Information processing system, information processing method, program, and recording medium

Abstract

To meet the need for an information processing system that makes it possible to properly identify an important explanatory variable and estimate material properties with high accuracy in a material development field, even when the acquirable number of data for learning is few.SOLUTION: Provided is an information processing system characterized by having an information processing unit. The information processing unit acquires a descriptor pertaining to a material, converts a structure descriptor included in the descriptor to a correlation-reduced descriptor in which a correlation is reduced by using a conversion coefficient, calculates the importance of the correlation-reduced descriptor, identifies an important descriptor from within the descriptor on the basis of the conversion coefficient used in conversion to the correlation-reduced descriptor of higher importance, creates a trained model by machine learning in which the important descriptor and the characteristic value of the material are used for training data, and estimates the characteristic value of a material different from the material by using the trained model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an information processing system used to estimate material properties of a resultant product from material recipes and processing conditions. [Background technology]

[0002] Materials informatics (MI) is a well-known technology that uses machine learning to generate a trained model (property estimation model) based on data on prototype conditions, such as material recipes and processing conditions, and data on the material properties (evaluation values) of the resulting product, and then uses the trained model to estimate the properties of untested materials. In material development using MI, the property estimation model is used to estimate the material properties obtained under untested prototype conditions. The results are used to identify the prototype conditions that are estimated to result in the desired material properties. Then, the material is actually prototyped under the identified prototype conditions to confirm whether the desired material properties were obtained. In this case, if the accuracy of the property estimation model's estimation is poor, the accuracy of the prototype conditions identified will also be poor, making it difficult to develop materials efficiently.

[0003] The training data used to create a physical property estimation model consists of explanatory variables and response variables. In materials development, explanatory variables represent prototype conditions, such as the amount and characteristics of raw materials used and processing conditions, while response variables indicate evaluation values, such as the characteristics of the developed material. To improve the accuracy of the estimation model, explanatory variables sometimes include descriptors, which are values ​​converted according to molecular structure and characteristics. Using calculations and databases, it is possible to create hundreds to thousands of descriptors from a single molecule's information, which increases the number of explanatory variables accordingly. However, when attempting to obtain training data, which are sets of explanatory variables and response variables, through experiments, there is a limit to the amount of data that can be obtained due to practical constraints. In such cases, the training data will have many explanatory variables and a small amount of data, making it difficult to improve the accuracy of the physical property estimation model.

[0004] In order to improve the accuracy of a physical property estimation model, explanatory variables are selected, but among the many explanatory variables, it is necessary to keep those that are relevant to the target variable and remove those that are not. If the mechanism of the phenomenon is known, it is possible to select only the relevant descriptors, but if such knowledge is not available, selection is made by analyzing the data.

[0005] In Patent Document 1, the importance of explanatory variables is estimated from data using permutation importance or the like, and explanatory variables with high importance are selected. In Patent Document 2, explanatory variables are transformed using principal component analysis. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Patent Publication No. 2021-174403 [Patent Document 2] Japanese Patent Publication No. 2022-108269 Summary of the Invention [Problem to be solved by the invention]

[0007] However, the method described in Patent Document 1 uses permutation importance, but when there is a correlation between explanatory variables, it is not possible to correctly evaluate the importance, and therefore it is not possible to make an appropriate selection.

[0008] Furthermore, in the method described in Patent Document 2, while it is possible to reduce the correlation between explanatory variables by principal component analysis, it is not possible to select important explanatory variables because the principal components are values ​​that have no physical meaning. As a result, in material development where the amount of data that can be obtained is small, it is not possible to obtain a highly accurate estimation model, and efficient material development is not possible.

[0009] Therefore, in the field of materials development, there is a need for an information processing system that can appropriately identify important explanatory variables and estimate material properties with high accuracy, even if the amount of learning data that can be obtained is small. [Means for solving the problem]

[0010] A first aspect of the present invention is an information processing method, characterized in that an information processing unit acquires descriptors related to a material, converts the acquired descriptors into reduced-correlation descriptors that reduce correlation using a conversion coefficient, calculates the importance of the reduced-correlation descriptors, identifies important descriptors from among the descriptors based on the conversion coefficients used for conversion to the reduced-correlation descriptors with high importance, creates a trained model by machine learning using the important descriptors and characteristic values ​​of the material as training data, and uses the trained model to estimate characteristic values ​​of a material different from the material.

[0011] Furthermore, a second aspect of the present invention is an information processing system comprising an information processing unit, which acquires descriptors related to a material, converts the acquired descriptors into reduced-correlation descriptors that have reduced correlation using a conversion coefficient, calculates the importance of the reduced-correlation descriptors, identifies important descriptors from among the descriptors based on the conversion coefficients used for conversion into the reduced-correlation descriptors with high importance, creates a trained model by machine learning using the important descriptors and characteristic values ​​of the material as training data, and uses the trained model to estimate characteristic values ​​of a material different from the material. [Effects of the Invention]

[0012] According to the present invention, it is possible to realize an information processing system that can appropriately identify important explanatory variables and estimate material properties with high accuracy even when the amount of learning data that can be acquired is small. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 2 is a schematic functional block diagram showing the configuration of functional blocks included in the information processing system according to the embodiment. [Figure 2] FIG. 4 is a diagram showing an example of learning data acquired in the processing of step S1. [Figure 3] FIG. 10 is a diagram showing an example of data obtained by the processing in step S2. [Figure 4] FIG. 10 is a diagram showing an example of data obtained by the processing in step S3. [Figure 5] FIG. 10 is a diagram illustrating an example of the importance of each correlation-reducing component obtained by the processing in step S4. [Figure 6] FIG. 10 is a diagram showing an example of conversion coefficients acquired in the process of step S5. [Figure 7] FIG. 10 is a diagram showing an example of learning data composed of important explanatory variables and a response variable extracted in the processing of step S6. [Figure 8] FIG. 10 is a diagram illustrating the relationship between explanatory variables and weighted averages of conversion coefficients obtained in the second embodiment. [Figure 9] 1 is a flowchart showing an information processing procedure according to an embodiment. [Figure 10] FIG. 1 is a schematic diagram showing an example of a hardware configuration of an information processing system according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0014] An information processing system according to an embodiment of the present invention will be described with reference to the drawings. Note that the embodiment shown below is an example, and those skilled in the art can appropriately modify and implement the detailed configuration, for example, without departing from the spirit of the present invention.

[0015] In the drawings referred to in the following description of the embodiments and examples, elements denoted by the same reference numerals have the same functions unless otherwise specified. When multiple identical elements are arranged in a drawing, the reference numerals and their descriptions may be omitted. Furthermore, since the drawings may be represented schematically for the convenience of illustration and explanation, the shape, size, arrangement, etc. of elements depicted in the drawings may not strictly correspond to the actual objects.

[0016] [Embodiment 1] (Function block configuration) FIG. 1 is a schematic functional block diagram showing the configuration of functional blocks included in an information processing system according to an embodiment. While FIG. 1 uses functional blocks to represent functional elements necessary for explaining the features of this embodiment, general functional elements that are not directly related to the problem-solving principle of the present invention are omitted. Furthermore, the functional elements shown in FIG. 1 are conceptual functional elements and do not necessarily have to be physically configured as shown. For example, the specific form of distribution and integration of the functional blocks is not limited to the illustrated example, and all or part of them can be functionally or physically distributed and integrated in any unit depending on the usage situation, etc.

[0017] Each functional block can be configured using hardware or software. These functional blocks can be configured by a CPU reading and executing a control program stored in, for example, a storage device or a non-transitory recording medium. Alternatively, some or all of the functional blocks can be configured using hardware such as an ASIC included in the information processing system.

[0018] The information processing unit 103 of the information processing system according to the embodiment includes a descriptor acquisition unit 111, an explanatory variable conversion unit 112, an importance calculation unit 113, an important explanatory variable selection unit 114, an important explanatory variable acquisition unit 115, a model learning unit 116, and a physical property estimation unit 117. The functions of each unit will be described later in relation to the procedure of the material property estimation process. The information processing unit 103 can acquire training data from a database 101 that stores training data 121, acquire estimation conditions 124 from the outside via a network interface 1607, and output estimated physical property values ​​to the outside via the network interface 1607.

[0019] (Hardware configuration) An example of the hardware configuration of an information processing system according to this embodiment is shown in schematic form in Fig. 10. The information processing system according to this embodiment may be a physical computer system, or may be a system built on a group of computing resources such as a cloud platform.

[0020] As shown in Fig. 2, the information processing system may include PC hardware equipped with a CPU 1601 as a main control means, a ROM 1602 as a storage device, and a RAM 1603. The ROM 1602 may store information such as a processing program for implementing an information processing method described below. The RAM 1603 is used as a work area for the CPU 1601 when executing the information processing method. An external storage device 1606 is also connected to the PC hardware. The external storage device 1606 may be an HDD, an SSD, or an external storage device of another system mounted via a network.

[0021] The processing program of the CPU 1601 for realizing the information processing device or information processing method according to the embodiment can be stored in a storage unit such as an external storage device 1606 including an HDD or SSD, or (for example, an EEPROM area) of the ROM 1602. In this case, the processing program of the CPU 1601 for realizing the information processing method (method for estimating material properties) can be supplied to each of the above storage units via the network interface 1607 and updated with a new program. Alternatively, the processing program of the CPU 1601 for realizing the information processing method can be supplied to each of the above storage units via various storage means such as magnetic disks, optical disks, and flash memories and their associated drive devices and updated. Various storage means, storage units, or storage devices storing a program capable of executing the processing of the CPU 1601 for realizing the information processing method are computer-readable recording media related to the information processing method or information processing device of the present invention.

[0022] The network interface 1607 can be configured using a communication standard for wired communication such as IEEE 802.3 or wireless communication such as IEEE 802.11 or IEEE 802.15. The CPU 1601 can communicate with the external devices 1104 and 1121 on the cloud 1608 via the network interface 1607. For example, the external devices 1104 and 1121 may be a supervising control device such as a PLC or sequencer arranged to control and manage the material prototyping device or material analysis device, or a management server.

[0023] 10, an operation unit 1604 as an input unit and a display device 1605 as an output unit are connected as a UI device (user interface device). The operation unit 1604 can be configured by a terminal such as a handy terminal, or devices such as a keyboard, a jog dial, a mouse, a pointing device, or a voice input device (or a control terminal equipped with these). The display device 1605 may be any device that can display information related to processing executed by the important explanatory variable selection unit 114, the important explanatory variable acquisition unit 115, the physical property estimation unit 117, etc., which will be described later with reference to FIG. 1, on a display screen, and for example, a liquid crystal display device can be used.

[0024] The hardware configuration of the information processing system according to the embodiment is not limited to the example of FIG. 2, and may include a processor (e.g., a graphics processing unit) suitable for generating a trained model by machine learning.

[0025] (Information Processing Procedures) The information processing procedure for estimating material properties will be described in relation to the functions of each unit of the information processing unit 103 (FIG. 1). FIG. 9 shows a flowchart illustrating the information processing procedure.

[0026] When the process starts, in step S1, the information processing unit 103 acquires the learning data 121 stored in the database 101. The learning data may be read from an internal or external database 101, or may be input from the outside via the network interface 1607.

[0027] Referring to the example shown in FIG. 2, the learning data 121 will be described. The learning data includes a set of explanatory variables and target variables for the number of trial productions. The explanatory variables include information such as the usage amount of raw materials, molecular structure, physical property values, and processing conditions. The molecular structure is preferably in the SMILES format for easy handling in calculations, but is not limited to this format as long as the molecular structure is accurately represented. The target variable includes the physical property values of the material obtained in the trial production, that is, the information of the evaluation values. The evaluation value may be one or more.

[0028] Next, in step S2, the descriptor acquisition unit 111 of the information processing unit 103 acquires descriptors from the information on the molecular structure included in the learning data acquired in step S1, and replaces the molecular structure item of the learning data with the descriptors (for example, structure descriptors).

[0029] FIG. 3 illustrates the data obtained by this process. Here, N is the number of data, and M is the number of descriptors. By this process, the explanatory variables increase, and M becomes several hundred to several thousand. At this time, when the relationship between the number of data N and the number of explanatory variables P (in the example of FIG. 3, P = 2M + 3) is such that N < P, the effects of implementing the present invention are particularly likely to be obtained. Also, even if N ≧ P, it is possible to implement the present invention.

[0030] Descriptors may be calculated from molecular structures or obtained by referencing a database containing molecular information and descriptors. When obtaining descriptors by calculation, open source libraries such as RDKit and Mordred may be used. Descriptors to be obtained are preferably those related to the evaluation value of the objective variable, but all obtainable descriptors may be used. In this case, the user may select any descriptor using an input device (not shown). The explanatory variables including the descriptors obtained here include those that are correlated with each other, and the importance of the explanatory variables cannot be correctly evaluated as is.

[0031] Therefore, in step S3, the explanatory variable transformation unit 112 of the information processing unit 103 performs a process of reducing the correlation. That is, the explanatory variables are replaced with components in which the correlation between the explanatory variables is reduced, that is, correlation-reduced components.

[0032] FIG. 4 shows an example of the learning data obtained by this process. It is desirable to standardize each explanatory variable of the learning data obtained by the descriptor acquisition unit 111, that is, to scale it so that the mean is 0 and the standard deviation is 1. The explanatory variables that have undergone this process are treated as standardized explanatory variables. This makes it possible to appropriately select important explanatory variables (important descriptors) even if the numerical ranges of the explanatory variables differ between explanatory variables. Note that if the numerical ranges of the explanatory variables are similar between explanatory variables, standardization does not need to be performed. Even if standardization is not performed, the explanatory variables are treated as standardized explanatory variables from this step onwards.

[0033] Next, the standardized explanatory variables are used to perform principal component analysis (PCA) to find eigenvectors, which are used as conversion coefficients A. Singular value decomposition (SVD) may be used instead of principal component analysis. Next, the conversion coefficients A are used to perform the conversion of the following formula (1).

[0034]

number

[0035] Here, E, A, and D are the matrix of correlation reduction components, the transformation matrix, and the matrix of standardized explanatory variables, respectively. Equation (1) can be decomposed into its elements and written as the following equation (2).

[0036]

number

[0037] where E i is the i-th correlation reduction component, D j is the jth standardized explanatory variable, A ij is E i D to calculate j is the coefficient of. By performing the transformation of Equation (2) on one prototype data, i.e., one row in Figure 3, the corresponding row in Figure 4 is obtained. Therefore, this transformation process is performed sequentially on all prototypes. By performing this transformation, the correlation-reducing components become orthogonal to each other, and a correlation-reducing descriptor, which is an explanatory variable with reduced correlation, is obtained.

[0038] Next, in step S4, the importance calculation unit 113 of the information processing unit 103 acquires the importance of the explanatory variables. Specifically, the importance is calculated by analyzing an estimation model created using the learning data in which the explanatory variables have been replaced with correlation-reduced descriptors by the explanatory variable conversion unit 112.

[0039] Figure 5 shows an example of the importance of each correlation-reducing component calculated by this process. It is desirable to use permutation importance to calculate the importance. Permutation importance is a method of evaluating the performance of an estimation model (trained model) created using training data, evaluating the performance of an estimation model created using data from the training data in which certain explanatory variables have been shuffled, and calculating the importance of the explanatory variables by evaluating the amount of performance degradation. Machine learning algorithms that can be used in this case include ridge regression, support vector regression, random forest, neural network, and gradient boosting regression tree. Alternatively, other machine learning algorithms that can express the relationship between explanatory variables and target variables may be used. Alternatively, estimation models (trained models) may be created using multiple algorithms, and the performance of the estimation models may be evaluated using methods such as tolerance verification, and the algorithm with the best performance may be adopted. After evaluating the performance of the estimation model, the user may select the appropriate one by referring to the evaluation results. When calculating the importance using a decision tree algorithm, feature importance based on the decision tree may be obtained. When calculating the importance using a Lasso regression algorithm, the coefficient of a linear combination may be used as the importance.

[0040] Next, in step S5, the important explanatory variable selection unit 114 of the information processing unit 103 selects important explanatory variables from the explanatory variables as important descriptors. The processing in step S4 obtains the importance of each correlation-reducing component, but each correlation-reducing component does not have a physical meaning. Therefore, it is not possible to select important explanatory variables as they are. Meanwhile, the relationship between the explanatory variables and the correlation-reducing components is described in the conversion coefficients used in the processing by the explanatory variable conversion unit 112. Therefore, by examining the conversion coefficients 122 related to the important correlation-reducing components, it is possible to convert the correlation-reducing components back into explanatory variables, and important explanatory variables (important descriptors) can be identified.

[0041] Using the above-mentioned method, the important explanatory variable selection unit 114 selects important explanatory variables. Specifically, first, from the processing result of step S4, the correlation reducing component E M Next, the most important correlation reducing component E among the conversions performed by the explanatory variable conversion unit 112 is selected. M The conversion factor A used to obtain M1 A MP Since the conversion coefficients can be positive or negative, the absolute values ​​are taken to compare the magnitudes.

[0042] Figure 6 shows the obtained conversion coefficient |A M1 |···|A MP An example of | is shown below. The explanatory variables corresponding to the conversion coefficients with large absolute values ​​are selected as important explanatory variables 123. Note that the conversion coefficient A M1 A MP are explanatory variables D1 D P Since the coefficients are , the two correspond.

[0043] As shown in Figure 6, there is usually more than one important explanatory variable (and its corresponding conversion coefficient). M1 |···|A MP Four methods for selecting the largest transform coefficient from | are described below.

[0044] (1) The number of conversion coefficients to be selected (a predetermined number) may be determined in advance, and descriptors may be selected in descending order of value. This allows for an appropriate number of explanatory variables to be selected in consideration of the number of data to be used, resulting in a highly accurate estimation model. Typically, it is desirable for the number of explanatory variables to be selected to be approximately 0.1N to 0.5N relative to the number of data N, but in some cases it may be acceptable to deviate from this value.

[0045] (2) Also, |A M1 |···|A MPAlternatively, the maximum value of | can be normalized to 1 and those that exceed a predetermined threshold can be selected. In method (1), if the number of truly important explanatory variables is less than the number of explanatory variables set, unnecessary explanatory variables will also be selected, resulting in a deterioration in the accuracy of the estimated model. Method (2) can avoid this.

[0046] (3) It is also desirable to combine methods (1) and (2). In other words, the number of conversion coefficients to be selected is determined in advance, and |A M1 |···|A MP The maximum value of | is normalized to 1, a threshold is also determined, and only explanatory variables that meet both conditions are selected. Method (1) has the problem mentioned above, and method (2) may select too many explanatory variables compared to the amount of data, making it difficult to improve the accuracy of the estimated model. Method (3) avoids both problems and allows for the creation of a highly accurate estimated model.

[0047] (4) As shown in Figure 6, |A M1 |···|A MP The value of | may be displayed on the display device 1605, and the user may manually select explanatory variables. This allows for flexible response, although the results obtained will depend on the user.

[0048] Next, in step S6, the important explanatory variable acquisition unit 115 of the information processing unit 103 extracts only explanatory variable data (material characteristic values) that match the selected important explanatory variables 123 from the post-descriptor acquisition learning data 127 output from the descriptor acquisition unit 111. Fig. 7 shows an example of learning data consisting of the extracted important explanatory variables and objective variables.

[0049] Next, in step S7, the model learning unit 116 of the information processing unit 103 creates an estimation model 125 (trained model) using learning data consisting of important explanatory variables and a target variable. A machine learning algorithm is used to create the estimation model 125. As with the calculation of permutation importance, algorithms such as ridge regression, support vector regression, random forest, neural network, and gradient boosting regression tree can be used. Alternatively, other machine learning algorithms capable of expressing the relationship between explanatory variables and a target variable may be used. Alternatively, estimation models may be created using multiple algorithms, and the performance of the estimation models may be evaluated using a method such as tolerance verification, and the algorithm with the best performance may be adopted. Alternatively, after evaluating the performance of the estimation models, the user may select a model by referring to the evaluation results. The created estimation model 125 is configured in the physical property estimation unit 117 as an executable trained model.

[0050] Next, in step S8, the estimation conditions 124 are input by the user to the physical property estimation unit 117 via the network interface 1607 or an input unit (not shown). The estimation conditions 124 are made up of the same explanatory variables as the training data 121, and do not require a target variable. In this case, there may be multiple conditions, not just one. Note that step S8 may be performed following step S1, allowing the user to perform the input work all at once. Next, in step S9, the descriptor acquisition unit 111 acquires a descriptor from the molecular structure information of the estimation conditions and replaces it with a structural descriptor.

[0051] Next, in step S10, the important explanatory variable acquisition unit 115 extracts those that match the important explanatory variables from the estimation conditions. Note that, in order to simplify the process, instead of steps S9 and S10, the descriptor acquisition unit 111 may acquire only descriptors that match the important explanatory variables.

[0052] Next, in step S11, when estimation conditions in which only important explanatory variables are extracted are input, estimated physical property values ​​126 (characteristic values ​​of a material different from the material used in the training data) are output by the estimation model 125 configured in the physical property estimation unit 117. Note that although the estimation conditions are the same as the explanatory variables of the training data 121, estimation conditions 128 in which only important explanatory variables are extracted may also be input directly after descriptor acquisition. Furthermore, although estimation conditions may be input directly, when multiple conditions are obtained, generation rules may be set and multiple conditions may be automatically generated according to the rules.

[0053] Next, in step S12, the calculated result, that is, the estimated physical property value 126, is displayed on the display device 1605 (FIG. 10). The estimated physical property value 126 may be output to the outside via the network interface 1607.

[0054] In this embodiment, by obtaining estimated physical property values ​​using only the important explanatory variables in this manner, it is possible to obtain an information processing system (material property estimation model) that can estimate physical property values ​​with high accuracy even if there is a small amount of learning data.

[0055] [Embodiment 2] An information processing system according to embodiment 2 will be described. Explanations of matters common to embodiment 1 will be simplified or omitted.

[0056] In the first embodiment, the important explanatory variable selection unit 114 selects the correlation reducing component E with the highest importance from the result of the importance calculation unit 113. M However, information for selecting important explanatory variables included in correlation-reducing components with second or lower importance is not used. Embodiment 2 differs from Embodiment 1 in that this information is also utilized to more appropriately select important explanatory variables. Two aspects of this embodiment will be described below.

[0057] (1) The important explanatory variable selection unit 114 selects the correlation-reducing components E1...E output from the importance calculation unit 113. NA predetermined number T of the correlation reducing components E are selected in descending order of importance. S1 ···E ST Absolute value of the conversion factor |A S11 |···|A S1P |, , |A ST1 |···|A STP | and select explanatory variables with large values, as in the first embodiment. At this time, since there is a possibility that explanatory variables may be selected more than once, the duplicates are removed to obtain important explanatory variables. Furthermore, duplicates may be considered more important, and the explanatory variables may be further narrowed down based on the number of times they have been selected.

[0058] (2) The important explanatory variable selection unit 114 uses the importance of each correlation-reducing component output from the importance calculation unit 113 as a weight to calculate the conversion coefficient A obtained by the explanatory variable conversion unit 112. ij Select important explanatory variables (important descriptors) based on the weighted average of the absolute values ​​of each explanatory variable D i The weighted average value S corresponding to i The calculation is carried out using the following formula (3).

[0059]

number

[0060] Figure 8 illustrates the results obtained from this calculation. i Explanatory variable D that increases i The selection method is the same as that of the conversion coefficient |A M1 |···|A MP This is the same as the method of selecting a conversion coefficient with a large value from |.

[0061] In the second embodiment, the explanatory variables are selected more appropriately, so that an information processing system (material property estimation model) can be obtained that can estimate property values ​​with higher accuracy even if there is a small amount of learning data.

[0062] [Example] An example will be described in which the information processing system according to the second embodiment is applied to the development of a material that achieves a high elastic modulus by mixing bismaleimide with an aromatic amine and curing the mixture.

[0063] First, the types of bismaleimide and aromatic amine materials and their mixing amounts were changed, and the elastic modulus was evaluated to create training data consisting of prototype conditions and evaluation values. One type of bismaleimide material (BMI-70 manufactured by K.I. Chemical Co., Ltd.) was used. Four types of aromatic amine materials (WANAMINE MDA-100H, MDA-100, Ethacure 100 Plus, Ethacure 300, and 4,4'-diaminodiphenyl sulfone manufactured by Mitsui Chemicals Fine Co., Ltd.) were used. The number of data points was 25. The explanatory variables for the training data were the ratio of the amount of aromatic amine to the amount of bismaleimide, and the molecular structure of each material with respect to the amount of aromatic amine, expressed in SMILES format.

[0064] The file containing this learning data was input into the information processing system according to embodiment 2 to obtain a material property estimation model (trained model). The following processing was performed inside the information processing system.

[0065] RDKit was used to obtain 400 descriptors related to the molecular structure from the SMILES format molecular structure information of each material. As a result, the explanatory variables were the ratio of the amount of aromatic amine to the amount of bismaleimide, and 401 descriptors for aromatic amine. Principal component analysis was performed on this data to obtain correlation-reducing components. At this time, the number of explanatory variables was 25.

[0066] For this data, material property estimation models were created using several machine learning algorithms, and tolerance verification was performed, with the coefficient of determination used as an indicator to select the material property estimation model. As a result, the method using LightGBM with gradient boosting was found to be the most accurate, with a coefficient of determination of 0.81. When Permutation Importance was calculated using this material property estimation model, the three most important correlation-reducing components were found to be principal component 3, principal component 15, and principal component 5.

[0067] The conversion coefficients for each relevant principal component were referenced, and three explanatory variables with the largest absolute values ​​were selected for each, for a total of nine. After removing duplicates, five explanatory variables were selected: the ratio of aromatic amine to bismaleimide, aromatic amine MolWt, aromatic amine TPSA, aromatic amine BertzCT, and aromatic amine SMR_VSA7.

[0068] The explanatory variables mentioned above were extracted from the training data after descriptor conversion, and estimation models were created using several machine learning algorithms. The model with the highest accuracy was selected using the coefficient of determination (CR) obtained through tolerance validation. The coefficient of determination for this estimation model was 0.92, indicating that a highly accurate estimation model was obtained.

[0069] The estimation conditions were six types of aromatic amine to bismaleimide ratios (0.05, 0.1, 0.15, 0.2, 0.25, and 0.3), and three types of aromatic amines (Bisaniline-M, Bisaniline-P, and Ethacure® 100 Plus, manufactured by Mitsui Fine Chemicals, Inc.) were combined to create 18 conditions. The data format was the same as the explanatory variables of the training data. A file containing these estimation conditions was entered into an information processing system, and the aforementioned material property estimation model was used to obtain estimated values ​​for the elastic modulus under each condition. An actual prototype was produced under the prototype condition with the highest elastic modulus, and the estimated value was compared with the measured elastic modulus. The error in the estimated value was 6.4%.

[0070] For comparison, the training data described above was converted into descriptors, and the training data with 401 explanatory variables was used as is to create a material property estimation model. Of the estimation conditions described above, the prototype conditions that yielded the highest estimated elastic modulus using the information processing system according to the embodiment were applied to the estimation model created here, and an estimated value for the elastic modulus was obtained. As a result, the error in the estimation value was 48%, which was significantly lower in accuracy than the information processing system according to the present embodiment.

[0071] In this way, it was shown that by using the information processing system according to the embodiment, it is possible to estimate material properties with sufficiently high accuracy even with a small amount of learning data, without using any special knowledge.

[0072] [Other embodiments] The present invention is not limited to the above-described embodiments and examples, and many modifications are possible within the technical spirit of the present invention. For example, the above-described different embodiments and examples may be combined in whole or in part.

[0073] The present invention can also be realized by supplying a program that realizes one or more functions of the embodiments to a system or device via a network or a storage medium, and having one or more processors in the computer of the system or device read and execute the program. It can also be realized by a circuit (e.g., ASIC) that realizes one or more functions.

[0074] The operation unit 1604 (FIG. 10) as an input unit has a function of accepting information (commands, data, selection results of options, etc.) input by the user that is necessary to execute processing (for example, steps S1 to S12 shown in FIG. 9).

[0075] The display device 1605 (FIG. 10) as a display unit has the function of displaying information related to the processing to be executed (processing content, progress, options, warnings, etc.) to the user when the information processing unit 103 executes each processing (steps S1 to S12) shown in FIG.

[0076] This specification discloses at least the following: [Item 1] The information processing section Obtaining descriptors related to the material; converting the obtained descriptor into a correlation-reduced descriptor using a conversion coefficient; Calculating the importance of the correlation reduction descriptor; Identifying important descriptors from among the descriptors based on the transformation coefficients used for transformation into the correlation-reduced descriptors with higher importance; A trained model is created by machine learning using the important descriptors and the characteristic values ​​of the material as training data; Using the trained model, estimate a property value of a material different from the material. An information processing method comprising: [Matter 2] In identifying the important descriptors, a predetermined number of the descriptors having the largest transform coefficients used for transforming into the correlation reduced descriptor are identified as the important descriptors. 2. The information processing method according to item 1, [Matter 3] In identifying the important descriptor, the descriptor whose transformation coefficient used for transformation into the correlation reduced descriptor exceeds a predetermined threshold is identified as the important descriptor. 2. The information processing method according to item 1, [Matter 4] In identifying the important descriptor, the important descriptor is identified based on a plurality of the transformation coefficients used for transformation into a plurality of the correlation reduced descriptors having high importance. 2. The information processing method according to item 1, [Matter 5] In identifying the important descriptor, the important descriptor is identified based on a result of weighted averaging of the transformation coefficients used in the transformation into the correlation reduced descriptor with weights of importance. 2. The information processing method according to item 1, [Matter 6] displaying information related to the processing executed by the information processing unit on a display unit; 6. The information processing method according to any one of items 1 to 5, [Matter 7] the information processing unit receives information necessary for executing the processing via an input unit; 7. The information processing method according to any one of items 1 to 6, [Matter 8] The obtained descriptors include structural descriptors. 8. The information processing method according to claim 1, wherein the first and second inputs are input to the first and second inputs. [Matter 9] 9. A program for causing the information processing unit to execute the information processing method according to any one of items 1 to 8. [Matter 10] 10. A computer-readable recording medium having the program described in item 9 recorded thereon. [Matter 11] An information processing unit is provided, The information processing unit Obtaining descriptors related to the material; converting the obtained descriptor into a correlation-reduced descriptor using a conversion coefficient; Calculating the importance of the correlation reduction descriptor; Identifying important descriptors from among the descriptors based on the transformation coefficients used for transformation into the correlation-reduced descriptors with higher importance; A trained model is created by machine learning using the important descriptors and the characteristic values ​​of the material as training data; Using the trained model, estimate a property value of a material different from the material. An information processing system comprising: [Matter 12] In identifying the important descriptors, a predetermined number of the descriptors having the largest transform coefficients used for transforming into the correlation reduced descriptor are identified as the important descriptors. 12. The information processing system according to item 11. [Matter 13] In identifying the important descriptor, the descriptor whose transformation coefficient used for transformation into the correlation reduced descriptor exceeds a predetermined threshold is identified as the important descriptor. 12. The information processing system according to item 11. [Matter 14] In identifying the important descriptor, the important descriptor is identified based on a plurality of the transformation coefficients used for transformation into a plurality of the correlation reduced descriptors with high importance. 12. The information processing system according to item 11. [Matter 15] In identifying the important descriptor, the important descriptor is identified based on a result of weighted averaging of the transformation coefficients used in the transformation into the correlation reduced descriptor with weights of importance. 12. The information processing system according to item 11. [Matter 16] a database that records the descriptors associated with the material; 16. The information processing system according to any one of items 11 to 15. [Matter 17] a display unit that displays information related to the processing executed by the information processing unit; 17. The information processing system according to any one of items 11 to 16, [Matter 18] an input unit that receives information necessary for the information processing unit to execute processing; 18. The information processing system according to any one of items 11 to 17. [Matter 19] The obtained descriptors include structural descriptors. 19. The information processing system according to any one of items 11 to 18, [Explanation of symbols]

[0077] 101 Database / 103 Information processing unit / 111 Descriptor acquisition unit / 112 Explanatory variable conversion unit / 113 Importance calculation unit / 114 Important explanatory variable selection unit / 115 Important explanatory variable acquisition unit / 116 Model learning unit / 117 Physical property estimation unit / 121 Learning data / 122 Conversion coefficient / 123 Important explanatory variable / 124 Estimation conditions / 125 Estimation Model / 126···Estimated physical property value / 127···Training data after descriptor acquisition / 128···Estimation condition after descriptor acquisition / 1104···External device / 1121···External device / 1601···CPU / 1602···ROM / 1603···RAM / 1604···Operation unit / 1605···Display device / 1606···External storage device / 1607··Network interface / 1608···Cloud

Claims

1. The information processing section Obtaining descriptors related to the material; converting the obtained descriptor into a correlation-reduced descriptor using a conversion coefficient; Calculating the importance of the correlation reduction descriptor; Identifying important descriptors from among the descriptors based on the transformation coefficients used for transformation into the correlation-reduced descriptors with higher importance; A trained model is created by machine learning using the important descriptors and the characteristic values ​​of the material as training data; Using the trained model, estimate a property value of a material different from the material.

1. An information processing method comprising:

2. In identifying the important descriptors, a predetermined number of the descriptors having the largest transform coefficients used for transforming into the correlation reduced descriptor are identified as the important descriptors.

2. The information processing method according to claim 1,

3. In identifying the important descriptor, the descriptor whose transformation coefficient used for transformation into the correlation reduced descriptor exceeds a predetermined threshold is identified as the important descriptor.

2. The information processing method according to claim 1,

4. In identifying the important descriptor, the important descriptor is identified based on a plurality of the transformation coefficients used for transformation into a plurality of the correlation reduced descriptors having high importance.

2. The information processing method according to claim 1,

5. In identifying the important descriptor, the important descriptor is identified based on a result of weighted averaging of the transformation coefficients used in the transformation into the correlation reduced descriptor with weights of importance.

2. The information processing method according to claim 1,

6. displaying information related to the processing executed by the information processing unit on a display unit; 6. The information processing method according to claim 1, wherein:

7. the information processing unit receives information necessary for executing the processing via an input unit; 6. The information processing method according to claim 1, wherein:

8. The obtained descriptors include structural descriptors.

6. The information processing method according to claim 1, wherein:

9. A program for causing the information processing unit to execute the information processing method according to any one of claims 1 to 5.

10. A computer-readable recording medium on which the program according to claim 9 is recorded.

11. An information processing unit is provided, The information processing unit Obtaining descriptors related to the material; converting the obtained descriptor into a correlation-reduced descriptor using a conversion coefficient; Calculating the importance of the correlation reduction descriptor; Identifying important descriptors from among the descriptors based on the transformation coefficients used for transformation into the correlation-reduced descriptors with higher importance; A trained model is created by machine learning using the important descriptors and the characteristic values ​​of the material as training data; Using the trained model, estimate a property value of a material different from the material. An information processing system comprising:

12. In identifying the important descriptors, a predetermined number of the descriptors having the largest transform coefficients used for transforming into the correlation reduced descriptor are identified as the important descriptors.

12. The information processing system according to claim 11.

13. In identifying the important descriptor, the descriptor whose transformation coefficient used for transformation into the correlation reduced descriptor exceeds a predetermined threshold is identified as the important descriptor.

12. The information processing system according to claim 11.

14. In identifying the important descriptor, the important descriptor is identified based on a plurality of the transformation coefficients used for transformation into a plurality of the correlation reduced descriptors with high importance.

12. The information processing system according to claim 11.

15. In identifying the important descriptor, the important descriptor is identified based on a result of weighted averaging of the transformation coefficients used in the transformation into the correlation reduced descriptor with weights of importance.

12. The information processing system according to claim 11.

16. a database that records the descriptors associated with the material; 16. The information processing system according to claim 11,

17. a display unit that displays information related to the processing executed by the information processing unit; 16. The information processing system according to claim 11,

18. an input unit that receives information necessary for the information processing unit to execute processing; 16. The information processing system according to claim 11,

19. The obtained descriptors include structural descriptors.

16. The information processing system according to claim 11,

Images