predetermined structure.

[0036] Moreover, the collection unit 101 may collect various data 150. The various data 150 may include experimental data. Specifically, for example, the various data 150 may include at least one of numerical data of past experiments, a material database available for a fee or free of charge, language data representing past experiments (for example, experimental contents and experimental results), material data obtained by calculation such as simulation, and language data of patent documents and papers. There may be one or more data sources for the various data 150. For example, the data source of the numerical data of past experiments and the data source of the language data of patent documents and papers may be different. The various data 150 may include sensor data that is data including measurements by a sensor. At least a part of the various data 150 may be stored in the data lake 102.

[0037] The feature amount calculation unit 103 calculates the feature amount of one or more predetermined types of data in the shaped experimental data. As a result, each of the one or more types of data is quantified, and as a result, processing by the AI ​​unit 108 becomes possible. For example, the one or more types of data may be image data and text data. The image data is analyzed by the image analysis unit 104, and the feature amount calculation unit 103 calculates the feature amount of the image data based on the result of the analysis. In addition, the text data is text mined by the natural language analysis unit 105, and the feature amount calculation unit 103 calculates the feature amount of the text data based on the result of the text mining. The above-mentioned neural network may be used for the feature amount calculation. A copy of the experimental data may be generated in the data lake 102, and each of the one or more predetermined types of data in the copy may be converted into a calculated feature amount (numerical value).

[0038] The organizer 106 organizes the data lake 102 into one or more data marts 107 as a data set that meets a predetermined condition. For example, a data mart 107 including organized data that can be presented as at least a part of the target data not via the AI ​​unit 108 and a data mart 107 including organized data that can be presented as at least a part of the target data via the AI ​​unit 108 may be generated.

[0039] In this way, organized data based on the experimental data collected and formatted in the MI platform system 100 is prepared, and the target data based on the organized data is presented to the materials researcher 5A by the MI platform system 100.

[0040] Also, data may be presented to the materials researcher 5A without going through the MI platform system 100. For example, another materials researcher 5B may use the trainee terminal 11B to obtain experimental data from the electronic experiment notebook 131, organize the experimental data, and input it to the AI ​​unit 140. The AI ​​unit 140 may be realized outside (or inside) the trainee terminal 11B. As with the AI ​​unit 108, the AI ​​unit 140 may also perform inference by inputting the organized data that has been input into a machine learning model. Data based on the results of this inference may be presented to the materials researcher 5A. The materials researcher 5A may perform an experiment based on this presented data. The data of this experiment may also be stored in the data server 120 from the experiment system 110 and collected in the MI platform system 100.

[0041] A predetermined function implemented inside or outside the researcher terminal 11A may promote experiments based on the presented target data. For example, the function may determine what kind of experiment is recommended to the materials researcher 5A, and present the determined items to the materials researcher 5A. Alternatively, the target data may include data indicating what kind of experiment is recommended.

[0042] According to the above-mentioned embodiment, it is expected that the cycle of presenting target data → experiment → collecting experimental data → converting the experimental data into organized data → presenting target data based on the organized data can be performed efficiently and quickly. In this way, MI based on experimental data is realized. Therefore, it is possible to support the creation of materials of a material type for which it is difficult to obtain highly accurate simulation data on the material properties, such as materials with complex high-order structures of ceramics.

[0043] Furthermore, according to the above-described embodiment, various data inside or outside the organization (e.g., a company) that provides the MI platform system 100 can be aggregated, structured, and managed in the MI platform system 100, and reflected in the data mart 107. This increases the likelihood that the presented target data will be useful for material creation.

[0044] The following expressions are possible based on the above description. The following expressions may include supplementary explanations or modified explanations of the above description. <Expression 1> a data presentation unit (50) that periodically or each time an inquiry is received, acquires data from a first data source, and presents target data that is at least one of the data and data based on an inference result obtained by inputting the data into a machine learning model; a data acquisition unit (60) that stores experimental data, which is data obtained by an experiment based on the target data, in a second data source; a data conversion unit (70) that acquires experimental data from the second data source and converts the experimental data into data contained in an existing or new first data source; A materials creation support system equipped with: <Expression 2> At least one of the data obtained from the first data source, the presented subject data, the experimental data, and the data included in the existing or new first data source includes data related to a ceramic material; The material creation support system according to expression 1. <Expression 3> (A) periodically or upon receipt of a query, acquiring data from a first data source, and presenting target data that is at least one of the data and data based on an inference result obtained by inputting the data into a machine learning model; (B) storing experimental data, which is data obtained by an experiment based on the subject data, in a second data source; (C) obtaining experimental data from the second data source and converting the experimental data into data contained in an existing or new first data source; Material creation support method. <Expression 4> The machine learning model acquires updated data contained in an existing or new first data source by acquiring the experimental data, and learns parameters. A material creation support method according to expression 3. Specifically, for example, it is as follows. Based on the experimental data acquired by (C) (e.g., experimental data acquired by the data conversion unit 70), the data in the existing data mart 107 is updated, or a new data mart 107 including data based on the acquired experimental data is generated. Therefore, the existing or new data mart 107 includes update data that is data based on the acquired experimental data. The AI ​​unit 108 may acquire the update data from the data mart 107 and learn (e.g., update) parameters of the machine learning model used by the AI ​​unit 108. <Expression 5> Collecting and converting existing or new experimental data into data contained in a first data source; periodically or upon receipt of a query, acquiring data from a first data source, and presenting target data that is at least one of the data and data based on an inference result obtained by inputting the data into a machine learning model; Material creation support method. For example, this material creation support method may be performed by the MI platform system 100. In other words, the material creation support system 10 may have the data presentation unit 50 and the data conversion unit 70, but may not have the data acquisition unit 60. <Expression 6> At least one of the data obtained from the first data source, the presented object data, the experimental data, the data included in the existing or new first data source, and the data based on the inference result includes data related to a ceramic material; A material creation support method according to any one of expressions 3 to 5. <Expression 7> (A) periodically or upon receipt of a query, acquiring data from a first data source, and presenting target data that is at least one of the data and data based on an inference result obtained by inputting the data into a machine learning model; (B) storing experimental data, which is data obtained by an experiment based on the subject data, in a second data source; (C) obtaining experimental data from the second data source and converting the experimental data into data contained in an existing or new first data source; (D) executing a learning process for a machine learning model that uses the learning data included in the first data source of (C) to output a recommended inference result indicating a recipe or a material composition; A method for generating a material creation support model. <Expression 8> At least one of the data acquired from the first data source, the presented target data, the experimental data, the data included in the existing or new first data source, the training data, and the data representing the recommended inference result includes data related to a ceramic material; A method for generating a material creation support model according to Representation 7. <Expression 9> A program for causing a computer to execute the material creation support method according to any one of expressions 3 to 6. <Expression 10> A program for causing a computer to execute the method for generating a material creation support model according to Representation 7 or 8.

[0045] The AI ​​unit 108 according to this embodiment will be described in detail below. In the following description, the material to be created is assumed to be ceramics. In the following description, the definitions of terms are as follows. A "dataset" is a logical block of electronic data seen by a program such as an application program, and may be, for example, a record, a file, a key-value pair, or a tuple. "Researcher input information" is information input by the researcher 5A. The researcher input information is input to the IF section 109. In this embodiment, the researcher input information is typically information that represents a manufacturing recipe or material properties. "Researcher output information" is information that is output (provided) to the researcher 5A. The researcher output information is output from the IF unit 109. In this embodiment, the researcher output information is typically information that represents material properties or manufacturing recipes. "Recipe data set" is a data set that represents a recipe for a material. When the researcher input information is information that represents a recipe, the recipe data set may be a data set based on the researcher input information. "Material property dataset" is a dataset that represents material properties. When the researcher output information is information that represents material properties, the researcher output information may be information based on the material property dataset.

[0046] FIG. 3 shows an example of a functional block of the AI ​​unit 108.

[0047] The AI ​​unit 108 may receive a dataset from the IF unit 109 (i.e., a dataset based on researcher input information may be generated by the IF unit 109), or may receive researcher input information from the IF unit 109 and generate a dataset based on the researcher input information. The AI ​​unit 108 may also output a dataset to the IF unit 109 (i.e., researcher output information based on the dataset may be generated by the IF unit 109), or may output researcher output information based on the dataset to the IF unit 109. The IF unit 109 may also be included in the AI ​​unit 108.

[0048] In the inference using the machine learning model, the AI ​​unit 108 inputs the manufacturing recipe data set 311 and outputs (predicts) the material property data set 313. In addition, in this embodiment, when the researcher input information is information representing material properties, the AI ​​unit 108 identifies the manufacturing recipe data set 311 of the manufacturing recipe that is expected to create a material having the material properties, and outputs the manufacturing recipe data set 311 (or the researcher output information based on the manufacturing recipe data set 311) to the IF unit 109. That is, if the researcher 5A inputs information representing the manufacturing recipe to the MI platform system 100, the researcher 5A can receive information representing the predicted material properties of a material assumed to be created by the manufacturing recipe from the MI platform system 100. In addition, if the researcher 5A inputs information representing the material properties to the MI platform system 100, the researcher 5A can receive information representing the manufacturing recipe of a material that is expected to create a material having the material properties from the MI platform system 100.

[0049] The AI ​​unit 108 has a learning unit 351 that learns a machine learning model, and an inference unit 352 that performs inference using the learned machine learning model (i.e., the inference described above). The machine learning models include first to third models 301 to 303. The first to third models 301 to 303 are stored in the storage device 202.

[0050] The first model 301 is a model that inputs a manufacturing recipe data set 311 and outputs a material property data set 313. Ceramics have a complex higher-order structure (e.g., grain size, grain size distribution, grain boundary phase, etc.) in which many crystal particles are gathered, and since material characteristics including such higher-order structure affect the material properties of ceramics, it is generally difficult to predict the material properties with high accuracy from the manufacturing recipe.

[0051] Therefore, in this embodiment, based on the fact that material features affect material properties, in other words, based on the fact that there is a correlation between material features and material properties, a second model 302 is prepared, which inputs a material feature dataset 312 (a dataset representing material features) and outputs a material property dataset 313. Then, in order to make it possible to predict the material property dataset 313 from the manufacturing recipe dataset 311 via an intermediate dataset called the material feature dataset 312, a third model 303 is prepared, which inputs the manufacturing recipe dataset 311 and outputs the material feature dataset 312.

[0052] The "recipe" represented by the recipe recipe dataset 311 may include material compositions and / or synthesis processes, as described above, and specifically may include, for example, one or more recipe items (e.g., synthesis temperature) and values ​​(typically numerical values) for each of the one or more recipe items.

[0053] The "material properties" represented by the material property dataset 313 refer to the properties exhibited by a substance (e.g., a structure) as a material, and specifically may include, for example, one or more property items (e.g., strength, thermal expansion coefficient, etc.) and values ​​(typically numerical values) for each of the one or more property items.

[0054] The "material characteristic" represented by the material characteristic dataset 312 means a physical and / or chemical state exhibited by a substance as a material, and specifically may include, for example, one or more characteristic items (e.g., crystal grain size, grain size distribution, etc.) and a value (typically a numerical value) for each of the one or more characteristic items. A feature amount may be included in the material characteristic dataset 312 for each characteristic item. For example, using the second model 302, it is possible to predict a material property including values ​​for one or more characteristic items from a material characteristic including values ​​for multiple characteristic items, or to predict a material property including values ​​for one or more characteristic items from a material characteristic including a value for one characteristic item.

[0055] The material feature dataset 312 may be a dataset representing features calculated by the feature calculation unit 103, and may be stored in the data mart 107 and acquired from the data mart 107. One or more features represented by the material feature dataset 312 are extracted by the feature calculation unit 103 from image data (e.g., SEM image) or spectrum data using data processing (e.g., processing using CNN). Note that SEM is an abbreviation for "Scanning Electron Microscope" and CNN is an abbreviation for "Convolutional Neural Network".

[0056] For each of the first to third models 301 to 303, the model may be a linear regression (e.g., Ridge regression, Lasso regression, Elastic Net regression), a logistic regression, a Support Vector Machine (SVM), a decision tree model (e.g., Random Forest, XGBoost (Xtreme Gradient Boosting), LightGBM (Light Gradient Boosting Machine)), a neural network (e.g., CNN, RNN (Recurrent Neural Network), ResNet (Residual Network)), Bayesian optimization, k-NN (k-Nearest Neighbor), or a combination of any two or more of these models. Preferably, for each of the first to third models 301 to 303, the model may be a random forest, XGBoost, LightGBM, CNN, Bayesian optimization, or a combination of any two or more of these models. For example, the following is expected. · Although the number of recipe data sets 311 used for learning is relatively small, Bayesian optimization is capable of making predictions that take into account prediction deviations, and is excellent at finding candidate points for adding data, and finding the optimal value while increasing the amount of data. Decision tree models (e.g., Random Forest, XGBoost, LightGBM) are a combination of simple binomial distribution problems, so they have low computational costs and high accuracy. For example, XGBoost or LightGBM creates multiple decision trees and weights them to deal with targets with prediction errors, preventing overfitting and increasing robustness. CNN has a high degree of freedom, can handle complex systems, and is useful when there are many features. By appropriately controlling the number of learning times (typically the number of epochs), it is possible to adjust the degree of overfitting.

[0057] In this embodiment, each of the first to third models 301 to 303 is a regression equation.

[0058] FIG. 4 shows an example of a flow of processing performed by the learning unit 351.

[0059] The learning unit 351 acquires learning data for the first model 301, specifically, one or more recipe recipe data sets 311 and one or more material property data sets 313 corresponding to the one or more recipe recipe data sets 311 (S401). These data sets 311 and / or 313 may be input from the researcher 5A in the learning process, may be prepared in advance as learning data, may be acquired from data such as experimental data and / or various data 150, or may be a recipe recipe data set 311 (a recipe recipe data set 311 newly generated by the learning unit 351) that represents a recipe with a recipe item or value range narrowed down in a recipe represented by a certain recipe recipe data set 311 (or a recipe with an expanded recipe item or value range).

[0060] The learning unit 351 learns the first model 301 using the acquired data sets 311 and 313 (S402). Specifically, for example, the learning unit 351 learns the first model 301 using the recipe data set 311 and a material property data set 313 (the recipe and the material property of the material created by the recipe) corresponding to the recipe data set 311.

[0061] The learning unit 351 judges whether or not to end the learning of the first model 301 (S403). For example, when the accuracy (prediction accuracy) of the first model 301 is equal to or greater than the first accuracy threshold (or when the accuracy of the first model 301 is higher than the accuracy of the second model 302 and the third model 303), the judgment result of S403 is true (S403: Yes). In this case, the processing of the learning unit 351 ends.

[0062] When the determination result of S403 is false (S403: No), the learning unit 351 acquires learning data for the second model 302, specifically, one or more material feature data sets 312 and one or more material property data sets 313 corresponding to the one or more material feature data sets 312 (S404). These data sets 312 and / or 313 may be input by the researcher 5A in the learning process, may be prepared in advance as learning data, may be acquired from data such as experimental data and / or various data 150, or may be a material feature data set 312 (a material feature data set 312 newly generated by the learning unit 351) that represents a material feature with a narrowed feature item or value range in the material feature represented by a certain material feature data set 312 (or a material feature with an expanded feature item or value range).

[0063] The learning unit 351 learns the second model 302 using the acquired data sets 312 and 313 (S405). Specifically, for example, the learning unit 351 learns the second model 302 using the material feature data set 312 and a material property data set 313 (material feature and material property of the material having the material feature) corresponding to the material feature data set 312. The purpose of this learning is to find a material feature that has a strong correlation with the material property, in other words, to identify a material feature that corresponds to the material property.

[0064] The learning unit 351 judges whether or not to end the learning of the second model 302 (S406). For example, when the accuracy (prediction accuracy) of the second model 302 is equal to or greater than the second accuracy threshold, the judgment result of S406 is true (S406: Yes). The accuracy of the second model 302 being equal to or greater than the threshold means that important feature items and / or values ​​(feature amounts) have been identified. In other words, when important feature items and / or values ​​(feature amounts) are not identified, typically the accuracy of the second model 302 does not become equal to or greater than the threshold.

[0065] When the judgment result of S406 is true, the learning unit 351 acquires learning data for the third model 303, specifically, one or more material characteristic data sets 312 identified in the learning of the second model 302, and one or more recipe data sets 311 corresponding to the one or more material characteristic data sets 312 (S407). The recipe data set 311 corresponding to the material characteristic data set 312 may be, for example, a recipe recipe data set 311 corresponding to the material property data set 313 corresponding to the material characteristic data set 312. These data sets 312 and / or 311 may be data sets acquired from the data sets used in this learning process, or may be recipe recipe data sets 311 (newly generated by the learning unit 351) representing a recipe with a recipe item or value range narrowed in a recipe represented by a certain recipe recipe data set 311 (or a recipe with an expanded recipe item or value range).

[0066] The learning unit 351 learns the third model 303 using the acquired data sets 312 and 311 (S408). Specifically, for example, the learning unit 351 learns the third model 303 using the material feature data set 312 and the recipe data set 311 (material feature and recipe for creating a material having the material feature) corresponding to the material feature data set 312.

[0067] The learning (S407 and S408) of the third model 303 may be performed in parallel with the learning (S404 and S405) of the second model 302. The material characteristic dataset 312 acquired in S407 may be a dataset representing feature amounts identified (narrowed down) in the learning of the second model 302, or may be the material characteristic dataset 312 before such identification.

[0068] The learning unit 351 judges whether or not to end the learning of the third model 303 (S409). If the accuracy (prediction accuracy) of the third model 303 is equal to or greater than the third accuracy threshold, the judgment result of S409 is true (S409: Yes). Note that the above-mentioned first to third accuracy thresholds may be the same value or different values.

[0069] When the determination result of S409 is true, the learning unit 351 performs S401 again. In S401, the recipe data set 311 identified (or newly generated) in the learning of the third model 303 and the material property data set 313 corresponding to the recipe data set 311 may be acquired.

[0070] 4, as described above, when the accuracy of the first model 301 becomes equal to or greater than the first accuracy threshold and the learning ends, the processing of the learning unit 351 ends. Note that, even if the determination in S403 is performed N times (N is an integer equal to or greater than 2), if the accuracy of the first model 301 is less than the first accuracy threshold (if the determination result in S403 is not true), the processing of the learning unit 351 may end.

[0071] In this embodiment, "relearning" means that the processing of the learning unit 351 is once terminated and then the processing of the learning unit 351 is performed again. The relearning may be started at a predetermined opportunity. The predetermined opportunity may be, for example, at least one of the following: For at least one of models 301 to 303, the amount of training data has increased sufficiently since the previous training (for example, the amount of training data has increased beyond a certain amount). The accuracy of at least one of the models 301 to 303 (for example, the first model 301) has deteriorated below a first accuracy threshold.

[0072] The inference by the inference unit 352 may be performed in parallel with the processing by the learning unit 351. Furthermore, a data set obtained in the inference by the inference unit 352 may be used for learning any one of the models 301 to 303. Specifically, for example, at least one of the following may be adopted. When the determination result of S403 is true (when the learning of the first model 301 is completed), inference is performed using the first model 301, and a manufacturing recipe may be provided appropriately. If the determination result in S406 is false (if the learning of the second model 302 is not completed), inference may be performed using the second model 302, and material characteristics may be provided appropriately. Inference using the second model 302 may be possible, for example, when the difference between the accuracy of the second model 302 and the second accuracy threshold is equal to or smaller than a predetermined difference. When the determination result of S409 is false (when the learning of the third model 303 is not completed), inference using the third model 303 may be performed, and a manufacturing recipe may be provided appropriately. Inference using the third model 303 may be possible, for example, when the difference between the accuracy of the third model 303 and a third accuracy threshold is equal to or smaller than a predetermined difference. If the judgment result of S409 is true (if the learning of the third model 303 has been completed) and if the judgment result of S403 is false (if the learning of the first model 301 has not been completed), inference is performed using the second model 302 and the third model 303, and a manufacturing recipe may be provided as appropriate. Based on the recipe provided by the researcher 5A after inference is performed, or based on the recipe estimated by the researcher 5A after inference is performed and provided, an experiment may be performed under new conditions determined by the researcher 5A, and experimental data representing the recipe, material properties, and material properties related to the experiment may be accumulated. The recipe recipe data set 311, material property data set 312, and material property data set 313 obtained from the experimental data may be used in learning or re-learning to make any of the judgment results of S403, S406, and S409 true. The new conditions may include conditions based on the inference results, or may include conditions determined based on the knowledge of the researcher 5A without any relation to the inference results.

[0073] FIG. 5 shows an example of the flow of processing performed by the inference unit 352.

[0074] The inference unit 352 judges whether or not re-learning has been performed (S501). When the judgment result of S501 is true (S501: Yes), the inference unit 352 invalidates the list (S502). The "list" referred to here is data generated or updated in inference, and is data in which the relationship between the material properties and the recipe and / or the material features is recorded. The list is stored in the storage device 202 (see FIG. 2) by the inference unit 352, and can be referred to by the IF unit 109. In addition, "invalidating" the list means making the list unreachable by the IF unit 109, and specifically, for example, may be deleting the list from the storage device 202, may be associating data indicating invalidity with the list, or may be saving the list from one storage area to another storage area. In addition, S502 may not be performed. For example, every time a list is newly generated or updated, the latest list may be made valid (a list referred to for providing the recipe or material properties).

[0075] The inference unit 352 judges whether the learning of the first model 301 is completed (S503). When the judgment result of S503 is true (S503: Yes), the inference unit 352 performs inference using the first model 301 (S504). Specifically, for example, the inference unit 352 may acquire a plurality of recipe recipe data sets 311, and may acquire the material property data set 313 by inputting the recipe recipe data set 311 to the first model 301 for each acquired recipe recipe data set 311. Each of the plurality of recipe recipe data sets 311 may be a data set acquired from newly added experimental data or various data 150, or may be a data set generated by changing the combination of recipe items or the value of each recipe item (for example, the inference unit 352 may increase the recipe recipe data set 311 by repeatedly changing the recipe recipe, such as increasing or decreasing the value of the synthesis temperature by a fixed value). The inference unit 352 records in a list, for each pair of the input recipe data set 311 and the acquired material property data set 313, the recipe represented by the recipe recipe data set 311 and the material property represented by the material property data set 313. In addition, the material property may be recorded in the list when the material property satisfies a predetermined condition (for example, when the value of a certain property item is within a predetermined value range).

[0076] When the judgment result of S503 is false (S503: No), the inference unit 352 judges whether the learning of the third model 303 has been completed (S505). When the judgment result of S505 is true (S505: Yes), the inference unit 352 performs inference using the second and third models 302 and 303 (S506). Specifically, for example, the inference unit 352 acquires the material property data set 313 by inputting the material property data set 312 to the second model 302. To create a material having material properties represented by the acquired material property data set 313 (i.e., to specify the recipe data set 311 from which the material property data set 312 inputted to acquire the material property data set 313 is the output), the inference unit 352 may acquire the material property data set 312 by inputting the recipe data 311 to the third model 303. The inference unit 352 records in a list the recipe represented by the recipe recipe data set 311, the material feature represented by the material feature data set 312, and the material property represented by the material property data set 313 for each pair of the input recipe recipe data set 311, the acquired material feature data set 312, and the acquired material property data set 313. The list may further record the material feature corresponding to the recipe recipe and the material property. In addition, the material property may be recorded in the list when the material property satisfies a predetermined condition (for example, when the value of a certain property item is within a predetermined value range). In addition, each of the multiple recipe recipe data sets 311 or material property data sets 312 may be a data set acquired from newly added experimental data or various data 150, or may be a data set generated by changing the combination of each item or the value of each item (for example, the inference unit 352 may increase the recipe recipe data set 311 by repeatedly changing the recipe recipe, such as increasing or decreasing the value of the synthesis temperature by a certain value).

[0077] In addition to determining whether or not the learning of the third model 303 has been completed, in S505, the inference unit 352 may determine whether or not the accuracy of the inference using the second and third models 302 and 303 is equal to or higher than a threshold. If the result of this determination is also true, the inference unit 352 may perform S506 (inference using the second and third models 302 and 303).

[0078] If the judgment result of S505 is false (S505: No), the inference unit 352 judges whether the learning of the second model 302 is completed (S507). If the judgment result of S507 is true (S507: Yes), the inference unit 352 performs inference using the second model 302 and the third model 303 (S508). Specifically, for example, the inference unit 352 first uses the second model 302 to identify material features (material feature data set 312 as input such that the material feature data set 313 representing the target material property is output) necessary to satisfy the target material property. Then, the inference unit 352 uses the third model 303 to infer a recipe for obtaining the identified material feature (specifically, identifies recipe recipe data set 311 as input such that the material feature data set 312 representing the identified material property is output). The inference unit 352 records the recipe, material features, and material properties in a list for each pair of the recipe recipe data set 311 as input, the acquired material feature data set 312, and the acquired material property data set 313. Each of the multiple recipe recipe data sets 311 or material property data sets 312 may be a data set acquired from newly added experimental data or various data 150, or may be a data set generated by changing the combination of recipe items or the value of each recipe item (for example, the inference unit 352 may increase the recipe recipe data set 311 by repeatedly changing the recipe recipe, such as increasing or decreasing the value of the synthesis temperature by a fixed value).

[0079] When the determination result of S507 is false (S507: No), the inference unit 352 performs inference using two models, the second and third models 302 and 303, or the first model 301 (S509). Specifically, for example, the inference unit 352 performs the following (S509-1) or (S509-2). More specifically, for example, when the prediction accuracy of the second model 302 and the third model 303 is higher than the prediction accuracy of the first model 301, the inference unit 352 may perform (S509-1), and when the prediction accuracy of the second model 302 and the third model 303 is the same as or lower than the prediction accuracy of the first model 301, the inference unit 352 may perform (S509-2). In the following, each of the multiple manufacturing recipe data sets 311 or material characteristic data sets 312 may be a data set obtained from newly added experimental data or various data 150, or may be a data set generated by changing the combination of each item or the value of each item (for example, the inference unit 352 may increase the number of manufacturing recipe data sets 311 by repeatedly changing the manufacturing recipe, such as increasing or decreasing the value of the synthesis temperature by a fixed value). (S509-1) The inference unit 352 inputs the material characteristic data set 312 to the second model 302 to obtain the material characteristic data set 313. To create a material having material characteristics represented by the obtained material characteristic data set 313 (i.e., to specify the recipe data set 311 from which the material characteristic data set 312 input to obtain the material characteristic data set 313 is the output), the inference unit 352 inputs the recipe data 311 to the third model 303 to obtain the material characteristic data set 312. For each pair of the input recipe data 311, the obtained material characteristic data set 312, and the obtained material characteristic data set 313, the inference unit 352 records in a list the recipe represented by the recipe data 311, the material characteristics represented by the material characteristic data set 312, and the material characteristics represented by the material characteristic data set 313. (S509-2) The inference unit 352 inputs the recipe data 311 to the first model 301 to obtain the material property data 313 (in an actual experiment, a material property data set 312 may also be collected for evaluation of material properties). The inference unit 352 records in a list the recipe represented by the recipe recipe data set 311 and the material property represented by the material property data set 313 for each pair of the input recipe recipe data set 311 and the obtained material property data set 313.

[0080] The inference unit 352 judges whether or not to end the inference process (S510). If the judgment result of S510 is true (S510: Yes), the process ends. If the judgment result of S510 is false (S510: No), the process returns to S501.

[0081] In the process shown in Fig. 5, a list 600 shown in Fig. 6 is generated or updated. When the IF unit 109 receives an inquiry from a researcher 5A specifying material properties, the IF unit 109 identifies a manufacturing recipe for creating a material having the material properties based on the list 600, and provides the identified manufacturing recipe to the researcher 5A.

[0082] Although one embodiment has been described above, this is merely an example for explaining the present invention, and the scope of the present invention is not limited to this embodiment. The present invention can be implemented in various other forms.

[0083] For example, the AI ​​unit 140 shown in FIG. 1 may have the same functions as the AI ​​unit 108 described above.

[0084] Furthermore, for example, the "experimental data" collected by the collection unit 101 may include at least some of the following data: data from combinatorial experiments, data acquired from the electronic experiment notebook 131 in which past experimental data is accumulated, and data acquired from an external site. Furthermore, the "experimental data" may include computational data (for example, data predicted by first-principles calculations or separate machine learning) or may include language data. Furthermore, at least one of the data sets 311 to 313 may also include computational data or language data. In addition, images may be generated during learning, but in the embodiment, image generation is not required during learning. This is expected to reduce the calculation load and simplify the model.

[0085] Furthermore, for example, the above description can be summarized as follows. The following summary may include supplementary explanations and explanations of modifications of the above description.

[0086] The material creation support system 10 may include, for example, an interface device 201 connected to a user device (for example, a researcher terminal 11A) that is a device having an input device and a display device, a storage device 202 in which models 301 to 303 are stored, and a processor 203 connected to the interface device 201 and the storage device 202. The processor 203 may implement an AI unit 108 and an IF unit (interface unit) 109 by, for example, executing a computer program. The AI ​​unit 108 performs inference using the first model 301 or inference using the second model 302 and the third model 303 depending on the accuracy (for example, prediction accuracy) of at least one of the first to third models 301 to 303, and generates or updates a list 600 (an example of recipe characteristic data that is data representing the association between a recipe and material characteristics). The IF unit 109 may receive an inquiry in which material characteristics are specified, identify a recipe corresponding to the characteristics specified in the inquiry based on the list 600, and provide the identified recipe.

[0087] A second model 302 is prepared based on the correlation between the material features and the material properties, and a third model 303 is prepared to enable prediction of the material property data set 313 from the manufacturing recipe data set 311 via an intermediate data set called the material feature data set 312. The existence of the second and third models 302 and 303 can contribute to the proposal of a manufacturing recipe for creating a material having the material properties desired by a researcher, even if the accuracy of the first model 301 is insufficient due to an insufficient amount of experimental data or other reasons.

[0088] When the accuracy of the first model 301 is equal to or greater than the first accuracy threshold, or when the accuracy of the first model 301 is higher than the accuracy of the second and third models 302 and 303, the AI ​​unit 108 may generate or update the list 600 in the inference using the first model 301. This allows the inference for generating or updating the list 600 used to provide the recipe to be performed with a smaller calculation load (without inputting and outputting the intermediate dataset as the material characteristic dataset 312). Note that the "accuracy of the second model 302 and the third model 303" does not refer to the individual accuracy of the second model 302 and the third model 303, but refers to the accuracy when both the second model 302 and the third model 303 are used.

[0089] The AI ​​unit 108 may train the first model 301 using the recipe recipe data set 311 and the material property data set 313 identified or generated in the learning or inference of the second and third models 302 and 303. For example, in the learning or inference of the second and third models 302 and 303, it may be identified that it is preferable that the recipe item of synthesis time is present or absent in the recipe recipe. In this way, in the learning or inference of the second and third models 302 and 303, the recipe items are expanded or reduced, and the values ​​of the recipe items are expanded or reduced, and as a result, it is expected that the recipe recipe data set 311 preferable for training the first model 301 will increase. Therefore, it is expected that the accuracy of the first model 301 will be improved.

[0090] For example, since a learning data set is acquired from experimental data, if the experimental data is insufficient, the learning data set is also insufficient, and as a result, the accuracy of the models 301 to 303 may also be insufficient. In addition, it takes time for the experimental data to become sufficient, and it is difficult to guarantee when the data will be sufficient. Therefore, the AI ​​unit 108 may perform inference using the first model 301 and / or inference using the second and third models 302 and 303 in parallel with the learning of the models 301 to 303. For example, when the learning is switched (depending on which model has finished learning), the AI ​​unit 108 may switch (select) the model used in the inference.

[0091] In the training of the second model 302, the AI ​​unit 108 may identify or generate one or more material feature data sets 312 from the multiple material feature data sets 312 as factors that cause the accuracy of the second model 302 to be equal to or greater than the second accuracy threshold. In the training of the third model 303, the AI ​​unit 108 may identify or generate one or more recipe recipe data sets 311 corresponding to the one or more material feature data sets 312 identified or generated in the training or inference of the second model 302 from the multiple recipe recipe data sets 311. The AI ​​unit 108 may train the first model using one or more recipe recipe data sets 311 identified or generated in the training or inference of the third model 303 and one or more material property data sets 313 corresponding to the one or more recipe recipe data sets 311. As a result, the material feature data sets 312 are narrowed down in the training or inference of the third model 303, so that the accuracy of the third model 303 can be improved with a small calculation load. Furthermore, since the manufacturing recipe data set 311 corresponding to such a narrowed-down material characteristic data set 312 is used for learning the first model 301, the accuracy of the first model 301 can be improved with a small calculation load.

[0092] The material feature data set 312 may be a data set composed of feature amounts as numerical values ​​for each feature item belonging to the material feature, which can reduce the computational load required for learning and inference, and can simplify the second and third models 302 and 303.

[0093] When re-learning is performed, the AI ​​unit 108 may discard the list 600 and newly generate or update the list 600 in inference using the first model 301 after re-learning and / or inference using the second model 302 after re-learning and the third model 303 after re-learning. This makes it possible to avoid the IF unit 109 providing, as a recipe, a list generated or updated in inference using an old model before accuracy is improved.

[0094] Also, as one modified example, the following modified example may be used. That is, for example, as shown in FIG. 7, when the accuracy of the first model 301 is less than the first accuracy threshold (for example, when the accuracy of the first model 301 does not become equal to or greater than the first accuracy threshold even after performing the learning of the first model 301 a predetermined number of times), the AI ​​unit 108 may learn the first model 301 (learn the correspondence between the set of recipe recipes and material features and the material properties) by using, as input to the first model 301, the recipe recipe data set 311 specified or generated in the learning or inference of the second model 302 in addition to the recipe recipe data set 311 specified or generated in the learning of the second model 302. This is expected to increase the possibility that the accuracy of the first model 301 can be made equal to or greater than the first accuracy threshold. In this case, the AI ​​unit 108 may generate or update a list 600 in which material features are associated with material properties in addition to the recipe (i.e., a list 600 in which the relationships between the recipe, material features, and material properties are recorded) in the inference using the first model 301. The IF unit 109 may receive a query in which material features are specified in addition to the material properties, identify a recipe corresponding to the material properties and material features specified in the query based on the list 600, and provide the identified recipe.

[0095] As one use case, for example, the following use case may be included. That is, the material characteristics may include (1) density and porosity quantified by Archimedes' method, and (2) SEM image feature amount (for example, composition area ratio and composition area deviation). Feature items (feature factors) that may affect the target characteristic (material characteristic desired by the researcher) may be porosity, crystal perimeter, composition area ratio, and composition area deviation ("composition area deviation" is a feature amount that represents the variation in composition ratio of a plurality of divided regions that constitute one image (here, SEM image)). The first model 301 may be Ridge regression (the input may be manufacturing conditions such as composition ratio, material relay, and firing temperature, and the output may be bending strength and thermal expansion coefficient). The second model 302 may be Ridge regression or LightGBM (the input may be the above-mentioned (1) density and porosity, and (2) SEM image feature amount, and the output may be bending strength and thermal expansion coefficient). The third model 303 may be a Ridge regression (the input may be manufacturing conditions such as composition ratio, material grain size, and firing temperature, and the output may be the above (1) density and porosity, and (2) SEM image feature quantity). In response to an inquiry from the researcher 5A (an inquiry specifying material properties), a manufacturing recipe may be provided (displayed on the researcher terminal 11A). Also, any of the models 301 to 303 may include a regression equation for each element (e.g., objective variable) as an output. That is, any of the models 301 to 303 may include one or more regression equations. Also, for any of the models 301 to 303, the data set as an input may be common, but typically, explanatory variables and weightings are different for each regression equation in the model.

[0096] The significance of the process illustrated in FIG. 5 is, for example, as follows.

[0097] That is, the purposes of inference can be (1) to find a recipe for producing a material with desired material properties, and (2) to find a candidate recipe for adding experimental data to improve model accuracy. The inference in S504 in Fig. 5 corresponds to the purpose of (1) (because model learning has been completed). On the other hand, the inferences in S506, S508, and S509 in Fig. 5 correspond to the purpose of (2) (because model learning has not been completed).

[0098] From among the combinations (e.g., combinations of recipes and material properties) identified in the inference corresponding to the objective of (2), a dataset that is useful or effective for model learning (e.g., recipe) is selected, and a dataset obtained by actually conducting experiments based on that dataset is added as a learning dataset, and model learning is performed using the added dataset. Since the dataset is augmented in this way, the model can be trained with greater accuracy.

[0099] If S505: Yes, the accuracy of the inference using the second model 302 and the third model 303 can be considered sufficient. Therefore, an additional learning dataset can be prepared based on the dataset input and output in the inference. In addition to the dataset input and output in the inference, a dataset thought up and generated by a researcher (e.g., an experimenter) may also be used to prepare the additional learning dataset.

[0100] On the other hand, if S505: No, it can be considered that the accuracy of the inference using the second model 302 and the third model 303 is insufficient. Therefore, it is difficult to prepare an additional learning dataset based only on the dataset input and output in the inference, and a dataset created by a researcher may be used to prepare an additional learning dataset.

[0101] Since the second model 302 is trained before the third model 303 is trained, it is more efficient to perform the determination in S505 before S507. Note that the order of the determinations is not limited to the order illustrated in FIG. 5 (i.e., the order of S503 → S505 → S507).

[0102] Based on the above explanation, the following expression is possible.

[0103] The AI ​​unit 108 may perform inference using the second model 302 and the third model 302 when the accuracy of the first model 301 is less than a first threshold or when the accuracy of the first model 301 is lower than the accuracy of the second model 302 and the third model 303.

[0104] The AI ​​unit 108 may train the second model 302 prior to the third model 303. The AI ​​unit 108 may prioritize the use of a data set input or output in inference using the second model 302 and the third model 303 for training at least one of the first to third models in the second case over the first case. First case: An inference made using the second model 302 and the third model 303 is an inference made when the accuracy of the third model 303 is less than a third threshold. Second case: An inference made using the second model 302 and the third model 303 is an inference made when the accuracy of the third model 303 is equal to or greater than a third threshold.

[0105] When the accuracy of the second model 302 is less than the second threshold, if the accuracy of the second model 302 and the third model 303 is higher than the accuracy of the first model 301, the AI ​​unit 108 may perform inference using the second model 302 and the third model 303, and if the accuracy of the second model 302 and the third model 303 is the same as or lower than the accuracy of the first model 301, the AI ​​unit 108 may perform inference using the first model 301. [Explanation of symbols]

[0106] 10...Material Creation Support System 100...MI Platform System

Claims

1. an AI unit that obtains an inference result by inputting data from a first data source into a second model that has material feature data representing material features as input and material property data as output; a data acquisition unit that stores experimental data obtained by an experiment based on target data, which is at least one of data from the first data source and data based on the inference result, in a second data source; a data conversion unit that acquires experimental data from the second data source and converts the experimental data into data contained in an existing or new first data source; A materials creation support system equipped with:

2. The material creation support system according to claim 1, the data conversion unit calculates feature quantities of one or more predetermined types of data in the experimental data stored in the second data source; Data including the feature amount calculated by the data conversion unit is input to the AI ​​unit. Materials creation support system.

3. The material creation support system according to claim 1, A material creation support system that uses ceramics as the material.

4. The material creation support system according to claim 1, The material creation support system, wherein the material characteristics include at least one selected from the group consisting of porosity, crystal perimeter, composition area ratio, and composition area deviation.

5. The material creation support system according to claim 1, a data presentation unit that presents a manufacturing recipe represented by the manufacturing recipe data as the target data; A material creation support system in which experimental data of an experiment conducted by an experimental system based on the manufacturing recipe presented by the data presentation unit is output, and the output experimental data is transmitted to and stored in a data server.

6. The material creation support system according to claim 2, the predetermined one or more types of data include image data, The data conversion unit analyzes the image data and calculates feature quantities of the image data based on the results of the analysis.

7. The material creation support system according to claim 1, The AI ​​unit selects whether to perform inference using the first model or the second model and the third model from among a first model, which is a model that receives manufacturing recipe data representing a manufacturing recipe as input and outputs material property data representing material properties, a third model, which is a model that receives manufacturing recipe data as input and outputs material property data, and the second model, and performs the selected inference.

8. obtaining an inference result by inputting data from the first data source into a second model having material feature data representing material features as input and material property data as output; storing experimental data obtained by an experiment based on target data, which is at least one of data from the first data source and data based on the inference result, in a second data source; Obtaining experimental data from the second data source and converting the experimental data into data contained in an existing or new first data source. A material creation support method that uses a computer system to perform the above steps.

9. obtaining an inference result by inputting data from the first data source into a second model having material feature data representing material features as input and material property data as output; storing experimental data obtained by an experiment based on target data, which is at least one of data from the first data source and data based on the inference result, in a second data source; Obtaining experimental data from the second data source and converting the experimental data into data contained in an existing or new first data source. A computer program that causes a computer to do something.

10. obtaining an inference result by inputting data from the first data source into a second model having material feature data representing material features as input and material property data as output; storing experimental data obtained by an experiment based on target data, which is at least one of data from the first data source and data based on the inference result, in a second data source; Obtaining experimental data from the second data source and converting the experimental data into data contained in an existing or new first data source. A recording medium that records a computer program that causes a computer to execute the following:

11. obtaining an inference result by inputting data from the first data source into a second model having material feature data representing material features as input and material property data as output; storing experimental data obtained by an experiment based on target data, which is at least one of data from the first data source and data based on the inference result, in a second data source; obtaining experimental data from the second data source and converting the experimental data into data contained in an existing or new first data source; a step of manufacturing materials according to manufacturing conditions represented by the manufacturing recipe data as the target data; A method for manufacturing a material comprising the steps of:

Images