Trained model generation program, setting program, trained model generation method, setting method, and trained model generator

Abstract

To provide a trained model generation program, a setting program, a trained model generation method, a setting method, and a trained model generator for estimating the mechanical characteristics of a molding quickly with high accuracy.SOLUTION: A trained model generation program causes a computer to execute a training step of: reading, from a storage unit, a plurality of data sets each composed of a mechanical property parameter obtained from a stress-strain curve of a resin composition and a mechanical characteristic of a molding obtained by numerical analysis of the mechanical property parameter and shape data on the molding; and generating a trained model having been trained using the plurality of data sets while adopting the mechanical property parameter as an explanatory variable and the mechanical characteristic of the molding as an objective variable.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a trained model generation program, a setting program, a trained model generation method, a setting method, and a trained model generation device. [Background technology]

[0002] In recent years, in order to improve the efficiency of material design, a technique has been disclosed in which design parameters are input and the desired properties of a material produced by the design parameters are estimated as a technique for selecting a combination of raw materials having desired physical properties from a wide variety of combinations of raw materials (see, for example, Patent Documents 1 and 2). In this technique, for example, design parameters are input and an estimated value of the glass transition temperature is output. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2021 / 015134 [Patent Document 2] Japanese Patent Publication No. 2022-13310 Summary of the Invention [Problem to be solved by the invention]

[0004] The mechanical properties of a molded article vary depending on its shape. For example, the impact absorption and fatigue properties of a test piece formed from a resin composition vary depending on its shape. The mechanical properties of a molded article are typically estimated through mechanical tests on test pieces specified by international standards, such as ISO test pieces and ASTM (American Society for Testing and Materials) test pieces. However, because mechanical properties vary depending on the shape, it is difficult to easily and accurately estimate the mechanical properties of a molded article of any shape.

[0005] Numerical analysis such as finite element simulation can be used to accurately estimate the mechanical properties of molded products of any shape. However, this type of numerical analysis requires a long run time to obtain results, making it unsuitable for short-term predictions or material design exploration.

[0006] The present invention has been made in consideration of the above, and aims to provide a trained model generation program, a setting program, a trained model generation method, a setting method, and a trained model generation device that can estimate mechanical properties of molded products of any shape quickly and with high accuracy. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems and achieve the objective, the trained model generation program of the present invention causes a computer to execute a learning step of reading from a memory unit multiple data sets consisting of mechanical property parameters obtained from the stress-strain curve of a resin composition and the mechanical properties of a molded product obtained by numerically analyzing the mechanical property parameters and shape data of the molded product, and using the multiple data sets to generate a trained model trained with the mechanical property parameters as explanatory variables and the mechanical properties of the molded product as target variables.

[0008] In addition, in the above invention, the trained model generation program of the present invention includes, as mechanical property parameters obtained from the stress-strain curve, four parameters that use the least squares method to approximate the stress-strain curve to two line segments: a first line segment having an endpoint at the origin, and a second line segment having one end at the breaking point and the other end connected to the endpoint of the first line segment on the opposite side from the origin, and that respectively represent the endpoint of the second line segment and the connecting point of the first and second line segments.

[0009] In addition, in the above invention, the trained model generation program of the present invention generates random numbers for the four parameters within a known range, and generates a predetermined number of patterns each consisting of a set of the four parameters.

[0010] In addition, in the trained model generation program of the present invention, in the above invention, the mechanical properties of the molded product include the energy absorbed per 1 g of resin composition obtained by impact analysis using the finite element method.

[0011] In addition, the setting program of the present invention causes a computer to execute a calculation step in which multiple mechanical property parameters are input into a trained model that has been trained using mechanical property parameters obtained from the stress-strain curve of a resin composition as explanatory variables and the mechanical properties of a molded product obtained by numerically analyzing the mechanical property parameters and shape data of the molded product as objective variables, and the computer calculates estimated values ​​of the mechanical properties of the molded product for each mechanical property parameter; and a setting step in which the mechanical property parameters that give estimated values ​​that comply with predetermined conditions, among the estimated values ​​obtained in the calculation step, are set as adapted mechanical property parameters.

[0012] In addition, the trained model generation method according to the present invention includes a learning step of reading from a storage unit a plurality of data sets each consisting of a set of mechanical property parameters obtained from the stress-strain curve of a resin composition and a set of mechanical properties of a molded product obtained by numerically analyzing the mechanical property parameters and shape data of the molded product, and using the plurality of data sets to generate a trained model trained using the mechanical property parameters as explanatory variables and the mechanical properties of the molded product as objective variables.

[0013] In addition, the setting method of the present invention includes a calculation step of reading out from a memory unit a trained model that has been trained using mechanical property parameters obtained from the stress-strain curve of the resin composition as explanatory variables and the mechanical properties of the molded product obtained by numerically analyzing the mechanical property parameters and the shape data of the molded product as objective variables, inputting multiple mechanical property parameters into the trained model, and calculating estimated values ​​of the mechanical properties of the molded product for each mechanical property parameter, and a setting step of setting, as an adapted mechanical property parameter, a mechanical property parameter that gives an estimated value that complies with predetermined conditions, from the estimated values ​​obtained in the calculation step.

[0014] In addition, the trained model generation device of the present invention includes a learning unit that uses multiple data sets consisting of mechanical property parameters obtained from the stress-strain curve of a resin composition and the mechanical properties of a molded product obtained by numerically analyzing the mechanical property parameters and shape data of the molded product to generate a trained model trained using the mechanical property parameters as explanatory variables and the mechanical properties of the molded product as target variables. [Effects of the Invention]

[0015] According to the present invention, it is possible to estimate the mechanical properties of a molded product of any shape quickly and with high accuracy. [Brief explanation of the drawings]

[0016] [Figure 1] FIG. 1 is a diagram showing a schematic configuration of a physical property estimation system according to one embodiment of the present invention. [Figure 2] FIG. 2 is a block diagram showing the configuration of a learning device included in a physical property estimation system according to one embodiment of the present invention. [Figure 3] FIG. 3 is a diagram for explaining stress-strain curves showing the relationship between stress and strain for each strain rate. [Figure 4] FIG. 4 is a diagram for explaining an example of a case where the stress-strain curve shown in FIG. 3 is approximated by two straight lines. [Figure 5]FIG. 5 is a diagram illustrating a data set used as learning data. [Figure 6] FIG. 6 is a block diagram showing the configuration of a setting device included in a physical property estimation system according to one embodiment of the present invention. [Figure 7] FIG. 7 is a diagram for explaining the flow of the estimation process performed by the physical property estimation system according to one embodiment of the present invention. [Figure 8] FIG. 8 is a flowchart showing an outline of the estimation process performed by the property estimation system according to one embodiment of the present invention. [Figure 9] FIG. 9 is a flowchart illustrating the learning process performed by the learning device according to one embodiment of the present invention. [Figure 10] FIG. 10 is a flowchart illustrating the parameter generation process performed by the learning device according to one embodiment of the present invention. [Figure 11] FIG. 11 is a diagram for explaining the parameter generation process performed by the learning device according to one embodiment of the present invention. [Figure 12] FIG. 12 is a diagram (part 1) for explaining an example of the setting process. [Figure 13] FIG. 13 is a diagram (part 2) for explaining an example of the setting process. [Figure 14] FIG. 14 is a diagram (part 3) for explaining an example of the setting process. DETAILED DESCRIPTION OF THE INVENTION

[0017] Hereinafter, an embodiment of a physical property 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 appropriate combination with each other.

[0018] (Embodiment) 1A to 1C are diagrams showing a schematic configuration of a physical property estimation system according to one embodiment of the present invention. The physical property 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, a setting device 3 that estimates physical properties 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 setting device 3, and an input device 5.

[0019] The physical properties estimated by the setting device 3 are mechanical properties of a molded article made of a resin composition, such as impact absorption properties and fatigue properties. In this embodiment, an example will be described in which the energy absorbed per gram of resin composition during an impact test (EA (Energy Absorption) Efficiency (J / g)) is used as the mechanical properties.

[0020] The learning device 2 is electrically connected to the setting device 3. The learning device 2 generates learning data, and generates and outputs a trained model by learning using the learning data. FIG. 2 is a block diagram showing the configuration of the learning device provided in the physical property estimation system 1 according to one embodiment of the present invention. The learning device 2 has a dynamic physical property parameter generation unit 21, a learning data generation unit 22, a learning unit 23, a control unit 24, and a memory unit 25.

[0021] The mechanical property parameter generator 21 generates mechanical property parameters (parameters) that represent the shape of the stress-strain curve. In this embodiment, the parameters generated by the mechanical property parameter generator 21 are values ​​that represent the endpoints obtained by two straight lines obtained by bilinear approximation of the stress-strain curve. Specifically, two line segments whose ends are connected to each other are generated by the bilinear approximation. The endpoint of one line segment is located at the origin, the endpoint of the other line segment is located approximately at the breaking point, and the junction point where the endpoints of the two line segments are connected is located approximately at the yield point. The mechanical property parameter generator 21 generates parameters that represent each of the two endpoints (junction point and second endpoint) of the three endpoints of this approximation line: the endpoint located at the origin (first endpoint), the endpoint where the endpoints of the two line segments are connected (junction point), and the endpoint on the opposite side of the origin (second endpoint). These parameters are represented by four values ​​of strain and stress that represent each endpoint.

[0022] FIG. 3 is a diagram for explaining a stress-strain curve showing the relationship between stress and strain for each strain rate. For example, FIG. 3 shows the relationship between stress and strain for each strain rate (s -1 ) are shown. As shown in Figure 3, even for the same material, the shape of the stress-strain curve changes depending on the strain rate.

[0023] FIG. 4 is a diagram illustrating an example of bilinear approximation of the stress-strain curve shown in FIG. 3. In the figure, points located near each line indicate values ​​on the stress-strain curve shown in FIG. 3. Two line segments shown in FIG. 4 are obtained by bilinear approximation of each stress-strain curve. For example, the approximation line for a strain rate of 230 ( / s) consists of two line segments (a first line segment L1 and a second line segment L2). The first line segment L1 extends from an end point S0 located at the origin to a junction point S2, which is the junction point between the first line segment L1 and the second line segment L2 and corresponds to the yield point. The second line segment L2 extends from an end point S1, which corresponds to the fracture point, to a junction point S2. In this case, the mechanical property parameters are a set of four parameters: strain (breaking strain) and stress (breaking stress) corresponding to the end point S1, and strain (yield strain) and stress (yield stress) corresponding to the connecting point S2.

[0024] The learning data generation unit 22 generates learning data that combines the parameters generated by the mechanical property parameter generation unit 21 with the energy absorbed per gram of resin composition (EA efficiency). The EA efficiency is obtained by numerically analyzing the mechanical property parameters and the shape data of the molded product. In this case, the numerical analysis can use the finite element method. For example, a finite element method model of the molded product is created using three-dimensional shell elements, and the characteristics are calculated using an explicit method. A known method can be used for the numerical analysis using the finite element method.

[0025] Fig. 5 is a diagram for explaining the data sets used as learning data. Fig. 5 shows 1 to N sets of data sets. Each data set is a set of mechanical property parameters (breaking strain, breaking stress, yield strain, yield stress) and the mechanical properties (EA efficiency) of the molded product obtained by numerically analyzing the mechanical property parameters and the shape data of the molded product. In this case, the breaking point and yield point can be set, for example, based on the above-mentioned end points S1 and S2.

[0026] The learning unit 23 generates a trained model that outputs a predicted value of EA efficiency based on input mechanical property parameters. The learning unit 23 obtains the trained model by learning the training data generated by the training data generation unit 22. In this case, the learning unit 23 performs training using the mechanical property parameters obtained from the stress-strain curve of the resin composition as explanatory variables and the mechanical properties of the molded product obtained by numerically analyzing the mechanical property parameters and the shape data of the molded product as objective variables.

[0027] The trained model is, for example, a neural network consisting of an input layer, an intermediate layer, and an output layer, 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.

[0028] 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.

[0029] The control unit 24 controls the overall operation of the learning device 2.

[0030] 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 also include a trained model generation program that uses training data to learn and generate a trained model.

[0031] 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.

[0032] 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 distributed widely. The learning device 2 can also acquire the various programs via a communications network. The communications 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 can be wired or wireless.

[0033] 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).

[0034] The setting device 3 is electrically connected to the learning device 2 and the display device 4. The setting device 3 estimates the mechanical properties of the molded product using the mechanical property parameters to be estimated and the trained model acquired from the learning device 2, and sets the mechanical property parameters based on the estimated values. FIG. 6 is a block diagram showing the configuration of the setting device 3 included in the physical property estimation system 1 according to an embodiment of the present invention. The setting device 3 has a calculation unit 31, a setting unit 32, a control unit 33, and a memory unit 34.

[0035] The calculation unit 31 calculates the mechanical properties of the molded product estimated using the mechanical property parameters acquired from the input device 5 and the trained model acquired from the learning device 2. In this embodiment, the EA efficiency is calculated for the mechanical property parameters obtained from the stress-strain curve. The calculation unit 31, for example, uses a plurality of preset parameter patterns to calculate the EA efficiency for each pattern.

[0036] The setting unit 32 extracts an EA efficiency that matches a preset condition from among the EA efficiencies calculated by the calculation unit 31, and sets the mechanical property parameters that match the EA efficiency as matched mechanical property parameters.

[0037] The control unit 33 comprehensively controls the operation of the setting device 3. The control unit 33 has a display control unit 331 that displays the setting results of the setting unit 32 on the display device 4. In addition to the setting results, the display control unit 331 may also display on the display device 4 the mechanical property parameters of the setting target, a stress-strain curve expressed by the mechanical property parameters, and information on the resin composition that shows this stress-strain curve.

[0038] The storage unit 34 stores various programs for operating the setting device 3 and data including various parameters necessary for the operation of the setting device 3. The various programs include a physical property estimation program executed using a trained model. The storage unit 34 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.

[0039] 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 distributed widely. The setting device 3 can also acquire the various programs via a communications network. The communications network referred to here is configured using, for example, an existing public line network, LAN, WAN, etc., and can be wired or wireless.

[0040] The setting 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, an FPGA, and the like.

[0041] 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 setting device 3. The display device 4 acquires and displays display data output from the setting device 3 under the control of the display control unit 331. The display device 4 may also have an audio output function such as a speaker.

[0042] The input device 5 receives input of various information including information such as settings related to the process of estimating physical properties, and outputs the received information to the learning device 2 and the setting device 3. The input device 5 is configured using user interfaces such as a keyboard, mouse, microphone, and touch panel.

[0043] 7 is a diagram illustrating the flow of estimation processing performed by the physical property estimation system 1 according to an embodiment of the present invention. The calculation unit 31 acquires a trained model 100 that has been trained using a plurality of pieces of training data IP and that estimates the EA efficiency of an estimation target. The calculation unit 31 uses this trained model 100 to output an estimated EA efficiency OP for the mechanical physical property parameters of the estimation target.

[0044] Fig. 8 is a flowchart showing an outline of the estimation process performed by the setting device 3. The estimation process shown in Fig. 8 shows a series of flows of the learning process and the estimation process, but if a trained model has been generated in advance and there is no need to generate or update a new model, the estimation process may be performed without executing the learning process.

[0045] First, the learning device 2 acquires setting conditions via the input device 5 (step S1). The setting conditions input here include at least shape data of the molded product. The shape data of the molded product is, for example, a "finite element method model of the molded product made with three-dimensional shell elements." In this embodiment, the EA efficiency for setting the adapted mechanical property parameters is also set as a setting condition. The EA efficiency for setting the adapted mechanical property parameters is, for example, set to the "predicted maximum EA efficiency."

[0046] The learning device 2 executes a learning process and generates a trained model (step S2).

[0047] FIG. 9 is a flowchart for explaining the learning process performed by the learning device 2 according to this embodiment.

[0048] In the learning process, first, the mechanical property parameter generating unit 21 generates mechanical property parameters using random numbers (step S11).

[0049] FIG. 10 is a flowchart illustrating the mechanical property parameter generation process performed by a learning device 2 according to an embodiment of the present invention. FIG. 11 is a diagram illustrating the mechanical property parameter generation process performed by a learning device 2 according to an embodiment of the present invention. Here, an example is described in which parameters expressed by four values ​​are generated: strain (parameter 3) and stress (parameter 1) indicating one end point of an approximation line obtained by bilinear approximation, and strain (parameter 4) and stress (parameter 2) indicating the other end point. As shown in FIG. 11, parameter P1, which corresponds to parameter 1, has a larger value than parameter P2, which corresponds to parameter 2. Furthermore, parameter P3, which corresponds to parameter 3, has a larger value than parameter P4, which corresponds to parameter 4.

[0050] The mechanical property parameter generating unit 21 first sets a parameter 1 (step S101). The mechanical property parameter generating unit 21 sets a parameter within a preset range. The mechanical property parameter generating unit 21 sets, as parameter 1, a value, for example, between 10 and 500 (MPa) as stress.

[0051] After setting parameter 1, the mechanical property parameter generation unit 21 sets parameter 2 based on parameter 1 (step S102). For example, the mechanical property parameter generation unit 21 selects a value greater than 0 and less than 1 as a coefficient, and sets the value obtained by multiplying the coefficient by parameter 1 as parameter 2. In this way, parameter 1 and parameter 2 of the stress are determined.

[0052] Furthermore, the mechanical property parameter generator 21 sets parameter 3 (step S103). The mechanical property parameter generator 21 sets parameter 3 within a preset range. The mechanical property parameter generator 21 sets parameter 3 to, for example, any value between 0.01 and 0.99 as strain.

[0053] After setting parameter 3, the mechanical property parameter generation unit 21 sets parameter 4 based on parameter 3 (step S104). For example, the mechanical property parameter generation unit 21 selects a value greater than 0 and less than 1 as a coefficient, and sets the value obtained by multiplying the coefficient by parameter 3 as parameter 4. In this way, parameters 3 and 4 of the strain are determined. Note that steps S103 and S104 may be executed before steps S101 and S102, or may be executed simultaneously with steps S101 and S102. The parameter setting ranges are known ranges that can be taken by the resin composition, and the setting range of each parameter is set according to the type of resin composition used.

[0054] Once parameters 1 to 4 are determined, the mechanical physical parameter generation unit 21 determines whether the pattern of the generated parameters 1 to 4 is a previously created pattern (step S105). The mechanical physical parameter generation unit 21 references the storage unit 25 and determines whether the currently generated pattern is the same as the pattern stored in the storage unit 25. If the mechanical physical parameter generation unit 21 determines that the generated pattern has already been created (step S105: Yes), the process proceeds to step S107. On the other hand, if the mechanical physical parameter generation unit 21 determines that the generated pattern has not yet been created (step S105: No), the process proceeds to step S106.

[0055] In step S106, the mechanical property parameter generating unit 21 stores the currently generated pattern (parameters 1 to 4) in the storage unit 25. After the storage process, the mechanical property parameter generating unit 21 proceeds to step S107.

[0056] In step S107, the mechanical physical parameter generation unit 21 determines whether the number of generated patterns is equal to or greater than a preset number. If the mechanical physical parameter generation unit 21 refers to the storage unit 25 and determines that the number of generated parameter patterns is less than the preset number (step S107: No), the process returns to step S101 and a new parameter pattern is generated. On the other hand, if the mechanical physical parameter generation unit 21 determines that the number of generated patterns is equal to or greater than the preset number (step S107: Yes), the parameter generation process ends.

[0057] 9, after the parameters are set, the learning data generation unit 22 reads out the mechanical property parameters generated by random numbers in step S11 and the shape data of the molded product from the storage unit 25, performs a numerical analysis, and calculates the EA efficiency (step S12). The learning data generation unit 22 performs, for example, an impact analysis using the finite element method as the numerical analysis.

[0058] The learning data generation unit 22 performs the numerical analysis multiple times to generate multiple data sets each including a pair of mechanical property parameters and EA efficiency (step S13). As a result, learning data consisting of multiple data sets each including a pair of mechanical property parameters and EA efficiency is generated, as shown in FIG.

[0059] Thereafter, the learning unit 23 executes a learning process (step S14). The learning unit 23 learns the learning data generated by the learning data generation unit 22, thereby generating a trained model.

[0060] 8, after the trained model is generated, the setting device 3 calculates an estimated value for the mechanical property parameter to be estimated (step S3). The calculation unit 31 inputs the mechanical property parameter to be estimated into the trained model generated by the learning device 2, thereby calculating the EA efficiency, which is an estimated value.

[0061] After calculating the estimated values, the setting unit 32 sets the mechanical property parameters based on the estimated values ​​(step S4). Of the estimated values ​​calculated by the calculation unit 31, an estimated value that becomes the EA efficiency for setting the adapted mechanical property parameters set in step 1 is extracted. Then, the setting unit 32 sets the combination of mechanical property parameters that gives the extracted estimated value as the adapted mechanical property parameter.

[0062] Next, the display control unit 331 outputs the setting result by the setting unit 32 to the display device 4, and performs display control to display the setting result on the display device 4 (step S5). At this time, the display device 4 may display information on the EA efficiency corresponding to the adapted dynamic property parameter together with the setting result.

[0063] An example of the setting process will now be described with reference to FIGS. 12 to 14. FIGS. 12 to 14 are diagrams for explaining an example of the setting process. FIGS. 12 to 14 show the results of predicting EA efficiency by randomly generating 1,000 mechanical property parameters, with the breaking strain set to 0.03 to 0.1, the breaking stress set to 10 to 30 MPa, the yield strain set to 1 to 100% of the breaking strain, and the yield stress set to 1 to 100% of the breaking stress. FIG. 12 shows an example of the breaking stress relative to the breaking strain. FIG. 13 shows an example of the yield stress relative to the yield strain. In both FIGS. 12 and 13, the shading of each plot indicates the magnitude of the EA efficiency. Furthermore, FIG. 14 shows an approximation line estimated to have the highest EA efficiency based on the results of FIGS. 12 and 13.

[0064] For example, in Figure 12, point E1 with the highest EA efficiency is extracted. Also, in Figure 13, point E2 with the highest EA efficiency is extracted. Based on this extraction result, parameters are set such that, for example, the yield strain is 0.013, the yield stress is 120 MPa, the breaking strain is 0.078, and the breaking stress is 126 MPa. For example, based on these parameters, approximate lines consisting of line segments L3 and L4 connecting the origin with the breaking point S3 and the yield point S4, respectively, are generated (see Figure 14), and this approximate line can be used to obtain an optimal stress-strain curve.

[0065] In the embodiment described above, a trained model for estimating the mechanical properties of a molded product is generated by learning using a plurality of data sets each including a set of mechanical property parameters and the mechanical properties of the molded product obtained by numerically analyzing the mechanical property parameters and shape data of the molded product, so that a trained model can be constructed for a molded product of any shape, thereby enabling the mechanical properties of the molded product to be estimated with high accuracy.

[0066] Furthermore, in the above-described embodiment, it is possible to extract mechanical property parameters that optimize the mechanical properties of a molded article of any shape, thereby enabling the establishment of material design guidelines for resin molded articles with excellent impact absorption properties for any molded article shape.

[0067] In the above-described embodiment, an example was described in which the mechanical properties of the molded product were values ​​obtained by numerical analysis as data sets, but some data sets may include values ​​in which the mechanical properties of the molded product are actually measured values.

[0068] (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 setting device may have the function of a learning unit. In this case, the setting device not only calculates the objective variable of the estimation target, but also generates learning data and sequentially updates the trained model. [Explanation of symbols]

[0069] 1. Physical property estimation system 2 Learning device 3 Setting device 4 Display device 5 Input Devices 21 Mechanical property parameter generation unit 22 Learning data generation unit 23 Learning Department 24, 33 Control section 25, 34 Storage section 31 Calculation section 32 Setting section 331 Display control unit

Claims

1. On the computer, a learning step of reading from a storage unit a plurality of data sets each including a set of mechanical property parameters obtained from a stress-strain curve of a resin composition and a set of mechanical properties of a molded article obtained by numerically analyzing the mechanical property parameters and shape data of the molded article, and generating a trained model trained using the plurality of data sets with the mechanical property parameters as explanatory variables and the mechanical properties of the molded article as objective variables; A trained model generation program that executes the above.

2. The mechanical property parameters obtained from the stress-strain curve are: The stress-strain curve is approximated by two line segments using the least squares method: a first line segment having an end point at the origin, and a second line segment having one end at the breaking point and the other end connected to the end point of the first line segment on the opposite side from the origin; and four parameters representing the end points of the second line segment and the connecting points of the first and second line segments, respectively. The trained model generation program according to claim 1 , comprising:

3. The four parameters are generated as random numbers within a known range; generating a preset number of patterns each having a set of the four parameters; The trained model generation program according to claim 2.

4. As the mechanical properties of the molded article, the energy absorbed per 1 g of the resin composition obtained by impact analysis using the finite element method, The trained model generation program according to claim 1 , comprising:

5. On the computer, a calculation step of inputting a plurality of mechanical property parameters into a trained model that has been trained using mechanical property parameters obtained from the stress-strain curve of the resin composition as explanatory variables and mechanical properties of the molded article obtained by numerically analyzing the mechanical property parameters and shape data of the molded article as objective variables, and calculating an estimated value of the mechanical property of the molded article for each mechanical property parameter; a setting step of setting a mechanical property parameter that gives an estimated value that satisfies a preset condition, among the estimated values ​​obtained in the calculation step, as an adapted mechanical property parameter; A configuration program that runs the following.

6. a learning step of reading from a storage unit a plurality of data sets each including a set of mechanical property parameters obtained from a stress-strain curve of a resin composition and a set of mechanical properties of a molded article obtained by numerically analyzing the mechanical property parameters and shape data of the molded article, and generating a trained model trained using the plurality of data sets with the mechanical property parameters as explanatory variables and the mechanical properties of the molded article as objective variables; A method for generating trained models, including:

7. a calculation step of reading from a storage unit a trained model that has been trained using mechanical property parameters obtained from the stress-strain curve of the resin composition as explanatory variables and mechanical properties of the molded article obtained by numerically analyzing the mechanical property parameters and shape data of the molded article as objective variables, inputting a plurality of mechanical property parameters into the trained model, and calculating an estimated value of the mechanical properties of the molded article for each mechanical property parameter; a setting step of setting a mechanical property parameter that gives an estimated value that satisfies a preset condition, among the estimated values ​​obtained in the calculation step, as an adapted mechanical property parameter; Including how to set it up.

8. a learning unit that generates a trained model using a plurality of data sets each including a mechanical property parameter obtained from a stress-strain curve of a resin composition and a mechanical property of a molded article obtained by numerically analyzing the mechanical property parameter and shape data of the molded article, the trained model being trained using the mechanical property parameter as an explanatory variable and the mechanical property of the molded article as a target variable; A trained model generation device comprising:

Images