Method for generating a physical property prediction model, apparatus for generating a physical property prediction model, method for predicting formulations, and apparatus for predicting formulations

Abstract

This technology can generate computational models that predict not only the physical properties of rubber compounding but also its sustainability rate, and further, it can predict the amount of compounding required. [Solution] The method for generating a physical property prediction model comprises: a data mart generation step of generating a data mart that includes the blending amounts of multiple materials used in the production of a rubber compound, the property values ​​of the materials, and the physical properties and sustainability rate of the rubber compound; a physical property prediction step of estimating the physical properties and sustainability rate using a machine learning type computational model with the blending amounts and property values ​​as explanatory variables; and a learning processing step of training the computational model based on the data mart generated in the data mart generation step.

Description

Technical Field

[0001] The present invention relates to a method for generating a physical property prediction model in a rubber formulation, a physical property prediction model generation device, a formulation prediction method, and a formulation prediction device.

Background Art

[0002] For example, a rubber formulation used for tires or the like mounted on a vehicle is a composite material in which a polymer, which is a rubber material as a main raw material, is added with a reinforcing agent and various chemicals. Rubber formulations are formulated based on various materials according to their uses, and those having different physical properties have been developed.

[0003] A conventional physical property prediction system disclosed in Patent Document 1 associates a value of the physical property of a predicted rubber predicted based on material information regarding a plurality of materials constituting the predicted rubber with a feature amount of a specific material among the plurality of materials and displays the same on a display unit. Further, a conventional tire management device disclosed in Patent Document 2 calculates a sustainable rate, which is the ratio of the weight of a sustainable material in the weight of a tire, based on the raw material information and manufacturing information of the tire.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] Incidentally, it is conceivable to construct a machine learning-type computational model that uses the proportions and characteristics of materials as explanatory variables and predicts the physical properties of rubber compound as the objective variable, and then predict the proportions and other factors that will yield the desired physical properties by performing an inverse analysis of this computational model. The inventors of the present invention have realized that by using the sustainability rate in rubber compound such as tires as the objective variable of the computational model, it is possible to predict the proportions and characteristics of materials that will yield the desired sustainability rate.

[0006] The present invention has been made in view of the above circumstances, and its objective is to provide a method for generating a physical property prediction model, a physical property prediction model generation apparatus, a formulation prediction method, and a formulation prediction apparatus that can generate a computational model that predicts the sustainability rate in addition to the physical properties of rubber compound, and further predict the formulation amount, etc. [Means for solving the problem]

[0007] One aspect of the present invention is a method for generating a physical property prediction model. The method for generating a physical property prediction model comprises: a data mart generation step of generating a data mart that includes the blending amounts of multiple materials used in the production of a rubber compound, the property values ​​of the materials, and the physical properties and sustainability rate of the rubber compound; a physical property prediction step of estimating the physical properties and sustainability rate using a machine learning type computational model with the blending amounts and the property values ​​as explanatory variables; and a learning processing step of training the computational model based on the data mart generated in the data mart generation step.

[0008] Another aspect of the present invention is a physical property prediction model generation device. The physical property prediction model generation device includes a data mart generation unit that generates a data mart including the blending amounts of multiple materials used in the production of a rubber compound, the property values ​​of the materials, and the physical properties and sustainability rate of the rubber compound, The system comprises a physical property prediction processing unit that estimates the physical properties and the sustainability rate using a machine learning type computational model with the aforementioned blending amount and the aforementioned property values ​​as explanatory variables, and a learning processing unit that trains the computational model based on the data mart generated by the data mart generation unit.

[0009] Another aspect of the present invention is a formulation prediction method. The formulation prediction method comprises: a target value acquisition step of acquiring target values ​​for the physical properties and sustainability rate of a rubber compound; and a formulation prediction step of using a trained machine learning type computational model that estimates the physical properties and sustainability rate, with the blending amounts and property values ​​of a plurality of materials used in the production of the rubber compound as explanatory variables, to predict the blending amounts and property values ​​from which the target values ​​acquired in the target value acquisition step can be obtained.

[0010] Another aspect of the present invention is a compounding prediction device. The compounding prediction device comprises a target value acquisition unit that acquires target values ​​for the physical properties and sustainability rate of a rubber compound, and a compounding prediction processing unit that uses a trained machine learning type computational model that estimates the physical properties and sustainability rate, with the blending amounts and property values ​​of a plurality of materials used in the production of the rubber compound as explanatory variables, and predicts the blending amounts and property values ​​from which the target values ​​acquired by the target value acquisition unit can be obtained. [Effects of the Invention]

[0011] According to the present invention, it is possible to generate a computational model that predicts not only the physical properties of rubber compound but also the sustainability rate, and furthermore, it is possible to predict the amount of compounding, etc. [Brief explanation of the drawing]

[0012] [Figure 1] This is a block diagram showing the functional configuration of the physical property prediction model generation device according to Embodiment 1. [Figure 2] This chart shows an example of how the sustainability rate of rubber compounds is calculated. [Figure 3]Figure 3(a) is a diagram showing an example of a base calculation table, and Figure 3(b) is a diagram showing an example of a feature calculated by the base calculation table. [Figure 4] Figure 4(a) is an example of a material group table that specifies the correspondence between material groups and material target numbers, and Figure 4(b) is an example of a material category table that specifies the correspondence between material target numbers and material categories. [Figure 5] Figure 5(a) is a diagram showing an example of an analysis group table that specifies the correspondence between numerical analysis groups and analysis target numbers, and Figure 5(b) is a diagram showing an example of an analysis category table that specifies the correspondence between analysis target numbers and numerical analysis categories. [Figure 6] Figure 6(a) is a diagram showing an example of a composite calculation table, and Figure 6(b) is a diagram showing an example of a feature calculated by the composite calculation table. [Figure 7] This is a flowchart showing the steps involved in generating the computational model. [Figure 8] This flowchart shows the procedure for calculating features based on a base calculation table. [Figure 9] This flowchart shows the procedure for calculating features based on a composite calculation table. [Figure 10] This is a block diagram showing the functional configuration of the physical property prediction device according to Embodiment 2. [Figure 11] This is a flowchart showing the procedure for predicting material properties. [Figure 12] This is a block diagram showing the functional configuration of the formulation prediction device according to Embodiment 3. [Figure 13] This is a flowchart showing the procedure for the formulation prediction process. [Modes for carrying out the invention]

[0013] Hereinafter, the present invention will be described based on preferred embodiments with reference to FIGS. 1 to 13. The same or equivalent components and members shown in each drawing are denoted by the same reference numerals, and redundant explanations will be omitted as appropriate. Also, the dimensions of the members in each drawing are shown enlarged or reduced as appropriate for ease of understanding. Further, some of the members that are not important for explaining the embodiments in each drawing are omitted from the display.

[0014] (Embodiment 1) FIG. 1 is a block diagram showing the functional configuration of a physical property prediction model generation device 100 according to Embodiment 1. The physical property prediction model generation device 100 includes a storage unit 10, an operation unit 31, a display unit 32, and an arithmetic processing unit 40, and generates an arithmetic model for predicting the physical properties and sustainability rate of a rubber compound used in, for example, tires and the like.

[0015] The physical property prediction model generation device 100 is an information processing device such as a PC (personal computer). Each part in the physical property prediction model generation device 100 can be realized hardware-wise by an electronic processing circuit or mechanical parts composed of electronic elements including the CPU of a computer, and software-wise by a computer program or the like. Here, functional blocks realized by their cooperation are depicted. Therefore, it is understood by those skilled in the art that these functional blocks can be realized in various forms by combinations of hardware and software.

[0016] The storage unit 10 is a storage device composed of, for example, an SSD (Solid State Drive), a hard disk, a CD-ROM, a DVD, or the like. In the storage unit 10, a data mart 11 used for predicting the physical properties and sustainability rate of a rubber compound, a table 21 used for arithmetic operations, a computer program executed by the arithmetic processing unit 40, and data used for the execution of the computer program are stored.

[0017] The data mart 11 includes formulation data 12, raw material data 13, processing condition data 14, physical property data 15, and sustainability rate data 16, each corresponding to a plurality of rubber compound. In the data mart 11, for example, identification information such as a sequential number is assigned to each of the plurality of rubber compound, and the formulation, raw material properties, processing conditions, physical properties, and sustainability rate corresponding to each identification information are included in the formulation data 12, raw material data 13, processing condition data 14, physical property data 15, and sustainability rate data 16.

[0018] Compounding data 12 is a set of data showing the proportions of each material in multiple rubber compoundings. Compounding data 12 includes the names and proportions of the main rubber materials, reinforcing agents, and various chemicals corresponding to each of the multiple rubber compoundings. Examples of rubber materials include natural rubber (NR), butadiene rubber (BR), styrene-butadiene rubber (SBR), and isobutylene-isoprene-styrene rubber (IIR). Examples of reinforcing agents (fillers) include carbon black and silica. Examples of various chemicals include sulfur, wax, and silanes. The proportions are expressed as volume, weight, and their ratios.

[0019] The raw material data 13 is a set of data such as property values ​​and categorical variables for each material used in each of the multiple rubber compound. Property values ​​include, for example, DBP (Dibutyl Phthalate) absorption and carbon surface area, while categorical variables include, for example, the ASTM grade of carbon. The processing condition data 14 is a set of data that includes conditions such as time, temperature, name of the machine used, and name of the processing method during each processing step such as kneading, molding, and vulcanization during the manufacturing of each of the multiple rubber compound. The data may be in the form of numerical variables or categorical variables.

[0020] The physical property data 15 consists of data for each of the multiple rubber compound, such as hardness Hs, tensile stress S, loss tangent tanδ, abrasion resistance, and rebound elasticity. The sustainability ratio data 16 is the ratio of the weight of sustainable material to the weight of the rubber compound.

[0021] Figure 2 is a chart showing an example of calculating the sustainability rate of a rubber compound. The sustainability rate is the ratio of the weight of sustainable materials to the weight of the rubber compound. Sustainable materials include, but are not limited to, materials that are environmentally friendly and not derived from fossil resources, renewable materials, recycled materials, and biodegradable materials. Renewable materials include renewable resources, such as bio-derived materials. Recycled materials are materials obtained through recycling, and biodegradable materials are materials obtained from substances decomposed by microorganisms.

[0022] In Figure 2, for example, the sustainability rate of polymer raw material b is 75%, and the blended PHR (PHR is parts by weight) is 25, so the sustainable portion of polymer raw material b in the rubber compound is 18.75. The total blended PHR of the rubber compound is 196.8, of which the total sustainable portion is 168.35. The sustainability rate of the rubber compound is calculated as 85.5%, which is obtained by dividing the total sustainable portion by the total blended PHR and multiplying by 100.

[0023] Thus, the sustainability rate of the rubber compound is calculated based on the amounts of multiple materials blended into the rubber compound and the sustainability rate of each material, and is stored in the storage unit 10 as sustainability rate data 16.

[0024] The data mart 11 may acquire compounding data 12, raw material data 13, processing condition data 14, physical property data 15, and sustainability rate data 16 for multiple rubber compound products from an external device 90. The external device 90 is, for example, an information processing device such as a PC, and has a storage device that stores compounding data 12, raw material data 13, processing condition data 14, physical property data 15, and sustainability rate data 16 for multiple rubber compound products.

[0025] Table 21 includes the base calculation table 22, material group table 23, material category table 24, analysis group table 25, analysis category table 26, and composite calculation table 27 used by the calculation processing unit 40. These tables will be described later.

[0026] The operation unit 31 has operable input devices such as a touch panel, switches, keyboard, and mouse device, and accepts user input. The operation unit 31 accepts user input in the generation of rubber compound property prediction models, property prediction, and compound prediction.

[0027] The display unit 32 has a display device such as a liquid crystal display and displays a screen for receiving various data and user inputs during the generation process of the rubber compound property prediction model. The display unit 32 also displays a screen for receiving various data and user inputs during the rubber compound property prediction and formulation prediction.

[0028] The arithmetic processing unit 40 includes a data mart generation unit 41, a base arithmetic unit 42, a composite arithmetic unit 43, a learning data generation unit 44, a physical property prediction processing unit 45, and a learning processing unit 46. The arithmetic processing unit 40 is an electronic circuit that performs arithmetic processing, such as a CPU, and functions by reading and executing computer programs and data stored in the storage unit 10. The data mart generation unit 41, base arithmetic unit 42, composite arithmetic unit 43, learning data generation unit 44, physical property prediction processing unit 45, and learning processing unit 46 in the arithmetic processing unit 40 may be constructed as a plurality of program modules formed by computer programs.

[0029] The data mart generation unit 41 acquires compounding data 12, raw material data 13, processing condition data 14, physical property data 15, and sustainability rate data 16 for multiple rubber compounding products from an external device 90 to generate a data mart 11. The data mart generation unit 41 may also set all or part of a database of compounding data for multiple rubber compounding products, property values ​​and categorical variables of raw materials, processing conditions, and physical properties and sustainability rates of rubber compounding products, which are stored in the storage unit 10 in advance, as the data mart 11 to be used for generating a physical property prediction model. The data mart generation unit 41 may also generate the data mart 11 by receiving user input via the operation unit 31 to specify the rubber compounding products to be included in the data mart 11.

[0030] The base calculation unit 42 generates a base calculation table 22 that describes the materials used in the compounding of the rubber compound and the calculation processing performed on those materials. The base calculation unit 42 calculates feature quantities based on the calculation processing performed using the generated base calculation table 22.

[0031] Figure 3(a) is a diagram showing an example of the base calculation table 22, and Figure 3(b) is a diagram showing an example of the feature quantities calculated by the base calculation table 22. As shown in Figure 3(a), the base calculation table 22 consists of an action number, a material group, a numerical analysis group, and an action description.

[0032] In the base calculation table 22 generated by the base calculation unit 42, for example, for action number A1, material group G1 is specified as numerical analysis group SG1, and the action content is specified as calculating a weighted average value W_avg based on the blending amount. For action number A2, etc., the action content is specified as calculating a sum value SUM by addition. For action number A4, the action content is specified as performing clustering by principal component analysis (PCA) and extracting principal components. The action content in the base calculation table 22 includes arithmetic operations, weighted averages, principal component analysis, and special numerical analysis calculations such as t-distribution type stochastic nearest neighbor embedding (t-SNE).

[0033] The base calculation unit 42 calculates the feature quantities shown in Figure 3(b) by performing calculation processing according to the base calculation table 22 shown in Figure 3(a). Figure 4(a) is a diagram showing an example of a material group table 23 that specifies the correspondence between material groups and material target numbers, and Figure 4(b) is a diagram showing an example of a material category table 24 that specifies the correspondence between material target numbers and material categories. Figure 5(a) is a diagram showing an example of an analysis group table 25 that specifies the correspondence between numerical analysis groups and analysis target numbers, and Figure 5(b) is a diagram showing an example of an analysis category table 26 that specifies the correspondence between analysis target numbers and numerical analysis categories.

[0034] As shown in Figure 4(a), material group G1 is associated with material target number M1. Material group G4 is associated with material target numbers M4 and M5. As shown in Figure 4(b), material classifications and material categories are associated with material target numbers. For example, material target number M1 is associated with carbon (material category), which is a reinforcing material (material classification). Material target number M2 is associated with carbon and silica, which are reinforcing materials. Material target number M3 is associated with oil, material target number M4 is associated with SBR, a type of rubber, and material target number M5 is associated with BR, a type of rubber. In addition to the material classifications and material categories shown in Figure 4(b), further subdivided classifications and categories may be created. Alternatively, material classifications and material categories may be identified by symbols, and a master table may be created for the materials corresponding to each symbol.

[0035] As shown in Figure 5(a), analyte number B1 is associated with numerical analysis group SG1, and analyte number B2 is associated with numerical analysis group SG2. As shown in Figure 5(b), the DBP absorption amount, which indicates the degree of carbon particle linkage and aggregation, is associated with analyte number B1 as a numerical analysis category. The numerical analysis categories associated with analyte number B2 are the amount of cis-1,4 bonds in the butadiene unit portion of SBR or BR (denoted as Cis%), the amount of trans-1,4 bonds in the butadiene unit portion of SBR or BR (denoted as Trans%), the styrene content in SBR (denoted as St%), and the amount of 1,2-vinyl bonds in the butadiene unit portion of SBR or BR (denoted as Vinyl%).

[0036] The calculation process in the base calculation unit 42 will be explained using action number A1 as an example. Returning to Figure 3(a), in action number A1, material group G1 is specified as material target number M1 (see Figure 4(a)), which corresponds to the reinforcing carbon (see Figure 4(b)). In action number A1, numerical analysis group SG1 is specified as analysis target number B1 (see Figure 5(a)), which corresponds to DBP absorption amount (see Figure 5(b)). The action content in action number A1 is to calculate a weighted average value W_avg based on the blending amount. As shown in Figure 3(b), the base calculation unit 42 calculates a weighted average value W_avg based on the blending amount for the DBP absorption amount of various carbons used in the rubber compound as a feature quantity through the calculation of action number A1.

[0037] The blending amounts of each type of carbon are included in the blending data 12 of the data mart 11. The DBP absorption amounts of each type of carbon are included in the raw material data 13 of the data mart 11. The base calculation unit 42 reads the necessary data from the data mart 11 and performs calculation processing in the base calculation table.

[0038] The base calculation unit 42 generates a base calculation table 22 describing the calculation process based on user input specifying the materials used in the rubber compound formulation and the calculation process applied to those materials, and stores it in the storage unit 10. The base calculation unit 42 also generates a material group table 23, a material category table 24, an analysis group table 25, and an analysis category table 26 based on user input specifying material groups, etc., and stores them in the storage unit 10. The base calculation table 22, material group table 23, material category table 24, analysis group table 25, and analysis category table 26 may be generated based on user input from the operation unit 31, or they may be generated using pre-prepared tables. The administrator of the physical property prediction model generation device 100 may also prepare an electronic file containing the base calculation table 22 and other tables in advance and store it in the storage unit 10. Note that the operator performing the processing related to the physical property prediction model, as well as the administrator of the physical property prediction model generation device 100, may be considered a user.

[0039] When a user adds a new feature, the base calculation unit 42 sequentially adds the new action number to the base calculation table 22 and updates the base calculation table 22 based on user input specifying the material group, numerical analysis group, and action content. The base calculation unit 42 updates the material group table 23, material category table 24, analysis group table 25, and analysis category table 26 based on user input specifying information related to the new action number.

[0040] The material group table 23 and the material category table 24 function as auxiliary tables that supplement the material information in the base calculation table 22. Similarly, the analysis group table 25 and the analysis category table 26 function as auxiliary tables that supplement the information related to the numerical analysis.

[0041] The composite calculation unit 43 selects at least two feature quantities calculated by the base calculation unit 42 and generates a composite calculation table 27 that describes the calculation process. The composite calculation unit 43 calculates feature quantities based on the calculation process using the generated composite calculation table 27. Based on user input specifying the calculation process by selecting at least two feature quantities calculated by the base calculation unit 42, the composite calculation unit 43 generates a composite calculation table 27 that describes the calculation process and stores it in the storage unit 10. The composite calculation table 27 may be generated based on user input from the operation unit 31, or it may be generated using a pre-prepared table. Alternatively, the administrator of the physical property prediction model generation device 100 may prepare an electronic file containing the composite calculation table 27 in advance and store it in the storage unit 10.

[0042] Figure 6(a) is a diagram showing an example of a composite calculation table 27, and Figure 6(b) is a diagram showing an example of a feature calculated by the composite calculation table 27. As shown in Figure 6(a), the composite calculation table 27 consists of an action number, an action number in the base calculation table 22, and an action description.

[0043] In the composite calculation table 27 generated by the composite calculation unit 43, for example, action number AM1 is specified as action numbers A2 and A3 in the base calculation table 22, and the action content is a subtraction operation. The action content in the composite calculation table 27 is an arithmetic operation on at least two of the feature quantities calculated by the base calculation table 22.

[0044] The composite calculation unit 43 calculates the feature quantities shown in Figure 6(b) by performing calculations using the composite calculation table 27 shown in Figure 6(a).

[0045] As shown in Figure 6(b), action number AM1 yields a feature value obtained by subtracting the total amount of oil from the total amount of filler. This value tends to correlate highly with rubber hardness. Action number AM2 yields a feature value obtained by dividing the total amount of oil by the total amount of filler. This value also tends to correlate highly with rubber hardness. Generally, rubber hardness is thought to be determined by the balance between oil and filler.

[0046] Action number AM3 yields a feature value that is the sum of the total filler amount and the total oil amount. This sum tends to correlate highly with fuel efficiency. Generally, fuel efficiency is determined by the energy loss components in the rubber compound, and oil and filler are considered to be energy loss components.

[0047] Action number AM4 yields a feature value obtained by multiplying the total filler amount by the amount of DBP absorbed by carbon. This product of the total filler amount and the amount of DBP absorbed by carbon tends to correlate highly with wear resistance. Generally, wear resistance depends on the reinforcing properties of the filler, and it is believed that the reinforcing properties of the filler can be represented by multiplying the total filler amount and the amount of DBP absorbed by carbon.

[0048] The training data generation unit 44 generates a training data mart consisting of at least one of the features calculated by the base calculation unit 42 and the features calculated by the composite calculation unit 43, and the data contained in the data mart 11. The training data mart may include at least one of the features calculated by the base calculation unit 42 and not include any features calculated by the composite calculation unit 43. The training data mart may not include any features calculated by the base calculation unit 42 and may include at least one of the features calculated by the composite calculation unit 43. The training data mart may include at least one of the features calculated by the base calculation unit 42 and at least one of the features calculated by the composite calculation unit 43.

[0049] The learning data generation unit 44 generates a learning data mart that includes some or all of the multiple rubber compounds included in the data mart 11. The rubber compounds included in the learning data mart may be pre-set, or they may be selected by user input received by the operation unit 31. The learning data generation unit 44 also generates a learning data mart that includes some or all of the compounding data 12 and physical property data 15. The learning data mart may also include some or all of the raw material data 13. The learning data mart may also include some or all of the processing condition data 14.

[0050] The learning data generation unit 44 may include pre-set data from among the formulation data 12, raw material data 13, processing condition data 14, physical property data 15, and sustainability rate data 16 in the learning data mart. Alternatively, the learning data generation unit 44 may receive user input via the operation unit 31 and select the data from among the formulation data 12, raw material data 13, processing condition data 14, physical property data 15, and sustainability rate data 16 to include in the learning data mart.

[0051] The physical property prediction processing unit 45 inputs data other than the physical property data 15 and sustainability rate data 16 from the learning data mart generated by the learning data generation unit 44 as explanatory variables into a learning-type computational model 45a to predict the physical properties and sustainability rate of the rubber compound. In the physical property prediction processing unit 45, the explanatory variables that become input data to the computational model 45a are the compounding data 12, the feature quantities calculated by the base computation unit 42 and the composite computation unit 43, the raw material data 13, and the processing condition data 14. The computational model 45a is a random forest type learning model, which is an ensemble learning algorithm using decision trees.

[0052] The learning processing unit 46 trains the computational model 45a based on the learning data mart generated by the learning data generation unit 44. The learning processing unit 46 compares the estimated physical properties and sustainability rate of the rubber compound, which are output data from the computational model 45a, with known physical property data 15 and sustainability rate data 16, which are training data, to train the computational model 45a. The computational model 45a may be, for example, a generalized linear regression model, a DNN (Deep Neural Network) model, a gradient boosting decision tree model, a random forest model, a nonlinear regression model, etc. Known validation methods such as random data sampling and cross-validation can be used to validate the computational model 45a.

[0053] Next, the operation of the physical property prediction model generation device 100 will be described. Figure 7 is a flowchart showing the procedure for generating the calculation model 45a. The data mart generation unit 41 of the calculation processing unit 40 generates a data mart 11 to be used for training the calculation model 45a (S1). The data mart 11 includes at least formulation data 12, raw material data 13, physical property data 15, and sustainability rate data 16. In addition to the formulation data 12, raw material data 13, physical property data 15, and sustainability rate data 16, the data mart 11 may also include processing condition data 14.

[0054] The base calculation unit 42 generates a base calculation table 22 that describes the materials of the rubber compound and the calculation processing for those materials (S2). In step S2, along with generating the base calculation table 22, the base calculation unit 42 generates a material group table 23, a material category table 24, an analysis group table 25, and an analysis category table 26.

[0055] The base calculation unit 42 calculates feature quantities based on the calculations performed by the base calculation table 22 (S3). In step S3, the base calculation unit 42 refers to the material group table 23, the material category table 24, the analysis group table 25, and the analysis category table 26, and performs the calculations performed by the base calculation table 22.

[0056] The composite calculation unit 43 selects at least two of the feature quantities calculated by the base calculation unit 42 and generates a composite calculation table 27 that describes the calculation process (S4). The composite calculation unit 43 calculates feature quantities based on the calculation process performed by the base calculation table 22 (S5).

[0057] The training data generation unit 44 generates a training data mart consisting of at least one of the feature quantities calculated by the base calculation unit 42 and the composite calculation unit 43, as well as data included in the data mart 11 (S6). The training data mart includes formulation data 12, raw material data 13, physical property data 15, and sustainability rate data 16. The training data mart may also include processing condition data 14.

[0058] The learning processing unit 46 trains a learning-type computational model 45a that predicts the physical properties of rubber compounds based on the training data mart (S7), and then terminates the process. The learning processing unit 46 compares the estimated values ​​of the physical properties and sustainability rate of the rubber compounds calculated by the computational model 45a with known physical property data 15 and sustainability rate data 16 as training data, and trains the computational model 45a.

[0059] Figure 8 is a flowchart showing the procedure for calculating features based on the base calculation table 22. The base calculation unit 42 selects one unexecuted action number in the base calculation table 22 (S11). Based on the material group of the selected action number, the base calculation unit 42 refers to the material group table 23 and the material category table 24 and extracts data from the formulation data 12 (S12).

[0060] The base calculation unit 42 determines whether or not there is a numerical analysis group corresponding to the action number in the base calculation table 22 (S13). If it is determined in step S13 that there is no numerical analysis group (S13: NO), the base calculation unit 42 executes the action content in the base calculation table 22 and calculates the feature quantities (S14).

[0061] If it is determined in step S13 that there is a numerical analysis group (S13: YES), the base calculation unit 42 refers to the analysis group table 25 and the analysis category table 26 and extracts data such as property values ​​from the raw material data 13 (S15). After step S15, the base calculation unit 42 proceeds to step S14 and executes the action contents in the base calculation table 22 to calculate feature quantities.

[0062] After step S14, the base calculation unit 42 determines whether or not it has performed all action numbers (S16). If it determines in step S16 that it has not performed all action numbers (S16: NO), the base calculation unit 42 returns to step S11 and repeats the process. If it determines in step S16 that it has performed all action numbers (S16: YES), the base calculation unit 42 terminates the process.

[0063] Figure 9 is a flowchart showing the procedure for calculating features based on the composite calculation table 27. The composite calculation unit 43 selects one unexecuted action number in the composite calculation table 27 (S21). The composite calculation unit 43 extracts the feature of the action number in the base calculation table 22 that corresponds to the selected action number (S22). The composite calculation unit 43 executes the action content in the composite calculation table 27 on the feature extracted in step S22 and calculates the feature (S23).

[0064] After step S23, the composite calculation unit 43 determines whether or not it has performed the actions for all action numbers (S24). If it determines in step S24 that it has not performed the actions for all action numbers (S24: NO), the composite calculation unit 43 returns to step S21 and repeats the process. If it determines in step S24 that it has performed the actions for all action numbers (S24: YES), the composite calculation unit 43 terminates the process.

[0065] The physical property prediction model generation method in this embodiment generates a data mart 11 in a data mart generation step, which includes the blending amounts of multiple materials used in the production of a rubber compound, the property values ​​of the materials, and the physical properties and sustainability rate of the rubber compound. The physical property prediction step in the physical property prediction model generation method estimates the physical properties and sustainability rate of the rubber compound using a machine learning type computational model, with the blending amounts and property values ​​of the materials as explanatory variables. The learning processing step in the physical property prediction model generation method trains the computational model 45a based on the data mart 11 generated in the data mart generation step. According to this physical property prediction model generation method, it is possible to generate a computational model 45a that predicts the sustainability rate in addition to the physical properties of the rubber compound.

[0066] The material property prediction model generation method further calculates features based on a base calculation table 22 that describes the material and the calculation process applied to the material, using a base calculation step. The material property prediction model generation method then trains the calculation model 45a by including the features calculated in the base calculation step as explanatory variables. As a result, the material property prediction model generation method can improve user convenience by standardizing the calculation process for generating features using the base calculation table 22.

[0067] In the material property prediction model generation method, the base calculation table 22 includes at least one of the four basic arithmetic operations. This allows the material property prediction model generation method to generate feature quantities for rubber compounds by specifying an arithmetic operation in the base calculation table 22.

[0068] The material property prediction model generation method further calculates features based on the computation process in the composite computation table 27, which describes the computation process for selecting at least two of the features calculated in the base computation step, through a composite computation step. The material property prediction model generation method includes the features calculated in the composite computation step as explanatory variables of the computation model 45a and trains the computation model 45a. As a result, the material property prediction model generation method can standardize the computation process for the compositely calculated features and further improve user convenience.

[0069] In the material property prediction model generation method, the composite calculation table 27 includes at least one of the four basic arithmetic operations. This allows the material property prediction model generation method to generate complex features in rubber compounds by specifying an arithmetic operation in the composite calculation table 27.

[0070] In the method for generating a physical property prediction model, the data mart 11 includes data on processing conditions for the rubber compound. This allows the physical property prediction model generation method to train the computational model 45a while reflecting the influence of the processing condition data 14.

[0071] In the material property prediction model generation method, the base calculation table 22 performs calculations within a material group containing multiple materials. This allows the material property prediction model generation method to generate features that take into account the overall blending amount and property values ​​of carbon materials, for example, when multiple carbon materials are blended.

[0072] The actions specified in the base calculation table 22 are not limited to arithmetic operations, but may also include numerical analyses such as DBP absorption, Cis%, Trans%, St%, and Vinyl%, as described above.

[0073] Furthermore, in the method for generating material property prediction models, the computational model 45a can use, for example, a random forest model, a gradient boosting decision tree model, and a generalized linear regression model. This allows the material property prediction model generation method to construct computational models that utilize the characteristics of each model.

[0074] The physical property prediction model generation device 100 comprises a data mart generation unit 41, a physical property prediction processing unit 45, and a learning processing unit 46. The data mart generation unit 41 generates a data mart 11 that includes the blending amounts of multiple materials used in the production of rubber compound, the property values ​​of the materials, and the physical properties and sustainability rate of the rubber compound. The physical property prediction processing unit 45 uses the blending amounts and property values ​​of the materials as explanatory variables and estimates the physical properties and sustainability rate of the rubber compound using a machine learning type computational model 45a. The learning processing unit 46 trains the computational model 45a based on the data mart 11 generated by the data mart generation unit 41. As a result, the physical property prediction model generation device 100 can generate a computational model 45a that predicts the sustainability rate in addition to the physical properties of the rubber compound.

[0075] (Embodiment 2) Figure 10 is a block diagram showing the functional configuration of the physical property prediction device 110 according to Embodiment 2. The physical property prediction device 110 has the same configuration as the physical property prediction model generation device 100 in Embodiment 1, but with the learning data generation unit 44 and the learning processing unit 46 removed. Furthermore, the calculation model 45a in the physical property prediction processing unit 45 of the physical property prediction device 110 uses the calculation model that has been learned by the physical property prediction model generation device 100.

[0076] The data mart generation unit 41 of the physical property prediction device 110 generates a data mart 11 consisting of formulation data 12, raw material data 13, and processing condition data 14 for a new rubber compound whose physical properties are to be predicted. Since the data mart 11 generated here is not used to train the calculation model 45a, it is sufficient to have the formulation data 12, raw material data 13, and processing condition data 14, and does not need to include physical property data 15 and sustainability rate data 16.

[0077] The base calculation unit 42 of the material property prediction device 110 generates a base calculation table 22, similar to Embodiment 1, and calculates feature quantities based on the calculations performed by the base calculation table 22. During the calculations performed by the base calculation table 22, the material group table 23, material category table 24, analysis group table 25, and analysis category table 26 are referenced. The composite calculation unit 43 generates a composite calculation table 27, similar to Embodiment 1, and calculates feature quantities based on the calculations performed by the composite calculation table 27.

[0078] The physical property prediction processing unit 45 uses at least one feature calculated for the new rubber compound by the base calculation unit 42 and the composite calculation unit 43, as well as the compounding data 12, raw material data 13, and processing condition data 14 of the new rubber compound, as explanatory variables. The physical property prediction processing unit 45 inputs the data used as explanatory variables into the trained calculation model 45a and predicts the physical properties and sustainability rate.

[0079] The explanatory variables used by the physical property prediction processing unit 45 as input to the learned calculation model 45a during physical property prediction are the same as the explanatory variables used during the learning of the calculation model 45a in the above-described embodiment 1. The physical property prediction processing unit 45 may store the physical properties and sustainability rate predicted for the new rubber compound by the calculation model 45a in the storage unit 10 as physical property data 15 and sustainability rate data 16.

[0080] Next, the operation of the physical property prediction device 110 will be described. Figure 11 is a flowchart showing the procedure for the physical property prediction process. The data mart generation unit 41 of the calculation processing unit 40 generates a data mart 11 for the new rubber compound (S31). The processes from step S32 to step S35 are equivalent to the processes from step S2 to step S5 in Figure 7, and are omitted for brevity. In addition, the calculation processes in steps S33 and S35 are equivalent to the calculation processes explained in Figures 8 and 9.

[0081] The physical property prediction processing unit 45 predicts the physical properties and sustainability rate of the new rubber compound using the computational model 45a with the feature quantities calculated in steps S33 and S35 (S36), and then terminates the process.

[0082] In this embodiment, the physical property prediction method generates a data mart 11 in a data mart generation step, which includes the blending amounts and material property values ​​of multiple materials used in the production of a rubber compound. The physical property prediction method generates a base calculation table 22 in a base calculation generation step, which describes the materials and the calculation processes applied to them. The physical property prediction method calculates feature quantities based on the calculation processes in the base calculation table 22 in a base calculation step.

[0083] The physical property prediction method predicts the physical properties and sustainability rate of a rubber compound by inputting at least one of the features calculated in the base calculation step and the data contained in the data mart 11 into a trained calculation model 45a in the physical property prediction step. This physical property prediction method can predict not only the physical properties but also the sustainability rate of a rubber compound, and improves user convenience by standardizing the calculation process for generating features.

[0084] The physical property prediction method further generates a composite calculation table 27 in a composite calculation generation step, selecting at least two of the features calculated in the base calculation step and describing the calculation process. The physical property prediction method calculates features based on the calculation process in the composite calculation table 27 in a composite calculation step. The physical property prediction method predicts the physical properties and sustainability rate of the rubber compound by inputting at least one of the features calculated in the composite calculation step and the data contained in the data mart 11 into a trained calculation model 45a in a physical property prediction step. This allows the physical property prediction method to standardize the calculation process of the compositely calculated features and further improve user convenience.

[0085] (Embodiment 3) Figure 12 is a block diagram showing the functional configuration of the formulation prediction device 120 according to Embodiment 3. The formulation prediction device 120 includes a target value acquisition unit 47 and a formulation prediction processing unit 48. The formulation prediction processing unit 48 has a calculation model 45a that has been learned by the physical property prediction model generation device 100.

[0086] The data mart generation unit 41 in the formulation prediction device 120 generates a data mart 11 based on constraints related to the rubber compound for which the formulation is to be predicted. The data mart 11 includes constraints related to formulation data 12, raw material data 13, and processing condition data 14 related to the rubber compound.

[0087] The base calculation unit 42, similar to Embodiment 1, generates a base calculation table 22 and has the function of calculating feature quantities based on the calculations performed by the base calculation table 22. During the calculations performed by the base calculation table 22, the material group table 23, material category table 24, analysis group table 25, and analysis category table 26 are referenced. The composite calculation unit 43, similar to Embodiment 1, generates a composite calculation table 27 and has the function of calculating feature quantities based on the calculations performed by the composite calculation table 27.

[0088] The target value acquisition unit 47 acquires target values ​​for the physical properties and sustainability rate of the rubber compound. The target value acquisition unit 47 may read target values ​​for physical properties and sustainability rate that have been previously stored in the physical property data 15 and sustainability rate data 16 of the storage unit 10, or it may acquire target values ​​for physical properties and sustainability rate input from the operation unit 31.

[0089] The formulation prediction processing unit 48 uses the computation model 45a to predict the formulation of a rubber compound as an inverse problem, based on the target physical properties and sustainability rate, as well as constraints regarding the amount of materials used in the rubber compound, raw material data, and processing condition data, for example, by using a genetic algorithm. Alternatively, the formulation prediction processing unit 48 may also predict the formulation of a rubber compound as an inverse problem using known optimization methods, such as gradient descent. The formulation prediction processing unit 48 may also predict a rubber compound that satisfies the physical properties, sustainability rate, and formulation cost by adding a target formulation cost in addition to the target physical properties and sustainability rate. In this case, the computation model 45a outputs the formulation cost, and the formulation cost as physical property data is pre-learned as training data. For example, by including the cost of each material in the raw material data 13, and calculating the formulation cost by taking a weighted average using the amount of material used and the cost of the material, and registering it in the physical property data 15, the formulation cost can be treated in the same way as physical properties.

[0090] The formulation prediction processing unit 48 may add the formulation data of the rubber compound assumed in the prediction process, raw material data, and processing condition data to the data mart 11, calculate feature quantities using the base calculation unit 42 and the composite calculation unit 43, and reflect them in the formulation prediction.

[0091] Figure 13 is a flowchart showing the procedure for the formulation prediction process. The target value acquisition unit 47 acquires the physical properties and sustainability rate of the target rubber compound (S41). The formulation prediction processing unit 48 acquires the constraints of the rubber compound (S42). Using the trained calculation model 45a, the formulation prediction processing unit 48 predicts the formulation of the rubber compound as an inverse problem using an optimization method based on the target physical properties and sustainability rate, as well as the constraints regarding the amount of material used and the material properties in the rubber compound (S43), and then terminates the process. Processing conditions may also be included in the constraints. Furthermore, the formulation prediction processing unit 48 may be configured to predict at least one of the following for a rubber compound having the target physical properties: the formulation (material ratio), the properties and categorical variables of the raw materials, and the processing conditions.

[0092] The formulation prediction method in this embodiment obtains target values ​​for the physical properties and sustainability rate of the rubber compound in the target value acquisition step. The formulation prediction step in the formulation prediction method uses the blending amounts and property values ​​of multiple materials used in the production of the rubber compound as explanatory variables, and uses a trained machine learning computation model 45a that estimates the physical properties and sustainability rate to predict the blending amounts and property values ​​of the rubber compound that will obtain the target physical properties and sustainability rate. According to this formulation prediction method, it is possible to predict the blending amounts and property values ​​of the rubber compound that will obtain the target physical properties and sustainability rate.

[0093] Furthermore, the formulation prediction step predicts the amount and properties of the rubber compound under constraints on the amount and properties of the materials in the rubber compound. This prevents the formulation prediction method from obtaining prediction values ​​that are inappropriate for, for example, the amount and properties of the rubber compound.

[0094] The formulation prediction method further includes a base calculation step, which calculates features based on a base calculation table 22 that describes materials and calculation processes applied to those materials. The formulation prediction step of the formulation prediction method predicts the blending amount and property values ​​of the rubber compound using a learned calculation model 45a in which at least one of the features calculated in the base calculation step is added as an explanatory variable. As a result, the formulation prediction method improves user convenience by standardizing the calculation process for generating features.

[0095] The formulation prediction method further includes a composite calculation step, which calculates features based on a composite calculation table 27 that describes the calculation process using at least two of the features calculated in the base calculation step. The formulation prediction step of the formulation prediction method predicts the blending amount and property values ​​of the rubber compound using a learned calculation model 45a in which at least one of the features calculated in the composite calculation step is added as an explanatory variable. As a result, the formulation prediction method can further improve user convenience by standardizing the calculation process of the compositely calculated features.

[0096] The compounding prediction device 120 includes a target value acquisition unit 47 and a compounding prediction processing unit 48. The target value acquisition unit 47 acquires target values ​​for the physical properties and sustainability rate of the rubber compound. The compounding prediction processing unit 48 uses the blending amounts and property values ​​of multiple materials used in the production of the rubber compound as explanatory variables and uses a trained machine learning type computational model 45a that estimates the physical properties and sustainability rate to predict the blending amounts and property values ​​of the rubber compound that will yield the target values ​​acquired by the target value acquisition unit 47. As a result, the compounding prediction device 120 can predict the blending amounts and property values ​​of the rubber compound that will yield the target physical properties and sustainability rate.

[0097] The technical ideas embodied in the above embodiments can be generalized to include the technical ideas described in the following items.

[0098] The first item is a data mart generation step that generates a data mart including the blending amounts of multiple materials used in the production of a rubber compound, the property values ​​of the materials, and the physical properties and sustainability rate of the rubber compound; and a physical property prediction step that estimates the physical properties and sustainability rate using a machine learning type computational model with the blending amounts and property values ​​as explanatory variables. The method for generating a physical property prediction model comprises a learning process step of training the computation model based on the data mart generated by the data mart generation step.

[0099] The second item is a method for generating a physical property prediction model as described in the first item, further comprising a base calculation step that calculates feature quantities based on the material and a base calculation table describing the calculation process for the material, and including the feature quantities calculated by the base calculation step as explanatory variables.

[0100] The third item is a method for generating a physical property prediction model as described in item 2, wherein the base calculation table includes at least one of the four basic arithmetic operations.

[0101] The fourth item is a method for generating a physical property prediction model as described in the second or third item, further comprising a composite calculation step which calculates features based on a composite calculation table which describes the calculation process by selecting at least two of the features calculated in the base calculation step, and includes the features calculated in the composite calculation step as explanatory variables.

[0102] The fifth item is a method for generating a physical property prediction model as described in item 4, wherein the composite calculation table includes at least one of the four basic arithmetic operations.

[0103] The sixth item is a physical property prediction model generation device comprising: a data mart generation unit that generates a data mart including the blending amounts of multiple materials used in the production of a rubber compound, the property values ​​of the materials, and the physical properties and sustainability rate of the rubber compound; a physical property prediction processing unit that estimates the physical properties and sustainability rate using a machine learning type computational model with the blending amounts and property values ​​as explanatory variables; and a learning processing unit that trains the computational model based on the data mart generated by the data mart generation unit.

[0104] The seventh item is a formulation prediction method comprising: a target value acquisition step of acquiring target values ​​for the physical properties and sustainability rate of a rubber compound; and a formulation prediction step of predicting the amount and property values ​​of a plurality of materials used in the production of a rubber compound, using a trained machine learning type computational model that estimates the physical properties and sustainability rate, with the amount and property values ​​of the materials used as explanatory variables, and the amount and property values ​​of the materials used to obtain the target values ​​acquired in the target value acquisition step.

[0105] The eighth item is the formulation prediction method described in item seven, in which the formulation prediction step predicts the formulation amount and the property value under constraints on the formulation amount and the property value.

[0106] The ninth item is a compounding prediction device comprising: a target value acquisition unit that acquires target values ​​for the physical properties and sustainability rate of a rubber compound; and a compounding prediction processing unit that uses a trained machine learning type computational model to estimate the physical properties and sustainability rate, with the blending amounts and property values ​​of multiple materials used in the production of the rubber compound as explanatory variables, and predicts the blending amounts and property values ​​from which the target values ​​acquired by the target value acquisition unit can be obtained.

[0107] The embodiments of the present invention have been described above. These embodiments are illustrative, and it will be understood by those skilled in the art that various modifications and changes are possible within the scope of the claims of the present invention, and that such modifications and changes are also within the scope of the claims of the present invention. Accordingly, the descriptions and drawings herein should be treated as illustrative rather than limiting. [Explanation of Symbols]

[0108] 11 Data marts, 22 Base calculation tables, 27 Compound calculation tables, 41 Data mart generation unit, 45 Physical property prediction processing unit, 45a Calculation model, 46 Learning processing unit, 47 Target value acquisition unit, 48 Formulation prediction processing unit, 100. Physical property prediction model generation device, 120. Formulation prediction device.

Claims

1. A data mart generation step that generates a data mart including the blending amounts of multiple materials used in the production of a rubber compound, the property values ​​of the materials, and the physical properties and sustainability rate of the rubber compound, A property prediction step in which the aforementioned blending amount and the aforementioned property values ​​are used as explanatory variables and the aforementioned physical properties and the aforementioned sustainability rate are estimated by a machine learning type computational model, A learning process step in which the computational model is trained based on the data mart generated in the data mart generation step, A method for generating a material property prediction model that includes the following features.

2. The system further comprises a base calculation step for calculating feature quantities based on the material and a base calculation table describing the calculation process applied to the material, The method for generating a physical property prediction model according to claim 1, which includes the feature quantities calculated by the base calculation step as explanatory variables.

3. The method for generating a physical property prediction model according to claim 2, wherein the base calculation table includes at least one of the four basic arithmetic operations.

4. The further step comprises a composite calculation step that calculates feature quantities based on calculations performed by a composite calculation table which describes calculations performed by selecting at least two of the feature quantities calculated by the base calculation step, The method for generating a physical property prediction model according to claim 2, which includes the feature quantities calculated by the above-mentioned complex calculation step as explanatory variables.

5. The method for generating a physical property prediction model according to claim 4, wherein the composite calculation table includes at least one of the four basic arithmetic operations.

6. A data mart generation unit generates a data mart that includes the blending amounts of multiple materials used in the production of a rubber compound, the property values ​​of the materials, and the physical properties and sustainability rate of the rubber compound. A physical property prediction processing unit that uses the aforementioned blending amount and the aforementioned property values ​​as explanatory variables and estimates the physical properties and the sustainability rate using a machine learning type computational model, A learning processing unit that trains the calculation model based on the data mart generated by the data mart generation unit, A physical property prediction model generation device equipped with the following features.

7. A target value acquisition step to obtain target values ​​for the physical properties and sustainability rate of rubber compound, A compounding prediction step that uses a trained machine learning type computational model that estimates the physical properties and sustainability rate, with the blending amounts and property values ​​of multiple materials used in the production of a rubber compound as explanatory variables, to predict the blending amounts and property values ​​that will yield the target value obtained in the target value acquisition step, A method for predicting formulations that includes the following features.

8. The formulation prediction method according to claim 7, wherein the formulation prediction step predicts the amount of formulation and the property values ​​under constraints on the amount of formulation and the property values.

9. A target value acquisition unit that acquires target values ​​for the physical properties and sustainability rate of rubber compounding, A compounding prediction processing unit predicts the compounding amounts and property values ​​from which the target values ​​obtained by the target value acquisition unit can be obtained, using a trained machine learning type computational model that estimates the physical properties and the sustainability rate, with the compounding amounts and property values ​​from which A formulation prediction device equipped with the following features.

Images