Method for generating a trained model, estimation method, trained model generation program, and estimation program

Abstract

To provide a method for generating a trained model, an estimation method, a trained model generation program, and an estimation program that can accurately predict a hardening curve showing the time change in the degree of hardening. [Solution] The present invention provides a method for generating training data, comprising: a data set generation step of generating a plurality of data sets, each set consisting of a formulation and heating temperature of a thermosetting resin and a curing curve showing the time change in the degree of curing at the formulation and heating time; and a learning step of generating a trained model by learning using the plurality of data sets, with the formulation and heating temperature of the thermosetting resin as explanatory variables and the curing curve at the formulation and heating time as the target variable.

Description

Technical Field

[0001] The present invention relates to a method for generating a learned model, an estimation method, a learned model generation program, and an estimation program.

Background Art

[0002] A fiber-reinforced composite material is manufactured by integrally combining reinforcing fibers and a matrix resin. For example, there are a method of pressure molding a prepreg in which reinforcing fibers are impregnated in advance with a thermosetting resin as a matrix resin using an autoclave or a press machine, and a method of injecting a thermosetting resin, which is a low-viscosity matrix resin, into a shaped reinforcing fiber base material and heating and curing it. In these methods, the process of curing the thermosetting resin is common, and it is necessary to appropriately set the heating and curing time from the viewpoints of productivity and quality control.

[0003] In recent years, along with the expansion of the application range of fiber-reinforced composite materials in the manufacturing industry, such as in addition to structural members of airplanes and automobiles, computer casings, and concrete repair for civil engineering and construction, it has been desired to construct a technology for predicting the curing time according to the application and the curing state during the molding process.

[0004] As a technology for predicting the curing state, for example, Patent Document 1 discloses a method of creating a learned model using the formulation information and physical properties of a material as feature amounts and predicting the physical properties of a polymer material. Further, Patent Document 2 discloses a method of improving the accuracy of reaction path calculation by adding data on reaction conditions obtained by simulation to the actually measured results obtained by actual measurement and performing learning. Further, Patent Document 3 describes a method of creating a learned model by adding parameters obtained from molecular orbital method calculations to the prediction of the reaction rate related to the synthesis of a resin material.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

[0006] Incidentally, as a manufacturing condition for fiber-reinforced composite materials, it is sometimes necessary to predict the curing curve, which shows the change in degree of curing over time, and then set the temperature and heating time based on the predicted curing curve. However, Patent Document 1 does not describe the curing reaction rate of the monomers that make up the resin. Furthermore, Patent Document 2 deals with low molecular weight synthesis reactions using solvents and cannot determine the reaction rate of resins that do not contain solvents, such as thermosetting resins. The reactivity prediction method described in Patent Document 3 does not target thermosetting resins and cannot derive the reaction time. Thus, even if the curing curve is predicted using the techniques described in Patent Documents 1 to 3, there is a risk that the prediction accuracy will decrease.

[0007] The present invention has been made in view of the above, and aims to provide a method for generating a trained model, an estimation method, a trained model generation program, and an estimation program that can predict with high accuracy a hardening curve showing the change in hardening degree over time. [Means for solving the problem]

[0008] To solve the above-mentioned problems and achieve the objective, the present invention provides a method for generating a trained model in which a computer generates a trained model for estimating the curing curve of a thermosetting resin, comprising: a dataset generation step of generating a plurality of datasets, each dataset consisting of a formulation and heating temperature of the thermosetting resin and a curing curve showing the time change in the degree of curing at the formulation and heating time; and a learning step of generating a trained model by learning using the plurality of datasets, with the formulation and heating temperature of the thermosetting resin as explanatory variables and the curing curve at the formulation and heating time as the target variable.

[0009] Furthermore, in the pre-trained model generation method according to the present invention, the hardening curve in the above invention shows the change in the degree of hardening from the time of heating start to the time when a predetermined degree of hardening is reached.

[0010] Furthermore, the pre-trained model generation method according to the present invention includes, in the above invention, a training step of distributing the dataset into a training dataset and a validation dataset; a model generation step of applying the training dataset to a plurality of different statistical models to generate a pre-trained model corresponding to each statistical model; and a selection step of inputting explanatory variables from the validation dataset into the pre-trained model corresponding to each statistical model, calculating the prediction error by cross-validation based on the output values ​​obtained, and selecting the statistical model that gives the smallest prediction error.

[0011] Furthermore, in the pre-trained model generation method according to the present invention, the training step generates the pre-trained model by machine learning or multivariate regression analysis.

[0012] Furthermore, in the pre-trained model generation method according to the present invention, the training step is generated by multivariate regression analysis to generate the pre-trained model which includes an interaction term between the type and amount of resin included in the dataset.

[0013] Furthermore, in the pre-trained model generation method according to the present invention, the formulation of the thermosetting resin comprises an epoxy resin and a curing agent component that reacts with the epoxy groups contained in the epoxy resin.

[0014] Furthermore, the estimation method according to the present invention is an estimation method in which a computer estimates the curing curve of a thermosetting resin, and includes a reaction time acquisition step in which a trained model generated by learning a plurality of datasets in which the formulation and heating temperature of the thermosetting resin are explanatory variables and the curing curve showing the time change of the degree of curing at the formulation and heating time is the objective variable is read from a storage unit, input reaction conditions including the formulation and heating temperature of the thermosetting resin to be estimated into the trained model and obtain the estimated degree of curing based on the reaction conditions, and a curing curve calculation step in which the curing curve of the thermosetting resin at a set heating temperature is calculated based on the estimated curing curve obtained in the reaction time acquisition step.

[0015] Furthermore, the estimation method according to the present invention includes, in the above invention, a reaction time acquisition step of inputting a plurality of reaction conditions with different heating temperatures into the learned model and acquiring a curing curve estimated from each reaction condition, and a curing curve calculation step of a first calculation step of applying the Arrhenius law to the curing curve for each heating temperature to calculate the activation energy and frequency factor at each heating temperature, and a second calculation step of using the activation energy and frequency factor calculated in the first calculation step to calculate a curing curve at a set heating temperature.

[0016] Furthermore, the trained model generation program according to the present invention is a trained model generation program that causes a computer to generate a trained model for estimating the curing curve of a thermosetting resin, and causes the computer to execute the following steps: a dataset generation step which generates a plurality of datasets, each of which sets the formulation and heating temperature of the thermosetting resin and the curing curve showing the time change in the degree of curing at the formulation and heating time; and a trained model training step which uses the plurality of datasets and trains a trained model by training with the formulation and heating temperature of the thermosetting resin as explanatory variables and the curing curve at the formulation and heating time as the target variable.

[0017] Furthermore, the estimation program according to the present invention is an estimation program that causes a computer to estimate the curing curve of a thermosetting resin, and causes the computer to execute the following steps: a reaction time acquisition step in which the formulation and heating temperature of the thermosetting resin to be estimated are input to the trained model, and a reaction time acquisition step in which the estimated curing degree is obtained based on the reaction time conditions, and a curing curve calculation step in which the curing curve of the thermosetting resin at a set heating temperature is calculated based on the estimated curing curve obtained in the reaction time acquisition step. [Effects of the Invention]

[0018] According to the present invention, it is possible to predict with high accuracy the curing curve, which shows the change in the degree of curing over time. [Brief explanation of the drawing]

[0019] [Figure 1] Figure 1 is a diagram showing a schematic configuration of an estimation system according to one embodiment of the present invention. [Figure 2] Figure 2 is a block diagram showing the configuration of a learning device included in an estimation system according to one embodiment of the present invention. [Figure 3] Figure 3 shows an example of a hardening curve that illustrates the degree of hardening over time. [Figure 4] FIG. 4 is a block diagram showing the configuration of an estimation device included in an estimation system according to an embodiment of the present invention. [Figure 5] FIG. 5 is a flowchart showing an outline of learning processing performed by a learning device according to an embodiment of the present invention. [Figure 6] FIG. 6 is a flowchart for explaining estimation processing performed by an estimation device according to an embodiment of the present invention. [Figure 7] FIG. 7 is a diagram for explaining an Arrhenius plot calculated when estimating a curing curve. [Figure 8] FIG. 8 is a diagram showing a curing curve calculated from the Arrhenius plot shown in FIG. 7. [Figure 9] FIG. 9 is a flowchart for explaining learning processing performed by a learning device according to a modification.

Embodiments for Carrying Out the Invention

[0020] Hereinafter, embodiments of a learned model generation method, an estimation method, a learned model generation program, and an estimation program according to the present invention will be described in detail based on the drawings. Note that the present invention is not limited by this embodiment. In addition, the individual embodiments of the present invention are not independent, and can be appropriately implemented in combination with each other.

[0021] (Embodiment) Figure 1 is a diagram illustrating the schematic configuration of an estimation system according to one embodiment of the present invention. The estimation system 1 shown in these figures comprises a learning device 2 that creates training data and generates a trained model using the created training data, an estimation device 3 that estimates the characteristic values ​​of the target to be estimated using the trained model generated by the learning device 2, a display device 4 that displays information including the estimation results of the estimation device 3, and an input device 5. The learning device 2, estimation device 3, display device 4, and input device 5 are connected to each other via a communication network. The communication network referred to here is configured using, for example, an existing public telephone network, LAN (Local Area Network), WAN (Wide Area Network), etc., and can be wired or wireless.

[0022] The characteristic estimated by the estimation device 3 is the curing curve of a thermosetting resin, which shows the change in the degree of curing with respect to the heating time (time elapsed since the start of heating) at a predetermined heating temperature.

[0023] The thermosetting resin in this embodiment is not particularly limited as long as it is a resin that hardens when heated. Examples of thermosetting resins include phenolic resins, epoxy resins, melamine resins, and polyurethane resins. These are generally used in combination with curing agents or curing catalysts. Furthermore, thermoplastic resins or additives may be mixed in as appropriate.

[0024] In this embodiment, epoxy resin is preferred as the thermosetting resin because it offers an excellent balance of heat resistance and mechanical properties, and its mechanical properties and curing time can be easily altered by selecting the resin to combine it with. Such epoxy resin is not particularly limited as long as it is a compound having epoxy groups in its molecule. Examples of epoxy resins include bisphenol A type epoxy resin, bisphenol F type epoxy resin, amine type epoxy resin, dicyclopentadiene type epoxy resin, and biphenyl type epoxy resin.

[0025] Commercially available bisphenol A type epoxy resins that can be used include "jER(registered trademark)" 825, "jER(registered trademark)" 828 (both manufactured by Mitsubishi Chemical Corporation), "Epotote(registered trademark)" YD-128, "Epotote(registered trademark)" YD-8125 (both manufactured by Nippon Steel Chemical & Material Co., Ltd.), and "DER(registered trademark)" 331, "DER(registered trademark)" 332 (both manufactured by Dow Chemical Ltd.).

[0026] Commercially available bisphenol F type epoxy resins include, for example, "jER(registered trademark)" 806, "jER(registered trademark)" 807, "jER(registered trademark)" 4004P (all manufactured by Mitsubishi Chemical Corporation), "EPICLON(registered trademark)" 830 (manufactured by DIC Corporation), "Epotote(registered trademark)" YD-170, "Epotote(registered trademark)" YDF-8170C (all manufactured by Nippon Steel Chemical & Material Co., Ltd.).

[0027] Examples of amine-type epoxy resins used in this embodiment include tetraglycidyldiaminodiphenylmethane, triglycidylaminophenol, and diglycidylaniline. Specific examples of such epoxy resins are listed below.

[0028] Commercially available tetraglycidyldiaminodiphenylmethane products include "SumiEpoxy®" ELM434 (manufactured by Sumitomo Chemical Co., Ltd.), YH434L (manufactured by Nippon Steel Chemical & Material Co., Ltd.), "jER®" 604 (manufactured by Mitsubishi Chemical Corporation), "Araldite®" MY720, and "Araldite®" MY721 (all manufactured by Huntsman Advanced Materials).

[0029] Commercially available triglycidylaminophenol products that can be used include "SumiEpoxy®" ELM100, "SumiEpoxy®" ELM120 (both manufactured by Sumitomo Chemical Co., Ltd.), "Araldite®" MY0500, "Araldite®" MY0510, "Araldite®" MY0600 (all manufactured by Huntsman Advanced Materials), and "jER®" 630 (manufactured by Mitsubishi Chemical Corporation).

[0030] Commercially available diglycidylaniline products include GAN (N,N-diglycidylaniline), GOT (N,N-diglycidyl-o-toluidine) (both manufactured by Nippon Kayaku Co., Ltd.), and "TOREP®" A-204E (diglycidyl-p-phenoxyaniline) (manufactured by Toray Fine Chemicals Co., Ltd.).

[0031] Examples of dicyclopentadiene-type epoxy resins include "EPICLON®" HP-7200L, "EPICLON®" HP-7200, "EPICLON®" HP-7200H, and "EPICLON®" HP-7200HH (all manufactured by DIC Corporation). Examples of biphenyl-type epoxy resins include "jER®" YX-4000 (manufactured by Mitsubishi Chemical Corporation). Examples of biphenyl aralkyl-type epoxy resins include NC-3000H, NC-3000, and NC-3000L (all manufactured by Nippon Kayaku Co., Ltd.).

[0032] Compounds capable of curing epoxy resins, i.e., curing agents, include stoichiometric compounds such as dicyandiamide, aromatic amines, aliphatic amines, and hydrazides, and catalytic compounds such as aromatic ureas, imidazoles, and Lewis acid complexes. Among curing agents, aromatic amines and dicyandiamide are preferably used because they produce fiber-reinforced plastics with excellent heat resistance and mechanical properties.

[0033] Aromatic amines are compounds in which an amino group is directly attached to an aromatic ring. Examples of aromatic amines include 4,4'-diaminodiphenylmethane, 4,4'-diaminodiphenylsulfone, 3,3'-diaminodiphenylsulfone, diphenylmethane-type amines, diethyltoluenediamine, and dimethylthiotoluenediamine. Commercially available aromatic amines that can be used include Seika Cure-S (manufactured by Seika Co., Ltd.), 3,3'DAS (manufactured by Mitsui Chemicals Fine, Inc.), "Lonzacure®" M-MIPA, "Lonzacure®" M-DEA, "Lonzacure®" M-CDEA, "Lonzacure®" M-DIPA (all manufactured by Lonza), "KAYAHARD®" AA (manufactured by Nippon Kayaku Co., Ltd.), "jER Cure®" W (manufactured by Mitsubishi Chemical Corporation), and "Ethacure®" 300 (manufactured by Albemarle).

[0034] The learning device 2 is electrically connected to the estimation device 3. The learning device 2 generates a dataset for training and generates and outputs a trained model by training using the generated dataset. Figure 2 is a block diagram showing the configuration of the learning device in the estimation system according to an embodiment of the present invention. The learning device 2 has a dataset generation unit 21, a learning unit 22, a control unit 23, and a storage unit 24.

[0035] The dataset generation unit 21 generates datasets consisting of a resin formulation, heating temperature, and curing curve at that heating temperature. The dataset generation unit 21 generates multiple datasets by, for example, reading various data from the storage unit 24. In this embodiment, the dataset generation unit 21 generates datasets for the same thermosetting resin formulation, consisting of a thermosetting resin formulation and curing curve for three different heating temperatures. The curing curves in these datasets are preferably measured data obtained through tests or the like, but they may also be data generated through analysis such as simulations.

[0036] Figure 3 shows an example of a curing curve that illustrates the degree of curing over time. The curing curve shown in Figure 3 shows an example of the change in the degree of curing (cd(%)) over time at heating temperatures of 140°C, 160°C, and 180°C for a certain thermosetting resin. As shown in Figure 3, even with the same formulation, the change in the degree of curing over time differs depending on the heating temperature.

[0037] Furthermore, the data included in the dataset may include conditions such as heating temperature changing in steps, pressure (MPa), and type of equipment.

[0038] The learning unit 22 generates a trained model by performing training using the dataset generated by the dataset generation unit 21. For example, the learning unit 22 reads the dataset from the memory unit 24 and generates a trained model by training with the thermosetting resin formulation and heating temperature as explanatory variables and the curing curve as the target variable. The training performed by the learning unit 22 can employ known machine learning methods such as Bayesian optimization, genetic algorithms, linear regression, nonlinear regression, multivariate regression, logistic regression, k-nearest neighbors, support vector machines, decision trees, random forests, gradient boosting trees, k-means algorithms, principal component analysis, and deep learning. The trained model is, for example, a neural network consisting of an input layer, hidden layers, and output layers, with each layer having one or more nodes. Information such as network parameters in the trained model is stored in the memory unit 24. Network parameters include information on the weights and biases between layers of the neural network.

[0039] Furthermore, when the learning unit 22 generates a trained model, for example, by learning using regularization, it is given multiple candidate values ​​for the hyperparameters of the regression model and performs learning for each of the given candidate values. After that, it calculates the prediction error using cross-validation or holdout validation with the training data for the model obtained by learning with each candidate value, and selects the regression model that gives the smallest prediction error. The selected regression model is output as the trained model. Here, hyperparameters refer to, for example, the number of layers in the neural network or the regularization coefficients.

[0040] The control unit 23 comprehensively controls the operation of the learning device 2.

[0041] The memory unit 24 stores data including various programs for operating the learning device 2, and various parameters necessary for the operation of the learning device 2. The various programs include a learning data generation program that generates training data for generating a trained model, and a trained model generation program that uses the training data to train and generate a trained model. The memory unit 24 also has a dataset storage unit 241 that stores multiple datasets.

[0042] The memory unit 24 is composed of a ROM (Read Only Memory) on which various programs are pre-installed, and RAM (Random Access Memory), HDD (Hard Disk Drive), SSD (Solid State Drive), etc., which store calculation parameters and data for each process.

[0043] Various programs can be recorded on computer-readable recording media such as HDDs, flash memory, CD-ROMs, DVD-ROMs, and Blu-ray® discs and widely distributed. Furthermore, the learning device 2 can acquire various programs via a communication network.

[0044] The learning device 2 having the above functional configuration is a computer composed of one or more hardware components such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), and an FPGA (Field Programmable Gate Array).

[0045] The estimation device 3 is electrically connected to the learning device 2 and the display device 4. The estimation device 3 outputs estimation results that include the conditions of the thermosetting resin to be estimated (formulation and heating temperature) and the characteristics (in this case, curing curve) estimated using a trained model acquired from the learning device 2. Figure 4 is a block diagram showing the configuration of the estimation device in the estimation system according to an embodiment of the present invention. The estimation device 3 has a calculation unit 31, a control unit 32, and a storage unit 33.

[0046] The calculation unit 31 calculates the curing curve of the thermosetting resin at a set heating temperature, which is estimated using the conditions of the target to be estimated obtained from the input device 5 and the trained model obtained from the learning device 2.

[0047] The control unit 32 comprehensively controls the operation of the estimation device 3. The control unit 32 includes a display control unit 321 that displays the calculation result (estimation result) of the calculation unit 31 on the display device 4. In addition to the estimation result, the display control unit 321 may also display information about the target of estimation (for example, details of the formulation or manufacturing conditions) on the display device 4.

[0048] The memory unit 33 stores data including various programs for operating the estimation device 3, and various parameters necessary for the operation of the estimation device 3. The various programs include estimation programs that are executed using trained models. The memory unit 33 is composed of a ROM with various programs pre-installed, and RAM, HDD, SSD, etc., for storing calculation parameters and data for each process.

[0049] Various programs can be recorded on computer-readable recording media such as HDDs, flash memory, CD-ROMs, DVD-ROMs, and Blu-ray® discs and widely distributed. Furthermore, the estimation device 3 can acquire various programs via a communication network.

[0050] The estimation device 3, having the above functional configuration, is a computer composed of one or more hardware components such as a CPU, GPU, ASIC, and FPGA.

[0051] The display device 4 is a display made of liquid crystal or organic EL (Electro-Luminescence), and is electrically connected to the estimation device 3. The display device 4 acquires and displays display data output from the estimation device 3 under the control of the display control unit 321. The display device 4 may also have an audio output function such as a speaker.

[0052] Input device 5 accepts various types of information, including settings related to the process of estimating characteristic values, and outputs the received information to learning device 2 and estimation device 3. Input device 5 is configured using a user interface such as a keyboard, mouse, microphone, and touch panel.

[0053] Next, the processes of the learning device 2 and the estimation device 3 will be described with reference to Figures 5 and 6. Figure 5 is a flowchart outlining the learning process performed by the learning device according to this embodiment. In the learning process shown in Figure 5, an example is described in which a trained model is generated using multivariate regression with at least the formulation of the thermoexchangeable resin and the heating temperature as explanatory variables, but it is also possible to generate a trained model using other methods.

[0054] First, the learning device 2 acquires learning conditions via the input device 5 (step S11). The learning conditions input here are conditions related to the curing curve of the thermosetting resin to be estimated. The learning conditions may include the formulation of the thermosetting resin, the heating temperature, and conditions related to the reaction (curing), such as the type of heating means.

[0055] The learning device 2 generates multiple datasets based on the learning conditions input via the input device 5 (step S12). The dataset generation unit 21 generates a dataset from the data stored in the storage unit 24, for example, based on the input learning conditions, consisting of a set of the thermosetting resin formulation, heating temperature, and curing curve at that heating temperature. For a single formulation, multiple (e.g., three or more) datasets of heating temperatures are generated.

[0056] Then, the learning unit 22 divides the multiple datasets into a training dataset and a validation dataset (step S13). The learning unit 22 distributes the datasets into training and validation in a predetermined ratio. For example, the distribution condition is that the number of training datasets is greater than the number of validation datasets, and the distribution is done randomly in this ratio.

[0057] After the dataset allocation, the learning unit 22 generates a trained model including interaction terms using the training dataset (step S14). For example, if the target variable is Y and the explanatory variables are x1, x2, ..., the regression model is expressed as Y = ax1 + bx2 + ... (where a, b, ... are regression coefficients). In this case, as an interaction term, for example, the ax1 * bx2 term, which is obtained by multiplying the x1 term and the x2 term by the above equation, is added. Furthermore, in this process, regularization methods such as Lasso regression may be used to eliminate explanatory variables with relatively low correlation. This allows, for example, interaction terms with high correlation to be selected.

[0058] Then, the learning unit 22 outputs the generated trained model (step S15). At this time, the learning unit 22 outputs the generated trained model to the memory unit 24 and the estimation device 3. For example, when a trained model is input to the memory unit 24, the memory unit 24 stores the trained model.

[0059] When a new curing curve is obtained through testing or other means, the trained model generated in this way is evaluated by comparing the measured curing curve with a predicted curing curve obtained by inputting the formulation and heating temperature of the thermosetting resin into the trained model. If there is a large discrepancy between the measured curing curve and the predicted curing curve, a new trained model is generated by retraining using a dataset that includes the measured curing curve.

[0060] Estimation device 3 estimates the curing curve of the thermosetting resin to be estimated. Figure 6 is a flowchart illustrating the estimation process performed by the estimation device according to this embodiment. The calculation unit 31 uses the trained model generated by the learning device 2 to calculate the curing curve of the thermosetting resin to be estimated at a predetermined heating temperature.

[0061] When the estimation device 3 receives an instruction to estimate the curing curve of a thermosetting resin, the control unit 32 first obtains the estimation conditions for the thermosetting resin via the input device 5 (step S21). The estimation conditions include the heating temperature to be estimated for the target thermosetting resin.

[0062] The calculation unit 31 obtains the curing curve of the thermosetting resin to be estimated using the trained model generated in step S13 (step S22). The calculation unit 31 inputs the formulation of the thermosetting resin to be estimated and the heating temperature into the trained model and obtains the curing curve. In this case, the number of heating temperatures may be based on a preset number, or it may be a number instructed, for example, via the input device 5. As for the number of heating temperatures, at least 3 heating temperatures are set for one formulation. There should be 3 or more heating temperatures, and preferably 3 to 15. It is preferable that there be 4 to 6 heating temperatures, as this number can suppress the computational load while improving prediction accuracy. In addition, the heating temperatures may be based on a preset temperature, or they may be a temperature instructed, for example, via the input device 5. In step S22, for the thermosetting resin to be estimated, predicted curing curves at multiple heating temperatures, such as those shown in Figure 3, are output.

[0063] The calculation unit 31 then applies the Arrhenius law to the curing curve for each heating temperature to calculate the activation energy and frequency factor for each heating temperature (step S23: first calculation step). At this time, the calculation unit 31 plots the natural logarithm of the curing degree set for each heating temperature on a coordinate space where the natural logarithm of the curing degree (In(Cd / t)) is on the vertical axis and the reciprocal of the temperature (1000 / T; T is the absolute temperature (K)) is on the horizontal axis, and generates an approximate straight line for this plot. The slope of the approximate straight line generated at this time is -E a The activation energy E is calculated from / R, and the frequency factor A is calculated from the intercept InA. The degree of hardening in this case may be a predetermined degree of hardening (e.g., cd 50%), or it may be a degree of hardening instructed via, for example, the input device 5.

[0064] Figure 7 illustrates the Arrhenius plot calculated when estimating the hardening curve. Figure 7 shows approximate straight lines plotted from the hardening curves for three heating temperatures (140°C, 160°C, and 180°C) based on each hardening degree (20%, 30%, 50%, 80%, 90%, and 95%). The activation energy and frequency factor are calculated from the slope and intercept of each approximate straight line.

[0065] Subsequently, the calculation unit 31 calculates the curing curve at the set heating temperature using the calculated activation energy and frequency factor (step S24: second calculation step). The calculation unit 31 calculates the curing curve at the set heating temperature based on the activation energy and frequency factor.

[0066] Figure 8 shows the hardening curve calculated from the Arrhenius plot shown in Figure 7. Figure 8 shows an example of estimating the hardening curve at a heating temperature of 150.2°C (1000 / T ≈ 2.36) based on the activation energy and frequency factor obtained from the approximation line shown in Figure 7. The calculation unit 31 determines the heating time (elapsed time from the start of heating) for each degree of hardening (20%, 30%, 50%, 80%, 90%, 95%) from the activation energy and frequency factor, calculates an approximation curve from the plot, and uses this as the hardening curve. In this way, it is possible to calculate hardening curves for heating temperatures that did not exist in the dataset.

[0067] Next, the display control unit 321 outputs the estimation result from the calculation unit 31 to the display device 4 and performs display control to display the estimation result on the display device 4 (step S25). The display device 4 displays, for example, the curing curve of the thermosetting resin to be estimated at a predetermined heating temperature.

[0068] In the embodiments described above, regarding the formulation of a thermosetting resin, a trained model obtained by learning a dataset consisting of heating temperature and curing curve can be used to accurately estimate the curing curve, which shows the change in the degree of curing over time from the start of heating, for the formulation of the thermosetting resin to be estimated.

[0069] Furthermore, in this embodiment, a curing curve is estimated by obtaining curing curves for multiple heating temperatures using a trained model and using the activation energy and frequency factor obtained from these curing curves. According to this embodiment, since the curing curve for the heating temperature to be estimated is calculated based on the curing patterns of multiple heating temperatures for the thermosetting resin to be estimated, the curing curve can be estimated with even greater accuracy compared to, for example, simply outputting the curing curve directly from a trained model. In this case, if the formulation of the thermosetting resin to be estimated and its heating temperature exist as a dataset, it is not necessary to perform the processing in steps S22 to S24 shown in Figure 6, and the corresponding dataset may be read and output. This embodiment is particularly effective for estimating curing curves for thermosetting resins for which measured curing curves have not been obtained, or for formulations of unknown thermosetting resins that are not yet known.

[0070] In the above-described embodiment, an example was explained in which a dataset is extracted by the dataset generation unit 21. However, if there is no need for extraction processing, such as using the dataset stored in the storage unit 24 as is, the system may be configured in a way that does not perform extraction processing by the dataset generation unit 21 (a configuration without the dataset generation unit 21).

[0071] (modified version) Next, a modified example of the embodiment of the present invention will be described with reference to Figure 9. The estimation system according to the modified example is the same as the estimation system 1 according to the embodiment. The following describes the processing content that differs from the embodiment.

[0072] In this modified example, the learning unit 22 generates a trained model by selecting the optimal statistical model from a plurality of different statistical models.

[0073] Figure 9 is a flowchart illustrating the learning process performed by the modified learning device. First, the learning device 2 generates multiple datasets based on the acquired learning conditions, in the same manner as in steps S11 to S13, and then divides the multiple datasets into a training dataset and a validation dataset (steps S31 to S33).

[0074] After the dataset is sorted, the learning unit 22 uses the training dataset to generate trained models using multiple different statistical models (step S34).

[0075] The learning unit 22 then selects the optimal trained model from the trained models generated by each statistical model (step S35). In this selection, cross-validation is used to evaluate the accuracy of the trained model using a validation dataset. The learning unit 22 calculates the prediction error by cross-validation by comparing the output value (hardened curve) obtained by inputting the explanatory variables of the validation dataset to the trained model obtained by training with each statistical model, and the target variable of the validation dataset, and selects the trained model (statistical model) that gives the smallest prediction error. If there are multiple models with the smallest prediction error, the learning unit 22 can select the statistical model that satisfies the pre-set conditions, such as the maximum or minimum value. Alternatively, the learning unit 22 may extract the trained model that gives the smallest prediction error and display the extraction result on the display device 4, allowing the user to select a trained model that meets the desired conditions.

[0076] After selecting a model, the learning unit 22 outputs the selected trained model to the memory unit 24 and the estimation device 3. The estimation process using this pre-trained model is the same as in the embodiment.

[0077] In the modified examples described above, similar to the embodiments, a trained model obtained by learning a dataset of heating temperature and curing curves for the thermosetting resin formulation can accurately estimate the curing curve showing the change in curing degree over time from the start of heating for the target thermosetting resin formulation. Furthermore, since the curing curve for the heating temperature to be estimated is calculated based on curing patterns at multiple heating temperatures for the target thermosetting resin, the curing curve can be estimated with even greater accuracy compared to, for example, simply outputting the curing curve directly from the trained model.

[0078] Furthermore, in this modified example, the optimal model is selected from multiple statistically generated trained models, thus enabling the acquisition of a trained model that can predict with even greater accuracy.

[0079] (Other embodiments) While embodiments for carrying out the present invention have been described so far, the present invention should not be limited to the embodiments described above. For example, the estimation device may also include the function of a learning unit. In this case, in addition to calculating the hardening curve of the target to be estimated, the estimation device sequentially updates the trained model. [Explanation of Symbols]

[0080] 1 Estimation System 2 Learning device 3 Estimation device 4 Display device 5 Input devices 21 Dataset Generation Unit 22 Learning Department 23, 32 Control Unit 24, 33 Storage section 31 Calculation Section 241 Dataset Storage Unit 321 Display Control Unit

Claims

1. A method for generating a trained model in which a computer generates a trained model for estimating the curing curve of a thermosetting resin, A dataset generation step that generates multiple datasets, each dataset comprising a combination of the formulation and heating temperature of the thermosetting resin and a curing curve showing the time change in the degree of curing for the formulation and heating time; A learning step in which a trained model is generated by learning using the above-mentioned multiple datasets, with the formulation and heating temperature of the thermosetting resin as explanatory variables and the curing curve at the said formulation and heating time as the objective variable, A method for generating a pre-trained model that includes this.

2. The curing curve shows the change in the degree of curing from the start of heating to the time when a predetermined degree of curing is reached. The method for generating a trained model according to claim 1.

3. The aforementioned learning steps are: A distribution step that divides the aforementioned dataset into a training dataset and a validation dataset, A model generation step involves applying a training dataset to multiple different statistical models to generate a trained model corresponding to each statistical model, A selection step involves inputting explanatory variables from a validation dataset into the pre-trained models corresponding to each statistical model, calculating the prediction error by cross-validation based on the output values ​​obtained, and selecting the statistical model that gives the smallest prediction error. A method for generating a trained model according to claim 1, including the following:

4. The learning step involves generating the trained model by machine learning or multivariate regression analysis. The method for generating a trained model according to claim 1.

5. The aforementioned learning step generates the trained model, which includes an interaction term between the type and amount of resin included in the dataset, generated by multivariate regression analysis. The method for generating a trained model according to claim 4.

6. The formulation of the thermosetting resin comprises an epoxy resin and a curing agent component that reacts with the epoxy groups contained in the epoxy resin. A method for generating a trained model according to any one of claims 1 to 5.

7. A computer estimation method for estimating the curing curve of a thermosetting resin, A reaction time acquisition step involves reading a trained model generated by training multiple datasets in which the formulation and heating temperature of the thermosetting resin are explanatory variables and the curing curve showing the time change in curing degree for the formulation and heating time is the objective variable, reading the trained model from the storage unit, inputting reaction conditions including the formulation and heating temperature of the thermosetting resin to be estimated into the trained model, and obtaining the estimated curing degree based on the reaction conditions. A curing curve calculation step, which calculates the curing curve of the thermosetting resin at a set heating temperature based on the estimated curing curve obtained in the reaction time acquisition step, An estimation method that includes [this].

8. The reaction time acquisition step is, Multiple reaction conditions with different heating temperatures are input into the trained model, and the hardening curves estimated from each reaction condition are obtained. The hardening curve calculation step is: The first calculation step involves applying the Arrhenius rule to the hardening curve for each heating temperature to calculate the activation energy and frequency factor at each heating temperature. A second calculation step involves calculating a hardening curve at a set heating temperature using the activation energy and frequency factor calculated in the first calculation step, The estimation method according to claim 7, including the following:

9. A pre-trained model generation program that causes a computer to generate a pre-trained model for estimating the curing curve of a thermosetting resin, A dataset generation step that generates multiple datasets, each dataset comprising a combination of the formulation and heating temperature of the thermosetting resin and a curing curve showing the time change in the degree of curing for the formulation and heating time; Using the aforementioned multiple datasets, a trained model is created by learning with the formulation and heating temperature of the thermosetting resin as explanatory variables and the curing curve for the formulation and heating time as the objective variable, as the training step. A trained model generation program that causes the aforementioned computer to execute.

10. An estimation program that causes a computer to estimate the curing curve of a thermosetting resin, A reaction time acquisition step involves reading a trained model generated by training multiple datasets in which the formulation and heating temperature of the thermosetting resin are explanatory variables and the curing curve showing the time change in curing degree for the formulation and heating time is the objective variable, reading the trained model from the storage unit, inputting reaction conditions including the formulation and heating temperature of the thermosetting resin to be estimated into the trained model, and obtaining the estimated curing degree based on the reaction conditions. A curing curve calculation step, which calculates the curing curve of the thermosetting resin at a set heating temperature based on the estimated curing curve obtained in the reaction time acquisition step, An estimation program that causes the aforementioned computer to execute.

Images