Learning data generation method, learned model generation method, characteristic value estimation method, learning data generation program, learned model generation program, and characteristic value estimation program

Abstract

Provided are a learning data generation method, a learned model generation method, a characteristic value estimation method, a learning data generation program, a learned model generation program, and a characteristic value estimation program capable of predicting a characteristic value of a fiber-reinforced resin molded article with high accuracy in a short time.SOLUTION: A learning data generation method according to the present invention includes a learning data generation step of reading, from a storage unit, a plurality of data sets each including a material condition, a molding condition, and a structural condition of a molded fiber-reinforced plastic article and a characteristic value of the molded fiber-reinforced plastic article manufactured under the material condition, the molding condition, and the structural condition, and generating learning data using the plurality of data sets, wherein an item belonging to the structural condition includes interlayer information of the molded fiber-reinforced plastic article, and the interlayer information includes two or more of presence or absence of an interlayer reinforcement material, geometric information of the interlayer reinforcement material, and geometric information of an interlayer resin.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a training data generation method, a trained model generation method, a characteristic value estimation method, a training data generation program, a trained model generation program, and a characteristic value estimation program. [Background technology]

[0002] Fiber-reinforced plastic (FRP) molded products are often made by laminating prepregs, which are sheets of reinforcing fibers aligned in one direction and then impregnating them with resin, in multiple directions, and then molding them by heating and pressurizing them. Incidentally, characteristic values ​​such as the amount of deformation of a fiber-reinforced plastic molded product are predicted by, for example, simulating the type of material (see, for example, Non-Patent Documents 1 to 3). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] OG Kravchenko et al. / Composites: Part A 99 (2017) 186-197 [Non-patent document 2] Panasonic Technical Journal Vol.55 No.4 Jan.2010 [Non-patent document 3] DENSO Technical Review Vol.4 No.1 1999 Summary of the Invention [Problem to be solved by the invention]

[0004] However, conventional simulations have the problem that calculations are complex and time-consuming, especially for laminated types such as the fiber-reinforced plastic molded products mentioned above. Also, high accuracy is required for the prediction results.

[0005] The present invention has been made in consideration of the above, and aims to provide a training data generation method, a trained model generation method, a characteristic value estimation method, a training data generation program, a trained model generation program, and a characteristic value estimation program that can predict the characteristic values ​​of fiber-reinforced plastic molded products with high accuracy in a short period of time. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems and achieve the objective, the training data generation method of the present invention is a training data generation method in which a computer generates training data for estimating characteristic values ​​of a fiber-reinforced plastic molded product, and includes a training data generation step of reading from a storage unit a plurality of data sets each including the material conditions, molding conditions, and structural conditions of the fiber-reinforced plastic molded product, as well as the characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and generating training data using the plurality of data sets, wherein items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, and the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin.

[0007] In the training data generating method according to the present invention, the characteristic values ​​are deformation characteristic values ​​or physical characteristic values.

[0008] In the learning data generating method according to the present invention, the items belonging to the structural conditions include geometric information on fiber arrangement.

[0009] In addition, in the training data generation method according to the present invention, the geometric information of the fiber arrangement includes one or more of the fiber volume content of the fiber-reinforced plastic molded product, the fiber volume content within a layer, and the degree of fiber distribution.

[0010] In the learning data generating method according to the present invention, the items belonging to the structural conditions include non-destructive inspection data.

[0011] In the learning data generation method according to the present invention, the characteristic values ​​included in the data set include values ​​measured on a fiber-reinforced plastic molded product having an asymmetric laminate structure.

[0012] In the learning data generating method according to the present invention, the measurement accuracy of the characteristic values ​​in an asymmetric laminated structure is higher than the measurement accuracy of the characteristic values ​​in a symmetric laminated structure.

[0013] Further, a trained model generation method according to the present invention is a trained model generation method in which a computer generates a trained model for estimating characteristic values ​​of a fiber-reinforced plastic molded product, and includes: a training data generation step of reading from a storage unit a plurality of data sets each including the material conditions, molding conditions, and structural conditions of the fiber-reinforced plastic molded product, as well as the characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and generating training data using the plurality of data sets; and a trained model generation step of generating a trained model by learning the training data using the material conditions, molding conditions, and structural conditions of the fiber-reinforced plastic molded product as explanatory variables and the characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions as objective variables, wherein items belonging to the structural conditions include interlayer information of the fiber-reinforced plastic molded product, and the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin.

[0014] In addition, the method for estimating characteristic values ​​of fiber-reinforced plastic molded products according to the present invention is a method for estimating characteristic values ​​of fiber-reinforced plastic molded products in which a computer estimates characteristic values ​​of a fiber-reinforced plastic molded product, and includes a calculation step of reading from a storage unit a plurality of data sets each containing material conditions, molding conditions, and structural conditions, as well as characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, to generate a trained model that has been trained using training data generated based on the plurality of data sets, and calculating the characteristic values ​​of the fiber-reinforced plastic molded product to be estimated based on the trained model, wherein items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, and the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin.

[0015] In addition, the training data generation program of the present invention is a training data generation program that causes a computer to generate training data for estimating characteristic values ​​of a fiber-reinforced plastic molded product, and causes the computer to execute a training data generation step of reading from a memory unit a plurality of data sets each including the material conditions, molding conditions, and structural conditions of the fiber-reinforced plastic molded product, as well as the characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and generating training data using the plurality of data sets, wherein items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, and the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin.

[0016] In addition, a 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 characteristic values ​​of a fiber-reinforced plastic molded product, and causes the computer to execute the following steps: a training data generation step of reading from a storage unit a plurality of data sets each including the material conditions, molding conditions, and structural conditions of the fiber-reinforced plastic molded product, as well as the characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and generating training data using the plurality of data sets; and a trained model generation step of generating a trained model by learning the training data using the material conditions, molding conditions, and structural conditions of the fiber-reinforced plastic molded product as explanatory variables and the characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions as objective variables, wherein the items belonging to the structural conditions include interlayer information of the fiber-reinforced plastic molded product, and the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin.

[0017] In addition, the characteristic value estimation program for fiber-reinforced plastic molded products according to the present invention is a characteristic value estimation program for fiber-reinforced plastic molded products that causes a computer to estimate the characteristic values ​​of a fiber-reinforced plastic molded product, and causes the computer to execute the following calculation steps: reading from a memory unit a trained model generated by learning a regression model using the multiple data sets generated by reading from the memory unit a trained model that includes the material conditions, molding conditions, and structural conditions of the fiber-reinforced plastic molded product, as well as the characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions; and calculating the characteristic values ​​of the fiber-reinforced plastic molded product to be estimated based on the trained model, wherein the items belonging to the structural conditions include interlayer information of the fiber-reinforced plastic molded product, and the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin. [Effects of the Invention]

[0018] According to the present invention, the characteristic values ​​of a fiber-reinforced plastic molded product can be predicted with high accuracy in a short time. [Brief explanation of the drawings]

[0019] [Figure 1] FIG. 1 is a diagram showing a schematic configuration of a characteristic value estimation system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a block diagram showing the configuration of a learning device included in a characteristic value estimation system according to an embodiment of the present invention. [Figure 3] Figure 3 shows a list of the contents of the dataset. [Figure 4] FIG. 4 is a diagram (part 1) for explaining an example of the temperature and pressure conditions (molding profile) among the molding conditions. [Figure 5] FIG. 5 is a diagram (part 2) for explaining an example of the temperature and pressure conditions (molding profile) among the molding conditions. [Figure 6] FIG. 6 is a diagram (part 3) for explaining an example of the temperature and pressure conditions (molding profile) among the molding conditions. [Figure 7] FIG. 7 is a block diagram showing the configuration of an estimation device included in a characteristic value estimation system according to an embodiment of the present invention. [Figure 8] FIG. 8 is a diagram for explaining the flow of the estimation process performed by the characteristic value estimation system according to an embodiment of the present invention. [Figure 9] FIG. 9 is a flowchart showing an outline of the learning process performed by the learning device according to an embodiment of the present invention. [Figure 10] FIG. 10 is a flowchart illustrating the estimation process performed by the estimation device according to an embodiment of the present invention. [Figure 11] FIG. 11 shows a list of the contents of the dataset. [Figure 12] FIG. 12 is a diagram showing a list of the extracted data sets. [Figure 13] FIG. 13 is a flowchart illustrating a learning data generation process performed by a learning device according to a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0020] Hereinafter, an embodiment of a characteristic value estimation system according to the present invention will be described in detail with reference to the drawings. Note that the present invention is not limited to this embodiment. Furthermore, the individual embodiments of the present invention are not independent and can be implemented in combination with each other as appropriate.

[0021] (Embodiment) 1A to 1C are diagrams showing a schematic configuration of a characteristic value estimation system according to an embodiment of the present invention. The characteristic value estimation system 1 shown in these figures includes a learning device 2 that creates learning data and generates a trained model trained using the created learning data, an estimation device 3 that estimates a characteristic value of an estimation target 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.

[0022] The characteristic value estimated by the estimation device 3 is a characteristic value of a fiber reinforced plastic molded product made using fiber reinforced plastics (FRP), such as the amount of thermal deformation. Fiber-reinforced resin molded products are made by laminating prepregs, which are sheets of reinforcing fibers aligned in one direction and impregnated with resin, in multiple directions, and then molding them by heating and pressurizing.

[0023] Examples of reinforcing fibers include glass fibers, carbon fibers, metal fibers, aromatic polyamide fibers, polyaramid fibers, alumina fibers, silicon carbide fibers, boron fibers, and basalt fibers. These may be used alone or in combination of two or more types. These reinforcing fibers may be surface-treated. Examples of surface treatments include metal deposition treatment, treatment with a coupling agent, treatment with a sizing agent, and treatment with an additive. In this specification, when reinforcing fibers have been surface-treated, the term "reinforcing fibers" includes fibers in a surface-treated state. These reinforcing fibers also include conductive reinforcing fibers. Carbon fibers are preferably used as reinforcing fibers because of their low specific gravity, high strength, and high elastic modulus. Commercially available carbon fibers include "TORAYCA (registered trademark)" T800G-24K, "TORAYCA (registered trademark)" T800S-24K, "TORAYCA (registered trademark)" T700G-24K, "TORAYCA (registered trademark)" T700S-24K, "TORAYCA (registered trademark)" T300-3K, and "TORAYCA (registered trademark)" T1100G-24K (all manufactured by Toray Industries, Inc.).

[0024] The form and arrangement of the reinforcing fibers can be appropriately selected from those in which the reinforcing fibers are arranged in one direction, a laminate of reinforcing fibers arranged in one direction, or a woven form. Fiber-reinforced plastic molded products can have a symmetrical laminate structure in which the fiber direction is symmetrical with respect to the central plane of the prepreg laminate, or an asymmetrical laminate structure in which the fiber direction is asymmetric with respect to the central plane. Here, the central plane corresponds to, for example, the plane between the third and fourth layers, or the face of the third layer facing the fourth layer, in a six-layer structure. The asymmetric laminated structure has a larger thermal deformation amount than the symmetric laminated structure.

[0025] As the resin, a thermoplastic resin, a thermosetting resin, a photocurable resin, or the like can be used.

[0026] Thermoplastic resins can be polymers with a carbon-carbon bond, amide bond, imide bond, ester bond, ether bond, carbonate bond, urethane bond, urea bond, thioether bond, sulfone bond, imidazole bond, or carbonyl bond in the main chain. Specific examples include polyacrylate, polyolefin, polyamide (PA), aramid, polyester, polycarbonate (PC), polyphenylene sulfide (PPS), polybenzimidazole (PBI), polyimide (PI), polyetherimide (PEI), polysulfone (PSU), polyethersulfone (PES), polyetherketone (PEK), polyetheretherketone (PEEK), polyetherketoneketone (PEKK), polyaryletherketone (PAEK), and polyamideimide (PAI). These can be polymers, or oligomers or monomers can be used for low viscosity and low-temperature application. Of course, depending on the purpose, these may be copolymerized, or various types may be mixed and used as a polymer blend / alloy.

[0027] Examples of thermosetting resins include epoxy resins, maleimide resins, polyimide resins, acetylene-terminated resins, vinyl-terminated resins, allyl-terminated resins, nadic acid-terminated resins, and cyanate ester-terminated resins. These resins can generally be used in combination with a curing agent or a curing catalyst. These thermosetting resins can also be used in combination as appropriate.

[0028] It is also suitable to use a mixture of a thermosetting resin and a thermoplastic resin. A mixture of a thermosetting resin and a thermoplastic resin provides better results than using a thermosetting resin alone. This is because thermosetting resins generally have the disadvantage of being brittle but can be molded at low pressures in an autoclave, while thermoplastic resins generally have the advantage of being tough but are difficult to mold at low pressures in an autoclave. These contradictory properties make it possible to balance physical properties and moldability by using a mixture of these resins. When using a mixture, it is preferable that the thermosetting resin be contained in an amount of more than 50 mass% from the viewpoint of mechanical properties.

[0029] The learning device 2 is electrically connected to the estimation device 3. The learning device 2 selectively extracts learning data, and generates and outputs a trained model by learning using the extracted learning data. FIG. 2 is a block diagram showing the configuration of a learning device provided in a characteristic value estimation system according to an embodiment of the present invention. The learning device 2 has an extraction unit 21, a learning data generation unit 22, a learning unit 23, a control unit 24, and a storage unit 25. In the first embodiment, the learning data generation device is configured by at least the extraction unit 21, the learning data generation unit 22, and the storage unit 25.

[0030] The extraction unit 21 extracts a data set from the data sets stored in the storage unit 25 based on the input conditions, and outputs the extracted data set to the learning data generation unit 22 as learning data.

[0031] The data set and its extraction will now be described. The data set includes material conditions for the material constituting the fiber-reinforced plastic molded product, molding conditions for producing the molded product, and structural conditions and characteristic values ​​for the molded product.

[0032] The extraction unit 21 reads out a plurality of data sets based on the fiber-reinforced plastic molded product to be estimated input via the input device 5, and extracts them according to the conditions. In this embodiment, the extraction unit 21 reads out a plurality of data sets that satisfy the material conditions, molding conditions, and structural conditions that are preset for the fiber-reinforced plastic molded product to be estimated input via the input device 5.

[0033] Fig. 3 is a diagram showing a list of the contents of the data set. Fig. 3 shows an example of a data set read by the extraction unit 21 that satisfies the molding conditions and structural conditions that are preset for the input fiber-reinforced plastic molded product. Fig. 3 shows an example of material conditions set as material conditions 1, 2, 3, 4, ..., molding conditions set as molding conditions 1, 2, 3, 4, ..., structural conditions set as structural conditions 1, 2, 3, 4, ..., and level 1, level 2, ..., level N-1, and level N, to which their characteristic values ​​are respectively associated.

[0034] Here, the material conditions include fiber type, resin type, compounding ratio, interlayer reinforcement type, CF density, resin density, interlayer reinforcement density, CF elastic modulus, resin elastic modulus, etc. As material conditions in the data set, any of the above types is assigned as material condition 1, 2, 3, 4, ..., and a value is set for each level.

[0035] The molding conditions include post-cure conditions, temperature rise temperature (cpm), temperature fall temperature (cpm), molding temperature (first stage), molding temperature (second stage), molding temperature (post-cure), molding pressure (MPa), post-cure pressure (MPa), molding equipment, and a molding profile including temperature and pressure conditions. Note that molding pressure is, for example, the difference from atmospheric pressure. As molding conditions in the data set, any of the above various types are assigned as molding conditions 1, 2, 3, 4, ..., and the corresponding values ​​are set for each level.

[0036] The molding profile includes, for example, a profile of molding temperature and molding pressure. 4 to 6 are diagrams for explaining examples of temperature and pressure conditions (molding profiles) among molding conditions. In FIGS. 4 to 6, solid lines indicate temperature profiles, and dashed lines indicate pressure profiles. FIGS. 4 to 6 explain straight cure (see FIG. 4), step cure (see FIG. 5), and precure + postcure (see FIG. 6). Note that the profiles shown in FIGS. 4 to 6 are examples, and other profiles can be set.

[0037] Straight cure has a one-stage temperature rise profile. Specifically, as shown in Figure 4, the temperature rises to a set temperature at a predetermined rate, is maintained at the set temperature for a predetermined time, and then is lowered. The pressure rises to a set pressure at a predetermined rate, is maintained at the set pressure for a predetermined time, and is then reduced. In this case, the pressure reaches the set pressure before reaching the set temperature, and can be reduced after the temperature drops below the set temperature.

[0038] Step cure has a two-stage temperature rise profile. Specifically, as shown in Figure 5, the temperature rises to a first set temperature at a predetermined rate, maintains the first set temperature for a predetermined time, then rises to a second set temperature (>first set temperature) at a predetermined rate, maintains the second set temperature for a predetermined time, and then drops. The pressure also rises to a set pressure at a predetermined rate, maintains the set pressure for a predetermined time, and then is reduced. In this case, the pressure can reach the set pressure before reaching the first set temperature, and then be reduced after the temperature drops below the second set temperature.

[0039] The pre-cure and post-cure process has a profile in which the temperature is raised, then lowered, and then raised again. Figure 6 shows an example of a profile in which the pre-cure is a step cure and the post-cure is a straight cure. Specifically, as shown in Figure 6, the temperature changes in two stages, similar to the step cure, and then changes in one stage, similar to the straight cure. Note that Figure 6 includes a first curing period T1 using the step cure and a second curing period T2 using the post-cure (straight cure). The pressure increases to a set pressure at a predetermined rate, and can be reduced after maintaining the set pressure for a predetermined time. Figure 6 illustrates a program that changes the pressure during the pre-cure (first curing period T1), but the pressure may also be changed during both the pre-cure (first curing period T1) and the post-cure (second curing period T2).

[0040] Structural conditions include FAW (basis weight), number of plies (number of layers), lamination configuration, plate thickness, interlayer information, geometric information on fiber arrangement, non-destructive testing data, etc. Interlayer information includes the presence or absence of interlayer reinforcement, geometric information on the interlayer reinforcement, and geometric information on the interlayer resin. Geometric information on the interlayer reinforcement includes the shape of the interlayer reinforcement and the average diameter of the interlayer reinforcement. Geometric information on the interlayer resin includes the thickness of the interlayer resin layer, the fiber volume content of the fiber-reinforced resin molded product, the fiber volume content within the layer, and the degree of fiber dispersion. Non-destructive testing data includes the void fraction, etc. As structural conditions in the dataset, one of the above various types is assigned as structural condition 1, 2, 3, 4, etc., and a value is set for each level.

[0041] In this embodiment, the extraction unit 21 extracts interlayer information as the structural conditions. That is, the items belonging to the structural conditions to be extracted include interlayer information. In this case, the interlayer information includes two or more of the presence or absence of interlayer reinforcing material, geometric information of the interlayer reinforcing material, and geometric information of the interlayer resin.

[0042] Preferably, the extracted structural conditions further include geometric information on fiber arrangement. The geometric information on fiber arrangement includes at least one of the fiber volume content of the fiber-reinforced plastic molded product, the fiber volume content in the layer, and the degree of fiber dispersion. The geometric information on fiber arrangement can improve the prediction accuracy of deformation characteristic values, such as the flexural modulus, which will be described later.

[0043] The characteristic values ​​include deformation characteristic values ​​and physical characteristic values. The deformation characteristic values ​​include longitudinal modulus E1 (GPa) in the 0-degree direction, longitudinal modulus E2 (GPa) in the 90-degree direction, flexural modulus, thermal deformation (mm), linear expansion coefficient α (1 / K) in the 90-degree direction, cure shrinkage strain β, transverse modulus, deformation resistance, and fracture strain. The physical characteristic values ​​include glass transition temperature Tg (°C), DoC (%), electrical conductivity, thermal conductivity, degree of hardening, and density. In this case, the characteristic values ​​include values ​​measured in fiber-reinforced plastic molded products with an asymmetric laminate structure. Asymmetric laminate structures have a larger amount of thermal deformation than symmetric laminate structures. In the case of deformation characteristic values, the measurement accuracy of the characteristic values ​​in an asymmetric laminate structure is higher than that in a symmetric laminate structure. This difference in measurement accuracy occurs because the magnitude of the deformation characteristic value is sufficiently large compared to the resolution (the maximum level of detail that can be measured) of the equipment used to measure the displacement, as well as the error (manufacturing error, measurement error) and variability.

[0044] Here, we will explain how to measure the thermal deformation of a fiber-reinforced plastic molded product. Hereinafter, we will explain how to prepare a molded product and have the thermal deformation occur in the molded product. In a molded product in which thermal deformation occurs, the thermal deformation W is the distance from a line passing through both ends to the top of the convex. The molded product is composed of one 0-degree layer and four 90-degree layers whose fiber direction is at 90 degrees to the fiber direction of the 0-degree layer. This molded product has an asymmetric laminated structure in which the fiber directions of the layers are asymmetric with respect to the center plane.

[0045] First, a prepreg is made by laminating reinforcing fibers impregnated with resin in the above-mentioned layer relationship, attaching it to a plate, covering it with a film or the like, sealing it, and then heating and pressurizing it in an autoclave to form it. The structural conditions for this process include the number of plies (layers) and the layering configuration. Furthermore, the molding conditions include, for example, the molding profile described above. After removing any resin flash or flowing 90-degree layers, the thermal deformation is measured at room temperature (23°C) and at high temperature (80°C) using a universal testing machine / thermostat.

[0046] The learning data generation unit 22 generates learning data using the data set extracted by the extraction unit 21.

[0047] The learning unit 23 performs learning using the learning data generated by the learning data generation unit 22 to generate a trained model. The trained model is a neural network consisting of an input layer, an intermediate layer, and an output layer, with each layer having one or more nodes. The trained model is generated by training. Information such as network parameters in the trained model is stored in the storage unit 25. The network parameters include information regarding the weights and biases between layers of the neural network.

[0048] The learning performed by the learning unit 23 can employ a known learning method. For example, when generating a trained model by learning using regularization, multiple candidate values ​​for the hyperparameters of a regression model are provided, and learning is performed for each of the provided candidate values ​​for the hyperparameters. Then, for the model obtained by learning using each candidate value, prediction errors are calculated by cross-validation or holdout validation using the learning data, and the regression model that provides the smallest prediction error is selected. The selected regression model is output as the trained model. Note that the hyperparameters referred to here include, for example, the number of layers in a neural network and the regularization coefficient.

[0049] The control unit 24 controls the overall operation of the learning device 2.

[0050] The storage unit 25 stores various programs for operating the learning device 2 and data including various parameters necessary for the operation of the learning device 2. The various programs include a training data generation program that generates training data for generating a trained model, and a trained model generation program that generates a trained model by learning using the training data. The storage unit 25 also has a dataset storage unit 251 that stores multiple datasets that make up the training data. In this embodiment, multiple levels that associate the material conditions, structural conditions, and molding conditions for manufacturing a fiber-reinforced plastic molded product with their characteristic values ​​are stored as datasets.

[0051] The storage unit 25 is configured using a ROM (Read Only Memory) in which various programs etc. are pre-installed, a RAM (Random Access Memory) for storing calculation parameters and data for each process, an HDD (Hard Disk Drive), an SSD (Solid State Drive), etc.

[0052] The various programs can also be widely distributed by recording them on computer-readable recording media such as HDDs, flash memories, CD-ROMs, DVD-ROMs, Blu-ray (registered trademark), etc. The communication network referred to here is configured using, for example, an existing public line network, a LAN (Local Area Network), a WAN (Wide Area Network), etc., and may be wired or wireless.

[0053] The learning device 2 having the above functional configuration is a computer configured using one or more pieces of hardware 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).

[0054] The estimation device 3 is electrically connected to the learning device 2 and the display device 4. The estimation device 3 outputs an estimation result including characteristic values ​​estimated using the conditions of the fiber-reinforced plastic molded product to be estimated and the trained model acquired from the learning device 2. FIG. 7 is a block diagram showing the configuration of the estimation device included in the characteristic value estimation system according to an embodiment of the present invention. The estimation device 3 includes a calculation unit 31, a control unit 32, and a storage unit 33.

[0055] The calculation unit 31 calculates the characteristic values ​​of the fiber reinforced plastic molded product that are estimated using each condition of the estimation target acquired from the input device 5 and the trained model acquired from the learning device 2.

[0056] 8 is a diagram illustrating the flow of estimation processing performed by a characteristic value estimation system according to an embodiment of the present invention. The calculation unit 31 acquires a trained model 100 trained using the training data IP extracted by the extraction unit 21. The calculation unit 31 uses this trained model 100 to output an estimated characteristic value OP under the conditions of the estimation target.

[0057] The control unit 32 comprehensively controls the operation of the estimation device 3. The control unit 32 has a display control unit 321 that causes the display device 4 to display the calculation result (estimation result) of the calculation unit 31. The display control unit 321 may cause the display device 4 to display information on the estimation target, such as each condition, in addition to the estimation result.

[0058] The storage unit 33 stores various programs for operating the estimation device 3 and data including various parameters necessary for the operation of the estimation device 3. The various programs include a characteristic value estimation program executed using a trained model. The storage unit 33 is configured using a ROM in which the various programs are pre-installed, and a RAM, HDD, SSD, etc. that store calculation parameters and data for each process.

[0059] The various programs can be recorded on computer-readable recording media such as HDDs, flash memories, CD-ROMs, DVD-ROMs, and Blu-ray (registered trademark) and widely distributed. The estimation device 3 can also acquire the various programs via a communication network. The communication network here is configured using, for example, an existing public line network, LAN, WAN, etc., and can be wired or wireless.

[0060] The estimation device 3 having the above-described functional configuration is a computer configured using one or more pieces of hardware such as a CPU, a GPU, an ASIC, and an FPGA.

[0061] The display device 4 is a display made of liquid crystal or organic EL (Electro Luminescence) or the like, 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.

[0062] The input device 5 receives input of various information including information such as settings related to the process of estimating characteristic values, and outputs the received information to the learning device 2 and the estimation device 3. The input device 5 is configured using user interfaces such as a keyboard, a mouse, a microphone, and a touch panel.

[0063] Next, each process of the learning device 2 and the estimation device 3 will be described with reference to Fig. 9 and Fig. 10. Fig. 9 is a flowchart showing an outline of the learning process performed by the learning device according to this embodiment.

[0064] First, the learning device 2 acquires estimation conditions via the input device 5 (step S11). The estimation conditions input here are the fiber reinforced plastic molded product to be estimated and its characteristic values.

[0065] The learning device 2 generates learning data based on the estimation conditions input via the input device 5 (step S12). The extraction unit 21 reads out, from the multiple data sets stored in the storage unit 25, multiple data sets that satisfy the conditions (material conditions, molding conditions, and structural conditions) previously set for the fiber-reinforced plastic molded product to be estimated, based on the input estimation conditions. At this time, the structural conditions include at least interlayer information. Specifically, the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin. Furthermore, it is preferable that the extracted structural conditions further include geometric information of the fiber arrangement. The learning data generation unit 22 generates learning data using the data set extracted by the extraction unit 21. Based on the extracted data set, the learning data generation unit 22 generates learning data in which the material conditions, molding conditions, and structural conditions are explanatory variables and the characteristic values ​​are objective variables.

[0066] Then, the learning unit 23 generates a trained model by learning using the training data (step S13).

[0067] The estimation device 3 estimates the characteristic values ​​of the fiber-reinforced plastic molded product to be estimated. Fig. 10 is a flowchart for explaining the estimation process performed by the estimation device according to this embodiment. The calculation unit 31 calculates the characteristic values ​​of the fiber-reinforced plastic molded product to be estimated using the trained model generated by the learning device 2.

[0068] The calculation unit 31 acquires each condition of the fiber reinforced plastic molded product to be estimated via the input device 5 (step S21).

[0069] The calculation unit 31 calculates the characteristic values ​​using the trained model generated in step S13 (step S22), thereby estimating the characteristic values ​​of the fiber-reinforced plastic molded product that is the estimation target.

[0070] Next, the display control unit 321 outputs the estimation result by the calculation unit 31 to the display device 4 and performs display control to display the estimation result on the display device 4 (step S23). The display device 4 displays, for example, the estimated characteristic values ​​together with each condition of the fiber-reinforced plastic molded product that is the estimation target.

[0071] In the embodiment described above, when the training data generation unit generates training data using a plurality of data sets each including the material conditions, molding conditions, and structural conditions of a fiber-reinforced plastic molded product, as well as the characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, the training data generation unit includes interlayer information of the fiber-reinforced plastic molded product as an item belonging to the structural conditions, and the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin. According to this embodiment, by using a trained model that has learned the structural conditions including the interlayer information of the fiber-reinforced plastic molded product, the characteristics are predicted taking into account the interlayer information of the reinforcing fibers, thereby making it possible to predict the characteristic values ​​of the fiber-reinforced plastic molded product with high accuracy in a short time.

[0072] In the above-described embodiment, an example of extracting a dataset by the extraction unit 21 has been described. However, if there is no need for extraction processing, for example, if the dataset stored in the memory unit 25 is used as is, the configuration may be such that extraction processing is not performed by the extraction unit 21 (a configuration without the extraction unit 21).

[0073] (Variation) Next, a modified example of the embodiment of the present invention will be described with reference to Fig. 11 to Fig. 13. The estimation system according to the modified example is similar to the physical property value estimation system 1 according to the embodiment. Below, processing contents that differ from the embodiment will be described.

[0074] In this modified example, when extracting a data set according to the blending ratio of ingredients, the extraction unit 21 extracts the data set by excluding some data from the data set according to the set number of levels.

[0075] FIG. 11 is a diagram showing a list of the contents of the datasets. FIG. 12 is a diagram showing a list of the datasets after extraction. The extraction unit 21 performs a process of excluding raw materials for which the number of levels having data is less than a predetermined number, and levels using such raw materials, from the multiple datasets, and extracts, from the multiple datasets read out, datasets consisting of levels that include a condition in which the number of levels having data is equal to or greater than a predetermined number. In this modified example, an example in which the predetermined number is 2 will be described, but this is not limiting. In FIG. 11, the data shown in hatching is the data excluded by extraction.

[0076] In the example shown in FIG. 11, for example, the number of levels using material conditions 2 and 3, molding condition 2, and structural condition 1 is assumed to be 1. Therefore, the extraction unit 21 excludes conditions that do not satisfy the number of levels, and also excludes levels that include the excluded conditions. The extraction unit 21 repeats the extraction process using the above conditions for each extracted level until there are no conditions or levels that match the excluded conditions. As a result, the extraction unit 21 extracts a data set consisting of levels that include conditions with a predetermined number of levels or more (see FIG. 12).

[0077] Next, the processing of the learning device 2 according to this modification will be described with reference to Fig. 13. In this modification, the processing of extracting a data set differs from that of the embodiment.

[0078] First, the learning device 2 acquires estimation conditions via the input device 5 in the same manner as in step S11 (step S11). Thereafter, the learning device 2 generates learning data based on the estimation conditions input via the input device 5 (step S12).

[0079] 13 is a flowchart for explaining the learning data generation process performed by the learning device according to this modification. The extraction unit 21 reads out, from the plurality of data sets stored in the storage unit 25, a plurality of data sets that satisfy the respective conditions (material conditions, molding conditions, and structural conditions) previously set for the fiber-reinforced plastic molded product to be estimated, based on the input estimation conditions (step S31).

[0080] Then, the extraction unit 21 performs a process of excluding from the plurality of read data sets levels that use conditions for which the number of levels having data is less than two, and extracts from the plurality of read data sets data sets consisting of levels including material conditions, molding conditions, and structural conditions for which the number of levels having data is two or more (step S32). For example, levels with rare data configurations, in which the number of levels having data for each item is less than two, are excluded from the data sets. In this modified example, the explanation is given on the assumption that the number of levels of at least the information between layers is two or more and that it is not an out-of-bounds target.

[0081] Thereafter, the extraction unit 21 determines whether the number of levels is two or more for all conditions in the data set after extraction (step S33). If the extraction unit 21 determines that the number of levels is not two or more for some conditions (step S33: No), the process returns to step S32 and performs re-extraction processing on the data set after extraction. On the other hand, if the extraction unit 21 determines that the number of levels is two or more for all conditions (step S33: Yes), the process proceeds to step S34.

[0082] In step S34, the learning data generation unit 22 generates learning data using the data set extracted by the extraction unit 21. Based on the extracted data set, the learning data generation unit 22 generates learning data in which the material conditions, molding conditions, and structural conditions are explanatory variables and the characteristic values ​​are objective variables.

[0083] After extracting the data set, the learning unit 23 generates a trained model by learning using the training data generated in step S34 (step S13).

[0084] In the modified example described above, as in the embodiment, the learning data generation unit generates learning data using multiple data sets each including the material conditions, molding conditions, and structural conditions of the fiber-reinforced plastic molded product, as well as the characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions.In this case, the items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, and the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin.Therefore, the characteristic values ​​of the fiber-reinforced plastic molded product can be predicted with high accuracy in a short period of time.

[0085] Furthermore, according to this modification, by extracting a data set consisting of levels including a condition where the number of levels is two or more from among a plurality of data sets for estimating characteristic values ​​of a fiber-reinforced plastic molded product, training data from which rare data has been excluded is generated, and the training data is trained to generate a trained model. According to this embodiment, by suppressing changes in estimated values ​​due to rare data, it is possible to suppress a decrease in accuracy when estimating characteristic values.

[0086] (Other embodiments) Although the embodiments of the present invention have been described above, the present invention should not be limited to the above-described embodiments. For example, the estimation device may have the function of a learning unit. In this case, the estimation device not only calculates the objective variable of the estimation target but also sequentially updates the trained model.

[0087] Furthermore, in the above-described embodiment, an example has been described in which a dataset to be extracted is selected based on the number of levels. However, the selection of datasets is not limited to the form of the above-described embodiment, as long as the structural conditions include interlayer information, such as by using all of the datasets stored in the memory unit 25 as training data without performing the extraction process, and the interlayer information includes two or more of the presence or absence of interlayer reinforcement, geometric information of the interlayer reinforcement, and geometric information of the interlayer resin.

[0088] Furthermore, the machine learning is not limited to the above-mentioned deep learning, and for example, a support vector machine, a decision tree, a random forest, or a gradient boosting tree may be used. [Explanation of symbols]

[0089] 1. Characteristic value estimation system 2 Learning device 3 Estimation device 4 Display device 5 Input Devices 21 Extraction part 22 Learning data generation unit 23 Learning Department 24, 32 Control section 25, 33 Storage section 31 Calculation section 251 Dataset Storage Unit 321 Display control unit

Claims

1. A learning data generation method in which a computer generates learning data for estimating characteristic values ​​of a fiber-reinforced plastic molded product, a learning data generation step of reading from a storage unit a plurality of data sets each including material conditions, molding conditions, and structural conditions of a fiber-reinforced plastic molded product, as well as characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and generating learning data using the plurality of data sets; Including, The items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, The interlayer information includes two or more of: presence or absence of an interlayer reinforcing material; geometric information of the interlayer reinforcing material; and geometric information of an interlayer resin. Method for generating training data.

2. The characteristic value is a deformation characteristic value or a physical characteristic value; The learning data generation method according to claim 1 .

3. The items belonging to the structural conditions include geometric information of fiber arrangement, The learning data generation method according to claim 1 .

4. The geometric information of the fiber arrangement includes one or more of the fiber volume content of the fiber reinforced resin molded product, the fiber volume content in the layer, and the fiber distribution degree; The learning data generation method according to claim 3 .

5. The items belonging to the structural conditions include non-destructive testing data. The learning data generation method according to claim 1 .

6. The property values ​​included in the data set include values ​​measured in a fiber-reinforced resin molded product having an asymmetric laminate structure. The learning data generation method according to claim 1 .

7. The measurement accuracy of the characteristic value in the asymmetric laminate structure is higher than the measurement accuracy of the characteristic value in the symmetric laminate structure. The learning data generating method according to claim 6 .

8. A trained model generation method in which a computer generates a trained model for estimating characteristic values ​​of a fiber-reinforced plastic molded product, a learning data generation step of reading from a storage unit a plurality of data sets each including material conditions, molding conditions, and structural conditions of a fiber reinforced plastic molded product, as well as characteristic values ​​of the fiber reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and generating learning data using the plurality of data sets; A trained model generation step of generating a trained model by learning in the training data using material conditions, molding conditions, and structural conditions of a fiber reinforced plastic molded product as explanatory variables and characteristic values ​​of a fiber reinforced plastic molded product manufactured under the material conditions, molding conditions, and structural conditions as objective variables; Including, The items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, The interlayer information includes two or more of: presence or absence of an interlayer reinforcing material; geometric information of the interlayer reinforcing material; and geometric information of an interlayer resin. How to generate a trained model.

9. A method for estimating a characteristic value of a fiber reinforced plastic molded product, in which a computer estimates a characteristic value of a fiber reinforced plastic molded product, A calculation step of reading from the storage unit a trained model trained using training data generated based on a plurality of data sets each including material conditions, molding conditions, and structural conditions, as well as characteristic values ​​of a fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and calculating characteristic values ​​of the fiber-reinforced plastic molded product to be estimated based on the trained model; Including, The items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, The interlayer information includes two or more of: presence or absence of an interlayer reinforcing material; geometric information of the interlayer reinforcing material; and geometric information of an interlayer resin. A method for estimating the characteristic values ​​of fiber-reinforced plastic molded products.

10. A learning data generation program that causes a computer to generate learning data for estimating characteristic values ​​of a fiber-reinforced plastic molded product, a learning data generation step of reading from a storage unit a plurality of data sets each including material conditions, molding conditions, and structural conditions of a fiber-reinforced plastic molded product, as well as characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and generating learning data using the plurality of data sets; causing the computer to execute The items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, The interlayer information includes two or more of: presence or absence of an interlayer reinforcing material; geometric information of the interlayer reinforcing material; and geometric information of an interlayer resin. A program for generating training data.

11. A trained model generation program that causes a computer to generate a trained model for estimating characteristic values ​​of a fiber-reinforced plastic molded product, a learning data generation step of reading from a storage unit a plurality of data sets each including material conditions, molding conditions, and structural conditions of a fiber reinforced plastic molded product, as well as characteristic values ​​of the fiber reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and generating learning data using the plurality of data sets; A trained model generation step of generating a trained model by learning in the training data using material conditions, molding conditions, and structural conditions of a fiber reinforced plastic molded product as explanatory variables and characteristic values ​​of a fiber reinforced plastic molded product manufactured under the material conditions, molding conditions, and structural conditions as objective variables; causing the computer to execute The items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, The interlayer information includes two or more of: presence or absence of an interlayer reinforcing material; geometric information of the interlayer reinforcing material; and geometric information of an interlayer resin. A trained model generator.

12. A characteristic value estimation program for a fiber reinforced plastic molded product that causes a computer to estimate a characteristic value of a fiber reinforced plastic molded product, A calculation step of reading a trained model from the storage unit, the trained model being generated by learning a regression model using a plurality of data sets generated by reading from the storage unit a plurality of data sets each including material conditions, molding conditions, and structural conditions of a fiber-reinforced plastic molded product, and characteristic values ​​of the fiber-reinforced plastic molded product produced under the material conditions, molding conditions, and structural conditions, and calculating characteristic values ​​of the fiber-reinforced plastic molded product to be estimated based on the trained model; causing the computer to execute The items belonging to the structural conditions include information between layers of the fiber-reinforced plastic molded product, The interlayer information includes two or more of: presence or absence of an interlayer reinforcing material; geometric information of the interlayer reinforcing material; and geometric information of an interlayer resin. A program for estimating the characteristic values ​​of fiber-reinforced plastic molded products.

Images