Material Property Prediction System and Material Property Prediction Method

Abstract

The present invention provides a method for effectively utilizing past data to improve the prediction accuracy of material properties. A system for processing case data including a plurality of records composed of materials, experimental conditions, and material properties to predict material properties. The system includes a material property prediction prompt unit, a case mutual feature quantity generation unit, and a material property prediction unit. The material property prediction prompt unit accepts the designation of first case data, which includes records with unknown material properties and becomes the prediction object of the material properties based on the first prediction model. The case mutual feature quantity generation unit uses a second prediction model to predict the feature quantity according to the material composition of the first case data. The material property prediction unit uses the material composition, experimental conditions, feature quantity, and known material properties of the first case data to generate the first prediction model. In addition, the material property prediction unit inputs the material composition, experimental conditions, and feature quantity of the records with unknown material properties in the first case data into the first prediction model to predict the unknown material properties.

Description

Technical Field

[0001] The present invention relates to a technique for assisting experiments in materials science and the like. Background Art

[0002] With the development of statistical processing techniques involved in data analysis, the demand for data analysis in materials science has been increasing. Especially in the field of materials science, in order to efficiently develop new materials, a method called screening is known in which candidates for the next experiment are selected based on known data.

[0003] In Patent Document 1, a method of assisting design is described in which knowledge in a nanoscale region is linked and structured with the same concept regardless of the material type, and this is used to assist in the design of new materials independent of the material type.

[0004] In Patent Document 2, the following is described: Using quantum statistical values obtained by statistically processing quantum thermodynamic state quantities inherent to the elements constituting a reaction system, only in the case of selecting the same physical property values of substances having the same number of elements but different numbers of elements or mixing ratios of the elements constituting the reaction system, a system of simultaneous linear equations of the number of elements of the substance or more is derived and its solution is obtained, whereby it is possible to perform materials design of metallic and non-metallic substances having target physicochemical properties and functions.

[0005] As a screening method, data of various experiments are input into an information system and machine learning is performed to construct a prediction model of experimental results, and screening based on the model prediction is performed. In this prediction, a method is known in which a function is obtained by regression analysis with various parameters related to materials design as independent variables and the properties of the material as the return value.

[0006] Prior Art Documents

[0007] Patent Documents

[0008] Patent Document 1: Japanese Unexamined Patent Application Publication No. 2003-178102

[0009] Patent Document 2: Japanese Unexamined Patent Application Publication No. 2004-086892 Summary of the Invention

[0010] Problems to be Solved by the Invention

[0011] In materials development, by improving the accuracy of predicting material properties, it is possible to more accurately evaluate the predictability of candidates for new materials, and it is expected that efficient materials development can be carried out by omitting unnecessary experiments.

[0012] In regression analysis, a variable corresponding to the independent variable of a function is called an explanatory variable, and a value corresponding to the return value of the function is called an objective variable. However, in the prediction of material properties, the material properties are used as the objective variable, and explanatory variables representing the characteristics of the material are selected in such a way as to be able to predict the objective variable. Depending on how the explanatory variables are selected, the prediction accuracy fluctuates. Therefore, it is particularly important to prepare changes in the explanatory variables in a way that can handle the prediction of various material properties.

[0013] In Patent Document 1 and Patent Document 2, attempts have been made to predict material properties using past data. However, in material development, the general process is to start development with a specific composition and manufacturing process at the beginning. For materials with effective properties obtained, further measures are taken with their associated composition and manufacturing process.

[0014] That is, at the initial stage of development, there is a problem that there is very little data that can be used at the initial stage of the case. When using the information of past data, the material properties as the target are different for each case. Therefore, the data with consistent material properties is almost only the data used for that case. In addition, even for experiments with the same property as the target, there are cases where the measurement methods are different, and in most cases, it is difficult to directly reuse them.

[0015] The problem of the present invention is to provide a method for effectively using past data to improve the prediction accuracy of material properties.

[0016] Means for solving the problem

[0017] A preferred aspect of the present invention is a system for processing case data including a plurality of records composed of material composition, experimental conditions, and material properties to predict material properties. The system includes a material property prediction prompt section, a case mutual feature quantity generation section, and a material property prediction section. The material property prediction prompt section accepts the designation of first case data, which includes records with unknown material properties and becomes the prediction object of the material properties based on the first prediction model. The case mutual feature quantity generation section uses the second prediction model to predict the feature quantity according to the material composition of the first case data. The material property prediction section uses the material composition, experimental conditions, feature quantity, and known material properties of the first case data to generate the first prediction model. In addition, the material property prediction section inputs the material composition, experimental conditions, and feature quantity of the records with unknown material properties in the first case data into the first prediction model to predict the unknown material properties.

[0018] The material composition is at least information related to the composition of the material, and more preferably information related to the structure of the material, such as a structural formula.

[0019] Another preferred aspect of the present invention is a method for predicting material properties by an information processing apparatus including an input device, a storage device, and a processor. In this method, when generating a first prediction model for predicting a first material property based on first data including a first feature quantity, the following steps are performed. That is, execute: a first step of preparing a second prediction model that predicts a second material property differently defined from the first material property based on the first feature quantity; a second step of applying the first data to the second prediction model to predict the second material property; and a third step of generating the first prediction model with the first feature quantity as a first explanatory variable, the second material property as a second explanatory variable, and the first material property as a target variable.

[0020] Advantages of the Invention

[0021] Past data can be effectively utilized to improve the prediction accuracy of material properties. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a functional block diagram showing an example of the schematic configuration of an embodiment.

[0023] Figure 2 It is a block diagram showing an example of the physical implementation structure of an embodiment.

[0024] Figure 3 It is a conceptual diagram showing an example of the usage steps of an embodiment.

[0025] Figure 4 It is a flowchart showing an example of the material DB update process of an embodiment.

[0026] Figure 5 It is a schematic diagram showing an example of the screen display for receiving experimental data of an embodiment.

[0027] Figure 6 It is a table showing an example of the structure of experimental data of an embodiment.

[0028] Figure 7 It is a table showing an example of the experimental data table of the material DB of an embodiment.

[0029] Figure 8 It is a conceptual diagram showing an example of case data.

[0030] Figure 9 It is an explanatory diagram showing the concept of mutual feature quantities between cases.

[0031] Figure 10 It is a flowchart showing an example of the material property prediction process of an embodiment.

[0032] Figure 11 It is a schematic diagram showing an example of the material property prediction display of an embodiment.

[0033] Figure 12 This is a table showing an example of the structure of data for predicting material characteristics of an embodiment. Detailed implementation manners

[0034] The embodiments will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to the descriptions of the embodiments shown below. Those skilled in the art can easily understand that the specific structure can be changed without departing from the spirit or gist of the present invention.

[0035] In the structures of the invention described below, for the same parts or parts having the same functions, the same reference signs are sometimes commonly used among different drawings, and repeated descriptions are omitted.

[0036] In the case where there are multiple identical or parts having the same functions, different subscripts are sometimes used for the same reference signs for explanation. However, in the case where it is not necessary to distinguish multiple elements, the subscripts are sometimes omitted for explanation.

[0037] Expressions such as "first", "second", "third", etc. in this specification are used for identifying structural elements, and do not necessarily limit the quantity, order or their contents. In addition, the numbers used for identifying structural elements are used in each context, and the numbers used in one context do not necessarily represent the same structure in other contexts. In addition, a structural element identified by a certain number does not prevent it from having the functions of a structural element identified by other numbers.

[0038] The positions, sizes, shapes, ranges, etc. of the respective structures shown in the drawings and the like sometimes do not represent the actual positions, sizes, shapes, ranges, etc. for the sake of easy understanding of the invention. Therefore, the present invention is not necessarily limited to the positions, sizes, shapes, ranges, etc. disclosed in the drawings and the like.

[0039] [Embodiment 1]

[0040] <1. System structure>

[0041] Figure 1 This shows an example of the material characteristic prediction device of Embodiment 1. The material characteristic prediction device 101 of this embodiment is a device that accepts the operations of the user 102, and includes an experimental data reception unit 111 that receives experimental data from the user, and a material database (DB: Data Base) 112 that stores the characteristics and properties of materials classified by case. Here, a case refers to a set of data that the user can freely define, for example, data obtained from experiments and developments with different generation subjects and generation purposes.

[0042] In addition, the material property prediction device 101 includes: a material property prediction unit 113 that generates a material property prediction model for predicting material properties, and further uses the material property prediction model to predict unmeasured material properties; and a material property prediction model DB 114 that stores the material property prediction model.

[0043] The material property prediction unit 113 generates a material property prediction model using the feature quantities obtained from the data for which the material property values in the material DB 112 have been measured and the feature quantities obtained from the inter-case mutual feature quantity generation unit 115, and predicts unknown properties. The inter-case mutual feature quantity generation unit 115 generates new feature quantities based on the data in the material DB 112 and the material property prediction model DB 114. The material property prediction prompt unit 116 presents the prediction result of the material property prediction unit 113 to the user 102.

[0044] In this embodiment, the material property prediction device 101 is constituted by an information processing device such as a server including an input device, an output device, a storage device, and a processing device. Functions such as calculation and control are realized in cooperation with other hardware by the processing device executing a program stored in the storage device to perform the determined processing. In Figure 1 functional blocks are shown in place of the hardware structure of the information processing device. As each functional block, a program executed by a computer or the like, its function, or a unit for realizing its function may be referred to as a "function", "means", "unit", "element", "module", etc.

[0045] Figure 2 An example of the physical implementation structure of Embodiment 1 is shown. The material property prediction device 101 can be implemented using a general computer. That is, it is a device including a processor 201 having arithmetic performance, a DRAM (Dynamic Random Access Memory) 202 as a volatile temporary storage area capable of high-speed reading and writing, a storage device 203 as a persistent storage area using an HDD (hard disk device), a flash memory, etc., an input device 204 such as a mouse or keyboard for operations, a monitor 205 for showing actions to the user, and an interface 206 such as a serial port for communicating with the outside.

[0046] Figure 1 The experimental data reception unit 111, the material property prediction unit 113, the inter-case mutual feature quantity generation unit 115, and the material property prediction prompt unit 116 can be realized by the processor 201 executing a program recorded in the storage device 203. The material DB 112 and the material property prediction model DB 114 can be realized by the processor 201 executing a program for accumulating data in the storage device 203.

[0047] Figure 2The structure can be composed of a single computer, or any part can also be composed of other computers connected via a network. That is, the same system can also be composed of multiple computers.

[0048] Figure 3 Schematically shows the utilization steps of the system of Example 1. Example 1 can execute two steps: material data input (S310) where the user inputs data regarding material property prediction, and prediction result viewing (S320) where the result of the material property prediction is confirmed.

[0049] Material data input (S310) is the step of inputting the data set of experimental data 600, which stores data of materials that have undergone experiments and data of materials to be experimented on next, into the material property prediction device 101. Correspondingly, the material property prediction device executes a material DB update process (S311) to update the information stored internally.

[0050] In prediction result viewing (S320), the material property prediction device executes a material property prediction prompt process (S321) according to the requirements of the user 102, and prompts the material property prediction display 322, which is a screen obtained by visualizing the result of the material property prediction.

[0051] <2. Material Data Input Process>

[0052] Figure 4 Shows an example of the processing steps of the material DB update process (S311). In the material DB update process (S311), first, the experimental data receiving unit 111 receives the experimental data 600 from the user 102 and determines or attaches a case ID (S401). Then, the material DB 112 is updated or appended by case (S402).

[0053] Figure 5 Shows an example of the screen displayed on the monitor 205 to receive the experimental data 600 from the user 102 in the initial step (S401) of the material DB update process 311. In Example 1, the user 102 stores the experimental data in a file in advance and transfers the experimental data 600 in the form of specifying the location of the file in the text box 501. In the transferred file, data in tabular form is described in a well-known (Comma Separated Value) form, and the result obtained by interpreting it into tabular form is displayed in the table screen 502.

[0054] In Figure 5Among them, the information described is exemplified as the identifier of the experiment, i.e., "ID", "Temp" representing the temperature during the experiment, "SOL" representing the water solubility at this time, and the string "SMILES" representing the structural formula of the material. In this example, the water solubility is the material property to be predicted, and the data with a blank SOL column represents untested conditions. In addition, this data transfer is an example, and as information that can be converted into a table form, as long as it is a method that can transfer experimental data including the structural formula of the material and the material properties, other methods can also be used. The information is displayed in the table screen 502 and saved in the material DB112 through the button 503.

[0055] Figure 6 An example of the structure of one record representing this experimental data 600 is shown. One record corresponds to a material obtained through a specific composition and preparation process. In this example, the experimental data 600 takes the material property 601, such as information representing the structural formula of the material in the form of SMILES, i.e., the material structural formula 602, and the experimental condition 603 representing the conditions during the experiment, such as temperature and pressure, as information for one record, and the experimental data 600 is data formed by collecting one or more of these records. These information correspond to Figure 5 each item of the table screen 502. In this embodiment, which element each item corresponds to is determined according to the correspondence with the predetermined item name. Regarding this correspondence relationship, the user 102 can also input it from the screen, etc. In addition, regarding the material property 601, the value determined through experiments, etc. is stored, and it is stored as blank in the case of untested. Other case names and other information can also be attached to the experimental data 600.

[0056] In Figure 4 the initial step (S401) of the material DB update process (S311), the above experimental data 600 is interpreted, shaped, and stored as the experimental data table of the material DB112.

[0057] Figure 7 Information representing one record of the experimental data table. This data includes: the experimental ID 701 numbered in a consecutive number or other way in a manner that can uniquely identify the experiment, the material property 702 derived from the material property 601 of the experimental data 600, the material structural formula 703 derived from the material structural formula 602 of the experimental data 600, and the experimental condition 704 derived from the experimental condition 603. They can also convert the information that becomes each source in terms of unit or form and convert it into a unified representation.

[0058] Case ID 700 is an identification number that uniquely identifies a case. In Example 1, since it is assumed that there is one file per case, the case ID corresponds to the file name of the substantive data file. The case ID 700 can be appended in consecutive numbers when registering with the material DB 112. In the case where the correspondence between the file and the case is not determined, it is also possible to prompt the user with the question "Which case does the file to be uploaded now correspond to?" when registering with the material DB 112, and perform registration after inputting the correspondence. The form of the experimental data table needs to be the same as the form of the additional quantity and the completion of registration. The material characteristics 702 and the experimental conditions 704 can be arbitrarily defined by the user, and the quantity can also be freely set.

[0059] <3. Inter-case mutual characteristic quantity>

[0060] One feature of this embodiment is that by using the data of existing cases, the prediction accuracy of material characteristics is improved even in the case of a small number of data. In the process of material development, very little data can be used in the initial stage. Before explaining the specific embodiments, the concept of this embodiment will be explained.

[0061] In Figure 8 an example of the case data stored in the material database 112 divided by case is shown. As Figure 8 shown, in the case of using the information of other cases, the material characteristics that are usually targeted are different for each case. Therefore, the data with consistent material characteristics is almost only the data for that case. In addition, even for experiments with the same characteristic as the goal, there are cases where the measurement methods are different, and it is difficult to directly use them in most cases.

[0062] In Figure 8 the example, the temperatures and humidities of the experimental conditions of the past case A and the past case B are different, and the material characteristics are also different between A and B. Therefore, they cannot be directly used for each other's characteristic prediction. In this embodiment, by using the data of past cases as "information for generating characteristic quantities", the explanatory variables can be increased. Here, the newly generated characteristic quantity is called "inter-case mutual characteristic quantity".

[0063] In Figure 9 the process of using the information related to past cases as "information for generating characteristic quantities" will be explained. First, using the data 901 of the past case A, setting the target variable as the known material characteristic A, and setting the explanatory variable as the structural formula, a prediction model 902 for predicting the material characteristic A based on the structural formula is generated (learned). This can be generated by known supervised machine learning such as regression tree, random forest, support vector regression, Gaussian process regression, neural network, etc.

[0064] Next, apply the structural formula of the data 903 of past case B to the prediction model 902 to predict material property A. Add material property A to the data of past case B to generate a new data set 904. If there is the same structural formula in past case A as in past case B, the material properties of past case A can also be directly added to the new data set. This material property A corresponds to the inter-case mutual feature quantity.

[0065] If the new data set 904 is obtained, use the data with known material property B (item numbers 1, 2, 3) in it as training data to generate a prediction model 905 for predicting material property B. At this time, the explanatory variables are the structural formula, experimental conditions (humidity), and material property A, and the target variable is material property B. The prediction model 905 can be generated by known supervised machine learning.

[0066] Input the data (item number 4) for which material property B is to be predicted into the generated prediction model 905 to obtain material property B. By adding material property A as a new feature quantity (inter-case mutual feature quantity), an improvement in prediction accuracy can be expected compared to the case of directly using the data of past case B. Especially when there is a correlation between material properties A and B, it is considered effective.

[0067] Based on the understanding of the above concepts, the process of the specific prediction result viewing process will be described.

[0068] <4. Prediction Result Viewing Process>

[0069] Use Figure 10 to describe the material property prediction prompt process (S321) during prediction result viewing (S320). In the description, the correspondence with the Figure 9 concept is also referred to using the reference signs of the 900 series of Figure 9 .

[0070] First, the material property prediction prompt unit 116 prompts the user 102 with the material property prediction display 322 to accept the designation of the experimental data table that is the object of the prediction property (S1001). At this time, the content of the experimental data table stored in the material DB 112 is specified using the case ID. Here, it is assumed that the experimental data has already been stored in the material DB 112.

[0071] Figure 11 Examples of the screen displayed on the monitor 205 that accepts instructions from the user 102 and the screen of the material property prediction display 322 obtained by visualizing the results of the material property prediction are shown.

[0072] In the drop-down box 1101 in the figure, the content of the experimental data table is displayed as a candidate. If a case ID is specified and the prediction value update button 1102 is pressed, the material property prediction prompt section 116 sends an instruction to the material property prediction section 113 to perform interpolation of the blank parts of the material properties 702 in the record of the experimental data table ( Figure 7 ) using the prediction value, and displays the result on the screen 1103. In Figure 11 , the underlined numerical values of the material properties indicate that the blank data has been interpolated.

[0073] When the material property prediction section 113 receives the instruction to perform the interpolation of the material property prediction prompt section 116, it obtains the data of the experimental data table specified by the case ID 700 from the material DB 112 (S1002). In addition, in Figure 11 's screen 1104, other cases used to generate the inter-case mutual feature quantities are selected. The material property prediction section 113 obtains the prediction model 902 of the selected other case from the material property prediction model DB 114 (S1003).

[0074] In Figure 10 's process description, the data obtained in process S1002 corresponds to Figure 9 's data 903 of the past case B. In addition, the prediction model of the case obtained in process S1003 corresponds to the prediction model 902 generated based on Figure 9 's data 901 of the past case A.

[0075] In the above description, it is assumed that the prediction model 902 has been generated and is retrieved from the material property prediction model DB 114 through the case ID 700. In the case where there is no prediction model 902 that matches the material property prediction model DB 114, as Figure 9 shows, the material structure formula of the data 901 of the past case A can be used as the explanatory variable, and the known material properties can be used as the target variable for learning and generating the prediction model 902.

[0076] Next, the material property prediction section 113 generates the material property prediction data (S1004). This process is equivalent to applying the structure formula of the data 903 of the past case B to the prediction model 902, predicting the material property A, adding the material property A to the data of the past case B, and generating a new data set 904. At this time, the inter-case mutual feature quantity generation section 115 uses the prediction model 902 obtained in process S1003 to perform the prediction of the material property A (inter-case mutual feature quantity).

[0077] Figure 12 Shows the structure of one record 1500 of the material property prediction data. The content of one record inherits the experimental data table of the data 903 of the past case B (Figure 7 ) case ID 700, experiment ID 701, material properties 702, and experimental conditions 704. It also includes the characteristic quantity 1201 derived from the structural formula. The characteristic quantity derived from the structural formula is calculated based on the material structural formula 703. As a method for calculating the characteristic quantity according to the structural formula, there are well-known methods such as the fingerprint method.

[0078] The data for predicting material properties includes the characteristic quantities 1202 and 1203, that is, the inter-case characteristic quantities, generated by the prediction model 902 of other cases. In Figure 9 the description, one of the other cases is the past case A, and the inter-case characteristic quantity is one of the predicted material properties A. However, the characteristic quantity generated in the prediction model 902 of other cases can be either one or any number of them. In addition, multiple other cases can also be used.

[0079] The material property prediction unit 113 removes the part where the material property 702 is not actually measured, that is, blank, from the data for predicting material properties. It uses the items other than the case ID 700, experiment ID 701, and material property 702 as explanatory variables, and the material property 702 as the target variable, and performs a well-known regression analysis to obtain a prediction function and learn the prediction model 905 (S1005). The generated prediction model 905 is stored in the material property prediction model DB114 together with the case ID of the data for which the prediction model 905 is generated.

[0080] This step means that if the prediction function is written as y = f(x1, x2,...), then y is the target variable, and x1, x2,... are the explanatory variables, and the functional form of f is determined in such a way that if x1, x2,... are determined, then y can be predicted. In the case of this embodiment, when using Figure 12 the data for predicting material properties, it is set as [material property 702] = f([characteristic quantity 1201 derived from the structural formula], [experimental conditions 704], [characteristic quantity 1202 of case [1]], [characteristic quantity 1203 of case [2]]...), and thus a regression analysis is learned to generate the prediction model 905.

[0081] This learning is equivalent to Figure 9 the generation of the lowest-level prediction model 905. In Figure 9 this case, the experimental condition 704 is only one type, which is humidity. However, on the premise that there is data, the number and type of experimental conditions are arbitrary. As experimental conditions, for example, there are the manufacturing conditions of the material, but they can also be omitted in the case of no data. In addition, as described above, in Figure 9 this case, the inter-case characteristic quantity only represents one of the predicted material properties A, but there can also be multiple as described above.

[0082] The algorithm for regression analysis can be a well-known algorithm, such as a regression tree, LASSO, random forest, Gaussian process, support vector regression, neural network, etc. In addition, in this embodiment, the explanatory variables are increased. However, when increasing the explanatory variables, compared with support vector regression, a regression tree or a random forest is preferred. In particular, a high-precision prediction can be expected through a non-linear random forest.

[0083] After generating the prediction model 905 in this way, the material property prediction unit 113 selects the part where the material property 702 is not actually measured, i.e., blank, and calculates the predicted value of the material property 702 using the above prediction function y = f(x1, x2, ···) (S1006).

[0084] Through the material property prediction prompt unit 116, the calculated predicted value is displayed on the screen of the monitor 205 as Figure 11 shown (S1007). In addition, in this embodiment, only the spatial structure feature quantity and experimental conditions are used as explanatory variables, but actually some other quantities (such as molecular weight, charge) can also be derived and used.

[0085] In the above embodiment, when generating the feature quantities of other cases, the structural formula is used. However, as long as it is data common in the case data, data other than the composition can also be used. In addition, a method that can directly use the structural formula for prediction is also well-known, and in this case, the mechanism is the same.

[0086] According to the embodiment described above, a model compatible with the current prediction is generated based on the data obtained from the material property prediction in other past cases, and the explanatory variables are increased via this model, thereby improving the accuracy. For example, at the beginning of research and development, the number of data in the first case ( Figure 9 past case B) is small. However, in this embodiment, for example, the data of a past case ( Figure 9 past case A) that has been developed and has a large amount of data can be used. Thus, when performing material property prediction, the problems of small data volume and low accuracy can be overcome. Therefore, in the prediction evaluation for screening experimental plans, a higher-precision prediction can be made. As a result, it becomes easier to formulate experimental plans, and high-quality materials can be developed with fewer experimental times. For example, the parameters for which the predicted properties become better can be investigated, and the experimental conditions can be preferentially recommended.

[0087] Description of Reference Numerals

[0088] 101 Material property prediction device,

[0089] 102 User,

[0090] 111 Experimental data reception unit,

[0091] 112 Material DB,

[0092] 113 Material Property Prediction Unit,

[0093] 114 Material Property Prediction Model DB,

[0094] 115 Inter-case Mutual Feature Quantity Generation Unit,

[0095] 116 Material Property Prediction Display Unit.

Claims

1. A material property prediction system is a system for processing case data that includes multiple records composed of materials as material structure formulas, representing experimental conditions such as temperature, pressure, or humidity conditions during experiments related to material properties and the material properties, to predict material properties. It is characterized in that the material property prediction system includes a material property prediction prompt section, a case mutual feature quantity generation section, and a material property prediction section. The material property prediction prompt section accepts the designation of first case data, which includes records with unknown material properties and becomes the prediction object of material properties based on a first prediction model. The case mutual feature quantity generation section generates a second prediction model by using the material composition and known second material properties in second case data different from the first case data. The case mutual feature quantity generation section uses the second prediction model to predict the feature quantity of the material properties defined by the second case data according to the material composition of the first case data. The material property prediction section uses the material composition, the experimental conditions, the feature quantity, and the known material properties of the first case data to generate the first prediction model, and inputs the material composition, the experimental conditions, and the feature quantity of the records with unknown material properties in the first case data into the first prediction model to predict the unknown material properties.

2. The material property prediction system according to claim 1, It is characterized in that the material property prediction system can obtain the case data from a material database. The material database stores multiple case data, and between the case data, the experimental conditions and the material properties include differently defined data. The material property prediction prompt section accepts the designation of second case data different from the first case data. The case mutual feature quantity generation section obtains the second case data from the material database to generate the second prediction model.

3. The material property prediction system according to claim 2, It is characterized in that the material property prediction system includes the material database, in which there are stored: the first case data, which includes multiple records composed of material composition, first experimental conditions, and first material properties; the second case data, which includes multiple records storing material composition and second experimental conditions differently defined from the first experimental conditions.

4. The material property prediction system according to claim 2, It is characterized in that the material property prediction system includes the material database, in which there are stored: the first case data, which includes multiple records composed of material composition, first experimental conditions, and first material properties; the second case data, which includes multiple records storing material composition and second material properties differently defined from the first material properties.

5. The material property prediction system according to claim 2, It is characterized in that The material property prediction system has a material property prediction model database that stores at least one of the first prediction model and the second prediction model.

6. The material property prediction system according to claim 5, wherein, the second prediction model is managed corresponding to the second case data.

7. The material property prediction system according to claim 1, wherein, the first prediction model is constituted by a random forest.

8. A method for predicting material properties, which is a method for predicting material properties by an information processing device including an input device, a storage device, and a processor, wherein, the first case data includes a plurality of records, which are composed of a material composition as a material structural formula, an experimental condition representing the temperature, pressure, or humidity condition during an experiment related to the first material property, and the first material property. When generating a first prediction model for predicting the first material property based on the first case data, the following steps are performed: The first step is to prepare a second prediction model by using the material composition and the known second material property in the second case data different from the first case data, and the second prediction model predicts a second material property defined differently from the first material property; The second step is to apply the first case data to the second prediction model to predict the second material property; The third step is to use the material structural formula as the first explanatory variable, the predicted second material property as the second explanatory variable, the experimental condition as the third explanatory variable, and the first material property as the target variable to generate the first prediction model; and The fourth step is to use the first prediction model and the first case data to predict the first material property for records with unknown first material properties.

9. The method for predicting material properties according to claim 8, wherein, a material database differentiated by cases is used, and the first case data related to the first case and the second case data related to the second case are stored in the material database.

10. The method for predicting material properties according to claim 9, wherein, the first case data further includes first information related to the manufacturing conditions of the material.

11. The method for predicting material properties according to claim 10, wherein, the second case data further includes second information, and the definition of the second information is different from the first information related to the manufacturing conditions of the material.

12. The method for predicting material properties according to claim 8, wherein, a random forest is used as the first prediction model.

Images