Prediction method, information processing device, and program

Abstract

Improve techniques for predicting the application performance of polymeric materials. [Solution] A prediction method executed by an information processing device includes: inputting feature quantities based on material design information into a trained first machine learning model that has undergone machine learning to predict physical properties related to a polymer material, thereby obtaining predicted values ​​of physical properties from the first machine learning model; and inputting feature quantities based on material design information and predicted values ​​of physical properties into a trained second machine learning model that has undergone machine learning to predict the applied performance of the polymer material, thereby obtaining the applied performance of the polymer material from the second machine learning model.

Description

[Technical Field]

[0001] The present disclosure relates to a prediction method, an information processing device, and a program. [Background technology]

[0002] Conventionally, there are known techniques for predicting the application performance of materials using machine learning, etc. For example, Patent Document 1 discloses a method for predicting the adhesive strength of a curable composition. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-038344 Summary of the Invention [Problem to be solved by the invention]

[0004] Patent Document 1 assumes that when training a learning model, information that can be known at the design stage of a polymer material, such as composition, functional group information, molecular weight information, and manufacturing conditions, is used. In other words, a conventional method for predicting the application performance of a polymer material has generally been considered to use the composition, structure (e.g., functional groups), molecular weight information, and manufacturing conditions of the polymer material as explanatory variables, and to generate a prediction model using the application performance of the polymer material as the target variable. However, conventional methods that use only information that can be known at the design stage of a polymer material, such as composition, structure, molecular weight information, and manufacturing conditions, as explanatory variables can sometimes result in low accuracy in the application performance of polymer materials, and there is room for improvement in the technology for predicting application performance.

[0005] In view of the above circumstances, an object of the present disclosure is to improve the technology for predicting the applied performance of polymer materials. [Means for solving the problem]

[0006] (1) A method according to one embodiment of the present disclosure includes: A prediction method executed by an information processing device, acquiring predicted values ​​of physical properties from a first machine learning model that has undergone machine learning to predict physical properties of a polymer material by inputting feature quantities based on material design information into the first machine learning model; The method includes inputting features based on the material design information and predicted values ​​of the physical properties into a trained second machine learning model that has undergone machine learning to predict the applied performance of the polymer material, thereby obtaining the applied performance of the polymer material from the second machine learning model.

[0007] (2) A prediction method according to an embodiment of the present disclosure is the prediction method according to (1), The feature amount is generated by compressing the analysis chart of each raw material of the polymer material into a predetermined dimension and adding up the analysis chart of each dimension according to the blending ratio of each raw material of the polymer material.

[0008] (3) A prediction method according to an embodiment of the present disclosure is the prediction method according to (2), The analysis chart is an analysis chart showing the molecular structure of each raw material of the polymer material, and an analysis chart including molecular weight information of each raw material of the polymer material.

[0009] (4) A prediction method according to an embodiment of the present disclosure is the prediction method according to (3), The analytical chart showing the molecular structure of each raw material of the polymer material is an IR chart, and the analytical chart containing the molecular weight information of each raw material of the polymer material is a GPC chart.

[0010] (5) A prediction method according to an embodiment of the present disclosure is the prediction method according to any one of (1) to (4), The predicted value of the physical property obtained from the first machine learning model is chart information, and the predicted value of the physical property input to the second machine learning model is information on a partial area of ​​the chart information.

[0011] (6) A prediction method according to an embodiment of the present disclosure is the prediction method according to (5), The predicted physical properties are the physico-rheological properties of the polymeric material.

[0012] (7) A prediction method according to an embodiment of the present disclosure is the prediction method according to (6), The physical rheological properties of the polymeric material are viscoelastic properties.

[0013] (8) A prediction method according to one embodiment of the present disclosure is the prediction method described in any one of (1) to (7), wherein the applied performance of the polymer material is adhesive strength, shear strength, and constant load strength.

[0014] (9) A prediction method according to an embodiment of the present disclosure is the prediction method according to any one of (1) to (8), wherein the performance data for training the first machine learning model includes data on materials used in adhesive applications and performance data on resin compositions used in applications other than adhesive applications.

[0015] (10) An information processing device according to an embodiment of the present disclosure includes: An information processing device for predicting applied performance of a polymer material, comprising: The control unit acquiring predicted values ​​of physical properties from a first machine learning model that has undergone machine learning to predict physical properties of a polymer material by inputting feature quantities based on material design information into the first machine learning model; The applied performance of the polymer material is obtained from the second machine learning model by inputting features based on the material design information and predicted values ​​of the physical properties into a trained second machine learning model that has undergone machine learning to predict the applied performance of the polymer material.

[0016] (11) A program according to an embodiment of the present disclosure includes: A program for predicting the application performance of a polymer material executed by an information processing device, the program comprising: acquiring predicted values ​​of physical properties from a first machine learning model that has undergone machine learning to predict physical properties of a polymer material by inputting feature quantities based on material design information into the first machine learning model; The applied performance of the polymer material is obtained from the second machine learning model by inputting features based on the material design information and predicted values ​​of the physical properties into a trained second machine learning model that has undergone machine learning to predict the applied performance of the polymer material. [Effects of the Invention]

[0017] According to the prediction method, information processing device, and program according to an embodiment of the present disclosure, it is possible to improve the technology for predicting the applied performance of polymer materials. [Brief explanation of the drawings]

[0018] [Figure 1] FIG. 1 is a diagram showing an example of an adhesive material. [Figure 2] FIG. 1 is a diagram showing an example of storage modulus. [Figure 3] FIG. 10 is a diagram showing an example of loss modulus. [Figure 4] FIG. 10 is a diagram illustrating an example of a loss tangent. [Figure 5] FIG. 1 is a diagram illustrating an overview of a prediction technique according to an embodiment of the present disclosure. [Figure 6] FIG. 1 illustrates an example of a prediction technique according to an embodiment of the present disclosure. [Figure 7] 1 is a block diagram illustrating a schematic configuration of an information processing device according to an embodiment of the present disclosure. [Figure 8] 1 is a flowchart illustrating a prediction method according to an embodiment of the present disclosure. [Figure 9] 10 shows a graph comparing predicted and experimental results of shear retention force using a prediction technique according to an embodiment of the present disclosure. [Figure 10] 10 is a graph comparing the predicted results of shear holding force by a prediction technique according to a comparative example with the experimental results. [Figure 11]10 shows a graph comparing predicted and experimental results of constant load holding force using a prediction technique according to an embodiment of the present disclosure. [Figure 12] 10 is a graph comparing the predicted results of constant load holding force by a prediction technique according to a comparative example with the experimental results. [Figure 13] 10 shows a graph comparing predicted and experimental results of 180° peel adhesive force using a prediction technique according to an embodiment of the present disclosure. [Figure 14] 10 is a graph comparing the predicted results of 180° peel adhesive strength by a prediction technique according to a comparative example with the experimental results. DETAILED DESCRIPTION OF THE INVENTION

[0019] Hereinafter, a prediction technique according to an embodiment of the present disclosure will be described with reference to the drawings.

[0020] In each drawing, the same or corresponding parts are denoted by the same reference numerals. In the description of this embodiment, the description of the same or corresponding parts will be omitted or simplified as appropriate.

[0021] (Prediction target) The target of the prediction technology according to an embodiment of the present disclosure is, for example, the application performance of a polymer material for use in an adhesive material. The application performance may be, for example, shear holding strength, constant load holding strength, and adhesive strength (e.g., 180° peel adhesive strength). However, the target of prediction is not limited to this. The material to be predicted may be, for example, an organic compound, an inorganic compound, a low molecular weight material, a polymer material such as a resin composition, or more specifically, a reinforced carbon fiber plastic. The material properties to be predicted may also be any of these properties. In the following embodiment, a case where the target of prediction is the application performance of a polymer material for use in an adhesive material will be mainly described.

[0022] First, the adhesive material will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of an adhesive material.

[0023] In this specification, the term "adhesive material" refers to a material that has the function of bonding objects together. Examples of the adhesive material include adhesive tape and adhesives. In the following, an explanation will be given using adhesive tape as an example of the adhesive material, but the adhesive material is not limited to adhesive tape.

[0024] The adhesive tape may be a single-sided tape or a double-sided tape. The application of the adhesive tape is not particularly limited, and the tape may be for industrial use or for home use. Examples of types of adhesive tape include coreless tapes, nonwoven fabric core tapes, film core tapes, foam core tapes, and metal foil core tapes.

[0025] 1 shows a single-sided adhesive tape 10A, which is an example of an adhesive material 10. The single-sided adhesive tape 10A is composed of, for example, a sheet-like core substrate 12, an adhesive 14, and a release agent 16.

[0026] There are no particular limitations on the type of core substrate 12. The core substrate 12 may be a plastic film, paper, a foam, a metal foil, or any other type of sheet.

[0027] An adhesive 14 is applied to one surface of the core substrate 12. A release agent 16 is applied to the other surface of the core substrate 12.

[0028] The release agent 16 covers the adhesive 14 when the single-sided adhesive tape 10A is wound in a roll, and protects the adhesive 14 so that the adhesive strength does not decrease.

[0029] The viscoelastic properties of a polymeric material that serves as an adhesive for a pressure-sensitive adhesive material are temperature-dependent. Such viscoelastic properties are shown, for example, by a viscoelastic property chart (hereinafter also referred to as a viscoelasticity chart). Specific examples of the viscoelasticity chart include charts of storage modulus, loss modulus, and loss tangent.

[0030] The storage modulus corresponds to the component of energy stored in the adhesive as elastic energy when the adhesive is deformed, and is an index representing the degree of hardness of the adhesive.

[0031] Fig. 2 is a diagram showing an example of the storage modulus. In the example of Fig. 2, the temperature is indicated by "T", and the storage modulus at the temperature T is indicated by "G1(T)".

[0032] The loss modulus corresponds to the component of the loss energy dissipated due to internal friction when the adhesive is deformed, and is an index representing the degree of viscosity of the adhesive.

[0033] Fig. 3 is a diagram showing an example of the loss modulus. In the example of Fig. 3, the temperature is indicated by "T", and the loss modulus at the temperature T is indicated by "G2(T)".

[0034] The loss tangent is expressed as the ratio of the storage modulus "G1(T)" to the loss modulus "G2(T)". Typically, the loss tangent "tanδ(T)" is expressed by the following formula (1):

[0035] tanδ(T)=G2(T) / G1(T) (1) Fig. 4 is a diagram showing an example of the loss tangent. In the example of Fig. 4, the temperature is indicated by "T", and the loss tangent at the temperature T is indicated by "tan δ(T)".

[0036] As will be described later, the prediction technique according to this embodiment utilizes information on basic properties (hereinafter also referred to as predicted physical properties) to accurately predict the application performance of the material. The predicted physical properties include the physiorheological properties of polymeric materials such as soft matter. Physorheological properties are properties related to the deformation and flow of a material, such as viscoelasticity, stress relaxation, creep, and thixotropy. The prediction technique according to this embodiment may also be used to predict application performance related to mechanical properties such as fracture toughness, tensile strength, tensile modulus, and tensile elongation, as well as application performance related to thermal properties such as glass transition temperature (Tg). In this case, the corresponding physical property information may be the indentation modulus, creep rate, elastic deformation rate, indentation depth, etc. The physical property information and application performance are not limited to these, and any physical property information and application performance can be predicted with high accuracy by the prediction technique according to this embodiment.

[0037] (Outline of this embodiment) First, an overview of the present embodiment will be described. A prediction method according to an embodiment of the present disclosure is executed by an information processing device. In addition, the prediction method according to an embodiment of the present disclosure acquires predicted values ​​of physical properties from a first machine learning model that has been trained through machine learning to predict physical properties of a polymer material by inputting feature quantities based on material design information into the first machine learning model. In addition, a prediction method according to an embodiment of the present disclosure acquires predicted values ​​of physical properties from the first machine learning model by inputting feature quantities based on material design information and predicted values ​​of physical properties into a second machine learning model that has been trained through machine learning to predict applied performance of the polymer material.

[0038] As described above, according to the prediction method of the embodiment of the present disclosure, first, predicted values ​​of physical properties are obtained by inputting feature quantities based on material design information from a first machine learning model. Then, the predicted values ​​of physical properties are also used as explanatory variables to obtain the applied performance of the polymer material from a second machine learning model. Therefore, the technology for predicting the applied performance of polymer materials is improved in that the applied performance can be predicted with high accuracy by taking into account the predicted values ​​of physical properties, which are information on basic physical properties.

[0039] (Outline of prediction technology) FIG. 5 illustrates an outline of a prediction technique according to an embodiment of the present disclosure. As illustrated in FIG. 5, feature quantities based on material design information of a polymer material (e.g., a resin composition) are input as explanatory variables to a first machine learning model 100. The material design information is, for example, the blending composition of the polymer material. In this case, the feature quantities based on the material design information may be a vector of feature quantities calculated based on the blending composition of the polymer material (hereinafter, referred to as a blending feature vector). Note that the material design information is not limited to this. For example, the feature quantities may be blending composition and manufacturing conditions, blending composition and characteristic values ​​of raw materials, characteristic values ​​of raw materials, characteristic values ​​of raw materials and manufacturing conditions, blending composition, and characteristic values ​​of raw materials and manufacturing conditions. The first machine learning model 100 outputs a predicted value of a physical property as a response variable. Furthermore, the feature quantities (e.g., blending feature vector) are input as first explanatory variables to a second machine learning model 200. Furthermore, the predicted value of the physical property output from the first machine learning model 100 is input as a second explanatory variable to the second machine learning model 200. In other words, the feature amounts related to the blend composition and the predicted values ​​of the physical properties are input to the second machine learning model 200. The second machine learning model 200 outputs the applied performance of the polymer material as a response variable.

[0040] FIG. 6 illustrates an example of a prediction technique according to an embodiment of the present disclosure. In the example of FIG. 6, feature quantities based on material design information are generated based on the blend composition and IR-GPC data of the raw materials. Specifically, the IR-GPC data of the raw materials is compressed and converted into feature quantities by a VAE encoder 110. In the example illustrated in FIG. 6, the feature quantities are six-dimensional vector data. This vector data is input to a first machine learning model 100. The first machine learning model 100 outputs predicted values ​​of physical properties. The predicted values ​​of the physical properties in FIG. 6 are the loss tangent (tanδ(T)), the storage modulus (G1), and the loss modulus (G2(T)). The predicted values ​​of the physical properties are input to a second machine learning model 200. In the example of FIG. 6, only a portion of the predicted values ​​of the physical properties is input to the second machine learning model 200. Any method may be used to select the information of the predicted values ​​of the physical properties to be input to the second machine learning model 200. For example, a genetic algorithm is used to select predicted values ​​of physical properties to be input to the second machine learning model 200. By limiting the predicted values ​​of physical properties to be input to the second machine learning model 200 to a partial chart region in this way, overlearning can be prevented and prediction accuracy can be improved. The second machine learning model 200 outputs the applied performance of the polymer material as a dependent variable.

[0041] (Configuration of information processing device) Next, a detailed description will be given of each component of the information processing device 300. The information processing device 300 is any device used by a user. For example, a personal computer, a server computer, a general-purpose electronic device, or a dedicated electronic device can be adopted as the information processing device 300.

[0042] As shown in FIG. 7, the information processing device 300 includes a control unit 301, a storage unit 302, an input unit 303, and an output unit 304.

[0043] The control unit 301 includes at least one processor, at least one dedicated circuit, or a combination thereof. The processor is a general-purpose processor such as a central processing unit (CPU) or a graphics processing unit (GPU), or a dedicated processor specialized for a specific process. The dedicated circuit is, for example, a field-programmable gate array (FPGA) or an application specific integrated circuit (ASIC). The control unit 301 executes processes related to the operation of the information processing device 300 while controlling each unit of the information processing device 300.

[0044] The storage unit 302 includes at least one semiconductor memory, at least one magnetic memory, at least one optical memory, or a combination of at least two of these. The semiconductor memory is, for example, a random access memory (RAM) or a read only memory (ROM). The RAM is, for example, a static random access memory (SRAM) or a dynamic random access memory (DRAM). The ROM is, for example, an electrically erasable programmable read only memory (EEPROM). The storage unit 302 functions as, for example, a main storage device, an auxiliary storage device, or a cache memory. The storage unit 302 stores data used in the operation of the information processing device 300 and data obtained by the operation of the information processing device 300.

[0045] The input unit 303 includes at least one input interface. The input interface may be, for example, a physical key, a capacitance key, a pointing device, or a touch screen integrated with a display. The input interface may also be, for example, a microphone that accepts voice input, or a camera that accepts gesture input. The input unit 303 accepts an operation to input data used for the operation of the information processing device 300. The input unit 303 may be connected to the information processing device 300 as an external input device instead of being provided in the information processing device 300. As a connection method, any method such as USB (Universal Serial Bus), HDMI (registered trademark) (High-Definition Multimedia Interface), or Bluetooth (registered trademark) may be used.

[0046] The output unit 304 includes at least one output interface. The output interface is, for example, a display that outputs information as a video. The display is, for example, an LCD (liquid crystal display) or an organic EL (electro luminescence) display. The output unit 304 displays and outputs data obtained by the operation of the information processing device 300. The output unit 304 may be connected to the information processing device 300 as an external output device instead of being provided in the information processing device 300. As a connection method, any method such as USB, HDMI (registered trademark), or Bluetooth (registered trademark) can be used.

[0047] The functions of the information processing device 300 are realized by executing a program according to this embodiment on a processor corresponding to the information processing device 300. That is, the functions of the information processing device 300 are realized by software. The program causes a computer to execute the operations of the information processing device 300, thereby causing the computer to function as the information processing device 300. That is, the computer functions as the information processing device 300 by executing the operations of the information processing device 300 in accordance with the program.

[0048] In this embodiment, the program can be recorded on a computer-readable recording medium. The computer-readable recording medium includes non-transitory computer-readable media, such as a magnetic recording device, an optical disc, a magneto-optical recording medium, or a semiconductor memory. The program can be distributed, for example, by selling, transferring, or lending a portable recording medium, such as a DVD (digital versatile disc) or a CD-ROM (compact disc read only memory), on which the program is recorded. The program can also be distributed by storing the program in the storage of an external server and transmitting the program from the external server to another computer. The program can also be provided as a program product.

[0049] Some or all of the functions of the information processing device 300 may be implemented by a dedicated circuit equivalent to the control unit 301. In other words, some or all of the functions of the information processing device 300 may be implemented by hardware.

[0050] Referring now to the flowchart of FIG. 8, a technique for predicting the application performance of polymeric materials according to one embodiment of the present disclosure is illustrated.

[0051] Step S101: The control unit 301 of the information processing device 300 inputs features based on the material design information into the first machine learning model 100, thereby acquiring predicted values ​​of physical properties from the first machine learning model 100. Specifically, for example, the control unit 301 inputs a blend feature vector into the first machine learning model 100 as a feature based on the material design information. As described above, the blend feature vector may be generated based on the blend composition and an analysis chart (e.g., IR / GPC data) of the raw materials. Specifically, for example, the control unit 301 converts the IR / GPC data of the raw materials into features of a predetermined dimension using the VAE encoder 110. For example, the control unit 301 compresses the IR data of each raw material into first- and second-dimensional data. Furthermore, the control unit 301 compresses the high molecular weight data of the GPC data of each raw material into third- and fourth-dimensional data. Furthermore, the control unit 301 compresses the low molecular weight data of the GPC data of each raw material into fifth- and sixth-dimensional data. That is, the control unit 301 compresses the analysis chart of each raw material into a predetermined dimension. In this way, the control unit 301 generates, for example, six-dimensional features for each raw material. Then, the control unit 301 generates a blend feature vector of the polymer material based on the blend composition ratio of each raw material for the six-dimensional feature of each raw material. Specifically, the control unit 301 generates a blend feature vector of the polymer material by multiplying the six-dimensional feature of each raw material by the blend composition ratio. In other words, the control unit 301 sums up the analysis charts of each dimension according to the blend ratio of the raw materials, and generates a blend feature vector based on the summation results for each dimension. The control unit 301 inputs the blend feature vector into the first machine learning model 100.

[0052] Although an example using IR and GPC data as the raw material analysis chart has been shown here, this is not limiting. The raw material analysis chart may include an analysis chart showing the molecular structure of any polymeric material and an analysis chart containing molecular weight information of the polymeric material. Specifically, for example, the analysis chart showing the molecular structure of the polymeric material may include IR, Raman, NMR, and X-ray spectra. Furthermore, for example, the analysis chart containing molecular weight information of the polymeric material may include GPC, SEC, FFF, and electrochemical oscillation.

[0053] When predicting the physical rheological properties of polymeric materials as physical properties, it is desirable to use an IR chart as the analytical chart. An IR chart is easy to measure as it shows the molecular structure of the raw material. Furthermore, it can reflect the functional group structure, crosslink density, and other factors important to the physical rheological properties of polymers as explanatory variables. It is desirable to use a GPC chart as the analytical chart containing molecular weight information of the raw material. As an analytical chart containing molecular weight information, a GPC chart is easy to measure and highly accurate. Furthermore, by using molecular weight distribution information as an explanatory variable rather than single information such as average molecular weight, it is possible to reflect information on the entire bulk of the polymeric material, enabling accurate prediction of viscoelastic properties. In other words, using an IR chart as information corresponding to the polymer structure and a GPC chart as information corresponding to the polymer size distribution can improve prediction accuracy.

[0054] Any method may be used for the learning process of the first machine learning model 100. For example, the first machine learning model 100 may be a neural network model. Specifically, the first machine learning model 100 may be a fully connected neural network, a convolutional neural network, a recurrent neural network, or a neural network with an attention mechanism. However, the machine learning algorithm used in the learning process is not limited to a neural network model. For example, various machine learning algorithms may be used, such as neural networks such as deep learning, support vector machines, Gaussian process regression models, or decision tree systems. When a Gaussian process regression model is used, information such as the standard deviation and variance of the prediction can be obtained along with the predicted value. This makes it possible to calculate the probability of achieving the target value from this numerical information. Therefore, when a Gaussian process regression model is used, it is also possible to formulate an experimental plan that satisfies the target characteristics based on the achievement probability.

[0055] Step S102: The control unit 301 inputs features based on the material design information and predicted values ​​of physical properties acquired from the first machine learning model into the second machine learning model 200, thereby acquiring the applied performance of the polymer material from the second machine learning model 200. Specifically, for example, the control unit 301 inputs a blend feature vector into the second machine learning model 200 as a feature related to the blend composition of the polymer material. As in the above-described step S102, the blend feature vector may be generated based on the blend composition and an analysis chart (e.g., IR / GPC data) of the raw materials. Specifically, for example, the control unit 301 converts the IR / GPC data of the raw materials into features of a predetermined dimension using the VAE encoder 110. For example, the control unit 301 compresses the IR data of each raw material into first- and second-dimensional data. Furthermore, the control unit 301 compresses the polymer data of the GPC data of each raw material into third- and fourth-dimensional data. The control unit 301 also compresses the low-molecular-weight data from the GPC data for each raw material into fifth- and sixth-dimensional data. In this way, the control unit 301 generates, for example, six-dimensional features for each raw material. The control unit 301 then generates a polymer material blend feature vector based on the blend composition ratio of each raw material, using the six-dimensional features of each raw material. Specifically, the control unit 301 generates a polymer material blend feature vector by multiplying the six-dimensional features of each raw material by the blend composition ratio. The control unit 301 inputs the blend feature vector into the second machine learning model 200. Similarly to the above, the raw material analysis chart may include an analysis chart showing the molecular structure of an arbitrary polymer material and an analysis chart containing molecular weight information of the polymer material. Specifically, for example, the analysis chart showing the molecular structure of the polymer material may include IR, Raman, NMR, and X-ray spectra. For example, the analysis chart containing molecular weight information of the polymer material may include GPC, SEC, FFF, and electrochemical oscillation.

[0056] As described above, the control unit 301 inputs the predicted values ​​of the physical properties acquired from the first machine learning model 100 to the second machine learning model 200 in addition to the blend feature vector. The input predicted values ​​of the physical properties may be, for example, chart information of the physical properties. Examples of the chart information of the physical properties include the loss tangent (tanδ(T)), the storage modulus (G1), and the loss modulus (G2(T)). As described above, the chart information input to the second machine learning model 200 may be limited to a portion of the chart area. Any method may be used to limit the portion of the chart area. For example, the chart information input to the second machine learning model 200 may be limited to a portion of the chart area using a genetic algorithm. By limiting the physical property information input to the second machine learning model 200 to a portion of the chart area, overlearning can be prevented and prediction accuracy can be improved.

[0057] The learning process of the second machine learning model 200 may employ any method. For example, the second machine learning model 200 may be a neural network model. Specifically, the second machine learning model 200 may be a fully connected neural network, a convolutional neural network, a recurrent neural network, or a neural network with an attention mechanism. However, the machine learning algorithm employed in the learning process is not limited to a neural network model. Various machine learning algorithms, such as neural networks such as deep learning, support vector machines, Gaussian process regression models, and decision tree systems, may be employed. When a Gaussian process regression model is used, information such as the standard deviation and variance of the prediction can be obtained along with the predicted value. This makes it possible to calculate the probability of achieving the target value from this numerical information. Therefore, when a Gaussian process regression model is used, it is also possible to formulate an experimental plan that satisfies the target characteristics based on the achievement probability.

[0058] Step S103: The control unit 301 outputs information related to the application performance. Any method can be used to output the information. For example, the control unit 301 may present the information related to the application performance through a user interface that is displayed and output by the output unit 304.

[0059] FIG. 9 shows a comparison graph of predicted values ​​of shear holding force predicted using the prediction technology according to this embodiment (hereinafter referred to as "Example") and actual measured values. As shown in FIG. 9, in the Example, it can be seen that the predicted results of shear holding force and the experimental results are generally in agreement. In addition, the coefficient of determination (R 2 ) is 0.93, which indicates high prediction accuracy.

[0060] Figure 10 shows a comparison graph of predicted values ​​of shear holding strength predicted using the prediction technology according to the comparative example and the actual measured values. Note that the prediction technology according to the comparative example does not use physical property information as an explanatory variable, but predicts the applied performance (shear holding strength in this case) using only the blend feature vector of the polymer material. As shown in Figure 10, in the comparative example, it can be seen that there is a slight deviation between the predicted results of shear holding strength and the experimental results. In addition, the coefficient of determination (R 2 ) is 0.73, and it can be seen that the prediction technique according to the embodiment has higher prediction accuracy.

[0061] FIG. 11 shows a comparison graph of the predicted values ​​of the constant load holding force predicted using the prediction technique according to the embodiment and the actual measured values. As shown in FIG. 11, it can be seen that the predicted results of the constant load holding force and the experimental results are generally in agreement. In addition, the coefficient of determination (R 2 ) is 0.89, indicating high prediction accuracy.

[0062] FIG. 12 shows a graph comparing the predicted values ​​of the constant load holding force predicted using the prediction technique according to the comparative example with the actual measured values. As shown in FIG. 12, in the comparative example, it can be seen that the predicted results of the constant load holding force are somewhat different from the experimental results. In addition, the coefficient of determination (R 2 ) is 0.34, and it can be seen that the prediction technique according to the embodiment has higher prediction accuracy.

[0063] FIG. 13 shows a comparison graph of the predicted values ​​of 180° peel adhesive strength predicted using the prediction technique according to the embodiment and the actual measured values. As shown in FIG. 13, it can be seen that the predicted results of 180° peel adhesive strength generally match the experimental results. In addition, the coefficient of determination (R 2 ) is 0.66, which indicates high prediction accuracy.

[0064] FIG. 14 shows a graph comparing the predicted values ​​of 180° peel adhesive strength predicted using the prediction technique according to the comparative example with the actual measured values. As shown in FIG. 14, in the comparative example, the predicted results of 180° peel adhesive strength are somewhat different from the experimental results. In addition, the coefficient of determination (R 2 ) is 0.49, and it can be seen that the prediction technique according to the embodiment has higher prediction accuracy.

[0065] Table 1 shows the prediction accuracy of each application performance (shear holding strength, constant load holding strength, and 180° peel adhesive strength) for the examples and comparative examples.

[0066] [Table 1]

[0067] As shown in Table 1, it can be seen that the prediction technique according to the embodiment has higher accuracy for each application performance.

[0068] As described above, in the prediction method according to the embodiment of the present disclosure, feature quantities based on material design information are input to a trained first machine learning model 100 that has undergone machine learning to predict physical properties of a polymer material, thereby obtaining predicted values ​​of the physical properties from the first machine learning model 100. In addition, in the prediction method according to the embodiment of the present disclosure, feature quantities related to the blend composition of the polymer material and predicted values ​​of the physical properties are input to a trained second machine learning model 200 that has undergone machine learning to predict the applied performance of the polymer material, thereby obtaining the applied performance of the polymer material from the second machine learning model 200.

[0069] According to this prediction method, first, predicted values ​​of physical properties are obtained by inputting feature quantities based on material design information from a first machine learning model 100. Then, the predicted values ​​of physical properties are also used as explanatory variables to obtain the applied performance of the polymer material from a second machine learning model 200. This improves the technology for predicting the applied performance of polymer materials in that the applied performance can be predicted with high accuracy by taking into account the predicted values ​​of physical properties, which are information on basic physical properties.

[0070] The performance data for training the first machine learning model 100 and the second machine learning model 200 may include data on polymer materials used for predetermined applications and performance data on polymer materials used for applications other than the predetermined applications. More specifically, the first machine learning model 100 and the second machine learning model 200 may be generated using data on polymer materials used for applications other than adhesives, and then the first machine learning model 100 and the second machine learning model 200 may be retrained using data on polymer materials used for adhesives. In this way, in the process of training the first machine learning model 100 and the second machine learning model 200, by including compositions with a wide range of applications in the training data used for training, it is possible to prevent a decrease in accuracy due to extrapolation.

[0071] Although the present disclosure has been described based on the drawings and examples, it should be noted that those skilled in the art can easily make various modifications and alterations based on the present disclosure. Therefore, it should be noted that these modifications and alterations are included in the scope of the present disclosure. For example, the functions included in each means or step can be rearranged so as not to be logically inconsistent, and multiple means or steps can be combined or divided into one. [Explanation of symbols]

[0072] 10 Adhesive material 10A single-sided adhesive tape 12 Core base material 14 Adhesive 16 Stripping agent 100 First Machine Learning Model 200 Second Machine Learning Model 110 VAE Encoder 300 Information processing device 301 Control Unit 302 Storage section 303 Input section 304 Output section

Claims

1. A prediction method executed by an information processing device, acquiring predicted values ​​of physical properties from a first machine learning model that has undergone machine learning to predict physical properties of a polymer material by inputting feature quantities based on material design information into the first machine learning model; A prediction method comprising: inputting features based on the material design information and predicted values ​​of the physical properties into a trained second machine learning model that has undergone machine learning to predict the applied performance of the polymer material, thereby obtaining the applied performance of the polymer material from the second machine learning model.

2. 2. The prediction method of claim 1, A prediction method in which the feature amount is generated by compressing an analysis chart of each raw material of the polymer material into a predetermined dimension and adding up the analysis charts of each dimension according to the blending ratio of each raw material of the polymer material.

3. 3. The prediction method according to claim 2, A prediction method, wherein the analysis chart is an analysis chart showing the molecular structure of each raw material of the polymer material, and an analysis chart including molecular weight information of each raw material of the polymer material.

4. 4. The prediction method according to claim 3, The prediction method, wherein the analytical chart showing the molecular structure of each raw material of the polymer material is an IR chart, and the analytical chart containing the molecular weight information of each raw material of the polymer material is a GPC chart.

5. 2. The prediction method according to claim 1, wherein the predicted value of the physical property obtained from the first machine learning model is chart information, and the predicted value of the physical property input to the second machine learning model is information of a partial area of ​​the chart information.

6. 6. The prediction method according to claim 5, The prediction method, wherein the predicted value of the physical property is a physico-rheological property of the polymeric material.

7. 7. The prediction method according to claim 6, A prediction method, wherein the physico-rheological properties of the polymeric material are viscoelastic properties.

8. 2. The prediction method of claim 1, The application performance of the polymeric material is adhesion strength, shear holding strength, and constant load holding strength.

9. 2. The prediction method according to claim 1, wherein the performance data for training the first machine learning model includes data on materials used in pressure-sensitive adhesive applications and performance data on resin compositions used in applications other than pressure-sensitive adhesive applications.

10. An information processing device for predicting applied performance of a polymer material, comprising: The control unit acquiring predicted values ​​of physical properties from a first machine learning model that has undergone machine learning to predict physical properties of a polymer material by inputting feature quantities based on material design information into the first machine learning model; An information processing device that obtains the applied performance of the polymer material from a second machine learning model by inputting features based on the material design information and predicted values ​​of the physical properties into a trained second machine learning model that has undergone machine learning to predict the applied performance of the polymer material.

11. A program for predicting the application performance of a polymer material executed by an information processing device, the program comprising: acquiring predicted values ​​of physical properties from a first machine learning model that has undergone machine learning to predict physical properties of a polymer material by inputting feature quantities based on material design information into the first machine learning model; and obtaining the applied performance of the polymer material from the second machine learning model by inputting features based on the material design information and the predicted values ​​of the physical properties into a trained second machine learning model that has undergone machine learning to predict the applied performance of the polymer material.

Images