A program for generating trained models, a program for predicting PVT curves, a program for splitting training data, a method for generating PVT curve prediction models, a method for predicting PVT curves, and a method for calculating fluid dynamics analysis.

Abstract

To provide a program and method for more accurately understanding the PVT properties of unknown resin materials. [Solution] The trained model generation program according to the present invention performs a PVT curve prediction model generation step, in which the computer is made to perform machine learning using training data in which the explanatory variables are the blending ratio of raw materials constituting a known crystalline resin composition, a plurality of pressures and a continuous range of temperatures, and the target variable is the specific volume of the resin composition at each pressure and a continuous range of temperatures, thereby generating a PVT curve prediction model.

Description

Technical Field

[0001] The present invention relates to a trained model generation program, a PVT curve prediction program, a training data division program, a PVT curve prediction model generation method, a PVT curve prediction method, and a flow analysis calculation method.

Background Art

[0002] For parts manufactured using resin materials, the shape and resin material (composition, manufacturing conditions, etc.) are designed based on prediction results of part characteristics such as mechanical strength. As a technique for predicting part characteristics, CAE (Computer Aided Engineering) analysis using the resin physical properties of the resin material constituting the part and the shape of the part is known (see, for example, Patent Document 1).

[0003] CAE analysis of resins requires input of CAD data defining the shape and physical property values of the resin material to be analyzed. In particular, injection molding analysis, which predicts the characteristics of a part shape during molding, requires input of the resin's melt viscosity, crystallization temperature, thermal conductivity, linear expansion coefficient, and PVT characteristics (Pressure-Volume-Temperature), so these characteristic values of the target resin material must be measured in advance.

[0004] Patent Document 1 discloses a method for estimating material physical properties varying between lots based on process parameters during injection molding, in order to reflect differences in resin physical properties between lots.

Prior Art Literature

Patent Literature

[0005]

Patent Document 1

Summary of the Invention

Problem to be Solved by the Invention

[0006] However, this approach assumed that the PVT properties of the resin material were known, and did not enable CAE analysis for unknown resin materials. Conventionally, CAE analysis of component properties using resins relied on the physical properties of known resin materials, but there was a need to accurately understand the physical properties of unknown resin materials and determine whether they were suitable for the required properties. In particular, the PVT properties necessary for flow analysis change in a complex manner depending on the properties of the resin, making them indispensable for predicting the physical properties of unknown resin materials.

[0007] The present invention has been made in view of the above, and aims to provide a trained model generation program, a PVT curve prediction program, a training data partitioning program, a PVT curve prediction model generation method, a PVT curve prediction method, and a flow analysis calculation method that can predict PVT properties with high accuracy in an unknown resin material. [Means for solving the problem]

[0008] To solve the above-mentioned problems and achieve the objective, the PVT curve prediction program of the present invention performs a PVT curve prediction model generation step, in which the computer is made to perform machine learning using training data in which the explanatory variables are the blending ratio of raw materials constituting a known crystalline resin composition, multiple pressures and a continuous range of temperatures, and the objective variable is the specific volume of the resin composition at each pressure and a continuous range of temperatures, thereby generating a PVT curve prediction model.

[0009] The PVT curve prediction program of the present invention includes, in the PVT curve prediction model generation step according to the above invention, a solid specific volume prediction model using only data where the measurement temperature is below the crystallization temperature as training data, a melt specific volume prediction model using only data where the measurement temperature is above the crystallization temperature, and a crystallization temperature prediction model generated by having a computer perform machine learning using training data in which the explanatory variables are the blending ratio of raw materials constituting a known crystalline resin composition, multiple pressures for one blending ratio, and the objective variable is the crystallization temperature of the crystalline resin composition at each pressure.

[0010] The PVT curve prediction program in the present invention performs an input step in which the mixing ratio of raw materials constituting a known crystalline resin composition, measured pressure and measured temperature are input to a PVT curve prediction model generated by performing machine learning using training data in which the explanatory variables are the mixing ratio of raw materials constituting a known crystalline resin composition, multiple pressures and a continuous range of temperatures, and the objective variable is the specific volume of the crystalline resin composition; and a PVT prediction curve output step in which a PVT curve is calculated based on the information output by the PVT curve prediction model.

[0011] The training data division program in the present invention provides training data in which the blending ratio of raw materials constituting a known crystalline resin composition, the measured pressure and measured temperature are explanatory variables, and the specific volume of the crystalline resin composition is the objective variable. The program then causes a computer to perform a division step in which it estimates the crystallization temperature of a known crystalline resin composition at each pressure from the rate of change of specific volume when the measured temperature changes at a constant pressure, and divides the training data into data above and below the estimated crystallization temperature.

[0012] The PVT curve prediction model generation method of the present invention generates a trained model by having a computer perform machine learning using training data in which the explanatory variables are the blending ratio of raw materials constituting a known crystalline resin composition, multiple pressures and a continuous range of temperatures, and the target variable is the specific volume of the resin composition at each pressure and a continuous range of temperatures.

[0013] The PVT curve prediction method of the present invention involves performing machine learning on a PVT curve prediction model generated by using training data in which the explanatory variables are the blending ratio of raw materials constituting a known crystalline resin composition, multiple pressures and a continuous range of temperatures, and the objective variable is the specific volume of the crystalline resin composition. The PVT curve prediction model is then input with the blending ratio of raw materials constituting the crystalline resin composition to be predicted, the measured pressure and the measured temperature, and an output PVT prediction curve based on the information output by the PVT curve prediction model.

[0014] The fluid analysis calculation method in the present invention involves performing an analysis based on the PVT model equation, using the PVT curve obtained by the PVT curve prediction method according to the above invention. [Effects of the Invention]

[0015] According to the present invention, even for unknown crystalline resin compositions, the PVT curve can be efficiently determined and used for resin flow analysis. [Brief explanation of the drawing]

[0016] [Figure 1] Figure 1 is a diagram showing the schematic configuration of the PVT prediction curve calculation system according to Embodiment 1 of the present invention. [Figure 2] Figure 2 is a block diagram showing the configuration of the learning device included in the PVT prediction curve calculation system according to Embodiment 1 of the present invention. [Figure 3] Figure 3 is a block diagram showing the configuration of the analysis device included in the PVT prediction curve calculation system according to Embodiment 1 of the present invention. [Figure 4] Figure 4 is a diagram illustrating the flow of the PVT prediction curve calculation process performed by the PVT prediction curve calculation system according to Embodiment 1 of the present invention. [Figure 5] Figure 5 is a flowchart showing an overview of the learning process performed by the learning device according to Embodiment 1 of the present invention. [Figure 6] Figure 6 is a flowchart outlining the PVT prediction curve calculation process performed by the PVT prediction curve calculation system according to Embodiment 1 of the present invention. [Figure 7] Figure 7 is a flowchart showing an overview of the PVT prediction curve calculation process performed by the PVT prediction curve calculation system according to Embodiment 2 of the present invention. [Modes for carrying out the invention]

[0017] Hereinafter, embodiments of a PVT curve prediction method, a PVT curve prediction program, and a PVT curve prediction apparatus according to the present invention will be described in detail with reference to the drawings. The present invention is not limited by these embodiments. Furthermore, individual embodiments of the present invention are not independent of each other, and can be implemented as appropriate in combination with each other.

[0018] (Embodiment 1) [System Configuration] FIG. 1 is a diagram showing a schematic configuration of a PVT prediction curve calculation system according to Embodiment 1 of the present invention. The PVT prediction curve calculation system 1 includes: a learning device 2 that creates learning data and generates a trained model by learning using the created learning data; an analysis device 3 that executes analysis processing of a PVT prediction curve using the trained model generated by the learning device 2; a display device 4 that displays information including an analysis result of the analysis device 3; and an input device 5.

[0019] The learning device 2 is electrically connected to the analysis device 3. The learning device 2 selectively extracts learning data, generates and outputs a trained model through learning using the extracted learning data. FIG. 2 is a block diagram showing a configuration of the learning device included in the analysis system according to Embodiment 1 of the present invention. The learning device 2 includes a learning data generation unit 21, a learning unit 22, a control unit 23, and a storage unit 24.

[0020] The learning data generation unit 21 generates learning data using data stored in the storage unit 24 and data input from an external source. The learning data generation unit 21 generates a data set that uses the blending ratio of a resin composition, a plurality of measured pressures, and measured temperatures in a continuous range as explanatory variables, and uses specific volumes of the resin composition at each measured pressure and measured temperature as an objective variable.

[0021] The learning unit 22 generates a trained model by performing training using training data. In this process, the learning unit 22 generates a trained model by performing machine learning using a dataset in which the blending ratio of the resin composition, multiple measurement pressures, and a continuous range of measurement temperatures are explanatory variables, and the specific volume of the resin composition at each measurement pressure and temperature is the target variable. The learning performed by the learning unit 22 can employ known learning methods. Examples of statistical models that can be used for learning include simple linear models, Ridge models, Lasso models, Elastic Net models, general additive models, random forest models, rule-fit models, gradient boosting tree models, extra-tree models, support vector models, Gaussian process models, k-nearest neighbor models, kernel ridge models, simple Bayesian models, and neural networks.

[0022] For example, when the learning unit 22 generates a trained model by learning using regularization, the learning unit 22 is given multiple candidate values ​​for the hyperparameters of the trained model, performs learning for each of the given candidate hyperparameter values, generates one trained model for one target variable (characteristic value), and stores it in the memory unit 24. Then, the learning unit 22 calculates the prediction error for the models obtained by learning with each candidate value using cross-validation or holdout validation with the training data, and selects the trained model that gives the smallest prediction error. Here, hyperparameters are parameters that the learning unit 22 sets in advance for learning, and include, for example, the regularization coefficient. In the case of a trained model using a neural network, the hyperparameters also include the number of layers in the neural network.

[0023] The control unit 23 comprehensively controls the operation of the learning device 2.

[0024] The memory unit 24 stores data including various programs for operating the learning device 2, and various parameters necessary for the operation of the learning device 2. The various programs include a learning data generation program that generates training data for generating a trained model, and a trained model generation program that uses the training data to train and generate a trained model.

[0025] The memory unit 24 is composed of a ROM (Read Only Memory) on which various programs are pre-installed, and RAM (Random Access Memory), HDD (Hard Disk Drive), SSD (Solid State Drive), etc., which store calculation parameters and data for each process.

[0026] Various programs can be recorded on computer-readable recording media such as HDDs, flash memory, CD-ROMs, DVD-ROMs, and Blu-ray® discs and widely distributed. The communication network referred to here is configured using existing public telephone networks, LANs (Local Area Networks), WANs (Wide Area Networks), etc., and can be wired or wireless.

[0027] The learning device 2 having the above functional configuration is a computer composed of one or more hardware components such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), and an FPGA (Field Programmable Gate Array).

[0028] Furthermore, the learning device 2 is connected to the analysis device 3 and the input device 5 via a communication network, enabling it to send and receive information.

[0029] The analysis device 3 is electrically connected to the learning device 2 and the display device 4. The analysis device 3 performs component characteristic prediction analysis processing using the trained model created by the learning device 2 and outputs the analysis results. Figure 3 is a block diagram showing the configuration of the analysis device included in the analysis system according to Embodiment 1 of the present invention. The analysis device 3 has a calculation unit 31, a control unit 32, and a storage unit 33.

[0030] The calculation unit 31 uses the predicted specific volume output by inputting the manufacturing conditions of the resin composition into a trained model to calculate the relationship between specific volume and temperature under multiple pressure conditions for a single crystalline resin composition in the form of a two-dimensional graph.

[0031] The control unit 32 comprehensively controls the operation of the analysis device 3. The control unit 32 includes a display control unit 321 that displays the calculation results (analysis results) of the calculation unit 31 on the display device 4. In addition to the analysis results, the display control unit 321 may also display information such as analysis conditions and manufacturing conditions on the display device 4.

[0032] The memory unit 33 stores data including various programs for operating the analysis device 3, and various parameters necessary for the operation of the analysis device 3. The various programs also include a component characteristic prediction program executed using a trained model. The memory unit 33 is composed of a ROM with various programs pre-installed, and RAM, HDD, SSD, etc., for storing calculation parameters and data for each process.

[0033] Various programs can be recorded on computer-readable recording media such as HDDs, flash memory, CD-ROMs, DVD-ROMs, and Blu-ray® discs and widely distributed. Furthermore, the analysis device 3 can acquire various programs via a communication network. This communication network can be, for example, an existing public network, LAN, or WAN, and can be wired or wireless.

[0034] The analysis device 3 having the above functional configuration is a computer composed of one or more hardware components such as a CPU, GPU, ASIC, and FPGA.

[0035] Furthermore, the analysis device 3 is connected to the learning device 2, the display device 4, and the input device 5 via a communication network, enabling it to send and receive information.

[0036] The display device 4 is a display made of liquid crystal or organic EL (Electro-Luminescence), and is connected to the analysis device 3 in a communication manner. The display device 4 acquires and displays display data output from the analysis 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.

[0037] Input device 5 is connected to the learning device 2 and the analysis device 3 via a communication network, enabling it to send and receive information. Input device 5 accepts various types of information, including settings related to the process of calculating the PVT prediction curve, and outputs the received information to the learning device 2 and the analysis device 3. Input device 5 is configured using a user interface such as a keyboard, mouse, microphone, and touch panel.

[0038] In this embodiment, the resin composition is, for example, a crystalline resin composition, and is a material mainly composed of crystalline resin. Below, an example of performing a process on a crystalline resin composition will be described.

[0039] There are no particular restrictions on the crystalline resins that make up the crystalline resin composition, but examples include fluororesins, polyoxymethylene, polyamides, polyesters, polyamide-imides, olefin resins, polyether ketones, polyether ether ketones, polyarylene sulfide, cellulose derivatives, liquid crystalline resins, tetrafluoroethylene resins, epoxy resins, and modified resins thereof. Two or more of these may be included. In addition, amorphous resins such as styrene resins, polyvinyl chloride, polyolefin elastomers, polyether ester elastomers, polyetheramide elastomers, polyacrylates, polyphenylene ethers, polycarbonates, polyethersulfones, polyetherimides, and acrylonitrile styrene may also be included, as long as the crystalline resins constitute the majority of the resin components.

[0040] The resin composition may also contain fibrous fillers, non-fibrous inorganic fillers, and other additives.

[0041] Examples of fibrous fillers include glass fibers, glass milled fibers, glass flat fibers, irregularly shaped cross-section glass fibers, glass cut fibers, flattened glass fibers, stainless steel fibers, aluminum fibers, brass fibers, rock wool, carbon fibers such as PAN (Polyacrylonitrile) and pitch-based fibers, carbon nanotubes, carbon nanofibers, calcium carbonate whiskers, wollastonite whiskers, potassium titanate whiskers, barium titanate whiskers, aluminum borate whiskers, silicon nitride whiskers, aramid fibers, alumina fibers, silicon carbide fibers, asbestos fibers, gypsum fibers, ceramic fibers, zirconia fibers, silica fibers, titanium oxide fibers, and silicon carbide fibers. Two or more of these can be used in combination. Among these, glass fibers and carbon fibers are preferred.

[0042] Non-fibrous inorganic fillers include talc, wollastonite, zeolite, sericite, mica, kaolin, clay, pyrophyllite, bentonite, asbestos, silicates such as alumina silicate and hydrotalcite, silicon dioxide, glass powder, magnesium oxide, aluminum oxide (alumina), silica (crushed / spherical), quartz, glass beads, glass flakes, crushed / irregularly shaped glass, glass microballoons, molybdenum disulfide, aluminum oxide (crushed), and translucent alumina (fibrous, plate-like, flaky, granular, irregularly shaped, crushed), acid Examples of inorganic fillers include titanium oxide (crushed), zinc oxide (fibrous, plate-like, flaky, granular, irregularly shaped, crushed), calcium carbonate, magnesium carbonate, zinc carbonate, and other carbonates, calcium sulfate, barium sulfate, and other sulfates, calcium hydroxide, magnesium hydroxide, aluminum hydroxide, and other hydroxides, silicon carbide, carbon black and silica, graphite, aluminum nitride, translucent aluminum nitride (fibrous, plate-like, flaky, granular, irregularly shaped, crushed), calcium polyphosphate, graphite, metal powders, metal flakes, metal ribbons, and metal oxides. Specific examples of metal species (metal powders, metal flakes, metal ribbons) include silver, nickel, copper, zinc, aluminum, stainless steel, iron, brass, chromium, and tin. Other inorganic fillers include carbon powder, graphite, carbon flakes, flaky carbon, fullerene, and graphene, which may be hollow, and it is also possible to use two or more of these inorganic fillers in combination. Among these, calcium carbonate, carbon black, and graphite are preferred.

[0043] Other additives include, for example, silane compounds, antioxidants and heat stabilizers, weathering agents, mold release agents and lubricants, pigments, dyes, nucleating agents, plasticizers, antistatic agents, flame retardants, heat stabilizers, lubricants, UV inhibitors, colorants, flame retardants and blowing agents, and other common additives.

[0044] [PVT curve prediction processing] Figure 4 is a diagram illustrating the flow of the PVT prediction curve calculation process performed by the PVT prediction curve calculation system according to Embodiment 1 of the present invention. In the PVT prediction curve calculation process, the learning device 2, analysis device 3, display device 4, etc., shown in Figure 1, cooperate to calculate the prediction curve OP from the input value set. The analysis device 3 acquires the trained model, the PVT curve prediction model 100, from the learning device 2. Using this PVT curve prediction model 100, the analysis device 3 calculates the predicted values ​​of each pressure and the specific volume at a continuous range of temperatures for the resin composition, from the input value set IP consisting of the blending ratio of the raw materials constituting the crystalline resin composition, multiple pressures, and a continuous range of temperatures. Therefore, the output prediction curve may also be formed by multiple points.

[0045] The blending ratios used in the input value set can be those of crystalline resin compositions whose PVT properties have not yet been acquired but whose other physical properties are known. Alternatively, blending ratios of hypothetically defined crystalline resin compositions for which no physical property information has been acquired may be used.

[0046] The multiple pressures and continuous temperature ranges used in the input value set can be set to appropriate ranges for each resin type, as long as they are within the upper and lower limits of the training data. These can be input separately using the input device 5, or pre-set values ​​can be used.

[0047] The display device 4 uses the predicted specific volume values ​​output by inputting the manufacturing conditions of the resin composition into a trained model to output a prediction curve OP for a single crystalline resin composition, with the specific volume under multiple pressure conditions on the vertical axis and temperature on the horizontal axis.

[0048] Next, the flow of the learning process performed by the learning device 2 will be explained with reference to Figure 5. Figure 5 is a flowchart showing an overview of the learning process performed by the learning device according to Embodiment 1 of the present invention.

[0049] First, the learning device 2 refers to the memory unit 24 and extracts a dataset to be used for learning (step S11). Here, the learning data generation unit 21 extracts a dataset consisting of known crystalline resin composition blending ratios and specific volumes corresponding to the blending ratios and ranges of multiple measurement pressures and measurement temperatures. This dataset corresponds to the learning data for generating a trained model.

[0050] The learning unit 22 generates a trained model (PVT curve prediction model) by performing machine learning using the training data generated in step S11, with the explanatory variables being the blending ratio of raw materials constituting a known crystalline resin composition, multiple pressures, and a continuous range of temperatures, and the objective variable being the specific volume of the resin composition at each pressure and a continuous range of temperatures (step S12).

[0051] Next, the flow of the PVT curve prediction process performed by the analysis device 3 will be explained with reference to Figure 6. An overview is shown below. Figure 6 is a flowchart showing the overview of the PVT curve prediction process performed by the analysis device according to Embodiment 1.

[0052] The calculation unit 31 acquires a trained model from the learning device 2, and, referring to the storage unit 33, inputs the blending ratio of the crystalline resin composition and the calculation range consisting of multiple pressures and a continuous range of temperatures, which are set in advance for each resin type, into the PVT curve prediction model to obtain the predicted value of the PVT curve (step S21: input step).

[0053] Subsequently, the calculation unit 31 obtains predicted values ​​from the PVT curve prediction model to calculate the specific volume corresponding to the blending ratio of the crystalline resin composition, multiple pressures, and a continuous temperature range.

[0054] When the display control unit 321 obtains the calculation result from the calculation unit 31, it outputs this setting result to the display device 4 and performs display control to display the calculation result on the display device 4. The display device 4 depicts and displays the change in specific volume under multiple pressures and continuous temperature ranges for a single crystalline resin composition in the form of a two-dimensional graph (temperature vs. specific volume) with specific volume under multiple pressures on the vertical axis and temperature on the horizontal axis (step S22: PVT prediction curve output step).

[0055] Subsequently, parameter fitting based on the PVT model equation is performed on the PVT prediction curve obtained as a 2D graph, and the obtained parameters are input as resin properties. This allows for resin flow analysis to be performed on a crystalline resin composition consisting of the blending ratios of the input value set. There are no particular restrictions on the PVT model equation, but examples include the Schmidt model and the Tait model.

[0056] In Embodiment 1 described above, the specific volume of the crystalline resin composition is estimated using a trained model, and a prediction curve is calculated from its correspondence with the measured pressure and temperature. According to Embodiment 1, for resin compositions whose PVT curve is unknown, it is possible to estimate the PVT curve using the trained model, and by performing parameter fitting in the same way as with measured data, the number of candidate crystalline resin compositions to be targeted for CAE analysis increases, making it easier to identify crystalline resin compositions that satisfy the required characteristics of the parts.

[0057] (Embodiment 2) Next, Embodiment 2 of the present invention will be described. In Embodiment 2, the PVT curve prediction process (see Figure 6) described above is different. Note that the configuration of the PVT prediction curve calculation system 1 is the same as in Embodiment 1, so its description will be omitted.

[0058] In Embodiment 2, the specific volume prediction model is divided into separate prediction models for temperatures above the crystallization temperature and below the crystallization temperature. A crystallization temperature prediction model is then prepared separately. This set of three trained models is treated as a PVT curve prediction model, and the predicted values ​​are combined to calculate the temperature dependence of the specific volume of each crystalline resin composition at multiple measurement pressures as a graph.

[0059] For example, in step S11 shown in Figure 5, the training data generation unit 21 takes the training data extracted as described above, in which the explanatory variables are the blending ratio of raw materials constituting the known crystalline resin composition, the measurement pressure and measurement temperature, and the objective variable is the specific volume of the crystalline resin composition, and estimates the crystallization temperature of the known crystalline resin composition at each pressure from the rate of change of specific volume when the measurement temperature changes at a constant pressure, and divides the training data into those above the estimated crystallization temperature and those below the estimated crystallization temperature (dividing step). Furthermore, the training data generation unit 21 generates and extracts training data in which the explanatory variables are the blending ratio of raw materials constituting the known crystalline resin composition and the measurement pressure, and the objective variable is the estimated crystallization temperature of the crystalline resin composition at each measurement pressure.

[0060] Subsequently, in step S12, the learning unit 22 generates PVT curve prediction models consisting of a solid specific volume prediction model, a melting specific volume prediction model, and a crystallization temperature prediction model by performing machine learning using the respective training data extracted by the training data generation unit 21.

[0061] Figure 7 is a flowchart showing an overview of the PVT curve prediction process according to Embodiment 2. The calculation unit 31 acquires multiple trained models (solid specific volume prediction model, melt specific volume prediction model, crystallization temperature prediction model) that constitute the PVT curve prediction model from the learning device 2, and inputs the blending ratio of the crystalline resin composition, multiple pressures set in advance for each resin type, and a calculation range consisting of a continuous temperature range into the PVT curve prediction model by referring to the storage unit 33 (step S31).

[0062] Subsequently, the calculation unit 31 obtains predicted values ​​from the PVT curve prediction model to calculate the specific volume corresponding to the blending ratio of the crystalline resin composition, multiple pressures, and a continuous temperature range.

[0063] When the display control unit 321 obtains the calculation result from the calculation unit 31, it outputs this setting result to the display device 4 and performs display control to display the calculation result on the display device 4. The display device 4 displays the change in specific volume over multiple pressures and a continuous range for a single crystalline resin composition in the form of a two-dimensional graph with specific volume under multiple pressures on the vertical axis and temperature on the horizontal axis (step S32).

[0064] In Embodiment 2 described above, similar to Embodiment 1, the specific volume of the crystalline resin composition is estimated using a trained model, and a prediction curve is calculated from its correspondence with the measured pressure and temperature. As a result, the number of candidate crystalline resin compositions that can be analyzed by CAE increases, making it easier to identify crystalline resin compositions that satisfy the required characteristics of the parts.

[0065] (Other embodiments) While embodiments for carrying out the present invention have been described so far, the present invention should not be limited to the embodiments described above. In the embodiments described above, an example was described in which the PVT prediction curve calculation system 1 is provided as separate devices: a learning device equipped with a learning function and an analysis device equipped with an analysis function. However, the learning unit and the analysis unit may be provided in the same device. [Explanation of symbols]

[0066] 1. PVT Prediction Curve Calculation System 2 Learning device 3 Analysis device 4 Display device 5 Input devices 21. Training Data Generation Unit 22 Learning Department 23, 32 Control Unit 24, 33 Storage section 31 Calculation Section 100 PVT curve prediction model 321 Display Control Unit

Claims

1. Using training data in which the explanatory variables are the blending ratios of raw materials constituting a known crystalline resin composition, multiple pressures, and a continuous range of temperatures, and the dependent variable is the specific volume of the resin composition at each pressure and a continuous range of temperatures, A PVT curve prediction model generation step involves generating a PVT curve prediction model by having a computer perform machine learning. A program that generates a pre-trained model to execute the following.

2. In the PVT curve prediction model generation step, the training data is as follows: A solid specific volume prediction model that uses only data with measurement temperatures below the crystallization temperature, A melting specific volume prediction model that uses only data where the measurement temperature is above the crystallization temperature, A crystallization temperature prediction model was generated by running machine learning on a computer using training data in which the explanatory variables were the blending ratios of raw materials constituting a known crystalline resin composition and multiple pressures for a single blending ratio, and the objective variable was the crystallization temperature of the crystalline resin composition at each pressure. A trained model generation program according to claim 1, including the following:

3. Using training data in which explanatory variables are the blending ratios of raw materials constituting a known crystalline resin composition, multiple pressures, and a continuous range of temperatures, and the objective variable is the specific volume of the crystalline resin composition, a PVT curve prediction model is generated by performing machine learning. An input step in which the mixing ratio of raw materials constituting the crystalline resin composition to be predicted, the measurement pressure and the measurement temperature are input, A PVT curve prediction curve output step, which calculates a PVT curve based on the information output by the PVT curve prediction model, A PVT curve prediction program that performs this operation.

4. The learning data consists of explanatory variables being the blending ratio of raw materials constituting a known crystalline resin composition, the measured pressure and temperature, and the objective variable being the specific volume of the crystalline resin composition. The learning data is divided into two parts: the crystallization temperature of the known crystalline resin composition at each pressure is estimated from the rate of change of specific volume when the measured temperature changes at a constant pressure, and the learning data is divided into parts above and below the estimated crystallization temperature. A training data splitting program consisting of the following.

5. Using training data in which the explanatory variables are the blending ratios of raw materials constituting a known crystalline resin composition, multiple pressures, and a continuous range of temperatures, and the dependent variable is the specific volume of the resin composition at each pressure and a continuous range of temperatures, A method for generating a PVT curve prediction model by having a computer perform machine learning to generate a pre-trained model.

6. For a PVT curve prediction model generated by performing machine learning using training data in which the explanatory variables are the blending ratios of raw materials constituting a known crystalline resin composition, multiple pressures, and a continuous range of temperatures, and the dependent variable is the specific volume of the crystalline resin composition, Input the blending ratio of the raw materials constituting the crystalline resin composition to be predicted, the measurement pressure, and the measurement temperature. A PVT curve prediction method that outputs a PVT prediction curve based on the information output by the PVT curve prediction model.

7. A method for calculating fluid dynamics analysis, which performs analysis based on a PVT model equation using a PVT curve obtained by the PVT curve prediction method described in claim 6.

Images