Generation method, information processing apparatus, and generation program

Abstract

There is a need for a technique for generating learning data that can be used for common machine learning from material data with different types and combinations of raw materials. 【Solution means】The method for generating a learning data set includes a step of obtaining a plurality of first source data used for generating the learning data set. Each of the plurality of first source data associates the blending ratio of the raw materials constituting the material, the analysis information indicating the characteristics of the various raw materials, and the first performance related to the material. The generation method further includes, for each of the plurality of first source data, a step of generating a first explanatory variable using the blending ratio defined in the first source data and the analysis information related to the raw materials defined in the first source data, and associating the first performance defined in the first source data with the first explanatory variable generated for each of the plurality of first source data as an objective variable, thereby generating a first learning data set.

Description

Technical Field

[0001] The present disclosure relates to a generation method, an information processing apparatus, and a generation program.

Background Art

[0002] Japanese Unexamined Patent Application Publication No. 2024-59737 (Patent Document 1) discloses a physical property data prediction method aimed at accurately predicting the physical property data of an adhesive composition. In the physical property data prediction method, the adhesive composition is used as explanatory variable data, the physical property data of the adhesive composition is used as objective variable data, and the relationship between the explanatory variable data and the objective variable data is machine-learned. Examples of the explanatory variable data include the physical property information of the raw materials of the adhesive composition, the information on the blending ratio of the raw materials, the information on the processing conditions of the adhesive composition, and the information on the test conditions for evaluating the physical properties of the adhesive composition.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In order to ensure a sufficient number of learning data, it is desired to share material data among different research and development themes. Researchers and developers make materials while changing the types of raw materials, combinations of raw materials, and blending conditions of raw materials according to their own research and development themes, and evaluate the performance of each material. As a result, a large number of material data associating the performance evaluation results of materials with the combinations of raw materials and the blending conditions of raw materials are obtained in each research and development theme.

[0005] In material data used for different research and development themes, even if the types of performance to be evaluated are common, the types and combinations of raw materials associated with the performance may not be common. It is difficult to use these material data as common machine learning for learning data.

[0006] Therefore, a technique for generating learning data that can be used for common machine learning from material data with different types and combinations of raw materials is desired.

Means for Solving the Problem

[0007] In an example of the present disclosure, a method for generating a learning data set is provided. The generation method includes a step of acquiring a plurality of first source data used for generating the learning data set. Each of the plurality of first source data associates the blending ratio of raw materials constituting the material, analysis information indicating the characteristics of the various raw materials, and a first performance related to the material. The generation method further includes, for each of the plurality of first source data, a step of generating a first explanatory variable using the blending ratio defined in the first source data and the analysis information related to the raw materials defined in the first source data, and a step of generating a first learning data set by associating the first performance defined in the first source data with the first explanatory variable generated for each of the plurality of first source data as an objective variable.

[0008] In one example of the present disclosure, the generation method further includes a step of obtaining a plurality of second source data used for generating a learning dataset. Each of the plurality of second source data associates the blending ratio of raw materials constituting the material, the structural information of the various raw materials, and the second performance related to the material. The generation method further includes, for each of the plurality of second source data, a step of generating a second explanatory variable using the blending ratio defined in the second source data and the structural information defined in the second source data, and a step of generating a second learning dataset by associating the second performance defined in the second source data with the generated second explanatory variable for each of the plurality of second source data as an objective variable.

[0009] In one example of the present disclosure, the step of generating the first explanatory variable includes compressing the analysis information related to each raw material defined in each of the plurality of first source data into a predetermined dimension, summing the analysis information of each dimension according to the blending ratio defined in the first source data, and calculating the sum result of each dimension as the first explanatory variable.

[0010] In one example of the present disclosure, the step of generating the second explanatory variable includes converting the structural information of each raw material defined in each of the plurality of second source data into a common representation format, summing the structural information converted into the common format according to the blending ratio defined in the second source data, and calculating the sum result as the second explanatory variable.

[0011] In one example of the present disclosure, the representation format is a format representing the structural information by molecular descriptors, a format representing the structural information by fingerprints, or a format representing the structure of raw materials by feature vectors.

[0012] In one example of the present disclosure, the analysis information is data obtained by analyzing raw materials by spectroscopy. The spectroscopy includes infrared spectroscopy, visible spectroscopy, ultraviolet spectroscopy, Raman spectroscopy, electron spin resonance spectroscopy, X-ray analysis, or nuclear magnetic resonance spectroscopy.

[0013] In an example of the present disclosure, the above analysis information is data obtained by analyzing a raw material by chromatography.

[0014] In an example of the present disclosure, the above material is an adhesive material.

[0015] In an example of the present disclosure, each of the above first performance and the above second performance includes at least one of a characteristic indicating the strength required for peeling of the adhesive material and the adherend, a characteristic indicating the difficulty of displacement of the adhesive material with respect to the adherend, and a characteristic indicating the ease of adhesion between the adhesive material and the adherend.

[0016] In an example of the present disclosure, the above generation method further includes a step of generating a first learning model by machine learning using the above first learning dataset, and a step of generating a second learning model by machine learning using the above second learning dataset.

[0017] In another example of the present disclosure, an information processing apparatus is provided. The information processing apparatus includes a control unit for controlling the information processing apparatus. The control unit executes a process of acquiring a plurality of first source data used for generating a learning dataset. Each of the plurality of first source data associates a blending ratio of raw materials constituting a material, analysis information indicating characteristics of the various raw materials, and first performance regarding the material. The control unit further executes, for each of the plurality of first source data, a process of generating a first explanatory variable using the blending ratio defined in the first source data and the above analysis information regarding the raw materials defined in the first source data, and a process of generating a first learning dataset by associating the first performance defined in the first source data with the generated first explanatory variable for each of the plurality of first source data as an objective variable.

[0018] In another example of the present disclosure, a program for generating a training dataset is provided. The program causes an information processing device to execute a process of acquiring a plurality of first original data used to generate a training dataset. Each of the plurality of first original data associates a blending ratio of ingredients constituting a material, analytical information indicating characteristics of the ingredients, and a first performance related to the material. The program further causes the information processing device to execute a process of generating, for each of the plurality of first original data, a first explanatory variable using the blending ratio defined in the first original data and the analytical information related to the ingredients defined in the first original data, and a process of generating a first training dataset by associating the first explanatory variable generated for each of the plurality of first original data with the first performance defined in the first original data as a dependent variable.

[0019] The above and other objects, features, aspects and advantages of the present invention will become apparent from the following detailed description of the invention taken in conjunction with the accompanying drawings. [Brief explanation of the drawings]

[0020]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

[0021] Hereinafter, each embodiment according to the present invention will be described with reference to the drawings. In the following description, the same parts and components are denoted by the same reference numerals. Their names and functions are also the same. Therefore, detailed description thereof will not be repeated. Note that each embodiment and each modified example described below may be selectively combined as appropriate.

[0022] [First embodiment] <A.概要> An overview of a method for generating a training dataset according to the first embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram schematically showing a process for generating a training dataset 124 from a material dataset 122.

[0023] The generation function of the learning dataset 124 is implemented, for example, in the information processing apparatus 100. The information processing apparatus 100 is, for example, a desktop PC (Personal Computer), a notebook PC, a tablet terminal, a smartphone, or other computers.

[0024] The material dataset 122 includes a plurality of material data 123 (first original data). The material data 123 is, for example, data obtained through experiments or the like in past research and development themes. As an example, the material data 123 associates, in addition to the material ID (Identification) 123A, the raw materials 123B that make up the material, the blending ratio 123C of the raw materials that make up the material, the analysis information 123D indicating the characteristics of the various raw materials, and the performance 123E related to the material.

[0025] The material ID 123A defined in the material data 123 is information for uniquely identifying a material. The material ID 123A may be defined, for example, by a numeric string such as an ID, or by a character string such as a material name.

[0026] The raw materials 123B defined in the material data 123 indicate the combination of raw materials that make up the material indicated by the material ID 123A. In addition, the blending ratio 123C at the time of manufacturing the material is associated with the raw materials 123B. The blending ratio 123C may be a ratio based on weight, a ratio based on mol, or a ratio based on other physical quantities.

[0027] The analysis information 123D defined in the material data 123 indicates the result of analyzing the characteristics of the raw materials 123B. Specific examples of the analysis information 123D will be described later. Note that the relationship between the raw materials 123B and the analysis information 123D may be defined in the material data 123 or in other databases.

[0028] The performance 123E defined in the material data 123 indicates the result of evaluating the properties of the material. Specific examples of the performance 123E will be described later.

[0029] Since each of the material data 123 is data used in a different research and development theme, even if the type of performance 123E to be evaluated is the same, the type of raw material 123B may be different. It is difficult to use the material data 123 with different types of raw material 123B as common learning data when the number of types of raw materials is large.

[0030] Therefore, the information processing device 100 generates a common explanatory variable 125A (first explanatory variable) from the material data 123 having different types of raw materials 123B. Then, the information processing device 100 associates the generated explanatory variable 125A with performance 123E (first performance) to generate a learning dataset 124 that can be used in common machine learning.

[0031] More specifically, for each of the material data 123, the information processing device 100 generates explanatory variables 125A using the blending ratio 123C defined in the material data 123 and the analysis information 123D related to the raw materials defined in the material data 123. Typically, the information processing device 100 generates explanatory variables 125A of the same type by summing the analysis information 123D of each raw material according to the blending ratio 123C.

[0032] Thereafter, the information processing device 100 associates the performance 123E defined in the material data 123 with the explanatory variables 125A generated for each of the material data 123 as the objective variables 125B, and generates the training dataset 124.

[0033] As described above, the information processing device 100 generates training data 125 that can be used in common machine learning from material data 123 having different types of raw materials 123B. This increases the number of training data 125, and allows material data 123 generated under different research and development themes to be effectively utilized.

[0034] <B.材料> Next, referring to FIG. 2, specific examples of the materials defined in the above-described material data 123 (see FIG. 1) will be described. The material is, for example, an adhesive material.

[0035] FIG. 2 is a diagram showing an adhesive material which is an example of the material. The "adhesive material" means a material having a function of joining objects. Examples of the adhesive material include adhesive tapes and adhesives. Hereinafter, the adhesive tape will be taken as an example to describe, but the adhesive material is not limited to the adhesive tape.

[0036] The adhesive tape may be a single-sided tape or a double-sided tape. Also, the use of the adhesive tape is not particularly limited and may be for industrial use or for household use. Examples of the types of adhesive tapes include coreless tapes, non-woven core tapes, film core tapes, foam core tapes, metal foil core tapes, and the like.

[0037] FIG. 2 shows a single-sided adhesive tape 10A which is an example of the adhesive material 10. The single-sided adhesive tape 10A is composed of, for example, a sheet-like core base material 12, an adhesive 14, and a release agent 16.

[0038] The type of the core base material 12 is not particularly limited. The core base material 12 may be a plastic film, paper, foam, metal foil, or other types of sheets.

[0039] The adhesive 14 is applied to one surface of the core base material 12. The release agent 16 is applied to the other surface of the core base material 12.

[0040] The release agent 16 covers the adhesive 14 in a state where the single-sided adhesive tape 10A is wound in a roll shape and protects the adhesive 14 so that the adhesive force does not decrease.

[0041] <C. Analysis Information 123D> Next, with reference to FIG. 3, a specific example of the above-described analysis information 123D (see FIG. 1) will be described. FIG. 3 is a diagram showing an example of the analysis information 123D.

[0042] When raw materials are purchased from other companies, characteristics such as the structure and properties of the raw materials are unknown. In this case, the manufacturer of the material analyzes the characteristics of the raw materials using some kind of analytical device.

[0043] The means for analyzing the raw materials is not particularly limited. As an example, the analysis information 123D is data obtained by analyzing the raw materials by spectroscopy. The spectroscopy includes infrared spectroscopy, visible spectroscopy, ultraviolet spectroscopy, Raman spectroscopy, electron spin resonance method, X-ray analysis method, or nuclear magnetic resonance spectroscopy. As another example, the analysis information 123D is data obtained by analyzing the raw materials by chromatography.

[0044] In the example of FIG. 3, as the analysis information 123D, IR data 123D1 obtained by analyzing the raw materials by infrared spectroscopy and GPC data 123D2 obtained by analyzing the raw materials by chromatography are shown.

[0045] The IR data 123D1 is, for example, the analysis result of the raw materials by an infrared spectrometer. The infrared spectrometer irradiates the raw materials with infrared light and measures the transmitted or reflected light to perform structural analysis of the raw materials. The horizontal axis of the IR data 123D1 indicates the wavelength of the transmitted or reflected light. The vertical axis of the IR data 123D1 represents the amount of infrared rays absorbed by the raw materials.

[0046] The GPC data 123D2 is, for example, the measurement result of a raw material obtained by a chromatography analyzer. A chromatography analyzer is a device for measuring the molecular weight distribution of a raw material. A chromatography analyzer separates raw material dissolved in a solvent based on molecular size using a porous gel in a column. Molecules with large molecular weights pass through the column quickly. On the other hand, molecules with small molecular weights enter the pores of the gel and take longer to pass through the column. The chromatography analyzer utilizes these differences in retention time to sequentially detect components separated by molecular weight. The horizontal axis of the GPC data 123D2 represents, for example, retention time. This retention time correlates with molecular weight. The vertical axis of the GPC data 123D2 represents, for example, the intensity of the detection signal. This intensity corresponds to the content of the component.

[0047] <D.性能123E> Next, a specific example of the performance 123E defined in the material data 123 (see FIG. 1) will be described.

[0048] The performance 123E indicates, for example, the properties relating to adhesion (hereinafter also referred to as "adhesion properties") of the adhesive material 10 (see FIG. 2). The adhesion properties are expressed by physical quantities indicating the adhesive strength of the adhesive material 10, etc.

[0049] The adhesive properties are actual values measured using various measuring devices such as testing machines. The adhesive properties include at least one of the strength required to peel the adhesive material 10 from the adherend (hereinafter also referred to as "adhesion strength"), the resistance of the adhesive material 10 to slippage relative to the adherend (hereinafter also referred to as "holding strength"), and the ease with which the adhesive material 10 adheres to the adherend (hereinafter also referred to as "tack"). As an example, the adhesive properties are defined by three factors: adhesive strength, holding strength, and tack.

[0050] Examples of adhesive strength include peel adhesive strength, shear adhesive strength, and split adhesive strength.

[0051] The peel adhesion is represented by, for example, the force required to peel the adhesive material in a direction perpendicular to the adherend surface (so-called 90° peel adhesion). Alternatively, the peel adhesion is represented by the force required to peel the adhesive material in a direction parallel to the adherend surface (so-called 180° peel adhesion).

[0052] The shear adhesion corresponds to the force when a force in the opposite direction (i.e., shear stress) is applied to the adherends joined by the adhesive material in a horizontal plane to break the joint.

[0053] The cleavage adhesion corresponds to the force when the adherends are peeled off by applying a force in the opposite direction in a vertical plane to the adherends joined by the adhesive material.

[0054] Examples of the holding force include the shear holding force and the constant load holding force. The shear holding force is represented by the distance by which the adhesive material is displaced when a load is applied in the shear direction to the adhesive material attached to the adherend for a certain period of time. Alternatively, the holding force is represented by the time from when a load is applied in the shear direction to the adhesive material attached to the adherend until the adhesive material falls off. On the other hand, the constant load holding force is represented by the distance by which the adhesive material is displaced when a load is applied in the vertical direction to the adhesive material attached to the adherend, or the time from when the load is applied until the adhesive material falls off.

[0055] Examples of tack include ball tack. The ball tack is measured by rolling balls with different diameters in sequence on an inclined surface to which the adhesive material is attached. The ball tack is represented by the maximum diameter of the ball that stops on the inclined surface.

[0056] <E. Hardware Configuration of Information Processing Apparatus 100> Next, with reference to FIG. 4, the hardware configuration of the information processing apparatus 100 will be described. FIG. 4 is a diagram showing an example of the hardware configuration of the information processing apparatus 100.

[0057] The information processing device 100 includes a control device 101 (control unit), a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, a communication interface 104, a display interface 105, an input interface 107, and an auxiliary storage device 120. These components are connected to a bus 115.

[0058] The control device 101 is configured, for example, by at least one integrated circuit. The integrated circuit may be configured, for example, by at least one central processing unit (CPU), at least one graphics processing unit (GPU), at least one application specific integrated circuit (ASIC), at least one field programmable gate array (FPGA), or a combination thereof.

[0059] The control device 101 controls the operation of the information processing device 100 by executing various programs. Upon receiving an execution command for one of the programs, the control device 101 reads the program to be executed from the auxiliary storage device 120 or the ROM 102 into the RAM 103. The RAM 103 functions as a working memory and temporarily stores various data required for executing the program.

[0060] A LAN (Local Area Network), an antenna, etc. are connected to the communication interface 104. The information processing device 100 exchanges data with external devices via the communication interface 104. The external devices include, for example, servers.

[0061] A display interface 105 is connected to a display device 106. The display interface 105 sends an image signal for displaying an image to the display device 106 in accordance with a command from the control device 101 or the like. The display device 106 is, for example, a liquid crystal display, an organic EL (Electro Luminescence) display, or other display. Note that the display device 106 may be integrally configured with the information processing apparatus 100 or may be configured separately from the information processing apparatus 100.

[0062] An input interface 107 is connected to an input device 108. The input device 108 is, for example, a mouse, a keyboard, a touch panel, or other device capable of receiving a user operation. Note that the input device 108 may be integrally configured with the information processing apparatus 100 or may be configured separately from the information processing apparatus 100.

[0063] The auxiliary storage device 120 is a storage medium such as a hard disk or a flash memory, for example. The auxiliary storage device 120 stores, for example, the above-described material data set 122, the above-described learning data set 124, the learning model 126, the generation program 128, and the learning program 130. These storage locations are not limited to the auxiliary storage device 120 and may be stored in a storage area (for example, a cache memory or the like) of the control device 101, the ROM 102, the RAM 103, an external device (for example, a server), or the like.

[0064] The generation program 128 is a program for generating the learning dataset 124 from the material dataset 122. Since the generation process is as described above, its description will not be repeated. The generation program 128 may be provided incorporated into a part of any program, rather than as a single program. In this case, the learning process by the generation program 128 is realized in cooperation with any program. Even a program that does not include such a part of the module does not deviate from the gist of the generation program 128 according to the present embodiment. Furthermore, some or all of the functions provided by the generation program 128 may be realized by dedicated hardware. Furthermore, the information processing apparatus 100 may be configured in a form such as a so-called cloud service in which at least one server executes a part of the processing of the generation program 128.

[0065] The learning program 130 is a program for generating the learning model 126 using the learning dataset 124. The learning process for generating the learning model 126 will be described later. The learning program 130 may be provided incorporated into a part of any program, rather than as a single program. In this case, the learning process by the learning program 130 is realized in cooperation with any program. Even a program that does not include such a part of the module does not deviate from the gist of the learning program 130 according to the present embodiment. Furthermore, some or all of the functions provided by the learning program 130 may be realized by dedicated hardware. Furthermore, the information processing apparatus 100 may be configured in a form such as a so-called cloud service in which at least one server executes a part of the processing of the learning program 130.

[0066] <F. Functional Configuration> Next, with reference to FIGS. 5 to 10, the functional configuration of the information processing apparatus 100 will be described. FIG. 5 is a diagram showing an example of the functional configuration of the information processing apparatus 100.

[0067] As shown in FIG. 5, the information processing apparatus 100 includes, as functional components, a generation unit 150, a learning unit 152, a prediction unit 154, and an output unit 156. These functional components will be described in order below.

[0068] Note that the generation unit 150, the learning unit 152, the prediction unit 154, and the output unit 156 do not necessarily have to be implemented in one information processing apparatus 100. Some of the functional components of the generation unit 150, the learning unit 152, the prediction unit 154, and the output unit 156 may be implemented in the information processing apparatus 100, and the remaining functional components may be implemented in another computer such as a server.

[0069] (F1. Generation Unit 150) First, with reference to FIGS. 6 to 8, the functions of the generation unit 150 shown in FIG. 5 will be described. As described with reference to FIG. 1, the generation unit 150 generates a common explanatory variable 125A by adding up the analysis information 123D of each raw material according to the blending ratio 123C. More specifically, the generation unit 150 compresses the analysis information 123D related to each raw material defined in each of the material data 123 into a predetermined dimension. Then, the generation unit 150 adds up the analysis information 123D compressed into a predetermined dimension according to the blending ratio 123C defined in the material data 123, and calculates the addition result as the explanatory variable 125A.

[0070] FIG. 6 is a diagram for explaining an example of generating an explanatory variable 125A1 from IR data 123D1 (see FIG. 3), which is an example of the analysis information 123D.

[0071] As shown in FIG. 6, the generation unit 150 compresses the IR data 123D1 into a feature amount of a predetermined dimension. The number of dimensions is not particularly limited. In the example of FIG. 6, the IR data 123D1 is compressed into a two-dimensional feature amount. The compression process can be realized by any method.

[0072] As an example, in the compression process, the encoder part of a VAE (Variational Auto Encoder) is used. A VAE is a type of neural network based on deep learning. A VAE is composed of an encoder part that compresses high-dimensional data into low-dimensional feature quantities, and a decoder part that restores the original data from the compressed feature quantities. In pre-training, the internal parameters of the encoder part and the decoder part are updated so that the compressed IR data 123D1 is restored to its original state. Such learning is performed in advance for each raw material. The information processing apparatus 100 can compress the IR data 123D1 into feature quantities of a predetermined dimension while retaining the characteristics of each raw material by using the learned encoder part.

[0073] Note that the compression process is not limited to VAE as long as it is a means capable of compressing the IR data 123D1 into low-dimensional feature quantities. As another example, the compression process may be realized by principal component analysis.

[0074] The generation unit 150 adds up the IR data 123D1 compressed into feature quantities of a predetermined dimension according to the blending ratio 123C of each raw material, and generates the addition result as the explanatory variable 125A1.

[0075] In the example of FIG. 6, for the feature quantity of dimension "1", the generation unit 150 performs the calculation of "4.6097×0.6 + 0.0269×0.2 + 5.4815×0.2", and generates the calculation result "3.86750" as the explanatory variable 125A1. Also, for the feature quantity of dimension "2", the generation unit 150 performs the calculation of "4.7769×0.6 + 0.0993×0.2 + 5.5131×0.2", and generates the calculation result "3.98862" as the explanatory variable 125A1.

[0076] FIG. 7 is a diagram for explaining an example of generating an explanatory variable 125A2 from GPC data 123D2, which is another example of the analysis information 123D.

[0077] As shown in FIG. 7, the generation unit 150 compresses the GPC data 123D2 into feature amounts of a predetermined dimension. The number of dimensions is not particularly limited. In the example of FIG. 7, the GPC data 123D2 is compressed into two-dimensional feature amounts. The compression process can be realized by any method.

[0078] As an example, in the compression process, the encoder part of the VAE is used. The VAE is a kind of neural network based on deep learning. The VAE is composed of an encoder part that compresses high-dimensional data into low-dimensional feature amounts and a decoder part that restores the original data from the compressed feature amounts. In pre-training, the internal parameters of the encoder part and the decoder part are updated so that the compressed GPC data 123D2 is restored to its original state. Such learning is executed in advance separately for each raw material. The information processing device 100 can compress the GPC data 123D2 into feature amounts of a predetermined dimension while retaining the features of each raw material by using the learned encoder part.

[0079] Note that the compression process is not limited to the VAE as long as it is a means capable of compressing the GPC data 123D2 into low-dimensional feature amounts. As another example, the compression process may be realized by principal component analysis.

[0080] The generation unit 150 sums up the GPC data 123D2 compressed into feature amounts of a predetermined dimension according to the blending ratio 123C of each raw material, and generates the sum result as the explanatory variable 125A2.

[0081] In the example of FIG. 7, for the feature amount of dimension "1", the generation unit 150 performs the calculation of "2.7107×0.7 + 0.301×0.2 + 3.2785×0.1", and generates "2.28554", which is the calculation result, as the explanatory variable 125A2. Also, for the feature amount of dimension "2", the generation unit 150 performs the calculation of "2.9329×0.7 + 0.0143×0.2 + 4.6202×0.1", and generates "2.51791", which is the calculation result, as the explanatory variable 125A2.

[0082] FIG. 8 is a diagram showing an example of the learning data set 124 generated by the generation unit 150. The generation unit 150 associates the explanatory variables 125A1 and 125A2 generated from each of the material data 123 with the performance 123E defined in the material data 123 as the objective variable 125B. Thereby, the generation unit 150 generates the learning data set 124 from the material data set 122.

[0083] (F2. Learning Unit 152) Next, with reference to FIG. 9, the functions of the learning unit 152 shown in FIG. 5 will be described. FIG. 9 is a diagram conceptually showing the learning process by the learning unit 152.

[0084] The learning unit 152 executes a learning process using the above-described learning data set 124 generated by the generation unit 150, and generates a learning model 126 (first learning model). The machine learning algorithm to be adopted is not particularly limited, and for example, various machine learning algorithms such as neural networks such as deep learning, support vector machines, or decision trees can be adopted. Hereinafter, the learning process using a neural network will be described.

[0085] The learning model 126 is composed of an input layer X, an intermediate layer H, and an output layer Y.

[0086] The input layer X receives the input of the explanatory variables defined in the learning data 125. As an example, the input layer X includes a unit group x1 and a unit group x2.

[0087] The unit group x1 is configured to receive the input of, for example, the explanatory variable 125A1 defined in the learning model 126. The number of units constituting the unit group x1 is the same as the number of dimensions of the explanatory variable 125A1. As an example, when the explanatory variable 125A1 is two-dimensional, the unit group x1 is composed of two units. Each unit constituting the unit group x1 outputs the input data to each unit in the first layer of the intermediate layer H.

[0088] The unit group x2 is configured to receive the input of the explanatory variable 125A2 defined in the learning model 126, for example. The number of units constituting the unit group x2 is the same as the number of dimensions of the explanatory variable 125A2. As an example, when the explanatory variable 125A2 is two-dimensional, the unit group x2 is composed of two units. Each unit constituting the unit group x2 outputs the input data to each unit in the first layer of the intermediate layer H.

[0089] The intermediate layer H is composed of a plurality of layers. The number of layers of the intermediate layer H is arbitrary. In the example of FIG. 9, the intermediate layer H is composed of N layers (N is a natural number). Each layer of the intermediate layer H includes a plurality of units. In the example of FIG. 9, the first layer of the intermediate layer H is the unit h A1 , h A2 ··· and is composed. The last layer of the intermediate layer H is the unit h N1 , h N2 ··· and is composed.

[0090] Each unit constituting each layer of the intermediate layer H is connected to each unit of the previous layer and each unit of the next layer. Each unit of each layer receives each output value from each unit of the previous layer, multiplies each output value by a weight, accumulates the multiplication results, adds (or subtracts) a predetermined bias to the accumulated result, inputs the addition result (or subtraction result) to a predetermined function (for example, a sigmoid function), and outputs the output value of the function to each unit of the next layer.

[0091] The output layer Y outputs the adhesion characteristics corresponding to the viscoelastic characteristics input to the input layer X. As an example, the output layer Y is composed of units y1 to y3. Hereinafter, the units y1 to y3 are also referred to as units y.

[0092] Each of the units y is each unit h of the last layer of the intermediate layer H N1 , h N2... is connected to. Each of the units in unit y receives the output values from each unit in the final layer of the intermediate layer H, multiplies each output value by a weight, accumulates the multiplication results, adds (or subtracts) a predetermined bias to the accumulated result, inputs the addition result (or subtraction result) into a predetermined function (for example, a sigmoid function), and outputs the output result of the function as an output value.

[0093] The number of units constituting the output layer Y is determined according to the number of target variables defined in the learning data 125. As an example, when predicting the above-mentioned adhesive force, the above-mentioned holding force, and the above-mentioned tack, the number of units constituting the output layer Y is three.

[0094] Unit y1 is configured to output the adhesive force as a prediction result. Unit y2 is configured to output the holding force as a prediction result. Unit y3 is configured to output the tack as a prediction result.

[0095] In the example of FIG. 9, one learning model 126 is configured to output three prediction values of the adhesive force, the holding force, and the tack. However, one learning model 126 may output one prediction value. As an example, the first learning model 126 is configured to receive the input of the viscoelastic properties of the adhesive material and output the adhesive force of the adhesive material. The second learning model 126 is configured to receive the input of the viscoelastic properties of the adhesive material and output the holding force of the adhesive material. The third learning model 126 is configured to receive the input of the viscoelastic properties of the adhesive material and output the tack of the adhesive material.

[0096] Next, the update process of the internal parameters of the learning model 126 by the learning unit 152 will be described.

[0097] The learning unit 152 inputs the explanatory variables 125A1 and 125A2 defined in the first learning data 125 into the learning model 126. As a result, the learning model 126 outputs the adhesion characteristics corresponding to the input explanatory variables 125A1 and 125A2. Next, the learning unit 152 compares the output prediction results "s1" to "s3" with the objective variable defined in the first learning data 125. The objective variable is represented by scores "sA" to "sC", for example.

[0098] The learning unit 152 calculates the error "Z" between the output results "s1" to "s3" of the learning model 126 and the objective variables "sA" to "sC". As an example, the error "Z" is calculated based on the following formula (1).

[0099] Z = {(s1 - sA) 2 + (s2 - sB) 2 + (s3 - sC) 2} / 3 ··· (1) Next, the learning unit 152 updates various parameters (for example, weights and biases) included in the learning model 126 so that the error "Z" becomes smaller. The update of the parameters is realized by, for example, the error backpropagation method.

[0100] The learning unit 152 repeatedly performs the update process of the internal parameters of the learning model 126 for each learning data 125 included in the learning data set 124. As a result, the learning model 126 will output accurate prediction results as learning progresses.

[0101] Note that the learning unit 152 does not necessarily need to use all the learning data 125 included in the learning data set 124 for the learning process, and the learning model 126 may be generated using some of the learning data 125 included in the learning data set 124. The remaining learning data 125 is used for evaluating the learning model 126 and so on.

[0102] (F3. Prediction Unit 154) Next, the function of the prediction unit 154 shown in Fig. 5 will be described with reference to Fig. 10. Fig. 10 is a diagram conceptually showing the processing by the prediction unit 154.

[0103] The prediction unit 154 receives user input of a combination of raw materials and the blending ratio of each raw material, and predicts the performance of the material according to the combination of raw materials and the blending ratio. The user input is received, for example, on an input screen displayed on the display device 106 described above. Operations on the input screen are realized, for example, using the input device 108 described above.

[0104] More specifically, the prediction unit 154 first acquires analytical information 123D for each ingredient input by the user. The ingredients input by the user may be the same as or different from the ingredients used during training. The prediction unit 154 then compresses the analytical information 123D for each ingredient into a feature quantity of a predetermined dimension using the method described above with reference to FIGS. 6 and 7. Next, the prediction unit 154 sums the analytical information 123D compressed to the predetermined dimension according to the blending ratio input by the user, and generates the summation results as explanatory variables 125A1 and 125A2.

[0105] Thereafter, the prediction unit 154 inputs the generated explanatory variables 125A1 and 125A2 into the learning model 126. As a result, the learning model 126 outputs the performance of the material as a prediction result. As a result, the user can evaluate the performance of various materials by inputting the combination of raw materials, the blending ratio, etc.

[0106] (F4. Output section 156) Next, the function of the output unit 156 shown in FIG. 5 will be described.

[0107] The output unit 156 outputs the prediction result by the prediction unit 154. The prediction result may be output to any destination. As an example, the prediction result is displayed on the display device 106 of the information processing device 100. As another example, the prediction result is saved as data.

[0108] <G. Flowchart> Next, with reference to FIG. 11, an example of the process flow for generating the learning dataset 124 from the material dataset 122 will be described. FIG. 11 is a flowchart showing the process flow for generating the learning dataset 124.

[0109] The control device 101 of the information processing apparatus 100 functions as the above-described generation unit 150 (see FIG. 5) by executing the above-described generation program 128 (see FIG. 4), and executes the process shown in FIG. 11. In other aspects, part or all of the process shown in FIG. 11 may be executed by circuit elements or other hardware.

[0110] In step S110, the control device 101 acquires the above-described material dataset 122 (see FIG. 1). The acquisition source of the material dataset 122 is arbitrary. As an example, the acquisition source of the material dataset 122 may be the auxiliary storage device 120 of the information processing apparatus 100 or an external device.

[0111] In step S112, the control device 101 initializes the variable "i". As an example, the variable "i" is initialized to "1".

[0112] In step S114, the control device 101 acquires the i-th material data 123 included in the above-described material dataset 122.

[0113] In step S116, the control device 101 acquires the analysis information 123D for the raw material 123B defined in the material data 123 acquired in step S114.

[0114] In step S118, the control device 101 compresses the analysis information 123D acquired in step S116 into a predetermined dimension. Since the compression process is as described above, the description thereof will not be repeated.

[0115] In step S120, the control device 101 adds up the analyzed information 123D after compression in step S118 according to the blending ratio 123C defined in the material data 123 acquired in step S114. Since the addition process is as described above, the description thereof will not be repeated.

[0116] In step S122, the control device 101 associates the addition result in step S120 with the explanatory variable 125A, and associates the performance 123E defined in the material data 123 acquired in step S114 with the objective variable 125B. Thereby, the i-th learning data 125 associating the explanatory variable 125A and the performance 123E is generated.

[0117] In step S130, the control device 101 determines whether to end the generation process of the learning data 125. As an example, when the variable "i" is greater than a predetermined value, the control device 101 determines to end the generation process. Alternatively, when all the material data 123 included in the material data set 122 are used for the generation process of the learning data 125, the control device 101 determines to end the generation process. When the control device 101 determines to end the generation process of the learning data 125 (YES in step S130), it ends the process shown in FIG. 11. Otherwise (NO in step S130), the control device 101 switches the control to step S132.

[0118] In step S132, the control device 101 increments the variable "i". That is, the control device 101 increases the variable "i" by 1. Thereafter, the control device 101 returns the control to step S114.

[0119] <H. Evaluation Experiment>

[0120] (H1. First Evaluation Experiment) Next, with reference to FIG. 12, the first evaluation experiment conducted by the inventors will be described. FIG. 12 is a diagram showing the experimental results G1 and G2 related to the first evaluation experiment.

[0121] More specifically, the inventors generated a prediction model (hereinafter also referred to as "learning model 126X") according to the comparative example using only the material data 123 used in a single research and development theme, and conducted a performance evaluation experiment on the learning model 126X. The experimental result G1 is a diagram showing the result of the performance evaluation experiment of the learning model 126X.

[0122] Twenty material data 123 were used for the learning of the learning model 126X. Also, the learning model 126X was learned to output the 180° peel adhesion as a prediction result.

[0123] The horizontal axis of the experimental result G1 indicates the measured value (correct value) of the 180° peel adhesion. The vertical axis of the experimental result G1 indicates the predicted value of the 180° peel adhesion by the learning model 126X. As shown in the experimental result G1, in the learning model 126X according to the comparative example, the coefficient of determination between the measured value and the predicted value was 0.49.

[0124] Also, the inventors generated a learning model 126 according to the example using the learning dataset 124 generated by the method shown in FIG. 11 above, and conducted a performance evaluation experiment on the learning model 126. The experimental result G2 is a diagram showing the result of the performance evaluation experiment of the learning model 126.

[0125] Two hundred and fifty learning data 125 generated from the material dataset 122 were used for the learning of the learning model 126. Also, the learning model 126 was learned to output the 180° peel adhesion as a prediction result.

[0126] The horizontal axis of the experimental result G2 indicates the measured value (correct value) of the 180° peel adhesion. The vertical axis of the experimental result G2 indicates the predicted value of the 180° peel adhesion by the learning model 126. As shown in the experimental result G2, in the learning model 126 according to the example, the coefficient of determination between the measured value and the predicted value was 0.70.

[0127] As described above, it was confirmed that the prediction accuracy of the learning model 126 according to the embodiment is higher than that of the learning model 126X according to the comparative example.

[0128] (H2. Second Evaluation Experiment) Next, referring to FIG. 13, a second evaluation experiment conducted by the inventors will be described. FIG. 13 is a diagram showing the experimental results related to the second evaluation experiment.

[0129] In this evaluation experiment, the inventors generated a learning model 126_1 according to the embodiment using the learning dataset 124 generated by the method shown in FIG. 11 above. For this learning, 200 learning data 125 included in the learning dataset 124 were used. Also, Gaussian process regression was used as the learning means. The learning model 126_1 was trained to output the 180° peel adhesion as a prediction result. Also, in this evaluation experiment, the inventors generated a learning model 126X_1 for comparison using the material data 123 as it is. For this learning, 20 material data 123 were used. Also, Gaussian process regression was used as the learning means in this experiment. The learning model 126X_1 was trained to output the 180° peel adhesion as a prediction result.

[0130] The learning models 126_1 and 126X_1, which are Gaussian process regression models, output the deviation of the predicted value together with the predicted value when receiving the input of the explanatory variable. The inventors used the evaluation data of different formulations 1 to 5 in which the types and blending ratios of the raw materials were randomly selected to obtain the deviation output from the learning model 126_1 according to the embodiment and the deviation output from the learning model 126X_1 according to the comparative example, and evaluated the reliability of the learning model based on these deviations.

[0131] The results of the deviations output from the learning models 126_1 and 126X_1 are shown in FIG. 13. As shown in FIG. 13, the deviation of the predicted value by the learning model 126_1 according to the embodiment is smaller than the deviation of the predicted value by the learning model 126X_1 according to the comparative example.

[0132] As described above, it was confirmed that the deviation output from the learning model 126_1 according to the embodiment is smaller than the deviation output from the learning model 126X_1 according to the comparative example. Thus, it was confirmed that the prediction result output from the learning model 126_1 according to the embodiment is more stable than the prediction result output from the learning model 126X_1 according to the comparative example. As a result, it was shown that the reliability of the prediction by the learning model 126_1 according to the embodiment is improved compared to the reliability of the prediction by the learning model 126X_1 according to the comparative example.

[0133] [Second Embodiment] <I. Overview> Next, with reference to FIG. 14, a method for generating the learning dataset 224 according to the second embodiment will be described. FIG. 14 is a diagram schematically showing the process of generating the learning dataset 224 from the material dataset 222.

[0134] The information processing apparatus 100 according to the above-described first embodiment generated the learning dataset 124 by using the analysis information 123D of the raw material for the raw material whose molecular structure was unknown. However, depending on the raw material, the molecular structure may be known. As an example, for raw materials manufactured in-house, the molecular structure is known.

[0135] Therefore, the information processing apparatus 100 according to the second embodiment generates the learning dataset by a different method for the material data composed of raw materials whose molecular structures are known. That is, the information processing apparatus 100 changes the method for generating the learning dataset according to whether the raw material structure is known or unknown. Thereby, the information processing apparatus 100 according to the present embodiment generates not only the learning dataset 124 from the material dataset 122 but also the learning dataset 224 from the material dataset 222.

[0136] 14, the material dataset 222 includes a plurality of material data 223 (second source data). The material data 223 is, for example, data obtained through experiments in past research and development themes. As an example, the material data 223 associates, for each material ID 223A, raw materials 223B constituting the material, a blending ratio 223C of the raw materials constituting the material, molecular structure information 223D of the various raw materials, and performance 223E related to the material.

[0137] The material ID 223A defined in the material data 223 is information for uniquely identifying a material. The material ID 223A may be defined, for example, by a numeric string such as an ID, or by a character string such as a material name.

[0138] The raw material 223B defined in the material data 223 indicates a combination of raw materials that constitute the material indicated by the identifier 223A. The raw material 223B is also associated with a blending ratio 223C used when producing the material. The blending ratio 223C may be a ratio based on weight, a ratio based on moles, or a ratio based on other physical quantities.

[0139] Molecular structure information 223D defined in material data 223 is data defining the molecular structure of a raw material. Material data 223 in this embodiment differs from material data 123 in the first embodiment in that molecular structure information 223D is defined instead of analysis information 123D (see FIG. 1). Molecular structure information 223D is defined by a chemical formula such as a molecular formula, for example.

[0140] Performance 223E defined in material data 223 indicates the results of evaluating the properties of the material. The type of performance 223E defined in material data 223 may be the same as or different from the type of performance 123E defined in material data 123 in the first embodiment. Specific examples of performance 223E are as described above, and therefore will not be described again.

[0141] In the present embodiment, for each of the material data 223, the information processing apparatus 100 generates an explanatory variable 225A (second explanatory variable) using the blending ratio 223C defined in the material data 223 and the molecular structure information 223D related to the raw materials defined in the material data 223. Typically, the information processing apparatus 100 generates the same type of explanatory variable 225A by adding up the molecular structure information 223D of each raw material according to the blending ratio 223C.

[0142] After that, for the explanatory variable 225A generated for each of the material data 223, the information processing apparatus 100 associates the performance 223E (second performance) defined in the material data 223 as the objective variable 225B, and generates a learning data set 224. In each of the learning data 225 included in the learning data set 224, the types of the explanatory variables 225A are common. Similarly, in each of the learning data 225 included in the learning data set 224, the types of the objective variables 225B are common.

[0143] As described above, the information processing apparatus 100 generates the learning data 225 that can be used in common machine learning from the material data 223 with different combinations of raw materials 223B. Thus, in the present embodiment, the learning data 225 can be generated without using the analysis information 123D of the raw materials, and it is not necessary to analyze the molecular structure of the raw materials with an analyzer.

[0144] <J. Functional Configuration> Next, with reference to FIGS. 15 and 16, the functional configuration of the information processing apparatus 100 according to the second embodiment will be described. FIG. 15 is a diagram showing an example of the functional configuration of the information processing apparatus 100.

[0145] As shown in FIG. 15, the information processing apparatus 100 includes, as functional components, a generation unit 150, a learning unit 152, a prediction unit 154, an output unit 156, a generation unit 250, a learning unit 252, a prediction unit 254, and an output unit 256.

[0146] Note that the functional configuration shown in FIG. 15 does not necessarily have to be implemented in the information processing apparatus 100. A part of the functional configuration shown in FIG. 15 may be implemented in the information processing apparatus 100, and the remaining functional configuration may be implemented in another computer such as a server.

[0147] The functions of the generation unit 150, the learning unit 152, the prediction unit 154, and the output unit 156 are as described in FIG. 5 above, so the descriptions thereof will not be repeated. Hereinafter, the functional configurations of the generation unit 250, the learning unit 252, the prediction unit 254, and the output unit 256 will be described in order.

[0148] (J1. Generation Unit 250) First, the function of the generation unit 250 shown in FIG. 15 will be described.

[0149] The generation unit 250 converts the molecular structure information 223D of each raw material defined in each of the material data 223 into a common expression format. Thereafter, the generation unit 250 sums up the molecular structure information 223D converted into the common format according to the blending ratio 223C defined in the material data 223, and calculates the summation result as the explanatory variable 225A.

[0150] More specifically, first, the generation unit 250 unifies each of the molecular structure information 223D in a predetermined notation. The notation is, for example, the SMILES (Simplified Molecular Input Line Entry System) notation or the graph notation. The SMILES notation is a notation for concisely expressing a molecular structure in text. In the SMILES notation, the structure of a molecule is linearly expressed as a character string. Graph is a method for mathematically expressing the relationship between atoms and bonds in a molecule.

[0151] Next, the generation unit 250 converts the molecular structure information 223D with unified notation into a common representation format and quantifies the molecular structure information 223D. The said representation format is, for example, a format representing the molecular structure information 223D with molecular descriptors, a format representing the molecular structure information 223D with fingerprints, or a format representing the molecular structure information 223D with feature vectors.

[0152] A molecular descriptor is a representation format for quantifying the features of a molecule as numerical values. In a molecular descriptor, the physicochemical properties of a molecule (for example, molecular weight and polarity) are represented numerically.

[0153] A fingerprint is a representation format for quantifying a molecular structure as a binary vector. In a fingerprint, for example, whether or not there are specific features or substructures within a molecule is represented by "0" and "1".

[0154] For the conversion from the molecular structure information 223D to a feature vector, for example, mol2vec, transformer, etc. are adopted. Mol2vec, transformer, etc. can be utilized as methods for converting a molecule into a numerical vector.

[0155] As described above, the generation unit 250 digitizes the molecular structure information 223D of each raw material defined in each of the material data 223 by converting the molecular structure information 223D of each raw material according to a common representation format. Thereafter, the generation unit 250 sums up the digitized molecular structure information 223D of each raw material according to the blending ratio 223C defined in the material data 223, and calculates the sum result as the explanatory variable 225A.

[0156] Next, the generation unit 250 associates the explanatory variable 225A generated from each of the material data 223 with the performance 223E defined in the material data 223 as the target variable 225B. Thereby, the generation unit 250 generates a learning dataset 224 from the material dataset 222.

[0157] (J2. Learning unit 252) Next, with reference to FIG. 16, the function of the learning unit 252 shown in FIG. 15 will be described. FIG. 16 is a diagram conceptually showing the learning process by the learning unit 252.

[0158] The learning unit 252 executes a learning process using the learning dataset 224 generated by the generation unit 250, and generates a learning model 226 (first learning model). The machine learning algorithm to be adopted is not particularly limited, and for example, various machine learning algorithms such as neural networks (deep learning), support vector machines, or decision trees can be adopted. Hereinafter, the learning process using a neural network will be described.

[0159] The learning model 226 is composed of an input layer X, an intermediate layer H, and an output layer Y.

[0160] The input layer X receives the input of the explanatory variable 225A defined in the learning data 225. Since other points of the learning model 226 are the same as those of the above-described learning model 126 (see FIG. 9), the description of the learning process of the learning model 226 will not be repeated.

[0161] (J3. Prediction Unit 254) Next, the function of the prediction unit 254 shown in FIG. 15 will be described.

[0162] The prediction unit 254 receives a user input of a combination of raw materials and a blending ratio of each raw material, and predicts the performance of the material according to the combination of the raw materials and the blending ratio. The user input is input, for example, to an input screen displayed on the above-described display device 106. The operation on the input screen is realized, for example, using the above-described input device 108.

[0163] More specifically, first, the prediction unit 254 acquires the molecular structure information 223D of each raw material (see FIG. 14) input by the user. The raw materials input at this time may be the same as or different from the raw materials used during learning. Then, the prediction unit 254 quantifies the molecular structure information 223D of each raw material in the same manner as the above-described generation unit 250. Next, the prediction unit 254 sums up the quantified molecular structure information 223D according to the mixing ratio input by the user, and generates the summation result as the explanatory variable 225A.

[0164] After that, the prediction unit 254 inputs the generated explanatory variable 225A into the learning model 226. As a result, the learning model 226 outputs the performance of the material as a prediction result. As described above, the user can evaluate the performance of various materials by inputting various combinations of raw materials and various mixing ratios.

[0165] (J4. Output unit 256) Next, the function of the output unit 256 shown in FIG. 15 will be described.

[0166] The output unit 256 outputs the prediction result by the prediction unit 254. The output destination of the prediction result is arbitrary. As an example, the prediction result is displayed on the display device 106 of the information processing apparatus 100. As another example, the prediction result is stored as data.

[0167] <K. Flowchart> As described above, the information processing apparatus 100 according to the second embodiment not only has the function of generating the learning data set 124 from the material data set 122 in which the raw material structure is not defined, but also has the function of generating the learning data set 224 from the material data set 222 in which the raw material structure is defined. In this way, the information processing apparatus 100 changes the method of generating the learning data set according to whether the raw material structure is known or unknown.

[0168] The flowchart of the process of generating the learning dataset 124 from the material dataset 122 is as described with reference to FIG. 11. Therefore, hereinafter, only the flowchart of the process of generating the learning dataset 224 from the material dataset 222 will be described with reference to FIG. 17. FIG. 17 is a flowchart showing the flow of the generation process of the learning dataset 224.

[0169] The control device 101 of the information processing apparatus 100 functions as the above-described generation unit 250 (see FIG. 15) by executing the above-described generation program 128 (see FIG. 4) and executes the process shown in FIG. 17. In other aspects, part or all of the process shown in FIG. 17 may be executed by a circuit element or other hardware.

[0170] In step S210, the control device 101 acquires the above-described material dataset 222 (see FIG. 14). The acquisition source of the material dataset 222 is arbitrary. As an example, the acquisition source of the material dataset 222 may be the auxiliary storage device 220 of the information processing apparatus 100 or an external device.

[0171] In step S212, the control device 101 initializes the variable "j". As an example, the variable "j" is initialized to "1".

[0172] In step S214, the control device 101 acquires the j-th material data 223 included in the above-described material dataset 222.

[0173] In step S216, the control device 101 acquires the molecular structure information 223D for the raw material 223B defined in the material data 223 acquired in step S214. Then, the control device 101 unifies the acquired molecular structure information 223D in a predetermined notation. The notation is, for example, the SMILES notation or the graph notation.

[0174] In step S218, the control device 101 quantifies the molecular structure information 223D unified in a predetermined notation in step S216. Since the method of this quantification is as described above, the description thereof will not be repeated.

[0175] In step S220, the control device 101 sums up the molecular structure information 223D quantified in step S218 according to the blending ratio 223C defined in the material data 223 acquired in step S214.

[0176] In step S222, the control device 101 associates the summation result in step S220 with the performance 223E defined in the material data 223 acquired in step S214, using the summation result as the explanatory variable 225A. Thereby, the j-th learning data 225 associating the explanatory variable 225A with the performance 223E is generated.

[0177] In step S230, the control device 101 determines whether to end the generation process of the learning data 225. As an example, the control device 101 determines to end the generation process when the variable "j" is greater than a predetermined value. Alternatively, the control device 101 determines to end the generation process when all the material data 223 included in the material data set 222 have been used in the generation process of the learning data 225. When the control device 101 determines to end the generation process of the learning data 225 (YES in step S230), it ends the process shown in FIG. 11. Otherwise (NO in step S230), the control device 101 switches the control to step S232.

[0178] In step S232, the control device 101 increments the variable "j". That is, the control device 101 increases the variable "j" by 1. Then, the control device 101 returns the control to step S214.

[0179] <L. Others> Next, another example of the above embodiment will be described.

[0180] In the example of FIG. 1 described above, in the material data 123, an example was described in which, in addition to the material ID 123A, the raw material 123B, the blending ratio 123C, the analysis information 123D, and the performance 123E are associated with each other. However, even more information may be added. The additional information includes, for example, the adhesive thickness and the substrate thickness. Note that the additional information may further include information on additives such as particles and fillers. Further, the additional information may further include conditions for measuring each adhesion performance (for example, information on an adherend used when measuring adhesive strength).

[0181] In this case, the information processing apparatus 100 directly reflects the additional information defined in the material data 123 as another explanatory variable different from the explanatory variable 125A in the learning data 125.

[0182] <M. Others> Next, still another example of the above-described embodiment will be described.

[0183] In the example of FIG. 14 described above, in the material data 223, an example was described in which, in addition to the material ID 223A, the raw material 223B, the blending ratio 223C, the molecular structure information 223D, and the performance 223E are associated with each other. However, even more information may be added. The additional information includes, for example, the adhesive thickness and the substrate thickness. Note that the additional information may further include information on additives such as particles and fillers. Further, the additional information may further include conditions for measuring each adhesion performance (for example, information on an adherend used when measuring adhesive strength).

[0184] In this case, the information processing apparatus 100 directly reflects the additional information defined in the material data 223 as another explanatory variable different from the explanatory variable 225A in the learning data 225.

[0185] The embodiments disclosed this time should be considered to be illustrative in all respects and not restrictive. The scope of the present invention is shown not by the above description but by the claims, and it is intended that all modifications within the meaning and scope equivalent to the claims are included.

Explanation of Symbols

[0186] 10 Adhesive material, 10A Single-sided adhesive tape, 12 Core base material, 14 Adhesive, 16 Release agent, 100 Information processing device, 101 Control device, 102 ROM, 103 RAM, 104 Communication interface, 105 Display interface, 106 Display device, 107 Input interface, 108 Input device, 115 Bus, 120 Auxiliary storage device, 122 Material data set, 123 Material data, 123A Material ID, 123B Raw material, 123C Blending ratio, 123D Analysis information, 123D1 IR data, 123D2 GPC data, 123E Performance, 124 Learning data set, 125 Learning data, 125A Explanatory variable, 125A1 Explanatory variable, 125A2 Explanatory variable, 125B Objective variable, 126 Learning model, 128 Generation program, 130 Learning program, 150 Generation unit, 152 Learning unit, 154 Prediction unit, 156 Output unit, 220 Auxiliary storage device, 222 Material data set, 223 Material data, 223A Material ID, 223B Raw material, 223C Blending ratio, 223D Molecular structure information, 223E Performance, 224 Learning data set, 225 Learning data, 225A Explanatory variable, 225B Objective variable, 226 Learning model, 250 Generation unit, 252 Learning unit, 254 Prediction unit, 256 Output unit.

Claims

1. A method for generating a learning dataset, which is executed by an information processing apparatus, comprising: obtaining a plurality of first source data used for generating the learning dataset; each of the plurality of first source data associates a blending ratio of raw materials constituting a material, analysis information indicating characteristics of the various raw materials, and a first performance related to the material; the generation method further includes: for each of the plurality of first source data, generating a first explanatory variable using the blending ratio defined in the first source data and the analysis information related to the raw materials defined in the first source data; generating a first learning dataset by associating the first performance defined in the first source data with the first explanatory variable generated for each of the plurality of first source data as an objective variable; the step of generating the first explanatory variable includes compressing the analysis information related to each raw material defined in each of the plurality of first source data into a predetermined dimension, summing up the analysis information of each dimension according to the blending ratio defined in the first source data for each dimension, and calculating the sum result for each dimension as the first explanatory variable, the generation method.

2. the generation method further includes obtaining a plurality of second source data used for generating the learning dataset; each of the plurality of second source data associates a blending ratio of raw materials constituting a material, structure information of the various raw materials, and a second performance related to the material; the generation method further includes: for each of the plurality of second source data, generating a second explanatory variable using the blending ratio defined in the second source data and the structure information defined in the second source data; generating a second learning dataset by associating the second performance defined in the second source data with the second explanatory variable generated for each of the plurality of second source data as an objective variable, the generation method according to claim 1.

3. the step of generating the second explanatory variable includes converting the structure information of each raw material defined in each of the plurality of second source data into a common expression form, summing up the structure information converted into the common form according to the blending ratio defined in the second source data, and calculating the sum result as the second explanatory variable, the generation method according to claim 2.

4. The generation method according to claim 3, wherein the representation form is a form representing the structural information by molecular descriptors, a form representing the structural information by a fingerprint, or a form representing the structure of the raw material by a feature vector.

5. The analysis information is data obtained by analyzing the raw material by a spectroscopic method, The spectroscopic method includes infrared spectroscopy, visible spectroscopy, ultraviolet spectroscopy, Raman spectroscopy, electron spin resonance method, X-ray analysis method, or nuclear magnetic resonance spectroscopy, and the generation method according to claim 1 or 2.

6. The generation method according to claim 1 or 2, wherein the analysis information is data obtained by analyzing the raw material by a chromatography method.

7. The generation method according to claim 1 or 2, wherein the material is an adhesive material.

8. Each of the first performance and the second performance includes at least one of a characteristic indicating the strength required for peeling the adhesive material and the adherend, a characteristic indicating the difficulty of displacement of the adhesive material with respect to the adherend, and a characteristic indicating the ease of adhesion between the adhesive material and the adherend, and the generation method according to claim 2.

9. The generation method further includes generating a first learning model by machine learning using the first learning dataset; and generating a second learning model by machine learning using the second learning dataset, and the generation method according to claim 2.

10. An information processing apparatus, comprising a control unit for controlling the information processing apparatus, the control unit executes a process of acquiring a plurality of first source data used for generating a learning dataset, each of the plurality of first source data associates a blending ratio of raw materials constituting the material, analysis information indicating characteristics of the various raw materials, and first performance related to the material, the control unit further generates a first explanatory variable for each of the plurality of first source data using the blending ratio defined in the first source data and the analysis information related to the raw material defined in the first source data; and generates a first learning dataset by associating the first performance defined in the first source data with respect to the first explanatory variable generated for each of the plurality of first source data as a target variable, and executes the process. The process of generating the first explanatory variable includes compressing the analysis information related to each raw material defined in each of the plurality of first source data into a predetermined dimension, summing up the analysis information of each dimension separately according to the blending ratio defined in the first source data, and calculating the sum result of each dimension as the first explanatory variable. An information processing apparatus.

11. A program for generating a learning data set, The generation program causes an information processing apparatus to execute a process of acquiring a plurality of first source data used for generating a learning data set, Each of the plurality of first source data associates the blending ratio of the raw materials constituting the material, the analysis information indicating the characteristics of the various raw materials, and the first performance related to the material. The generation program further causes the information processing apparatus to For each of the plurality of first source data, a process of generating a first explanatory variable using the blending ratio defined in the first source data and the analysis information related to the raw materials defined in the first source data, For each of the first explanatory variables generated for each of the plurality of first source data, a process of generating a first learning data set by associating the first performance defined in the first source data as a target variable. The process of generating the first explanatory variable includes compressing the analysis information related to each raw material defined in each of the plurality of first source data into a predetermined dimension, summing up the analysis information of each dimension separately according to the blending ratio defined in the first source data, and calculating the sum result of each dimension as the first explanatory variable. A generation program.

Images